PHD Thesis MarcoManfredi

Department of Energy
Doctoral Program in Energy and Nuclear Science and Technology
NUMERICAL AND EXPERIMENTAL CHARACTERIZATION

OF SMALL-SCALE ORC TURBINES
Doctoral dissertation of:

Marco Manfredi
Supervisors:
Prof. Andrea Spinelli, Prof. Marco Astolfi
Tutor:
Prof. Vincenzo Dossena
The Chair of the Doctoral Program:
Prof. Vincenzo Dossena
September 2023– Cycle XXXV

Copyright © 2023 Marco Manfredi
All rights reserved.

SUMMARY
The worldwide interest in reducing CO2 and greenhouse gas emissions, improving the efficiency
of fossil-fueled plants, and increasing the utilization of renewable energy systems is crucial for
achieving a sustainable growth. In this thesis the attention is posed on the efficiency improvement
of bottoming Organic Rankine cycles (ORCs), which are energy-conversion systems designed
to convert thermal energy into electrical energy. The research primarily focuses on small-power
organic Rankine cycles (mini-ORCs) as they feature unique design challenges, including complex
integration with the main plant, part-load and transient operations, and difficulties associated with
the design of small-size, highly loaded turbo-expanders. The efficiency of mini-ORC systems,
the optimal plant layout, and the working fluid selection are indeed closely tied to the achievable
performance of the turbomachinery components. However, relying solely on design practices and
guidelines developed for conventional steam and air-operating machines may lead to limitations
in the design of mini-ORC turbines, owing to the reduced-dimensions and the unconventional
gas dynamics underlying mini-ORCs. The primary objective of this thesis, subdivided into two
main parts, is therefore to investigate, analyze and characterize ORC turbines for small-power
applications, both numerically and experimentally. The original contribution of this research
is twofold: first, to enrich the dedicated design rules that can enable highly efficient mini-ORC
turbines and optimal plant layouts, and second, to provide relevant experimental data for turbine
components operated with organic fluids in non-ideal conditions.
The first part focuses on the design of bottoming ORCs for small-power applications, with
specific attention given to the performance characterization of small-scale turbo-expanders and their
impact on the selection of ORC optimal working fluid and cycle layout. A methodology is indeed
developed to integrate the turbo-expander performance evaluation within the thermodynamic
cycle optimization. To this end, a dedicated reduced-order model for the preliminary sizing
of single-stage radial inflow turbines (RITs) is developed and validated against numerical and
experimental data available in the open literature. To tailor the model to small-power high-pressure
ratio applications, a meticulous literature review and a corresponding sensitivity analysis on existing
loss correlations are performed. To integrate the turbine sizing model within the ORC design
and optimization procedure a performance map for single-stage RITs is exploited. This approach
allows therefore to design the thermodynamic cycle considering a realistic performance of the
expander and to retrieve the best cycle architectures and turbine geometry depending on heat source
characteristics and active constraints. To demonstrate the potential of the proposed methodology,
a benchmark small-power application is considered, namely the ORC-based waste heat recovery
(WHR) from the internal combustion engine (ICE) of long-haul trucks. The performance map
reports the maximum RIT total-to-static efficiency 𝜂𝑇𝑆 as function of two relevant cycle-dependent
parameters: the Size Parameter (SP), a dimensional variable accounting for machine dimension,
and the isentropic Volume ratio (Vr), namely the outlet to inlet volumetric flow ratio in isentropic
hypothesis. According to similarity theory, the effects related to the fluid molecular properties and
inter-molecular forces, defining the thermodynamic state of the fluid, are investigated to assess the
validity of SP and Vr as unique independent parameters. The results of such analyses demonstrate
minor effects of the real-gas thermo-fluid-dynamics on the attainable maximum RIT efficiency,
proving the effectiveness SP and Vr. The proposed methodology is then applied to compare the
iii
optimal cycle configurations and working fluids obtained adopting turbine performance maps
within the cycle optimization procedure rather than assuming constant efficiency values. Results
show that the assumption of constant turbine efficiency may lead to a non-optimal choice of the
working fluid for a specific plant size and to the definition of thermodynamic cycles extremely far
from the optimal ones that, in turn, can result in an unfeasible design of the turboexpander.
The second part details the experimental campaigns on an ORC supersonic linear cascade,
assembled within the Test Rig for Organic VApors (TROVA) at Politecnico di Milano. The
experiments are conceived to provide relevant experimental data for turbine components operated
with organic fluids in non-ideal conditions and, consequently, to contribute to fill the lack currently
existing in the open literature. The main objective of the campaigns is to characterize the flow
field developing through supersonic ORC nozzles in linear cascade configuration, including the
measurement of total pressure losses. The TROVA facility is firstly described, together with its
operation procedures and the available instrumentation. All the steps required to design from
scratch such novel experiments are detailed. The investigated thermodynamic conditions are chosen
to carry out expansion processes characterized by highly non-ideal fluid thermodynamic states. The
linear cascade features three complete blades, defining two central converging-diverging channels,
and two side-walls integrating partial blade profiles. This configuration is devised to make the
linear cascade representative of stator nozzles of axial/radial ORC turbines. Therefore, both blades
and side-walls are designed and optimized exploiting high-fidelity CFD simulations in order to
reproduce all the important features of these flows, namely the trailing-edge shock system, the
wake generation and recovery, and the periodicity between the two central channels. The devised
measurement strategies include the acquisition of the static pressure evolution along the centerline
of the two central channels, the employment of a double-passage schlieren equipment to visualize
the morphology of shock/fan wave structures, and a downstream traversing system to retrieve total
pressure loss pitch-wise distribution. A long-duration experimental campaign exploiting nitrogen as
working fluid is firstly carried out to demonstrate the technical relevance of the experiment and the
validity of the measurement system in a simplified thermodynamic condition. These experiments
prove an excellent repeatability and a remarkable periodicity, regardless the cascade was operated
off-design, both in terms of working fluid and inlet total conditions. Comparison with CFD
simulations of the cascade prove a satisfactory agreement both in terms of retrieved static pressures
and density gradient field versus schlieren visualizations. Conversely, the blade wakes retrieved
experimentally result shifted compared to those predicted by numerical calculations, despite a
good agreement between experimental and numerical total pressure loss peak. Following the
corrective actions taken to fix an unwanted leakage, which negatively affected the flow periodicity
and the cascade back-pressure, a long-duration experimental campaign exploiting the siloxane MM
(hexamethyldisiloxane) as working fluid is currently underway. The preliminary experiments are
carried out without implementing the total pressure probe and in off-design conditions, at lower
inlet total pressure and temperature, featuring a minimum compressibility factor of 0.65 at the
cascade inlet. The attained level of periodicity results fairly improved, the shocks reflected by
the bottom side-wall remaining the major source of non-periodicity. The comparison with CFD
predictions is outstanding both in terms of retrieved static pressures and density gradient field
versus schlieren visualization, for all the values of the inlet compressibility factor investigated.
These preliminary outcomes prove the effectiveness of the cascade design in representing the
actual flow field of supersonic ORC nozzle vanes. The remarkable agreement with high-fidelity
CFD predictions, for both nitrogen and MM, validates the theoretical tools used in the design
phase. Furthermore, it demonstrates the suitability of the employed CFD models for the design and
analysis of ORC turbines, as well as for the validation of reduced-order models and loss correlations.
The collected data aim at serving as a valuable reference for future experimental investigations on
ORC cascades, also considering the lack of data currently existing in the open literature.
iv
CONTENTS
1 INTRODUCTION 1
1.1 Research motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 State of the art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Research objectives and outcomes . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
I Mini-ORC design and characterization 11

2 RITML: A Mean-line model for single-stage RITs 15
2.1 Concept of the mean-line model . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Mean-line method structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Vaned nozzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.2 Semi-bladed Nozzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.3 Vaneless space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.4 Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.5 Diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Loss models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1 Nozzle losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.2 Vaneless Space losses . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.3 Rotor losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Test case radial inflow turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 Loss sensitivity analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6 Validation strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.7 Concluding remarks and key findings . . . . . . . . . . . . . . . . . . . . . . . . 38
3 Turbine Performance Map integration within ORC optimization procedure 41

3.1 Concept of proposed approach: integration of RIT and ORC design tools . . . . . 41
3.2 Numerical routines description . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.1 ORC design and optimization tool . . . . . . . . . . . . . . . . . . . . . 43
3.2.2 RIT optimization strategy . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 Preliminary considerations on ORC design . . . . . . . . . . . . . . . . . . . . . 46
3.4 Assessment of SP and Vr as unique independent parameters . . . . . . . . . . . . 49
3.4.1 Sensitivity analysis of optimum RIT efficiency: effects of real-gas . . . . 50
3.4.2 Sensitivity analysis of optimum RIT efficiency: effects of 𝛤 at fixed 𝑍 . . 51
3.5 RIT maximum efficiency map definition . . . . . . . . . . . . . . . . . . . . . . 53
3.6 ORC comprehensive design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.6.1 Performance investigation with working fluid R1233zd(E) . . . . . . . . 56
3.6.2 Potential fluid screening - cycle performance assessment . . . . . . . . . 58
v
II Experiments on supersonic ORC linear cascade 63

4 Experimental Facility and Instrumentation 67
4.1 Experimental facility: Test Rig for Organic VApors (TROVA) . . . . . . . . . . . 67
4.2 Adopted measurement instrumentation . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.1 Temperature sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.2 Pressure transducers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.3 Schlieren equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Data acquisition system and data post-processing . . . . . . . . . . . . . . . . . 76
5 Design of Experiment 79
5.1 Concept of the cascade experiment . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 Test section design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.2.1 Blade design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.2.2 Linear cascade adaptation . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3 Measurement strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.3.1 Measurement strategies overview . . . . . . . . . . . . . . . . . . . . . 96
5.3.2 Total pressure probe design . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4 Test section final arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6 Experimental Campaigns 109

6.1 Nitrogen experimental campaign . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.1.1 Experiments repeatability . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.1.2 Probe disturbance effects . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.1.3 Flow field periodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.1.4 CFD comparison and total pressure losses . . . . . . . . . . . . . . . . . 114
6.2 MM experimental campaigns . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.2.1 First long-duration campaign . . . . . . . . . . . . . . . . . . . . . . . . 118
6.2.2 Latest experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7 CONCLUSION AND OUTLOOK 131

7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
7.2 Outlooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
References 141
Nomenclature 151
List of figures 155
List of tables 157
vi
CHAPTER 1
INTRODUCTION
1
Chapter 1. INTRODUCTION
1.1 Research motivation

The achievement of net-zero carbon dioxide emissions by the middle of 21st century and the
reduction of climate change effects are main objectives set by the Paris Agreements and represent
pillars of the United Nations Sustainable Development Goals (SDGs), towards which the scientific
community, governments, and industries are directing common efforts. The worldwide interests in
reducing CO2 and greenhouse gas emissions, in improving fossil-fueled plants efficiency, and in
increasing the usage of renewable-energy systems are unavoidable to pursuit a sustainable growth
and presently represent the cornerstone of every research activity in the energy and environmental
field. However, as reported in the World Energy Outlook by the International Energy Agency
(2019), the global energy-related CO2 emissions would grow steadily from today’s levels before
plateauing after the mid-2040s due to the rise of energy demand (the Stated Policies Scenario
in Figure 1.1). More efforts and actions are therefore required by all those involved to limit
the temperature increase with respect to pre-industrial levels well below 2 ◦ C - the long-term
temperature goal set by the Paris Agreements - as shown by the Sustainable Policies Scenario in
Figure 1.1. The required actions to achieve above goals would compel all sectors of the economy
to significantly accelerate the uptake of low-carbon technologies, including energy efficiency and
renewables as well as nuclear and carbon capture, utilization, and storage (CCUS).
In this work the attention is posed on the energy efficiency improvement, which accounts for
about the 30% of the required CO2 emission reduction, according to the policy scenario depicted in
Figure 1.1. In particular, this work focuses on energy conversion systems that produce mechanical
energy, then possibly converted into electricity, exploiting low grade thermal energy sources, such
as biomass, waste heat recovery from energy-intensive process, solar energy, and geothermal fluid.
Historically, electric power production from thermal energy mostly relied on open Joule-Brayton
cycles operating with air and closed steam-based Rankine cycles. Although these technologies
are well established, high temperature sources are needed to achieve sufficiently high efficiency to
justify the considerable capital investment required for such complex plants. This is highlighted in
Figure 1.2, in which the second low efficiency (𝜂) is reported for Joule-Brayton cycles operating at
different limited values of 𝑇𝑚𝑎𝑥 for both recuperative and non-recuperative cycles, together with
typical values of the efficiency losses associated with the cycle components (𝛥𝜂).
Figure 1.1: CO2 emission reduction as foreseen by a Sustainable Development Scenario (i.e.
to meet the temperature goal established by the Paris Agreement) relative to the Stated Policies
Scenario (i.e. including existing policy framework and announced policy intentions) in the period
2010–2050 (International Energy Agency, 2019).
2
1.1. Research motivation
Figure 1.2: Joule-Brayton gas cycle: second-law efficiency losses at various 𝑇𝑚𝑎𝑥 (250 ◦ C, 300 ◦ C,
and 400 ◦ C) for (a) simple (b) and recuperative cycles, from Macchi (2017).
For applications characterized by low grade heat sources (𝑇𝑚𝑎𝑥 ∼ 150 − 400◦ C), Organic
Rankine Cycles (ORCs) represent an effective and promising solution that can outperform
conventional technologies based on steam and air. ORCs are phase-transition cycles based on
organic fluids, which are compounds characterized by high molecular complexity, high molar mass,
lower critical temperature with respect to water and usually retrograde or nearly vertical saturated
vapor curve in the temperature-specific entropy per unit mass plane. The possibility to choose the
operating fluid among many different organic compounds, together with their thermo-physical
properties, allow for an optimal design of the turbo-expander and for a better match between the
temperature profiles of the low heat grade hot source and of the corresponding cycle heating
(Macchi, 2017). Moreover, the typical shape of the saturated vapor curve of organic fluids leads to
dry expansions both for super- and sub-critical cycles, also for absent or very low vapor superheating.
This characteristic leads to plant simplicity and relatively high cycle thermodynamic efficiency,
and it avoids the turbine to be exposed to wetness losses and droplet damages (Macchi, 2017). The
aforementioned features of ORCs are well distinguishable in Figure 1.3, in which an example of
supercritical regenerative ORC implementing MM (hexamethyldisiloxane, C6 H18 OSi2 ) as working
fluid is reported in a temperature - specific entropy (𝑇 − 𝑠) diagram (a), together with a sketch of
ORC plant main components (b). As a drawback, organic fluids are characterized by low thermal
stability limits, either affecting or preventing their usage for high-temperature applications.
Although organic Rankine cycles stand out as an effective and mature energy conversion
technology and are successfully applied in several energy fields in the 10−1 − 101 MW power
range (Tartière & Astolfi, 2017), from geothermal energy (Astolfi et al., 2014), to biomass
(Bini & Manciana, 1996), from solar energy (Wang et al., 2011b) to stationary thermal recovery
(Campana et al., 2013), economic and technical challenges prevent their penetration in the few
kW range (Yang & Li, 2021). On the other hand, ORCs represent a promising solution in
the framework of small-power applications, such as the waste heat recovery (WHR) from the
industrial and transportation sector, due to their remarkable recovery potential. This work tackles
the investigation of small-power organic Rankine cycles (mini-ORCs) and corresponding turbo-
expanders considering a reference application: the waste heat recovery from internal combustion
engines (ICEs) on-board of innovative long-haul vehicles. In fact, ICEs release a large fraction
of the fuel energy at high-temperature (350-550°C) through the engine hot exhaust (about 30%
of the fuel inlet power) and the exhaust gas recirculation (EGR) system (about 5-10%), and at
low-temperature (100-100°C) from the charge-air cooler (about 5-10%) and the engine cooling
3
(a) (b)
Figure 1.3: (𝑎) Example of supercritical regenerative ORC implementing MM (hexamethyldis-

iloxane, C6 H18 OSi2 ) as working fluid, represented on the temperature - specific entropy diagram
together with the compressibility factor Z contour in the single phase region (Z is defined in
Equation 3.3. (𝑏) sketch of the power plant main components.
system (about 10-15%), ase detailed by Dolz et al. (2012). Moreover, ICEs of long-haul trucks
represent a challenging environment for an effective design of ORC-based WHR systems since
they feature relatively high-temperature heat sources and reduced plant size (ORC power between
10 − 50 kW), and their operating point vary almost continuously throughout the vehicle driving
cycle.
Despite the large market potential, design methods for onboard small-scale ORCs are not
trivial, requiring a meaningful working fluid and heat sources selection and an optimal design
and performance quantification of the thermodynamic cycle and of the small-scale components in
both nominal, part load and transient conditions (Xu et al., 2019; Amicabile et al., 2015). At the
same time, the intrinsic reduced dimensions of the plant and the thermo-physical characteristics of
the working fluid complicate the turbo-expander design. Organic compounds normally combine
low enthalpy drop (hence, low specific work) with low speed of sound, leading to turbines highly
sensitive to load changes and featuring few stages characterized by high expansion ratios and,
hence, by transonic or supersonic flows (Persico & Pini, 2017). The turbine fluid-dynamic design is
even more complicated when the expansion occurs in the proximity of the liquid-vapor saturation
curve, where the ideal gas law does not hold and the flow field is affected by non-ideal gas effects.
Further investigations and insights on small-power organic Rankine cycles (mini-ORCs), for
instance intended to recover the thermal power discarded as hot fluid streams from ICEs of
long-haul trucks, acquire remarkable importance to boost their competitiveness. In addition, due
to the lack of consolidated expertise and design guidelines, utmost attention must be payed to
the turbo-expander design, since this component usually operates in non-ideal compressible fluid
regimes and its efficiency strongly impacts on the overall cycle performance, plant profitability, and
optimal working fluid and design parameters definition (Colonna et al., 2015; Macchi & Astolfi,
2017b).
1.2 State of the art

Despite their remarkable recovery potential from ICEs, especially in the perspective of power pro-
duction on-board of innovative long-haul trucks, ORC systems are still far from commercialization
due to technical and economical issues, even if market leader companies (e.g. Mahle-Bosch, AVL,
Cummins, Volvo) have already proved the concept effectiveness (Glensvig et al., 2016).
4
1.2. State of the art
While ORCs exploit turbo-expanders in the power range (10−1 to 101 MW) typical of current
commercial systems (Colonna et al., 2015), small-power and low-temperature applications are
currently dominated by volumetric expanders (Lemort et al., 2013; Cipollone et al., 2014). Scroll,
screw and reciprocating machines are characterized by low specific cost, high reliability and
robustness, but they are affected by high leakage losses and by the need of lubricating oil, which
may lead to a reduction of the cycle thermal efficiency due to the oil outflows within the cycle
(Imran et al., 2016). Volumetric machines are also characterized by low maximum theoretical
efficiencies at medium/high optimal specific speed (> 10) and they are constrained by limited
volumetric expansion ratios, making unfeasible their efficient employment in high-temperature
applications (𝑇𝑚𝑎𝑥 ∼ 200 − 300◦ C) (Sprouse III & Depcik, 2013).
On the contrary, turbo-expanders do not require lubrication except for bearings, they can bear
larger expansion ratios with higher maximum theoretical isentropic efficiencies, and feature larger
power density. However, turbine performance may easily drop if the specific diameter is constrained
to sub-optimal values for a given specific speed and, additionally, consolidated design guidelines
are not yet available for small-power output ORC machines. Axial or radial-outflow turbines
are generally used in single or multistage architectures for large power range, while radial inflow
turbines (RIT) are usually preferred for high-temperature mini and micro-ORC power systems
(Persico & Pini, 2017). In this kind of applications, an optimal design of the thermodynamic
cycle usually implies reduced condensation pressure and, hence, large expansion ratios across the
expander. This motivates the widespread choice of implementing RITs, which are characterized by
high compactness, large power density, and capability to handle high pressure ratios (PRs) within a
single-stage configuration.
Many procedures were proposed in the open literature concerning design of bottoming ORC
power plants for on-board WHR applications. Most simple design methods perform either
thermodynamic or techno-economic single-objective ORC optimization (Dai et al., 2009; Astolfi
et al., 2014; Tian et al., 2012), often performing parametric analyses to identify the best working
fluids for a given application (Saleh et al., 2007; Wang et al., 2011a). More exhaustive analyses
include sizing models of the heat exchangers and multi-objective optimization procedures to find
the best trade-off between thermodynamic and economic performance, such that a very high
conversion efficiency can be achieved still limiting the system cost and/or dimensions (Quoilin
et al., 2011; Rosset et al., 2018). The design method may be structured to integrate a unique
or a discrete number of heat thermal inputs (Hountalas et al., 2012; Dolz et al., 2012; Ringler
et al., 2009) and the fluid selection may be automatically included in the optimization problem
(Papadopoulos et al., 2013). Although the ORC power plant design is often based on a specific
set of hot source mass flow rate and temperature, related to a single engine operating point, some
studies take into account off-design conditions to globally improve ORC performance throughout
the engine driving cycle (Lin et al., 2019).
The design procedures of bottoming ORCs available in the open literature are often more
simplified with respect to current industrial state-of-art and usually adopts simplifying assumptions
relative to the performance of the components. In particular, the turbine efficiency is generally kept
constant (Dai et al., 2009) even though its quantification during the cycle optimization procedure is
of paramount importance because of the large impact on the definition of optimal working fluid
and design parameters (e.g. maximum and minimum pressure of the cycle). In the framework
of small-scale ORC design for onboard WHR, assuming a constant turbine efficiency may result
into misleading optimal cycle configurations having turbine expansion ratios and dimensions
non compatible with the assumed turbine efficiency, especially if the number of stages and the
rotational speed are constrained by rotodynamic, available space and economic factors. Similarly,
a coupled sensitivity analysis on turbine efficiency (Bademlioglu et al., 2018) results ineffective to
optimize the ORC design variables, since they are only slightly affected by the specific value of
5
the turbine efficiency. An apparently more reliable and complex iterative procedure, in which the
ORC and turbine optimization routines are iteratively alternated (Meroni et al., 2018a) turns out
to be also inadequate, the turbine efficiency being still constant at each ORC optimization step,
despite estimated with a dedicated design tool. A more reliable method would instead integrate
the turbine efficiency quantification into the ORC optimization process to account for realistic
turboexpander performance as function of the inlet/outlet thermodynamic states at each ORC
design iteration step. This is usually achieved through turbine reduced-order (or mean-line) models,
capable of providing preliminary sizing and performance evaluation with a limited computational
cost. The most rigorous method would require the implementation of a dedicated turboexpander
optimization routine within the ORC cycle optimization tool, however this approach implies a
non-feasible computational cost (even using mean line models) since several hundreds of turbine
mean-line model evaluations should be performed at each ORC design iteration step. Alternatively,
a higher-order optimization may be performed by varying concurrently the ORC design parameters
and the geometric variables of the turbine model (Bahamonde et al., 2017a). However, this approach
may converge to local minima, due to the different effects of geometric and thermodynamic design
variables on the turbine performance and may lead to suboptimal solutions. A low number of
turbine and cycle optimized parameters is indeed usually adopted in order to reduce the complexity
of the numerical problem. Finally, the turbine design tool may be exploited to retrieve performance
maps as a function of certain cycle-dependent non-dimensional parameters, to be included in the
cycle optimization routine. This approach was used for both axial (Astolfi & Macchi, 2015; La Seta
et al., 2016) and radial-inflow turbines (Perdichizzi & Lozza, 1987; Da Lio et al., 2017a; White &
Sayma, 2019).
As previously stated, the expander represents a crucial component in organic Rankine cycle
power systems and its efficiency strongly impacts the definition of optimal working fluid, cycle
layout and overall performance, and plant profitability. Utmost attention is therefore usually payed
to its fluid dynamic design and high-fidelity design/optimization tools based on computational fluid
dynamics (CFD) are nowadays largely employed, especially in the detailed design phase. However,
during preliminary system design (involving fluid selection, cycle layout definition, specification of
component architecture, as heat exchangers and fluid machines, and their initial sizing), the high
computational effort typical of high-fidelity expander design tools prevent their integration in the
overall optimization process, due to the considerable number of simulations initially required. A
viable option to integrate the expander preliminary design within the ORC system optimization
consists in the adoption of reduced-order methods, capable of providing machine preliminary
geometry, efficiency, and off-design performance with a reduced computational cost. Referring to
turbo-expanders, preferred for high temperature/high expansion ratio ORCs and under study in this
work, such tools are the so called mean-line methods, which are low-fidelity approaches based
on the solution of 1D continuity, energy, and momentum equations, alongside loss correlations.
Several reduced-order models are available in the open literature for both axial (Persico & Pini,
2017; Salah et al., 2022) and radial-inflow turbines (Glassman, 1995; Meroni et al., 2018b).
The capability and reliability of mean-line models in correctly predicting turbine efficiency
depend on the effectiveness and ranges of validity of the adopted loss correlations. Several loss
models, spanning a wide range of empiricism level, are available in literature for both axial and
radial turbine architectures (Vavra, 1969; Craig & Cox, 1970; Ainley & Mathieson, 1951; Glassman,
1976a; Baines et al., 1998). The weakness of such relations is that they were developed for machines
operating with conventional fluids and validated against test cases concerning standard gas or
steam turbines, while their verification against experimental data over ORC expanders is largely
lacking. An approach commonly adopted is to validate reduced-order models and loss correlations
using high-fidelity CFD simulations (Fiaschi et al., 2012; De Servi et al., 2019). However, the
verification of CFD simulations relies only on the accuracy of state-of-the-art thermodynamic
6
1.3. Research objectives and outcomes
fluid models (Span & Wagner, 2003) or, at most, on experimental data related to paradigmatic
compressible flows, such as isentropic nozzle expansions (Spinelli et al., 2018; Robertson et al.,
2019; Beltrame et al., 2021) or flows around aerodynamic (Zocca et al., 2019) or bluff bodies
(Reinker et al., 2021), where loss impact on the flow field substantially differs from the case of
turbine channel flows. The only exception is the work by (Baumgärtner et al., 2019), in which
total pressure measurements were performed downstream of a micro supersonic annular cascade
operated with air, carbon dioxide, and refrigerant R134a. However, the total pressure was measured
downstream of the probe-induced shock, while the actual cascade-exit total pressure was retrieved
numerically. Moreover, the measurement space resolution was limited by the miniaturization of the
cascade and, in the test conditions, air and carbon dioxide exhibited an ideal-gas behavior, while
only R134a showed mild non-ideal effects (minimum compressibility factor 𝑍 = 0.88, where Z
quantifies the deviation of the fluid behavior from the ideal gas one and it is defined as 𝑍 = 𝑃/𝜌𝑅𝑇,
where 𝑃 is the pressure, 𝑇 the temperature, 𝜌 the density and R the specific gas constant). As
a result, the experiments documented in (Baumgärtner et al., 2019) were essentially devoted to
evaluate the effect of fluid molecular complexity in wake losses. Therefore, experimental data
concerning non-ideal and supersonic flows representative of ORC turbine cascades are still missing
in the open literature. Experiments reproducing all the important features of these flows, namely
the trailing-edge shock system, the wake generation and recovery, and the total pressure loss, would
result instrumental for constructing and validating high-fidelity CFD models implemented into
ORC turbine design and analysis tools.
1.3 Research objectives and outcomes

Referring to on-board waste heat recovery from ICEs as a benchmark small-power application,
several factors complicate the design of mini-ORCs: i) the integration of the bottoming power plant
with the main system, in terms for instance of optimal heat sources selection, management of the
additional mechanical power produced, and sealing of the WHR unit, ii) the choice of an optimal
working fluid, which needs to maximize cycle performance and concurrently ensure zero ODP,
very small GWP, and low safety issues, iii) the optimization of the cycle layout, and iv) the optimal
design of all the small-scale components, which need to effectively convert the available thermal
energy both in nominal, part-load, and transient conditions. Moreover, the small-power levels
involved, usually resulting from quite large expansion-ratios through the expander in concurrence
with very small mass flow rates of the organic fluid, may prevent to design turbines featuring
sufficiently high values of the efficiency both in nominal and off-design conditions to outperform
other expander architectures, such as volumetric machines.
In this framework, it results clear that an estimation of the actual turbine efficiency during the
preliminary steps of the ORC design is of paramount importance, as well as the assessment of
its effects on optimal fluid and thermodynamic cycle design variables selection. To this end, the
first part of this work focuses on the definition of an effective procedure to integrate the turbine
performance prediction tool within the ORC optimization process, based on the development of a
maximum efficiency performance map for single-stage radial inflow turbines.
This map reports the maximum RIT total-to-static efficiency 𝜂𝑇𝑆 as function of turbine Size
Parameter (SP), a dimensional parameter accounting for machine dimension, and isentropic Volume
ratio (Vr), namely the outlet to inlet volumetric flow ratio in isentropic hypothesis, accurately
defined in (Macchi, 1977) and in Equation 3.1 of Section 3.1. These parameters were chosen
since their definition accounts to some extent for the effects of fluid properties (complexity and
non-ideality) and they were identified as the most adequate independent parameters from previous
studies on ORC axial and radial turbines (Macchi & Perdichizzi, 1981a; Masi et al., 2020; Astolfi &
Macchi, 2015). Differently from the approach suggested by (White & Sayma, 2019), in which the
7
map was retrieved by means of a scaling procedure applied to the results published by (Perdichizzi
& Lozza, 1987), the performance map proposed in this work was obtained performing expander
optimizations for a discrete number of SP and Vr values. Specifically, in the proposed method
the RIT optimization was carried out exploiting an in-house reduced-order model, which adopts
many design variables and several constraints, that have been included to obtain feasible turbine
geometries. This represents a further improvement with respect to the approach of (Da Lio
et al., 2017a), where maps were obtained exploiting a bivariate optimization procedure based
on specific speed and velocity ratio. Therefore, the map here reported defines an ultimate upper
bound for the RIT efficiency. Also, the ranges of Vr and SP investigated are wider than those
reported in literature (Da Lio et al., 2017a) and shifted towards higher Vr and lower SP, due
to the high-temperature and small-power applications here investigated. This required to focus
and analyze converging-diverging nozzles in choking conditions, while previous studies only
investigated converging-nozzle architectures (Perdichizzi & Lozza, 1987; Da Lio et al., 2017a).
Since in mini-ORC power systems the turbo-expander involves the major technical challenges,
as the demand for compactness and flexibility of operation couples with severe compressibility and
non-ideal gas effects, the primary objective of this work is to investigate, analyze and characterize
ORC turbines for small-power applications. This is the reason why the RIT reduced-order model
exploited to derive the performance maps just described was developed in-house. Even though
several turbine mean-line models are available in the open literature, the reduced-order model
for single-stage RIT described in this work results from dedicated analyses over existing loss
models, which allowed to identify their limits of validity, making the mean-line model tailored
to machines operating with organic fluids for small-power, high-pressure ratio applications. The
model is intended to be employed not only to integrate turbine performance estimation within the
ORC design tool, but also for an early and complete sampling of the design space, which allows
to adopt solutions typically beyond the standard design practices and to derive design guidelines
for small-scale ORC RITs, targeting an efficiency improvement both in design and off-design
conditions.
As previously anticipated, another crucial aspect is the lack in the open-literature of relevant
experiments over turbines or linear/annular cascades operated with organic fluids in non-ideal
conditions. In fact, a combination of reduced-order models and high-fidelity CFD-based tools
is usually employed to design and investigate such kind of machines. This is to be ascribed to
the complexity of the flow fields, to the technical challenges of performing measurements on
supersonic flows at thermodynamic conditions close to saturation, and to the relatively high costs
and difficulties associated with the management of wind tunnels operating with organic compounds,
since the fluid is stored and operated at usually high temperatures (150 − 250◦ C) and pressures
(10 − 25bar). To fill this gap, the second part of this work focuses on the design and the subsequent
execution of an experimental campaign on a supersonic ORC nozzle linear cascade, assembled
within the Test Rig for Organic VApors (TROVA) at Politecnico di Milano. The test section consists
in a supersonic converging-diverging nozzle blade row, designed to operate at nominal outlet
mach number of 1.6 (typical of highly loaded stages) with siloxane MM (hexamethyldisiloxane), a
working fluid employed in high temperature ORCs. Firstly, this work describes the conception
of such a novel class of experiments together with all steps required to design the campaign,
from the choice of operating thermodynamic conditions, to the set-up of suitable measurement
strategies, from the blade design and optimization, to the final test section arrangement, which
needed to comply with space constraints dictated by the test section geometry, while keeping at
the same time the high spatial resolution needed for CFD validation. Secondly, the experimental
data resulting from execution of different campaigns exploiting either nitrogen or MM as working
fluids are described and compared with high-fidelity CFD results. Thanks to the heterogeneity
of implemented measurement techniques, these experiments allow to characterize the flow field
8
1.4. Outline of the thesis
associated with the expansion of an organic fluid in non-ideal conditions through a supersonic
linear cascade representative of stator nozzles of axial/radial ORC turbines, and they aim at serving
as a reference for future experimental investigations on ORC cascades.
1.4 Outline of the thesis

The thesis is structured such that two main parts composing the research activity can be distinguished:
the mini-ORC power plant design and optimization, with a particular focus on the turboexpander
and on the analysis of its performance characteristics; the design and execution of the experimental
campaign on supersonic ORC nozzles in linear cascade configuration.
Part I concerns the design of bottoming ORCs for small power applications and delineates
the methodologies developed to integrate the turbo-expander performance evaluation within the
thermodynamic cycle optimization. To demonstrate the effectiveness of the proposed approach a
challenging environment such as ORC-based on-board waste heat recovery from ICEs is considered
as the reference small-power application. First, the reduced-order model developed to preliminary
size the turbo-expander and assess its performance is described in Chapter 2. To tailor the
turbine mean-line model to small-power high-pressure ratio applications, a sensitivity analysis on
existing loss correlation is performed, while a wide and strong validation strategy is proposed to
asses the robustness and reliability of the model. The numerical routine developed to design the
thermodynamic cycle is then described in Chapter 3 and the cycle-turbine design tools integration
is tackled, delineating the approach employed to derive the RIT maximum performance map. This
procedures allow also to retrieve some useful guidelines concerning the optimal design variation
of single-stage RIT at different SP and Vr values. Before integrating the expander performance
estimation, several hundreds of cycle are designed at constant turbine efficiency to assess the
design space in terms of SP and Vr for the specific small-power application considered. The
effectiveness of SP and Vr as the solely independent parameters for the development of the map is
investigated. To assess the dependence of the maximum efficiency on other parameters, different
turbine geometries are indeed optimized for a fixed set of SP and Vr, by varying either the inlet
compressibility factor or the working fluid. Finally, the effects of integrating turbine performance
prediction tool within the cycle optimization routine are analyzed, by adopting several working
fluids and investigating different amounts of available thermal power (plant sizes). These analyses
highlight the differences between the constant turbine efficiency approach and the performance
map integration method, in terms of optimal cycle layout and working fluid selection
Part II details the experimental campaign on an ORC supersonic linear cascade, assembled within
the Test Rig for Organic VApors (TROVA) at Politecnico di Milano. Firstly, TROVA facility is
described in Chapter 4, together with its operation procedures and the commercial off-the-shelf
(COTS) instrumentation available to be employed in the campaign. Secondly, the conception
and design of the experiment is reported in Chapter 5, reviewing all the steps required to design
such a novel experimental campaign. After the description of the concept of the experiments
and the discussion of chosen thermodynamic conditions, the detailed blade and side-walls design
is investigated, including profile shape-optimization and cascade periodicity assessment. The
set of employed measurement techniques is then detailed, together with a thorough description
of the measured strategy employed to retrieve total pressure downstream the cascade. In fact,
the total pressure probe has to comply with the reduced dimensions of the test section without
excessively affecting the upstream flow field and to concurrently withstand the large aerodynamic
loads resulting by the high pressures involved. In the last part of Chapter 5 an overview of the
final test section arrangement is reported, including 3D CAD models of the main components
and practical solutions to deal with test section reduced dimensions. The last part of the thesis,
9
Chapter 6 details all the experimental campaigns carried out, exploiting both nitrogen and MM
as working fluids, and delineates their outcomes both in terms of flow field characterization and
comparison with high-fidelity CFD simulations. The first section presents the findings from a
long-duration campaign conducted exploiting nitrogen at moderate inlet pressure. The experimental
data thoroughly described are extremely encouraging. In the second section, the first long campaign
with MM is introduced, emphasizing the challenges encountered during the experiments. Finally,
the corrective measures undertaken are described, along with the latest experiments conducted with
both nitrogen and MM. The collected experimental data are meticulously analyzed and represent a
valuable reference for future experimental investigations on ORC cascades.
10
PART I
MINI-ORC DESIGN AND
CHARACTERIZATION
This part of the thesis illustrates the strategies developed to design ORCs for small-power
applications (mini-ORCs) and to integrate the turbo-expander performance evaluation within the
thermodynamic cycle optimization, considering ORC-based waste heat recovery from internal
combustion engines of long-haul trucks as the reference small-power application. In fact, ICEs of
long-haul trucks represent a challenging environment for an effective design of ORC-based WHR
systems since they feature relatively high-temperature heat sources and reduced plant size (ORC
power between 10 − 50 kW), and their operating point varies during the vehicle driving cycle. This
characteristics complicate the design of the turbo-expander, possibly preventing to obtain values of
the efficiency both in nominal and off-design conditions large enough to outperform other expander
architectures, such as volumetric machines. In this framework, the estimation of the actual turbine
efficiency, the assessment of design feasibility, and the development of specific design guidelines for
mini-ORC turbines are of paramount importance during the preliminary steps of the ORC design.
Such a focus on the turbo-expander and on its performance is indeed required for mini-ORC power
plants, most likely characterized by miniaturized machines, especially for those applications, such
as on-board WHR, in which the turbine number of stages and/or rotational speed cannot be freely
chosen. The numerical routines exploited to design and optimize both the thermodynamic cycle and
the turbo expander, a single-stage radial inflow turbine, are described, as well as the methodology
developed to integrate the turbine performance estimation within the ORC design tool. Several
analyses are undertaken to show the effects of such integration, highlighting the impact on optimal
cycle layout definition and optimal working fluid selection based on the power-plant size.
12
Some contents of this part are also discussed in:
Manfredi, M., Alberio, M., Astolfi, M., & Spinelli, A., (2021) A reduced-order model for the
preliminary design of small-scale radial inflow turbines, Turbo Expo: Power for Land, Sea, and
Air, vol. 84966, American Society of Mechanical Engineers, doi: 10.1115/GT2021-59444
Manfredi, M., Spinelli, A., & Astolfi, M., (2023) Definition of a general performance map for
single stage radial inflow turbines and analysis of the impact of expander performance on the
optimal ORC design in on-board waste heat recovery applications, Applied Thermal Engineering,
vol. 224: 119857, doi: 10.1016/j.applthermaleng.2022.119857
13
14
CHAPTER 2
RITML
A Mean-line model for single
stage radial inflow turbines
2.1 Concept of the mean-line model

The small size associated with low-power applications together with the peculiar characteristics
of organic compounds, which normally combine low enthalpy drop (hence, low specific work)
with low speed of sound, and with the non-ideal effects in thermodynamic conditions of interest
for ORC greatly complicate the design of the expander. ORC turbines generally feature very few
stages (from one to three), transonic or supersonic flows, and large cross section variation, required
to deal with the relevant increase of volumetric flow rate (density reduction) as the flow expands
through the machine. Axial turbines are generally used in single or multistage architectures for
the larger power range, while radial inflow turbines (RIT) are usually preferred for small-scale,
high-temperature power systems (Persico & Pini, 2017). In this kind of applications, an optimal
design of the thermodynamic cycle usually implies low condensation pressure, which depending
on temperature of the cold sink and on the organic fluid considered may reach sub-atmospheric
values, and large expansion ratios across the expander. This motivates the widespread choice of
implementing RITs, which are characterized by high compactness, large power density, capability
to handle high pressure ratios within a single-stage configuration and to operate at high flow
coefficients. However, these features generally lead to transonic-to-supersonic radial nozzle vanes
and transonic mixed-flow rotors.
Although experimental data on ORC turbines are mandatory - and still missing in the open
literature - for the validation of computational tools, reduced order (or mean-line) turbine’s models
15
Chapter 2. RITML: A Mean-line model for single-stage RITs
are usually employed during the ORC preliminary system sizing and optimization. These design
procedures involve fluid selection, cycle layout definition, specification of components’ (e.g. heat
exchangers and fluid machines) architecture, and preliminary sizing. This is due to the fact that the
large computational cost associated with CFD-based design tools prevent their employment in this
phase of the design process. In this framework, reduced-order methods based on the solution of 1D
continuity, energy, and momentum equations, alongside loss correlations (Persico & Pini, 2017),
are capable of providing machine preliminary geometry, efficiency, and off-design performance
with a reduced computational cost. The capability and reliability of mean-line models in correctly
predicting turbine efficiency depends on the effectiveness and range of validity of the adopted loss
correlations. Several loss models, spanning a wide range of empiricism level, are available in
literature for both axial and radial turbine architectures (Ainley & Mathieson, 1951; Vavra, 1969;
Craig & Cox, 1970; Glassman, 1976a; Baines et al., 1998). The major conceptual weakness of
such relations is that they were developed for machines operating with conventional fluids and
validated against test cases concerning standard gas or steam turbines, while their verification
against experimental data over ORC expanders is largely lacking. For this reason, empirical design
guidelines, coming from the experience gained in steam- and air-based applications, should be
applied with extreme caution to this novel class of turbomachinery, especially when small-power
miniaturized machines are involved. Such guidelines may indeed possibly lead to sub-optimal
designs featuring different (and generally lower) performance than expected, as many aerodynamic
features are usually neglected in the process. The development of tailored design/analysis tools for
mini-ORC turbines is therefore mandatory to improve this technology beyond what achievable by
the direct transfer of the design guidelines developed for conventional gas or steam turbines.
Several reduced-order models have been proposed in literature, covering a wide range of design
power outputs, mass flow rates and expansion ratios. However, mean-line models for small size
RITs operating with organic fluids and related validation strategies are limited. To the author’s best
knowledge, most of the existing tools have been validated over experimental data for low-pressure
ratio air operating turbines, also due to the lack of published experimental data concerning ORC
turbines. Developed models can be divided into two main categories, depending on the set of
design variables (DVs) and constraints selected. Many authors (Ventura et al., 2012; Fiaschi et al.,
2015; Alshammari et al., 2017) employed the methodology proposed by Moustapha (Moustapha
et al., 2003), in which several non-dimensional parameters – such as work and flow coefficients –
are provided as inputs (DVs) to compute the RIT geometry which meets the design requirements.
This approach ensures a direct link between the geometry and the main performance parameters,
leading to a fast investigation of the allowable design space. As a drawback, these methods are
not well-suited to study the effect on performance of turbine’s configuration (e.g. stage counts)
or geometry (e.g. converging-diverging instead of simply converging nozzles). he other major
design methodology, exploited for example by Meroni and Da Lio (Meroni et al., 2018b; Da Lio
et al., 2017b), provides most of the turbine geometric parameters as DVs to evaluate each loss
contribution and, consequently, the efficiency. This approach is more flexible and allows for a
step-by-step solution of the flow field within the turbine components, but, depending on the chosen
geometry, may lead to sub-optimal efficiencies and, in the worst-case scenario, to non-physical
flow fields.
Fiaschi et al. (2015) developed a mean-line model to investigate the effects of fluid characteristics
and rotor inlet blade configuration on the RIT performance. The author designed six turbines to
be operated with different organic fluids, with a 50 kW target power output. For all the fluids
considered, radial rotor blades (at inlet) resulted less efficient than backswept ones, which are
characterized by a positive inlet blade angle, namely concordant with the peripheral speed direction.
In a subsequent study Fiaschi et al. (2016) exploited the same mean-line method to design a 5 kW
turbine operating with R134a. The model results were validated over numerical data extracted
16
2.1. Concept of the mean-line model
from three-dimensional (3D) RANS simulations of the same geometry. The reduced-order method
proposed by the author is targeted to medium expansion ratios (about 4), considers subsonic to
transonic flows (only converging nozzles are included in the analysis), and is not validated over
other geometries, fluids, or flow conditions. Alshammari et al. (2017) developed a mean-line
model for the preliminary design of a RIT for WHR from an off-highway heavy-duty Diesel engine
(HDDE). The turbine was designed to operate with NOVEC 649 (a fluoroketone commercialized
by 3M) and to produce a power output of 18 kW for a pressure ratio of 13. Similarly to Fiaschi
et al. (2015), also this work proved an efficiency benefit from backswept blades configuration. The
geometry resulting from the mean-line model was manufactured and tested (Alshammari et al.,
2018) in off-design conditions only, leading to a measured turbine efficiency of 35%, against the
81.3% resulting from CFD computations at the design point. A validation strategy of the model
over other geometries, fluids, or flow conditions is missing.
The model developed by Meroni et al. (2018b) focuses on RITs operating with air and
characterized by pressure ratio up to 5. The authors thoroughly described the geometry generation
process, the performance evaluation strategy at nominal design point and the method to compute
the performance maps in off-design conditions. However, the model structure conceives only
radial-type rotor inlet blades, a purely converging nozzle and does not allow for choking conditions
at the design point. As final step of the model development, a calibration procedure of the
loss correlation parameters was performed over the experimental data concerning six different
air-operating RIT geometries. This procedure may lead to a mean-line model excessively tailored to
turbines featuring similar characteristics to those employed for tuning the loss coefficients, possibly
implying lower accuracy when different fluids and/or applications are considered. However, the
calibrated mean-line method was afterwards tested over three additional turbines, two of which
operating with organic fluids.
The preliminary design model devloped by Da Lio et al. (2017a) can be classified as a direct
approach, even though the fluid-dynamic design variables comprise the specific rotational speed and
the velocity ratio, defined as 𝑁 𝑠 = 𝑁 𝑄¤ 0.5
𝑜𝑢𝑡 /𝛥ℎ𝑖𝑠 and 𝑉𝑠 = 𝑈/(2𝛥ℎ𝑖𝑠 ) , respectively. Boundary
0.5
conditions for the model are the total inlet thermodynamic state, the operating fluid, the mass flow
rate, and the desired total-to-static expansion ratio. The geometric design variables are mostly
defined as non-dimensional ratios, while the constraints introduced by the author are mainly related
to the rotor geometry. The reduced-order model was validated over existing ORC turbines of large
size (12 MW nominal power), theoretical studies carried out by Perdichizzi & Lozza (1987), and a
5 kW turbine resulting from 3D CFD-simulations (Fiaschi et al., 2016). The loss model proposed
by Aungier (2006a) was employed for both nozzle and rotor, resulting in an overestimation of
the total-to-static efficiency, according to the validation outcomes. The model however proved a
satisfactory capability of predicting the optimal 𝑁 𝑠 and 𝑉𝑠 required to maximize the efficiency.
The optimization of several turbine geometries revealed that the optimal 𝑁 𝑠 is comprised between
0.41 and 0.42 regardless of the required expansion ratio. On the contrary, the optimal 𝑉𝑠 turned
out to be a function of the required expansion ratio: it approaches 0.70 for lower values, while it
decreases down to 0.60 for higher expansion ratios.
Pini et al. (2017a) performed a preliminary design of a RIT operating with MM and sized for
10 kW gross power output, by means of the mean-line code presented in Bahamonde et al. (2017b).
3D RANS CFD computations were then performed for code verification on the resulting geometry,
featuring a converging-diverging nozzle architecture, as reported by De Servi et al. (2019). Other
validation test cases are missing in the open literature.
This chapter focuses on the development of a reduced-order model for the preliminary design of
single-stage radial inflow turbines, named RITML and focused on machines operating with organic
fluids for small-power, high-pressure ratio applications. The reason why the mean-line model was
developed in-house and specifically targeted to small-power ORC turbines is twofold. Firstly, the
17
tool is intended to be employed for an early and complete sampling of the design space, aiming
at the development of specific design guidelines for mini-ORC turbines that allow engineers to
adopt solutions typically beyond the standard design practices. Secondly, the model coupled with
an external optimization algorithm, aiming at maximizing turbine efficiency, can be exploited to
integrate the RIT performance estimation within the ORC design and optimization routine. In this
way an optimal value of the turbine efficiency for each set of inlet/outlet conditions is considered
rather than assuming a constant (and often non justified) value.
The method, described in Section 2.2, allows to include in the analysis both converging and
converging-diverging nozzle geometries, often required by ORC applications. Different correlations
for stator and rotor passage loss contributions are investigated in detail in Section 2.3 due to
their large contribution to the overall RITs’ efficiency, highlighting their level of empiricism.
Additionally, passage loss models are tested against published data on both air-operating or ORC
turbines. These loss sensitivity analyses, reported in Section 2.5, underline the validity, robustness,
and intrinsic limitations of the various correlations, leading to a final choice of physically-based
passage loss models targeted to small-scale, high-expansion ratio turbines. The ensuing mean-line
model results to be reliable, robust, and characterized by good performance prediction capabilities.
These features are confirmed by testing the reduced-order method (see Section 2.6) over published
experimental/numerical data of seven turbines, operating with different fluids, in both design and
off-design conditions, and covering a wide range of target expansion ratio and gross power output.
2.2 Mean-line method structure

The RIT mean-line model RITML is implemented in a routine developed in the Python language
and adopting CoolProp (Bell et al., 2014) for the calculation of fluid thermodynamic properties,
eventually recalling the NIST REFPROP backend (Lemmon et al., 2018). This feature makes the
model flexible to deal with any fluid available in CoolProp (or REFPROP) database.
The reduced-order method developed belongs to the geometry-based approaches (see Section
2.1) and, consequently, the geometrical parameters, together with mass flow rate, rotational speed,
and inlet total conditions, represent model inputs. This choice enhances the code flexibility and
allows to analyse both subsonic and supersonic flows through the nozzle. As standard practice and
without any loss of generality, the model is based on the hypotheses of steady state conditions and
of adiabatic machine. The design includes the nozzle, either converging or converging-diverging,
the vaneless interspace, the rotor, with the option of either radial or backswept blades at the inlet,
and a conical diffuser that can be optionally included downstream of the rotor. The schematic
reference geometry together with the employed nomenclature are reported in Figure 2.1, in which
the two possible nozzle architectures are also sketched. Notice that for subsonic flows within the
stator, the velocity and thermodynamic state at the nozzle exit station (labeled as 3 in Figure 2.1)
are supposed to be equal to those existing at the last section of the nozzle bladed channel (i.e. the
throat section, labeled as 3∗ in the left side of Figure 2.1). Conversely, for either a sonic converging
nozzle or a supersonic converging-diverging one, the two stations are investigated separately, since
a post-expansion (or post-compression) may occur for 𝑝 3 ≠ 𝑝 ∗3 when the mass flow rate is choked,
as illustrated in the detail of Figure 2.1.
As shown in the flow-chart reported in Figure 2.2a, the code can be used in two different modes.
In the analysis mode, the geometry must be fully provided by the user - even if the code is robust
enough to set some geometrical paramters if not provided (such as blade trailing edge thicknesses
and clearances) - and the code evaluates the turbine performance, in terms of losses, efficiency
and total-to-static pressure ratio 𝛽𝑒,𝑇𝑆 . This mode can be exploited to carry out the validation
process and to evaluate the off-design performance of a turbine designed from scratch, leading
to the development of the performance maps. When the nozzle is choked, the turbine rotational
18
2.2. Mean-line method structure
*
M*3 3 b1
A STATOR
3* C
o3* 3 b3
o2
s3 * b b4
o3* 3* 3 a
s3
B
4
3 M3 3
r6tip
3 r
BACKPLATE
r6rms
r4 b6
r1 ROTOR
b6 r3 r6hub
r6 Lz
Figure 2.1: Schematic reference geometry. Left: view normal to the axis of rotation. Right:
meridional plane view Middle: detail illustrating post-expansion process.
Table 2.1: Mode-dependent inputs to the mean-line model.
Parameter Analysis Design

𝛽𝑒,𝑇𝑆 Computed Input
Nozzle Geom. Input Different geometries may be investigated
𝛽4,𝑏𝑙 Input Computed to minimize rotor incidence loss
𝛽6,𝑏𝑙 Input Computed to meet the 𝛽𝑇𝑆 required
speed and the mass flow rate are not sufficient to define the operating point (i.e. 𝛽𝑒,𝑇𝑆 ), the mass
flow rate being independent on machine outlet pressure. As better explained in Sections 2.2.2
and 2.6, the post-expansion static pressure ratio 𝑝 ∗3 /𝑝 3 has been selected as additional parameter
required for the model closure. In the design mode, 𝛽𝑒,𝑇𝑆 is additionally assigned and the code
evaluates the rotor inlet blade angle 𝛽4,𝑏𝑙 to minimize incidence losses and the rotor outlet blade
angle 𝛽6,𝑏𝑙 required to achieve such expansion ratio, with all the other geometrical parameters
provided by the user. Hereafter, 𝛼 and 𝛽 indicate the flow angles in the absolute and relative
reference frame respectively, both intended as measured from the meridional direction and positive
when the tangential component of the velocity to which they refer has the same orientation as the
rotor peripheral speed. The same symbols, when coupled with the subscript 𝑏𝑙 indicates the blade
metal angle. Design or analysis mode is automatically activated depending on whether or not the
total-to-static pressure ratio 𝛽𝑒,𝑇𝑆 is provided as input. A list of mode-dependent inputs is reported
in Table 2.1.
If design mode is selected, a preliminary calculation (Pre Processing block in Figure 2.2a) is
carried out to compute the nozzle choking mass flow rate 𝑚¤ 𝑐ℎ corresponding to a throat section
∗ 𝑠𝑢𝑏
equal to 𝑜∗3 , the nozzle outlet subsonic and supersonic isentropic Mach numbers, respectively 𝑀3,𝑖𝑠
∗ 𝑠𝑢 𝑝
and 𝑀3,𝑖𝑠 , and the throat opening and area (𝑜 𝑐ℎ and 𝐴𝑐ℎ respectively) leading to nozzle choke at
the given mass flow rate. This means that for supersonic converging-diverging architectures, the
codes adapts the nozzle throat section 𝐴𝑡 ℎ - setting 𝐴𝑡 ℎ = 𝐴𝑐ℎ - to process the required mass flow
rate given the operating fluid and the turbine inlet conditions, assuming an isentropic flow between
nozzle leading edge and sonic section. Following a preliminary check on the values just introduced,
which allows to prevent the code from converging to unfeasible solutions, either one, two, or three
geometries may be contemporary evaluated, two of which characterized by a subsonic converging
19
(a) (b)
Figure 2.2: Synoptic view of the mean-line model developed: (𝑎) allowed computational paths
and optional optimization algorithm, (𝑏) modular calculation path of main module.
nozzle and by a supersonic converging-diverging nozzle respectively. In addition, for values of

∗ 𝑠𝑢 𝑝
𝑀3,𝑖𝑠 lower than 1.3, the routine may evaluate a third geometry, a choked converging nozzle
with post-expansion. The selected Mach number threshold is consistent with standard practice
(Persico & Pini, 2017). In this case, the nozzle throat section 𝑜∗3 is reduced up to 𝑜∗3 = 𝑜 𝑐ℎ and
∗ 𝑠𝑢 𝑝
the post-expansion ratio 𝛽𝑒, 𝑃𝐸 is computed to ensure 𝑀3 = 𝑀3,𝑖𝑠 at the stator outlet 3. More
technically, to decrease the nozzle throat section 𝑜∗3 , which theoretically results from user-defined
inputs (nozzle inlet/outlet radii, blade heights, and blade angles), the routine iteratively overwrites
the stator outlet angle 𝛼3∗ to minimize the residual function 𝑟 (𝛼3∗ ) = 𝑜 ∗3 − 𝑜 𝑐ℎ .
As sketched in Figure 2.2b and detailed in Sections 2.2.1-2.2.4, an independent routine for
each component is sequentially solved leading to a flexible and modular design procedure. Notice
that the Post-Expansion routine is not run in case of subsonic nozzles, for which stations 3 and 3∗
coincide. Regardless the Analysis or Design Mode operation, RITML can be optionally coupled
with an external optimization algorithm to find the combination of the predefined design variables -
to be chosen among the code inputs - that maximizes the turbine performance (e.g. the total-to-static
efficiency 𝜂𝑇𝑆 ). A typical RIT optimization strategy is described in detail in Section 3.2.2.
2.2.1 Vaned nozzle

To solve the flow field within the vaned nozzle the control volume extending from station 1 to
the nozzle exit section 3∗ is considered, and the following system of continuity, energy and loss
equations is numerically solved
𝑚¤ = 𝜌3∗𝑉3∗ 𝑜∗3 𝑏 3 𝑍 𝑠 , (2.1a)

𝑉3∗ 2
ℎ∗𝑡3 = ℎ∗3 + , (2.1b)
2
𝑛
∑︁
ℎ∗3 − ℎ∗3,𝑖𝑠 = 𝛥ℎ𝑖𝑣,𝑙𝑜𝑠𝑠 , (2.1c)
𝑖=1
20
2.2. Mean-line method structure
where ℎ∗𝑡3 = ℎ𝑡1 , 𝑏 3 = 𝑏 ∗3 , and 𝑜∗3 = 𝑠3 cos 𝛼𝑏3 , being 𝑏 the blade height and 𝑍 𝑠 the stator
blade counts respectively. The subscript 𝑖𝑠 indicates an isentropic process starting from the
thermodynamic point 1 (stator inlet section). The term 𝛥ℎ𝑖𝑣,𝑙𝑜𝑠𝑠 represents the generic 𝑖-th
enthalpy loss contribution occurring through the vaned nozzle (the other symbol are defined in the
nomenclature included at the end of the thesis). A comprehensive description of loss mechanisms
associated with each component is provided in Section 2.3 together with the respective correlations,
all of them rearranged to express the loss contributions in terms of static enthalpy drop. Once the
geometrical parameters are given as inputs, Equations 2.1 represent a system of three non-linear
equations in the unknowns ℎ∗3 , 𝜌3∗ , and 𝑉3∗ , since all the loss contributions 𝑖=1
Í𝑛
𝛥ℎ𝑖𝑣,𝑙𝑜𝑠𝑠 can be
computed through the respective loss correlations.
2.2.2 Semi-bladed Nozzle
The Post-Expansion/Post-Compression routine through the semi-bladed nozzle is not run in case of
subsonic nozzles, for which stations 3 and 3∗ coincide. This routine activates only for converging
vanes in choked conditions, in which post-expansion may occur depending on the value of 𝑝 3 , and
for de Laval nozzles either over- or under-expanded. To solve the flow field in the semi-bladed
region, the trilateral control volume, delimited by points A, B, and C in Figure 2.1, is considered.
The approach suggested by Vavra (1969) is employed, which makes use of continuity equation,
energy balance, momentum balance along the direction perpendicular to 𝑜∗3 , and a thermodynamic
model of the fluid. The system writes as
𝑚¤ = 𝜌3𝑉3 cos 𝛼3 2𝜋𝑟 3 𝑏 3 , (2.2a)

𝑉32
ℎ𝑡3 = ℎ3 + , (2.2b)
2
¤ 3 cos 𝛼3∗ − 𝛼3 + 𝑝 3 cos 𝛼3∗ 2𝜋𝑟 3 𝑏 3 ,
¤ 3∗ + 𝑝 ∗3 𝑜∗3 𝑏 3 𝑍 𝑠 = 𝑚𝑉

𝑚𝑉 (2.2c)
ℎ3 = ℎ ( 𝑝 3 , 𝜌3 ) . (2.2d)
The flow deviation at the nozzle exit section 3∗ is neglected, such that flow and blade angle
may be considered coincident, hence 𝛼3∗ = 𝛼3,𝑏𝑙 . Conversely, the post-expansion process (or
post-compression in case of over-expanded converging-diverging nozzle) produces a rotation of the
streamlines, leading to a flow angle 𝛼3 different from 𝛼3∗ . To solve Equations 2.2, a supplemental
parameter defining the post-expansion extent is required, which allows to fill the lack of physical
constraints arising when the flow becomes choked. In the present work, the post-expansion pressure
ratio 𝛽𝑒, 𝑃𝐸 = 𝑝 ∗3 /𝑝 3 is provided as input to quantify the pressure variation in the semi-bladed
region. Therefore, Equations 2.2 represent a system of four non-linear equations in the four
unknowns ℎ3 , 𝜌3 , 𝑉3 , and 𝛼3 and can be solved numerically.
2.2.3 Vaneless space
The control volume between radius 𝑟 3 and 𝑟 4 is considered to solve the flow field in the vaneless
gap by applying the conservation of mass, angular momentum, and energy, and a thermodynamic
model of the fluid. As thoroughly described in Section 2.3, a friction term is added in the angular
momentum equation to account for the viscous dissipation at the two end-walls, in accordance with
the approach suggested by Whitfield & Baines (1990a). Hence, the set of equations considered is
21
the following:
𝑚¤ = 𝜌4𝑉4 cos 𝛼4 2𝜋𝑟 4 𝑏 4 , (2.3a)

𝑉42
ℎ𝑡4 = ℎ4 + , (2.3b)
2
𝑉3 sin 𝛼3 𝑟 4 2𝜋 𝑟 32 − 𝑟 3 𝑟 4 𝐶 𝑓 𝜌4𝑉4 sin 𝛼4
= + , (2.3c)
𝑉4 sin 𝛼4 𝑟 3 𝑚¤
ℎ4 − ℎ4,𝑖𝑠𝑠 = 𝛥ℎ 𝑓 ,𝑙𝑜𝑠𝑠 , (2.3d)
ℎ4 = ℎ (𝜌4 , 𝑝 4 ) , (2.3e)
where 𝛥ℎ 𝑓 ,𝑙𝑜𝑠𝑠 is the friction loss contribution and 𝐶 𝑓 is the friction coefficient defined in Whitfield
& Baines (1990a), as better explained in Section 2.3, and the subscript 𝑖𝑠𝑠 an isentropic expansion
from the previous thermodynamic point. Equations 2.3 represent a system of five non-linear
equations in the five unknowns ℎ4 , 𝜌4 , 𝑉4 , 𝛼4 , and 𝑝 4 and is numerically solvable.
2.2.4 Rotor
As anticipated, the code structure allows to deal with either radial or backswept blades at rotor
inlet. To solve the flow field within the rotor, continuity equation, rothalpy conservation, and loss
equation are considered
𝑊42 𝑈42 𝑊 2 𝑈2
ℎ4 + − = ℎ6 + 6 − 6 , (2.4a)
2 2 2 2
2 2
𝑚¤ = 𝜌6𝑊6 cos 𝛽6 𝜋 𝑟 6𝑡𝑖 𝑝 − 𝑟 6ℎ𝑢𝑏 , (2.4b)
𝑛
∑︁
ℎ6 − ℎ6,𝑖𝑠𝑠 = 𝛥ℎ𝑟𝑖 ,𝑙𝑜𝑠𝑠 , (2.4c)
𝑖=𝑖
where 𝛥ℎ𝑟𝑖 ,𝑙𝑜𝑠𝑠 indicates the generic 𝑖-th loss contribution within the rotor, also including the
incidence losses as a function of 𝛽4 and 𝛽4,𝑏𝑙 in Analysis mode.
In the current version, supersonic relative flow at the rotor inlet are not accepted by RITML and
hence the relative mach number 𝑀𝑤4 is limited to 1. The reason under this limitation is twofold.
From one side, it was supposed that the relevant entropy rise and flow field complexity associated
with the interaction between supersonic relative flows and rotor blade complicates supersonic
rotor designs. From the other side, it was judged that the scarcity of validated reduced-order
models dealing with RIT supersonic rotors and corresponding loss production would have raised
strong reliability issues on the prediction of supersonic rotor performance. Moreover, to ease the
optimization process, the present meanline method limits to 1 the maximum relative Mach number
within the rotor,avoiding the need to account for possible post-expansion downstream of the rotor
outlet station. Therefore, throat section (that would be labeled as 5) and rotor outlet station (labeled
as 6) are considered identical in both velocity and thermodynamic state.
Equations 2.4 represent a system of three non-linear equations in the three unknowns ℎ6 , 𝜌6 ,
and 𝛽6 and can be solved numerically. Notice that 𝛽6 represents an unknown parameter only in
design mode, in which 𝑝 6 is a known variable since 𝛽𝑒,𝑇𝑆 is provided as input, while in analysis
mode 𝛽6 can be evaluated from the rotor outlet blade angle 𝛽𝑏,6 , which has to be provided as
input. To link the flow and blade angle, the user can optionally employ the well-established method
developed for axial turbines by Ainley & Mathieson (1951) to account for deviation effects at
rotor outlet, which produce a flow that is more meridional than the blade direction. This empirical
correlation links the metal angle 𝛽𝑏,6 to the flow angle 𝛽6 as a function of the relative Mach number
22
2.3. Loss models
𝑀𝑤6 and of the rotor outlet pitch distance 𝑠6 = 2𝜋𝑟 6 /𝑍𝑟 , where 𝑍𝑟 represents the number of rotor
blades. If the option of computing the deviation through Ainley and Mathieson method is not
activated the code simply sets 𝛽6 = 𝛽𝑏,6 .
2.2.5 Diffuser
The diffuser is a stationary component which may be optionally included downstream of the rotor
to reduce the discharged kinetic energy, hence reducing the rotor outlet pressure (for the same
pressure imposed downstream of the diffuser) and increasing the total-to-static efficiency 𝜂𝑇𝑆 .
Although different geometries can be selected for the diffuser (such as flat-squared or annular type),
a conical diffuser shape is adopted in RITML since typically employed in radial-inflow-turbines
(Nithesh & Chatterjee, 2016; Zou et al., 2018). If the calculation for this component is activated,
two additional inputs have to be provided by the user, namely the semi-opening angle 𝜀 and the
area ratio 𝐴𝑅 = 𝐴7 /𝐴6 . Exploiting these two inputs, the outlet radius 𝑟 7 and the axial length 𝐿 𝑑 of
the diffuser can be computed as ‘
√︂
𝑟7 = 2
𝐴𝑅 𝑟 6,𝑡𝑖 𝑝
− 𝑟 6,ℎ𝑢𝑏
2 , (2.5a)
𝑟 6 − 𝑟 6,𝑡𝑖 𝑝
𝐿𝑑 = , (2.5b)
tan 𝜀
To solve the flow field within the diffuser a control volume extending from rotor outlet (station 6)
to the diffuser exit section 7 is considered, and the following system of momentum and energy is
numerically solved
𝑝 7 = 𝑝 6 + 𝐶 𝑝 ( 𝑝 𝑡6 − 𝑝 6 ) , (2.6a)
ℎ7,𝑖𝑠𝑠 − ℎ6
ℎ7 = ℎ6 + , (2.6b)
𝜂𝑑
ℎ7,𝑖𝑠𝑠 = ℎ7,𝑖𝑠𝑠 ( 𝑝 7 , 𝑠6 ) , (2.6c)
where 𝐶 𝑝 and 𝜂 𝑑 are respectively the diffuser pressure recovery coefficient and isentropic efficiency,
the latter being included to account for the non-negligible entropy production related to the diffusion
process, which is characterized by strong adverse pressure gradients. 𝐶 𝑝 is estimated as a function
of the turbine outlet properties (such as 𝛼6 , 𝑉6 and 𝜌6 ) and diffuser geometry, by coupling the
models proposed by Mimic et al. (2018) and Thompson (1979). The former model allows to
compute the ideal pressure recovery coefficient 𝐶 𝑝,𝑖𝑑 for non-axial inlet flows, while the latter
takes into account viscous dissipation and blockage associated with the diffusion process to derive
the relation between the actual and ideal pressure recovery coefficients. Finally, the enthalpy loss
associated with the diffusion process and the corresponding isentropic efficiency 𝜂 𝑑 is estimated
resorting to the method proposed by Osnaghi (2020), which computes 𝜂 𝑑 as a function of the
diffuser architecture, semi-opening angle 𝜀, and area ratio 𝐴𝑅.
2.3 Loss models

The objective of this section is twofold: firstly, to provide an exhaustive view of the main loss
models available in literature; secondly, to describe the loss correlations employed in the present
analysis. While the chosen loss models are thoroughly described, the detailed description of all the
other loss correlations mentioned throughout the manuscript is not included in the present work.
Only the different loss models associated with stator and rotor passage losses are detailed, due
to the large impact these loss contributions has on the overall efficiency of RITs. Loss models
23
available in literature for RITs were developed for and validated over conventional gas turbines,
hence their validity for machines operating with organic fluids needs to be verified. This makes
the validation process a crucial step towards the development of an accurate and robust mean-line
model.
2.3.1 Nozzle losses

Entropy generation mechanisms through subsonic vaned nozzles can be grouped into two main
contributions: trailing edge losses and passage losses. The former contribution takes into account
the entropy production due to an abrupt opening of the cross section at the blade outlet, and a
correlation to model it was recommended by Glassman (1976b). According to the author, the
enthalpy drop corresponding to trailing edge losses may be computed as
𝛾∗

3
𝛾3∗ − 1
2 1−𝛾 ∗
𝑍 𝑠 𝑡3 1 ∗2
𝑀3∗ 2
3
𝛥ℎ𝑡𝑒 = 𝑉 1+ , (2.7)
2𝜋𝑟 3 cos 𝛼3∗ 2 3 2
where 𝑡 3 is the nozzle trailing edge thickness. Concerning passage losses, which account for
friction and blade loading effects, existing correlations can be traced back to two main approaches
proposed respectively by Rodgers (1987) and Glassman (1976a). Both the authors evidenced that
the blade loading contribution is negligible with respect to friction effects, since the camber line
deflection in radial nozzles is usually small. The method developed by Rodgers empirically relates
friction loss to blade geometrical parameters and to the average outlet Reynolds number 𝑅𝑒 3 by
means of a best-fit of experimental data, such that
∗ ∗
0.05 3 tan 𝛼3 𝑠3 cos 𝛼3 1 ∗2
𝛥ℎ 𝑝𝑎𝑠𝑠 = + 𝑉 , (2.8)
𝑅𝑒 0.2
3
𝑠3 /𝑐 𝑠 𝑏3 2 3
where 𝑠3 = 2𝜋𝑟 3 /𝑍 𝑠 and 𝑐 𝑠 are the geometrical pitch at the nozzle outlet and the nozzle chord
respectively. According to Glassman (1976a), the nozzle chord may be defined as
√︂
𝑐 𝑠 = −𝑟 3 cos 𝛼𝑏3 + 𝑟 32 (cos 𝛼𝑏3 ) 2 − 1 + 𝑟 12 , (2.9)
for uncambered blades, and as

v
u
u v
u
t t !2
2 𝑟 12 − 𝑟 32
𝑐𝑠 = 𝑟 12 + 𝑟 32 − 𝑟 12 + 𝑟 32 − , (2.10)
cos 𝜙 𝑐𝑚
for cambered ones, where 𝜙 𝑐𝑚 is the average blade angle and is defined as
𝛼𝑏1 + 𝛼𝑏3
𝜙 𝑐𝑚 = . (2.11)
2
This approach was adopted by several authors, such as Meroni et al. (2018b), Fiaschi et al. (2015),
and Ventura et al. (2012). Differently, the method developed by Glassman exploits boundary layer
theory and relates the kinetic energy loss coefficient 𝜀, defined as a fraction of the ideal kinetic
energy of the blade row in the clean region, to the main boundary layer integral parameters. In
particular, the two-dimensional kinetic energy loss coefficient 𝜀 2𝐷 is defined as
!
𝑉3∗
𝜀 2𝐷 = 1 − ∗ . (2.12)
𝑉3,𝑖𝑠
24
2.3. Loss models
More specifically, the author considered the three-dimensional kinetic energy loss coefficient 𝜀 3𝐷 ,
which accounts for end-walls contribution and can be related to its 2D counterpart through
!
𝐴2𝐷 𝑟 13 − 𝑟 32
𝜀 3𝐷 = 𝜀 2𝐷 = 𝜀 2𝐷 1 + . (2.13)
𝐴3𝐷 2𝑟 3 𝑏 3 𝑐𝑙𝑠 𝑐𝑠3𝑠
where 𝐴2𝐷 , 𝐴3𝐷 are the two- and three-dimensional nozzle wetted surface, while 𝑙 is the geometric
length of the blade, which is equal to the chord 𝑐 𝑠 for uncambered blade geometries, while can be
expressed in function of the camber angle 𝛩 𝑐𝑎𝑚 for cambered ones, such that
𝛩 𝛩 𝑐𝑎𝑚
𝑙= 𝑐𝑠 = 𝑐𝑠 , (2.14)
𝛩𝑐𝑎𝑚
2 sin 2 2 sin 𝛩𝑐𝑎𝑚
2
where !
𝑟 12 + 𝑟 32 − 𝑐2𝑠
𝛩 𝑐𝑎𝑚 = 𝛼3∗ − 𝛼1 − arccos . (2.15)
2𝑟 1 𝑟 3
Accordingly to Glassman, 𝜀 3𝐷 can be approximated as
−0.2
𝐸 𝜗𝑇𝑂𝑇 𝑙
𝑅𝑒 𝑙 𝐴3𝐷
𝑟 𝑒 𝑓 𝑅𝑒𝑟𝑒 𝑓 𝑠 𝐴2𝐷
𝜀 3𝐷 = −0.2 , (2.16)
cos 𝛼3∗ − 𝑠𝑡33 − 𝐻 𝜗𝑇𝑂𝑇 𝑙
𝑅𝑒
𝑅𝑒𝑟𝑒 𝑓
𝑙
𝑠
𝑟𝑒 𝑓
where the boundary layer shape factor 𝐻 and energy factor 𝐸 for turbulent flows can be expressed
as a power series expansion of the Mach number 𝑀3∗ squared, while 𝑅𝑒 and 𝜗𝑇𝑂𝑇 are the
Reynolds number and the boundary layer total momentum thickness respectively. For further
details concerning the parameters in Equation 2.16 the reader is referred to Glassman (1976a).
𝜀3𝐷 is therefore a function of the nozzle geometry (G), exit section Mach number 𝑀3∗ , velocity 𝑉3∗ ,
and density 𝜌3∗ , the Reynolds number depending on the last two parameters. Therefore, the static
enthalpy drop due to passage losses 𝛥ℎ 𝑝𝑎𝑠𝑠 according to Glassman may be written as

𝛥ℎ 𝑝𝑎𝑠𝑠 = ℎ𝑡1 − ℎ∗3,𝑖𝑠 𝜀 3𝐷 = ℎ𝑡1 − ℎ∗3,𝑖𝑠 𝑓 𝑀3∗ , 𝑉3∗ , 𝜌3∗ , G .

(2.17)
For nozzle flows featuring post-expansion phenomena (either choked converging or under-expanded
de Laval nozzles), the correlation developed by Glassman provides better agreement with reference
data if the flow properties in Equation 2.17 are evaluated at the outlet station 3 instead of at the
exit section 3∗ , as noticed during the validation strategies described in Section 2.6. Although
post-expansion/post-compression losses need to be included in the analysis for supersonic nozzles,
in the present reduced-order model this loss contribution directly results from the solution of
conservation equations, as discussed in Section 2.2.2. The proposed approach avoids therefore the
implementation of any external empirical model, allowing for a physical description of the actual
flow field. The classification of stator loss contributions reported at the beginning of this section is
not unique and other subdivisions of the stator losses have been employed in the open literature.
Stewart (1955) considered the losses occurring through a stator blade row as the sum of boundary
layer and mixing losses, while Aungier (2006a) encompassed profile and incidence losses. More
specifically, the approach developed by Aungier (2006a) includes a comprehensive set of loss
models for each component of RITs and is often used by other meanline models, such as the one
developed by Da Lio et al. (2017b). Incidence losses for the stator have not been considered in the
present work since single-stage RITs usually features very low Mach numbers and velocities at the
stator inlet, whose leading edge is commonly designed to be highly tolerant to the severe change of
incidence angle (as also proved in Section 5.2.1).
25
2.3.2 Vaneless Space losses

Concerning the vaneless space between the stator and the rotor, a friction term is added in the
angular momentum equation to account for viscous effects, as described in Section 2.2.3. The
corresponding static enthalpy drop 𝛥ℎ 𝑓 ,𝑙𝑜𝑠𝑠 in Equation 2.3d can be computed, using the approach
suggested by Kastner & Bhinder (1975), as function of the velocity squared , such that
2
𝐿 ℎ 1 𝑉3 + 𝑉4
𝛥ℎ 𝑓 ,𝑙𝑜𝑠𝑠 = 𝐶 𝑓 , (2.18)
𝐷ℎ 2 2
where 𝐶 𝑓 is the same friction coefficient exploited in Equation 2.3c, while 𝐿 ℎ and 𝐷 ℎ are the
hydraulic length and diameter of the stator-rotor interspace gap respectively, defined as
𝐿 ℎ = 𝑟3 − 𝑟4 , (2.19a)
𝑏3 + 𝑏4
𝐷ℎ = . (2.19b)
2
(2.19c)
In particular, the Fanning formulation of 𝐶 𝑓 reported by Whitfield & Baines (1990a) is used in the
present analysis, which models the friction coefficient as
0.02
1.8 ∗ 105
𝛥𝐶 𝑓 = 𝐾 , (2.20)
𝑅𝑒
where the multiplicative constant 𝐾 is set equal to 0.01 in RITML, as suggested by Japikse (1982).
2.3.3 Rotor losses

Regarding the rotor, the loss models employed in literature can be traced back to three main
comprehensive approaches, which have been proposed between 1976 and 1998 respectively by
Rodgers (1987), Glassman (1976a) and Wasserbauer & Glassman (1975), afterwards improved
by Baines et al. (1998). Each of these loss models accounts for the different entropy production
mechanisms occurring in the rotor, namely incidence, passage, trailing edge, clearance and windage
losses. In the present work, the correlation developed by Wasserbauer & Glassman (1975) is
adopted to account for possible incidence losses, which model the corresponding enthalpy drop as
1
𝛥ℎ𝑖,𝑙𝑜𝑠𝑠 = 𝑊42 sin2 𝛽4 − 𝛽4,𝑜 𝑝𝑡 , (2.21)
2
where 𝛽4,𝑜 𝑝𝑡 is the optimal relative flow angle at rotor inlet, which is computed depending on
the rotor blades configuration. If radial, the method proposed by Baines (1996), who extended to
radial turbines the concept of slip factor developed by Stanitz (1952) for centrifugal compressors, is
adopted. Otherwise, the optimum inlet angle may be estimated according to Meitner & Glassman
(1983), who reported a previous investigation performed by Wiesner (1967). In particular, the two
approaches suggest different correlations to compute the slip factor, which can be estimated as
1.98
𝜇 =1− , (2.22)
𝑍𝑟
for radial blades according to Baines, and through
( 𝑟
𝐴 𝐶𝐵 if 6,𝑟𝑟4𝑚𝑠 > 𝜀𝑙𝑖𝑚
𝜇= 𝐴 𝑟 , (2.23)
𝐶 if 6,𝑟𝑟4𝑚𝑠 < 𝜀𝑙𝑖𝑚
26
2.3. Loss models
for backswept blades according to Meitner & Glassman (1983), where the patameters 𝐴,𝐵,𝐶, and
𝜀𝑙𝑖𝑚 are detailed in Equations 2.24:
√︁
cos 𝛽𝑏,4
𝐴 =1− , (2.24a)
𝑍𝑟0.7
𝑟6,𝑟 𝑚𝑠
!3
𝑟4 − 𝜀 𝑙𝑖𝑚
𝐵 =1− , (2.24b)
1 − 𝜀𝑙𝑖𝑚
tan 𝛽𝑏,4
𝐶 =1− , (2.24c)
tan 𝛼4
1
𝜀𝑙𝑖𝑚 = . (2.24d)
8.16 cos 𝛽𝑏,4
exp 𝑍𝑟
Once the slip factor is computed, the optimal relative flow angle at rotor inlet 𝛽4,𝑜 𝑝𝑡 can be
computed as
𝑉𝑡4,𝑜 𝑝𝑡 = 𝜇 𝑈4 , (2.25a)
𝑊𝑡4,𝑜 𝑝𝑡 = 𝑉𝑡4,𝑜 𝑝𝑡 − 𝑈4 , (2.25b)

𝑊𝑡4,𝑜 𝑝𝑡
𝛽4,𝑜 𝑝𝑡 = arctan . (2.25c)
𝑊𝑚4
Although the models described above are the most used in the open literature, alternative approaches
can be exploited to estimate either the slip factor 𝜇 or the optimum angle 𝛽4,𝑜 𝑝𝑡 (Bridle & Boulter,
1967; Dixon & Hall, 2013).
Each of the three approaches introduced at the beginning of the section provides a correlation
to estimate rotor passage losses. This entropy generation mechanism is produced by viscous
forces, which both decelerate the flow within blade boundary layers, and generate an unbalance
between pressure gradients and centripetal forces within end-wall boundary layers, resulting in the
so-called secondary or passage vortices. Glassman’s procedure (Glassman, 1976a) is similar to the
one developed for the nozzle and previously described, still resorting to boundary layer theory.
However, this method does not take into account the blade loading contribution. Accordingly to
the approach proposed by Baines et al. (1998), the enthalpy drop associated to rotor passage losses
con be computed as
( " # )
𝑟 6,𝑟 𝑚𝑠 2 cos 𝛽6,𝑟 𝑚𝑠 1 2

𝐿ℎ 2
𝛥ℎ 𝑝𝑎𝑠𝑠 = 𝐾 𝑝 + 0.68 1 − 𝑊4 + 𝑊6,𝑟 𝑚𝑠 , (2.26)
𝐷ℎ 𝑟4 𝑏 6 /𝑐𝑟 2
where the
√︃ subscript 𝑟𝑚𝑠 indicates properties evaluated at the root mean square radius, defined as
𝑟 𝑟 𝑚𝑠 = 𝑟 ℎ𝑢𝑏
2 + 𝑟 𝑡𝑖2 𝑝 , while 𝑐𝑟 , 𝐿 ℎ , and 𝐷 ℎ are the rotor equivalent chord, hydraulic lenght, and
hydraulic diameter respectively, which are defined as
√︄
2
𝑏4 2
𝑐𝑟 = 𝑙𝑧 − + 𝑟 4 − 𝑟 6,𝑟 𝑚𝑠 , (2.27a)
2

𝜋 𝑏4 𝑏6
𝐿ℎ = 𝑙𝑧 − + 𝑟 4 − 𝑟 6,𝑡𝑖 𝑝 − , (2.27b)
4 2 2
 
© 2𝜋 𝑟 6,𝑡𝑖 𝑝 − 𝑟 6,ℎ𝑢𝑏
2 2
1  4𝜋𝑟 4 𝑏 4 ª
𝐷ℎ =  + ®  , (2.27c)
2  2𝜋𝑟 4 + 𝑍𝑟 𝑏 4 𝜋 𝑟 6,𝑡𝑖 𝑝 − 𝑟 6,ℎ𝑢𝑏 + 𝑍𝑟 𝑏 6 ®

 « ¬
27
where the geometrical parameters 𝑙 𝑧 and 𝑏 are the rotor axial length and bade height respectively.
Validation against experimental data suggested a value of 𝐾 𝑝 in Equation 2.26 equal to 0.11 if the
rotor channel curvature is limited, that means
𝑟 4 − 𝑟 6,𝑡𝑖 𝑝
< 0.2 . (2.28)
𝑏6
Conversely, the value of 𝐾 𝑝 must be doubled to account for the large increase in secondary loss
induced by the very small curvature radius at the tip shroud. As can be recognized in Equation
2.26, the method proposed by Baines comprises both a term to account for frictions (first one in
brackets) and one to model blade loading losses (the second one in brackets). This is the reason
why this last method is extensively used in the literature by many authors (Ventura et al., 2012;
Meroni et al., 2018b; De Servi et al., 2019). Also the approach developed by Rodgers (1987) takes
into account both the contributions. Strong uncertainties are however present in literature regarding
this correlation since it underwent many modifications over the years. A first modification was
reported by Whitfield (1990), who introduced a correction to the original curvature friction term
proposed by Rodgers to have a better fit with experimental results. An additional adjustment of the
blade loading term was also carried out by Whitfield and later adopted by Fiaschi et al. (2015).
Ventura et al. (2012), followed by Alshammari et al. (2017), partially exploited Rodgers correlation
with some adjustments; the friction term was computed according to Musgrave (1979), while the
original blade loading term by Rodgers was revised, apparently without providing any explanation.
The approach proposed by Ventura expresses the enthalpy drop due to rotor passage losses as
2
𝐿 ℎ 𝑊4 + 𝑊6,𝑟 𝑚𝑠 𝐷4 2
𝛥ℎ 𝑝𝑎𝑠𝑠 = 𝑓𝑡 + 𝑉 , (2.29)
𝐷ℎ 2 𝑍𝑟 𝑟 𝑐 4
where 𝐷 4 is the rotor inlet diameter while 𝑓𝑡 is a modified friction factor to account for rotor
inherent curvature effects (Suhrmann et al., 2010), defined as
" # 0.05
2
𝑟4
𝑓𝑡 = 𝑓𝑐 𝑅𝑒 , (2.30)
𝑟𝑐
and 𝑓𝑐 is the skin friction factor for curved pipes (Suhrmann et al., 2010):
√︂ !
0.25 𝐷ℎ
𝑓𝑐 = 𝑓 1 + 0.075𝑅𝑒 . (2.31)
2𝑟 𝑐
In the last equation, 𝑓 is the Fanning friction factor, calculated depending on the flow regime and
assuming the upper limit for turbomachinery wall roughness, i.e. 0.2 mm (Aungier, 2006b). The
parameter 𝑟 𝑐 in Equation 2.29 is the curvature radius and can be computed as

1 𝑏4 𝑏6
𝑟𝑐 = 𝑙𝑧 − + 𝑟 4 − 𝑟 6,𝑡𝑖 𝑝 − . (2.32)
2 2 2
The uncertainties related to the passage loss model originally proposed by Rodgers and the several
modifications it underwent during the years are the reasons why the employment of this correlation
is not recommended in the present work, as further confirmed by results reported in Section
2.5, in which the three passage loss correlations have been compared and tested over both air
operating and ORC turbines. Similarly to the vaned nozzle, trailing edge losses through the rotor
are evaluated exploiting the correlation proposed by Glassman (1976b). According to Ventura, this
loss contribution should not be included if the approach suggested by Rodgers is adopted to model
passage losses (Ventura et al., 2012).
28
2.4. Test case radial inflow turbines
Table 2.2: Test cases for radial inflow turbine geometries. C and C-D stand for converging and
converging-diverging nozzles respectively. The power output refers to the turbine nominal design
point. For turbine #6, * N649 stands for NOVEC649. ** Turbine #7 was investigated with 3
different fluids: R134a, Heptane, and Propane.
Author Label Ref. Data fluid 𝛽𝑒,𝑇𝑆 𝑊¤ 𝑡 [kW] Nozzle

Geom.
Jones (1996) #1 Exp. air 6.1 82 C
Spence & Artt (1997) #2 Exp. air 1.7 - 2.7 35 C
Mclallin & Haas (1980) #3 Exp. air 1.6 - 3.3 23 C
De Servi et al. (2019) #4 CFD MM 23.9-46.1 10.6 C-D
Alshammari et al. (2017) #5 CFD N649* 13 20.5 C-D
Sauret & Gu (2014) #6 CFD R143a 2.68 422 C
Schuster et al. (2020) #7 CFD 3 fluids** 1.4-2.6 10.8-573 C
The presence of a tip gap together with pressure gradients between the blade pressure and
suction sides and the blade interaction with the casing boundary layer are responsible for the
generation of the so-called tip-clearance flows and the development of a tip-vortex leading to
tip-clearance losses. In the present work, this loss contribution is accounted for by means of the
empirical model proposed by Spraker (1987) and modified by Baines to achieve a better fit with
experimental data (Baines et al., 1998), accordingly to which the corresponding enthalpy drop may
be computed as
𝑈43 𝑍𝑟 √︁
𝛥ℎ 𝑐𝑙 = 𝐾 𝑎 𝜀 𝑎 𝐶𝑎 + 𝐾𝑟 𝜀𝑟 𝐶𝑟 + 𝐾 𝑎𝑟 𝜀 𝑎 𝜀𝑟 𝐶𝑎 𝐶𝑟 , (2.33)
8𝜋
where 𝜀 𝑎 , and 𝜀𝑟 are the axial and radial tip gaps respectively, 𝐶𝑎 , and 𝐶𝑟 are the axial and radial
rotor chords respectively, and 𝑍𝑟 is the rotor number of blades. As reported in Equation 2.33, the
correlation proposed by Spraker includes both axial and radial clearances, together with a mutual
interaction term, required to model the complex flow pattern in the rotor gaps. The flow is indeed
firstly pushed from the suction side to the pressure side by the Coriolis force acting at the rotor
inlet, while the opposite occurs in the mixed and axial flow region, where blade-to-blade pressure
gradients are dominant. Despite the completeness of the model proposed by Spraker, there are
several approaches in the open literature to estimate tip-clearance losses, most of them accounts
only for the radial clearance 𝜀𝑟 (Glassman, 1976b; Whitfield & Baines, 1990b).
Windage losses, due to the flow shearing between the turbine casing and the rotor backplate,
were estimated according to Daily & Nece (1960) but other approaches are available in the open
literature (Glassman, 1976b; Dixon & Hall, 2013). Since the reduced order-model here described
is structured to design single stage machines, the kinetic energy discharged at rotor or diffuser
outlet is considered lost while computing the total-to-static efficiency 𝜂𝑇𝑆 .
2.4 Test case radial inflow turbines

Altogether, seven test cases have been selected in the open literature to carry out both the loss
sensitivity analyses described in Section 2.5) and the validation strategy detailed Section 2.6),
whose main characteristics are listed in Table 2.2. The first three turbines investigated (test case
geometries #1 to #3) feature air as working fluid and they are characterized by a converging nozzle
29
(Jones, 1996; Spence & Artt, 1997; Mclallin & Haas, 1980). Their selection is motivated by the
wide availability of experimental data in terms of pressure ratio (1.6 < 𝛽𝑒,𝑇𝑆 < 6.1), rotational
speed (30000-70000 rpm) and power output (20-80 kW). Due to the lack of well-established ORC
experimental data, the last four turbines chosen for validation (test case geometries #4 to #7) have
been deeply investigated by means of high-fidelity 3D-steady CFD simulations (De Servi et al.,
2019; Alshammari et al., 2017; Sauret & Gu, 2014). Overall, the set of machines here described
is characterized by pressure ratio and power output in the range of 1.4 to 46.1 and 10 to 570 kW
respectively. Test case geometries #6 and #7 implement a converging nozzle, while the other two
turbines (#4 and #5) feature a de Laval nozzle architecture. Besides test cases #2 and #3, off-design
operating maps are also available for turbines #4 and #7.
The choice of different fluids, sizes and pressure ratio was intentionally made in order to inspect
the widest design space available, pursuing a validation procedure as general and unbiased as
possible. In order to perform the validation, the reduced-order model is used in analysis mode,
requiring as inputs for subsonic turbines the entire geometry, the inlet total conditions (𝑝 𝑡1 , 𝑇𝑡1 ),
the rotational speed, and the mass flow rate. In case of supersonic flows within the nozzle, the
post-expansion ratio 𝛽𝑒, 𝑃𝐸 must be also provided as input. Its value was chosen to unambiguously
meet the same operating point (and consequently the same 𝛽𝑒,𝑇𝑆 ) of the reference data. Turbines
#1 and #6 consider a diffuser in the experimental and numerical set-up respectively. Therefore, a
pre-processing of the reference data was performed to derive the static thermodynamic properties
at the rotor outlet, which are required to compute the values of 𝛽𝑒,𝑇𝑆 and 𝜂𝑇𝑆 needed for the
validation. In case of reference geometry #1, this was accomplished exploiting the given diffuser
pressure recovery coefficient 𝐶 𝑝 , while for turbine #6 the values of 𝑀6 , 𝑝 𝑡6 , 𝑇𝑡6 provided at rotor
outlet were used.
2.5 Loss sensitivity analyses

This section describes the loss sensitivity analyses carried out to assess the validity, robustness,
and intrinsic limitations of the main stator and rotor passage loss models, commonly used by many
authors in the past years. In these procedures, the passage loss correlations described in Section
2.3 are tested over published data on both air-operating and ORC turbines, leading to a final choice
of physically-based loss models targeted to small-scale, high-expansion ratio ORC turbines and to
the complete definition of the meanline model RITML.
The sensitivity analyses described in the following focus only on passage loss since this
contribution, including both profile and blade loading effects, accounts for most of the losses
typically occurring through stator and rotor blade rows of RITs. Consequently, passage losses have
a strong impact on the overall efficiency of radial-inflow-turbines, as also proved by the outcomes
described in Section 3.5. The other loss contributions (e.g. vaneless space friction losses, rotor
tip-clearance losses, and windage losses) are less relevant, at least for the application and the turbine
geometries considered, and represent a smaller fraction of the overall RITs’ losses. Additionally,
different models for these loss items proved very similar loss prediction capability. When this
happened, the choice of the most suitable correlation for a specific loss item was done relying on
those models featuring a physically-based set of equations, as less depending on empirical relations
as possible.
The effectiveness of the blade passage loss models, described in Section 2.3.1 for the stator and
in Section 2.3.3 for the rotor, is assessed by testing them over two reference test case geometries,
namely Jones (1996) and De Servi et al. (2019) (see Table 2.2). Indeed, these turbines operate
with distinct fluids, air and MM (hexamethyldisiloxane), and they are characterized by reduced
dimensions and high expansion ratios, which are peculiar features of small-power high-temperature
applications.
30
2.5. Loss sensitivity analyses
Table 2.3: Rotor loss models effectiveness analysis. Comparison of three different approaches to
compute blade passage losses.
Benchmark RIT
Loss model #1 Jones (1996) #4 De Servi et al. (2019)
𝛽𝑒,𝑇𝑆 𝛥𝛽 𝜂𝑇𝑆 % 𝛥𝛽 𝜂𝑇𝑆 %
Ref. 6.103 26° 84.40 98.60° 84.1
Ventura et al. (2012) RITML 6.214 28.32° 83.68 - -
Dev 1.82% 2.32° -0.72 - -
Ref. 6.103 26° 84.40 98.60° 84.1
Glassman (1976a) RITML 5.480 17.22° 86.11 97.82° 85.23
Dev -10.22% -8.78° 1.71 -0.78° 1.13
Ref. 6.103 26° 84.40 98.60° 84.10
Baines et al. (1998) RITML 6.066 27.63° 84.10 98.15° 84.52
Dev -0.62% 1.63° -0.30 -0.45° 0.42
Firstly, the rotor passage loss contribution is investigated and the correlations proposed by
Ventura et al. (2012) (on the basis of Rodgers), Glassman (1976a), and Baines et al. (1998) are
compared. Throughout this analysis, the set of correlations employed for all the other loss items
are those described in Section 2.3, while the correlation proposed by Glassman (1976a) is used
to estimate stator passage losses. The outcomes of the analysis are shown in Table 2.3 for both
reference turbines, in terms of 𝛽𝑒,𝑇𝑆 , 𝛥𝛽, and 𝜂𝑇𝑆 for test case #1, and 𝛥𝛽 - the flow deviation angle
across the rotor - and 𝜂𝑇𝑆 for test case #4. The turbine nozzle designed by De Servi is choked at the
design point and the overall pressure ratio 𝛽𝑒,𝑇𝑆 is imposed to have a fair comparison with CFD
computations. For each loss model, the reference value (experimental for Jones’ turbine, numerical
for De Servi’s turbine), the value predicted by the mean-line model here described (RITML), and the
related deviation are reported in Table 2.3. The latter is defined as a relative percentage deviation
for 𝛽𝑒,𝑇𝑆 , as expressed in Equation 2.34, while the absolute deviation (difference between predicted
and reference data) is given for 𝜂𝑇𝑆 and 𝛥𝛽.
𝑋𝑅𝐼𝑇 𝑀 𝐿 − 𝑋𝑅𝑒 𝑓
Dev = · 100% . (2.34)
𝑋𝑅𝑒 𝑓
Focusing on turbine #1, the comparison highlights how the loss models proposed by Ventura and
Baines provide a satisfactory agreement with experimental results both in terms of 𝛽𝑒,𝑇𝑆 , 𝛥𝛽, and
𝜂𝑇𝑆 . On the contrary, Glassman correlation under-estimates the entropy generation through the
rotor due to the lack of the blade loading term, leading to lower pressure, much lower deflection
(𝛥𝛽) and larger efficiency. Concerning De Servi’s turbine, comparison against Ventura correlation
cannot be carried out because the passage loss term 𝛥ℎ 𝑝,𝑙𝑜𝑠𝑠 , which depends on rotor inlet absolute
velocity and average relative velocity through the rotor, becomes too large when high pressure
ratios are involved (as for turbine # 4). This prevents the system of Equations 2.4 from having a
mathematical solution. This limitation, together with the discrepancies found in literature and
described in Section 2.3, discourage the employment of the passage loss model developed by
Rodgers (and derived ones) when high-pressure ratio turbines operating with organic fluids are
considered. Moreover, the higher accuracy of Baines model compared to Glassman correlation
is evident also when test case turbine #4 is analyzed. The higher loss prediction capabilities
associated to Baines’ model are to be ascribed to the completeness of this correlation, which
encompasses three contributions (sse Equation 2.26). The first term account for friction losses,
simply considering a rotor-equivalent duct, while the second and third contributions model the
31
Table 2.4: Nozzle loss models effectiveness analysis. Comparison of two different approaches to
compute blade passage losses.
Benchmark RIT
Loss model #1 Jones (1996) #4 De Servi et al. (2019)
𝛽𝑒,𝑇𝑆 𝛥𝛽 𝜂𝑇𝑆 % 𝛥𝛽 𝜂𝑇𝑆 %
Ref. 6.103 26° 84.40 98.60° 84.10
Rodgers (Rodgers, 1987) RITML 3.886 15.9° 84.16 103.76° 89.58
Dev -36.3% -10.1° -0.24 5.16° 5.48
Ref. 6.103 26° 84.40 98.60° 84.10
Glassman (Glassman, 1976a) RITML 6.066 27.63° 84.10 98.15° 84.52
Dev -0.62% 1.63° -0.30 -0.45° 0.42
blade loading and the hub-to-shroud loading, respectively. The above considerations, along with
the objectives of this work, lead to the selection of Baines correlation as reference method to
estimate rotor passage losses.
This approach is hereinafter exploited to carry out both the stator loss models effectiveness
analysis and the validation procedure reported below. However, the author noticed that Baines
correlation leads to an impossible solution of Equations 2.4 when very large pressure ratios together
with sub-atmospheric rotor outlet pressures occur, introducing some limitations on the employment
of this model, or establishing some feasibility limits in terms of maximum pressure ratio for
the effective design of small-scale single-stage ORC RIT. The 3 rotor passage loss models just
investigated have been tested also employing Rodgers’ correlation to estimate stator passage losses.
As better detailed below, Rodgers’ model features scarce prediction capabilities, producing a flow
at the rotor inlet that deviated significantly from the actual reference data. This ineffectiveness
of Rodgers’ correlation prevents to conduct a fair comparison of the different rotor passage loss
models. Conversely, the adoption of Glassman’s correlation for stator passage losses predicts flow
properties at the rotor inlet that closely match the reference data. This allows to effectively isolate
and fairly compare the performance of the different rotor passage loss correlations , the rotor inlet
conditions predicted by RITML being very close to reference data. This is the reason why the rotor
passage loss sensitivity analysis has been presented relying on Glassman’s correlation to estimate
stator passage losses.
Concerning the nozzle passage loss contribution, two different methods were compared; the
approach proposed by Rodgers (1987) and the one developed by Glassman (1976a), which was
introduced in Section 2.3.1 (Equations 2.16 and 2.17). Throughout this analysis, the set of
correlations employed for all the other loss items are those described in Section 2.3, while the
correlation proposed by Baines et al. (1998) is used to estimate rotor passage losses. The results of
the comparison are shown in Table 2.4. Regarding turbine #1, although the deviation for 𝜂𝑇𝑆 is
quite small, the mean-line model exploiting Rodgers correlation largely under-predicts the overall
pressure ration and the flow turning through the rotor. In fact, the correlation under-estimates the
passage loss contribution, leading to higher pressure levels at the stator outlet. This produces also
a higher rotor outlet static pressure (explaining the large discrepancies in the value of 𝛽𝑒,𝑇𝑆 ) and a
lower flow turning through the rotor, but on the other hand it results in lower relative velocities
within the rotor, leading to higher values of tip-clearances losses, which explain the good agreement
obtained for 𝜂𝑇𝑆 . On the contrary, Glassman model leads to an accurate estimation both in terms
of 𝜂𝑇𝑆 , 𝛥𝛽, and 𝛽𝑒,𝑇𝑆 , since it is able to correctly predict nozzle passage loss contribution. This
occurs because Glassman correlation takes into account compressibility effects and is physically
32
2.6. Validation strategy
based, relying on boundary layer theory. Conversely, Rodgers correlation does not take into account
compressibility - since it depends on the velocity - and it is based on a best-fit of experimental data.
The inaccuracy of Rodgers model is further proved by results on De Servi’s turbine. Imposing
the same outlet pressure for both the passage loss models tested, produces a similar prediction
of rotor losses, while leading to different results in terms of 𝜂𝑇𝑆 (Dev. of 5.48% and 0.42% for
Rodgers and Glassman correlations respectively) and 𝛥𝛽 (Dev. of 5.16°and -0.45°for Rodgers and
Glassman correlations respectively). This clearly highlights the inability of Rodgers approach to
predict nozzle passage loss contribution when high Mach numbers are involved.
These outcomes lead to the selection of Glassman correlation as reference method to estimate
nozzle passage losses for the design of high-pressure ratio single-stage RITs operating with organic
fluids. This kind of machines are indeed usually characterized by low momentum flows at the
stator inlet and by transonic/supersonic flows at the stator outlet. This implies that the flow turning
(usually limited for nozzle vanes of RITs) mostly occurs in the front part of the blades, where
the flow velocity is still limited. On the contrary, most of the entropy production occurs in the
second part of the blade channel, downstream the leading edge, due to the strong expansion the
flow undergoes, which leads to high velocities/Mach numbers and, consequently, to high profile (or
friction) losses. These concepts are confirmed also by the inlet angle sensitivity analysis carried out
in Part II for the designed linear cascade (see Section 5.2.1). The features just described, typical
of single-stage high-expansion ratio RITs, make Glassman’s correlations extremely suitable to
estimate stator passage losses, as the loss model is based on compressible boundary layer theory.
It should also be noticed that Glassman’s correlation proved to be effective for both turbines #1
and #4, which feature two different reaction degrees of 0.6 and 0.3, respectively. Conversely, the
validity of Glassman’s model needs to be assessed when other applications and turbine geometries
are considered, such as highly loaded stator vanes in multi-stage machines or subsonic nozzles in
low-pressure ratio turbines. However, it is important to note that these cases are not within the
scope of the present work.
The set of loss models investigated and those finally adopted to tune RITML for small-scale
high-pressure ratio ORC turbines are summarized in Table 2.8, reported at the end of this chapter.
2.6 Validation strategy

This section outlines the validation strategy carried to assess the robustness and performance
prediction capability featuring RITML when applied outside the context of small-size high-pressure
ratio turbines, which have been used to select the most effective passage loss models in Section 2.5.
In particular, the mean-line model resulting from the chosen set of correlations (see Sections 2.3
and 2.5) is further tested over the published experimental/numerical data of the seven benchmark
turbines reported in Table 2.2, which operate with different fluids, in both design and off-design
conditions, and covering a wide range of target expansion ratio and gross power output.
The validation over turbines #1, #3, #6, and #7 was performed at same 𝛽𝑒,𝑇𝑆 , as available
datume, while Spence’s turbine (#2) was tested at same mass flow rate. This choice is motivated
by significant uncertainty featured by pressure ratio experimental data, as illustrated below. The
validation over choked supersonic turbines, namely #4 and #5, has been carried out both at same
mass flow rate and at the same 𝛽𝑒,𝑇𝑆 , which was matched by modifying the required post-expansion
pressure ratio 𝛽𝑒,𝑃𝐸 to meet the test case operating point. Thanks to CFD/experimental data
availability, validation vs turbines #2, #3, #4, and #7 has been performed for different operating
points, evaluating off-design performance for different turbine pressure ratios and rotational speeds.
Experimental investigations over turbine #2 (see Table 2.2) were carried out at fixed inlet total
temperature 𝑇𝑡1 = 400 K, by adopting an abrupt opening downstream the turbine rotor to discharge
in ambient the flow outgoing the turbine. Different rotational speed (between 50 to 70 kRPM)
33
90
82
ηT S %
75
68
60
3.00
RIT-ML 53033 rpm
EXP 53033 rpm
2.62
RIT-ML 70710 rpm
EXP 70710 rpm
βe,T S
2.25
1.88
1.50
0.25 0.30 0.35 0.40 0.45
ṁ [kg/s]
Figure 2.3: Validation over Spence’s turbine for two different rotational speeds. Black and gray
symbols refer to mean-line model and experimental results respectively.
and pressure ratios (between 1.7 and 2.7) have been investigated, the latter by acting on the inlet
total pressure 𝑝 𝑡1 . The validation has been performed in the whole range of 𝛽𝑒,𝑇𝑆 experimentally
investigated (1.7-2.7) and for 2 different rotational speeds, namely 53033 and 70710 RPM. The
results of the validation over Spence’s turbine are shown in Figure 2.3. Concerning 𝜂𝑇𝑆 , very
good agreement with experimental data can be noticed in terms of both absolute value (maximum
deviation lower than 2%) and trend prediction. Contrarily, although the dependence on the mass
flow rate is correctly reproduced by the mean-line model, significant discrepancies occur for the
absolute value of the overall pressure ratio 𝛽𝑒,𝑇𝑆 . Such deviations are most probably to be ascribed
to inaccuracies in the outlet static pressure measurements, performed through taps located in the
backward plate of the sudden enlargement introduced to connect the turbine exit to the downstream
pipe discharging into the environment. As confirmed by Fearn et al. (1990), the taps are likely to
be in a recirculation zone, leading to discrepancies between the measured pressure (often named
as base pressure) and the one at the rotor outlet 𝑝 6 , introducing unknown uncertainties on the
claimed 𝛽𝑒,𝑇𝑆 . To take into account this phenomenon, the outlet static pressure measurements
may be corrected following the results reported in Sethuraman & Khan (2016) concerning base
pressure drops in recirculating regions. This procedure seems to largely reduce relative deviations
on 𝛽𝑒,𝑇𝑆 (up to 0.5%-3% depending on the operating point). These considerations motivate the
choice mentioned before of performing the comparison at given mass flow rate.
Experiments over turbine #3 (Mclallin & Haas, 1980) were carried out at fixed inlet total
conditions (𝑝 𝑡1 = 1.379 bar and 𝑇𝑡1 = 322.2 K) and by varying turbine pressure ratio (acting on
the downstream conditions) and rotational speed. The corresponding validation procedure has
been carried out at two different rotational speed, namely 29936 and 33262 rpm (90% and 100%
of nominal rotational speed), with the pressure ratio ranging from 1.6 to 3.4. As highlighted by
validation results shown in Figure 2.4, the variation of both mass flow rate and 𝜂𝑇𝑆 as a function of
the pressure ratio is accurately reproduced by the mean-line model. The predicted absolute values
are in good agreement with experimental data, since the maximum deviation is about 3% for the
mass flow rate (relative), and lower than the 3.5% for 𝜂𝑇𝑆 , except for low values of 𝛽𝑒,𝑇𝑆 at 33262
RPM, where it reaches 4.5%. The results do not take into account measurement uncertainties,
since they are not provided in Mclallin & Haas (1980).
34
90
82
ηT S %
75
68
60
0.32
0.29
ṁ [kg/s]
0.26
RIT-ML 29936 rpm
EXP 29936 rpm
0.23
RIT-ML 33262 rpm
EXP 33262 rpm
0.20
1.50 1.80 2.10 2.40 2.70 3.00 3.30
βe,T S
Figure 2.4: Validation over Mc Lallin’s turbine for two different rotational speeds. Black and gray
symbols refer to mean-line model and experimental results respectively.
Table 2.5: Fluid properties of Schuster’s Turbine.
FLUID
R134a Heptane Propane
𝑝 𝑇1 [𝑏𝑎𝑟] 1.379 26 65.6
𝑇𝑇1 [𝐾] 322.2 593.15 423
𝑚¤ [𝑘𝑔/𝑠] 0.393-0.557 6.333-7.971 13.707-18.425
𝛽𝑒,𝑇𝑆 1.50-2.61 1.387-1.982 1.389-2.192
𝜔[𝑟 𝑝𝑚] 15726 15726 18876
Turbine #7 was investigated by Schuster through 3D-CFD steady state simulations for three
different fluids, namely R134a, Heptane, and Propane, and for several turbine pressure ratios
ranging between about 1.4-2.7, the exact range of off-design conditions being a function of the
working fluid considered. Table 2.5 details the total inlet conditions, the rotational speed, and
the ranges of mass flow rate and pressure ratio investigated for all the three fluids considered. As
highlighted by validation results shown in Figure 2.5, the mass flow rate values on the y axis
have been normalized for each working fluid to the corresponding maximum mass flow rate 𝑚¤ 𝑐𝑚
indicated by Schuster et al. (2020). This choice is motivated by the huge difference in terms of
mass flow rate and power level that occurs operating the turbine with R134a and Propane (10.8 kW
for R134a vs 573 kW for Propane at nominal design point). The variation of both mass flow rate
and 𝜂𝑇𝑆 as a function of the pressure ratio is accurately reproduced by RITML for all the three
operating fluids. Concerning 𝜂𝑇𝑆 , the deviation is lower than 2% for most the operating points
associated with all the three fluids and becomes larger only for strong off-design operations (smaller
pressure ratios), reaching a maximum value of 4.7% for R134a. The average relative deviation for
the mass flow rate is about 3.5%, while the largest discrepancies occur when the turbine is operated
with R134a, for which mass flow rate discrepancies range between 3.62%-6.39%.
The comparison against Jones’ turbine (Jones, 1996) has been already carried out at fixed mass
flow rate in the loss sensitivity analyses detailed in Section 2.5, in order to evaluate the effectiveness
of the selected passage loss models in predicting the pressure ratio. However, in this context, the
35
90
82
ηT S %
75
68
60
1.10
0.97
ṁ/ṁem
0.85
RIT-ML R134a
CFD R134a
RIT-ML Heptane
0.72 CFD Heptane
RIT-ML Propane
CFD Propane
0.60
1.40 1.80 2.20 2.60
βe,T S
Figure 2.5: Validation over Schuster’s turbine for three different fluids. Black and gray symbols
refer to mean-line and CFD data results respectively.
90
82
ηT S %
75 Design point
68
RIT-ML
CFD
60
25.00 30.00 35.00 40.00 45.00
βe,T S
¤ values. Black and gray

Figure 2.6: Validation over De Servi’s turbine for different 𝑝 𝑇1 and 𝑚
symbols refer to mean-line model and CFD results respectively.
validation has been performed at fixed 𝛽𝑒,𝑇𝑆 to evaluate efficiency and mass flow rate deviations.
The deviation for the efficiency is 0.34% (84.06% vs 84.40% resulting from experiments), and for
the mass flow rate the relative one is 2.59%. As described below, the validation results concerning
efficiency and mass flow rate prediction capability are summarized in Figure 2.7 and in Table 2.7
respectively.
Regarding turbine #4 (De Servi et al., 2019), besides the nominal point (𝛽𝑒,𝑇𝑆 = 40.84) has
been already investigated at the beginning of this section, the off-design operating map reported
in De Servi et al. (2019) has been used for a further comparison. Being the turbine choked in
the entire pressure ratio range (24-46), the choking mass flow rate has been estimated through
a routine implementing an isentropic flow model as a function of the total inlet conditions (𝑝 𝑇1
changes case by case) and of the geometric throat area, due to the lack of published data concerning
choking mass flow rate in off-design conditions. The trend, depicted in Fig.2.6, is overall well
reproduced with deviations on the total-to-static efficiency below 1.05%, also considering that
pressure ratio varies between 60%-115% of the nominal one. Additionally, for this test case, a
dedicated loss breakdown analysis has been carried out at nominal design point and results are
reported in Table 2.6. The efficiency is correctly reproduced by the reduced-order model here
36
Table 2.6: De Servi’s test case turbine, loss breakdown analysis. Comparison of mean-line model
(RITML) and CFD results.
𝜟𝜼𝑻 𝑺 %
𝜼𝑻 𝑺 %
Nozzle Rotor Exit KE
CFD 84.10 8.40 3.70 3.80
RITML 84.52 7.48 4.02 3.98
Dev 0.42 -0.92 0.32 0.18
¤ for the subsonic turbine test cases whose validation

Table 2.7: Relative percentage deviation on 𝑚
has been performed at same 𝛽𝑒,𝑇𝑆 . Symbols represent the corresponding point in Figure 2.7
Test case #1 #3 #3 #6
Symbol ◀ ■ ▶
Dev. 2.59% -2.87% -3.05% 7.53%
described, featuring a deviation lower than 1%, while the entropy production through the nozzle is
slightly under-predicted by around 0.9 percentage points. This proves a satisfactory agreement
between RITML and CFD results for the loss generation through each component, even though the
0.9 percentage point difference may indicates that the nozzle is the component that would benefit
the most from a better modelling of passage loss for highly-supersonic flows.
Moving to ORC turbine #5 (Alshammari et al., 2017), the present mean-line analysis yields a
Mach number at nozzle outlet 𝑀3 of 1.63, close to the one obtained from 3D CFD simulations,
namely 1.60. The predicted efficiency is 83.7% against the 81.3% resulting from CFD computations,
with a deviation of 2.4%. Nonetheless, Alshammari declares a deviation by up to 4% on the total-
to-total enthalpy drop estimated by means of the Ansys built-in thermodynamic library employed,
hence entailing an intrinsic uncertainty on the declared value of the total-to-static efficiency.
Concerning the validation over turbine #6 (Sauret & Gu, 2014), the predicted efficiency is
82.88% vs the 83.1% resulting from CFD computations, with a deviation lower than 0.25%.
Conversely, the relative deviation on the mass flow rate is 7.53%. The synoptic view of the
validation results, in terms of predicted vs reference 𝜂𝑇𝑆 , is depicted in Figure 2.7 for turbines #1,
#4, #5, and #6 at design point, together with two operating points at two different rotational speeds
for test cases #2 and #3. Table 2.7 details the relative deviations on the mass flow rate for the
subsonic turbine test cases whose validation has been performed at same 𝛽𝑒,𝑇𝑆 .
The validation outcomes over the seven benchmark turbines considered highlight the reduced-
order model effectiveness to predict, with acceptable deviations, the total-to-static efficiency and
the mass flow rate (or overall pressure ratio) regardless of the fluid, power, and expansion ratios
involved. In particular, the deviation between RITML-predicted 𝜂𝑇𝑆 and reference data is always
lower than 2.5 percentage points when the design point of test case turbines is considered, being the
largest discrepancy (2.4 percentage point) associated to Alshammari et al. (2017) turbine, whose
numerical results are however affected by uncertainties concerning the total-to-total enthalpy drop.
Moreover, the mean-line model was also employed to estimate off-design performance of 4 turbines,
two air-operated and two of ORC type. The results prove also the effectiveness of the model in
predicting the performance variations at off-design conditions as well, being the discrepancies
always lower than 3.5 percentage points, other than at strong off-design conditions for Mclallin &
Haas (1980) and Schuster et al. (2020) turbines, for which the maximum discrepancy is about 4.5%.
37
88
McLallin @33262 rpm
McLallin @29936 rpm
De Servi
Alshammari
Jones
Sauret
84 Spence @64818 rpm
Spence @70710 rpm
T S %- Reference
80
76
0%deviation
2.5
3%deviation
72
72 76 80 84 88
T S %- RIT-ML
Figure 2.7: Summary of validation procedure over test case turbines in terms of 𝜂𝑇𝑆 . The two
operating point of Mc Lallin’s turbine are respectively: (𝛽𝑒,𝑇𝑆 = 3.25, 𝜔 = 33262 RPM) and
(𝛽𝑒,𝑇𝑆 = 1.6, 𝜔 = 29936 RPM). The two operating point of Spence’s turbine are respectively:
(𝑚¤ = 0.40, 𝜔 = 64818 RPM) and (𝑚¤ = 0.35, 𝜔 = 70710 RPM).
2.7 Concluding remarks and key findings

In the present work a reduced-order model was developed for the preliminary design of single-stage
radial inflow turbines, specifically tailored to machines operating with organic fluids and featuring
reduced size (small power) and large pressure ratio. Moreover, a comprehensive review of suitable
loss models available in literature was carried out, by retracing correlation development pattern up
to the original approaches, as thoroughly discussed.
The mean-line model, described in the first part of the section, allows to select both converging
and de Laval-type nozzles, which are often adopted in applications involving large expansion ratios.
The methodology used to accomplish the design process features flexibility and modularity, since
independent routines are used to solve the flow within each turbine component, namely the nozzle,
the vaneless interspace, the rotor, and optionally a downstream conical diffuser.
Seven benchmark radial inflow turbines operating with different fluids and covering a wide
range of target expansion ratio, size, and gross power output are considered to carry out a validation
procedure as general and unbiased as possible. Among these geometries, two turbines characterized
by reduced dimensions and high pressure ratios were chosen to test the effectiveness of different
passage loss correlations. Concerning the nozzle, this analysis revealed the poor prediction
capability of the correlation by Rodgers (1987) when high Mach number flows are involved, leading
to the selection of the model by Glassman (1976a) as the reference to estimate stator passage
losses. Focusing on the rotor, the employment of the method proposed by Ventura et al. (2012)
on the basis of Rodgers is discouraged when high-pressure ratio turbines operating with organic
38
2.7. Concluding remarks and key findings
fluids are involved, also considering the uncertainties related to this loss model and the several
modifications it underwent during the years. Moreover, the outcomes of the analysis revealed the
poor prediction capability when applied to the rotor of the correlation by Glassman (1976a), which
tends to under-predict the entropy generation neglecting the blade loading effects. Therefore, the
aforementioned analysis suggests the adoption of the model by Baines et al. (1998) as the reference
to estimate rotor passage losses, although its employment may suffer from some limitations when
very large pressure ratios together with sub-atmospheric rotor outlet pressures occur.
In the last part, the outcomes of the general validation procedure over the seven benchmark
turbines considered are discussed, which evidence the robustness and prediction capability of the
proposed mean-line model, in terms of total-to-static efficiency and mass flow rate (or overall
pressure ratio), independently of the machine investigated. In particular, RITML usually tends to
over-predict turbine efficiency, coherently with the intrinsic limitations of reduced-order models,
which do not consider some possible critical aspects such as detailed shock patterns within the
channels, leakage flows, and loss contributions interaction. The deviation between RITML-predicted
𝜂𝑇𝑆 and reference data is always lower than 2.5 percentage points when the design point of test case
turbines is considered, being the largest discrepancy (2.4 percentage point) associated to turbine
#5, whose numerical results are however affected by uncertainties concerning the total-to-total
enthalpy drop. Concerning the mass flow rate, the relative deviation between RITML and reference
data is mostly lower than 3.5%, with some exceptions for which the deviation reaches the 7%.
Moreover, the mean-line code proved its effectiveness also in reproducing the performance trends at
off-design conditions for 4 turbines, two air-operated and two of ORC type. The deviations remain
indeed limited to 3.5 percentage points for the efficiency, except for strong off-design conditions of
turbines #3 and #7, for which the maximum deviation is about 4.5%.
The proposed reduced-order model in design mode can be readily employed within the design
and optimization process of ORCs. Moreover, thanks to the low computational cost, the design
mode can be exploited for an early and complete sampling of a wide design space, prior to resorting
to time-consuming high-fidelity CFD simulations. This allows to easily retrieve the optimal
values of the main dimensionless quantities as a function of specific parameters defining the
application considered (such as size parameter, volume ratio and rotor dimensionless rotational
speed, see Section 3.4), contributing to the future development of tailored design guidelines.
Chapter 3 illustrates all the potentials associated to RITML, applying the method for a systematic
optimization of several ORC turbines, aiming at developing a maximum efficiency performance
map for single-stage radial inflow turbine to be integrated within the ORC thermodynamic design
and optimization.
39
Table 2.8: Loss models analyzed and adopted in RITML.
FINAL SET OF CHOSEN CORRELATIONS

Loss item Nozzle Reference
𝐴 −0.2
𝜗𝑇𝑂𝑇 𝑅𝑒 𝑙 3𝐷
𝐸 𝑅𝑒𝑟𝑒 𝑓 𝑠 𝐴2𝐷
𝑙
Passage ∗
𝛥ℎ 𝑝𝑎𝑠𝑠 = ℎ𝑡1 − ℎ3,𝑖𝑠 𝑟𝑒 𝑓
Glassman (1976a)
−0.2
𝑡 𝜗𝑇𝑂𝑇
cos 𝛼∗ − 𝑠3 −𝐻 𝑅𝑒
𝑅𝑒𝑟𝑒 𝑓
𝑙
𝑠
3 3 𝑙
𝑟𝑒 𝑓
𝛾∗
!
3
𝛾 ∗ −1
2
1−𝛾 ∗

𝑍 𝑡
Trailing-edge 𝛥ℎ𝑡𝑒 = 2 𝜋𝑟 𝑠cos3 𝛼∗ 12 𝑉3∗ 2 1 + 32 𝑀3∗ 2 3 Glassman (1976a)
3 3
2
𝐿 𝑉3 +𝑉4
Vaneless friction 𝛥ℎ 𝑓 = 𝐶 𝑓 𝐷ℎ 12 2 Kastner & Bhinder (1975)
ℎ
Loss item Rotor Reference

Incidence 𝛥ℎ𝑖𝑛𝑐 = 12 𝑊42 sin2 𝛽4 − 𝛽4,𝑜 𝑝𝑡 Wasserbauer & Glassman
(1975)
𝑊 2 +𝑊 2
𝑟 2
𝐿ℎ 6,𝑟 𝑚𝑠 cos 𝛽6,𝑟 𝑚𝑠 4 6,𝑟 𝑚𝑠
Passage 𝛥ℎ 𝑝𝑎𝑠𝑠 = 𝐾 𝑝 𝐷ℎ + 0.68 1 − 𝑟4 𝑏6 /𝑐𝑟 2 Baines et al. (1998)
!2
𝛾6
𝑍𝑟 𝑡6 1 𝛾6 −1 1−𝛾6
Trailing-edge 𝛥ℎ𝑡𝑒 = 2
2 𝑊6,𝑟 𝑚𝑠 1+ 2
2 𝑀𝑤6,𝑟 𝑚𝑠 Glassman (1976a)
𝜋 𝑟6,ℎ𝑢𝑏 +𝑟6,𝑡𝑖 𝑝 cos 𝛽6,𝑟 𝑚𝑠
3
𝑢 𝑍𝑟 √
Clearance 𝛥ℎ𝑐𝑙 = 48 𝜋 𝐾𝑎 𝜀𝑎 𝐶𝑎 + 𝐾𝑟 𝜀𝑟 𝐶𝑟 + 𝐾𝑎𝑟 𝜀𝑎 𝜀𝑟 𝐶𝑎 𝐶𝑟 Spraker (1987)
¯ 3𝑟2
𝜌𝑢
Windage 𝛥ℎ𝑤𝑖𝑛 = 𝑘 𝑓 24𝑚¤ 4 Daily & Nece (1960)
1 2
Kinetic energy 𝛥ℎ 𝑘𝑒 = 2 𝑉6 [-]
DISCARDED CORRELATIONS
Loss item Nozzle Reference
3 tan 𝛼∗ 𝑠3 cos 𝛼∗

0.05 1 ∗,2
Passage 𝛥ℎ 𝑝𝑎𝑠𝑠 = 𝑠3
3 + 3
2 𝑉3 Rodgers (1987)
𝑅𝑒0.2 𝑐𝑠
𝑏3
3
Loss item Rotor Reference
h 𝑊 +𝑊 i2
𝐿ℎ 4
6,𝑟 𝑚𝑠 𝐷
Passage 𝛥ℎ 𝑝𝑎𝑠𝑠 = 𝑓𝑡 𝐷ℎ 2 + 𝑍𝑟 𝑟4𝑐 𝑉42 Ventura et al. (2012)
−0.2
𝜗𝑇𝑂𝑇 𝑅𝑒 𝑙 𝐴3𝐷
𝐸 𝑅𝑒𝑟𝑒 𝑓 𝑠 𝐴2𝐷
𝑙
𝑟𝑒 𝑓
Passage 𝛥ℎ 𝑝𝑎𝑠𝑠 = ℎ𝑡6𝑟 − ℎ6,𝑖𝑠𝑠 −0.2 Glassman (1976a)
𝑡 𝜗𝑇𝑂𝑇 𝑅𝑒 𝑙
cos 𝛽6,𝑟 𝑚𝑠 − 𝑠6 −𝐻 𝑅𝑒𝑟𝑒 𝑓 𝑠
6 𝑙
𝑟𝑒 𝑓
40
CHAPTER 3
Turbine Performance Map
integration within ORC
optimization procedure
3.1 Concept of proposed approach: integration of RIT and ORC design tools
As introduced in Section 1.2, several aspects complicate the design of bottoming ORC plants for
small-power applications. These factors include the reduced ORC system dimensions, which often
couple with a challenging integration with the main plant (e.g. the ICE), strong possibility of
non-stationary operations, and difficulties associated with the design of feasible and efficient turbo-
expanders. In the past, volumetric machines were preferred as ORC expanders for small-power
applications (Lemort et al., 2013; Cipollone et al., 2014), since the miniaturized dimensions and
high rotational speeds associated with turbo-expanders lead to several complexities related to their
design and integration with the main plant, as well as to large manufacturing and production costs
(Colonna et al., 2015). These factors limited the large-scale diffusion of turbo-expanders within
mini ORC-based plants. However, there are several applications (such as WHR from the industrial
and transportation sector) featuring medium to high temperature sources and, concurrently, small
power, for which the employment of ORCs would be a promising solution to efficiently recover the
available thermal energy. For such cases the use of volumetric machines, particularly orbital (e.g.
scroll) and rotary (e.g. vane and screw) expanders, is prevented by the relatively high temperatures
and the high pressure ratios (> 15) usually resulting from the thermodynamic optimization of
ORCs characterized by medium to high temperature hot sources (Imran et al., 2016).
The current work deals with on-board waste heat recovery from internal combustion engines of
long-haul trucks as reference small-power high-temperature applications. In this framework, the
41
Chapter 3. Turbine Performance Map integration within ORC optimization procedure
bottoming ORC and main plants integration is further complicated by the choice of optimal hot
sources, whose usage should not jeopardize engine performance, as well as by a non-straightforward
management of the additional mechanical power produced and by the intrinsic unsteady operations
typical of truck engines. These features impact also the turbo-expander design, which has to
efficiently provide large pressures ratios, coupled with low mass flow rates, both at nominal and
off-design conditions. As a result of the aforementioned difficulties, it is clear that the assumption
of constant turbine efficiency during the design of small-scale ORC plant for onboard WHR may
result into misleading optimal cycle configurations having turbine expansion ratios and dimensions
non compatible with the assumed turbine efficiency. The estimation of the actual turbo-expander
performance efficiency during the preliminary steps of the ORC design is therefore of paramount
importance. This allows indeed to properly characterize the ORC-turbine integration and to assess
the effects of realistic and non-constant turbine efficiency values on optimal thermodynamic cycle
layout and optimal working fluid selection.
Several approaches can be exploited to integrate the turbine performance estimation tool within
the ORC design optimization procedure, whose main advantaged and drawbacks are detailed
in Section 1.2. In this work, a performance map is exploited, which provides the maximum
total-to-static efficiency 𝜂𝑇𝑆 of single-stage radial inflow turbines as a function of cycle-dependent
parameters, namely the turbine Size Parameter (SP) and the isentropic Volume ratio (Vr). The
former is a dimensional parameter accounting for machine dimension and relates the turbine outlet
volumetric flow rate with the machine work in isentropic conditions, while the latter accounts
for the fluid compressibility via the volumetric flow rate variation through the machine. More
specifically, these two parameters are defined as
√︁
𝑄¤ 𝑜𝑢𝑡
𝑆𝑃 = , (3.1a)
𝛥ℎ0.25
𝑖𝑠
𝑄¤ 𝑜𝑢𝑡 ,𝑖𝑠
𝑉𝑟 = , (3.1b)
𝑄¤ 𝑖𝑛
where 𝑄¤ and 𝛥ℎ𝑖𝑠 are the volumetric flow rate and the isentropic specific work of the respectively,
while the subscripts 𝑖𝑛 and 𝑜𝑢𝑡 refer to the turbine inlet and outlet stations. Vr and SP were selected
as independent parameters as a function of which retrieve the RIT maximum efficiency, since
they account for scale and fluid properties (molecular complexity and gas-dynamic deviation from
ideality). Mostly because of this reason their use was suggested by previous studies on ORC axial
and radial turbines (Macchi & Perdichizzi, 1981a; Masi et al., 2020; Astolfi & Macchi, 2015). In fact,
the usage of SP and Vr is particularly convenient also for combined cycle/turbomachinery studies,
as they only depend on quantities already available from the cycle thermodynamic optimization
and, hence, can be evaluated a priori with respect to the turbine design. The reader should bear in
mind that this approach holds as long as the maximum turbine efficiency is considered, that is as
long as the selection of the turbine’s optimal shape, size, and rotational speed, all of which have the
major impact on the performance, has been already carried out. Furthermore, the negligibility of
the effects related to other parameters, such as the fluid molecular properties and inter-molecular
forces, need to be verified, as stated by the turbine’s flow similarity law (Masi et al., 2020). This
assessment, which allows the maximum RIT efficiency to be expressed as a function of these 2
parameters only, is reported in Section 3.4.
The approach of embedding the optimum efficiency map within the ORC optimization procedure
is not novel and was already used for both axial (Astolfi & Macchi, 2015; La Seta et al., 2016) and
radial-inflow turbines (Perdichizzi & Lozza, 1987; Da Lio et al., 2017a; White & Sayma, 2019).
As detailed in Section 1.3, the novelties introduced in the present work concern the procedure
exploited to build the maximum efficiency map, the RIT nozzle geometry considered, and the
ranges of SP and Vr investigated, all necessary to deal with the high-temperature and small-power
42
3.2. Numerical routines description
applications object of this thesis. The performance map was obtained exploiting the RITML model
to carry out high-order RIT optimizations for a discrete number of SP and Vr values, considering
a large number design variables (up to 10) and several constraints, required to ensure feasible
geometries for the optimized machines. In fact, the ranges of Vr and SP investigated here are wider
than those reported in literature (Da Lio et al., 2017a) and shifted towards higher Vr and lower SP,
as described in Section 3.3. This required to focus on and analyze converging-diverging nozzles in
choked conditions, differently from previous studies, which limit the investigations to converging
nozzle architectures (Perdichizzi & Lozza, 1987; Da Lio et al., 2017a).
The following sections of this chapter describe the methodology adopted to integrate the RIT
performance estimation tool within the ORC optimization procedure, including the selection of
suitable ranges of SP and Vr, and the verification of the effectiveness of their use as the only two
independent parameters for the turbine analyses. A thorough description of the optimizations
carried out to build the turbines’ maximum efficiency map is reported, complemented by a
discussion on the main parameters driving loss generation through RIT components and on the
variation of the losses for different SP and Vr values, with the aim of defining some general
guidelines for the preliminary design of small-size ORC RITs. Finally, the effects of a realistic
and non-constant expander efficiency values during the cycle optimization process are described,
focusing on their impact on optimal thermodynamic cycle layout and working fluid selection.
To assess the potentials of the methodology developed, a reference diesel ICE is selected,
namely the MAN D2676 (Yang et al., 2018), having 316 kW nominal brake power output at 1800
rpm. The engine exhaust is selected as single hot stream providing the thermal power to the ORC
unit, since this thermal source typically features large availability, representing around the 30% of
the fuel inlet power (Dolz et al., 2012), and leads to exergy efficiencies that are potentially high
due the large temperature levels involved (Zhu et al., 2013; Ringler et al., 2009). A rigorous and
dedicated study concerning an optimal integration of the ORC system with the main plant (i.e.
the ICE of long-haul trucks), including an optimization of the number and kind of hot sources,
lies outside the scope of the present work. Preliminary analyses are carried out considering as
reference exhaust thermodynamic conditions those obtained at 50% of the engine load. In this
condition, the hot exhaust stream, modelled as ideal gas with constant specific heat (1.035 kJ/kgK),
has mass flow rate equal to 0.2 kg/s and temperature at the primary heat exchanger (PRHE) inlet
equal to 351°C. In Section 3.6, different exhaust mass flow rate values are investigated to simulate
either a different engine operating point or a smaller-sized ICE, this aspect being of paramount
importance to modify the ORC power output and to assess the dependency of the optimal design
and working fluid on the plant size.
3.2 Numerical routines description

Two numerical models have been developed and integrated in the present work: the first one
implements a reduced-order method for the preliminary sizing of single-stage RIT (RITML) and is
thoroughly described in Chapter 2, while the second one deals with the ORC cycle design and
optimization procedure. Section 3.2.1 describes the latter numerical routine, while Section 3.2.2
details the optimization algorithms coupled with RITML and the strategy implemented to carry out
the optimizations required to retrieve the turbine maximum efficiency map.
3.2.1 ORC design and optimization tool

The numerical code developed for the design of small-scale ORC power systems implements
several features that make it a flexible tool for system optimization and techno-economic analysis.
However, the potential of the code is not fully exploited in this work according to the scope of
43
Table 3.1: ORC model assumptions. Phase-transition pressure drops are introduced as temperature
difference 𝛥𝑇 before and after the transition.
𝛥𝑇 heat exchangers Heat exchangers pressure drops

𝛥𝑇 𝑝 𝑝, 𝑃𝑅𝐻 𝐸 , °C 10 𝛥𝑇𝑐𝑜𝑛𝑑 , °C 0.5
𝛥𝑇 𝑝 𝑝,𝑐𝑜𝑛𝑑 , °C 2 𝛥𝑃 𝑑𝑒𝑠ℎ , % 0.02
Components efficiency 𝛥𝑃𝑒𝑐𝑜 (sub), bar 0.5
𝜂 𝑝𝑢𝑚 𝑝 0.5 𝛥𝑇𝑒𝑣𝑎 (sub), °C 1
𝜂𝑡𝑢𝑟 𝑏 Fixed or Calculated 𝛥𝑃𝑠ℎ (sub), bar 0.5
𝜂𝑒𝑙,𝑚𝑒𝑐𝑐,𝑔𝑒𝑛 0.98 𝛥𝑃 𝑃𝑅𝐻 𝐸 (sup), bar 1
𝜂𝑒𝑙,𝑚𝑒𝑐𝑐, 𝑝𝑢𝑚 𝑝 0.98 𝛥𝑃𝑟 𝑒𝑐 (sup), % 0.002
𝜂 𝑓 𝑎𝑛 0.7 𝛥𝑃 𝑎𝑖𝑟 , mbar 22
the thesis which is to investigate the strong link between turbine performance and cycle design
rather than determine optimal cycle architecture from techno-economic perspective. Despite the
numerical code can analyze ORC systems integrated with multiple thermal power inputs, a single
heat source is considered, namely the exhaust gases of the reference diesel engine (MAN D2676),
as already discussed in Section 3.1. The ORC design method is implemented in a numerical routine
developed in the Python language and adopting CoolProp (Bell et al., 2014) for the calculation
of fluid thermodynamic properties, an open-source fluids library that can optionally recall the
NIST REFPROP backend (Lemmon et al., 2018). This feature makes the model flexible to deal
with any fluid available in CoolProp (or REFPROP) database. A single pressure level cycle is
selected in both subcritical and supercritical configurations, being saturated cycles considered as a
subclass of subcritical ones with no vapor superheating, and the employment of the recuperator can
be optionally activated. The ORC optimization procedure aims at minimize/maximize a selected
objective function by varying a certain number of design variables. In this research work, the
selected objective function is the ORC plant system efficiency, which is proportional to the net
power output and can be expressed as the product of cycle thermodynamic efficiency and heat
recovery factor, as reported in Equation 3.2. However, the model can be easily adapted for more
complex analyses as techno-economic (minimization of system specific cost or levelized cost of
electricity) and multi-objective optimizations.
𝑂𝑏 𝑗 𝐹𝑢𝑛𝑐. : 𝑊𝑒𝑙 = 𝑊𝑔𝑒𝑛 − 𝑊 𝑝𝑢𝑚 𝑝 − 𝑊 𝑓 𝑎𝑛 , (3.2a)

𝑊𝑒𝑙 𝑄 𝑖𝑛
𝑂𝑏 𝑗 𝐹𝑢𝑛𝑐. : 𝜂 𝑠𝑦𝑠𝑡𝑒𝑚 = 𝜂 𝑐𝑦𝑐𝑙𝑒 𝜒𝑟 𝑒𝑐𝑜𝑣𝑒𝑟 𝑦 = (3.2b)
𝑄 𝑖𝑛 𝑄 𝑖𝑛,𝑚𝑎𝑥
Four parameters can be currently selected by the user as DVs: (i) condensation temperature, cycle
(ii) maximum pressure and (iii) maximum temperature and (iv) recuperator pinch point temperature
difference. In the analyses described in the following sections, the first three parameters are
always referred to as DVs while different optimizations are carried out both implementing or not
the recuperator. Table 3.1 reports all plant quantities other than design variables needed for the
complete cycle definition, together with the corresponding numerical values considered in the
present study. These values are derived from literature review (Astolfi et al., 2017) and are kept
constant during the plant optimization procedure independently of the working fluid selected and
the maximum/minimum pressure of the cycle. Among them, the turbine efficiency can be set
equal to a fixed value or calculated as function of SP and Vr by means of the performance map
described in Section 3.5, in order to have a clear link between cycle DVs and expander performance.
44
3.2. Numerical routines description
Table 3.2: Optimization procedure DVs and corresponding variation ranges.
DVs Variation Ranges DVs Variation Ranges

𝜔𝑠 0.7-1.3 𝛽𝑒,𝑃𝐸 0.9-1.1
𝛼3 , deg 70-81 𝑟 6,ℎ𝑢𝑏 , m 0.006-0.1
𝑟 6,ℎ𝑢𝑏 /𝑟 6,𝑠ℎ 0.1-0.7 𝑟 4 /𝑟 6,𝑠ℎ 1.15-1.5
𝑟 3 /𝑟 4 1.005-1.1 𝑟 1 /𝑟 3 1.1-1.5
𝑏 4 /𝐷 4 0.03-0.09 𝐿 𝑧 /𝑟 4 1.1-1.5
Some constraints are considered in the optimization procedure, among which the most limiting
are: avoidance of two-phase flow expansion (i.e. vapor quality larger than one along the whole
expansion) and maximum volume ratio Vr equal to 60 in order to avoid unfeasible single-stage
turbine designs. Moreover, a minimum condensing pressure of 0.1 bar is considered to avoid a
strong pressure difference with respect to ambient conditions, which would strongly complicate
the design of the sealing system and, consequently, the integration of ORC unit with primary
power plant. ORC is condensed in a direct dry air-cooled heat exchanger featuring a design
resembling an engine radiator, with nominal air ambient temperature assumed as equal to 25 °C.
A high-rotational speed generator is considered in order to allow for effective turbine efficiency
optimizations avoiding the use of gearbox. As detailed in Section 3.2.2, the optimization routine
includes a global gradient-free optimization procedure implementing a self-adaptive Differential
Evolution algorithm and a local one exploiting SLSQP, a sequential quadratic programming (SQP)
algorithm for nonlinearly constrained gradient-based optimizations.
3.2.2 RIT optimization strategy

As stated in Section 3.1, the turbine performance map to be integrated within the ORC design
procedure is obtained exploiting the in-house developed mean-line model RITML. To compute
the maximum RIT efficiency for a specific set of inlet conditions - namely inlet total pressure
and temperature, mass flow rate, and outlet pressure, resulting from ORC design outcomes - the
reduced-order method needs to be coupled with and outer optimization algorithm. The number and
kind of DVs to be included in the optimization problem may be freely set by the user among the set
of inputs required by RITML, making the code flexible and suitable for multi-level design strategies.
Table 3.2 lists the parameters considered as DVs in the optimizations developed to retrieve the RIT
maximum efficiency map and the corresponding ranges of variation, which define the design space
investigated by the optimization algorithms. In Table 3.2 the subscripts 3, 4, 6 refer to nozzle outlet,
rotor inlet and rotor outlet respectively. 𝛽𝑒, 𝑃𝐸 is the post-expansion/post-compression pressure
ratio, defined in Section 2.2.2 as the ratio between the pressure at the inlet of the semi-bladed
region over the one at the nozzle outlet (i.e. at radius 𝑟 3 ). Several constraints are considered in
the present analysis to supervise those parameters which are not directly optimized and to avoid
non-physical solutions. It is worthwhile to highlight that the lower limit of each physical dimension
of the turbine was constrained to a certain minimum value (e.g. rotor hub radius > 6 mm, rotor
axial length > 5 mm, stator heights and rotor inlet height > 2 mm, trailing edge > 0.15 − 0.25
mm for stator and rotor respectively, axial, radial and backplate clearance > 0.1 mm, nozzle radial
chord > 5 mm) to ensure the feasibility of the proposed geometry. Other constraints concern
the flow field, such as for example the maximum Mach number at the nozzle outlet (< 2.5), the
maximum flow turning within the rotor (< 120°), or the maximum relative Mach number at the
rotor inlet/outlet (< 1).
The optimization strategy comprises a global gradient-free procedure, which allow to scan
45
Table 3.3: R1233zd(E) fluid main parameters.
𝑇𝑐𝑟𝑖𝑡 , °C 𝑃𝑐𝑟𝑖𝑡 , bar 𝜌 𝑐𝑟𝑖𝑡 , kg/m3 𝑇 Max, °C Molar Mass, kg/kmol

165.6 35.709 476.31 240 (250) 130.5
the entire design space to identify the region where the global optimum is located, followed by
a local gradient-based optimization routine, which allows to fine-tune the optimum values and
may help in the identification of a feasible champion for highly-constrained problems. To scan the
whole design space, the self-adaptive Differential Evolution algorithm DE 1220 available within
the open-source package PaGMO (Biscani & Izzo, 2020) is employed. This algorithm could not
be employed for constrained optimization problems since it does not include built-in functions
to handle equality and/or inequality constraints. Therefore, an augmented objective function is
employed by adding penalties proportional to the number and amount of violated constraints such
that the aforementioned algorithm can be used also for constrained optimizations. DE 1220 is a
variant of the basic Differential Evolution optimizer (Storn & Price, 1997) - probably the most
effective among meta-heuristic evolutionary strategies - that makes use of self-adaptation schemes
for both the weight coefficient (f), the crossover probability (CR), and the mutation variant. The
number of generations and the size of the population, whose product define the total number of
objective function evaluation and RITML runs, represent therefore the only parameters left to be
specified by the user. In the current work, 100 to 200 generations and 30 to 60 individuals are
considered depending on turbine inlet conditions. Particularly, both number of generations and
population sizes need to be increased for lower values of SP, this parameter being an indication of
turbine physical dimension. Indeed, small SP values are associated with miniaturized machines
and imply larger efforts of the optimization algorithm to find feasible optimal geometries that
concurrently meet all the constraints related to minimum turbine dimensions.
The local optimization routine exploits SLSQP, a routine available in the open-source library
NLopt that implements a sequential quadratic programming (SQP) algorithm for non-linearly
constrained gradient-based optimizations (supporting both inequality and equality constraints). If
not available, the algorithm provides the gradient estimates evaluating the objective function as
many times as the number of DVs involved in the optimization problem. Moreover, a multi-starting
feature can be optionally activated specifying the number of initial design variable vectors to
generate randomly starting from the one provided by the user and from a perturbation relative
percentage. For larger SP values the stand-alone global optimization strategy with a relatively low
number of generations and individuals resulted effective to find a feasible optimum set of DVs.
Contrarily, when small SP values were investigated, the addition of the local optimization routine
with 3 to 6 multi-starting individuals turned to be necessary. The combination of these techniques
allowed to find optimum turbine geometries fulfilling all the constraints activated for SP values up
to 7 mm, as better described in Section 3.5, greatly exceeding the minimum SP investigated by the
other works in literature adopting the turbine maximum performance map.
3.3 Preliminary considerations on ORC design

The ORC numerical code has been adopted for a preliminary analysis in which several hundreds of
cycles are designed considering different constant turbine efficiency values. Hydro-fluoro-olefin
R1233zd(E) has been selected as working fluid in this analysis as an environmentally friendly
drop-in substitute fluid for R245fa with zero ODP, extremely low GWP, and low safety issues
(e.g. non-flammable), and therefore widely suggested as a promising candidate for automotive
46
3.3. Preliminary considerations on ORC design
Figure 3.1: Domain of the three design variables (𝑇𝑐𝑜𝑛𝑑 , 𝑃𝑚𝑎𝑥 , and 𝑇𝑚𝑎𝑥 ) together with 𝑇 − 𝑠
diagram of R1233zd(E).
climatization as it is compatible with present and future regulation (Committee, 2020). Table 3.3
resumes the R1233zd(E) fluid main properties, in which the maximum fluid temperature is to be
intended as equal to the limit reported in the equation of state available in the REFPROP database.
The objective of such analyses is to retrieve useful ranges of SP and Vr to set as bounds for the
development of the expander maximum efficiency map. Both recuperative (𝛥𝑇 𝑝 𝑝,𝑟 𝑒𝑐 = 2°C) and
non-recuperative cycles are investigated while three turbine efficiency values are adopted, namely
70%, 80% and 90%. The others three design variables defined in Section 3.2.1 are varied in the
following ranges in order to cover the whole design space for the ORC cycle:
• Cycle maximum pressure. Both subcritical and supercritical cycles are investigated in the
range between 7.4 bar (reduced pressure 𝑃𝑟 = 𝑃/𝑃𝑐𝑟 around 0.2), corresponding to a low
evaporating temperature of 85°C, and 107.2 bar, corresponding to 3 times the critical pressure
(𝑃𝑟 = 3);
• Cycle condensation temperature. Minimum value (29°C) corresponds to very small pinch
point temperature difference and wide heat transfer area at the condenser and implies a
minimum pressure of 1.5 bar, compatible with the avoidance of non-condensable gases
inward leakage. Maximum value (80°C) is selected in order to investigate also supercritical
cycles with non-excessive pressure ratio;
• Turbine inlet temperature. Minimum value is fixed by the saturation condition for medium-
low evaporation temperature subcritical cycles while it is affected by the constraint of
single-phase flow expansion for high pressure subcritical and supercritical cycles. Maximum
value is defined by the fluid thermal stability temperature.
The investigated design space for the six cycles considered - corresponding to 3 turbine efficiency
values applied to a one-pressure level cycle either regenerative or not - is represented in Figure 3.1,
together with the R1233zd(E) saturation curve. The optimal design of the six cycles is reported
in Table 3.4. It is of interest to note that all the optimal configurations are supercritical, adopt a
similar condensation temperature (from 35.5 °C to 38.5 °C), which is higher than the lower bound,
and push the turbine inlet temperature up to the maximum value (240 °C). Optimal cycle efficiency
is strongly affected by the turbine efficiency while working fluid mass flow rate is mainly affected
by the presence of the recuperator. Note that optimal cycle design parameters (i.e. condensation
temperature and cycle maximum pressure and temperature) are only slightly affected by the turbine
47
Table 3.4: Optimal cycle results for the six investigated plants in the preliminary analyses with
R1233zd(E).
Cycle Base Regenerative (𝛥𝑇 𝑝 𝑝,𝑟 𝑒𝑐 =2°C)

𝜂𝑡𝑢𝑟 𝑏 70% 80% 90 % 70% 80% 90 %
Case A B C D E F
𝑇𝑐𝑜𝑛𝑑 , °C 38.5 37.7 37.2 36.6 36.1 35.4
𝑃𝑚𝑎𝑥 , bar 57 60.2 62.7 53.6 56.0 58.7
𝑇𝑚𝑎𝑥 , °C 240 240 240 240 240 240
𝜂 𝑠𝑦𝑠 12.6 15.0 17.4 15.0 17.8 20.6
𝑊𝑒𝑙 , kW 6.8 8.1 9.4 8.1 9.6 11.1
𝑚¤ 𝑂𝑅𝐶 , g/s 169.6 171.4 172.9 201.0 201.6 201.6
𝑆𝑃, mm 8.2 8.3 8.4 9.1 9.2 9.2
𝑉𝑟 30.7 33.7 36.1 29.0 31.8 34.2
Figure 3.2: Maximum system efficiency for Case B: (a) as function of cycle maximum pressure
and condensation temperature at optimal turbine inlet temperature, together with SP and Vr isolines
(b) as function of SP and Vr, together with the envelope of SP and Vr for the different investigated
cycles (lines) and location of optimal designs (markers).
assumed efficiency. This result proves that the execution of several ORC optimizations imposing
different, but constant, values of the turbine’s efficiency (as proposed by Bademlioglu et al. (2018))
may lead to misleading results in terms of optimal ORC power plant architecture. As additional
proof of the ineffectiveness of such approach, the higher the imposed RIT efficiency is, the higher
the turbine Vr associated with the optimum solution results. This is unrealistic if single-stage
turbines are adopted, for which higher Vr values should be associated with lower RIT efficiencies.
Figure 3.2.a depicts the maximum system efficiency as function of condensation temperature
𝑇𝑐𝑜𝑛𝑑 and maximum pressure 𝑃𝑚𝑎𝑥 always at the optimal cycle maximum temperature for a
non-recuperative cycle with 80% turbine efficiency (case B of Table 3.4). Contours of iso-SP and
iso-Vr are also reported together with the optimal design point (red dot). Note that the region of
48
3.4. Assessment of SP and Vr as unique independent parameters
maximum efficiency is extremely wide and extends up to very high cycle maximum pressure and
high condensation temperatures, corresponding to low-intermediate values of SP (0.007 < SP
< 0.009) and quite large Vr (15 < Vr < 55), both spanning in a wide range . Figure 3.2.b depicts
the system efficiency as function of turbine SP and Vr for case B and concurrently delimits the
ranges of SP and Vr for the other cases, reporting the corresponding envelopes (colored lines).
Finally, marker locations in Figure 3.2.b define the optimal solution for each case reported in Table
3.4. Results clearly highlight that the envelops of Vr and SP are quite overlapped and they are
thus not particularly affected by both the adoption of recuperator and the actual value of turbine
efficiency. This suggests that the choice of Vr and SP ranges based on the envelopes reported in
Figure 3.2.b is appropriate and would lead to a turbine performance map definition that can be
used during the cycle optimization independently of the cycle architecture.
3.4 Assessment of SP and Vr as unique independent parameters

Size parameter SP and volume ratio Vr were introduced in Macchi & Perdichizzi (1981b) and
have been recently used to correlate the performance of ORC turbines (Macchi & Astolfi, 2017a).
As clearly discussed by Masi et al. (2020), in the context of optimized machines (for that the
optimization guarantees to identify the optimal specific speed and specific diameter), the efficiency
of an optimized ORC turbine operating in nominal conditions depends on the Reynolds number, on
the compressibility of the fluid, on the machine size, and on the molecular properties of the working
fluid. The effects of Reynolds number can be generally neglected in the present context, due to the
relatively high-speed flows in ORC turbines and the typically low kinematic viscosity of organic
fluids (the Reynolds numbers are of the order of 107 or larger). The effects of compressibility,
usually expressed in terms of Mach number when analyzing conventional machines, in ORC
turbine sizing studies are more effectively correlated by the volume ratio Vr, which is basically the
ratio between the fluid specific volume at the outlet and at the inlet of the machine, as specified
in Equation 3.1. In this way, Vr provides an indication of the compressibility of the fluid and
also of its impact on the change in the cross section necessary to accommodate the change in
volumetric flow rate, which is a crucial parameter in turbines composed by a few stages. The effects
related to the machine size may be effectively correlated by the Size Parameter SP. In fact, this
parameter is an index of the so-called size effects, related to the inherent violation of geometrical
similarity arising when scaling machines from large to small size, due for example to roughness and
clearance values, fixed by the manufacturing technology. Since such effects depend on the actual
size of the machine (and they typically penalize small-scale machines with respect to large-scale
ones), they are properly represented by a dimensional quantity as SP. As reported in Masi et al.
(2020), two non-dimensional parameters are required to properly represent the effects related to the
inter-molecular forces (which contribute to the definition of the thermodynamic state of the fluid),
and to the molecular complexity of the working compound (which contribute to enhance non-ideal
gas-dynamic effects). Suitable parameters are the compressibility factor 𝑍 and the fundamental
derivative of gas dynamics 𝛤. The former quantifies the deviation of the fluid’s thermodynamics
from that one of an ideal gas, and it is defined as
𝑃
𝑍= , (3.3)
𝜌𝑅𝑇
while the latter specifies the variation of the speed of sound along isentropic processes (Thompson,
1971), and its low values, entailed by high fluid molecular complexity, enhances non-ideal effects
with quantitative (and possibly qualitative) impact on the expansion flow field (Romei et al.,
2020). According to Masi et al. (2020), the effects of fluid-state-related parameters on maximum
attainable turbine efficiency turned to be negligible only when very low values of Vr are dealt with.
49
Figure 3.3: 𝑇 − 𝑠 diagram of R1233zd with contour lines for compressibility factor Z. Expansion
lines refers to the turbines X,Y, K, and W. Continuous lines: 𝑆𝑃 = 0.011 m and 𝑉𝑟 = 14; dashed
lines: 𝑆𝑃 = 0.009 m and 𝑉𝑟 = 35.
Consequently, their impact is investigated in the following sections since the thermodynamics of
the ORC applications considered in this thesis involve also expansion processes characterized by
high Vr, and strong non-ideal gas effects are expected to occur. To this end, the impact of real gas
dynamics on the optimal RIT efficiency for fixed SP and Vr is firstly assessed, referring to the
values of 𝑍 at the start and end of the expansion as indicators of the level of non-ideal gas-dynamics
associated with each optimal design chosen for comparison. Subsequently, the effects of 𝛤 at fixed
𝑍 on the RIT maximum efficiency are assessed. This is done by fixing a set of SP, Vr, and 𝑍 and
by concurrently referring to different working fluids, as done by Da Lio et al. (2016) for axial flow
turbines.
3.4.1 Sensitivity analysis of optimum RIT efficiency: effects of real-gas compressibility

In order to check the consistency of adopting SP and Vr as independent parameters for the turbine
analyses, four RITs working with R1233zd(E) and having the same SP (0.011 m) and Vr (14) but
characterized by different expansion inlet and outlet conditions and, consequently, by different
inlet and outlet compressibility factors are optimized adopting RITML code and the optimization
strategy described in Section 3.2.2. The corresponding expansion processes considered to infer
the boundary conditions for the four optimizations are reported in the 𝑇 − 𝑠 diagram in Figure 3.3
(continuous lines) together with the vapor saturation curve of R1233zd(E). Turbine X expands from
saturated conditions, turbine Y is representative of a superheated subcritical cycle while turbines
K and W expand from supercritical region, leading to inlet compressibility factor values ranging
from 0.41 to 0.82. The outcomes of the four optimization procedures are summarized in Table 3.5.
It is possible to note that not only the non-dimensional parameters but also the physical geometric
dimensions of the four turbines are rather similar, all of them being characterized by the same SP.
Although the rotational speed is quite different, the corresponding specific speed 𝜔 𝑠 is similar for
all the cases. The specific speed is defined as
√︁
𝑄¤ 𝑜𝑢𝑡 𝑆𝑃
𝜔 𝑠 = 𝜔 0.75 = 𝜔 √ . (3.4)
𝛥ℎ𝑖𝑠 𝛥ℎ𝑖𝑠
Moreover, the four turbines exhibit very similar efficiencies that lie in an interval ranging between
±0.65% with respect to the average value of 90.35%. This deviation is within the range of accuracy
50
3.4. Assessment of SP and Vr as unique independent parameters
Table 3.5: Analysis over four RITs with the same SP = 0.011 m and Vr = 14, but characterized
by different expansion inlet and outlet conditions and compressibility factors. 𝑇𝑟 and 𝑃𝑟 refer to
reduced temperature and pressure respectively.
Parameter Turbine X Turbine Y Turbine K Turbine W

𝑇𝑖𝑛 - 𝑃𝑖𝑛 , °C - bar 130.3, 19.2 205.9, 22.8 213.4, 45 192.5, 50
𝑇𝑖𝑛,𝑟 - 𝑃𝑖𝑛,𝑟 0.92, 0.54 1.09, 0.64 1.26, 1.11 1.40, 1.06
𝑇𝑜𝑢𝑡 - 𝑃𝑜𝑢𝑡 , °C - bar 46.6, 1.6 121.9, 1.6 116.7, 3.8 84.3, 5.8
𝑚 𝑂𝑅𝐶 , g/s 219.9 205.1 471.3 681.9
𝑍𝑖𝑛 - 𝑍 𝑜𝑢𝑡 0.66, 0.95 0.82, 0.98 0.63, 0.94 0.41, 0.87
𝜂𝑡𝑢𝑟 𝑏 , % 90.1 89.7 90.5 91
Nozzle
𝐷 𝑖𝑛 - 𝐷 𝑜𝑢𝑡 , mm 44.4, 34.4 45.2, 35.2 44.3, 34.3 44.3, 34.3
ℎ𝑖𝑛 -ℎ𝑜𝑢𝑡 , mm 2.6, 2.6 2.7, 2.7 2.7, 2.7 2.9, 2.9
Rotor
𝜔 𝑠 - 𝜔, \- kRPM 0.97, 96.8 0.94, 112 0.97, 105 0.97, 85
𝐷 𝑖𝑛 - 𝐷 𝑜𝑢𝑡 , mm 33.4, 22.2 34.2, 22.6 33.3, 22.1 33.3, 22.2
𝑏 𝑖𝑛 - 𝑏 𝑜𝑢𝑡 , mm 2.5, 8.5 2.6, 8.8 2.6, 8.5 2.8, 8.5
featuring the adopted mean line method. The effects related to real-gas compressibility have been
investigated also for a pair of SP and Vr, namely 𝑆𝑃 = 0.009 m and 𝑉𝑟 = 35, leading to large
blade loading (isentropic work), smaller machines and, consequently, to more severe optimization
conditions. The four expansion processes considered start from the same thermodynamic states
as those already described (turbines X, Y, K, and W) and are represented in Figure 3.3 through
dashed lines. Also in this case, the non-dimensional parameters (including the specific speed)
and the physical geometric dimensions of the four turbines turn out to be rather similar. On the
other hand, the achieved optimum efficiencies lie in a larger interval with respect to the previous
case, ranging between ±1.2% with respect to the average value of 86.55%. This larger deviation
is associated with the specific choice of quite low SP and large Vr, which lead the optimization
procedure in an highly-constrained region of the design space, characterized by very high relative
and absolute Mach numbers, large blade loading, and very small machine dimensions. To find
allowable solutions, the optimization algorithm is forced to push some of the design variables at
their lower/upper boundaries. Being in a region of the design space where the constraints are more
apt to be violated may prevent the optimization process to find the actual maximum efficiency,
leading to slightly sub-optimal solutions that however meet all the imposed constraints. This
explains the larger scatter in the efficiency values for 𝑆𝑃 = 0.009 m and 𝑉𝑟 = 35 with respect to
the one observable in Table 3.5. Such larger scatter is however limited to ±1.2%, a value still
comparable with the range of accuracy featuring the adopted mean line method. The analysis just
described suggests negligible effects of the real-gas compressibility for fixed values of SP and Vr,
even if a larger impact is observed for thermodynamic boundary conditions leading to more severe
expansion ratios and smaller machine dimensions.
3.4.2 Sensitivity analysis of optimum RIT efficiency: effects of 𝛤 at fixed 𝑍

The effectiveness of SP and Vr as independent parameters was also tested considering a set of
working fluids. Several turbines were in fact optimized adopting RITML code and imposing the
51
Table 3.6: Analysis over various RITs with the same SP (0.008 m), Vr (50), and Z (0.63) but
characterized by different operating fluids. Results are obtained optimizing the turbines with
RITML. 𝑇𝑟 and 𝑃𝑟 refer to reduced temperature and pressure respectively.
Parameter R1233zd R134a Cyclopentane MM

𝑇𝑖𝑛 - 𝑝 𝑖𝑛 , °C - bar 255.2, 70 180.2, 85 280.3, 49 286.3, 22
𝑇𝑖𝑛,𝑟 - 𝑃𝑖𝑛,𝑟 1.2, 1.93 1.21, 2.09 1.08, 1.07 1.08, 1.13
𝑇𝑜𝑢𝑡 ,𝑖𝑠 - 𝑝 𝑜𝑢𝑡 , °C - bar 119.7, 1.6 31.2, 1.7 162, 1.2 229.4, 0.62
𝑚 𝑂𝑅𝐶 , g/s 128.2 144.9 66 45.6
𝑍𝑖𝑛 - 𝑍 𝑜𝑢𝑡 0.63, 0.98 0.63, 0.97 0.63, 0.98 0.63, 0.99
𝜂𝑡𝑢𝑟 𝑏 , % 77.3 76.9 77.4 78
Nozzle
𝐷 𝑖𝑛 - 𝐷 𝑜𝑢𝑡 , mm 36.1, 25.1 39.8, 26.6 36.1, 25.1 34.3, 24.3
ℎ𝑖𝑛 -ℎ𝑜𝑢𝑡 , mm 2.1, 2.1 2.1, 2.1 2.1, 2.1 2.1, 2.1
Rotor
𝜔 𝑠 - 𝜔, - kRPM 0.99, 189 0.96, 187 0.98, 268 1, 185
𝐷 𝑖𝑛 - 𝐷 𝑜𝑢𝑡 , mm 23.8, 16.9 24.5, 17.3 23.9, 17 23.3, 16.6
𝑏 𝑖𝑛 - 𝑏 𝑜𝑢𝑡 , mm 2, 4.3 2, 4.6 2, 4.4 2.1, 4.1
same values of size parameter, volume ratio, and compressibility factor but considering different
working fluids. This means that, in the framework of ORC turbines’ similarity, the differences
resulting from this analysis in RIT optimum design and efficiency are due to the effects of variations
in the value of the fundamental derivative, which is associated with the selection of different
working fluids for turbine operation at fixed compressibility factor. Particularly, two refrigerants
(R1233zd and R134a), a hydrocarbon (cyclopentane), and a siloxane (MM) were selected to span
wide ranges of molecular complexity and weight. The optimizations were carried for a fixed inlet
compressibility factor 𝑍 of 0.63 and for 3 different combinations of SP and Vr (namely SP = 0.007
m and Vr = 5, SP = 0.008 m and Vr = 50, SP = 0.01 and VR = 50). The results of the optimization
procedures for the most critical condition of SP = 0.008 m and Vr = 50 are summarized in Table 3.6.
Although operated with different fluids and characterized by large Vr and small SP, the optimized
turbines are characterized by similar non-dimensional parameters and geometric dimensions, as
well as by comparable values of the specific speed 𝜔 𝑠 . Moreover, the turbines exhibit very similar
efficiencies that lie in an interval ranging between ±0.55% with respect to the average value of
77.45%. The deviation in terms of efficiency is even lower for the other combinations of SP and Vr
analyzed, namely ±0.25% for SP = 0.007 m and Vr = 5 and ±0.5% for SP = 0.01 and VR = 50. As
pointed out in Chapter 2, these deviations are within the range of accuracy featuring the adopted
mean line method, confirming the outcomes by Masi et al. (2020), according to which different 𝛤
values at fixed 𝑍 play a minor role on the optimum RIT efficiency at fixed SP-Vr pairs.
In light of the outcomes just described in the last two paragraph, and considering the Reynolds
number self-similarity, the global sizing problem for the optimized ORC turbines adopted in
following analyses is judged to be quite accurately formulated in terms of VR and SP only. This
allows to identifying a 2D ‘explored space’ of configurations (see Section 3.5) and to develop a 2D
perfomance map, which provides the value of maximum RIT efficiency in function of SP and Vr.
52
3.5. RIT maximum efficiency map definition
0.9
Constant maximum efficiency 90% 3
0.89
80% 2
3
Turbine e ciency
0.88
70%
0.87
2
60%
0.86 1
*
1 **
** 0.85
50%
Extrapolation possible
0.84
0.7 0.8 0.9 1.0 1.1 1.2
Stage speci c speed, s
Figure 3.4: (a) turbine maximum efficiency as function of SP and Vr, markers are the turbines
optimized with RITML mean line code. (b) variation of turbine efficiency for markers 1, 2, and 3
as function of turbine 𝜔 𝑠 .
3.5 RIT maximum efficiency map definition

Single-stage radial inflow turbine maximum efficiency map is defined by optimizing 22 different
turbines (markers in Figure 3.4.a) covering most of the region of the investigated SP and Vr values,
obtained through the preliminary analysis described in Section 3.3. The 22 optimized turbines, all
using R1233zd(E) as working fluid, represent the support grid over which the performance map
is retrieved through regression. The resulting interpolating relation 𝜂 𝑚𝑎𝑥 = 𝜂 𝑚𝑎𝑥 (𝑆𝑃, 𝑉𝑟) (see
Equation 3.5) therefore allows to calculate the maximum efficiency in any point within, and in
proximity, of the selected domain (colored point in Figure 3.4.a) . Large SP and small Vr values are
not interesting for the applications considered since SP > 0.012 m have been scarcely encountered
during the design of several cycles described in Section 3.3, while for Vr < 5 the ORC plant
efficiency becomes very small (see Figure 3.2). On the contrary, the lower SP bound (0.007 m <
SP < 0.0075 m) is imposed by the constraints included in the optimization procedure conerning the
RIT minimum physical dimensions: SP values below the bound indicated in Figure 3.4.a prevent to
obtain feasible RIT geometries fulfilling all the activated constraints. Similarly, for Vr values larger
than 50 feasible solutions of the optimizations are prevented by the establishment of supersonic
flows in the rotor relative reference frame, feature that considerably reduce machine efficiency
and operability. To provide an overview, Figures 3.4.b and 3.5 report some details concerning
the optimization of the 22 RITs. Figure 3.4.b depicts the maximum attainable turbine efficiency
against the specific speed for the three designs enumerated in Figure 3.4.a. For turbines 1 and
2, the efficiency curve is fairly flat, indicating that the specific speed 𝜔 𝑠 (and consequently the
rotational speed 𝜔) can be selected over a quite wide range still obtaining efficiency values close to
the maximum. On the contrary, for turbine 3 (more critical in terms of Vr) the efficiency variation
with 𝜔 𝑠 is steeper, highlighting a lower flexibility in terms of rotor speed required to get high RIT
performance. Additionally, Figure 3.5 reports the loss breakdown for six test case optimized
turbines, those represented by grey circles in Figure 3.4.a, whose choice allows to isolate and
investigate separately the effects of Vr and SP on the variation of the different loss contributions.
The following observations can be reported:
53
SP=0.009 m Vr=14
20% 20%
Cumula ve enthalpy loss
Kine c Energy
Cumula ve enthalpy loss

15% 15% Windage
Rotor Clearance
10% 10% Passage + Trailing Edge
Mixing
Stator
5% 5% Passage + Trailing Edge
0% 0%
5 14 30 50 0.012 0.009 0.007
Vr SP, m
Figure 3.5: Loss breakdown for (left) turbines with the same SP = 0.009m and different Vr =[5-50]
and (right) turbines with same Vr = 14 and different SP = [0.007-0.012m].
• Turbine SP and Vr strongly affect turbine maximum efficiency which cannot be assumed
constant in the whole design space. Calculated values range between 67.1% and 91.9% but
lower efficiency is expected for Vr higher than 50 and SP lower than 0.007 m;
• Rotational speed has a large impact on turbine performance and must be carefully optimized.
On the contrary, although depending on SP and Vr, the optimal specific speed ranges from
0.9 to 1.04 for most of the optimized turbines, with the exception of 3 of them for which
0.8 < 𝜔 𝑠 < 0.9. This analysis confirms that a good first attempt value of 𝜔 𝑠 is around 1
(𝜔 = 𝜔𝑜 𝑝𝑡 ), as also suggested by Perdichizzi & Lozza (1987);
• An increment of Vr leads to an efficiency penalization mainly due to an increase of stator

passage loss contribution. In fact, higher Vr values lead to lower pressure levels at the
nozzle outlet, which in turn imply larger Mach number values and, consequently, higher
losses associated with blade boundary layers development. Performance penalization is
more marked at smaller SP;
• A reduction of SP produces a pronounced increase of the nozzle passage loss contribution.

A possible explanation of this effect is likely to be ascribed to the geometry dimensions
reduction that results decreasing the SP value. In fact, the turbine size shrinking is limited by
the constraints previously mentioned concerning the minimum values of main geometrical
parameters. Therefore, to reach the desired pressure level at the nozzle outlet, mainly affected
by the expansion ratio Vr, the optimizer is forced to rise the nozzle outlet angle value, leading
an increase of the flow wetted surface and in turn of the friction loss contribution. The
increment of clearance losses for lower SP values is to be ascribed to the increase of axial and
radial gap relative size occurring when the geometry is shrunk, a classical example of the so
called size effect. Moreover, the constraint concerning the maximum flow deviation through
the rotor leads to larger absolute flow angles at the rotor outlet, explaining the increase of
kinetic energy loss contribution when SP is reduced;
• To develop the map, the rotational speed has not been limited. As result, values of rotational
speed around 200000 RPM have been found for turbine labelled with * in Figure 3.4.a, a
strong efficiency penalization is expected for machines in this region if maximum shaft
rotational speed is constrained to lower values because of bearings, lubrication and frequency
conversion issues. In fact a reduction of 𝜔 coupled with large absolute Mach numbers and
flow angles at the nozzle outlet would imply larger relative Mach numbers at the rotor inlet,
with a subsequent increase of losses due to both profile and shock losses.
54
3.6. ORC comprehensive design
Table 3.7: Coefficients of interpolating relation for the turbine maximum efficiency map.
Coeff. Value Coeff. Value Coeff. Value

𝐴00 9103.26 𝐴12 28.10 𝐴31 -60.21
𝐴01 -1512.62 𝐴20 3962.80 𝐴32 0.4798
𝐴02 41.35 𝐴21 -421.09 𝐴40 85.22
𝐴10 9510.39 𝐴22 6.36 𝐴41 -3.22
𝐴11 -1305.07 𝐴30 823.11 𝐴50 3.52
Computed maximum efficiency values have been interpolated with a best fit correlation having a
corrected R2 equal to 0.993 and average absolute deviation equal to 0.38%, the difference being
above 1% only for the 2 turbines labelled with ** in Figure 3.4.a. The interpolating relation is
reported below and the corresponding coefficients 𝐴𝑖 𝑗 are listed in Table 3.7.
5 ∑︁
∑︁ 2
𝜂𝑡𝑢𝑟 𝑏 = 𝐴𝑖 𝑗 ln𝑖 (𝑆𝑃) ln 𝑗 (𝑉𝑟) with 𝑖 + 𝑗 ⩽ 5 (3.5)
𝑖=0 𝑗=0
Extrapolation of Equation 3.5 for Vr > 50 and SP < 7 mm, though possible, is not recommended
because it is not supported by any experimental or numerical evidence and can likely result in
turbine performance overestimation. On the contrary, for Vr < 5 and SP above 12 mm the efficiency
can be kept equal to the efficiency value obtained at the closer point on the map boundary, which
however would result in a slight underestimation of turbine performance according to the very flat
trend of the performance map for small Vr and large SP. It is worth to highlight that the value of
efficiency obtained through the map is the upper bound achievable from numerical optimization
based on mean-line model, which do not consider some possible critical aspects such as detailed
shock patterns within the channels, leakage flows, and loss contributions interaction. Therefore,
any turbine design must be confirmed with detailed 3D high-fidelity CFD analyses. In addition, the
calculated efficiencies strongly depend (especially for small SP and large Vr) on the investigated
design variable ranges and constraints adopted. Lower efficiency values can be obtained increasing
the lower bound of relevant geometrical features, such as trailing edge thicknesses and clearances,
investigating a narrower design space, or limiting rotational speed maximum value because of
limits of generator and bearings capabilities. Finally, the optimizations carried out to develop the
RIT performance map highlighted that single-stage RIT cannot be designed for SP < 0.007 m and
Vr > 50 without violating the constraints concerning minimum turbine dimensions, blade loading,
maximum Mach numbers at the nozzle outlet, and subsonic flows within the rotor.
3.6 ORC comprehensive design

Once defined, the turbine performance map was implemented in the ORC design tool to assess
the effects of a realistic expander efficiency values during the cycle optimization process. For
this purpose, two analyses were carried out: the first one focuses on R1233zd(E) and aims at
investigating how the turbine performance map affect the optimal values of the cycle design variable,
while the second one extends the analysis to different working fluids.
55
Case Turbine Map 𝜂𝑡𝑢𝑟 𝑏 = 0.8 Turbine Map (-33%) 𝜂𝑡𝑢𝑟 𝑏 = 0.8 (-33%)
𝑇𝑐𝑜𝑛𝑑 ,°C 35.1 37.7 35.1 37.7
𝑃𝑚𝑎𝑥 , bar 50.6 59.4 29 59.7
𝑇𝑚𝑎𝑥 , °C 240 240 189.3 240
𝜂 𝑠𝑦𝑠 , % 16.2 15 13.7 15
𝑊𝑒𝑙 , kW 8.7 8.1 5.0 5.4
𝑚¤ 𝑂𝑅𝐶 , g/s 163.1 170.8 123.0 114.6
𝑆𝑃, mm 8.5 8.3 7.6 6.8
𝑉𝑟 28.7 33 16.2 33.3
𝜂𝑡𝑢𝑟 𝑏 , % 85.7 80 83.6 80
Table 3.8: Optimal results for the non-recuperative cycle both considering a constant turbine
efficiency and adopting the turbine performance map. Results for both normal and reduced size
(-33%) engine.
3.6.1 Performance investigation with working fluid R1233zd(E)

Figure 3.6b depicts the maximum cycle efficiency map retrieved for the non-recuperative cycle
with R1233zd and by adopting the turbine efficiency correlation calibrated in previous section.
This map reports the cycle efficiency as a function of condensation temperature and maximum
pressure always at the optimal cycle maximum temperature, and it is to be compared against
the map of Figure 3.2.a obtained for a constant turbine efficiency equal to 80%, for simplicity
reported in Figure 3.6a. This comparison allows indeed to understand the effects on cycle design
and optimization of computing realistic and non-constant values of turbine efficiency, rather than
assuming a fixed value regardless of the cycle thermodynamics. It also helps to catch the trade-off
between optimizing cycle theoretical performance (Figure 3.6a) and the need of maximizing
turbine performance. Calculated turbine efficiency is reported in Figure 3.6c as a function of
condensation temperature and cycle maximum pressure. As evidenced by Figure 3.6c, adopting
high condensation temperature and high maximum pressures would strongly penalize the turbine
efficiency due to SP reduction. On the contrary, the lower the maximum pressure and the condensing
temperature are the higher the turbine performance is. This behavior of the expander reflects on
cycle efficiency, which tends to decrease for high values of 𝑝 𝑚𝑎𝑥 and 𝑇𝑐𝑜𝑛𝑑 when the turbine map
is integrated in the design process, as can be inferred by Figure 3.6b. Also the optimal design
obtained using the turbine efficiency map (green dot in Figure 3.6b) reflects these considerations
and selects an optimal turbine inlet pressure (𝑃𝑚𝑎𝑥 = 50.59 bar) and a condensation temperature
(𝑇𝑐𝑜𝑛𝑑 = 35°C) lower than those obtained for the 80% turbine efficiency case (CASE B in Table
3.4, with 𝑃𝑚𝑎𝑥 = 60.2 bar and 𝑇𝑐𝑜𝑛𝑑 = 37.7°C, and red dot in Figures 3.6a and 3.6b) in order to
improve turbine performance thanks to lower Vr (28.76) and a larger SP (8.54 mm). The difference
between the cycle performance calculated implementing turbine efficiency correlation and the case
with fixed turbine efficiency equal to 80% (refer to Figures 3.6a and 3.6b) is reported in Figure 3.6d.
It is interesting to note that in most of the design space the difference is limited to around 1.5%-2%
points of efficiency that corresponds to a relative difference of around 10%-15%. Although only
non-recuperative cycles are considered for sake of brevity and clarity in the analysis just described
and in the following ones, similar observations are valid also for the recuperative cycles.
A final analysis concerning the working fluid R1233zd(E) is reported in Table 3.8 considering
also an engine of a smaller size by simply reducing the exhaust mass flow rate down to 0.134 kg/s
(-33%), still considering non-recuperative cycle configuration. Particularly, the optimal cycles with
fixed turbine efficiency of 80% are compared to the optimal configurations obtained implementing
56
Figure 3.6: Comprehensive analysis results for the non-recuperative case as function of 𝑃𝑚𝑎𝑥 and
𝑇𝑐𝑜𝑛𝑑 at optimized turbine inlet temperature. (a) cycle efficiency for a constant turbine efficiency
equal to 80%, (b) cycle efficiency with integrated turbine map, (c) turbine efficiency from integrated
map, (d) difference between optimal cycle efficiency adopting turbine performance map vs constant
turbine efficiency (b Vs a). Red marker is the optimal design point with 80% constant turbine
efficiency while green one corresponds to the ORC optimization with turbine map implemented.
In this case the turbine efficiency in the optimal point result equal to 83.6 %.
the turbine efficiency map, at both nominal (0.2 kg/s) and reduced (0.134 kg/s) exhaust mass flow
rates. As can be noted in Table 3.8, when a fixed turbine efficiency is considered the optimum cycle
DVs and efficiency are not affected by the engine exhaust mass flow rate, which only impacts the
organic fluid mass flow rate 𝑚¤ 𝑂𝑅𝐶 and related properties (e.g. turbine power, and SP). Concerning
the comparison between implementing or not the turbine map in the reduced-size case, Table 3.8
highlights a huge variation of optimum DVs, especially in terms of maximum pressure 𝑃𝑚𝑎𝑥 and
temperature 𝑇𝑚𝑎𝑥 . In fact, if the design point is kept the same as that obtained with the fixed
turbine efficiency assumption, the lower organic fluid mass flow rate would result in a lower SP and
consequently in a very low realistic turbine efficiency value. In this case the optimization algorithm
reduces the maximum pressure in order to reduce the turbine enthalpy drop, increasing the turbine
SP by concurrently reducing the Vr. This allows to increase the attainable turbine efficiency and,
consequently, lead to an overall increase of cycle performance. Optimal cycle for this reduced size
57
Figure 3.7: 𝑇 − 𝑠 diagram for 4 optimal non recuperative cycles: with fixed turbine 80% and
adopting the turbine map with both nominal (0.2 kg/s) and reduced (0.134 kg/s) exhaust mass flow
rate. Red and grey lines represent the thermodynamic process in the four cases for the heat source
and the cooling air respectively.
case when implementing the turbine map becomes subcritical (29.5 bar) and superheated with an
optimal turbine inlet temperature considerably lower than the maximum value (189.3°C).
The outcomes reported in Table 3.8 can be visualized also in Figure 3.7, which outlines
the temperature-specific entropy (𝑇 − 𝑠) diagrams for the four optimal non recuperative cycles
investigated. Figure 3.7 proves the negligible effects of exhaust mass flow rate when a fixed turbine
efficiency is considered and concurrently highlights the impact of the adoption of realistic and
non-constant turbine efficiency values on optimal cycle layout, especially when the primary plant
(in this case the engine) size is reduced. These results show the importance of adopting a turbine
efficiency map in the design of small scale ORC based on single-stage turbine architecture. In fact,
without a proper link between turbine efficiency and cycle parameters (specifically SP and Vr) the
optimization leads to result based on a high-pressure ratio supercritical cycle independently of
the size of the plant. In case of nominal flue gas mass flow rate this result is rather close to the
real optimum obtained implementing the turbine efficiency map while in case of reduced mass
flow rate (-33%) this is dramatically misleading and would lead to the adoption of a cycle with an
excessive pressure ratio that in turn would cause a non feasible design of the turbine.
3.6.2 Potential fluid screening - cycle performance assessment

As final analysis, the thermodynamic cycle was optimized by means of the ORC design tool
implemented with the turbine efficiency map for a set of potential working fluids. In particular,
the list of fluids chosen includes 5 hydrocarbons (n-Pentane, cyclopentane, hexane, cyclohexane,
Toluene), 4 refrigerants (R134a, R1233zd, R1234yf, R245fa), and 2 siloxanes (MM,MDM).
The set aims at including most of state-of-the-art ORC working fluids, still bewaring of current
environmental and safety regulations, and concurrently at investigating wide ranges of compound
molecular complexity, molar mass, and critical point conditions.
The non-recuperative cycle configuration was optimized for each working fluid mentioned
above, exploiting as design variables the three parameters reported in Section 3.3, namely 𝑃𝑚𝑎𝑥 ,
𝑇𝑐𝑜𝑛𝑑 , and 𝑇𝑚𝑎𝑥 . Most relevant constraints applied to the optimization process concern the
58
Table 3.9: Optimal plant efficiencies for the whole set of potential working fluids and for the four
cases investigated, sorted by decreasing system efficiency.
CASE R CASE S CASE M CASE L

Constant 𝜂𝑡𝑢𝑟 𝑏 𝑚¤ 𝑒𝑔 = 0.1 kg/s 𝑚¤ 𝑒𝑔 = 0.2 kg/s 𝑚¤ 𝑒𝑔 = 0.4 kg/s
Fluid 𝜂 𝑠𝑦𝑠 , % Fluid 𝜂 𝑠𝑦𝑠 , % Fluid 𝜂 𝑠𝑦𝑠 , % Fluid 𝜂 𝑠𝑦𝑠 , %
Cyclopen 20.15 Toluene 20.61 Cyclopen 20.96 Cyclopen 21.43
Toluene 19.44 Cyclohex 20.17 Toluene 20.63 R1233zd 20.73
Cyclohex 19.04 Cyclopen 17.86 Cyclohex 20.18 Toluene 20.61
Pentane 17.01 Hexane 17.45 Hexane 17.84 Cyclohex 20.19
Hexane 16.79 MM 14.73 Pentane 16.67 Pentane 18.08
R1233zd 16.22 Pentane 13.29 R1233zd 16.17 Hexane 17.84
R245fa 16.15 R1233zd 11.69 MM 14.65 R245fa 16.90
R134a 14.23 MDM 11.51 R245fa 14.05 MM 14.72
MM 13.85 R245fa 9.90 MDM 11.47 MDM 11.47
R1234yf 11.37 R1234yf 1.31 R1234yf 6.09 R134a 10.97
MDM 10.81 R134a 0.79 R134a 5.68 R1234yf 10.13
avoidance of two-phase flow through the expander, a maximum Vr equal to 60, to avoid unfeasible
single-stage turbine geometries and excessively large nozzle outlet Mach numbers, and minimum
condensing pressure equal to 0.1 bar to avoid inward leakage issues. The optimizations were
carried out implementing the turbine efficiency map for three different exhaust mass flow rates in
order to simulate a variation of the engine (primary power plant) size: CASE S, M and L, which
adopts an exhaust gas mass flow rate equal to 𝑚¤ 𝑒𝑔 = 0.1 kg/s, 𝑚¤ 𝑒𝑔 = 0.2 kg/s, and 𝑚¤ 𝑒𝑔 = 0.4 kg/s
respectively. The optimizations were carried out also for and additional case (referred to as CASE
R), in which a constant turbine efficiency 𝜂𝑡𝑢𝑟 𝑏 = 80% is considered. This condition represents the
reference case since, as proved in Table 3.8 and in Figure 3.7, the optimized DVs are independent
of the specific engine exhaust mass flow rate in the case of constant turbine efficiency.
The results of this analysis are summarized in Table 3.9, in which the plant efficiency values
are sorted for the 4 cases investigated, and in Figure 3.8, which depicts the optimal thermodynamic
cycles for cyclopentane and R1234yf for cases S, M and L. Comparing CASE S, M and L it is
possible to appreciate the impact of using turbine performance map on the cycle definition and
performance, depending on the working fluid and plant size, which turns out to play a huge impact
on optimal cycle and working fluid selction. Fluids characterized by high molecular complexity
and high molar mass (cyclopentane, hexane, toluene) can attain good turbine efficiency also for
small scale systems thanks to the combined effect of low condensing pressures and low specific
enthalpy drop, which result in high SP values. Increasing the size for these fluids is beneficial
but the gain in performance is rather limited. However, as evidenced in Figure 3.8.a, the cycle
configuration for cyclopentane changes varying the engine size. In particular, for CASE M and L
the cycle adopts very high evaporation pressure close to the critical one, saturating the constraint
on maximum turbine volume ratio (60), and the cycle architecture and turbine efficiency (about
89%) result rather independent on the size. On the contrary, for CASE S the cycle adopts a lower
evaporation and maximum temperatures thus leading to lower volume ratio (39.6) and enthalpy
drop in order to limit the reduction of SP and get good turbine performance also in this case of
smaller engine size. For fluids characterized by even higher molecular complexity and molar
mass (MM and MDM), cycle efficiency is limited by the maximum volume ratio achievable with
single-stage machines, which is limited to 60 in this analysis. These fluids typically have low
specific enthalpy drops and large outlet volumetric flow rates, resulting in very high SP values.
However, an unconstrained cycle optimization would lead to extremely high volume ratios that are
59
(a) (b)
Figure 3.8: 𝑇 −𝑠 diagram of optimal non-recuperative cycles implementing the turbine performance
map with 3 different exhaust mass flow rate values Working fluids are: (a) Cyclopentane, (b)
R1234yf.
not compatible with single-stage radial-inflow turbines. Thus, turbine performance does not limit
the cycle’s optimum efficiency for such complex and high-molar mass fluids. Rather, the resulting
optimum cycle configuration exhibits low efficiency when the optimization is subject to constraints.
These effects are much more pronounced for low molecular complexity and low molecular
mass fluids that are characterized by higher condensation pressure and thus by lower volumetric
flow rates, leading to low values of SP also for larger exhaust mass flow rates and, consequently, to
miniaturized machines that suffer of low efficiencies. Considering R1234yf (Figure 3.8.b), the
optimal cycle changes dramatically for the different cases. For CASE L the high working fluid
mass flow rate allows to have reasonable SP also in case of supercritical cycles. On the contrary,
reducing the engine size the optimal cycle moves towards subcritical configurations, becoming
a nearly saturated cycle at very low evaporation temperature for CASE S. In the latter case the
optimization algorithm pushes the condensing temperature close to the minimum one in order to
increase the volumetric flow rate at turbine outlet, and reduces both the maximum temperature and
pressure in order to reduce the turbine enthalpy drop leading to reasonable SP. The final cycle in
CASE S has an extremely low efficiency (about 1%), due to the combination of very small pressure
ratios and a penalized turbine efficiency (69.9%). However, this result is compliant with the use of
a single stage radial inflow turbine and it is remarkably different from the result attainable in CASE
R (fixed turbine efficiency) that adopts a supercritical cycle with a combination of very low SP (4.5
mm) and moderate Vr (10.6), which are non-compatible with the selected turbine architecture and
that in turn would result in an unfeasible turbine design. A final consideration can be highlighted
for R1233zd(e), the fluid already selected as a promising candidate for this application. If no map
of turbine efficiency is adopted (CASE R) R1233zd(e) is certainly not the most promising fluid
ranking in 6th position with an efficiency gap of around 4 points with respect to the best fluid
(16.22% vs 20.15%). This positioning however is not representative of the potential this fluid; in
fact, if turbine efficiency dependency on cycle parameters is accounted for, the results are radically
different. In CASE S, the small engine size produces a penalization of turbine efficiency which in
turn leads to a subcritical cycle with strongly penalized efficiency (11.7%) which is 9 points of
efficiency lower than the one associated with the best fluid. Increasing the engine size (CASE M
and CASE L) the performance of R1233zd increases remarkably, approaching the the best fluid
(cyclopentane) efficiency in the larger size case (20.73% vs 21.34%) making it a very attractive
option considering that it is non-flammable, differently from hydrocarbons.
60

In the present chapter the methodology adopted to integrate the evaluation of turbine performance
within the ORC thermodynamic design and optimization is thoroughly described. In particular
a single-stage RIT maximum performance map is used, through which the expander efficiency
may be computed in function of two cycle-dependent parameters: the size parameter SP and the
volume ratio Vr. Main outcome of this chapter is the demonstration of the importance of adopting
reliable non-constant efficiency values for turbomachinery especially in those applications, like on
board waste heat recovery, where miniaturized single stage radial inflow turbines are adopted as a
consequence of the small power levels involved and where machine number of stages cannot be
varied to keep efficiency stable independently of the volume ratio.
Firstly, the numerical routines employed are described, and preliminary investigations are
carried out to retrieve the SP-Vr domain within which develop the turbine performance map. This
analysis revealed that the ranges of SP and Vr associated with WHR from internal combustion
engines of long-haul trucks are significantly different from those investigated in literature, being
characterized by very small SP (up to 7 mm) and quite large Vr (up to 60). The validity of SP and Vr
as unique independent parameters affecting the optimum efficiency of ORC RITs is then assessed,
reducing to some extent the cautiousness suggested in the use of this simplified approach by the
results reported in Masi et al. (2020) for axial-flow turbines case. In particular, since the dynamic
self-similarity can be assumed due to the relatively high Reynolds numbers usually associated with
ORC turbines (> 107 ), the effects related to non-ideal thermodynamic states and to the usage of
different working fluids are investigated. The variation of maximum RIT efficiency for the same
SP and Vr but different inlet compressibility factor values and different operating fluids turns out
to be limited below the range of accuracy associated with the reduced-order model employed to
carry out the optimizations. However, attention is needed when an highly constrained region of
the design space is investigated (e.g. low SP and large Vr) since the activation of the constraints
may prevent the optimization algorithm to find the actual maximum efficiency. The amount and
number of violated constraints may however vary as a function of either the inlet compressibility
factor or the operating fluid, possibly leading to a non-negligible maximum efficiency variation in
highly-constrained region of the SP-Vr domain, making the 2D performance map a less general
tool in those severe conditions. However, these analyses demonstrated that a maximum efficiency
map can be used with sufficient accuracy (±1.2% in the worst case scenario, as demonstrated in
Section 3.4, plus the intrinsic prediction uncertainty of the mean-line model), independently of the
specific values of pressure ratio, inlet compressibility factor, and adopted working fluid.
The turbine performance map, as well as the outcomes just described, have been obtained
thanks to an extensive numerical activity on turbine geometry optimization for a large range of
SP and Vr. The fitted equation can be thus adopted as a very general fluid-independent turbine
performance map representing the maximum efficiency attainable with an optimization of both
machine geometry and rotational speed for single stage radial inflow turbine architecture. Turbine
performance reduces for increasing Vr and reducing SP values. The former efficiency drop is due
to the presence of highly supersonic flows at the nozzle outlet and the need of adopting very large
blade span variation within the rotor. On the contrary, reduced SP values lead to small turbine
sizes and to the consequent increase of loss contributions related to those geometrical features
that cannot be freely scaled down, such as clearances and trailing edge thicknesses. Additionally,
turbine shrinking to comply with a SP reduction cannot be always achieved due to the physical
constraints considered to ensure the mechanical feasibility of the final proposed geometry. In this
case the optimizer is forced to change nozzle outlet angle and rotor deviation to reach the desired
volume ratio, leading to an increase of passage losses in both the components. Additionally, the
optimizations performed pointed out strong difficulties in designing a feasible single-stage RIT
61
for Vr over 60 and SP below 7 mm due to the limit of subsonic relative flow at the rotor inlet, the
increase of profile and secondary losses, and the miniaturized physical dimensions of the machine.
Concerning the comprehensive ORC thermodynamic design, the implementation of a map for
expected turbine performance allows to guide the optimization algorithm towards solution which
are more representative of the correct optimal design of the system. The results show that the
adoption of a fixed turbine efficiency can be very misleading in terms of optimal cycle definition and
selection of the optimal fluids. Focusing on R1233zd(E) as working fluid, the turbine performance
map adoption leads to a reduction of the optimal cycle maximum pressure 𝑃𝑚𝑎𝑥 of about the
15% − 52%, depending on the available heat power input, while the maximum temperature 𝑇𝑚𝑎𝑥
undergoes a 50°C reduction for the small-size engine case. In the framework of a potential fluid
screening, for low molecular complexity and low molar mass compounds (mainly refrigerants) the
use of constant turbine efficiency can lead to a large overestimation of their potential in small scale
applications where they are not recommended and to exclude them for WHR from large engines
where on the contrary they can get (especially R1233zde) very competitive performance with
respect to hydrocarbons. The difference between optimal cycle efficiency adopting hydrocarbons
and refrigerants decreases from from about 5-10 (small-size case) up to 1-3 percentage points
(large-size case), depending on the specific working fluids considered.
Finally, it is underlined that the use of constant turbine efficiency assumption for these fluids
results in optimal solutions that may strongly affect the final feasibility of the power plant. In
fact, such approach may lead to select a cycle thermodynamic not compatible with the effective
usage of a single stage radial inflow turbine. The performance map can be easily integrated in
numerical models for cycle optimization and allows to create a direct link between expander
efficiency and cycle design parameters, leading to the definition of optimal cycles that do not result
in an unfeasible turbine design.
62
PART II
EXPERIMENTS ON SUPERSONIC
ORC LINEAR CASCADE
In organic Rankine cycle (ORC) power systems, the turbo-expander strongly impacts the definition
of optimal working fluid, cycle layout and performance, and plant profitability. Additionally, ORC
turbine involves the major technical challenges as the demand for compactness and flexibility
of operation couples with severe compressibility and non-ideal gas effects. For these reasons
the design of this component combines reduced-order methods and high-fidelity CFD-based
aerodynamic models, whose validation is crucial but still limited due to a lack of experimental data.
This is to be ascribed to the complexity of the flow fields, to the technical challenges of performing
measurements on supersonic flows for a vapor at thermodynamic conditions close to saturation,
and to the relatively high costs and difficulties associated with the management of wind tunnels
operating with organic compounds. To fill this gap, an experimental campaign on supersonic ORC
turbine linear cascades has been carried out at Politecnico di Milano, whose main objective is to
characterize the flow field associated with the expansion of an organic fluid in non-ideal conditions
through a linear cascade representative of the statoric part of axial/radial ORC turbine stages.
This part of the thesis describes the conception and set-up of such a novel class of experiments,
discusses the associated technical challenges, all steps required to design the campaign, and the
devised measurement strategies. Finally, the outcomes and experimental data resulting from the
execution of different campaigns exploiting either nitrogen or a siloxane (MM) as working fluids
are described. The collected experimental data confirms the effectiveness of the designed linear
cascade in replicating the flow field of actual supersonic turbine stators. The remarkable agreement
observed between the experimental results and high-fidelity CFD predictions, for both nitrogen
and MM, validates the theoretical tools used in the design phase. Furthermore, it demonstrates the
suitability of the employed CFD models for the design and analysis of ORC turbines, as well as for
the validation of reduced-order models and loss correlations. Due to the lack currently existing
in the open literature, the collected data aims at serving as a reference for future experimental
investigations on ORC cascades.
64
Some contents of this part are also discussed in:
Manfredi, M., Persico, G., Spinelli, A., Gaetani, P., & Dossena, V., (2023) Design and commission-
ing of experiments for supersonic ORC nozzles in linear cascade configuration, Applied Thermal
Engineering, vol. 224: 119996, doi: 10.1016/j.applthermaleng.2023.119996
Manfredi, M., Persico, G., Spinelli, A., Gaetani, P., & Dossena, V., (2023) Nitrogen Experiments
on a Supersonic Linear Cascade For ORC Applications, XXVI Biennial Symposium in Measuring
Techniques in Turbomachinery, Pisa
Manfredi, M., Persico, G., Spinelli, A., Gaetani, P., & Dossena, V., (2023) Loss measurement
strategy in ORC supersonic blade cascades, Proceedings of the 4th International Seminar on
Non-Ideal Compressible Fluid Dynamics for Propulsion and Power, NICFD2022, Cham: Springer
Nature Switzerland, London 2023. p. 217-228.
Oliveti, M., Manfredi, M., Persico, G., Spinelli, A., Gaetani, P., & Dossena, V., (2023) Experiments
on Supersonic ORC Nozzles in Linear Cascade Configuration, Proceedings of the 7th International
Seminar on ORC Power Systems, Seville
65
66
CHAPTER 4
Experimental Facility and
Instrumentation
4.1 Experimental facility: Test Rig for Organic VApors (TROVA)

The Test Rig for Organic VApors (TROVA) is a facility located at the Laboratory of Compressible
fluid dynamics for Renewable Energy Applications (CREALab) of Politecnico di Milano. Below a
brief description of the plant is reported, together with an introductory explanation concerning its
standard operation and sequences during nominal experiments with organic fluids. The interested
reader is referred to Spinelli et al. (2010, 2013) and Pini et al. (2011) for further details. The
facility represents an innovative apparatus, conceived and developed to investigate the non-ideal gas
thermo-fluid dynamics of organic vapor flows. The plants was therefore designed and instrumented
to operate with high-pressure and high-temperature organic compounds, which present several
challenges in terms of operation and storage. In fact, these fluids decompose when heated beyond
their thermal stability limit temperature, especially if contaminated with ambient moist air, they
are expensive, and usually flammable. These features require dedicated actions to be taken in
order to safely and effectively run experiments with organic compounds, such as an ad hoc sealing
apparatus - compatible with high temperatures and non-reacting with the working fluid - and the
establishment of an environment within the test rig as uncontaminated as possible.
The TROVA is a blow-down wind tunnel implementing a phase-transition batch organic Rankine
cycle where a stationary test section replaces the expander and an heating phase alternates with
a discharging process. The facility is capable of operating organic compounds with maximum
pressure and temperature of 50 bar and 400 °C respectively, which allow to reach highly non-ideal
conditions for a large variety of fluids for ORC applications. Main components of the plant,
sketched in Figure 4.1, are:
67
Chapter 4. Experimental Facility and Instrumentation
• High Pressure Vessel (HPV), a 1 m3 vessel where the fluid is charged, typically in two-phase
conditions, with a mass defined by the desired operating pressure and temperature. To attain
test conditions starting from ambient temperature and the related saturation pressure, the
fluid is isochorically heated via an external electrical tracing heating system;
• Plenum, a settling chamber located just upstream of the test section that hosts an honeycomb
section implemented to improve the flow uniformity. The large cross section of the plenum
strongly reduces fluid velocity, allowing to reasonably assume the total flow quantities
(pressure, temperature, and enthalpy) equal to the static ones, as demonstrated for the total
pressure in Conti et al. (2023). The plenum is connected trough a pipeline to the HPV;
• Test Section, a 270 mm × 250 mm rectangular housing cut out by a cylindrical frame, which
may accommodate a custom steel plate to be assembled with the desired test component
(nozzle, aerodynamic profile, blade cascade, etc.). The test section, the plenum and all the
pipes up to the HPV are also heated in order to avoid fluid condensation during experimental
runs. To strongly limits heat dissipation toward the environment, the heated portion of the
plant and the pipes downstream of the test section are insulated with a double layer made of
ceramic fiber and rock wool;
• Low Pressure Vessel (LPV), a 6 m3 vessel that collects the vapor flow outgoing the test
section, avoiding the discharge of the working fluid into the environment and hence any
problem concerning high temperature, toxicity, and flammability. In the LPV the vapor is
de-superheated and condensed through a thermal oil heat exchanger, located at the shell of
the vessel and integrated with a cooling tower circuit and a plate oil/water heat exchanger.
• Fluid-Transferring Unit, a combination of pipes and pneumatically-actuated valves that,
together with a metering pump able to sustain a discharge pressure of 32 bar, allows liquid
transfer from LPV to HPV, closing the cycle and enabling next test preparation;
• Vacuum Pump, a liquid ring volumetric pump placed above the LPV and linked through
pipes and valves to both the HPV, the test section, and the LPV. This component reduces the
pressure in the LPV up to the level desired as boundary condition downstream of the test
section, allowing to increase the single experiment run duration. In fact, the lower is the
LPV starting pressure the longer is the time required to rise the LPV pressure up to match
the one upstream the test section. Additionally, the vacuum pump allows to remove moist
air and other non-condensable gases in the plant, which are undesired as they operates as
catalysts for thermal decomposition of the working fluid.
• Nitrogen system, composed by bundles with a storing capacity of 800 liters, where the
nitrogen is stored at an initial pressure of ≈ 200 bar, and by a combination of pneumatic
lines and valves, which connect the bundles with different sections of the facility (HPV, LPV,
and test section). The nitrogen serves for multiple scopes: first, it is used to pressurize one or
more components of the facility at the end of the testing day or between two subsequent tests,
in order to prevent any inward leakage of ambient air. This system complements the vacuum
pump in the setup of a test environment as uncontaminated as possible. Second, nitrogen
is used during the execution of experiments with organic fluids to flush the pneumatic
lines connecting the test section with the remotely assembled transducers. As explained
in Section 4.2.2, this procedure avoids the condensation of organic vapors within the lines.
Finally, experiments can be carried out exploiting nitrogen as working fluid to provide
a preliminary verification of the effectiveness of both test section design and adopted
measurement techniques. As detailed in the following paragraph, the execution of a single
test run is indeed much faster and easier exploiting nitrogen as working fluid.
68
4.1. Experimental facility: Test Rig for Organic VApors (TROVA)
Figure 4.1: TROVA plant layout
69
TROVA operations
A single experiment exploiting an organic vapor as working fluid may be divided in three phases:
loading of the working fluid in the HPV and pressurization of the tank, pressure reduction in
both test section and LPV and flow discharging process (actual test run), fluid transferring and
preparation for the subsequent test. At the beginning of a specific campaign, the organic compound
selected as working fluid is manually charged in the LPV, where it is also stored between two
subsequent test runs. The first phase displaces the fluid, usually in two-phase conditions, to the
HPV by means of the metering pump. As previously mentioned, the total mass of fluid to be
charged in the HPV is computed based on the ambient temperature and the corresponding saturation
pressure. This calculation ensures that the desired test operating conditions, in terms of pressure
and temperature, are achieved after the fluid is heated isochorically. The heating process takes from
4 to 6 hours depending on the target fluid temperature and loaded mass amount, making extremely
difficult and unlikely the execution of 2 test runs in the same day.
In the subsequent phase the pressure in the components downstream the HPV (LPV, plenum,
test section, and in-between pipelines) is reduced - up to about 50 − 80 mbar - through the vacuum
pump. This is done: i) to setup a pressure ratio between the inlet and outlet of the test section
higher than the critical value and suitable to enable the development of supersonic flows; ii) to
avoid the mixing of moist air or non-condensable gases with the working fluid; iii) to increase
the test duration. Once test conditions are reached and the pressure downstream the HPV is
sufficiently low, the opening of a series of valves (V3 and main control valve in Figure 4.1) starts
the actual test run. Valves opening indeed enables the progressive discharge of the HPV, which
feeds the test section with organic vapor flow at saturated, superheated, or supercritical conditions.
Tests typically last from tens to hundreds of seconds depending on operating conditions and test
section passage area, both of them affecting the discharged mass flow rate. Despite of the transient
operation, a quasi-steady flow field establishes in the test section, featuring a characteristic time at
least 2 order of magnitude lower than the one associated with HPV emptying process. Moreover,
the instrumentation adopted has a frequency response sufficiently high to follow the low-speed
dynamics typical of the emptying process (usually below 1 Hz). The transient discharging of
organic vapors that characterizes the TROVA facility entails a relevant advantage: it allows for the
exploration of a wide range of operating conditions in a single run during the HPV emptying process.
This includes investigating highly non-ideal states at the beginning of the test and approaching
almost ideal conditions as inlet pressure and temperature decrease.
Once the HPV emptying process is terminated, the organic fluid collected in the LPV in
two-phase conditions is cooled down and condensed before being transferred back to the HPV by
means of the metering pump. The full duration of a single test with an organic fluid ranges from 5
to 7 hours and, as previously anticipated, it is mainly dictated by the time required to heat-up the
fluid transferred in the HPV.
To have relatively easy, fast, and safe test runs, TROVA facility was conceived also to allow for
experiments exploiting nitrogen as working fluid. Since nitrogen tests are carried out at ambient
temperature, the heating up of both the HPV and the test section is not required, leading to a
drastic reduction of the first phase duration from 4-6 hours up to 20-30 minutes. The nitrogen
stored in the bundles is indeed simply charged in the HPV through an apposite regulator and a
system of pipelines and valves, which remain open until the desired pressure level in the HPV
is reached. The second phase, that is the pressure reduction by means of the vacuum pump and
the actual test run, is not dependent on the working fluid but only on the actual mass flow rate
featuring the test. Conversely, once the HPV emptying process is terminated (third phase) the
nitrogen is firstly collected in the LPV and then discharged in ambient through the opening of a
manual relief valve. Experiments with nitrogen are extremely useful since they allow for an easy
and rapid assessment of several aspects in simplified thermodynamic conditions. This includes the
70
4.2. Adopted measurement instrumentation
assessment of instrumentation possible malfunctioning, sealing system verification (leaking tests),

characterization of the dynamic response of pneumatic lines, and assessment of the effectiveness
of both test section design and adopted measurement strategies.
4.2 Adopted measurement instrumentation

In order to safely perform fruitful experiments involving the expansion of organic vapors, often in
transonic or supersonic conditions, the TROVA facility is equipped with instrumentation that allows
to keep under control the pressure and temperature levels throughout the test rig. Although this
instrumentation is of relevant importance to ensure correct and effective test runs, it is not related
only with the experimental campaigns designed and carried out in the present research activity and
therefore its description lies outside the scope this work. The interested reader is referred to Spinelli
et al. (2010, 2013) and Pini et al. (2011) for further details on such instrumentation. This section
reports a brief description of the commercially available off-the-shelf (COTS) instrumentation
employed to characterize the flow field developing within the test section, together with an overview
of the procedures adopted for their calibration. On the contrary, the measurement instruments
specifically designed for this experimental campaign (such as the total pressure probe), their
validation, and the procedures adopted to assess their measuring specifications are described in
Section 5.3. Similarly, the measurement strategies specifically conceived and adopted to gather a
comprehensive set of data from the flow field - such as sensors and probes number and position -
are also described in Section 5.3.
The COTS instrumentation employed in the experimental campaigns described in this work may
be split into three categories that are singularly described in the following subsections: temperature
sensors, pressure transducers, and a double-passage schlieren equipment.
4.2.1 Temperature sensors

All temperature measurements are performed resorting to thermocouples, whose well-known
working principle can be deepened in Arts & Buchlin (2018), as characterised by high reliability,
low cost and large availability, wide temperature measuring range, and possibility of miniaturisation.
Two thermocouples of different type (type J and K) are used for the temperature measurements
within the test section, in order to have a redundant double check of the acquired temperature values.
Particularly, the type J thermocouples are composed by Iron-Costantan wires, have a temperature
range from -210°C to 760 °C and a grounded junction of 0.5 mm diameter, while the type K ones
are composed by Chromium-Nickel, have a temperature range from -270 °C to 1260 °C and an
exposed junction of ≈ 0.20 mm diameter. The characteristic time of both types is small enough to
closely follow the temperature variation during HPV emptying process (Spinelli et al., 2018).
The thermocouples used to characterize the flow field in the test section have been calibrated
in-house before the beginning of the experimental campaign, according to the standard procedure
described in the following. Each sensor under scrutiny is inserted inside an oven, whose temperature
is monitored by a primary reference K type thermocouple, featuring an expanded uncertainty (95%
confidence level) equal to 0.5°C. A finite number of set point temperature values for the oven is
selected in order to cover the temperature range expected during the experiments. For this specific
case, the values chosen were: ambient temperature, 80°C, 130°C, 180°C, 230°C, and 280°C. For
each temperature, the voltage provided by the thermocouple to be calibrated is amplified and
linearized by a conditioning module, and then acquired at 250 Hz for 1 second, leading to a total
number of 250 data acquisition points, which allows to obtain meaningful values of mean and
variance. This permits to determine the coefficients of the temperature-voltage (T-V) calibration
curve, which can be considered linear due to the implementation of the conditioning module, such
71
Figure 4.2: Calibration curve of a K-type thermocouple obtained through Montecarlo method.
that
𝑇 = 𝑎𝑉 + 𝑏 ,
where 𝑎 and 𝑏 are the calibration coefficients. The post-processing algorithm exploited is based on
the adaptive Montecarlo method (Kirkup & Frenkel, 2006) and requires as inputs the temperatures
provided by the reference thermocouple, its expanded uncertainty, and the mean and the variance
of the acquired voltage values. The algorithm provides as output the two calibration coefficients
(𝑎 and 𝑏) and the corresponding uncertainties (𝛿𝑎 and 𝛿𝑏). Additionally, the routine returns also
the Sum of Squared Errors (SSE), the coefficient of determination 𝑅 2 , which is representative
of the linear regression ability to well predict new observations, and the thermocouple expanded
uncertainty. For each calibrated thermocouple, the computed 𝑅 2 is greater than 0.9999, indicating
an excellent prediction capability, and the expanded uncertainty lower than or equal to ±1°C. In
Figure 4.2 it is represented an illustrative calibration curve of a K-type thermocouple selected to be
used during the experiments.
4.2.2 Pressure transducers

The set of COTS pressure sensors exploited belongs to two main categories: absolute and differential
transducers. Regardless of the sensor considered, the uncertainty is usually proportional to the
full scale (FS) pressure, and the dynamic frequency response of the transducer is generally above
200 kHz. However, the dynamic frequency response of a specific pressure sensor can be lower
as it depends on the system of pneumatic lines and cavities associated with the sensor itself. As
detailed in Section 5.4, most of the absolute pressure sensors are fastened directly to the rear side
of the test section and the lines-cavities system consists simply in a hole connecting the front
and rear sides of the test section (see Figure 5.33.b). In this case, the dynamic response of the
pressure sensor is surely more than sufficient to represent the slow dynamics associated with HPV
emptying process (usually below 1 Hz). Conversely, remotely assembled pressure sensors feature a
more complex system of pneumatic lines and cavities, whose dedicated dynamic calibration is
thoroughly described in Section 5.3.2.
The absolute transducers belong to the XTEH series by Kulite® . They have a nominal full
scale that varies between 1.75 bar and 40 bar and an uncertainty of ≈ 0.1% FS, as declared by
the manufacturer. These sensors consist of a miniaturised piezoresistive sensing element able to
withstand up to 454 °C, making possible their integration with the heated test section. While the
72
description of the general working principle of such transducers lies outside the scope of the present
work and can be found in (Arts et al., 2018), it is worth to highlight that a thermal compensation
was implemented within the measuring circuit due to the high sensitivity of the piezoelectric
transducers to temperature. This was done inserting a resistor with known resistance value in the
voltage supply circuit, where temperature is constant. In this way, the voltage drop through the
resistor is function only of the transducer temperature, allowing to account for the thermal effects
when converting the measured voltage signal into pressure. As better detailed below, the sensitivity
to the temperature requires transducers calibration in both pressure and temperature.
The COTS differential pressure transducers available are GP:50® . These sensors embed a
membrane-type piezoresistive sensing element, have a declared uncertainty equal to 0.05% FS, and
do not withstand high-temperature levels, constraining their usage far from the heated part of the
plant, including the test section. This imposes the employment of more complex pneumatic lines
whenever a differential pressure transducer is needed. The pneumatic lines used in the experiments
described in the present work - composed by capillary pipes and T- or I-shaped junctions - connect
the total pressure probe and the plenum with differential and absolute pressure transducers remotely
mounted. The whole design of the lines set-up had as main objective the achievement of a dynamic
response of the system as fast as possible, concurrently ensuring simplicity during the assembly,
avoidance of gas/vapor leakages, and capability to withstand high pressure values. For these
reasons, polytetrafluoroethylene (PTFE) pipes of 2 mm internal diameter were selected, capable
to withstand pressures as high as 25 bar. All the connections between different components are
made through brass compression fittings and T-shaped junctions, which combine high resistance to
pressure (up to 125 bar) and an easy handling during assembly/disassembly, while NPT adaptors
were chosen to connect the PTFE pipes to the transducers and electric valves. The dynamic
response of the pneumatic lines is investigated in Section 5.3.2, in which the total pressure probe
design is detailed. This choice is motivated since the probe turned out to be the bottleneck that
limits the maximum achievable cut-off frequency of the pneumatic system. All the components
not integrated with the heated test section, including the pneumatic lines, need to be flushed with
nitrogen during experiments employing organic fluids. This procedure is required to avoid the
condensation of organic vapors along the line, which are exposed to lower temperature levels
moving away from the heated region. The fluid phase transition would indeed affect the pressure
measurements and would produce an unacceptable promptness reduction. The interested reader is
referred to Conti et al. (2022) for additional details about nitrogen flushing procedure.
The pressure sensors required to characterized the flow field within the test section have been
calibrated in-house before the beginning of the experimental campaign, following a standard
procedure that differs depending whether absolute or differential transducers are considered. Since
absolute pressure sensors are sensitive to temperature they are inserted inside an oven to control the
temperature (measured through a primary reference K type thermocouple) and connected through
a pneumatic line with a vessel, whose pressure can be controlled with a regulator and measured
through a reference primary transducer. A separate pressure-based calibration was therefore made
at different temperature levels, namely: ambient temperature, 80°C, 130°C, 180°C, 230°C, 280°C,
and then, again, at ambient temperature, to mitigate the risk of any hysteresis phenomenon. The
pressure-based calibration at a fixed temperature follows a well-known standard procedure: 10-15
equally spaced pressure values between 1 bar and the transducer FS value are considered and for
each of them the output voltage provided by the sensor under scrutiny 𝑉𝑃 and the one read across
the compensation resistance 𝑉𝑇 are acquired at 250 Hz for 1 second, leading to a total number
of 250 data acquisition points for both 𝑉𝑃 and 𝑉𝑇 . The collected values shape a matrix, each
element of which represents the 250𝑥2 acquired voltage values for a particular pair of pressure
and temperature. These values, together with the pressure provided by the primary transducer,
can be used to determine the coefficients of the calibration curve linking the pressure 𝑃 with the
73
Figure 4.3: Calibration surface achieved through adaptive Montecarlo method for an absolute
pressure transducer with 25 bar FS.
two acquired voltage signals 𝑉𝑃 and 𝑉𝑇 , which for such absolute pressure transducers may be
considered linear with respect to 𝑉𝑃 and quadratic with respect to 𝑉𝑇 , such that

𝑃 = 𝑎 + 𝑏𝑉𝑇 + 𝑐𝑉𝑇2 𝑉𝑃 + 𝑑 + 𝑒𝑉𝑇 + 𝑓 𝑉𝑇2 , (4.1)
where 𝑎, 𝑏, 𝑐, 𝑑, 𝑒, and 𝑓 are the calibration coefficients, which are computed, together with
the corresponding uncertainties, by means of a post-processing routine based on the adaptive
Montecarlo method for multiple outputs Kirkup & Frenkel (2006). The algorithm requires as
inputs the mean and the standard deviations of both the pressure read by primary transducer, 𝑉𝑃 ,
and 𝑉𝑇 , as well as the covariance matrix between 𝑉𝑃 and 𝑉𝑇 . Since for each calibration point
𝑛 = 250 voltage values are acquired for both 𝑉𝑃 and 𝑉𝑇 , the mean 𝑉, variance 𝜎𝑉 and covariance
𝜎𝑉𝑇 ,𝑉𝑃 can be computed as
𝑛
1 ∑︁ 𝑖
𝑉𝑇,𝑃 = 𝑉 , (4.2)
𝑛 𝑖=1 𝑇 , 𝑃
𝑛 2
1 ∑︁ 𝑖
𝜎𝑉𝑇,𝑃 = 𝑉𝑇 , 𝑃 − 𝑉 𝑇 ,𝑃 , (4.3)
𝑛 𝑖=1
𝑛
1 ∑︁
𝜎𝑉𝑇 ,𝑉𝑃 = 𝑉𝑇𝑖 − 𝑉𝑇 𝑉𝑃𝑖 − 𝑉𝑃 . (4.4)
𝑛 𝑖=1
Figure 4.3 reports an illustrative calibration surface obtained through the mentioned method for
a transducer with 21 bar full scale. Additionally to the coefficients highlighted in Equation 4.1
and their corresponding uncertainties, the algorithm provides as output the Sum of Squared Errors
(SSE), the coefficient of determination 𝑅 2 , and the transducer expanded uncertainties. Similarly to
thermocouples, the 𝑅 2 of all the absolute pressure sensors calibrated is greater than 0.9999, while
the expanded uncertainty is around 0.1% FS for most of them.
The calibration procedure for differential transducers is easier and faster since they are designed
for low temperature applications and thus remotely mounted. Consequently, these sensors do not
require the calibration at different temperature values and the standard pressure-based calibration
at a ambient temperature can be performed. The procedure is similar to the one just described
74
except for the need to place the transducers in the oven. In fact, they were simply connected to
the vessel, whose pressure is measured through the reference primary sensor. In this case, the
pressure is function of only 𝑉𝑃 and the calibration curve linking pressure 𝑃 and voltage 𝑉𝑃 is
well-approximated by a linear function (linear transducers), such that
𝑃 = 𝑎𝑉𝑃 + 𝑏 , (4.5)
where 𝑎 and 𝑏 are the calibration coefficients, which are also computed through the same routine
based on adaptive Montecarlo method and describe above. The algorithm provides as output 𝑎 and
𝑏, their uncertainties, the coefficient of determination 𝑅 2 , and the expanded uncertainty, which
resulted around 0.05% for all the differential transducers calibrated.
As revealed in previous experimental campaigns, the value of the calibration curve intercept
coefficient, namely 𝑑 in Equation 4.1 and 𝑏 in Equation 4.5, depends on the ambient conditions at
which the pressure transducers are exposed. This may lead to a modification after each test of the
intercept coefficient, especially for those sensors integrated with the heated test section. To avoid
the execution of a new calibration of all the sensors after each test, a procedure defined as on-line
zero is performed before each test run (detailed in Conti et al. (2022)), which allows to compute the
current value of the intercepts. During this procedure the plenum pressure is measured before the
test run through a calibrated absolute pressure transducer, whose intercept is, in turn, determined
with respect to the atmospheric pressure, provided by a reference primary sensor. The knowledge
of the plenum pressure before the test execution, together with the acquired voltage signals, allow
to compute the intercept value for each transducer employed.
4.2.3 Schlieren equipment

TROVA facility is equipped with a double-passage schlieren equipment in order to visualize the
flow field density gradients, highlighting the morphology of shock/fan waves originated within
the test section with an almost continuous space resolution. The schlieren method is an optical
visualization technique that exploits the deflection of light rays from their undisturbed path when
they pass through a medium in which there is a component of the density gradient (entailing a
refractive index gradient) normal to the rays direction. The light rays deflection depends on the
intensity of normal density gradients and therefore may be exploited to visualise such gradients
through the usage of an opaque obstacle (knife edge) filtering defected ray. The final outcomes of
this flow visualisation procedure are grey-scale images highlighting normal density gradients. The
equipment arrangement in this research was such that compression pattern (density rise) appear
dark while expansions (density drop) result to be brighter. For further details concerning the
schlieren method working principles the interested reader is referred to Degrez et al. (2018). As
sketched in Figure 4.4, the main components of the TROVA schlieren equipment are:
• Light source: a monochromatic light-emitting diode that, together with a miniaturised system
of lenses, is able to focus the generated light beam in a single point, resembling a point
source and therefore improving image quality;
• Beam splitter: a glass cube that is crossed twice by the light beam, allowing its re-direction
to the image-recording element;
• Schlieren head: an achromatic lens that collimates the light beam from the source and
focuses the return light beam on the knife;
• Quartz window: a transparent element that covers and closes the test section. This component
is crossed twice by the light beam, which cross the test section and is reflected by the mirror-
polished surface place on the rear plate of the test section;
75
Figure 4.4: Schlieren bench functioning scheme.
• Knife: a razor blade able to move along z direction (as in Figure 4.4), which makes density
gradients in direction x visible by obstructing the return light beam;
• Camera: a sensing element with changeable lenses and fast acquisition rate that captures the
refractive index variations induced by density gradients;
• Video system: a PC that receives the sequence of images recorded by the camera and produces
a synchronised video of the performed test, being triggered by the test execution start.
The components just listed are arranged according to a double-passage configuration: this layout
guarantees two important advantages, as the system gradients sensitivity is doubled by the crossing
the test section twice and it allows to fully instrument the rear plate side of the test section since all
the optical equipment is placed on a single bench. The schlieren method is extremely useful in
the framework of organic vapors supersonic flows since it allows to visualize density gradients
and, consequently, to highlighting the morphology of shock waves and expansion fans. The
understanding of shock/fan structures is of primary importance to characterize the flow field
developing within the test section and to compare it with numerical results. Under the assumption
of knowing the flow direction, the schlieren output can be also used for a direct Mach number
measurement, as detailed in Cammi et al. (2021b), even if this scenario is not relevant for the
experimental campaigns described in this work.
4.3 Data acquisition system and data post-processing

The thermocouples and pressure transducers exploited to characterize the flow field developing
within the test section, described in Section 4.2, are electrically connected to an high-frequency data
acquisition (DAQ) system, which in turn is linked to a data acquisition PC (pcACQ). These sensors
are analogically powered and their output voltage signals are appropriately amplified to match the
0 − 10 V range required by the DAQ system. Particularly, the amplification of differential pressure
transducers is internal, while the absolute pressure transducers and the thermocouples are amplified
by means of a specifically designed signal conditioning system. Despite the high-frequency DAQ
system, the actual acquisition frequency is limited to 1000 Hz due to the implementation of a
multiplexer required to acquire the electrical outputs provided by all sensors adopted. Voltage
signals are afterwards converted in .asc and .dat files through dedicated LabVIEW® routines. Data
recording is synchronized with the test run thanks to a trigger signal associated with the V3 valve
opening, which marks the start of a test run.
76
4.3. Data acquisition system and data post-processing
Output .asc and .dat files are post-processed through a dedicated routine, which firstly
associates each column in the files (representing an output voltage signal) to a specific transducer
and subsequently averages raw data according to a specific time window chosen by the user,
usually 100 ms (10 Hz). The regression relations resulting from the calibration procedure are then
applied, allowing to convert voltage average data in either pressure or temperature depending on
the specific sensor considered. The routine provides also the uncertainty associated with each
measured data point, computed combining the uncertainties related to the coefficients of calibration
regression relations (expressed in terms of covariance matrices) and the standard deviation resulting
from the raw voltage data averaging operation. This process allows to associate to each directly
measured quantity (pressure or temperature) a corresponding uncertainty, which can be graphically
represented through error bars.
To compute the uncertainties associated to non-directly measured quantities the theory of
the propagation of measurement errors is applied (Arts, 2018). Assuming that a general derived
quantity 𝑦 is a function of 𝑛 measurements 𝑥𝑖 , such that 𝑦 = 𝑓 (𝑥 1 , 𝑥2 , . . . , 𝑥 𝑛 )), then the general
form of the expression for the uncertainty in the derived quantity 𝛿 𝑦 as function of its measured
parameters (assuming independent 𝑥𝑖 ) is
v
t 𝑛 2
∑︁ 𝜕 𝑓 (𝑥1 , 𝑥2 , . . . , 𝑥 𝑛 )
𝛿𝑦 = · 𝛿 𝑥𝑖 , (4.6)
𝑖=1
𝜕𝑥𝑖
where 𝛿 𝑥𝑖 is the uncertainty related to the 𝑖-th measured quantity 𝑥𝑖 . Equation 4.6 results actually
exploitable if 𝑦 may be related to measured parameters through one or more explicit analytical
relations. However, this is not always the case, especially in the framework of non-ideal organic
flows, for which the adoption of thermodynamic libraries to compute flow properties is largely
widespread. Whenever analytical relations are not available, or too complex to be used explicitly,
the uncertainties are usually propagated associating a probability distribution function (PDF) to
the measured parameters 𝑥 𝑖 (often Gaussian) and considering the relation between inputs 𝑥 𝑖 and
output 𝑦 as a black box. Randomly sampling the parameters 𝑥𝑖 is possible to set-up a certain
number 𝑁 of input vectors 𝑥¯ = {𝑥1 , 𝑥2 , . . . , 𝑥 𝑛 } that can be used to evaluate the mathematical
model under scrutiny and generate 𝑁 times the output 𝑦. If the number of model evaluation 𝑁 is
sufficiently large is then possible to estimate the mean, variance, and final propagated uncertainty
of the derived quantity 𝑦. One of the most common approach resorting to such procedure is the
Monte Carlo method, thoroughly described in Kirkup & Frenkel (2006). This method usually
requires a very large number of sample input vectors, named Monte Carlo trials, to achieve a 95%
confidence interval on the propagated uncertainty, at least 106 for the standard Monte Carlo routine
and 104 − 105 for the adaptive one, according to Kirkup & Frenkel (2006). It is evident that the
computational burden for Monte Carlo, whether it is standard or adaptive, becomes extremely high
when the black box contains a relatively complex mathematical model, such as the solution of a
system of non-linear equations. In this case the computation of the propagated uncertainty for all
the data point acquired during a single test run may easily take 8 to 10 hours.
This is the reason why in the present a work a routine was developed to compute the propagated
uncertainty of non-trivial relations based on the Polynomial Chaos Expansion (PCE) method. The
PCE is a metamodelling (or surrogate modelling) technique that aims at providing a functional
approximation of a computational model (the black box) through its spectral representation on
a suitably built basis of polynomial functions (Marelli et al., 2022). Consider the input vector
𝑥¯ ∈ R𝑛 with independent and stochastic components described by corresponding probability
density functions, the polynomial chaos expansion of the PDF 𝑌 - or finite variance computational
model - of the output 𝑦 may be written as
∑︁
𝑌= 𝑏 𝛼𝛹 𝛼 𝑋¯ , (4.7)
𝛼∈N𝑛
77
Figure 4.5: Comparison between Monte Carlo and PCE methods: computed uncertainties related
to the total pressure loss through the cascade against test running time.

where 𝛹 𝛼 𝑋¯ are multivariate polynomials orthogonal to the probability density function character-
ising the input variables, 𝛼 ∈ N𝑛 is a multi-index that identifies the components of the multivariate
polynomials, and 𝑏 𝛼 are the corresponding expansion coefficients. Notice that the series in
Equation 4.7, potentially composed by an infinite number of addends, is truncated at the 𝑀-th
order such that 𝛼 ∈ A with A ⊂ N𝑛 , where 𝑀 represents also the order of the polynomials 𝛹 𝛼 𝑋¯ .
The multivariate orthogonal polynomial basis is chosen accordingly to the probability distribution
function of the inputs concerned, Hermite orthonormal polynomials being usually selected for
Gaussian distributed input variables. Once the polynomial basis is chosen, the computation of
the coefficients 𝑏 𝛼 - whose number depends on the input vector size 𝑛, order of the polynomials
𝑀,and series truncation scheme adopted - allows to retrieve an analytical function to describe
the output PDF 𝑌 . It is then straightforwardly possible to retrieve both the average, the variance,
and the standard deviation of the output quantity 𝑦, as well as all the other probabilistic moments.
The description of the methods developed to compute of coefficients lies outside the scope of the
present work, the interested reader is referred to Marelli et al. (2022) and Ko & Wynn (2016) for
further details. However, it is worth to highlight that much lower number of sampled input vectors
𝑥¯ are needed to compute the coefficients, and consequently the propagated uncertainty 𝛿𝑦 (standard
deviation of derived quantity), with respect to those required by the Monte Carlo method. In fact,
the total number of inputs is in the order of hundreds of samples for low order polynomial degree
(𝑀 ≈ 3 − 6) and of thousands of samples for high-order multivariate polynomials (𝑀 > 10). This
allows to reduce the computational time required to propagate the uncertainty through non-trivial
models - for all the data point acquired during a single test run - from the 8 to 10 hours associated
with the Monte Carlo approach to a couple of minutes when exploiting an algorithm based on the
PCE method. Figure 4.5 reports, as illustrative example to compare Monte Carlo and PCE methods,
the uncertainty related to the total pressure through the cascade, computed solving a system of
non-linear equations, as detailed in Section 5.1. Specifically, Figure 4.5 depicts the uncertainties
computed by means of the two methods against the running time of the reference test considered.
Notwithstanding the huge differences in terms of computational time between the two methods,
Figure 4.5 proves the effectiveness of the PCE in correctly computing the propagated uncertainty.
78
CHAPTER 5
Design of Experiment
5.1 Concept of the cascade experiment

Despite experimental data related to canonical compressible flows of organic fluids are available
and mentioned in Section 1.2, experiments focused on the detailed investigation of non-ideal and
supersonic flows through ORC turbine cascades, including total pressure loss estimation, are still
missing in the open literature. This was the main motivation driving the choice of conceiving, and
subsequently designing, the experimental campaign here described, as already stated in Section 1.3.
By reproducing all the important features of these flows, namely the trailing-edge shock system,
the wake generation and recovery, and the total pressure loss, such experiments would also be
instrumental for constructing and validating high-fidelity CFD models implemented into ORC
turbine design and analysis tools. With the aim of generating benchmark data, the experiment
arrangement was conceived to match multiple requirements. Designing a cascade geometry
representative of actual high-temperature and high-expansion ratio ORC turbines was deemed
key, as well as operating the blade row with a fluid at thermodynamic conditions relevant for
state-of-the-art high-temperature ORC power plants. To be representative of the application above
mentioned, the linear cascade should exhibit all the relevant features associated with annular turbine
stators and needs to operate at sufficiently high nominal Mach number (> 1.4) with an organic
fluid featuring non-ideal thermodynamic states - at least at the cascade inlet. Therefore, all the
design choices and measurement strategies have been devised to meet such requirements and to
concurrently comply with the reduced test section dimensions (270 mm × 250 mm rectangular
housing) and TROVA facility operations, described in Chapter 4, seeking solutions as simple as
possible.
This chapter reports the full process of experiment design, which substantially differs from
the case of conventional blade cascades (typically tested in cold-flow facilities using air at low
temperature and pressure) due to challenging operating conditions and the need of complying
with space constraints dictated by the test section geometry, while keeping at the same time
79
Chapter 5. Design of Experiment
αs
Stat
ion 0
Station 1
Station 2
Stat
Mea ion 1
Bowed Shock sure
ment
Total Pressure Line
Static Pressure Probe
Tap
Figure 5.1: Test section sketch.
the high spatial resolution needed for CFD validation. The design of experiment originates
from the fluid dynamic concept and includes blade shape definition and optimization in annular
configuration, followed by its adaptation in planar cascade within the TROVA test section (including
boundary walls) and structural verifications. The design of the experimental campaigns also
encompasses the measuring strategies conceived to characterize the flow field: number and type of
instruments, their eventual design, and their arrangement within the test section. CFD simulations
were largely employed throughout the procedure of experiment design, both for blade profile and
side-walls fluid-dynamic optimization, assessment the flow field characteristics - such as shock
patterns and attained degree of periodicity - and comparison with experimental data resulting from
preparatory tests. All calculations described in this chapter, including reduced-order models and
CFD simulations, were performed by applying a state-of-the-art Helmoltz energy equation of state
(Span & Wagner, 2003) for MM (Thol et al., 2016), implemented in the NIST RefProp database
(Lemmon et al., 2018).
The typically high value of the turbine stage expansion ratio, coupled with the low speed of
sound of organic compounds, generally lead to supersonic flows at the first stator outlet, with an
average Mach number that easily exceeds 1.4. Therefore, a converging-diverging geometry was
chosen as representative of ORC turbine nozzle. An illustrative sketch of the test section is reported
in Figure 5.1, while the design details are discussed in Sections 5.2.1 and 5.2.2. The cascade
features three complete blades, defining two central channels, together with two partial profiles at
the boundaries of the domain, sketched in grey in the figure. The chosen number of blades resulted
from a trade-off between obtaining a good flow field periodicity and complying with the reduced
dimensions of the test section, whose back-side end-wall is obtained with a 93x180 mm rectangular
plate (red box in Figure 5.1). In the final cascade layout, the two partial profiles are replaced by
side-walls (see Section 5.2.2 and Figure 5.14), designed to improve flow periodicity and reduce
shock disturbances at the measurement points. The blade profile of Figure 5.1 is to be considered
a baseline geometry, derived from an axial ORC stator investigated in (Romei et al., 2020) and
adapted to the present experiment to obtain a linear cascade configuration. The baseline shape
is optimized to minimize cascade entropy production, as described in Section 5.2.1. A stagger
angle 𝛼𝑠 ≈ 75° (see Figure 5.1) was selected for the cascade to be consistent with the application
considered and, at the same time, to limit the extent of the semi-bladed channel portion, leading
to a geometric pitch of 45 mm. The chosen stagger angle allows to avoid a markedly stretched
configuration, which would conflict with geometrical constraints of the test section, due to the
choice of designing the row with inlet and outlet parallel flows (no flow deflection). This option
not only notably simplifies the cascade assembly in the TROVA test section, but also turns out
to be representative of radial inflow nozzles, which typically receive rather tangential flow from
the volute and are characterized by camberlines with almost constant angle with respect to radial
direction, producing small flow deflections. Moreover, as proven in Section 5.2.1, this configuration
80
5.1. Concept of the cascade experiment
is also relevant for axial turbine supersonic stators, which are mostly front-loaded, with flow turning
(even remarkable) almost completed through the converging portion, while most of the entropy
production occurs in the diverging part of the channel. Additionally, as long as supersonic flows
are established, the flow rotation upstream of the throat does not affect the downstream flow field
topology.
The evaluation of space-resolved total pressure losses through the cascade is one of the primary
objective of the experimental campaigns here described, therefore the total pressure needs to be
either measured or retrieved both upstream (𝑃𝑇0 ) and downstream (𝑃𝑇1 ) of the cascade. Since
the cascade-exit flow is supersonic, a detached bowed shock sets up in front of the total pressure
probe adopted, whose design mainly affects the shock structure in terms of detachment distance
and far-field opening angle (Baumgärtner et al., 2021). The detached shock is sketched in Figure
5.1, where the head of an overhung probe is illustrated. The shock formation prevents from a
direct measure of the total pressure downstream of the cascade 𝑃𝑇1 as the actual pressure retrieved
through the probe is 𝑃𝑇2 (the total pressure downstream of the bowed probe-induced shock), which
is obviously lower than 𝑃𝑇1 due to shock losses. Under the assumptions of adiabatic conditions
within the test section (ℎ𝑡0 = ℎ𝑡1 = ℎ𝑡2 ) and of a normal shock in front of the probe head tap
(reasonable assumption for a flow-aligned probe) , 3 additional flow properties other than 𝑃𝑡2 need
to be concurrently measured to unambiguously retrieve 𝑃𝑡1 . This is demonstrated by the three sets
of equations reported in Equation (5.1), which specifies, from left to right, the thermodynamic
model of the fluid (where the double subscript indicates that those equations are accounted for both
in station 1 and 2), the definition of total enthalpy, and the integral balance equations through a
normal shock.

 ℎ = ℎ 𝜌1,2 , 𝑃1,2 (  𝜌 𝑉 = 𝜌2𝑉2
 1,2  1 12
 

 ℎ𝑇0 = ℎ (𝑇𝑇0 , 𝑃𝑇0 ) 

2𝑉2 − 𝜌1𝑉1 = 𝑃1 − 𝑃2
2
𝑠1,2 = 𝑠 𝜌1,2 , 𝑃1,2 , 𝑉12 , 𝜌 . (5.1)

 𝑠1,2 = 𝑠 ℎ𝑇0 , 𝑃𝑇1,𝑇2
 ℎ 1 + 2 = ℎ 𝑇0
 2
 ℎ + 1 = ℎ + 𝑉2
 𝑉 2
  1 2 2 2
In Equations 5.1 subscript 𝑇 refers to total conditions, subscripts 0, 1, and 2 specify respectively
section upstream/downstream of the blade row and downstream of the probe-induced shock (as in
Figure 5.1), while 𝑉 is the velocity, assumed fully normal to the central part of the bowed shock
being the probe a priori flow-aligned. Equations 5.1 form a system of 11 mathematical relation in
15 unknowns, namely ℎ1,2 , 𝑃1,2 , 𝜌1,2 , 𝑠1,2 , 𝑃𝑡1,𝑡2 𝑉1,2 , ℎ𝑡0 , 𝑃𝑡0 and 𝑇𝑡0 . Therefore, 4 independent
flow properties need to be measured to permit the solution of the mathematical problem. In
the present work the inlet total conditions 𝑃𝑡0 and 𝑇𝑡0 , the static (𝑃1 ) and total pressure (𝑃𝑇2 )
respectively upstream and downstream of the probe-induced shock are chosen as flow properties
to be measured. The devised methodology to concurrently measure 𝑃1 and 𝑃𝑇2 is described in
Section 5.3.1, while the design of the overhung total pressure probe is detailed in Section 5.3.2,
together with the corresponding assessment and validation procedures carried out.
Other crucial aspects were the selection of intake conditions relevant for state-of-the-art
high-temperature ORC power plants. Total inlet conditions (station 0 in Figure 5.1) in both design
and off-design operations were chosen to carry out expansion processes featuring highly non-ideal
fluid thermodynamic states, as pointed out in Romei et al. (2020). The nominal inlet total state
(labelled as DES in Figure 5.2a) was computed determining an inlet specific entropy as close as
possible to vapour saturation to enhance non-ideal effects along the expansion and concurrently
to allow the execution of tests in off-design conditions in both temperature and pressure (labels
OFF in Figure 5.2a) without the formation of liquid droplets within the test section. The design
entropy and total pressure were set to 𝑠 𝑑 = 0.883 kJ/(kgK) and 𝑃𝑇 𝑑 = 25 bar respectively, as
shown by line DES in Figure 5.2a. The figure shows that the expansion crosses a thermodynamic
region exhibiting low values of both compressibility factor 𝑍 and fundamental derivative of gas
dynamics 𝛤, the latter experiencing also significant gradients. 𝛤 specifies the variation of the
81
280
DES Γ
OFF-P
0.5
OFF-T+
0.4
1.10
OFF-T-
0.3
6
260 1.00
0.
0.90
7
0.80
0.
0.2 0.70
T (°C)
240
8
0.60
0.
0.50
9
0.40
0.
=
Z
0.30
220
0.20
0.10
0.00
200 sd -0.10
0.6 0.7 0.8 0.9 1

s (kJ/(kgK))
(a) (b)
Figure 5.2: (a) Temperature - specific entropy per unit mass diagram, with contour of 𝛤, 𝑍
isoline, and reference isentropic expansions in both design and off-design conditions for MM vapor.
(b) Mach number vs reduced pressure 𝑃/𝑃𝑐 distribution for MM vapor isentropic expansions
characterized by the same entropy 𝑠 𝑑 but different stagnation pressure.
speed of sound along isentropic processes (Thompson, 1971), and its low values, entailed by high
fluid molecular complexity, enhances non-ideal effects with quantitative (and possibly qualitative)
impact on the expansion flow field (Romei et al., 2020). This choice of conditions allows to widen
the explored range of non-ideality within a single test run, while keeping at the same time sufficient
safety margin with respect to both test section maximum pressure and fluid thermal stability limit.
Figure 5.2b shows several isentropic expansions for different values of the inlet total pressure and
temperature, the latter being computed from 𝑃𝑇 and entropy 𝑠 𝑑 . The expansions depicted clarify
as the design inlet total pressure was selected as an acceptable trade-off between the purpose of
attaining remarkable non-ideal gas-dynamic effects and the need to comply with safety constraints.
The corresponding inlet total temperature and compressibility factor result equal to 261.5 °C and
0.3 respectively. The outlet static pressure, marked with a circle in Figure 5.2b, was chosen to
have a high outlet Mach number, typical of high expansion-ratio applications, without relegating
the non-ideal effects in a small region close the nozzle throat. The design outlet Mach number
was thus set equal to about 1.6, leading to an adapted outlet static pressure of 5.77 bar, computed
through a reduced-order model based on the one-dimensional isentropic nozzle theory. Three
conditions, reported in Figure 5.2a, were identified as plausible choices to investigate cascade
off-design performance; the first one, referred to as OFF-P, is characterized by a higher inlet total
pressure 𝑃𝑇 = 28 bar and 𝑠 = 𝑠 𝑑 , while conditions labelled as OFF-T+ and OFF-T− , consider a
higher and lower inlet total temperature (±1 °C) respectively, at the design total pressure level
𝑃𝑇 𝑑 . For such class of off-design conditions non-ideal effects are expected to trigger significant
variability in the flow condition, as shown in Romei et al. (2020).
5.2 Test section design

5.2.1 Blade design
A further crucial step in the design of such a novel experiment is the identification of a blade shape
optimized for operating in the linear cascade, in terms of thermodynamic condition and cascade
set-up. To carry out a tailored blade design, computational fluid dynamics (CFD) is systematically
applied, at first to derive the baseline profile - reported in Figure 5.1 - starting from a reference
82
5.2. Test section design
Figure 5.3: Comparison between starting blade from Romei et al. (2020) and baseline blade,
characterized by a re-designed front-section and pressure side.
axial ORC stator, investigated in Romei et al. (2020). The design of the axial stator blade in Romei
et al. (2020) needs indeed to be modified in order to avoid flow separation when operating the
cascade without flow deflection. Subsequently, CFD simulations are also exploited to simulate the
baseline blade, aiming at assessing its performance for different value of the inlet flow angle, and
finally to perform a shape-optimization of the blade. These preliminary design studies are carried
out considering a periodic annular row configuration, while further calculations are performed
to simulate the flow in the actual linear cascade and to assess the degree of periodicity among
adjacent channels, as detailed in Section 5.2.2.
Regardless of the blade profile under investigation, annular cascade calculations are performed
using the CFD flow model developed at Politecnico di Milano for turbomachinery flow simulations,
based on the ANSYS-CFX solver, whose formulation specific for supersonic flows of non-ideal
gases is fully detailed in Persico et al. (2019) and Romei et al. (2020). The reliability of the
numerical model was previously assessed against experiments performed within a research turbine
installed at Politecnico di Milano (Gaetani & Persico, 2021): the model was able to reproduce
the detailed aerodynamics and loss mechanisms of the blade rows, and it provided estimates of
stage efficiency within 1% of the experimental datum, i.e., comparable to the uncertainty of the
measurement technique. The same model was also assessed for canonical supersonic flows, such as
the shock tube problem and the 2D compression corner (Mushtaq et al., 2022), providing excellent
results when compared to solutions and/or data available in the literature. Moreover, the suitability
of this computational set-up for non-ideal flow simulations was also investigated and confirmed
in Pini et al. (2017b). Simulations of each blade profile are performed considering a quasi-3D
slice, which is intended to be representative of the midspan section of the cascade. The solver
integrates the RANS equations, complemented by the turbulence model equations according to
the k-𝜔 SST formulation, by resorting to a look-up table approach to introduce the non-ideal
thermodynamics of the fluid. The equations were integrated over structured multi-block meshes
composed by about 220000 hexahedral elements, defined after a dedicated mesh-dependence
analysis, whose description and results lie outside the scope of the present work, and by assigning
𝑦 + ⩽ 1 at the blade walls. Calculations are performed by assigning total pressure, total temperature,
flow direction, and turbulence quantities at the inlet, and static pressure at the outlet. The total
conditions and the expansion ratio are defined as illustrated in Section 5.1 for the design point (line
DES in Figure 5.2a), while the turbulence quantities are assigned following (Romei et al., 2020).
Concerning the inlet flow angle, different values are assigned as boundary condition depending on
the specific type of analysis being conducted.
The first difficulty encountered was to adapt the blade investigated in Romei et al. (2020) -
referred to as starting blade in the following - to be operated in a linear cascade configuration
83
Figure 5.4: Mach number contour of the starting (left) and baseline (right) blades with an inlet
incidence angle of 75°.
with null flow deflection. In fact, the starting blade is part of the supersonic annular stator row of
an axial ORC turbine and it has been designed and optimized to operate with a purely axial inlet
flow and to provide a 75°flow deflection toward the tangential direction. On the other hand, the
chosen linear cascade configuration is quite different from the stator of conventional axial turbines
since the blades do not impart any rotation to the flow (as shown in Figure 5.1). In a conventional
configuration, no flow deflection corresponds to operate the cascade with highly negative incidence,
which in turn may lead to large flow separation regions if the front part of the blade is not properly
designed. This is the reason why the starting profile was modified by re-designing its front-section,
specifically the region of the leading edge in proximity of the stagnation point, as well as the initial
portion of the blade pressure side. These modifications led to the baseline profile, which can be
compared with the starting blade in Figure 5.3.
To verify their performance when operated with null deflection (−75°incidence angle), the two
blade geometries have been simulated in annular row configuration. Even though the numerical
set-up employed and the boundary conditions assigned are those aforementioned, the inlet flow
angle is set equal to 75°for both the annular rows, in order to reproduce the null-deflection condition
that features the linear cascade. The results of the numerical simulations over the annular rows
implementing the two blade geometries are reported - in terms of Mach number contour - in Figure
5.4 for both the starting and baseline profiles. As can be noted in the left contour of Figure 5.4, the
annular stator row implementing the starting blade features a large separation region in the airfoil
pressure side when operated at high negative incidence. In fact, the stagnation point drifts towards
the suction side and the flow undergoes a large turning around the blade leading edge (where the
stagnation point for an axial inlet flow occurs) and experiences adverse pressure gradients, which
produce the separation region. On the contrary, the annular row implementing the baseline profile
does not present any separation when operated at high negative incidence, as can be seen in the
right contour of Figure 5.4. This finding support the effectiveness of the re-designed front-section
and pressure side of the blade for such inlet conditions.
Further investigations are required to assess whether the use of a null-deflection configuration
could jeopardize the effectiveness of the linear cascade in representing a statoric annular row of an
axial ORC turbine in addition to the nozzle of radial-inflow expanders, which are characterized by
a very small camberline deflection. A dedicated analysis on the baseline profile concerning the
inlet flow angle is then required in order to investigate the effects of such boundary condition, in
particular on the post-throat flow evolution. Therefore, the baseline geometry was numerically
84
Figure 5.5: Mach number field (contour) with overlapped streamlines (grey solid lines) for the
baseline blade. Axial (left) and 75° (right) inlet flow comparison.
analyzed considering the two limiting conditions of inlet flow angle, measured with respect to the
annular row axis, equal to 0° (axial inlet flow) and 75° respectively. Despite the mass flow rate is
different in the two cases, due to the assignment of the outlet static pressure as boundary condition,
the Mach contour reported in Figure 5.5 highlights how the two flowfields downstream of the sonic
throat are almost identical in terms of Mach number value, shock and expansion wave patterns, and
wake thickness. Obviously, this is not the case of the leading edge region, where the variation of
the inlet flow angle leads to different stagnation points, pressure and suction side boundary layers
development, and blade pressure distribution. However, thanks to the large radius of curvature at
the leading edge, the baseline blade is highly tolerant to the severe change of incidence angle, and
no flow detachment is produced.
These outcomes are confirmed by the results reported in Figure 5.6, in which the blade loading
(on the right) and the inlet-to-outlet pitchwise mass-averaged pressure distribution (on the left) are
compared for the two cases. The pictures highlight how the differences in terms of blade loading
and pressure distribution occur only in the front part of the blade, upstream of the sonic throat.
The present comparison fully corroborates the re-design of the front-section of the blade, with
respect to the one described in Romei et al. (2020). By virtue of its peculiar leading edge shape,
the present blade is suitable to accept the meridional flow typical of axial machines as well as the
highly tangential swirling flow entering stators of radial-inflow turbines. In fact, in radial-inflow
turbines the distributor provides a high tangential flow to the nozzle, whose deflection results small
and thus close to the one reproduced in the proposed linear cascade experiment. Moreover, the
outcomes of the analysis just described prove the effectiveness of the concept of the experiment
and confirm the null-deflection linear cascade as representative of supersonic stators of both axial
and radial-inflow turbine.
Even though the baseline blade configuration was already satisfactory, a shape-optimization
process is performed to refine the blade shape in the rear part of the channel, especially from the
throat section to the outlet, as most of the entropy production in supersonic cascades occurs in this
region. The blade design is performed by applying the in-house shape-optimization tool FORMA
(Fluid-dynamic OptimizeR for turboMachinery Aerofoils), which optimizes the shape of blade
profiles by resorting to surrogate evolutionary strategies, formulated in constrained, multi-point and
multi-objective fashion. FORMA, whose details can be found in Persico et al. (2019), combines
a B-spline representation of the blade profiles, the CFD model previously presented, a Kriging
formulation as surrogate (Jones, 2001), and genetic algorithms, among which the differential
85
30 Inlet-to-Outlet 30 Blade Loading

27 A ial Inlet 27 A ial Inlet
75° Inlet 75° Inlet
24 24
21 21
18 18
P [bar]
P [bar]
15 15
12 12
9 9
6 6
3 3
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
X [-] S [-]
Figure 5.6: Baseline blade: axial and 75° inlet flow comparison. Inlet-to-outlet pitchwise
mass-averaged pressure distribution (left). Pressure distribution over the blade pressure and suction
side (right).
Baseline
Optimal
23 24 25 26
21 22
(a) (b)
Figure 5.7: Blade optimization procedure. (a) Baseline geometry parametrization; red and green
dots represent fixed and movable CPs respectively, the light green region represents the design
space. (b) Comparison of baseline geometry and optimal blade shape.
evolution approach is chosen (Storn & Price, 1997). Following the analysis just described, a inlet
flow angle of 75°is set as boundary condition. The baseline blade is, at first, parametrized using 30
control points (CPs), and only six of them, covering most of the blade suction side, were considered
movable during the optimization process; the mobility of such CPs being the design space of the
optimization problem. Figure 5.7a reports both fixed (red dots) and movable (green dots) CPs, as
well as the investigated design space (light green region). In this way, the optimization was focused
to the rear section of the blade suction side, with the aim of modifying the diverging portion of
the channel between two adjacent blades to minimize the loss generation during the supersonic
expansion process. The four CPs defining the trailing edge are kept fixed to concurrently avoid its
rigid translation and deformation. The two CPs on the suction side preceding the trailing edge
were instead kept fixed to avoid a thinning of the profile, which may lead to non-negligible blade
deformations in this region.
The global optimization strategy, which is applied for the present optimization problem,
involves an initial sampling of the design space for surrogate model interpolation including 60
evaluations of the blade annular cascade, followed by several training iterations that progressively
refine the surrogate as the optimization proceeds towards the optimum. The mass flow-averaged
entropy rise through the blade row is selected as the objective function to be minimized by the
optimization algorithm. Additionally, a constrain is imposed to ensure a flow angle at the domain
outlet within 74°and 76°, implying a flow deflection within ±1°. The optimization procedure has
been stopped after 48 iterations of the training phase, leading to 108 total cascade evaluations,
since a stable convergence between the surrogate and high-fidelity CFD results was reached. The
optimization convergence history during the training phase is reported in Figure 5.8, in which the
optimum objective function predicted by the surrogate model (blue curve) and the actual entropy
86
Entropy Rise J/(kg K)
Number of Training Iterations
Figure 5.8: Convergence history of the optimization training phase: optimum objective function
predicted by the surrogate model (blue curve) and by the high-fidelity CFD model (orange curve)
versus the number of training iterations.
Figure 5.9: Mach number field. Baseline (left) and optimal (right) blades comparison.
rise evaluated simulating numerically the corresponding training individual (orange curve) are
reported in function of the number of training iterations.
The baseline and optimal blade shapes are compared from a geometrical point of view in
Figure 5.7b, while Figure 5.9 reports the corresponding Mach fields and Table 5.1 summarizes
the pitchwise mass-averaged values of the main flow properties at the domain outlet for the two
configurations. In the table 𝑌𝑡𝑜𝑡 represents the total pressure loss coefficient and it is defined as
𝑃𝑇 ,𝑢 𝑝 − 𝑃𝑇 ,𝑑𝑛
𝑌𝑡𝑜𝑡 = , (5.2)
𝑃𝑇 ,𝑑𝑛 − 𝑃 𝑑𝑛
where the subscripts 𝑢 𝑝 and 𝑑𝑛 refer to upstream and downstream blade conditions respectively.
Besides the outlet Mach number is slightly larger, the optimal blade is characterized by lower Mach
number values within the supersonic diverging portion of the channel. This contributes to prove
that in the optimal configuration the flow undergoes a milder pressure drop up to the design outlet
one, leading to weaker fishtail shocks and, consequently, to lower entropy production and total
pressure losses, as highlighted in Table 5.1 by the values of 𝛥𝑠 and 𝑌𝑡𝑜𝑡 .
87
Table 5.1: Comparison of baseline and optimal blade geometries. Pitchwise mass-averaged flow
properties are evaluated at the domain outlet.
Baseline Blade Optimal Blade

𝛥𝑠 [J/(KgK)] 1.757 1.467
𝑀 [-] 1.569 1.574
𝛼 [deg] 75.8 74
𝑌𝑡𝑜𝑡 % 15.27 12.66
Figure 5.10: Pressure field acting on the optimal blade suction and pressure sides.
Since the optimal blade (as well as the baseline one) is characterized by a very thin tail, being
the thickness lower than 3 mm for a non-negligible portion of the profile up to 0.5 mm at the
trailing edge, its structural integrity and the expected deformation due to the blade loading need
to be verified. A further reason for such verification is that the aerodynamic loads involved are
remarkable, due to the large pressure difference occurring between the suction and pressure sides
of the blade. This is highlighted in Figure 5.10, in which the blade pressure distribution resulting
from CFD simulations is plotted together with the geometry of the profile.
The combination of the two features just mentioned motivated the choice of performing finite
element method analyses (FEM) of the optimal blade geometry. FEM simulations are carried out
using the Abaqus/CAE solver, a module of Abaqus FEA software (Smith, 2009). The geometry has
been imported from the CAD and constrained according to the actual dowel-clamping mechanism
employed in the experiment. The resulting structured mesh through which the blade solid body is
discretized is shown in Figure 5.11a, where it is also possible to observe the two holes that host the
dowels fixing the blades to the test section. The mesh is more refined in proximity of the trailing
edge since this represents the thinnest region where the largest displacements are expected. The
pressure distribution retrieved from CFD results and reported in Figure 5.10 has been imported in
the software and applied as an external load acting on the blade surface.
The results of the FEM simulation in terms of blade displacements are reported in Figure
5.11b, in which an amplification factor of 75 is applied for a more effective visualisation. As
expected, the trailing edge is the only part of the blade affected by non-negligible deformations,
which however produce a vertical displacement lower than 0.09 mm. The displacements of the
blade surface are reported in Figure 5.12 against the 𝑋 coordinate (refer to Figure 5.10), together
88
(a) (b)
Figure 5.11: (a) Optimal blade 3D structured mesh for FEM analysis; (b) Visualisation of the
deformed blade resulting through FEM analysis (amplification factor = 75).
Figure 5.12: Distribution of the maximum equivalent Von Mises stress and of the blade surfaces
displacement.
with the distribution of the maximum equivalent Von Mises stress. The resulting maximum stress
(𝜎𝑚𝑎𝑥 = 65 MPa) occurs in proximity of the hole closer to the trailing edge and it turns out to be
much lower than the material yielding limit (𝜎𝑌 𝑆 = 320 MPa), leading to a safety factor of about 5.
To investigate the effect of the blade deformation on the resulting flow field downstream the
throat section, the deformed blade geometry was numerically investigated in annular cascade
configuration, exploiting the same numerical set-up previously described. The results, in terms of
pitchwise mass-averaged flow properties at the domain outlet, are reported in Table 5.2, where
they are compared to those of the baseline and of the optimal undeformed blade. Clearly, the
deformation field caused by the blade pressure distribution does not produce any remarkable
variation on the resulting flow field. Therefore, the loss reduction attained by the optimization
process is not compromised, which proves the effectiveness of the blade design process.
The final optimized blade geometry is shown in Figure 5.13 where both the 3D model (a) and
the manufactured component are shown (b). The hole visible in Figure 5.13 is needed to extract the
blade from the test section rear end-wall, while the holes hosting the dowel-clamping mechanism
(see Figure 5.11) are machined in the other side of the blade and are not visible in Figure 5.13. A
horseshoe-shaped groove is also added on the surface of the blade to host a Gore-Tex string which
avoids a direct contact between the component and the glass placed in front of the test section.
89
Table 5.2: Comparison of baseline geometry with undeformed and deformed optimal blades.
Pitchwise mass-averaged flow properties are evaluated at the domain outlet.
Baseline Blade Optimal Blade Deformed Optimal Blade

𝛥𝑠 [J/(KgK)] 1.757 1.467 1.468
𝑀 [-] 1.569 1.574 1.574
𝛼 [deg] 75.8 74 73.95
𝑌𝑡𝑜𝑡 % 15.27 12.66 12.67
(a) (b)
Figure 5.13: Final optimal blade geometry: (a) 3D CAD model, (b) Manufactured blade.
5.2.2 Linear cascade adaptation

The optimal blade profile described in Section 5.2.1 and the corresponding annular row need to be
adapted in a linear cascade configuration. As introduced in Section 5.1, three complete blades
are allocated into the cascade in order to maximize flow field periodicity, still complying with
the reduced dimensions of the test section. The three blades are arranged horizontally to attain
a geometric blade-to-blade pitch of 45 mm and a stagger angle of 74°. The stagger angle has
been reduced by 1° to match the cascade outlet angle resulting from CFD simulation of the
optimized blade profile in an annular row configuration (see Table 5.2). The linear cascade final
layout is shown in Figure 5.14. The 3-blades row is further complemented by partial pressure side
(bottom right in Figure 5.14) and suction side (top left in Figure 5.14) profiles, integrated into
two side-walls bounding the flow field upstream and downstream the cascade. The side-walls are
required to accurately reproduce the periodicity and the flow field features observed in actual stator
annular rows. To attain a high level of periodicity and to minimize disturbances at the downstream
measuring points due to fishtail shocks departing from trailing edges and their reflections, the
design of side-walls was carried out reproducing the flow streamlines simulated in the annular row
configuration. To this end, Figure 5.15 highlights the velocity streamlines around 2 blade profiles
resulting from the CFD simulations in annular row configuration (corresponding to the Mach field
in the right-hand side of Figure 5.9). The bold red and blue streamlines in Figure 5.15 are therefore
extracted to define the tailored shape of the top and bottom cascade side-walls respectively, aiming
at maximizing the cascade periodicity and concurrently avoiding upstream recirculating regions.
The front part of both the side-walls was properly shaped to smoothly connect with the test rig pipe
connected at the test section inlet.
CFD-based investigations of the full linear cascade geometry have been extensively carried out to
thoroughly characterize the flow field. The numerical results indeed allow for the assessment of the
attained fluid-dynamic periodicity and, concurrently, for the properly refinement of the measuring
grid spatial distribution. Simulations were performed by means of the ANSYS-Fluent solver,
considering a 2D numerical domain representative of the midspan section of the cascade. Although
90
Figure 5.14: Final linear cascade arrangement, including 3 complete blades and the side-walls
design details.
Figure 5.15: Numerical results for the annular row configuration in terms of velocity streamlines
around 2 blade profiles: selected streamlines to design top (red) and bottom (blue) side-walls.
the same CFD flow model detailed in Section 5.2.1 is exploited (i.e., RANS equations complemented
by the k-𝜔 SST turbulence model and by look-up tables to introduce the thermophysical properties
of the fluid), these simulations are run with ANSYS-Fluent since ANSYS-CFX does not accept
two-dimensional geometries. This choice allows to avoid the implementation of fictitious layers
in the span-wise direction, thus reducing the computational cost. Preliminary comparisons have
been however made between the two solvers for both the annular and linear cascade configurations,
finding full qualitative and quantitative correspondence between the results obtained. Regardless
of the simulation under investigation, the pressure-based solver type is selected in ANSYS-Fluent
together with the built-in third order numerical scheme MUSCL.
The equations were integrated over structured multi-block meshes created with ANSYS-ICEM
CFD and composed composed by about 345000 hexahedral elements. Differently from the
physical one, the numerical domain implements inlet and outlet mesh blocks extending upstream
and downstream for about 0.75 and 5 blade chords respectively. The blocking approach is
shown in Figure 5.16 together with the full numerical domain. To ensure a constant near-wall
distance (𝑦 + ⩽ 1) and concurrently avoid extremely high computational costs, O-grid blocks
are implemented around both blades and side-walls. The mesh dimension was chosen after a
dedicated mesh-dependence analysis, in which also a coarser and a finer versions of the mesh
were tested, having around 168000 and 767000 cells respectively. Without going into exhaustive
91
Figure 5.16: Mesh blocking approach and full numerical domain.
Figure 5.17: Linear cascade: detail of the 2D mesh with 168000 cells.
details about the grid sensitivity analysis, the 345 thousand cells mesh was chosen as final grid to
perform the computations, since its difference with the finest one in terms of total pressure loss
coefficient resulted lower than the 0.07%. A detail of the grid with 168000 cells resulting from
the meshing procedure is shown in Figure 5.17. Since the flow axial Mach number is supersonic,
calculations were performed by assigning at the inlet total pressure and temperature, flow direction,
and turbulence. To assess the cascade periodicity, inlet conditions were first set accordingly to the
experiment design point, that is 𝑇𝑡 = 261.5°C and 𝑃𝑡 = 25 bar, while the inlet flow was considered
axial (see arrow in Figure 5.1) and the turbulence quantities are provided according to Romei et al.
(2020).
The numerical simulation results are shown in terms of Mach number and total pressure
loss fields in Figure 5.18a and Figure 5.18b respectively. In particular, the total pressure loss is
defined as the non-dimensional difference between inlet 𝑃𝑇0 and local 𝑃𝑇 total pressure. Both
the distributions prove the generation of wakes and trailing edge shock systems, highlighting the
absence of detached flows and large reticulation regions. Figure 5.18b demonstrates also that the
largest total pressure loss occur in the wakes, whose displacement thickness is relatively high due
to the high Mach number values involved. Moreover, both the numerical distributions provide a
first proof of the good periodicity of the downstream flow field, the reflection waves generated
at the bottom side-wall being the major source of non-periodicity. Figure 5.18 also reports (in
black and in white in Figure 5.18a and 5.18b respectively) the downstream measuring line, which
has be chosen parallel to the trailing edge locus and placed at a distance of about the 17% of the
blade chord (see also Figures 5.1). The numerical results reported in Figure 5.18 were exploited to
accurately locate a discrete set of measurement points along the measurement line. As detailed
in Section 5.3, the downstream measuring grid results non-uniformly spaced, but conveniently
refined in the vicinity of the wakes where higher gradients are expected.
The numerical results over the black line of Figure 5.18a were extracted to thoroughly investigate
the downstream flow field periodicity, as shown in the example reported in Figure 5.19a, in which
the Mach number distribution over the measuring line is plotted against the non-dimensional
pitch-wise distance 𝑥 𝑠 from the bottom blade trailing edge (see Figure 5.18). Figure 5.19a provides
92
(a) (b)
Figure 5.18: Two-dimensional CFD simulation of the full linear cascade: resulting Mach (a) and
total pressure loss (b) fields.
1 1
0.8
RXX - Mach Number
RXX - Entropy
0.5 0.6
0.4
0 0.2
-0.5 -0.2
0 0.5 1 1.5 2 2.5 3
Non-dimensional spatial shift along the xs-wise direction
(a) (b)
Figure 5.19: Periodicity analysis results - Distributions over the measuring line. (a) Mach number
and vertical dashed lines representing measuring points position. (b) Auto-correlation function for
entropy and Mach number: the unity value 1 corresponds to perfect periodicity.
a qualitative proof of the attained periodicity and it reports also vertical dashed lines preempting
the position of the points that constitute the downstream measuring grid, which is thoroughly
described in Section 5.3.
To quantify the level of periodicity, the auto-correlation function was applied to the distributions
over the measuring line of the most relevant flow properties (such as the Mach number, reported in
Figure 5.19a), allowing to assess the degree of overlapping of the various signals with their delayed
copies as function of the delay. The non-dimensional auto-correlation functions for entropy and
Mach number are reported in Figure 5.19b. As clearly shown, the peak distance, representing the
fluid dynamic pitch, nearly coincides with the geometrical one (deviation lower than 0.6 mm, or
≈ 1.3% of the pitch). Additionally, the two central auto-correlation peak values, which represent
the points of maximum overlapping, are very close to 1 (≈ 0.97). The attained level of periodicity
is further emphasized by the results reported in Figure 5.20, in which the suction and pressure side
pressure distributions related to the three full blades is reported. All these outcomes certify the
desired periodicity between the two central channels, at least for the design inlet conditions.
Since the TROVA facility operates as blow-down wind tunnel, total conditions at cascade inlet
are continuously changing as the HPV empties. Therefore, the full cascade was also tested at
several off-design conditions by means of CFD simulations implementing the numerical set-up
already described, but with different inlet conditions. A discrete number of calculations were
performed by varying 𝑃𝑡 from 25 bar to 5 bar and 𝑇𝑡 from 261.5°C to 230.3°C, following a
typical HPV emptying curve, as found in Cammi et al. (2021a). This analysis was carried out
mainly to assess the flow field periodicity throughout the whole experiment as well as to exclude
the occurrence of phenomena that may lead to misleading measurements, such as separations or
unexpected shock patterns. The results of the off-design analysis are summarized in Figure 5.21, in
93
25
20
Pressure [bar]
15
10
5 Top Blade
Central Blade
Bottom Blade
0
0 0.2 0.4 0.6 0.8 1
Non-dimensional axial position
Figure 5.20: Periodicity analysis results: suction and pressure side pressure distributions associated
with the three complete blades.
1 3 5 7 9 11 13 15 17 19 21
2 4 6 8 10 12 14 16 18 20
2
1.5
Mach
Pt = 5.0 bar || Tt = 503.5 K Pt = 17.4 bar || Tt = 518.8 K

0.5 Pt = 8.3 bar || Tt = 506.6 K Pt = 20.9 bar || Tt = 525.2 K

Pt = 15.2 bar || Tt = 515.7 K
0
-0.5 0 0.5 1 1.5 2 2.5
Non-dimensional pitch-wise distance from bottom blade trailing edge xs
Figure 5.21: Mach distribution over the measuring line for design and off-design conditions
(specified in the legend). Vertical dashed lines indicate measuring points position.
which the Mach number distribution along the measuring line is reported for different inlet total
conditions, representing the whole set of off-design operations numerically tested. The exact values
of off-design inlet total pressure 𝑃𝑇0 and temperature 𝑇𝑇0 are reported in the legend of Figure 5.21.
The Mach number increases as the total pressure reduces and more ideal conditions are attained,
as foreseen by theory and proven experimentally in (Spinelli et al., 2018). Even though the flow
field varies as the HPV empties, periodicity remains extremely satisfactory, as depicted in Figure
5.21 and confirmed by the corresponding autocorrelation-based analysis that revealed a difference
between the fluid-dynamic and geometrical pitches always lower than 1.5 mm (≈ 3.3% of the pitch)
for most relevant flow properties.
Further analyses were made by resorting to 3D numerical simulations of the full cascade layout.
The 3D domain was simulated in order to keep into account and investigate three-dimensional
effects, such as end-wall boundary layers and their interaction with the shock pattern, and to assess
they do not jeopardize the positive outcomes obtained through the 2D analyses. The meshing
procedure for the 3D domain still relied on ANSYS-ICEM CFD but the number of elements greatly
increased (9500000 cells) to guarantee the proper resolution in the endwall boundary layers. The
94
Figure 5.22: Three-dimensional CFD simulation of the full linear cascade: (left) Mach field at the
midspan section and (right) total pressure loss distribution in a plane passing through the measuring
line and perpendicular to the end-walls (along the span-wise direction).
Figure 5.23: Mach distribution over the measuring line for both 2D and 3D numerical domains in
off-design conditions (specified in the legend). Vertical dashed lines indicate measuring points.
numerical set up is identical to the one used for the 2D analyses except for inlet total pressure and
temperature, which have been set equal to 𝑃𝑡 = 9 bar and 𝑇𝑡 = 229°C to numerically investigate
another off-design flow condition occurring during a typical test run. The simulation results are
reported in Figure 5.22, which displays the Mach distribution at the cascade mid-span and the
total pressure loss field in a plane passing through the measuring line and perpendicular to the
end-walls (along the span-wise direction). In addition to the 3 wakes, the loss distribution clearly
highlights end-wall boundary layers and their interaction with the fish-tail shocks originating from
the blade trailing edges. Overall, the results shown in Figure 5.22 prove that 3D effects are of
minor importance also in strongly off-design conditions and do not affect the overall flow field
topology and resulting maximum Mach number values.
To further assess 3D effects, the Mach number distributions at the cascade midpsan and over
the measuring line for both 2D and 3D numerical domains are reported in Figure 5.23 for the same
total inlet conditions. This comparison provides and additional proof that relevant 3D effects are
not present since only very tiny differences between the 2 Mach distributions can be spotted along
the measuring line, mostly in terms of wake position and depth. However, such discrepancies are
sufficiently small not to challenge and jeopardize the validity of the 2D results previously described
or to question the credibility of the design of the experiments.
95
Figure 5.24: Test section layout with overlapped Mach field.
5.3 Measurement strategies

While the COTS instrumentation employed has been reported in Section 4.2, and Section 5.1
describes the methodology devised to retrieve total pressure loss through the cascade, this section
focuses on the measuring strategies developed to characterize the flow field as comprehensively as
possible. The first part reports the number, position, and purpose of the sensors employed while
the second subsection details the design of the total pressure probe, focusing on the associated
difficulties and assessing the probe-flow interaction.
5.3.1 Measurement strategies overview

The detailed investigation of the flow field within the supersonic cascade, particularly in the blade
trailing edge region, and the measurement of total pressure losses are the primary objectives of
the experimental campaign. According to the design procedures described above, the cascade
implements three complete blades and two side-walls, as shown in Figure 5.24, in which the final
cascade arrangement and the Mach field resulting from 2D CFD simulation are overlapped. The
flow delivered by two complete blade channels, including three wakes, is expected to be available
for downstream characterization over two full blade pitches.
As described in Section 5.1, the direct measurement of total pressure downstream of the cascade
𝑃𝑇1 (refer to Figure 5.1 for the subscript notation) is prevented by the formation of a bowed shock
at the probe head. Four additional thermodynamic properties are then required to retrieve 𝑃𝑇1 and,
consequently, the total pressure loss. In particular, since the flow may be considered adiabatic
due to the action of heating system and of insulation, the thermodynamic properties chosen in
the present work are: total inlet conditions, namely 𝑃𝑇0 and 𝑇𝑇0 , total pressure downstream
the probe-induced shock 𝑃𝑇2 and static pressure upstream the shock 𝑃1 . Total conditions 𝑃𝑇0
and 𝑇𝑇0 are measured in the plenum, the settling chamber upstream the test section in which
the flow is smoothly and moderately accelerated such that it can be considered uniform at the
cascade inlet. The total temperature is provided through 2 thermocouples (of J and K type) with
expanded uncertainty of 1°C. Due to the negligible kinetic energy in the settling chamber, 𝑃𝑇0
is measured by a wall flush-mounted absolute piezo-resistive transducer for high-temperature
applications with expanded uncertainty of approximately 0.1% of the full scale. The total pressure
value is double-checked through another similar transducer placed at the end of a pneumatic line
outgoing the test section and, therefore, not directly facing the flow. More generally, all the pressure
96
5.3. Measurement strategies
Figure 5.25: Test section arrangement with overlapped Mach field and measurement strategies:
downstream measuring grid (red dots), total pressure probe sketch (blue body), and in-channel
static pressure taps (yellow dots).
measurements described in the following are performed using the same kind of piezoresistive
transducer for high-temperature applications, described in Section 4.2. Each transducer is either
flush-mounted or connected through a pneumatic-line and characterized by a different full scale
pressure value.
Due to the inevitable non-uniformity, a measurement traversing system is required to characterize
the downstream flow field, including the pitch-wise distribution of total pressure losses. As
previously introduced, a measuring grid parallel to the trailing edge locus and placed at a distance
of 0.17 times the blade chord is therefore adopted. The grid is reported in Figure 5.25 and
features 21 points (red enumerated dots in Figure 5.25) not uniformly spaced along the two central
blade pitches, but conveniently refined close to the wakes where higher gradients are expected.
These points represent the position where the head of an ad-hoc designed total pressure probe is
located (sketched in blue in Figure 5.25) and therefore they constitute the measuring grid where
pressure 𝑃𝑇2 is measured. More specifically, 𝑃𝑇2 is measured using 𝑃𝑇0 as reference, exploiting a
differential pressure transducer (of 0.05% FS expanded uncertainty) connected to the probe by
means of a pneumatic line. As explained in Section 4.2, all the pneumatic lines connecting the
test section with remotely mounted transducers need to be flushed with nitrogen, in order to avoid
line condensation and the consequent unacceptable promptness reduction. Conversely, the static
pressure upstream the shock 𝑃1 is provided by a locally-mounted sensor connected to the flow
through static pressure taps of 0.3 mm diameter, which are machined at the same positions along
the whole measuring grid (red points in Figure 5.25). The probe head, pre-aligned with the flow, is
15 mm long, features a recessed stem, and is located at blade mid-span, while the taps are machined
at the rear wall such that they are located exactly below the probe head tip. The motivations
underlying this choice and a detailed description of the probe-tap configuration are reported in
Section 5.3.2, along with a description of the challenges associated with the total pressure probe.
The total pressure probe access is machined in the steel back plate, as well as the required sealing
and rotation/blade-span translation system.
Thanks to an already proven test repeatability (Cammi et al., 2021a), the probe traversing
is obtained with multiple test runs, with one single measuring point accessed at each run. This
extends the experimental campaign duration, but entails two indubitable advantages. The first one
is to simplify the implementation of the sealing system, since no pitch-wise probe movement is
required. The second one is the chance of exploring a wide range of operating conditions in a
97
single test, from highly non-ideal to almost ideal, during the HPV emptying.
Beside the evaluation of total pressure losses, two main techniques were adopted to investigate
the flow field developing through the cascade, crucial for CFD development and validation. Four
static pressure taps were manufactured on the end-wall plate, along the center-line of the diverging
portion of the two central channels, complemented by an equal number of pressure transducer.
The 8 taps are represented by the enumerated yellow dots in Figure 5.25. The importance of
these sensors is twofold: first, the processed pressure signals provide a quantitative estimation
of the obtained periodicity between the flow developing within the two central channels; second,
they contribute to characterize the flow expansion in the diverging portion of the channel and
in the semi-bladed region, both upstream and downstream the fishtail shock stemming from the
pressure side, as observable in Figure 5.25. Finally, a double-passage schlieren equipment was
also employed to visualize the flow field density gradients. This allows to highlight with an almost
continuous space resolution the pattern of fan/shock waves originated both within the diverging
channels and at blade trailing edges, the morphology of the probe-induced shock structure, and the
waves reflected at the bottom side-wall.
5.3.2 Total pressure probe design

The designed total pressure probe includes an overhung head with a length of 15 mm, which
requires pre-alignment with the flow, as well as a recessed stem for connection to the pneumatic
lines. The probe design presented several challenges related to four main requirements:
• Necessity of local (point-like) measurements and minimization of disturbances to the flow
field;
• A dynamic response fast enough to follow the (quasi-steady) experiment transients originating
from the HPV emptying;
• A suitable structural stiffness to withstand high-pressure loads and avoid stem/head oscilla-
tions;
• A smart configuration to ease plug-in, orientation, and plug-out procedures.
The possibility of triggering oscillations and/or head tip displacements when a probe is subject to
high aerodynamic loads was noticed in previous experimental campaigns carried out at CREALab
and described in Conti (2021). While deformations of the probe head and/or stem can be
identified through post-experiment inspections, the acquired schlieren videos are required to easily
recognize probe oscillations. However, the acquired pressure signals can be exploited to confirm
the oscillations and to assess if they affect the measured total pressure. Figure 5.26a reports a
comparison between 2 signals acquired in a previous experimental campaign (Conti, 2021): the
total pressure retrieved with a steel probe and the static pressure measured by a reference pressure
tap. The comparison with static pressure (in orange) allows to gather that the low-frequency
oscillations characterizing the total pressure signal (in blue) are produced by the mechanical
displacements of the probe head and not by electronic disturbances or flow field unsteadiness.
This outcome has been also confirmed by the evident fluctuations of the probe that can be spotted
analyzing the acquired schlieren visualisations. As also depicted in Figure 5.26a, probe oscillations
and displacements may remarkably jeopardize the acquired measurements, leading to modifications
of the dynamic pressure retrieved by the probe and to an increase of measurement uncertainties.
To guarantee resistance and rigidity, a cobalt-chrome alloy probe produced by means of
stereolithography (SLA) was considered. This material is characterised by a relatively high
hardness and, additionally, the employment of stereolithography provides a greater design freedom.
It enables to include reduced-size details that would be challenging to machine through traditional
98
(a) (b)
Figure 5.26: Comparison between static pressure measured by a reference tap and the total pressure
acquired by: (a) a steel probe employed in Conti (2021) and (b) the devised cobalt-chrome alloy
probe.
manufacturing technique. Conversely, the SLA technique has the drawback of limiting the minimum
dimensional tolerances achievable (around ±0.1 mm). The devised probe was tested in preparatory
experiments and it did not show any oscillations or deformations independently on the inlet
conditions investigated (up to 25 bar). A prove of this outcome is provided in Figure 5.26b, in
which the total pressure signal retrieved by the probe is compared with static pressure measured by
a reference tap, analogously to the plot shown in Figure 5.26a. The SLA probe indeed provides a
much more smooth signal, strongly reducing the oscillations observed in Figure 5.26a for the steel
probe.
The internal and external diameters of the probe head represent critical parameters due to
the impact they have on flow disturbances and measurement resolution. This is the reason why
the choice of their values was carried out to meet all the four requirements previously listed. In
particular, the probe external diameter was chosen equal to 1.6 mm as a compromise between
instrument stiffness, spatial resolution, and minimum probe encumbrance, in order to avoid
mechanical displacements/deformations and concurrently limiting the disturbances on the flow
field. Conversely, the chosen value of 1 mm for the internal diameter resulted mainly from a
trade-off between ensuring a fast-enough dynamic response and reducing the integration area as
much as possible. For this reason a dedicated dynamic characterization of the remotely mounted
measuring system - including pneumatic lines, transducers, and the probe - was carried out.
The dynamic calibration consists in two steps: firstly the pneumatic line between the probe
head tip and the remotely mounted differential transducer exploited to retrieve the total pressure
(using 𝑃𝑇0 as reference) is pressurized, secondly the electric valve closing the pneumatic circuit is
opened, allowing the release of the accumulated air. Comparing the electrical step signal provided
to open the valve with the pressure measured by the transducer during the pneumatic line emptying
process is possible to characterize the dynamic response of the measuring system. The emptying
phase is chosen for the dynamic characterization since during actual test with MM pneumatic lines
are flushed (so filled) with nitrogen to avoid MM condensation - as detailed in Section 4.2.2 - and
therefore the flow outgoes the lines as the inlet total pressure decreases due to the HPV dynamics.
The dynamic calibration has been carried out considering different lengths and diameters of the
pneumatic line as well as different probes, including 2 cobalt-chrome SLA probes with 0.6 mm and
1 mm internal diameter respectively and a 0.6 mm internal diameter steel probe. While both the
diameter and the length (reduced as much as possible) of the pneumatic line resulted to have very
limited effects, the probe turned out to be the component most affecting the dynamic promptness
of the pneumatic system. This can be explained by the small values of the probe internal diameter,
which represents a bottleneck for the outgoing flow, especially in the proximity of the 90° elbow
where an excessive shrinking is present, even worsened by the inaccurate dimension tolerances for
99
Figure 5.27: Pneumatic lines dynamic responses to an input step for three different probes.
the SLA probes. The dynamic response of the pneumatic line implementing the three probes is
compared with the input step in Figure 5.27, in which the pressure measured by the transducer 𝑃 is
normalized (Amplitude: (𝑃 − 𝑃 𝑎𝑚𝑏 )/(𝑃𝑚𝑎𝑥 − 𝑃 𝑎𝑚𝑏 )) and plotted versus the time following the
valve opening.
Figure 5.27 highlights that, for a fixed internal diameter, the dynamic response associated to
the cobalt-chrome probe is much slower compared to the steel probe. This is due to the inaccurate
dimension tolerances of the stereolithography technique, which produces an internal diameter that
is actually lower than 0.6 mm, especially in the proximity of the 90° elbow. Although preferable in
terms of dynamic promptness, the steel probe cannot be implemented in high-pressure experiments
with MM, since it is not enough stiff to withstand the large aerodynamic loads without causing
oscillations and displacements of the head tip. The input step and the normalized measured signals
can be used to model the dynamic system. In particular, different analytical transfer functions,
approximating the system response for the three probes, have been computed exploiting the input
and output time-domain signals and their sampling rate. The analytical transfer functions revealed
that the 99.9% cut-off frequency results equal to about 𝑓𝑐,𝑜 𝑓 𝑓 = 1.3 Hz and 𝑓𝑐,𝑜 𝑓 𝑓 = 0.2 Hz for
the pneumatic systems implementing the SLA probes with 1 mm and 0.6 mm internal diameter
respectively. These outcomes lead to the choice of the 1 mm head probe, being its 𝑓𝑐,𝑜 𝑓 𝑓 larger
than the typical frequency content of the HPV emptying dynamics, for which the 99.995% of the
pressure signal energy content is always contained below 0.1 Hz. This factor 10 margin between
the cut-off frequency and the typical energy content of the HPV emptying dynamics is indeed
considered acceptable. Conversely, the choice of the 0.6 mm internal diameter probe would have
produced an unacceptable drop in the cut-off frequency, which results comparable to the frequency
content of the experiment dynamics.
The 3D model of the final cobalt-chrome probe design is reported in Figure 5.28 where some
details have been highlighted. The probe features a cylindrical shape characterised by an external
diameter of 3 mm which is narrowed to 1.6 mm in the part of the probe interacting with the flow
field. A rib is included in the point where the probe presents the 90° elbow, in order to stiffen this
region and concurrently reduce the mechanical stresses. Following the dynamic analyses previously
described the internal diameter is set to 1 mm. In the probe stem a small cut, highlighted in Figure
5.28, is designed to ensure the presence of a flat surface. This insertion is indeed beneficial during
the probe fastening process and concurrently it eases the probe-related operations before the test
run, including the pre-alignment with the flow and the positioning at the cascade mid-span.
100
Figure 5.28: 3D model of the final cobalt-chrome total pressure probe, with details on the probe
stem and head.
As introduced in Section 5.3.2, 𝑃1 is measured through static pressure taps machined in the rear
end-wall plate at the 21 positions depicted in Figure 5.25 and exactly below the total pressure probe
head tip. This configuration is schematically represented in Figure 5.29a, in which the abscissa 𝑥
represents the axial distance from the probe head tip and the ordinate 𝑦 is the span-wise distance
from the rear end-wall, the test section height being equal to 18.7 mm. In Figure 5.29a the probe
is sketched in black and the static pressure tap is represented by a yellow circle. To retrieve the
correct value of total pressure downstream of the cascade (upstream of the probe-induced shock), it
is of primary importance to assess if the wall measured value is fully representative of the static
pressure upstream the bowed shock. The representativeness is indeed theoretically supported by the
boundary layer theory and by the fact that a planar cascade is involved, meaning that no span-wise
pressure gradient are expected in the clean flow region. Moreover, the flow is supersonic, meaning
that disturbances cannot be propagated upstream except for those within the boundary layers.
However, the formation of a bowed shock at the probe head may affect the pressure measured at the
wall if the shock-end wall boundary layer interaction occurs sufficiently close to the wall tap, such
that the distance travelled upstream by the disturbances is enough to modify the static pressure read
by the static tap. The model developed by Moeckel Moeckel (1949) was firstly applied to estimate
the bowed shock detachment distance from the probe head, its inclination in the oblique shock
region, and, consequently, the intersection point between the shock and the rear end-wall. The
results, reported in Figure 5.29a for different values of the shock-upstream Mach number, seems
to exclude the possibility of an interaction between the bowed shock and the retrieved wall static
pressure. In fact, the Mach number at the cascade measuring station (i.e. upstream the bowed
shock) is about 1.6 for the design conditions and increases when the inlet total pressure continues
decreasing as the test proceeds on (see also Figure 5.21).
To further assess the effectiveness of the measured wall pressure in representing the one
upstream the shock, a numerical investigation relying on CFD simulations was carried out. Some
assumptions were needed to ease the meshing procedure and to investigate the physical problem
without considering the full three-dimensional cascade layout with the total pressure probe as
numerical domain. Therefore, a 3D prismatic domain was considered, characterized by a square
cross section of 18.7 mm edge (equal to the cascade height) extruded in the axial, or stream-wise,
direction for 120 mm. For simplicity the probe was approximated by a cylinder parallel to the
prism, with an external diameter of 1.6 mm, and placed in the middle of the domain. Exploiting the
double symmetry of the problem to lighten the computational effort, only a quarter of the numerical
101
(a) (b)
Figure 5.29: (a) Schematic representation of probe and static pressure tap (yellow cicrle)
configuration. Bowed shock shape and location according to Moeckel (1949) as a function of the
upstream Mach. (b) Mach distribution in the section defined by axial and span-wise direction
and passing from the probe center. Comparison between the bowed shock resulting from CFD
simulation and Moeckel’s model.
domain was actually meshed leading to a total number of cell elements of about 1 M. Calculations
were performed using the ANSYS-Fluent solver and the same numerical set up described above.
To set the boundary conditions (in terms of inlet total pressure and temperature, flow angle, and
turbulence) the results of the planar cascade numerical simulations have been exploited, extracting
the quantities of interest in one of the measuring points (red dots in Figure 5.25). The resulting
Mach number field, in the section defined by axial, or stream-wise, and span-wise direction and
passing from the probe center, is shown in Figure 5.29b. As it can be observed, the numerical
results prove the lack of any interaction between the bowed shock and a pressure tap placed below
the total pressure head tip (same axial position). In fact, the resulting pressure at the wall is equal
to the one just upstream the probe-induced shock, confirming the possibility of using the measured
pressure to retrieve 𝑃𝑡1 . In Figure 5.29 the bowed shock shape predicted by Moeckel theory for
the exact value of the Mach number is also reported and compared with 3D high-fidelity results,
highlighting a satisfactory agreement in terms of shock detachment distance from the probe head.
On the contrary, some discrepancies concerning the bowed shock opening are observable moving
far away from the normal shock region.
5.4 Test section final arrangement

In the previous sections of this chapter the concept of the experiment was presented together with a
detailed description of the design of the main components and of the devised measuring strategies
to characterize the flow field throughout the test run. This section reports an overview of the final
test section arrangement required to combine the designed components and the devised measuring
techniques. The final layout needs also to comply with the space and geometric constraints
dictated by the TROVA facility and concurrently to facilitate an easy plug-in/plug-out of various
components, including the probe.
Figures 5.30 and 5.31 report the front and rear isometric views of the test section assembly
respectively. The linear cascade geometry previously described, including 3 complete blades and
2 side-walls, is clearly recognizable in Figure 5.30. The cascade is closed on the front and rear
102
5.4. Test section final arrangement
Figure 5.30: 3D model of the test section assembly: front view
Figure 5.31: 3D model of the test section assembly: rear view
103
Static Tap
Pressure
Transducer
Holes
Threaded
Holes
Probe Holes
(a) (b)
Figure 5.32: 3D model of the probe insert: (a) view of the surface facing the flow (b) rear view.
side through planar end-walls. On the rear steel plate, blades and sidewalls are dowel-fixed and
pressure taps and instrument accesses are machined, whose other-side end part is shown in Figure
5.31. The plate is mirror polished and the front end-wall is made by a quartz window (not shown in
Figure 5.31), allowing schlieren visualization in double-passage configuration. Main dimensions
characterizing the cascade are a pitch of 45 mm, a blade chord of 66.5 mm, and a blade height of
about 18.7 mm. The passage throat width in the blade-to-blade plane is approximately 5.5 mm.
The presence of an O-ring is needed around the test section, as proved by its proper cave in Figure
5.30, to avoid gas leakages from and toward the ambient environment.
As shown in Figure 5.31, the available space in the back-side of the test section is limited,
especially downstream of the cascade, where many measurement points are required to ensure
an acceptable space resolution of the total pressure loss pitch-wise distribution. This prevents
the possibility of machining in the rear side end-wall as many static pressure taps and probe
access holes as required for an acceptable spatial resolution. The minimum distance between
two consecutive measurement positions is indeed constrained by the dimensions of the O-ring
groove ensuring gas tightness around the probe stem, which features a diameter of around 8 mm.
Therefore, the holes and taps cannot be manufactured directly in the test section rear plate and the
design of an additional component that satisfies the following requirements is mandatory:
• Allow for as many measurement points as possible;
• Facilitate an easy variation of the probe position in between different experiments;
• Include both the static pressure taps to measure 𝑃1 and the holes to host the probe. In
particular the tap has to be machined exactly below the probe head tip for each position
where the total pressure need to be measured (as described in Section 5.3.2).
The component chosen to meet these requirements is the insert reported in Figure 5.32, in which
both the side of the insert facing the flow (a) and the rear view (b) are shown. This insert needs to
be embedded in the test section, in which a suitable cave is therefore machined (whose perimeter
on the cascade side can be spotted in Figure 5.30 together with the insert surface). While each
insert can accommodate a limited number of measuring point accesses (up to 5), the use of such
components theoretically enables the possibility of an infinite number of measuring positions.
This can be achieved by manufacturing multiple inserts, each of them characterized by a distinct
distribution of the access points. To limit the costs and the duration of the test campaign, 5 different
inserts are designed, leading to the 21 measuring points reported in Figure 5.25.
In Figure 5.32a the five holes that host the probe can be easily spotted, while the zoomed
detail highlights the corresponding static pressure tap of 0.3 mm diameter. In the rear view of the
insert (Figure 5.32b) the static pressure holes are more clear since their diameter is enlarged to 3
mm in the last part to allow the fastening of the pressure transducers. As shown in Figure 5.32,
104
(a) (b)
Figure 5.33: (a) Ad-hoc designed filling stem and (b) Section view of one static pressure tap
channel.
the probe access holes and the corresponding static pressure taps have a misalignment of 74°, in
order to guarantee the probe head tip to be exactly above the tap when pre-aligned with the flow.
Since multiple holes are machined in each insert while a single total pressure probe is employed, it
becomes necessary to cover the empty holes during a test run. This is to prevent any leakages and to
avoid undesired discontinuities in the surface facing the flow, which would lead to additional shock
patterns and disturbances to the flow. For this purpose an ad-hoc designed filling stem, shown in
Figure 5.33a, is developed. As previously anticipated, the test section plate is mirror-polished to
allow for high-quality double-passage schlieren visualizations and therefore the side facing the
fluid needs to undergo a lapping procedure, together with the all the inserts and fitting stems.
Similarly to the downstream total pressure measurements, also the maximum number of
in-channel pressure taps (yellow dots in Figure 5.25) is limited by the space required to assemble
the corresponding transducers in the rear side of the test section, as represented in Figure 5.31. In
order to incorporate 4 static pressure measurements within the diverging portion of the two central
channels, two triangular extrusions are necessary on the rear surface of the test section. These
extrusions, clearly visible in Figure 5.31, allow to accommodate all the corresponding 8 transducers
simultaneously. As a drawback, a conventional design of the pressure tap channels is not feasible,
as depicted in the section view of one channel shown in Figure 5.33b, leading to a more complex
manufacturing process. A design of straight tap channels is indeed prevented by the presence of
the probe insert, whose position is defined by the stagger of the cascade. Three different sections
of the pressure tap channel can be recognized in Figure 5.33b: firstly, a 0.3 mm diameter hole
perpendicular to the flow surface to avoid the measurement of any fraction of the dynamic pressure;
secondly, a tilted hole of 1.5 mm diameter to facilitate the manufacturing process; and finally, a 5
mm diameter final section that accommodates the head of the pressure transducer.

This chapter reports the full process of experiment design, which substantially differs from the
case of conventional blade cascades, mainly due to challenging operating conditions and the lack
of validated measuring techniques. With the aim of generating benchmark data, the experiment
arrangement was conceived to match multiple requirements. Designing a cascade geometry
representative of actual high-temperature and high-expansion ratio ORC turbines was deemed key,
including the measurement of total pressure losses. An important requirement is also to operate the
105
blade row with a fluid at thermodynamic conditions relevant for state-of-the-art high-temperature
ORC power plants. Another objective of the design of the experiment is to devise a set of
measurement strategies to have a characterization of the flow field as detailed as possible. Therefore
all the design choices and measurement strategies have been devised to meet such requirements and
to concurrently comply with the reduced test section dimensions (270 mm × 250 mm rectangular
housing) and TROVA facility operations, seeking solutions as simple as possible.
The linear cascade is designed to operate with the siloxane MM (hexamethyldisiloxane,
C6 H18 OSi2 ) and features 3 complete blades, defining two converging-diverging central channels,
together with two side-walls integrating two partial profiles. The number of blades allocated
into the cascade is the maximum amount that allows to comply with the reduced dimensions of
the test section. A further increase of the number of blades would indeed imply an excessive
miniaturization of blades and instrumentation, which would not be able to withstand the large
aerodynamic loads involved. To be consistent with the application considered, the blades are
designed to provide an outlet Mach number of 1.6 and a stagger angle of ≈ 75°is selected for the
cascade, leading to a geometric pitch of 45 mm. The linear cascade is designed to have parallel
inlet and outlet flow, thus not producing any flow deflection. This design choice not only notably
simplifies the cascade assembly in the TROVA test section, but also turns out to be representative
of radial inflow and axial nozzles, as proven by the analysis reported in Section 5.2.1. Total inlet
conditions in both design and off-design operation are chosen to carry out expansion processes
featuring highly non-ideal fluid thermodynamic states. Therefore, the design inlet total pressure
and temperature have been set to 𝑃𝑇0 = 25 bar and 𝑇𝑇0 = 261.5°C respectively, corresponding to
an inlet compressibility factor 𝑍 ≈ 0.3.
To design and optimize both the blade and side-walls, high-fidelity CFD simulations are
systematically applied, exploiting a numerical model developed in-house and based on the
commercial ANSYS-CFX and ANSYS-Fluent solvers. The baseline blade profile derives from a
reference axial ORC stator investigated in Romei et al. (2020), whose leading edge and pressure
side have been modified to avoid flow separations when operated in a linear cascade with null flow
deflection. In a conventional annular configuration, no flow deflection corresponds to operate the
cascade with highly negative incidence, which in turn may lead to large flow separation regions if
the front part of the blade is not properly designed. The results reported in Section 5.2.1 prove the
effectiveness of modifications applied, since the baseline profile does not present any separation
when operated at high negative incidence (no flow deflection). A dedicated analyses concerning the
inlet flow angle has been then conducted to assess whether the use of a null-deflection configuration
could jeopardize the effectiveness of the linear cascade in representing a statoric annular row of an
axial ORC turbine. The results reported in Section 5.2.1 highlight that the flow field downstream
the throat section and the entropy production through the cascade do not depend on the specific
value of the inlet angle set as boundary condition. The baseline blade is indeed highly tolerant to
the severe change of incidence angle, due to the large radius of curvature at the leading edge. These
outcomes prove the effectiveness of the concept of the experiment and confirm the null-deflection
linear cascade as representative of supersonic stators of both axial and radial-inflow turbine.
Subsequently, a shape-optimization process has been performed to refine the baseline blade shape
in the rear part of the channel, especially from the throat section to the outlet, as most of the
entropy production in supersonic cascades occurs in this region. The optimization is performed
by applying an in-house shape-optimization tool, which combines high-fidelity CFD calculations
and a surrogate evolutionary strategy, formulated in constrained, multi-point and multi-objective
fashion. The baseline blade is parametrized using 30 control points (CPs), and only six of them,
covering most of the blade suction side, are considered movable during the optimization process.
The optimizations procedure converged after 108 cascade CFD evaluations, leading to a reduction
of the entropy rise through the cascade of the 16%. The optimal blade is indeed characterized by
106
lower Mach number values within the supersonic diverging portion of the channel and features a
milder pressure drop up to the design outlet conditions, leading to weaker fishtail shocks. Since
the optimal blade is characterized by a very thin tail, being the thickness lower than 3 mm for
a non-negligible portion of the profile up to 0.5 mm at the trailing edge, its structural integrity
and the expected deformation have been assessed through dedicated FEM analyses. The resulting
maximum displacement, occurring at the trailing edge, is 0.09 mm, while the maximum stress
is around 65 MPa and occurs in proximity of the hole hosting the dowel-clamping mechanism
adopted to fasten the blade with the test section. As demonstrated by numerical simulations of
the deformed geometry, the deformation field caused by the blade pressure distribution does not
produce any remarkable variation on the resulting flow field.
To attain a high level of periodicity and to minimize disturbances at the downstream measuring
positions, the design of side-walls was carried out reproducing the flow streamlines simulated in
CFD calculations. The effectiveness of this choice is proven in Section 5.2.2, in which CFD-based
analyses of the full linear cascade, aiming at fully characterizing the flow field, are detailed. The
flow fields predicted by numerical simulations prove the generation of wakes and trailing edge
shock systems, highlighting the absence of detached flows and large recirculation regions. In
addition to the qualitative investigations of the numerical results, the auto-correlation function has
been applied to the flow properties distribution downstream the cascade, in order to assess the
attained level of fluid-dynamic periodicity. This analysis confirms a remarkable flow periodicity,
demonstrating that the fluid dynamic pitch nearly coincides with the geometrical one (deviation
lower than 0.6 mm, or ≈ 1.3% of the pitch) and that the degree of overlapping is very close to 1
for all the fluid properties investigated. Since total conditions at cascade inlet are continuously
changing as the HPV empties during the experiments, the full cascade has been numerically
investigated also at several off-design conditions, ranging from 𝑃𝑇0 = 25 bar up to 𝑃𝑇0 = 5 bar. A
periodicity of the downstream flow field is confirmed also in off-design conditions, as proven by
the results reported in Section 5.2.2, the reflection waves generated at the bottom side-wall being
the major source of non-periodicity.
The measuring strategies have been devised to characterize the flow field as comprehensively
as possible, still complying with the limited amount of available space, resulting from the reduced
dimensions of the test section. Following the linear cascade design just described, the flow
delivered by two complete blade channels, including three wakes, is available for downstream
characterization over two full blade pitches. Due to the inevitable non-uniformity, a measurement
traversing system is required to characterize the downstream flow field, including the pitch-wise
distribution of total pressure losses. A measuring grid parallel to the trailing edge locus and placed
at a distance of 0.17 times the blade chord has been therefore designed, exploiting the results of the
CFD simulations over the full linear cascade. The grid features 21 points (red enumerated dots in
Figure 5.25) not uniformly spaced along the two central blade pitches, but conveniently refined
close to the wakes where higher gradients are expected. These points represent the position where
the head of the total pressure probe is located. Due to the formation of a bowed shock in front of
the probe head tip, total pressure loss assessment through the cascade demands for the concurrent
measurements of the inlet total conditions 𝑃𝑇0 , 𝑇𝑇0 , of the static pressure downstream the cascade
𝑃1 , and of the total pressure at the probe head 𝑃𝑇2 . The static pressure upstream the shock 𝑃1 is
provided by a locally-mounted sensor connected to the flow through static pressure taps of 0.3 mm
diameter, which are machined at the same positions along the whole measuring grid. The probe
head, pre-aligned with the flow, is 15 mm long, features a recessed stem, and is located at blade
mid-span, while the taps are machined at the rear wall such that they are located exactly below
the probe head tip. Dedicated analyses are described in Section 5.3.2, which aim to assess if the
static pressure retrieved at the rear wall below the probe head tip is representative of the pressure in
front of the probe-induced bowed shock. The outcomes of both theoretical (Moeckel, 1949) and
107
numerical analyses prove that the probe-induced shock does not interact with the static pressure
tap, whose measurement results therefore representative of the pressure 𝑃1 . Thanks to an already
proven test repeatability (Cammi et al., 2021a), the probe traversing is obtained with multiple test
runs. Although this choice extends the experimental campaign duration, it allows to explore a wide
range of operating conditions in a single test, from highly non-ideal to almost ideal, during the
HPV emptying.
As detailed in Section 5.3.2, the recessed-stem probe design is the results of a trade-off between
probe stem stiffness, pneumatic line dynamic promptness, and the necessity of local (point-like)
measurements. To guarantee resistance and rigidity, a cobalt-chrome alloy probe produced by means
of stereolithography (SLA) has been chosen. Probe oscillations and deformations have been indeed
observed in previous campaigns implementing a conventional steel probe. Similarly, the probe
external diameter has been set to 1.6 mm in order to avoid mechanical displacements/deformations
and concurrently limiting the disturbances on the flow field. Conversely, the chosen value of 1
mm for the internal diameter resulted mainly from a trade-off between ensuring a fast-enough
dynamic response and reducing the integration area as much as possible. The dedicated dynamic
calibration revealed indeed that the 99.9% cut-off frequency is equal to about 𝑓𝑐,𝑜 𝑓 𝑓 = 1.3 Hz and
𝑓𝑐,𝑜 𝑓 𝑓 = 0.2 Hz for the pneumatic systems implementing the SLA probes with 1 mm and 0.6 mm
internal diameter respectively. These outcomes lead to the choice of the 1 mm head probe since the
99.995% of the pressure signal energy content for the HPV emptying dynamics is always contained
below 0.1 Hz.
Beside the evaluation of total pressure losses, four static pressure taps are include along the
center-line of the diverging portion of the two central channels, to characterize the flow expansion
within the channels and in the semi-bladed regions. Finally, a double-passage schlieren equipment
was also employed to visualize the flow field density gradients, evidencing the fan/shock wave
structures originated both within the diverging channels and at blade trailing edges, the morphology
of the probe-induced shocks, and the waves reflected at the bottom side-wall.
108
CHAPTER 6
Experimental Campaigns
6.1 Nitrogen experimental campaign

This section documents the first long-duration experimental campaign conducted, consisting in
preliminary tests performed by operating the facility with nitrogen as working fluid. The campaign
entails a comprehensive set of experiments that systematically explore all the aspects related
to the objectives described in Section 5.1, the execution of a single test run being much faster
and easier exploiting nitrogen as working fluid. Tests with nitrogen are therefore carried out to
assess the effectiveness of the designed cascade in reproducing the flow field of actual supersonic
turbine stators, both in terms of expansion-fan and shock-wave patterns and of cascade periodicity.
Moreover, such experiments are also useful to test the instrumentation and the measurement
technique conceived for computing total pressure losses from acquired pressure signals.
Around 60 tests were performed to investigate different configurations, in which the total
pressure probe is either absent or placed in one of the 21 downstream measuring positions. For
the label and the location of the measuring points the reader is referred to Figure 5.25, which has
been reported also in this chapter for the sake of clarity (Figure 6.1). Experiments without the
probe are of outstanding importance to assess and characterize the flow morphology, including the
several configurations of fan/shock waves, without the disturbances arising from the introduction of
a solid body (the probe) in a supersonic flow. Most of the experiments are carried out at moderate
inlet total pressure (𝑃𝑇0 ≈ 4 bar) in order to reduce nitrogen consumption and, consequently,
the cost of the campaign. This choice does not involve any loss of generality since main flow
features, such as shock patterns and Mach number spatial distribution, do not depend on the
absolute value of total pressure but mainly on cascade geometry. This is because, under the
thermodynamic conditions characterizing the tests, nitrogen can be considered a perfect gas with
excellent approximation. Supersonic flows are indeed established also for moderate inlet total
pressures since the downstream pressure can be reduced to values as low as 50 − 80 mbar, allowing
to overcome the critical pressure ratio, as explained in Section 4.1. However, some experiments
109
Chapter 6. Experimental Campaigns
Figure 6.1: Test section arrangement with measuring positions: downstream measuring grid (red
dots), total pressure probe sketch (blue body), and in-channel static pressure taps (yellow dots).
were carried out varying the inlet total pressure to check for any unexpected difference in the flow
field produced by such a variation. The pressure signals and the schlieren videos acquired from
nitrogen experimental campaign are exploited to investigate four important features: repeatability
of tests, probe-induced disturbances on the flow, fluid-dynamic periodicity between the two central
channels, and agreement of experimental data with CFD results. All these aspects are individually
addressed in the following subsections.
6.1.1 Experiments repeatability

Verify the repeatability of the experiments, with and without the probe, is of utmost importance
since the characterization of total-pressure loss pitch-wise distribution relies on this assumption,
the latter being constructed by combining different tests. Since an accurate control of the total
pressure at the test section inlet is not possible, there are differences in the values of the inlet
total pressure between different tests in the order of about ±0.5 bar. This makes meaningless the
comparison of acquired pressure signals from different experiments as function of time. More
relevant insights may be instead obtained comparing the ratios between the acquired pressures and
the inlet total pressure 𝑃𝑇0 as function of a suitable non-dimensional value (𝑃𝑇 ,𝑟 ) of the inlet total
pressure itself, defined as
𝑃𝑇0,𝑚𝑎𝑥 − 𝑃𝑇0
𝑃𝑇 ,𝑟 = , (6.1)
𝑃𝑇0,𝑚𝑎𝑥 − 𝑃𝑇0,𝑚𝑖𝑛
where 𝑃𝑇0,𝑚𝑎𝑥 and 𝑃𝑇0,𝑚𝑖𝑛 are , respectively, the maximum and minimum values of the inlet total
pressure 𝑃𝑇0 occurring during each test run.
First of all, to verify the signals repeatability for the 8 in-channel static pressure taps (yellow
labeled circles in Figure 6.1), 3 experiments without the total pressure probe were considered,
in the following referred to as TNP1, TNP2, and TNP3. The results in terms of pressure ratios
against 𝑃𝑇 ,𝑟 are reported in Figure 6.2. From left to right, data are ordered from the most upstream
to the most downstream tap, having in the first and second rows the signals from the upper and
lower channels respectively. It is worth to notice that 𝑃𝑇 ,𝑟 increases as the test proceeds (HPV is
emptying and 𝑃𝑇 ,𝑟 → 1) and therefore the plots in Figure 6.2 has to be read from left to right. The
considerable increase of pressure ratios - and consequently of acquired static pressures - at the
end of the experiment can be attributed to the continuous rise in the back-pressure of the facility.
110
6.1. Nitrogen experimental campaign
Test TNP1 Test TNP2 Test TNP3

Tap 1T Tap 2T Tap 3T Tap 4T
0.9 0.5 0.5 0.4
0.35
0.4
0.8 0.4
0.3
0.3
0.7 0.3 0.25
0.2
0.2
0.6 0.2
0.1
0.15
0.5 0.1 0 0.1

P/PT0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.1 0.2 0.3 0.4
Tap 1B Tap 2B Tap 3B Tap 4B
0.9 0.5 0.4 0.4
0.35
0.8 0.4 0.3
0.3
0.7 0.3 0.2 0.25
0.2
0.6 0.2 0.1
0.15
0.5 0.1 0 0.1

0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.1 0.2 0.3 0.4
(PT0,max - PT0)/(PT0,max - PT0,min)
Figure 6.2: Comparison of static pressure signals acquired by in-channel taps for 3 different tests
without total pressure probe.
This causes the shocks to migrate upstream and enter the bladed region once the critical ratio is
exceeded, leading to flow unchoking and to the consequent establishment of a completely subsonic
regime.
As proved in Figure 6.2, the pressure ratios retrieved from every tap position are not only
perfectly overlapped, indicating an outstanding repeatability, but they are also constant as the test
proceeds on and the inlet total pressure decreases (𝑃𝑇 ,𝑟 → 1). This was an expected outcome for
those taps located along the centerline of the diverging part of the 2 central channels, upstream
of the fishtail shock originating from the blade trailing edges, namely positions from 1 to 3 as
shown in Figure 6.3, in which a schlieren frame from test TNP2 is reported. In this region an
almost isentropic flow is established, resembling the flow field through converging-diverging
nozzles, in which the value of 𝑃/𝑃𝑇0 does not depend on the absolute values of 𝑃𝑇0 but only on
the geometrical area ratio. However, the same results - constant values of the ratio 𝑃/𝑃𝑇0 - can be
observed also for pressure taps 4T and 4B, indicating that the pressure rise occurring through the
fishtails do not depend on the absolute value of inlet total pressure. This outcome is confirmed also
by the acquired schlieren videos, depicting a quasi-steady position of the fishtail shocks which
do not vary their opening angle as the test proceeds except for very small oscillations due to the
intrinsic unsteadiness of the physical process.
To assess the repeatability when the total pressure probe is installed, at least two tests are
performed for all the 21 downstream measuring positions (red circles in Figure 6.1). In particular,
the pressure signals acquired in different experiments by the probe and by the static pressure tap
placed below the probe head tip are considered and compared. The results of such analysis are
reported in Figure 6.4, where the static and total pressure acquired during tests TP3, TP10, and
TP21 (probe in position 3, 10, and 21 respectively) are compared with those acquired during the
repetitions of those experiments (tests TP3bis, TP10bis, and TP21bis). As previously motivated,
the ratios between the measured and the inlet total pressure are plotted against a non-dimensional
pressure ratio, this time defined as (𝑃𝑇0,𝑚𝑎𝑥 − 𝑃𝑇0 )/𝑃𝑇0,𝑚𝑎𝑥 . Figure 6.4 highlights the good
repeatability featured by both the static and total pressure signals, regardless of the considered
measuring position. Also in this case the ratio 𝑃/𝑃𝑇0 results constant for the static pressure, while
111
Figure 6.3: Frame from a schlieren visualization of a nitrogen test without the probe (TNP2).
Total Pressure Static Pressure Total Pressure Repetition Static Pressure Repetition
Test TP21 & TP21bis Test TP10 & TP10bis Test TP3 & TP3bis
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4

P/PT0
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.1 0.2 0.3 0.4
(PT0,max - PT0)/PT0,max
Figure 6.4: Total and static pressure measurements downstream the cascade for 3 different probe
positions (3, 10, and 21) and corresponding test repetitions.
an extremely small (negligible) variation as 𝑃𝑇0 decreases can be observed for the ratio of measured
total pressure (i.e. the one downstream the probe-induced shock) over inlet total pressure, 𝑃𝑇2 /𝑃𝑇0 .
Although not reported here, very similar results were obtained for the other measuring positions.
6.1.2 Probe disturbance effects

The assessment of possible probe disturbances on the static pressure measured by the tap placed
below its head tip is mandatory to validate the theoretical and numerical results reported in Section
5.3.2. This verification is indeed a crucial aspect to corroborate the strategy devised to retrieve the
total pressure downstream of the cascade. To this end, 10 experiments without the total pressure
probe are considered, as the downstream static pressures measured in these tests encompass all
the 21 positions that constitute the measuring grid. The pressures measured in each downstream
position can be compared to the corresponding ones acquired in the experiments implementing the
total pressure probe, namely tests from TP1 to TP21. As in the previous section, the label TP𝑖
112
Static Pressure with the Probe Static Pressure without the Probe
Test TP1 Test TP6 Test TP9
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
P/PT0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Test TP13 Test TP17 Test TP21
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
(PT0,max - PT0)/(PT0,max-PT0,min)
Figure 6.5: Static pressure measurements downstream the cascade with and without implementing
the total pressure probe for 6 different measuring positions (1, 6, 9, 13, 17, and 21).
indicates a test with the probe in the 𝑖-th downstream measuring position (see Figure 6.1). The
results of such a comparison are shown in Figure 6.5, showing the pressure ratio 𝑃/𝑃𝑇0 against
𝑃𝑇 ,𝑟 for a subset of the downstream measuring positions (TP1, TP6, TP9, TP13, TP17, and TP21).
This selection can be considered a representative benchmark as it encompasses measurements in
both the wake and middle-channel.
As highlighted in Figure 6.5, the static pressures retrieved by the downstream taps remain
unchanged in the experiments implementing the total pressure probe. This unequivocally demon-
strates that the probe has no influence on the static pressure measured at rear end-wall plate, thereby
confirming its effectiveness in representing the static pressure upstream of the bowed shock.
6.1.3 Flow field periodicity

As detailed in Section 5.2, a primary target of the test section design was to obtain a satisfactory
periodicity between the two central channels when the cascade is operated with the organic fluid
MM. However, it is worthwhile to assess the degree of periodicity also when the selected working
fluid is nitrogen. To this end, the experimental data from the 10 tests without the probe are
considered, in which both the in-channel static pressures (taps 1T, 2T, 3T, 4T, 1B, 2B, 3B, and
4B in Figure 6.1) and the static pressures downstream of the cascade (with a particular focus on
measuring positions 1-11-21, 2-12, 3-13, 4-14, 5-15, 6-16, 8-17, and 10-19) are measured. Note
that measuring positions belonging to the same set (e.g. 1-11-21) are simultaneously investigated
in a single test run since they have been intentionally designed to be geometrically periodic.
The results are shown in Figure 6.6, in which the different pressure ratios 𝑃/𝑃𝑇0 are plotted
against the non-dimensional inlet pressure 𝑃𝑇 ,𝑟 . The results highlight a remarkable level of attained
periodicity, with a non-perfect overlapping observable only for taps 4-14 and 8-17. It is worth
highlighting that all the experimental results described, including the level of periodicity, are
significant only before the shocks migrate upstream and enter the bladed region. These outcomes
contribute to prove the effective design of the cascade, which is characterized by a remarkable
degree of periodicity between the two adjacent central channels also when operated in strongly
off-design conditions both in terms of operating fluid and inlet total conditions.
113
Tap 1T & 1B Tap 2T & 2B Tap 3T & 3B Tap 4T & 4B

0.9 0.4 0.4 0.4
Upper Channel Upper Channel Upper Channel Upper Channel
0.8 Lower Channel 0.3 Lower Channel 0.3 Lower Channel 0.3 Lower Channel
0.7 0.2 0.2 0.2
0.6 0.1 0.1 0.1
0.5 0 0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Downstream Tap 1, 11 & 21 Downstream Tap 2 & 12 Downstream Tap 3 & 13 Downstream Tap 4 & 14
0.4 0.4 0.4 0.4
Tap 1 Tap 2 Tap 3 Tap 4
0.3 Tap 11 0.3 Tap 12 0.3 Tap 13 0.3 Tap 14
Tap 21
P/PT0
0.2 0.2 0.2 0.2
0.1 0.1 0.1 0.1
0 0 0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Downstream Tap 5 & 15 Downstream Tap 6 & 16 Downstream Tap 8 & 17 Downstream Tap 10 & 19
0.4 0.4 0.4 0.4
Tap 5 Tap 6 Tap 8 Tap 10
0.3 Tap 15 0.3 Tap 16 0.3 Tap 17 0.3 Tap 19
0.2 0.2 0.2 0.2
0.1 0.1 0.1 0.1
0 0 0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Figure 6.6: Periodicity assessment within the two central blade channels (4 top charts) and
downstream the cascade (8 bottom charts).
6.1.4 CFD comparison and total pressure losses

Experimental data in terms of measured static pressures were also compared with the results of CFD
simulations. In particular, the same 2D domain and numerical set-up described in Section 5.2.2
are considered, still relying on the ANSYS-Fluent solver. Nitrogen is modelled as non-polytropic
ideal gas and look-up tables are exploited to introduce its thermophysical properties. To provide a
fair comparison with experimental data, numerical simulations are performed by setting the inlet
total pressure and temperature equal to those measured at a specific instant during the test run.
Experimental data from the 10 tests without the probe are considered.
Results of the comparison in terms of pressure ratios 𝑃/𝑃𝑇0 are reported in the top charts of
Figure 6.7 for the in-channel taps and in the bottom plot for the measurement stations downstream
of the cascade. The pressure values numerically computed are reported in Figure 6.7 together with
the related error-bars, which take into account the pressure variation over a circle with a radius of
0.3 mm (equal to the radius of the manufactured taps). These outcomes highlight a satisfactory
agreement between experimental and numerical data for all the pressure taps considered, revealing
a perfect matching for in-channel taps (except 4T) and for downstream taps from 1 to 5 and 21.
Although very small discrepancies occur for tap 4T and in the central part of the downstream
measuring grid between taps 6 and 16, they become appreciable only over a small portion of the
grid, between taps 16 and 20. In fact, this region is characterized by large pressure gradients, which
are associated to the presence of the shocks reflected by the bottom side-wall. This explanation
is confirmed by the results reported in Figure 6.8, in which the Mach number field processed
from numerical simulation results is compared with a frame extracted from the schlieren video.
Dashed black lines are added in Figure 6.8 to enhance the visibility of schlieren-detected shocks.
As can be inferred from Figure 6.8, small mispredictions of the shocks opening angles produce
incrementally larger discrepancies as we move along the shock path, leading to a final position
of the reflected shocks that differs between experimental and numerical results. Therefore, the
experimental reflected shocks result closer to the taps from 16 to 20, producing the aforementioned
discrepancies in Figure 6.7, in which the static pressures measured by taps 16-20 are slightly higher
than their numerical counterpart. However, measurements and inspections from an experimental
campaign (reported in Section 6.2.1) performed after the completion of the experiments reported
114
Upper channel Lower channel

0.6 0.6
Experimental data Experimental data
Numerical data Numerical data
0.4 0.4
0.2 0.2
0 0
1T 2T 3T 4T 1B 2B 3B 4B
P/PT0
Static Pressure Tap Static Pressure Tap

Downstream Measuring Positions
0.6
Experimental data
Numerical data
0.4
0.2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Downstream Static Pressure Tap
Figure 6.7: Comparison between experimental and numerical results: pressure taps (top) in-
channel and (bottom) downstream of the cascade.
here, demonstrate that the discrepancies highlighted in Figures 6.7 and 6.8 are mostly due to an
inward air leakage, which sets the back-pressure downstream of the cascade to a level higher than
the nominal one predicted by numerical calculations. The back-pressure is only slightly larger for
low-medium inlet total pressures (as those targeted in the experiments described in this section)
while the discrepancy with numerical results amplifies as the inlet total pressure increases, as
described in Section 6.2.1. This is the reason why this phenomenon has been clearly evidenced
and isolated only during experimental campaigns following the one described in the present
section. This issue of an inward air leakage producing an unexpected raise in the experimental
back-pressure is thoroughly described in Section 6.2.1, together with the reason why it happens
and the corresponding measures undertaken to reduce it.
To conclude the characterization of the experimental campaign with nitrogen, the losses
through the cascade may be assessed elaborating the total pressure retrieved by the probe and the
static pressure measured by the tap below the probe head tip, following the measurement strategy
thoroughly described in Section 5.1. The parameter selected to characterize the loss generation
through the cascade is the total pressure loss coefficient 𝑌𝑡𝑜𝑡 , defined as
𝑃𝑇0 − 𝑃𝑇1
𝑌𝑡𝑜𝑡 = , (6.2)
𝑃𝑇0 − 𝑃1
where the subscripts 0 and 1 refers to the stations upstream and downstream the cascade respectively
(see Figure 5.1). The experimentally-retrieved total pressure loss coefficient is compared in
Figure 6.9 with its distribution over the measuring line resulting from numerical simulation. The
mismatch characterizing taps from 16 to 21 originates from both the explanations reported above,
making evident that regions densely populated with reflected shock waves are more challenging to
characterize when calculating the total pressure loss. Focusing therefore on the channel defined by
bottom and central blades, between pressure taps 1 and 15, the results in the top plot of Figure
6.9 highlight a good agreement between experiments and numerical simulations in term of total
pressure loss peak (i.e. maximum values of 𝑌𝑡𝑜𝑡 ). On the other hand, the experimental wakes appear
shifted in the pitch-wise direction with respect to the numerical ones. Although the unexpected raise
in the experimental back-pressure due the inward air leakage may partially contribute (especially in
115
Figure 6.8: Comparison between experimental and numerical results: numerical Mach field vs
schlieren frame. Experimental shocks were highlighted (black lines) to enhance their visibility.
the upper part of the measuring grid), the reason why this mismatch occurs - in terms of wakes shift
- is still not fully clear. Most likely, also the design of the probe may has a non-negligible effect.
The discrepancies between the experimentally-retrieved and numerically-predicted 𝑌𝑡𝑜𝑡 indeed
persist also close to the bottom side-wall, where the effects of the inward leakage are negligible,
as demonstrated in Figure 6.7. In fact, the probe head external and internal diameters were set
equal to 1.6 mm an 1 mm respectively, as a trade off between probe stem stiffness, pneumatic line
dynamic promptness, and the necessity of local (point-like) measurements, as thoroughly described
in Section 5.3.2. However, the dimension of the probe head internal diameter (1 mm) is comparable
with the wake thickness at the measuring location, as evaluated by processing the numerical results.
This feature not only may prevent to measure the cascade total pressure loss distribution with
sufficient spatial resolution, but also may originate a complex interaction between the induced
bowed shock and the blade wake, jeopardizing the hypotheses at the basis of the measurement
strategy describe in Section 5.1. Accordingly, further analyses are required to assess the effects
of the probe head diameter and characterize the interaction between the bowed shock and the
blade wake. To preliminary search for possible effects related to the probe internal diameter, the
bottom plot of Figure 6.9 reports the numerical 𝑌𝑡𝑜𝑡 resulting from the integration of its distribution
over a 1 mm diameter circular area (black circles). The differences with respect to the point-wise
distribution (orange line) are relatively small but tend to reduce the discrepancies with experimental
data, justifying any additional effort required to better characterize the probe-wake interaction.
To summarize, the experiments with nitrogen showed an excellent repeatability and the
pressure data acquired exhibit a remarkable periodicity both between the two central channels
and downstream the cascade, confirming the analyses described in Section 5.2.2. These results
acquire even more importance considering that the cascade was operated in strong off-design
conditions, both in terms of working fluid and inlet total conditions. Moreover, no effects of
the probe on the static pressures acquired below the probe head tip are observed, confirming the
measurement strategy presented in Section 5.1 and the numerical analyses describe din Section
5.3.2. Measurements are also compared with CFD simulations performed with the same flow
model applied to design the cascade. The calculations reproduced in very good approximation
the flow configuration within the channel and downstream fishtail shock system. The agreement
between experimental data and CFD results is indeed satisfactory both in terms of pressure ratios
and Mach field versus schlieren visualizations. Discrepancies can be attributed to a misprediction
116
6.2. MM experimental campaigns
Downstream Measuring Positions

0.8
Experimental Data
Numerical Data
(PT0-PT1)/(PT0-P1 )
0.6
0.4
0.2
-0.2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
0.8
Experimental Data
Numerical Data
(PT0-PT1)/(PT0-P1 )
0.6
Average Numerical Data
0.4
0.2
-0.2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Downstream Static Pressure Taps
Figure 6.9: Total pressure loss coefficient: (top) comparison between experimental and numerical
pointwise distribution; (bottom) comparison between experimental and numerical averaged
distribution.
of the position of reflected shock waves, which, in turn, is caused by a slightly higher experimental
back-pressure downstream of the cascade compared to its numerical counterpart. This phenomenon
is fully addressed in Section 6.2. The blade wakes retrieved experimentally result shifted compared
to those predicted by numerical calculations, despite the agreement between the total pressure
peak retrieved from CFD simulations and experimental data. Ultimately, the outcomes of these
preparatory experiments with nitrogen are surely encouraging since they prove the effectiveness of
the designed linear cascade in reproducing the flow field and shock patterns typical of supersonic
stators and the validity of devised measurement strategies. The results presented in this chapter
can be considered therefore as the basis for the experimental campaign exploiting MM as working
fluid, whose results are described in the following section.
6.2 MM experimental campaigns
The following subsections describe two experimental campaigns carried out adopting the siloxane
MM as working fluid. Section 6.2.1 details the first long-duration campaign, in which test
repeatability and flow field periodicity with MM are investigated. In these tests an unexpected
experimental recompression through the fishtail shocks was observed, much stronger than the
one predicted by numerical computations. Following the experimental identification of this
significant back-pressure, further analyses were focused on trying to shed light on the observed
experimental/numerical discrepancies. The identified cause of the differences between experimental
data and CFD results is described in Section 6.2.1, together with the counteracting solution adopted.
Section 6.2.2 describes the latest set of experiments with MM (and also nitrogen), which are
currently ongoing. In these tests, the discrepancies between experimental and numerical back-
pressure values are largely reduced (basically nullified), as proven by the outcomes described in
Section 6.2.2.
117
Figure 6.10: Schlieren images acquired at different time instants: (left) test midpoint, 𝑃𝑇 = 8.63
and 𝑍 = 0.78, (right) final part of the experiment, 𝑃𝑇 = 3.09 and 𝑍 = 0.93.
6.2.1 First long-duration campaign

The first long-duration experimental campaign exploiting MM as working fluids encompasses
about 30 tests with and without the total pressure probe, covering only a subset of the 21 measuring
positions indicated in Figure 6.1. These experiments target total pressure and temperature at the
cascade inlet equal to 𝑃𝑇 = 18 bar 𝑇𝑇 = 243.5 °C respectively, leading to an initial compressibility
factor at the cascade inlet of about 𝑍 ≈ 0.47. This allows to span a wide range of inlet non-ideal
conditions since 𝑍 tends to increase during a single test run, as the HPV empties, up to reach
almost ideal conditions (𝑍 ≈ 1) at the end of the test. The first objective of this experimental
campaign is to prove the effectiveness of the cascade design in representing the actual flow field
developing in the bladed and semi-bladed regions of supersonic ORC nozzle vanes, both in terms
of shock/fan patterns and periodicity. Figure 6.10 shows two schlieren images acquired at different
time instants for a reference test without the probe, together with a map of the taps corresponding
to the acquired and processed pressure signals. The left-side frame corresponds to 𝑃𝑇 = 8.63 bar
and 𝑍 = 0.78 (test midpoint), while the right-sided one to 𝑃𝑇 = 3.09 bar and 𝑍 = 0.93 (final part
of the experiment).
Both the images clearly show the fishtail shocks departing from bottom and central blades, as
well as the wave patterns developing within the diverging portion of middle and top channels.
Focusing on the left frame, the dark area observable upstream of the throat (in front of the
leading edge of both blades) features relatively high positive pressure gradients and preludes the
flow stagnation points. Conversely, the dark areas within the throat result from very high negative
density gradients such that the schlieren laser beam deflection oversteps the measuring range. In
fact, in these regions the flow features high-acceleration dynamics. Another interesting aspect
emerges from the comparison of the two frames in figure 6.10. In fact, the variation of the fishtail
opening angle, which tends to rotate downstream with an increasing leaning as the test proceeds,
can be inferred by looking at the shocks position with respect to pressure taps marked as 3T and 3B
in the two frames of Figure 6.10. This phenomenon is due to a progressive increase of the Mach
number at blade channel outlet, just upstream of the fishtail shock, associated with the reduction
of gas non-ideality expected to occur as the test develops, as confirmed by the computed Mach
distributions reported in Figure 5.21.
Being interested in characterizing the flow field developing within the bladed channels, the
signals acquired by the 8 in-channel pressure taps are firstly processed to investigate the degree of
cascade periodicity and effectiveness of numerical simulations in representing the experimental
results. The measured static pressures are reported in Figure 6.11 for both the channels as a
function of the test running time. These plots provide a direct assessment of attained periodicity
and give an example of the blow-down wind tunnel emptying dynamics, characterized by slowly
reducing pressure levels. The pressure signals acquired by taps 1T/1B and 2T/2B exhibit a very
118
Upper Channel Lower Channel

Taps 1T & 1B Taps 2T & 2B Taps 3T & 3B Taps 4T & 4B
10 3.5 2.5 5
3
8 2 4
2.5
Pressure [bar]
6 1.5 3
2
1.5
4 1 2
1
2 0.5 1
0.5
0 0 0 0
0 50 100 0 50 100 0 50 100 0 50 100
Time [s]
Figure 6.11: Comparison between pressure signals acquired by top (T, blue colored) and bottom
(B, orange colored) channel pressure taps.
good periodicity, being the lower and upper channel pressures always almost overlapped. On the
contrary, the signals acquired by taps 3B/3T and 4B/4T present larger discrepancies for most of the
test run, up to about 60 seconds. This difference can be explained by a non-perfect periodicity
between the two central channels, leading to a different slope of the fishtail shocks, especially those
stemming from the blade pressure side and directed toward the adjacent blade suction side and
their reflections. In fact, as shown in both frames Figure 6.10, the fishtail shock branches enter the
channels passing very close to the 3B and (to a lower extent) 3T taps, in a region characterized
by very large pressure gradients. Thus even a small variation of the tap-shock distance produces
a considerable variation of the measured pressure. The same shocks, once reflected away from
the suction sides of the adjacent blades, pass through the channels, rather close to the respective
4B and (to a lower extent) 4T taps, respectively. In particular, the upper channel taps 3T and 4T
continue to lie enough upstream of both the shock and its reflection throughout all the test duration.
This justifies the regular (blue colored) signal plotted in Figure 6.11, as well as the lower initial
pressure. On the contrary, in the lower channel the fishtail shock and its reflection overstep 3B
and 4B taps, respectively. This justifies the pressures measured at the beginning of the test (i.e.,
when the shock overlaps to tap 3B but not 3T, as shown in the left frame of Figure 6.10), which
are higher than the corresponding ones measured by taps 3T and 4T, respectively. On the other
hand, when the distance of the shock and its reflection from taps 3B and 4B increase enough (i.e.,
moving towards the end of the experiment, as for the right frame in Figure 6.10) to let the taps
outside the high-pressure gradient regions associated to the shocks, the discrepancies between the
two signals largely reduce (as depicted in Figure 6.11 starting from about 60 seconds). At this
point the pressure becomes essentially a function of the geometric area ratio for taps 3T and 3B.
The pressures retrieved by the 8 in-channel taps are compared with the numerical simulation
results in Figure 6.12. In particular, two time instants are considered; the one for which the
measured inlet total pressure and temperature are, respectively, 𝑃𝑇0 = 8.63 bar and 𝑇𝑇0 = 224.8 °C
(left frame in Figure 6.10) and the one for which 𝑃𝑇0 = 3.09 bar and 𝑇𝑇0 = 216.4 °C (right frame
in Figure 6.10). The boundary conditions for the two numerical simulations are set accordingly,
still exploiting the same numerical schemes and settings described in Section 5.2.2. The results
highlighted in Figure 6.12 depict an outstanding agreement between measured and computed
pressure values within the diverging quasi-isentropic portion of both the two central bladed channels,
with the only exception of tap 3B for the high-pressure case. The maximum relative deviation,
defined as
|𝑃𝑒𝑥 𝑝 − 𝑃𝐶𝐹𝐷 |
𝑃 𝑑𝑒𝑣 = ,
𝑃𝑇0
119
PT = 8.63 bar & TT = 224.8 °C

6 6
5 Numerical data 5 Numerical data
Pressure [bar]
4 4
3 3
2 2
1 1
1T 2T 3T 4T 1B 2B 3B 4B
PT = 3.09 bar & TT = 216.4 °C
2 2
Numerical data Numerical data
1.5 1.5
Pressure [bar]
1 1
0.5 0.5
0 0
1T 2T 3T 4T 1B 2B 3B 4B
Figure 6.12: Comparison between measured and numerical in-channel static pressures, for two
different time instants:(top) test midpoint, 𝑃𝑇0 = 8.63 and 𝑍 = 0.78, (bottom) final part of the
experiment, 𝑃𝑇0 = 3.09 and 𝑍 = 0.93.
is indeed always lower than 1.7%, reaching the 2.6% only for tap 3B at high inlet pressures. On the
contrary, the measurements provided by taps 4T and 4B highlight an unexpected experimental
recompression through the fishtail shocks, stronger than the one predicted by numerical computa-
tions. Although much more remarked at high inlet total pressure (top charts in Figure 6.12), the
discrepancies with CFD results persist throughout the test duration, even when 𝑃𝑇0 decreases up to
about 3 bar, in the final part of the experiment (𝑡 > 60 seconds). The relative deviation for taps 4B
and 4T is indeed extremely large at high inlet total pressure, about 12% − 15%, and decreases up
to 3.5% − 5% for 𝑃𝑇0 = 3 bar. This implies that the larger experimental recompression, amplified
for high inlet total pressures, persists also when a satisfactory periodicity between taps 3B/3T and
4B/4T is achieved and the acquired pressure signals are smooth and regular (see Figure 6.11). The
occurrence of a larger back-pressure during the experiments contributes to explain the discrepancies
with numerical results observed for tests with nitrogen (see Figures 6.7, 6.8, and 6.9), as well as
the uncommon shock pattern reported Figure 6.10 and, consequently, the non-periodic pressure
signals featuring taps 3B/3T and 4B/4T, showcased in Figure 6.11.
The experimental identification of such large and unexpected re-compression suggested to
suspend the planned analyses of measurements repeatibility and total pressure loss quantification,
and to search for explaining the observed experimental/numerical discrepancies.
Firstly, potential improvements of the numerical model and set-up employed have been
considered. Several different CFD models have been indeed developed implementing different
equations (RANS, URANS), numerical domains (3D; incorporation of the duct downstream of
the cascade, actual geometry obtained via coordinate measuring machines), and schemes. None
of the investigated solutions led to appreciable improvements, resulting in minimal distinctions
between the various numerical models investigated in terms of generated flow field and downstream
back-pressure. The efforts have been then focused toward the investigation of the experimental
set-up and toward the assessment of the reliability of measured pressure values. Only in the last
120
Figure 6.13: Test section layout with overlapped non dimensional pressure field (𝑃𝑇0 − 𝑃)/𝑃𝑇0 .
months the cause producing a large experimental back-pressure was discovered. The large pressure
gradients associated with the flow expansion through the cascade lead to an high-pressure region
upstream the cascade (top-right corner of the test section) and a low-pressure area downstream
the blades (bottom-left corner of the test section). This feature is highlighted in Figure 6.13,
where the non dimensional pressure field (𝑃𝑇0 − 𝑃)/𝑃𝑇0 for the design total pressure (𝑃𝑇0 = 25
bar) is overlapped to the test section. The pressure field, in turn, produces a force distribution
acting on both the front (glass window) and rear (steel plate) end-walls of the test section. The
resulting set of forces induces a small rotation of the front glass due to a small gap (≈ 0.3 − 0.4
mm) existing between the blades tip and the glass itself, which results from dimensional tolerances
and a non-perfect assembly of the test section in the supporting frame. This rotation implies a
relative movement between the front glass and the o-ring of the external frame used to fasten the
glass itself. The gap between the front glass and the o-ring produces a leakage of ambient air
within the test section, whose mass flow rate increases as the inlet total pressure raises, the resultant
momentum being proportional to the pressure difference between the inlet and the outlet of the
cascade. Consequently, the large back-pressure observed during the experiments is due to ambient
air inward leaking within the test section, which produces an increases of the (very low) pressure
set downstream of the cascade. Although leakage check tests were carried out before each test
run with both nitrogen and MM, it was not possible to detect any inward/outward leakages since
these tests were performed fixing a uniform pressure within the entire test section (both above and
below the ambient pressure). This issue has been solved by slightly increasing the thickness of the
glass window. As proved by the experimental data reported in Section 6.2.2, this corrective action
successfully nullifies - or strongly reduces - any inward leakage, preventing the unexpected raise in
the measured back-pressure.
6.2.2 Latest experiments

The identification of the inward leakage causing the unexpected raise in the back-pressure enables
the continuation of research activities on the linear cascade. First of all, a set of preparatory
tests with nitrogen is carried out to assess the corrective action adopted. Unlike the experimental
campaign described in Section 6.1,the inlet total pressure is increased up to 8 bar in order to
amplify potential inward/outward leakages and back-pressure raises. Figure 6.14 compares a
frame extracted from the schlieren visualization at the timestep corresponding to 𝑃𝑇0 = 7 bar
(top) with the corresponding non-dimensional density gradient contour ∇ 𝜌˜ (bottom), obtained by
post-processing of the CFD results, and defined as
∇𝜌
∇ 𝜌˜ = , (6.3)
𝜌/𝑐
121
Figure 6.14: (top) Schlieren frame at 𝑃𝑇0 = 7 bar and (bottom) numerical non-dimensional density
gradient contour. The dashed black rectangle in the bottom figure represents the physical boundary
of the schlieren frame.
where 𝑐 is the blade chord. The CFD simulation was carried out with the same boundary conditions
measured at the cascade inlet and exploiting the same numerical model described in Section 5.2.2.
Both frames of Figure 6.14 depict the in-channel taps, while the dashed black rectangle in the
bottom frame represents the physical boundary of the schlieren frame. Differently from Figure 6.3,
the complex shock pattern of the flow field is clearly visible in Figure 6.14, due to the larger density
values involved. Three main shock structures can be recognized: the fishtails shocks originating
from the trailing edge of the blades, an oblique compression wave departing from the suction side
of each blade, particularly where the profile curvature changes concavity, and passing between taps
3 and 4, and the reflections of both aforementioned shock structures against adjacent blades and
side-walls. The shock structures, including the fishtail shocks, oblique compression waves, and
shock reflections, are overall excellently reproduced by the numerical predictions, demonstrating
a remarkable agreement between the experimental observations and CFD simulations. Small
discrepancies concern only the opening angle of the reflected shocks, possibly due to a non-perfect
numerical prediction of the boundary layer development. Moreover, even a small deviation in the
prediction of fishtail opening angles results in progressively larger discrepancies concerning the
position of the reflected shocks as we move along the shock path.
This qualitative analysis is confirmed by the results reported in Figure 6.15, where the pressure
ratios associated with in-channel taps are compared to their numerical counterpart for different
values of 𝑃𝑇0 . The agreement between experimental data and CFD results is excellent for all
122
Numerical Data Experimental Data

0.6
P T0 = 7 bar P T0 = 7 bar
0.4
0.2
0
0.6
0.4
0.2
P/PT0
0
0.6
0.4
0.2
0
0.6
0.4
0.2
0
1T 2T 3T 4T 1B 2B 3B 4B
Pressure Tap Number
Figure 6.15: Comparison between experimental and numerical in-channel pressure ratios for
different values of inlet total pressure.
the taps and regardless of the inlet total pressure considered. The maximum relative deviation
𝑃 𝑑𝑒𝑣 = |𝑃𝑒𝑥 𝑝 − 𝑃𝐶𝐹𝐷 |/𝑃𝑇0 is indeed always lower than 2.5 % and it occurs in 1B, the tap in
the bottom channel placed in in proximity of the throat, a region featuring very high pressure
gradients. Therefore, this measuring position is characterized by a larger uncertainty, resulting
from a non-negligible pressure variation over the integration area, a circle with a radius of 0.3
mm. Without considering the tap 1B, the relative deviation is always below 1.5 %, confirming the
remarkable agreement between predicted and measured pressures. The schlieren frame presented
in Figure 6.14, along with the comparison with CFD results shown in Figure 6.15, demonstrate a
notable improvement compared to the findings discussed in Section 6.1. In the previous campaign
at a lower total inlet pressure, larger discrepancies were indeed observed both in the pressure
retrieved by tap 4T, featuring a relative deviation of 2.7% (see Figure 6.7), and in the comparison
between the visualized shock structures and the numerical Mach field (see Figure 6.8). The
outcomes presented in this section are therefore encouraging and prove the lack of any significant
inward leakages producing an increase of the back-pressure downstream the cascade.
Additional experiments and analyses with nitrogen were not conducted, as the main focus was
on performing a long-duration experimental campaign exploiting MM as working fluid. While
this campaign with MM is currently underway, approximately 10 experiments have already been
completed, which target an inlet total pressure and temperature of 𝑃𝑇0 ≈ 13 bar and 𝑇𝑇0 ≈ 235 °C
respectively, resulting in an initial compressibility factor at the cascade inlet of about 𝑍 ≈ 0.65.
These first experiments and the corresponding results hereinafter reported aim at verifying the tests
repeatability and proving the effectiveness of the cascade design in representing the actual flow
field of supersonic ORC nozzle vanes, both in terms of periodicity and shock/fan patterns, without
the additional complexity and disturbances arising from the employment of the total pressure probe.
123
TNP1 TNP2
Tap 1T Tap 2T Tap 3T Tap 4T
1 1 1 1
0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0 0 0 0
P/PT0
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1

Downstream Tap 3 Downstream Tap 8 Downstream Tap 13 Downstream Tap 18
1 1 1 1
0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0 0 0 0
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
Figure 6.16: Comparison of static pressure signals acquired in 2 different tests without total
pressure probe: (top) taps in the upper channel and (bottom) taps in 4 different downstream
measuring positions (3, 8, 13, and 18).
This is the reason why the implementation and traversing of the probe to retrieve the pitch-wise
loss distribution is planned only in upcoming experiments.
To investigate test repeatability, two experiments without the probe (TNP1 and TNP2) are
considered. The measured pressures are plotted in Figure 6.16 in terms of pressure ratio 𝑃/𝑃𝑇0
against the non dimensional inlet total pressure 𝑃𝑇 ,𝑟 . In particular, Figure 6.16 depicts the signals
acquired by the taps in the upper channel (4 top charts) and by four downstream measuring positions
(4 bottom charts), namely taps 3, 8, 13, and 18 (refer to Figure 6.1 for the notation). A satisfactory
repeatability is attained also with MM as working fluid, especially within the blade channels and
in the two upper downstream measurements (taps 13 and 18). Conversely to what happens with
nitrogen, the pressure ratios 𝑃/𝑃𝑇0 are not constant as the inlet total pressure decreases (𝑃𝑇 ,𝑟 → 1)
throughout the blowdown process. Similarly to the variation of the fishtail opening angles described
in Section 6.2.1, this phenomenon is due to a progressive increase of the Mach number within the
diverging portion of the channel, which depends on inlet total conditions for flow expansions in
non-ideal thermodynamic conditions. This is an expected behavior, as confirmed by the computed
Mach distributions reported in Figure 5.21.
To investigate the cascade periodicity within the bladed channels another test without the probe
is considered (TNP3). The signals acquired by all the 8 in-channel static pressure taps in terms of
pressure ratio 𝑃/𝑃𝑇0 are reported in Figure 6.17 as function of 𝑃𝑇 ,𝑟 . The pressure signals acquired
by taps 1T/1B and 2T/2B exhibit a remarkable periodicity, being the lower and upper channel
pressures always overlapped. Although small discrepancies can be noticed, also the pressures
measured by taps 3T and 3B feature a satisfactory level of attained periodicity. Conversely, the
signals acquired by taps 4T and 4B present larger discrepancies in the first part of the test run (up
to 𝑃𝑇0 = 8 bar), which tend to decrease up to reach a satisfactory periodicity level for 𝑃𝑇0 ⩽ 5 bar.
Both the non-perfect repeatability of the signals acquired by taps 3 and 8 (Figure 6.16) and
the discrepancies retrieved for taps 4T and 4B in terms of periodicity (Figure 6.17) are motivated
by the several shock structures developing within the test section. As previously described, the
shock pattern includes the fishtails departing from the blade trailing edges, the compression waves
originating from blade suction sides, and the several shocks reflected by side-walls and adjacent
124

Taps 1T & 1B Taps 2T & 2B Taps 3T & 3B Taps 4T & 4B
1.1 1.2 1.2 1.2
1 1
1 1
0.8 0.8
0.9 0.8
P/PT0
0.6 0.6
0.8 0.6
0.4 0.4
0.7 0.4
0.2 0.2
0.6 0.2 0 0
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
(PT0,max - PT0)/(PT0,max -PT0,min )
Figure 6.17: Comparison between pressure signals acquired by top (T) and bottom (B) channel
pressure taps.
blades. All the structures are clearly distinguishable in in the top images of Figure 6.18, which
shows two frames extracted from the schlieren visualizations of TNP3, corresponding to 𝑃𝑇0 = 10
bar and 𝑃𝑇0 = 4 bar respectively. The in-channel taps and the downstream measuring positions
previously investigated are also reported in the 2 schlieren frames of Figure 6.18. The shock waves
originating from the partial blade profile integrated in the bottom side-wall, which are labelled and
highlighted with white dashed lines in the schlieren frames of Figure 6.18, represent the major
source of non-periodicity between the 2 central channels and introduce uncertainties between
different experiments. In particular, the shock branch referred to as S1 in Figure 6.18 departs from
the partial profile trailing edge and it is reflected twice before entering the lower central channel:
once by the bottom blade and once by the bottom side-wall. Conversely, the shock branch referred
to as S2 corresponds to the triple reflection (two reflections by the bottom side-wall and one by the
bottom blade) of the compression wave originating from the bottom blade suction side. As the test
proceeds, the shock S1 approaches and oversteps tap 4B , while a similar periodic shock in the
upper channel is not present. This dynamic of the shock implies that tap 4B measures a larger
pressure with respect to tap 4T at the beginning of the test (shock S1 upstream tap 4B), while
the difference decreases as the shock S1 oversteps tap 4B, leading to a satisfactory periodicity, as
proved also by Figure 6.17. A similar explanation can be inferred for the pressures retrieved by the
downstream taps 3 and 8, which experience the crossing of shocks S1 and S2 respectively as 𝑃𝑇0
decreases. In other word, as the HPV empties the opening angle of the shocks reduces and they
rotate downstream, towards the flow direction, leading to sudden pressure variation as they overstep
a measuring position. The dynamics of the shocks can be clearly appreciated by comparing the
two schlieren frames in Figure 6.18. This transient behavior of the shock structures, particularly of
S1 and S2, motivates the peculiar trends reported in Figure 6.16 and contribute to explain the non
perfect repeatability between 2 different tests, as the specific shock pattern and boundary layers
evolution are highly sensitive to operating conditions.
Nonetheless, the experimental schlieren visualisations show a remarkable agreement with CFD
simulation results, reported in the bottom images of Figure 6.18 in terms of non-dimensional
density gradient field. The numerical contours result from CFD simulations carried out with the
same boundary conditions measured at the cascade inlet and exploiting the same numerical model
described in Section 5.2.2. In both the bottom images the in-channel taps are also reported while
the dashed black rectangles represent the physical boundary of the corresponding schlieren frames.
Figure 6.18 prove that the shock structures, including the fishtail shocks, oblique compression
waves, and shock reflections, are overall excellently reproduced by the numerical predictions.
125
Figure 6.18: (top) Schlieren frame at 𝑃𝑇0 = 10 bar (left) and 𝑃𝑇0 = 4 bar (right); (bottom)
corresponding numerical non-dimensional density gradient contours. The dashed withe lines in
the top figures represent 2 reflected shocks: S1 originates from the partial profile trailing edge,
while S2 is the triple reflection of the compression wave departing from the bottom blade suction
side. The dashed black rectangles in the bottom figures represent the physical boundary of the
corresponding schlieren frames.
This qualitative analysis is confirmed by the results reported in Figure 6.19, where the pressure
ratios associated with in-channel taps are compared to their numerical counterpart for different
values of 𝑃𝑇0 . As can be noticed, the plots of Figure 6.19 corroborate the results discussed so far,
showing a very accurate CFD prediction of the experimental data. The largest discrepancies occur
for tap 4B at the beginning of the test (𝑃𝑇0 = 10 bar), for which the relative deviation is about
3.2%. Tap 4B is indeed just downstream the reflected shock S1 for large inlet total pressure values,
highlighting a relative position different from the one numerically predicted, as can be gathered
comparing the two left images of Figure 6.18. Nonetheless, the relative deviation associated with
the other pressure taps is always lower than 2.5% independently on both the spatial position and
the inlet total quantities considered.
Similarly, the pressure ratios associated with downstream measuring positions 3, 8, 13, and 18
are compared to their numerical counterpart for different values of 𝑃𝑇0 in Figure 6.20. Also in
this case the agreement between numerically predicted and measured static pressures is excellent
independently on the inlet conditions. At high inlet pressure (𝑃𝑇0 = 10 bar) the larger deviations
are equal to about 2.5% and 2%, and they occur for taps 13 and 18 respectively, which are indeed
very close to two shocks reflected by the bottom side-wall (see top-left image of Figure 6.18. For
𝑃𝑇0 < 10 bar the largest deviation is associated to tap 8 but it is mostly lower than 3%, other than
for 𝑃𝑇0 = 6 bar for which it reaches the 3.3%. This evolution of the relative deviation for tap 8
as 𝑃𝑇0 decreases is consistent with the shock dynamics observable in the two frames of Figure
6.18. In particular 𝑃𝑇0 ≈ 6 bar coincides with the time frame at which the shock S2 oversteps the
downstream measuring position 8.
126

0.8
P T0 = 10 bar P T0 = 10 bar
0.6
0.4
0.2
0
0.8
0.6
0.4
0.2
P/PT0
0
0.8
0.6
0.4
0.2
0
0.8
0.6
0.4
0.2
0
1T 2T 3T 4T 1B 2B 3B 4B
Pressure Tap Number
Figure 6.19: Comparison between experimental and numerical in-channel pressure ratios for
different values of inlet total pressure.

0.3
P T0 = 10 bar
0.2
0.1
0
0.3
P T0 = 8 bar
0.2
0.1
P/PT0
0
0.3
P T0 = 6 bar
0.2
0.1
0
0.3
P T0 = 4 bar
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Downstream Static Pressure Tap
Figure 6.20: Comparison between experimental and numerical pressure ratios for different values
of inlet total pressure: downstream measuring positions 3, 8, 13, and 18.
127

The first long-duration experimental campaign involved a comprehensive set of approximately 60
experiments performed by operating the facility with nitrogen as working fluid. The experiments
were mostly carried out at moderate inlet total pressure (𝑃𝑇0 ≈ 4 bar) and primarily aimed to
assess the effectiveness of the designed cascade in accurately representing actual supersonic turbine
stators and to evaluate the measurement techniques conceived to characterize the flow field. This
experimental campaign included tests without the total pressure probe, to characterize the flow
morphology, and experiments in which the probe was used to investigate all the 21 downstream
measuring positions.
The acquired schlieren visualizations of the flow-field showed shock structures typical of
supersonic ORC cascades, such as the trailing-edge shock system (fishtails) and the compression
waves originating at the suction side of the blades, where the profile curvature changes concavity.
The experiments exhibit remarkable repeatability of the static and total pressure measurements.
Additionally, the ratio between acquired pressure data and the inlet total pressure (𝑃/𝑃𝑇0 ) remains
constant as HPV empties, regardless of the absolute value of the inlet total pressure. This finding
demonstrates that the isentropic-like flow in the diverging portion of the channels, the pressure rise
across the fishtail shock originating at the blade trailing edge, and the downstream Mach number,
all depend mainly on the geometric area ratio. The static pressure acquired by the downstream taps
results not affected by the placement of the probe into the flow-field, excluding the possibility of
any disturbance effect induced by the probe on the static pressure data measured at rear end-wall
plate. This contributes to corroborate the measurement strategy devised to retrieve the cascade
total pressure loss (described in Section 5.1) and to validate the numerical and theoretical analyses
detailed in Section 5.3.2. Furthermore, the acquired measurements demonstrate a remarkable level
of fluid-dynamic periodicity achieved for the two central channels of the cascade, regardless of the
strongly off-design conditions imposed to the inlet total pressure and temperature, and the change
of the working fluid.
The experimental data were also compared with CFD simulations performed with the same
flow model applied to design the cascade. The calculations demonstrated very good accuracy in the
prediction of the flow configuration within the blade channel and of the downstream fishtail shock
system. The CFD simulations showed an acceptable agreement with the experimental data, both in
terms of pressure ratios and schlieren visualizations of the shocks’ system. However, appreciable
discrepancies occur on a small portion of the downstream measuring grid. Such discrepancies
result from the moderate differences existing between predicted and measured shock waves and
location of their reflection, which highlighted some features of the flow field not perfectly matching
the ideal periodicity of the cascade geometry. Moreover, the pitch-wise distribution of the cascade
total pressure loss coefficient 𝑌𝑡𝑜𝑡 , derived following the measuring strategy described in Section
5.1, when compared with numerical results highlighted the good agreement achieved for the values
of the total pressure loss peak. Conversely, it was found an unexpected difference in the position of
the blade wakes resulting from the experiments with respect to the one predicted by CFD, being
the wakes shifted to each other in pitch-wise direction.
While the probe design could affect the magnitude of the discrepancies existing between
the measured and numerically calculated wake positions, the differences in the position of the
reflected shocks, observed even in the tests performed without the probe, is primarily caused by
the discrepancies between the measured values of the back-pressure downstream of the cascade
and those predicted by CFD. Such differences are slight for low-medium inlet total pressures and
amplifies as the inlet total pressure increases. As a matter of fact, this phenomenon was clearly
evidenced and isolated only during the first long-duration experimental campaign exploiting MM
as working fluid, for which the target total inlet pressure was set to 18 bar. Around 30 tests have
128
been carried out during this campaign, all being affected by an unexpectedly high value of the
measured back-pressure. The comparison with the corresponding CFD simulations showed large
relative deviations between the pressures measured by the in-channel taps downstream of the fishtail
shocks (taps 4B and 4T) and their numerical counterparts, in the range from 12%-15% at high inlet
pressure and down to 3.5%-5% at the end of the test (i.e. low inlet pressure, 𝑃𝑇0 ≈ 3 bar). The
meticulous investigations performed to explain this unexpected finding led to the detection of an
inward leakage of ambient air, entering the test section downstream of the cascade during the test
run. The leakage was due to the pressure distribution developing within the test section during the
experiment, which induces a small rotation of the front glass with partial detachment from the
o-ring placed in the external frame designed to fasten the glass itself. The inward leakage resulted
proportional to the pressure difference between the inlet and outlet sections of the cascade and,
consequently, to the total pressure set at the cascade inlet. This issue was finally resolved by a
slight increase of the glass window’s thickness.
Afterwards, before the execution of a second long-duration campaign with MM, an additional
set of preparatory experiments has been carried out using nitrogen as working fluid, to assess
the effectiveness of the solution adopted to avoid inward leakages of ambient air. To this end,
these experiments feature an inlet total pressure of 8 bar, to amplify potential back-pressure raises.
The experimental data, compared with CDF results, proved that the corrective action undertaken
successfully nullified - or greatly reduced - any inward leakage, thereby preventing the unexpected
rise in the measured back-pressure. All the shock structures - characterized by the experiments
performed after revision of the glass window locking system - including the fishtail shocks, oblique
compression waves, and shock reflections, excellently reproduced the numerical predictions, as
demonstrated by the comparison of schlieren frames and density gradient contours obtained by
post-processing CFD results. Furthermore, the pressure signals retrieved by in-channel pressure
taps feature an outstanding agreement with numerical results. The relative deviation between
measured and predicted pressures is indeed lower than 1.5% for most of the investigated taps,
regardless of the value of 𝑃𝑇0 . Thus these preparatory experiments with nitrogen finally allowed
to demonstrate the effectiveness of the designed linear cascade in reproducing the flow field and
shock patterns typical of supersonic stators.
Although the second experimental campaign with MM is not yet completed, interesting findings
have been already achieved. In particular, the tests were carried out targeting an inlet total pressure
and temperature of 𝑃𝑇0 ≈ 13 bar and 𝑇𝑇0 ≈ 235° C respectively, corresponding to an initial
compressibility factor at the cascade inlet of about 𝑍 ≈ 0.65. The experimental data demonstrates
a satisfactory repeatability, especially within the blade channels and in the upper part of the
downstream measurement grid. In contrast to what measured during the tests with nitrogen, the
pressure ratios 𝑃/𝑃𝑇0 are not constant as the HPV empties and the inlet total pressure decreases.
Similarly, the fishtail shock opening angle tends to increase along the stream-wise direction with
the opening of the shock lines that reduces as the test proceeds. This phenomenon is due to the
reduction in non-ideality of the gas associated with the increase of the compressibility factor during
the progress of the test, which induces a progressive increase of the Mach number within the
diverging portion of the channel. The Mach number distribution is indeed function of the inlet
total conditions for flow expansions in non-ideal thermodynamic conditions, as also proved by
the theoretical studies reported in Section 5.2.2. The static pressures retrieved by the in-channel
taps prove the remarkable level of periodicity attained for the two central channels, especially for
the taps upstream of the fishtails (1B-1T, 2B-2T, and 3B-3T). Conversely, the signals acquired by
taps 4T and 4B present larger discrepancies at high inlet total pressures which tend to decrease
up to reach a satisfactory periodicity level for 𝑃𝑇0 ⩽ 5 bar. Both the non-perfect repeatability of
the signals acquired in the lower part of the downstream measuring grid (taps 3 and 8) and the
non-perfect periodicity retrieved for taps 4T and 4B are caused by the shock waves originating from
129
the partial blade profile integrated in the bottom side-wall. These shocks are the major source of
non-periodicity between the 2 central channels, since they interact only with the taps placed in the
lower channel for large values of the inlet total pressure. Furthermore, the shock waves originating
from the bottom side-wall introduce uncertainties between different experiments, as well. Their
transient dynamics and the evolution of the boundary layers with which they interact are indeed
highly sensitive to the specific inlet conditions of each test. These outcomes suggest that increasing
the number of blades would be beneficial to guarantee a central region featuring measurements
repeatability and high degree of periodicity. Unfortunately, this is not allowed by the current
facility set-up, due to to the limited space available for the test section and the corresponding
instrumentation. However, a re-design of the bottom side-wall is currently ongoing to limit the
disturbances produced by the reflected shock waves in the downstream measuring positions.
The small relative deviations between measured and predicted static pressures, as well as the
remarkable similarity between schlieren frames and numerically-predicted density gradient fields,
demonstrate noticeable improvements with respect to the first experimental campaign with MM,
definitely proving the lack of any significant inward leakages. The outcomes of this first set of
experiments with MM, in which the total pressure probe has not been implemented, are extremely
encouraging. The collected experimental data prove indeed the effectiveness of the designed
linear cascade in reproducing the flow field and shock patterns typical of actual supersonic stators
of ORC turbines with a satisfactory repeatability and periodicity level between the two central
channels. The remarkable agreement observed between the experimental results and high-fidelity
CFD predictions, for both nitrogen and MM, validates the theoretical tools used in the design
phase. Furthermore, it demonstrates the suitability of the employed CFD models for the design
and analysis of ORC turbines, as well as for the validation of reduced-order models and loss
correlations. Although additional experiments with MM are currently underway, the collected data
aim at serving as a valuable reference for future experimental investigations on ORC cascades. This
aspect is extremely important considering the lack of data currently existing in the open literature.
130
CHAPTER 7
CONCLUSION AND OUTLOOK
131
Chapter 7. CONCLUSION AND OUTLOOK
7.1 Conclusions
The analyses documented in this thesis offer a comprehensive characterization of organic Rankine
cycle (ORC) turbines for small-power applications. Despite the large market potential, economical
and technical challenge prevent the penetration ORC power systems in the few kW energy sector.
Several factors indeed complicate the design of bottoming mini-ORCs, including their integration
with the main plant (e.g. additional power management and sealing of the WHR unit), the selection
of the optimal organic working fluid, and the efficient design of all the small-scale components
that usually need to operate in nominal, part-load, and transient conditions. All these design
challenges actively constraint the optimization of the cycle layout. At the same time, the intrinsic
reduced dimensions of the plant and the thermo-physical characteristics of the working fluids
complicate the design of the turbo-expander, whose efficiency strongly impacts the definition
of optimal working fluid, cycle layout and overall performance, and plant profitability. Organic
compounds indeed normally combine low specific work with low speed of sound, leading to
turbines highly sensitive to load changes and featuring few stages characterized by high expansion
ratios and, hence, by transonic or supersonic flows. Additionally, the demand for compactness and
flexibility of operation, couples with high compressibility and severe non-ideal gas effects typical
of organic compounds. All these features motivate the focus of this work in the investigation and
characterization of small-scale ORC turbines.
The first part of the thesis focuses on the design of bottoming ORCs for small-power applications,
with specific attention given to the performance characterization of small-scale turbo-expanders
and their impact on the selection of ORC optimal working fluid and cycle layout. The second part
thoroughly details the experimental campaigns carried out at Politecnico di Milano on supersonic
ORC nozzles in linear cascade configuration. The thesis documents both the full process of design
of the experiments and all the experimental campaigns carried out, aiming at providing valuable
benchmark experimental data to partially fill the lack presently existing in the open literature.
Part I illustrates the strategies developed to design ORCs for small-power applications (mini-
ORCs) and to integrate the turbo-expander performance evaluation within the thermodynamic cycle
optimization. Such strategies include as first step the in-house development of a reduced-order
model, referred to as RITML, for the preliminary sizing of single-stage radial inflow turbines. The
mean-line model RITML is thoroughly described in Chapter 2, together with a comprehensive
review of suitable loss models among those available in literature and with the methodologies
adopted to specifically tailor the reduced-order method to machines operating with organic fluids
and featuring reduced size (small power) and large pressure ratio. To carry out a validation
procedure for RITML as general and unbiased as possible, seven benchmark radial inflow turbines
operating with different fluids and covering a wide range of target expansion ratio, size, and gross
power output have been selected. The following findings were obtained throughout the selection of
most suitable loss correlations and the final validation strategy:
• Concerning the nozzle, the loss sensitivity analysis revealed the poor prediction capability of
the correlation by Rodgers (1987) when high Mach number flows are involved, leading to
the selection of the model by Glassman (1976a) as the reference to estimate stator passage
losses, whose equations are absed on compressible boundary layer theories;
• Concerning the rotor, the employment of the method proposed by Ventura et al. (2012) on
the basis of Rodgers is discouraged when high-pressure ratio turbines operating with organic
fluids are involved, also considering the uncertainties related to this loss model and the
several modifications it underwent during the years. Moreover, the outcomes revealed the
poor prediction capability of the correlation by Glassman (1976a) when applied to the rotor,
which tends to under-predict the entropy generation neglecting the blade loading effects.
132
7.1. Conclusions
Therefore, it is suggested to adopt the model by Baines et al. (1998) as the reference to
estimate rotor passage losses, although its employment may suffer from some limitations
when very large pressure ratios together with sub-atmospheric rotor outlet pressures occur;
• The validation over the seven benchmark turbines evidenced the robustness and prediction
capability of the proposed mean-line model, in terms of total-to-static efficiency and mass
flow rate;
• The deviation between RITML-predicted 𝜂𝑇𝑆 and reference data is always lower than 2.5
percentage points as far as the design point of test case turbines is considered. Concerning
the mass flow rate, the relative deviation between RITML and reference data is mostly lower
than 3.5%, with some exceptions for which the deviation reaches the 7%;
• The mean-line code proved its effectiveness also in reproducing the performance trends
at off-design conditions. The deviations remain limited to 3.5 percentage points for the
efficiency, except for strong off-design conditions, for which the maximum deviation is about
4.5%.
The proposed reduced-order model can be readily employed within the design and optimization
process of ORCs. Moreover, thanks to the low computational cost, it can be exploited for an early
and complete sampling of a wide design space, prior to resorting to time-consuming high-fidelity
CFD simulations, contributing to the future development of tailored design guidelines.
Chapter 3 details the methodology developed to integrate a real estimation of turbine performance
within the mini-ORC design and optimization procedure, rather than assuming constant efficiency
values. To assess its potential, the proposed methodology is applied to the case of ORC-based waste
heat recovery from internal combustion engines of long-haul trucks. To integrate the evaluation of
turbine performance within the cycle optimization tool, an approach based on a single-stage RIT
maximum performance map has been chosen, which systematically exploits RITML to optimize
RIT geometries. The map allows to compute the maximum efficiency of single-stage RITs as
a function of two cycle-dependent parameters: the size parameter SP and the volume ratio Vr,
which are assumed as unique independent variables affecting the maximum achievable efficiency of
optimum design ORC RITs. The validity of this map was assessed based on the designs resulting
from all the optimizations carried out, which include the adoption of different working fluids and
the investigation of expansions in a thermodynamic state far from that one of an ideal gas. The
several optimizations carried out for different sets of SP-Vr allow also to draw useful guidelines
concerning the efficiency variation of optimized RITs. The main outcomes of all the analyses
aforementioned are:
• The ranges of SP and Vr associated with WHR from internal combustion engines of long-haul
trucks are significantly different from those investigated in literature, being characterized by
very small SP (up to 7 mm) and quite large Vr (up to 60);
• The assessment of SP and Vr as the independent parameters sufficient to assure the fluid-
dynamic similarity of optimal RIT designs smooths the limitations and concerns reported in
Masi et al. (2020). The variation of maximum RIT efficiency for the same SP and Vr but
different real-gas compressibility (i.e different inlet/outlet values of both 𝑍 and 𝛤) turns out
to be limited below the range of accuracy associated with RITML, the maximum deviation
from the average efficiency being ±1.2% for 𝑆𝑃 = 0.009 m and 𝑉𝑟 = 35. The variation of
the optimum efficiency for the same SP, Vr, and 𝑍 is even less affected by the adoption of
different working fluids (that means by different values of 𝛤), the maximum deviation from
the average efficiency being ±0.55% for 𝑆𝑃 = 0.008 m, 𝑉𝑟 = 50, and 𝑍 = 0.63;
133
• The maximum efficiency map can be used with sufficient accuracy (±1.2% in the worst case
scenario, plus the intrinsic prediction uncertainty of the mean-line model), regardless of the
specific value of pressure ratio, inlet compressibility factor, and adopted working fluid. The
fitted equation can be thus adopted as a very general fluid-independent turbine performance
map representing the maximum efficiency attainable with an optimization of both machine
geometry and rotational speed for single stage radial inflow turbine architecture;
• Attention is needed when an highly constrained region of the design space is investigated (e.g.
low SP and large Vr) since the activation of the constraints may prevent the optimization
algorithm to find the actual maximum efficiency, making the 2D performance map a less
general tool in those severe conditions;
• Performance of optimized turbines reduces for increasing Vr. This efficiency drop is due to
the presence of highly supersonic flows at the nozzle outlet and the need of adopting very
large blade span variation within the rotor;
• Performance of optimized turbines drops for decreasing SP. Reduced SP values indeed lead
to small turbine sizes and to the consequent increase of loss contributions related to those
geometrical features that cannot be freely scaled down, such as clearances and trailing edge
thicknesses. Additionally, turbine shrinking to comply with a SP reduction cannot be always
achieved due to the physical constraints considered to ensure the mechanical feasibility of
the final proposed geometry. In this case the optimizer is forced to change nozzle outlet
angle and rotor deviation to reach the desired volume ratio, leading to an increase of passage
losses in both the components;
• The optimizations performed pointed out strong difficulties in designing a feasible single-
stage RIT for Vr over 60 and SP below 7 mm due to the limit of subsonic relative flow at
the rotor inlet, the increase of profile and secondary losses, and the miniaturized physical
dimensions of the machine.
Concerning the effects of integrating turbine performance prediction tool within the cycle opti-
mization routine, with reference to the WHR from long haul trucks ICE application, the following
findings are obtained:
• The implementation of a map for expected turbine performance allows to guide the optimiza-
tion algorithm towards solutions which are more representative of the correct optimal design
of the system. In fact, the adoption of a fixed turbine efficiency can be very misleading in
terms of optimal cycle definition and selection of the optimal fluids;
• Focusing on R1233zd(E) as working fluid, the turbine performance map adoption leads
to a reduction of the optimal cycle maximum pressure 𝑃𝑚𝑎𝑥 from approximately 15% to
52%, depending on the available heat power input, while the maximum temperature 𝑇𝑚𝑎𝑥
undergoes a 50°C reduction for the small-size engine case;
• In the framework of a potential fluid screening, the use of constant turbine efficiency for low
molecular complexity and low molar mass compounds (mainly refrigerants) can lead either
to a large overestimation of the potential of such compounds (in small scale applications
where they are not recommended) or to exclusion of such fluids for WHR from ICEs (in
large engines where they can get - especially R1233zde - very competitive performance with
respect to hydrocarbons). In fact, the difference between optimal cycle efficiency adopting
hydrocarbons or refrigerants decreases from from about 5-10 (small-size case) down to 1-3
percentage points (large-size case), depending on the working fluid considered;
134
7.1. Conclusions
The use of constant turbine efficiency assumption for these fluids results in optimal solutions that
may strongly affect the final feasibility of the power plant, possibly leading to the selection of a
cycle thermodynamic not compatible with the effective usage of single-stage RITs. Finally, the
performance map can be easily integrated in numerical models for cycle optimization and allows
to create a direct link between expander efficiency and cycle design parameters, leading to the
definition of optimal cycles that do not result in an unfeasible turbine design.
Part II focuses on the experimental campaigns on supersonic ORC nozzles in linear cascade
configuration. The experiments are conceived to provide valuable experimental data over turbines
operated with organic fluids in non-ideal conditions and, consequently, to partially fill the lack
of experimental data concerning non-ideal and supersonic flows representative of ORC turbine
cascades still existing in the literature. Two main topics are meticulously documented: first, the
description of all the steps required to design from scratch such novel experiments; Second, the
documentation of all the experimental data collected in different campaigns, either with nitrogen or
MM, which are thoroughly analyzed and compared with high-fidelity CFD predictions.
With the aim of generating benchmark data, the experiment arrangement was conceived to
match multiple requirements:
• Design a cascade geometry representative of actual high-temperature and high-expansion
ratio ORC turbines;
• Operate the cascade with a fluid, and at thermodynamic conditions, both relevant for
state-of-the-art high-temperature ORC power plants;
• Devise a set of measurement strategies to have a characterization of the flow field as detailed
as possible, including the measurement of total pressure losses.
In accordance with the requirements and with the constraints provided by the reduced test section
dimensions and TROVA facility operations, the preliminary design choices include:
• The linear cascade is designed to operate with the siloxane MM and features 3 complete
blades, defining two converging-diverging central channels, together with two side-walls
integrating two partial profiles;
• To be consistent with the application considered, the blades of the cascade are designed to
provide an outlet Mach number of 1.6 and a stagger angle of ≈ 75°, and to have parallel inlet
and outlet flows, thus producing null flow deflection through the cascade;
• Total inlet conditions in both design and off-design operation are chosen to carry out
expansion processes featuring highly non-ideal fluid thermodynamic states. Therefore, the
design inlet total pressure and temperature have been set to 𝑃𝑇0 = 25 bar and 𝑇𝑇0 = 261.5°C
respectively, corresponding to an inlet compressibility factor 𝑍 ≈ 0.3.
To design and optimize both the blade and side-walls, high-fidelity CFD simulations are system-
atically applied, exploiting a numerical model developed in-house and based on the commercial
ANSYS-CFX and ANSYS-Fluent solvers. The main finding of the design process are listed below:
• The baseline blade profile derives from a reference axial ORC stator investigated in Romei
et al. (2020), whose leading edge and pressure side have been modified to avoid flow
separations when operated in a linear cascade with null flow deflection;
• The flow field downstream of the throat section as well as the entropy production through
the cascade do not depend on the specific value of the inlet angle set as boundary condition.
This outcome proves the effectiveness of the concept of the experiment and confirm the
null-deflection linear cascade as representative of supersonic stators of both axial and
radial-inflow turbines;
135
• The shape-optimization process, performed to refine the baseline blade shape in the rear part
of the channel, produced a reduction of the entropy rise through the cascade of the 16%,
leading to lower Mach number values within the supersonic diverging portion of the channel
and weaker fishtail shocks;
• The structural integrity and the expected deformation of the optimal geometry have been
assessed through dedicated FEM analyses, demonstrating that the deformation field caused
by the blade pressure distribution does not produce any remarkable variation on the resulting
flow field;
• The streamlined design of side-walls highlights the absence of detached flows and large
reticulation regions, as well as a remarkable flow periodicity;
• The high degree of periodicity of the downstream flow field is confirmed also at off-design
conditions, the reflection waves generated at the bottom side-wall being the major source of
non-periodicity.
The measuring strategies devised and adopted to characterize the flow field delivered by the two
complete blade channels, including three wakes, are listed in the following:
• Due to the formation of a bowed shock in front of the probe head tip, total pressure loss
assessment through the cascade demands for the concurrent measurements of the inlet total
conditions 𝑃𝑇0 , 𝑇𝑇0 , the static pressure downstream of the cascade 𝑃1 , and the total pressure
at the probe head 𝑃𝑇2 ;
• A measurement traversing system is required to characterize the downstream flow field,
including the pitch-wise distribution of the total pressure losses. The measuring grid has
been designed exploiting the results of the CFD simulations over the full linear cascade. It is
parallel to the trailing edge locus, at a stream-wise distance equal to 0.17 times the blade
chord. The grid features 21 points not uniformly spaced along the two central blade pitches,
conveniently refined in the proximity of the wakes where higher gradients are expected.
Each grid point represents one of the positions where the head of the total pressure probe is
located during the experiments;
• The static pressure upstream of the shock is provided by a locally-mounted sensor connected
to the flow through static pressure taps of 0.3 mm diameter, which are machined at the
rear wall such that they result located exactly below the probe head tip. The outcomes of
dedicated theoretical (Moeckel, 1949) and numerical analyses prove that the probe-induced
shock does not interact with the static pressure tap, whose measurement results therefore
representative of the pressure in front of the probe-induced bowed shock;
• The probe is 15 mm long, features a recessed stem, and is located at blade mid-span. Its
design is the results of a trade-off between probe stem stiffness, pneumatic line dynamic
promptness, and the necessity of local (point-like) measurements. To guarantee resistance and
rigidity, the probe is made of a cobalt-chrome alloy and manufactured by stereolithography
(SLA). The probe external diameter has been set to 1.6 mm in order to avoid mechanical
displacements/deformations and limit the disturbances on the flow field. The internal
diameter has been set equal to 1 mm as a trade-off between ensuring a fast-enough dynamic
response and reducing the integration area as much as possible;
• Four static pressure taps are include along the center-line of the diverging portion of the
two central channels, to characterize the flow expansion within the channels and in the
semi-bladed regions;
136
7.1. Conclusions
• A double-passage schlieren equipment was also employed to visualize the flow field density
gradients, evidencing the fan/shock wave structures originated both within the diverging
channels and at blade trailing edges, the morphology of the probe-induced shocks, and the
waves reflected at the bottom side-wall.
The first long-duration experimental campaign involved a comprehensive set of approximately

60 experiments performed by operating the facility with nitrogen at moderate inlet total pressure
(𝑃𝑇0 ≈ 4 bar) and covering all the 21 downstream measuring position. The main findings of this
campaign are detailed below:
• The acquired schlieren visualizations demonstrate the formation of shock structures typical
of supersonic ORC cascades, such as the trailing-edge shock systems (fishtails) and the
compression waves originating at the suction side of the blades;
• The collected experimental data exhibits remarkable repeatability of the static and total
pressure measurements. Additionally, all the acquired pressure ratios 𝑃/𝑃𝑇0 remain constant
as HPV empties, regardless of the absolute value of the inlet total pressure. This findings
demonstrates that the isentropic-like flow in the diverging portion of the channels, as well
as the pressure rise through the fishtail shocks and the downstream Mach number, are a
function of the sole geometric area ratio;
• The pressure acquired by the downstream taps results not affected by the presence of the
probe, excluding the possibility of any disturbance effect produced by the probe itself on the
static pressures measured at rear end-wall plate;
• The acquired measurements demonstrate a remarkable level of fluid-dynamic periodicity,

even when the cascade is operated under strongly off-design conditions both in terms of
operating fluid and inlet total conditions;
• The calculations reproduced rather satisfactorily the flow configuration within the channel
and downstream fishtail shock systems, highlighting an acceptable agreement between
experimental and numerical data both in terms of pressure ratios and Mach field versus
schlieren visualizations. Appreciable discrepancies in the position of reflected shock waves
occur only on a small portion of the downstream measuring grid;
• The experimental pitch-wise distribution of the losses through the cascade is in good
agreement with numerical simulations in term of total pressure loss peak. Conversely, the
blade wakes resulting from the experiments appear shifted compared to those predicted by
numerical calculations.
The mismatch in the position of the reflected shocks, observed even in tests without the probe,
was accompanied by an experimental back-pressure downstream of the cascade higher than its
numerical counterpart. This phenomenon has been clearly evidenced and isolated only during the
first long-duration experimental campaign exploiting MM as working fluid, for which the target
total inlet pressure has been set to 18 bar. It was found that this phenomenon is originated by
an inward leakage of ambient air, entering the test section downstream of the cascade during the
test run. This issue has been resolved by slightly increasing the thickness of the glass window
and improving the assembly of the test section in the supporting frame. Subsequently a set of
preparatory experiments with nitrogen at 8 bar of inlet total pressure has been carried out to assess
the corrective measures adopted, followed by a second long-duration campaign with MM, which is
currently underway. The main findings both of the preparatory tests with nitrogen and of the first
set of experiments with MM are reported below:
137
• Concerning preparatory experiments with nitrogen at high inlet pressure: all the shock
structures generating within the test section, including the fishtail shocks, oblique compression
waves, and shock reflections, are overall excellently reproduced by the numerical predictions.
The pressure signals retrieved by in-channel pressure taps feature an outstanding agreement
with numerical results, the relative deviation being lower than 1.5% for most of the investigated
taps, regardless of the inlet total pressure level. These outcomes prove the effectiveness
of the designed linear cascade in reproducing the flow field and shock patterns typical of
supersonic stators;
• The second long-duration campaign with MM was carried out targeting inlet total pressure
and temperature of 𝑃𝑇0 ≈ 13 bar and 𝑇𝑇0 ≈ 235° C, respectively, and a corresponding
initial compressibility factor at the cascade inlet of about 𝑍 ≈ 0.65. The experimental data
demonstrate a satisfactory repeatability of the measurements, especially within the blade
channels and in the upper part of the downstream measurement grid;
• The pressure ratios 𝑃/𝑃𝑇0 are not constant as the HPV empties and the inlet total pressure
decreases. Similarly, the fishtail opening angles tend to rotate downstream with an increasing
leaning as the test proceeds. These phenomena are caused by the reduction in non-ideality
along the test, resulting in a progressive increase of the Mach number within the diverging
portion of the channel. The Mach number distribution is indeed function of the inlet total
conditions for flow expansions in non-ideal thermodynamic conditions;
• The static pressures retrieved by the in-channel taps prove a remarkable level of periodicity,
which is very slightly altered by the shock waves originating from the partial blade profile
integrated in the bottom side-wall. These shock waves originating from the bottom side-wall
also introduce uncertainties between different experiments. Their transient dynamics and
the evolution of the boundary layers with which they interact are indeed highly sensitive to
the specific inlet conditions of each test;
• CFD simulation results overall excellently reproduce all the shock structures developing
within the test section, regardless the inlet total conditions. Similarly, the static pressure
signals retrieved both by in-channel taps and by 4 downstream measuring positions (taps 3,
8, 13, and 18) show an outstanding agreement with the corresponding CFD predictions. The
relative deviation associated with the in-channel taps is mostly lower than 2.5%, and that
associated with the investigated downstream measuring positions is mostly lower than 3%,
regardless of both the spatial position and the inlet total quantities considered.
The outcomes of this first set of experiments with MM are extremely encouraging. The collected
experimental data prove indeed the effectiveness of the designed linear cascade in reproducing the
flow field and shock patterns typical of actual supersonic stators of ORC turbines, with a satisfactory
repeatability and periodicity level between the two central channels. The remarkable agreement
observed between the experimental results and high-fidelity CFD predictions, for both nitrogen and
MM, validates the theoretical tools used in the design phase. Furthermore, it demonstrates the
suitability of the employed CFD models for the design and analysis of ORC turbines, as well as for
the validation of reduced-order models and loss correlations. Although additional experiments
with MM are currently underway, the collected data aim at serving as a valuable reference for future
experimental investigations on ORC cascades. This aspect is extremely important considering the
lack of this kind of data currently existing in the open literature.
138
7.2. Outlooks
7.2 Outlooks
Notwithstanding the advancements in the characterization of small-scale highly-loaded ORC
turbines illustrated in this work, there are still a lot of open questions that must be addressed in the
future.
Concerning the research activities addressed in Part I, the methodology applied to optimize
the ORC power plant may surely be refined. A possible research path consists in the integration
within the cycle optimization tool of reduced-order models to preliminary size the heat exchangers.
This approach would allow to evaluate the space requirements associated with the implemented
heat exchangers and, therefore, to perform multi-objective optimizations to find the best trade-off
between high conversion efficiency and limited system cost and/or dimensions.
In the context of small-power applications, particularly with regard to waste heat recovery
(WHR) from diesel engines in transportation vehicles, the assessment of the off-design performance
of an ORC power plant is of significant importance. This is because the operating point of
the engine continuously varies throughout the vehicle’s driving cycle, making it challenging to
determine the nominal conditions of the selected single or multiple hot sources. To maximize the
overall performance of the ORC power plant over a typical engine driving cycle, a methodology
could be developed to properly select the nominal conditions of the hot source(s), to be provided as
input for the ORC optimization procedure. This would permit to optimize the overall performance
of the ORC system, intended as integrated over a typical engine driving cycle.
Additionally, the developed reduced-order model for single-stage RITs could be exploited for
a complete sampling of a wide design space, in order to develop design guidelines tailored to
machine operating with organic fluid and to the specific application under investigation.
Concerning the research activities addressed in Part II, a long-duration camping with MM
is currently underway. This implies that several tests need still to be performed to cover all the
aspects that are presently missing. Furthermore, efforts can be focused on improving the existing
experimental setup. The next planned experiments with MM aims firstly at characterizing the
entire downstream measuring grid without the probe, to avoid the additional complexities and
uncertainties arising from the usage of this component. The implementation and traversing of the
probe to retrieve the pitch-wise loss distribution is planned in upcoming experiments. Additional
tests may be also performed at higher inlet total pressure and temperature, aiming at increasing the
non-ideal behavior of the organic vapor during the expansion process.
The discrepancy between the experimental and numerical data regarding the position of wakes
observed during the initial campaign with nitrogen needs to be further investigated. The probe
design however cannot be straightforwardly improved since it is the results of a trade-off between
probe stem stiffness, pneumatic line dynamic promptness, and the necessity of local (point-like)
measurements. Further analyses are therefore needed to evaluate the impact of the probe head
diameter on the flow and to characterize the interaction between the bowed shock and the blade
wake. These analyses can offer valuable insights and guidance before directing all efforts towards
redesigning the probe.
Finally, the huge amount of reflected shock waves interacting with the main flow field in
proximity of the downstream measuring grid suggests that increasing the number of blades would
be advantageous. Since this is not possible with the current facility set-up, due to to the limited
space available for the test section and the corresponding instrumentation, a methodology to
re-design of the bottom side-wall could be investigated in order to limit the disturbances produced
by the reflected shock waves near the downstream measuring positions.
To conclude, the TROVA facility holds great potential for the experimental characterization
of organic vapor flows and, more specifically, of ORC turbines. Therefore, it should be operated
continuously as there are practically endless possibilities for research activities in this field.
139
140
REFERENCES
Ainley, DG & Mathieson, G Cr 1951 A method of performance estimation for axial-flow turbines.
Tech. Rep.. AERONAUTICAL RESEARCH COUNCIL LONDON (UNITED KINGDOM).
Alshammari, Fuhaid, Karvountzis-Kontakiotis, A., Pesiridis, A. & Minton, Timothy 2017

Radial expander design for an engine organic rankine cycle waste heat recovery system. Energy
Procedia 129, 285–292.
Alshammari, Fuhaid, Pesyridis, Apostolos, Karvountzis-Kontakiotis, Apostolos,

Franchetti, Ben & Pesmazoglou, Yagos 2018 Experimental study of a small scale or-
ganic rankine cycle waste heat recovery system for a heavy duty diesel engine with focus on the
radial inflow turbine expander performance. Applied Energy 215, 543–555.
Amicabile, Simone, Lee, Jeong-Ik & Kum, Dongsuk 2015 A comprehensive design methodology
of organic rankine cycles for the waste heat recovery of automotive heavy-duty diesel engines.
Applied Thermal Engineering 87, 574–585.
Arts, T. 2018 Measurement chain: Definitions, error and uncertainty. Von Karman Institute for
Fluid-dynamics, lecture series on measurement techniques for fluid dynamics. .
Arts, T., Boerrigter, H.L., Carbonaro, M. & Van den Braembussche, R.A. 2018 Pressure
measurements. Von Karman Institute for Fluid-dynamics, lecture series on measurement
techniques for fluid dynamics. .
Arts, T. & Buchlin, J.M. 2018 Temperature measurements. Von Karman Institute for Fluid-
dynamics, lecture series on measurement techniques for fluid dynamics. .
Astolfi, Marco & Macchi, Ennio 2015 Efficiency correlations for axial flow turbines working
with non-conventional fluids. In 3rd International Seminar on ORC Power Systems, pp. 12–14.
Astolfi, Marco, Martelli, Emanuele & Pierobon, L 2017 Thermodynamic and technoeconomic
optimization of organic rankine cycle systems. In Organic Rankine cycle (ORC) power systems,
pp. 173–249. Elsevier.
Astolfi, Marco, Romano, Matteo C, Bombarda, Paola & Macchi, Ennio 2014 Binary orc
(organic rankine cycles) power plants for the exploitation of medium–low temperature geothermal
sources–part a: Thermodynamic optimization. Energy 66, 423–434.
Aungier, Ronald H 2006a Turbine aerodynamics. ASME, New York .
Aungier, R. H. 2006b Turbine aerodynamics: Axial-flow and radial-flow turbine design and
analysis. In ASME, p. 420.
141
Bademlioglu, AH, Canbolat, AS, Yamankaradeniz, N & Kaynakli, O 2018 Investigation
of parameters affecting organic rankine cycle efficiency by using taguchi and anova methods.
Applied Thermal Engineering 145, 221–228.
Bahamonde, Sebastian, Pini, Matteo, De Servi, Carlo, Rubino, Antonio & Colonna, Piero
2017a Method for the preliminary fluid dynamic design of high-temperature mini-organic
rankine cycle turbines. Journal of Engineering for Gas Turbines and Power 139 (8).
Bahamonde, Sebastian, Pini, Matteo, De Servi, Carlo, Rubino, Antonio & Colonna, Piero
2017b Method for the preliminary fluid dynamic design of high-temperature mini-organic
rankine cycle turbines. Journal of Engineering for Gas Turbines and Power 139 (8).
Baines, Nicholas C. 1996 Flow development in radial turbine rotors. In Turbo Expo: Power for
Land, Sea and Air, , vol. 78729, p. V001T01A021. American Society of Mechanical Engineers,
arXiv: ASME 1996 International Gas Turbine and Aeroengine Congress and Exhibition.
Baines, N. C., of Mechanical Engineers, Institution & Group, Combustion Engines 1998
A meanline prediction method for radial turbine efficiency. pp. 45–56. Bury St Edmunds:
Professional Engineering Publishing for the Institution of Mechanical Engineers;.
Baumgärtner, David, Otter, John J. & Wheeler, Andrew P. S. 2021 The effect of isentropic
exponent on supersonic turbine wakes. In Proceedings of the 3rd International Seminar on
Non-Ideal Compressible Fluid Dynamics for Propulsion and Power: NICFD 2020, pp. 153–161.
Springer International Publishing.
Baumgärtner, D., Otter, J.J. & Wheeler, A.P.S. 2019 The effect of isentropic exponent on
transonic turbine performance. , vol. 2C-2019.
Bell, Ian H., Wronski, Jorrit, Quoilin, Sylvain & Lemort, Vincent 2014 Pure and pseudo-
pure fluid thermophysical property evaluation and the open-source thermophysical property
library coolprop. Industrial & Engineering Chemistry Research 53 (6), 2498–2508, arXiv:
http://pubs.acs.org/doi/pdf/10.1021/ie4033999.
Beltrame, F., Head, A.J., De Servi, C., Pini, M., Schrijer, F. & Colonna, P. 2021 First
experiments and commissioning of the orchid nozzle test section. ERCOFTAC Series 28,
169–178.
Bini, R & Manciana, E 1996 Organic rankine cycle turbogenerators for combined heat and power
production from biomass. In Proceedings of the 3rd Munich Discussion meeting.
Biscani, Francesco & Izzo, Dario 2020 A parallel global multiobjective framework for opti-
mization: pagmo. Journal of Open Source Software 5 (53), 2338.
Bridle, EA & Boulter, RA 1967 Paper 42: A simple theory for the prediction of losses in the
rotors of inward radial flow turbines. In Proceedings of the institution of mechanical engineers,
conference proceedings, , vol. 182, pp. 393–405. SAGE Publications Sage UK: London, England.
Cammi, G., Conti, C.C., Spinelli, A. & Guardone, A. 2021a Experimental characterization of
nozzle flow expansions of siloxane mm for orc turbines applications. Energy 218.
Cammi, G., Spinelli, A., Cozzi, F. & Guardone, A. 2021b Automatic detection of oblique shocks
and simple waves in schlieren images of two-dimensional supersonic steady flows. Measurement:
Journal of the International Measurement Confederation 168.
142
Campana, Francesco, Bianchi, Michele, Branchini, Lisa, De Pascale, Andrea, Peretto,
Antonio, Baresi, Marco, Fermi, Alessandro, Rossetti, Nicola & Vescovo, Riccardo
2013 Orc waste heat recovery in european energy intensive industries: Energy and ghg savings.
Energy Conversion and Management 76, 244–252.
Cipollone, Roberto, Bianchi, Giuseppe, Di Battista, Davide, Contaldi, Giulio & Murgia,
Stefano 2014 Mechanical energy recovery from low grade thermal energy sources. Energy
Procedia 45, 121–130.
Colonna, Piero, Casati, Emiliano, Trapp, Carsten, Mathijssen, Tiemo, Larjola, Jaakko,
Turunen-Saaresti, Teemu & Uusitalo, Antti 2015 Organic Rankine Cycle Power Systems:
From the Concept to Current Technology, Applications, and an Outlook to the Future. Journal
of Engineering for Gas Turbines and Power 137 (10).
Committee, European FluoroCarbons Technical 2020 Hcfo-1233zd(e) chillers receive
environment awards around the world. http://precog.iiitd.edu.in/people/anupama.
Conti, Camilla Cecilia 2021 Non-ideal compressible fluid dynamics of organic vapors: from
nozzle flows to pressure probes. Ph.D. Thesis .
Conti, Camilla C., Fusetti, Alberto, Spinelli, Andrea, Gaetani, Paolo & Guardone,
Alberto 2022 Pneumatic system for pressure probe measurements in transient flows of non-ideal
vapors subject to line condensation. Measurement p. 110802.
Conti, Camilla C., Fusetti, Alberto, Spinelli, Andrea & Guardone, Alberto 2023 Shock
losses and pitot tube measurements in non-ideal supersonic and subsonic flows of organic vapors.
Energy 265, 126087.
Craig, H. R. M. & Cox, H. J. A. 1970 Performance estimation of axial flow tur-
bines. Proceedings of the Institution of Mechanical Engineers 185 (1), 407–424, arXiv:
https://doi.org/10.1243/PIME PROC 1970 185 048 02.
Da Lio, Luca, Manente, Giovanni & Lazzaretto, Andrea 2016 Predicting the optimum design
of single stage axial expanders in orc systems: Is there a single efficiency map for different
working fluids? Applied Energy 167, 44–58.
Da Lio, Luca, Manente, Giovanni & Lazzaretto, Andrea 2017a A mean-line model to predict
the design efficiency of radial inflow turbines in organic rankine cycle (orc) systems. Applied
Energy 205, 187–209.
Da Lio, Luca, Manente, Giovanni & Lazzaretto, Andrea 2017b A mean-line model to predict
the design efficiency of radial inflow turbines in organic rankine cycle (orc) systems. Applied
Energy 205, 187–209.
Dai, Yiping, Wang, Jiangfeng & Gao, Lin 2009 Parametric optimization and comparative
study of organic rankine cycle (orc) for low grade waste heat recovery. Energy Conversion and
Management 50 (3), 576–582.
Daily, J. W. & Nece, R. E. 1960 Chamber dimension effects on induced flow and frictional
resistance of enclosed rotating disks. Journal of Fluid Engineering 82 (1), 217–230, aSME J.
Basic Eng.
De Servi, Carlo M., Burigana, Matteo, Pini, Matteo & Colonna, Piero 2019 Design method
and performance prediction for radial-inflow turbines of high-temperature mini-organic rankine
cycle power systems. Journal of Engineering for Gas Turbines and Power 141 (9).
143
Degrez, G., Riethmuller, M.L. & Fletcher, D. 2018 Optical measurements. Von Karman
Institute for Fluid-dynamics, lecture series on measurement techniques for fluid dynamics. .
Dixon, S Larry & Hall, Cesare 2013 Fluid mechanics and thermodynamics of turbomachinery.
Butterworth-Heinemann.
Dolz, V, Novella, R, Garcı́a, A & Sánchez, J 2012 Hd diesel engine equipped with a bottoming
rankine cycle as a waste heat recovery system. part 1: Study and analysis of the waste heat
energy. Applied Thermal Engineering 36, 269–278.
Fearn, R. M., Mullin, T. & Cliffe, K. A. 1990 Nonlinear flow phenomena in a symmetric sudden
expansion. Journal of Fluid Mechanics 211, 595–608.
Fiaschi, Daniele, Innocenti, Gianmaria, Manfrida, Giampaolo & Maraschiello, Francesco
2016 Design of micro radial turboexpanders for orc power cycles: From 0d to 3d. Applied
Thermal Engineering 99, 402–410.
Fiaschi, Daniele, Manfrida, Giampaolo & Maraschiello, Francesco 2012 Thermo-fluid
dynamics preliminary design of turbo-expanders for orc cycles. Applied Energy 97, 601–608.
Fiaschi, Daniele, Manfrida, Giampaolo & Maraschiello, Francesco 2015 Design and
performance prediction of radial orc turboexpanders. Applied Energy 138, 517–532.
Gaetani, Paolo & Persico, Giacomo 2021 Influence of the rotor-driven perturbation on the
stator-exit flow within a high-pressure gas turbine stage. International Journal of Turbomachinery,
Propulsion and Power 6 (3), 28.
Glassman, A.J. 1995 Enhanced analysis and users manual for radial-inflow turbine conceptual
design code rtd. Tech. Rep.. NASA-Contractor Report CR-195454, Washington D.C., USA.
Glassman, A. J. 1976a Computer program for design analysis of radial-inflow turbines. Tech. Rep..
NASA Technical Note, TN D-8164, Washington D.C., USA.
Glassman, Arthur J 1976b Computer program for design analysis of radial-inflow turbines. Tech.
Rep.. NASA Technical Note, TN D-8164.
Glensvig, Michael, Schreier, Heimo, Tizianel, Mauro, Theissl, Helmut, Krähenbühl,
Peter, Cococcetta, Fabio & Calaon, Ivan 2016 Testing of a long haul demonstrator vehicle
with a waste heat recovery system on public road. Tech. Rep.. SAE Technical Paper.
Hountalas, Dimitrios T, Mavropoulos, Georgios C, Katsanos, Christos & Knecht, Walter
2012 Improvement of bottoming cycle efficiency and heat rejection for hd truck applications by
utilization of egr and cac heat. Energy Conversion and Management 53 (1), 19–32.
Imran, Muhammad, Usman, Muhammad, Park, Byung-Sik & Lee, Dong-Hyun 2016 Vol-
umetric expanders for low grade heat and waste heat recovery applications. Renewable and
Sustainable Energy Reviews 57, 1090–1109.
International Energy Agency 2019 World Energy Outlook 2019.
Japikse, D. 1982 Advanced diffusion levels in turbocharger compressors and component matching.
In: 1st international conference on turbocharging and turbochargers, 26-28 april 1982, london,
UK .
Jones, A. C. 1996 Design and test of a small, high pressure ratio radial turbine. Journal of
Turbomachinery 118 (2), 362–370.
144
Jones, Donald R 2001 A taxonomy of global optimization methods based on response surfaces.
Journal of global optimization 21, 345–383.
Kastner, L. J. & Bhinder, F. S. 1975 A method for predicting the performance of a centripetal
gas turbine fitted with a nozzle-less volute casing. In Turbo Expo: Power for Land, Sea, and Air,
, vol. 79771, p. V01BT02A003. American Society of Mechanical Engineers, arXiv: ASME
1975 International Gas Turbine Conference and Products Show.
Kirkup, Les & Frenkel, Robert B 2006 An introduction to uncertainty in measurement: using
the GUM (guide to the expression of uncertainty in measurement). Cambridge University Press.
Ko, Jordan & Wynn, Henry P. 2016 The algebraic method in quadrature for uncertainty
quantification. SIAM/ASA Journal on Uncertainty Quantification 4 (1), 331–357.
La Seta, Angelo, Meroni, Andrea, Andreasen, Jesper Graa, Pierobon, Leonardo, Persico,
Giacomo & Haglind, Fredrik 2016 Combined turbine and cycle optimization for organic
rankine cycle power systems—part b: Application on a case study. Energies 9 (6), 393.
Lemmon, E. W., , Bell, Ian H., Huber, M. L. & McLinden, M. O. 2018 NIST Standard Reference
Database 23: Reference Fluid Thermodynamic and Transport Properties-REFPROP, Version
10.0, National Institute of Standards and Technology.
Lemort, Vincent, Guillaume, Ludovic, Legros, Arnaud, Declaye, Sébastien & Quoilin,
Sylvain 2013 A comparison of piston, screw and scroll expanders for small scale rankine cycle
systems. In The 3rd international conference on microgeneration and related technologies.
Lin, Shan, Zhao, Li, Deng, Shuai, Ni, Jiaxin, Zhang, Ying & Ma, Minglu 2019 Dynamic
performance investigation for two types of orc system driven by waste heat of automotive internal
combustion engine. Energy 169, 958–971.
Macchi, E. 1977 Design criteria for turbines operating with fluids having a low-speed of sound. In
Lecture Series 100 on Closed Cycle Gas Turbines, Von Karman Institute for Fluid Dynamics.
Macchi, Ennio 2017 Theoretical basis of the organic rankine cycle. In Organic Rankine Cycle
(ORC) Power Systems, pp. 3–24. Elsevier.
Macchi, Ennio & Astolfi, Marco 2017a Axial flow turbines for organic rankine cycle applications.
In organic Rankine cycle (orc) power systems, pp. 299–319. Elsevier.
Macchi, E. & Astolfi, M. 2017b Organic Rankine Cycle (ORC) Power Systems. Technologies
and applications. New York: Woodhead Publishing Series in Energy: Number 107, Elsevier.
Macchi, Ennio & Perdichizzi, A 1981a Efficiency prediction for axial-flow turbines operating
with nonconventional fluids .
Macchi, E. & Perdichizzi, A. 1981b Efficiency Prediction for Axial-Flow Tur-

bines Operating with Nonconventional Fluids. Journal of Engineering for Power
103 (4), 718–724, arXiv: https://asmedigitalcollection.asme.org/gasturbinespower/article-
pdf/103/4/718/5804948/718 1.pdf.
Marelli, S., Lüthen, N. & Sudret, B. 2022 UQLab user manual – Polynomial chaos expansions.
Tech. Rep.. Chair of Risk, Safety and Uncertainty Quantification, ETH Zurich,Switzerland,
report UQLab-V2.0-104.
145
Masi, Massimo, Da Lio, Luca & Lazzaretto, Andrea 2020 An insight into the similarity
approach to predict the maximum efficiency of organic rankine cycle turbines. Energy 198,
117278.
Mclallin, K. L. & Haas, J. E. 1980 Experimental performance and analysis of 15.04-centimeter-

tip-diameter, radial-inflow turbine with work factor of 1.126 and thick blading. Tech. Rep.. NASA
Technical Paper TP-1730.
Meitner, Peter L & Glassman, Arthur J 1983 Computer code for off-design performance
analysis of radial-inflow turbines with rotor blade sweep. Tech. Rep.. NASA Technical Paper,
2199.
Meroni, Andrea, Andreasen, Jesper Graa, Persico, Giacomo & Haglind, Fredrik 2018a
Optimization of organic rankine cycle power systems considering multistage axial turbine design.
Applied Energy 209, 339–354.
Meroni, Andrea, Robertson, Miles, Martinez-Botas, Ricardo & Haglind, Fredrik 2018b
A methodology for the preliminary design and performance prediction of high-pressure ratio
radial-inflow turbines. Energy 164, 1062–1078.
Mimic, Dajan, Jätz, Christoph & Herbst, Florian 2018 Correlation between total pressure
losses of highly loaded annular diffusers and integral stage design parameters. Journal of the
Global Power and Propulsion Society 2, 388–401.
Moeckel, Wolfgang E 1949 Approximate method for predicting form and location of detached
shock waves ahead of plane or axially symmetric bodies. Tech. Rep..
Moustapha, H., Zelesky, M. F., Baines, N. C. & Japikse, D. 2003 Axial and radial turbines.
Concepts NREC, White River Junction, VT (2003).
Musgrave, D. S. 1979 The prediction of design and off-design efficiency for centrifugal compressor
impellers. In Performance Prediction of Centrifugal Pumps and Compressors, pp. 185–189.
AA(Chrysler Corp., Highland Park, Mich).
Mushtaq, Noraiz, Colella, Gabriele & Gaetani, Paolo 2022 Design and parametric analysis
of a supersonic turbine for rotating detonation engine applications. International Journal of
Turbomachinery, Propulsion and Power 7 (1), 1.
Nithesh, K. G. & Chatterjee, Dhiman 2016 Numerical prediction of the performance of radial
inflow turbine designed for ocean thermal energy conversion system. Applied Energy 167, 1–16.
Osnaghi, C. 2020 Teoria delle Turbomacchine. Società Editrice Esculapio.
Papadopoulos, Athanasios I, Stijepovic, Mirko, Linke, Patrick, Seferlis, Panos & Voute-
takis, Spyros 2013 Toward optimum working fluid mixtures for organic rankine cycles using
molecular design and sensitivity analysis. Industrial & Engineering Chemistry Research 52 (34),
12116–12133.
Perdichizzi, Antonio & Lozza, Giovanni 1987 Design criteria and efficiency prediction for
radial inflow turbines. In ASME 1987 International Gas Turbine Conference and Exhibition.
American Society of Mechanical Engineers Digital Collection.
Persico, G. & Pini, M. 2017 Fluid dynamic design of organic rankine cycle turbines. In Organic
Rankine Cycle (ORC) Power Systems, pp. 253–297. Elsevier.
146
Persico, G., Rodriguez-Fernandez, P. & Romei, A. 2019 High-fidelity shape optimization of non-
conventional turbomachinery by surrogate evolutionary strategies. Journal of Turbomachinery
141 (8).
Pini, M., De Servi, C., Burigana, M., Bahamonde, S., Rubino, A., Vitale, S. & Colonna, P.
2017a Fluid-dynamic design and characterization of a mini-orc turbine for laboratory experiments.
Energy Procedia 129, 1141–1148.
Pini, M., Spinelli, A., Dossena, V., Gaetani, P. & Casella, F. 2011 Dynamic simulation of a
test rig for organic vapours. pp. 1977–1988.
Pini, Matteo, Vitale, S, Colonna, Piero, Gori, Giulio, Guardone, ALBERTO, Economon, T,
Alonso, JJ & Palacios, F 2017b Su2: the open-source software for non-ideal compressible
flows. In Journal of Physics: Conference Series, , vol. 821, p. 012013. IOP Publishing.
Quoilin, Sylvain, Declaye, Sébastien, Tchanche, Bertrand F & Lemort, Vincent 2011
Thermo-economic optimization of waste heat recovery organic rankine cycles. Applied thermal
engineering 31 (14-15), 2885–2893.
Reinker, F., Wagner, R., Hake, L. & der Wiesche, S. 2021 High subsonic flow of an organic
vapor past a circular cylinder. Experiments in Fluids 62 (3).
Ringler, J, Seifert, M, Guyotot, V & Hübner, W 2009 Rankine cycle for waste heat recovery of
ic engines. SAE International Journal of Engines 2 (1), 67–76.
Robertson, M.C., Newton, P.J., Chen, T. & Martinez-Botas, R.F. 2019 Development and
commissioning of a blowdown facility for dense gas vapours. , vol. 3.
Rodgers, C. 1987 Mainline performance prediction for radial inflow turbines. Von Karman Institute
for Fluid-dynamics, Lecture Series 07 on Small High Pressure Ratio Turbines .
Romei, A., Vimercati, D., Persico, G. & Guardone, A. 2020 Non-ideal compressible flows in
supersonic turbine cascades. Journal of Fluid Mechanics 882, A121–A1226.
Rosset, Kevin, Mounier, Violette, Guenat, Eliott & Schiffmann, Jürg 2018 Multi-objective
optimization of turbo-orc systems for waste heat recovery on passenger car engines. Energy 159,
751–765.
Salah, Salma I, White, Martin T & Sayma, Abdulnaser I 2022 A comparison of axial turbine
loss models for air, sco2 and orc turbines across a range of scales. International Journal of
Thermofluids p. 100156.
Saleh, Bahaa, Koglbauer, Gerald, Wendland, Martin & Fischer, Johann 2007 Working
fluids for low-temperature organic rankine cycles. Energy 32 (7), 1210–1221.
Sauret, Emilie & Gu, Yuantong 2014 Three-dimensional off-design numerical analysis of an
organic rankine cycle radial-inflow turbine. Applied Energy 135, 202–211.
Schuster, Sebastian, Markides, Christos N. & White, Alexander J. 2020 Design and
off-design optimisation of an organic rankine cycle (orc) system with an integrated radial turbine
model. Applied Thermal Engineering 174, 115192.
Sethuraman, Vigneshvaran & Khan, Sher Afghan 2016 Effect of sudden expansion for varied
area ratios at subsonic and sonic flow regimes. International Journal of Energy, Environment
and Economics 24 (1), 99.
147
Smith, Michael 2009 ABAQUS/Standard User’s Manual, Version 6.9. United States: Dassault
Systèmes Simulia Corp.
Span, R. & Wagner, W. 2003 Equations of state for technical applications. i. simultaneously
optimized functional forms for nonpolar and polar fluids. International Journal of Thermophysics
24 (1), 1–39.
Spence, S. W. T. & Artt, D. W. 1997 Experimental performance evaluation of a 99.0 mm

radial inflow nozzled turbine with different stator throat areas. Proceedings of the Institution of
Mechanical Engineers, Part A: Journal of Power and Energy 211 (6), 477–488.
Spinelli, A., Cammi, G., Gallarini, S., Zocca, M., Cozzi, F., Gaetani, P., Dossena, V. &
Guardone, A. 2018 Experimental evidence of non-ideal compressible effects in expanding flow
of a high molecular complexity vapor. Experiments in fluids 59 (126).
Spinelli, A., Dossena, V., Gaetani, P., Osnaghi, C. & Colombo, D. 2010 Design of a test rig for
organic vapours. In Proceedings of ASME Turbo Expo, Glasgow, UK.
Spinelli, A., Pini, M., Dossena, V., Gaetani, P. & Casella, F. 2013 Design, Simulation, and
Construction of a Test Rig for Organic Vapors. Journal of Engineering for Gas Turbines and
Power 135.
Spraker, W.A. 1987 Contour clearance losses in radial inflow turbines for turbochargers. ASME
Paper 871.
Sprouse III, Charles & Depcik, Christopher 2013 Review of organic rankine cycles for internal
combustion engine exhaust waste heat recovery. Applied thermal engineering 51 (1-2), 711–722.
Stanitz, J. D. 1952 One-dimensional compressible flow in vaneless diffusers of radial- and

mixed-flow centrifugal compressors, including effects of friction, heat transfer and area change.
Tech. Rep.. National Advisory Committee for Aeronautics, Technical Note 2610.
Stewart, Warner L 1955 Analysis of two-dimensional compressible-flow loss characteristics

downstream of turbomachine blade rows in terms of basic boundary-layer characteristics. Tech.
Rep..
Storn, Rainer & Price, Kenneth 1997 Differential evolution-a simple and efficient heuristic for
global optimization over continuous spaces. Journal of global optimization 11 (4), 341.
Suhrmann, Jan F., Peitsch, Dieter, Gugau, Marc, Heuer, Tom & Tomm, Uwe 2010 Validation
and development of loss models for small size radial turbines. Turbo Expo: Power for Land, Sea
and Air pp. 1937–1949, arXiv: ASME Turbo Expo 2010: Power for Land, Sea, and Air.
Tartière, Thomas & Astolfi, Marco 2017 A world overview of the organic rankine cycle market.
Energy Procedia 129, 2–9.
Thol, Monika, Dubberke, Frithjof H, Rutkai, Gabor, Windmann, Thorsten, Köster,

Andreas, Span, Roland & Vrabec, Jadran 2016 Fundamental equation of state correlation
for hexamethyldisiloxane based on experimental and molecular simulation data. Fluid Phase
Equilibria 418, 133–151.
Thompson, Philip A 1971 A fundamental derivative in gasdynamics. The Physics of Fluids 14 (9),
1843–1849.
148
Thompson, R. G. 1979 Performance correlations for flat and conical diffusers. Turbo Expo: Power
for Land, Sea, and Air 1A:Gas Turbines (V01AT01A052), arXiv: ASME 1979 International
Gas Turbine Conference and Exhibit and Solar Energy Conference.
Tian, Hua, Shu, Gequn, Wei, Haiqiao, Liang, Xingyu & Liu, Lina 2012 Fluids and parameters
optimization for the organic rankine cycles (orcs) used in exhaust heat recovery of internal
combustion engine (ice). Energy 47 (1), 125–136.
Vavra, M.H. 1969 Axial flow turbines. Von Karman Institute for Fluid-dynamics, lecture series
15. .
Ventura, Carlos A. M., Jacobs, Peter A., Rowlands, Andrew S., Petrie-Repar, Paul &
Sauret, Emilie 2012 Preliminary design and performance estimation of radial inflow turbines:
An automated approach. Journal of Fluids Engineering 134 (3).
Wang, EH, Zhang, HG, Fan, BY, Ouyang, MG, Zhao, Y & Mu, QH 2011a Study of working
fluid selection of organic rankine cycle (orc) for engine waste heat recovery. Energy 36 (5),
3406–3418.
Wang, XD, Zhao, L & Wang, JL 2011b Experimental investigation on the low-temperature solar
rankine cycle system using r245fa. Energy Conversion and Management 52 (2), 946–952.
Wasserbauer, C. & Glassman, A. J. 1975 Fortran program for predicting off-design performance
of radial-inflow turbines. Tech. Rep.. NASA Technical Note, TN D-8063, Washington D.C.,
USA.
White, Martin T & Sayma, Abdulnaser I 2019 Simultaneous cycle optimization and fluid
selection for orc systems accounting for the effect of the operating conditions on turbine efficiency.
Frontiers in Energy Research 7, 50.
Whitfield, A. 1990 The preliminary design of radial inflow turbines. Journal of Turbomachinery
112 (1), 50–57.
Whitfield, A. & Baines, N. C. 1990a Design of radial turbomachines. New York, NY, USA: John
Wiley and Sons Inc.
Whitfield, Arnold & Baines, Nicholas C 1990b Design of radial turbomachines .
Wiesner, F. J. 1967 A review of slip factors for centrifugal impellers. Journal of Engineering for
Power 89, 558–566.
Xu, Bin, Rathod, Dhruvang, Yebi, Adamu, Filipi, Zoran, Onori, Simona & Hoffman, Mark
2019 A comprehensive review of organic rankine cycle waste heat recovery systems in heavy-duty
diesel engine applications. Renewable and Sustainable Energy Reviews 107, 145–170.
Yang, Can & Li, Yuhang 2021 Fuel-saving performance and main losses of an organic-rankine-
cycle-based exhaust heat recovery system in heavy truck application scenarios. Applied Thermal
Engineering 193, 117025.
Yang, Kangyi, Bargende, Michael & Grill, Michael 2018 Evaluation of engine-related
restrictions for the global efficiency by using a rankine cycle-based waste heat recovery system
on heavy duty truck by means of 1d-simulation. Tech. Rep.. SAE Technical Paper.
Zhu, Sipeng, Deng, Kangyao & Qu, Shuan 2013 Energy and exergy analyses of a bottoming
rankine cycle for engine exhaust heat recovery. Energy 58, 448–457.
149
Zocca, M., Guardone, A., Cammi, G., Cozzi, F. & Spinelli, A. 2019 Experimental observation
of oblique shock waves in steady non-ideal flows. Experiments in fluids 60 (101).
Zou, Aihong, Sauret, Emilie & Gu, YuanTong 2018 Numerical comparisons of conical and
annular radial diffusers performance for high density radial inflow turbines. In Proceedings of
the 21st Australasian Fluid Mechanics Conference (ed. T C W Lau & R M Kelso), pp. 1–4.
Australia: Australasian Fluid Mechanics Society.
150
NOMENCLATURE
Acronyms/Abbreviations
CCS Carbon-capture and storage

COTS Commercial Of-The-Shelf
CFD Computational fluid dynamics
CP Control points
DoE Design of experiments
DV Design Variables
EGR Exhaust Gas Recirculation
FEM Finite Element Method
GWP Global Warming Potential
HPV High Pressure Vessel
ICE Internal Combustion Engine
LE Leading edge
LPV Low Pressure Vessel
LUT Look up table
MDM Octamethyltrisiloxane
MM Hexamethyldisiloxane
ODP Ozone Depletion Potential
ORC Organic Rankine cycle
PC Polynomial chaos
R1233zd(E) trans-1-Chloro-3,3,3-trifluoropropene
R1234yf 2,3,3,3-Tetrafluoroprop-1-ene
R134a 1,1,1,2-tetrafluoroetano
R245fa 1,1,1,3,3-Pentafluoropropane
RANS Reynolds-averaged Navier-Stokes
RIT Radial Inflow Turbine
RITML Radial Inflow Turbine Mean-Line model
SLA Stereolithography
TE Trailing edge
TROVA Test Rig for Organic VApors
TP𝑖 Test with Probe in position 𝑖-th
TNP Test without the probe
𝑇−𝑠 Temperature - specific entropy
UQ Uncertainty quantification
WHR Waste Heat Recovery
151
Latin alphabet
𝐴 Area
𝐴𝑅 Area ratio
𝑏 Blade height
𝑐 Speed of sound / chord
𝐶𝑓 Friction coefficient
𝐶𝑃 Pressure recovery coefficient
𝐷 Diameter
𝐷ℎ Hydraulic diameter
𝐸 Energy factor
𝑓 Frequency
ℎ Specific enthalpy
𝐻 Shape factor
𝑖 Incidence
𝑙 Geometric length of the blade
𝐿𝑑 Diffuser axial length
𝐿ℎ Hydraulic length
𝑚¤ Mass-flow rate
𝑀 Mach number
𝑀𝑤 Relative Mach number
𝑜 opening distance
𝑃 Pressure
𝑃𝑇 ,𝑟 Non-dimensional inlet total pressure
𝑄 Heat flux
𝑄¤ Volumetric flow rate
𝑟 Radius
𝑅 Specific gas constant
𝑅2 Coefficient of determination
𝑅𝑒 Reynolds number
𝑠 Specific entropy / pitch distance
𝑆𝐹 Split factor
𝑆𝑃 Size parameter
𝑡 thickness
𝑇 Temperature
𝑈 Rotor linear speed
𝑉 Absolute velocity / Volt
𝑉𝑟 Volume ratio
𝑉𝑠 Velocity ratio, rotor linear speed over isentropic velocity
𝑊 Relative velocity / power
𝑦+ Dimensionless wall distance
𝑌𝑡𝑜𝑡 Total pressure loss coefficient
𝑍 Compressibility factor / number of blades
Greek alphabet
𝛼 Absolute angle
𝛼𝑠 Stagger angle
𝛽 Pressure ratio / Relative angle
𝛽𝑏𝑙 Blade angle
152
𝛽𝑒, 𝑃𝐸 Post-expansion / post-compression pressure ratio
𝛽𝑒,𝑇𝑆 Total-to-static pressure ratio
𝛾 Ratio of specific heats
𝛤 Fundamental derivative of gas dynamic
𝛿 Uncertainty
𝜀 Tip clearance / semi-opening angle
𝜂 Efficiency
𝜂𝑇𝑆 Total-to-static efficiency
𝜗 Boundary layer momentum thickness
𝛩 𝑐𝑎𝑚 Camber angle
𝜇 Slip factor
𝜌 Density
𝜌˜ Non-dimensional density
𝜎 Variance / covariance / stress
𝜙 𝑐𝑚 Average flow angle
𝜔 Rotational speed
Subscripts
a Axial
bl Blade
ch Choking
cl Clearance
cond Condenser
crit Critical conditions
d Diffuser / design
dev Deviation
desh De-superheater
eco Economizer
eg Exhaust gas
el Electrical
exp Experimental
eva Evaporator
f Friction
hub Hub
id Ideal
in Inlet
is Isentropic
loss Losses
m Meridional
max Maximum
min Minimum
out Outlet
p Passage
pp Pinch point
PRHE Primary heat exchange
r Rotor / radial / reduced
rec Recuperator
rms Root mean square
153
s Stator
sub Subsonic
sup Supersonic
sh Superheater / shroud
t Turbine / total / thickness / tangential
te Trailing edge
tip Tip
th Throat
TS Total-to-static
v Vaned
1 Stator inlet
2 Stator throat
3 Semi-bladed stator outlet
3∗ Stator outlet
4 Rotor inlet
5 Rotor throat
6 Rotor outlet
Mathematical symbols
𝛥 Difference (out - in)
∇ Gradient
154
LIST OF FIGURES
1.1 CO2 emission reduction by World Energy Outlook 2019 . . . . . . . . . . . . . 2

1.2 Second-law efficiency losses for Joule-Brayton gas cycles . . . . . . . . . . . . . 3
1.3 ORCs power systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Schematic RIT reference geometry . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Synoptic view of the mean-line model developed . . . . . . . . . . . . . . . . . 20
2.3 Validation over Spence’s turbine . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Validation over Mc Lallin’s turbine . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Validation over Schuster’s turbine . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 Validation over De Servi’s turbine . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7 Summary of validation procedure over test case turbines in terms of 𝜂𝑇𝑆 . . . . . 38
3.1 Domain of the three design variables . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2 Preliminary results: maximum system efficiency map . . . . . . . . . . . . . . . 48
3.3 Expansion lines for four different Z . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4 Turbine efficiency map definition . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Loss breakdown for different SP and Vr . . . . . . . . . . . . . . . . . . . . . . 54
3.6 Comprehensive analysis results for R1233zd(E) . . . . . . . . . . . . . . . . . . 57
3.7 𝑇 − 𝑠 diagram for 4 optimal non recuperative cycles . . . . . . . . . . . . . . . . 58
3.8 𝑇 − 𝑠 diagram of optimal cycles for Cyclopentane and R1234yf . . . . . . . . . . 60
4.1 TROVA plant layou . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Thermocouple T-V calibration curve . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 Calibration surface for absolute pressure sensors . . . . . . . . . . . . . . . . . . 74
4.4 Schlieren equipment sketch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Uncertainty quantification: Monte Carlo Vs PCE . . . . . . . . . . . . . . . . . 78
5.1 Test section sketch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2 Reference isentropic expansions . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Starting Vs baseline blades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4 Starting Vs baseline blade: Mach contour . . . . . . . . . . . . . . . . . . . . . 84
5.5 Baseline blade Mach contour: axial Vs 75°inlet angle . . . . . . . . . . . . . . . 85
5.6 Baseline blade properties distribution: axial Vs 75°inlet angle . . . . . . . . . . 86
5.7 Baseline geometry parametrization and optimal Vs baseline profiles . . . . . . . 86
5.8 Optimization convergence history . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.9 Optimal Vs baseline blade: Mach contour . . . . . . . . . . . . . . . . . . . . . 87
5.10 Pressure field acting on the optimal blade suction and pressure sides. . . . . . . . 88
5.11 Optimal blade FEM analysis: mesh and displacement field . . . . . . . . . . . . 89
5.12 Optimal blade FEM analysis: results . . . . . . . . . . . . . . . . . . . . . . . . 89
5.13 Final optimal blade geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.14 Final linear cascade arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.15 Streamlines in annular row configuration . . . . . . . . . . . . . . . . . . . . . . 91
5.16 2D Mesh blocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.17 2D Mesh detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
155
5.18 Full cascade CFD simulation: contours . . . . . . . . . . . . . . . . . . . . . . 93
5.19 Periodicity analysis results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.20 Periodicity analysis: blades loading . . . . . . . . . . . . . . . . . . . . . . . . 94
5.21 Mach distribution over the measuring line: off-design conditions . . . . . . . . . 94
5.22 3D CFD simulation of the full cascade: contours . . . . . . . . . . . . . . . . . 95
5.23 Mach distribution over the measuring line: 2D Vs 3D . . . . . . . . . . . . . . . 95
5.24 Test section layout with overlapped Mach field . . . . . . . . . . . . . . . . . . . 96
5.25 Test section layout with Mach field and measuring positions . . . . . . . . . . . . 97
5.26 Probe oscillations assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.27 Pneumatic lines dynamic responses . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.28 3D model of the final cobalt-chrome total pressure probe . . . . . . . . . . . . . 101
5.29 Probe-static pressure tap interaction assessment . . . . . . . . . . . . . . . . . . 102
5.30 3D model of the test section assembly: front view . . . . . . . . . . . . . . . . . 103
5.31 3D model of the test section assembly: rear view . . . . . . . . . . . . . . . . . 103
5.32 3D model of the probe insert . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.33 Filling stem and static pressure tap channel . . . . . . . . . . . . . . . . . . . . 105
6.1 Final test section arrangement and measuring positions . . . . . . . . . . . . . . 110
6.2 Repeatability of tests with nitrogen: static pressures . . . . . . . . . . . . . . . . 111
6.3 Schlieren frame with nitrogen at low inlet total pressure . . . . . . . . . . . . . . 112
6.4 Repeatability of tests with nitrogen: total pressures . . . . . . . . . . . . . . . . 112
6.5 Assessment of total pressure probe disturbance effects . . . . . . . . . . . . . . . 113
6.6 Assessment of attained periodicity with nitrogen . . . . . . . . . . . . . . . . . . 114
6.7 Nitrogen experiments: static pressure CFD comparison . . . . . . . . . . . . . . 115
6.8 Schlieren Vs Mach number field for nitrogen experiments . . . . . . . . . . . . . 116
6.9 Total pressure loss coefficient pitch-wise distribution . . . . . . . . . . . . . . . 117
6.10 Schlieren frames acquired during first campaign with MM . . . . . . . . . . . . 118
6.11 Assessment of attained periodicity in the first campaig with MM . . . . . . . . . 119
6.12 Comparison with CFD results for the first campaign with MM . . . . . . . . . . 120
6.13 Non dimensional pressure field within the test section . . . . . . . . . . . . . . . 121
6.14 Schlieren Vs density gradient field for nitrogen experiments . . . . . . . . . . . . 122
6.15 Comparison with CFD results for the latest experiments with N2 . . . . . . . . . 123
6.16 Assessment of repeatability for tests with MM . . . . . . . . . . . . . . . . . . . 124
6.17 Assessment of attained periodicity with MM . . . . . . . . . . . . . . . . . . . . 125
6.18 Schlieren Vs density gradient field for MM experiments . . . . . . . . . . . . . . 126
6.19 Comparison with CFD results for the latest tests with MM: in-channel taps . . . . 127
6.20 Comparison with CFD results for the latest tests with MM: downstream taps . . . 127
156
LIST OF TABLES
2.1 Mode-dependent inputs to the mean-line model . . . . . . . . . . . . . . . . . . 19

2.2 Test cases radial inflow turbines . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3 Rotor - passage loss models effectiveness analysis . . . . . . . . . . . . . . . . . 31
2.4 Nozzle - passage loss models effectiveness analysis . . . . . . . . . . . . . . . . 32
2.5 Fluid properties of Schuster’s Turbine. . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Loss breakdown analysis for De Servi’s test case turbine . . . . . . . . . . . . . 37
2.7 Summary of validation procedure over test case turbines in terms of 𝑚¤ . . . . . . 37
2.8 Loss models analyzed and adopted in RITML . . . . . . . . . . . . . . . . . . . 40
3.1 ORC model assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 Optimization design variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 R1233zd(E) fluid main parameters . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Optimal cycles in the preliminary analyses with R1233zd(E) . . . . . . . . . . . 48
3.5 RIT efficiency sensitivity analysis: effect of Z . . . . . . . . . . . . . . . . . . . 51
3.6 RIT efficiency sensitivity analysis: effects of working fluid . . . . . . . . . . . . 52
3.7 Coefficients of interpolating relation for turbine 𝜂𝑇𝑆 . . . . . . . . . . . . . . . . 55
3.8 Optimal results for the non-recuperative cycles with and without the map . . . . . 56
3.9 Optimal plant efficiencies for all potential working fluids . . . . . . . . . . . . . 59
5.1 Optimal Vs baseline blade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.2 Optimal deformed blade Vs baseline blade . . . . . . . . . . . . . . . . . . . . . 89
157

PHD Thesis MarcoManfredi

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PHD Thesis MarcoManfredi

Uploaded by

Copyright:

Available Formats

Department of Energy

Doctoral Program in Energy and Nuclear Science and Technology

NUMERICAL AND EXPERIMENTAL CHARACTERIZATION

Doctoral dissertation of:

September 2023– Cycle XXXV

All rights reserved.

I Mini-ORC design and characterization 11

3 Turbine Performance Map integration within ORC optimization procedure 41

II Experiments on supersonic ORC linear cascade 63

6 Experimental Campaigns 109

7 CONCLUSION AND OUTLOOK 131

List of figures 155

List of tables 157

1.1 Research motivation

Figure 1.3: (𝑎) Example of supercritical regenerative ORC implementing MM (hexamethyldis-

1.2 State of the art

1.3 Research objectives and outcomes

1.4 Outline of the thesis

2.1 Concept of the mean-line model

2.2 Mean-line method structure

Table 2.1: Mode-dependent inputs to the mean-line model.

Parameter Analysis Design

nozzle and by a supersonic converging-diverging nozzle respectively. In addition, for values of

2.2.1 Vaned nozzle

𝑚¤ = 𝜌3∗𝑉3∗ 𝑜∗3 𝑏 3 𝑍 𝑠 , (2.1a)

2.2.2 Semi-bladed Nozzle

𝑚¤ = 𝜌3𝑉3 cos 𝛼3 2𝜋𝑟 3 𝑏 3 , (2.2a)

2.2.3 Vaneless space

𝑚¤ = 𝜌4𝑉4 cos 𝛼4 2𝜋𝑟 4 𝑏 4 , (2.3a)

2.3 Loss models

2.3.1 Nozzle losses

for uncambered blades, and as

2.3.2 Vaneless Space losses

2.3.3 Rotor losses

Author Label Ref. Data fluid 𝛽𝑒,𝑇𝑆 𝑊¤ 𝑡 [kW] Nozzle

2.4 Test case radial inflow turbines

2.5 Loss sensitivity analyses

2.6 Validation strategy

Table 2.5: Fluid properties of Schuster’s Turbine.

¤ values. Black and gray

¤ for the subsonic turbine test cases whose validation

2.7 Concluding remarks and key findings

Table 2.8: Loss models analyzed and adopted in RITML.

FINAL SET OF CHOSEN CORRELATIONS

3.2 Numerical routines description

3.2.1 ORC design and optimization tool

𝛥𝑇 heat exchangers Heat exchangers pressure drops

𝑂𝑏 𝑗 𝐹𝑢𝑛𝑐. : 𝑊𝑒𝑙 = 𝑊𝑔𝑒𝑛 − 𝑊 𝑝𝑢𝑚 𝑝 − 𝑊 𝑓 𝑎𝑛 , (3.2a)

Table 3.2: Optimization procedure DVs and corresponding variation ranges.

DVs Variation Ranges DVs Variation Ranges

3.2.2 RIT optimization strategy

Table 3.3: R1233zd(E) fluid main parameters.

𝑇𝑐𝑟𝑖𝑡 , °C 𝑃𝑐𝑟𝑖𝑡 , bar 𝜌 𝑐𝑟𝑖𝑡 , kg/m3 𝑇 Max, °C Molar Mass, kg/kmol

3.3 Preliminary considerations on ORC design

Cycle Base Regenerative (𝛥𝑇 𝑝 𝑝,𝑟 𝑒𝑐 =2°C)

3.4 Assessment of SP and Vr as unique independent parameters

3.4.1 Sensitivity analysis of optimum RIT efficiency: effects of real-gas compressibility

Parameter Turbine X Turbine Y Turbine K Turbine W

3.4.2 Sensitivity analysis of optimum RIT efficiency: effects of 𝛤 at fixed 𝑍

Parameter R1233zd R134a Cyclopentane MM

3.5 RIT maximum efficiency map definition

Cumula ve enthalpy loss

• An increment of Vr leads to an efficiency penalization mainly due to an increase of stator