Variable-speed centrifugal compressors are used in a wide variety of industries and applications. Performance maps describe how a compressor's head and power vary with volumetric suction fowrate and rotational speed for a specifc set of suction conditions. Integrating the turbomachinery controls into the dynamic simulation of a compressor system offers many tangible and intangible benefts.
Variable-speed centrifugal compressors are used in a wide variety of industries and applications. Performance maps describe how a compressor's head and power vary with volumetric suction fowrate and rotational speed for a specifc set of suction conditions. Integrating the turbomachinery controls into the dynamic simulation of a compressor system offers many tangible and intangible benefts.
Variable-speed centrifugal compressors are used in a wide variety of industries and applications. Performance maps describe how a compressor's head and power vary with volumetric suction fowrate and rotational speed for a specifc set of suction conditions. Integrating the turbomachinery controls into the dynamic simulation of a compressor system offers many tangible and intangible benefts.
Variable-speed centrifugal compressors are used in a wide variety of industries and applications. Performance maps describe how a compressor's head and power vary with volumetric suction fowrate and rotational speed for a specifc set of suction conditions. Integrating the turbomachinery controls into the dynamic simulation of a compressor system offers many tangible and intangible benefts.
Computational Methods V ariable-speed centrifugal compressors are used in a wide variety of industries and applications, including natural gas pipelines and processing plants, oil refn- eries and chemical plants, air-separation units, refrigeration and air conditioning equipment refrigerant cycles, gas tur- bines and auxiliary power generators, and many more. Some natural gas plants, for example, employ a propane-precooled, mixed-refrigerant natural gas liquefaction process consisting of a classical propane liquefaction cycle that precools both the natural gas feed and the mixed refrigerant followed by a mixed-refrigerant liquefaction cycle that provides the low- temperature refrigeration needed to liquefy the natural gas. Compression is a key element of both refrigeration cycles. Dynamic process simulation offers many tangible and intangible benefts (1). Integrating the turbomachinery controls (TMC) into the dynamic simulation of a compressor system enables the engineer to select properly sized equip- ment that can cope with transient conditions (surge conditions in particular), reduces commissioning time through pre tuning of the regulatory controls, improves operator training by exposing operators to realistic simulated operating scenarios, and provides verifed startup and shutdown procedures during site acceptance testing before startup takes place (2). Dynamic simulators for modeling processes that involve gas compression require a rigorous unit-operation model of a centrifugal compressor. The performance maps provided by compressor manufacturers are a critical element of this model. This article explains how to use the performance maps for variable-speed, fxed-inlet guide-vane centrifugal compressors for the dynamic simulation of a compressor system. Performance maps and their use in simulation A performance map describes how a compressors poly- tropic head and power vary with volumetric suction fowrate and rotational speed for a specifc set of suction condi- tions (i.e., fuid molecular weight, pressure, temperature, compressibility, and isentropic exponent at the inlet of the compressor). Manufacturers typically provide performance maps as two sets of performance curves a series of plots of polytropic head vs. volumetric suction fowrate, and a cor- responding series of power vs. fow plots. Each pair of curves corresponds to a different rotational speed, and all pertain to the same suction conditions. Performance maps may be constructed for more than one set of suction conditions. Figure 1 presents two performance maps for a high-pressure mixed-refrigerant compressor for a natural gas liquefaction process. The top performance map corresponds to design operation, and the bottom map corresponds to derime operation, which removes rime (the granular ice that forms when supercooled droplets freeze rapidly on contact with a cold surface) from the compressor. These maps were generated from data obtained by manually digitizing the performance maps provided by the manufac- turer; the irregularities in the curves are a result of the manual digitization process. Process simulators combine the performance maps with material and energy balances and thermodynamic relation- ships to predict the performance of the compressor under various operating scenarios. However, the performance maps supplied by the manufacturer are not broadly useful for process simulation because, strictly speaking, they apply only at the suction conditions and rotational speeds for which they Transforming the performance maps to a reduced coordinate system that is independent of suction conditions and rotational speeds allows these curves to be accurately incorporated into a process simulator. Grant Stephenson Honeywell Process Solutions Integrate Compressor Performance Maps into Process Simulation Copyright 2011 American Institute of Chemical Engineers (AIChE) CEP June 2011 www.aiche.org/cep 43 were constructed. Process simulation requires prediction of the compressors performance not just at these suction condi- tions and speeds, but over the compressors entire operating range, from surge to stonewall, at varying suction conditions and rotational speeds. Some process simulators ignore the fact that the perfor- mance maps relate to specifc suction conditions, or allow users to enter multiple performance maps in tabular form and synthetically transition between them as the suction condi- tions change. However, as shown in Figure 2, the perfor- mance maps for different suction conditions can be quite different. In this instance, the scales for power differ by more than an order of magnitude. Using the performance map for design operation at all operating conditions will certainly yield poor results for derime operation. Varying rotational speed can be handled somewhat bet- ter than varying suction conditions, for example, by linear interpolation and extrapolation on an irregularly spaced nonrectangular grid. Although linear interpolation can yield reasonable results, linear extrapolation breaks down as the rotational speed moves further away from the lowest and highest speeds for which performance curves were con- structed. This limits the utility of the performance maps to normal operating speeds. Volumetric flowrate, cfm 0 5,000 10,000 15,000 20,000 25,000 0 5,000 10,000 15,000 20,000 25,000 30,000 3,760 rpm 3,850 rpm 3,950 rpm 4,000 rpm P o l y t r o p i c
H e a d ,
f t
Design Operation 0 5,000 10,000 15,000 20,000 25,000 30,000 0 5,000 10,000 15,000 20,000 25,000 30,000 P o w e r ,
h p
Volumetric flowrate, cfm 3,760 rpm 3,850 rpm 3,950 rpm 4,000 rpm Design Operation 0 5,000 10,000 15,000 20,000 25,000 0 5,000 10,000 15,000 20,000 25,000 P o l y t r o p i c
H e a d ,
f t
3,760 rpm 3,850 rpm 3,960 rpm 4,000 rpm Volumetric flowrate, cfm Derime Operation 0 100 200 300 400 500 600 700 800 900 0 5,000 10,000 15,000 20,000 25,000 P o w e r ,
h p 3,760 rpm 3,850 rpm 3,960 rpm 4,000 rpm Volumetric flowrate, cfm Derime Operation p Figure 1. These performance maps represent design (top) and derime (bottom) operation of a high-pressure, mixed-refrigerant compressor. 0 5,000 10,000 15,000 20,000 25,000 0 5,000 10,000 15,000 20,000 25,000 30,000 P o l y t r o p i c
H e a d ,
f t
Volumetric flowrate, cfm Design Derime Design and Derime Operation 0 5,000 10,000 15,000 20,000 25,000 30,000 0 5,000 10,000 15,000 20,000 25,000 30,000 P o w e r ,
h p Design Derime Volumetric flowrate, cfm Design and Derime Operation p Figure 2. Superimposing performance maps reveals that the power vs. suction owrate relationships for design and derime operation are very different. Article continues on next page Copyright 2011 American Institute of Chemical Engineers (AIChE) 44 www.aiche.org/cep June 2011 CEP Computational Methods If the unit-operation model of the compressor does not adequately account for variation in the suction conditions and rotational speed, the predictions of the process model in which it is used for example, the speed and shaft power needed to achieve a specifed (controlled) discharge pres- sure will certainly be fawed, and may even be misleading. If the process model is used in a dynamic simulation study, the trajectories for the operating variables will be incorrect, possibly signifcantly, which could lead to false conclusions. Simulation-based training using a model that misrepresents the actual process behavior could cause an operator to make incorrect operating decisions. Transforming the performance maps The challenge, therefore, is to fnd a way to make the performance maps applicable over the entire range of suc- tion conditions and speeds encountered in the operation of the compressor. This can be achieved by transforming the performance maps to a coordinate system that is independent of both suction conditions and rotational speed. This transfor- mation involves two steps. First, dimensional analysis is used to construct a coordi- nate system that is independent of suction conditions (3). A key aspect of this analysis is the identifcation of a set of dimensionless parameters in which the pertinent forces can be accurately represented. Functional relationships are set up in which polytropic head (H p ) and shaft power (P) are the dependent variables and volumetric fowrate (Q) and rotational speed (w) are the independent variables. The suction conditions and the compressor size are incorporated into the functions by includ- ing viscosity (m), density (r), and acoustic velocity (a) at the suction of the compressor and the diameter (d) of the com- pressor as additional independent variables. These functional relationships are represented as: H p = f 0 (Q, w, m, r, a, d) (1) P = f 1 (Q, w, m, r, a, d) (2) Each of these relationships involves seven dimensional parameters. Using the Buckingham Pi Theorem and assum- ing turbulent fow, they can be simplifed to: H p,n = f 2 (Q n , w) (3) P n = f 3 (Q n , w) (4) where the subscript n denotes a dimensionless variable. Although this dimensionless coordinate system was developed for compressor control, it is equally applicable to process simulation. Next, a modifed form of the Fan Laws is applied to further transform the dimensionless coordinate system to one that is also independent of rotational speed. The Fan Laws of centrifugal pumps or fans, also known as the Affnity Laws, express the infuence of changes in rotational speed, diameter, and density on volumetric fowrate, head (H) (or pressure), and shaft power. When the fan diameter and the fuid density at the fan suction are constant, the Fan Laws can be expressed as: Q 1 /Q 2 = w 1 /w 2 (5a) H 1 /H 2 = (w 1 /w 2 ) 2 (5b) P 1 /P 2 = (w 1 /w 2 ) 3 (5c) It follows from these relationships that when H/w 2 and P/w 3 are plotted against Q/w, the performance curves for a given set of suction conditions reduce to a single curve for head and a single curve for power. However, because the Fan Laws do not strictly apply to compressors, a modifed form of the Fan Laws must be used. The resulting coordinate system, referred to as the reduced coordinate system, has as its dependent variables the transformed (reduced) polytropic head (H p,r ) and reduced power (P r ), and as its independent variable the reduced volu- metric suction fow (Q r ). In the reduced coordinate system, the performance curves for polytropic head and for power at different suction conditions are each represented by a single p Figure 3. A reduced performance map aggregates a series of performance curves into a single curve in the reduced coordinate system. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 Reduced Volumetric Suction Flow R e d u c e d
P o l y t r o p i c
H e a d Design Derime 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 R e d u c e d
P o w e r Reduced Volumetric Suction Flow Design Derime Copyright 2011 American Institute of Chemical Engineers (AIChE) CEP June 2011 www.aiche.org/cep 45 aggregate curve, as shown in Figure 3. These aggregate curves are expressed as: H p,r = f 4 (Q r ) (6) P r = f 5 (Q r ) (7) Transforming the performance maps to the reduced coordinate system is very powerful. Consider the following modeling scenario. Given the suction conditions, rotational speed, and volumetric suction fow for a particular operat- ing condition (real or simulated), the reduced volumetric suction fow is calculated by applying the transforma- tions, calculating frst Q n and w n , and then Q r . The reduced polytropic head and reduced power corresponding to the reduced volumetric suction fow are then determined from the aggregate performance curves, i.e., from f 4 (Q r ) and f 5 (Q r ), respectively. The polytropic head and power in engineering units are then easily calculated by inverting the transformations, frst calculating H p,n and P n , and then H p
and P. Transforming the performance maps provided by the compressor manufacturer, which strictly apply only at their corresponding suction conditions and rotational speeds, to reduced coordinates in this way enables performance curves for polytropic head and power that are particular to different operating conditions (i.e., suction conditions and rotational speed) to be constructed. To make a reduced performance map (i.e., the aggre- gate curves of reduced polytropic head and reduced power) more useful, the aggregate curves need to be represented by appropriate functions whose parameters are ftted to the reduced data. Functional representation of the reduced performance map Based on a visual inspection of Figure 3, a quadratic function appears to be a good choice for representing the aggregate performance curves for both polytropic head, f 4 (Q r ), and power, f 5 (Q r ). However, experience has proven otherwise. A quadratic function is not a suitable representa- tion of the aggregate power curve, which is not symmetric about the perpendicular line through its apex. In addition, a quadratic function ftted to the data typically will not extrap- olate through the origin a necessary physical constraint. A quadratic function is also not a suitable representation of the aggregate curve of polytropic head at low values of reduced volumetric suction fow, where it becomes nearly linear. The next likely choice of functional form would be a higher-order polynomial. However, high-order polynomials can wobble, even within the range of the data being ftted, and they may not extrapolate well both signifcant disad- vantages for process simulation. Experience has shown that a very good ft is obtained when the aggregate performance curves are divided into three segments representing the surge, normal, and stone- wall regions of the performance curves and the segments are represented by the functions listed in Table 1. The functions parameters and the segments end points are determined by ftting the curves to the reduced data. This yields smooth, piecewise-ftted curves, as shown in Figure 4. The functions ft the curve segments well, and they extrapolate well for both low and high reduced volumetric suction fows. Although fowrates below the reduced surge fow are not physically meaningful, reasonable extrapola- tion to low fows is required to ensure robust solution of the compressor unit-operation model. Reasonable extrapolation in the stonewall region facilitates robust solution up to the stonewall fow. Design Derime Fitted 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.2 0.4 0.6 0.8 R e d u c e d
P o l y t r o p i c
H e a d Reduced Volumetric Suction Flow 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.2 0.4 0.6 0.8 R e d u c e d
P o w e r Reduced Volumetric Suction Flow Design Derime Fitted p Figure 4. The surge, normal, and stonewall operation segments of the aggregate performance curves can be represented by different functions. Table 1. The aggregate performance curves can be divided into segments, each represented by a different functional form. Curve Segment Polytropic Head Power Surge Exponential Quadratic Normal Power Power Stonewall Quadratic Quadratic Article continues on next page Copyright 2011 American Institute of Chemical Engineers (AIChE) 46 www.aiche.org/cep June 2011 CEP Computational Methods Using the reduced performance map in dynamic process simulation As stated previously, accurately simulating the behavior of processes that include centrifugal compressors requires process simulators that rigorously model these devices over their entire range of suction conditions and rotational speeds, from surge through stonewall operation. Figure 5 is a typical compressor loop fowsheet devel- oped using such a simulator. Figure 6 presents details for compressor unit operation K-100, which has been confgured as a centrifugal compressor using the reduced coordinate sys- tem representation of the performance maps. The polytropic head performance map for this compressor, Figure 7, was constructed from the reduced aggregate performance curve for polytropic head at the existing suction conditions. The speeds for which performance curves are plotted were speci- fed by the modeling engineer using the speed curves selector on the right-hand side of the window. The current operating point, represented by the red dot, is superimposed on the performance map. In dynamic simula- tion, the operating point moves as the suction conditions, rotational speed, and/or discharge pressure vary. For the purpose of illustration, a transfer function (the K-100 speed block of the fowsheet) was applied to cause the rotational speed to oscillate, as depicted by the green line on the strip chart in Figure 8. The red and blue lines in Figure 8 represent the dynamic responses of the volumetric suction fow and the polytropic head. 15 17 1 3 4 6 5 7 10 11 12 13 14 8 9 16 2 VLV-103 VLV-100 VLV-102 VLV-101 V-101 MIX-100 MIX-101 K-100 E-100 V-100 TEE-100 RCY-1 R PIC-100 LIC-100 Q-100 Q-101 K-100 Speed K-100 Surge Controller p Figure 5. This compressor loop owsheet was created by a process simulator that rigorously models compressors over their entire operating range. p Figure 6. The Parameters page of the Design tab for compressor K-100 provides design details for this equipment. 3.5e+004 3.0e+004 2.5e+004 2.0e+004 1.5e+004 1.0e+004 5.0e+004 0.00 0 5,000 10,000 15,000 20,000 25,000 H e a d ,
m Flowrate, actual m 3 /h Head Curves 2,800 rpm 3,200 rpm 3,400 rpm 2,800 rpm Operating Point p Figure 7. Compressor K-100s performance map for polytropic head was constructed from the reduced aggregate polytropic head curve at the existing suction conditions, at various speeds. Copyright 2011 American Institute of Chemical Engineers (AIChE) CEP June 2011 www.aiche.org/cep 47 In addition to integrating reduced performance maps into its mathematical model of the centrifugal compressor unit operation, it is crucial that the process simulator also provide a tool to assist the engineer in transforming the dig- itized performance curves into the reduced coordinate sys- tem, that is, in creating the aggregate performance curves and ftting the parameters of the functions that represent these curves. Without such a tool, this task would simply be too daunting, and the value associated with the reduced coordinate system representation of the performance maps would not be realized. 1.800e +004 1.500e +004 1.200e +004 9,000 6,000 2,995 3,000 3,360 (rpm) 2.152e+004 (m) 1.328e+004 (m 3 ) 3,005 V o l u m e t r i c
F l o w r a t e ,
m 3 / h Minutes p Figure 8. This strip chart records the dynamic responses of polytropic head (blue) and volumetric suction owrate (red) to variations in rotational speed (green). Literature Cited 1. Stephenson, G., et al., Proft More from Process Simulation, Chem. Processing, 72 (8), pp. 2326 (Aug. 2009). 2. Willetts, I., and A. Nair, Using High-Fidelity Dynamic Simu- lation to Model Compressor Systems, Chem. Eng. Progress, 106 (4), pp. 4448 (Apr. 2010). 3. Batson, B. W., Invariant Coordinate Systems for Compressor Control, presented at the International Gas Turbine and Aeroen- gine Congress and Exhibition, Birmingham, U.K. (June 1996). GRANT STEPHENSON is an Engineering Fellow with Honeywells Automa- tion Control Solutions business, where he serves as the global process simulation architect for Honeywell Process Solutions. He has worked in the eld of process simulation for more than 35 years, with particular interest in dynamic simulation, equation-oriented modeling and simultaneous solution of owsheet models, and the application of modeling and optimization to plant operations. He is the originator of the dynamic simulation engine of the Shadow Plant simulator and is a pioneer of the hybrid solution architecture and its application to large-scale dynamic simulation. Before joining Honeywell, he held positions with DuPont Canada, Atomic Energy of Canada, the Univ. of Western Ontario (in the engineering facultys Systems Analysis Control and Design Activity, SACDA, group), and SACDA Inc. He holds an MSc degree in applied mathematics from the Univ. of Western Ontario. CEP Copyright 2011 American Institute of Chemical Engineers (AIChE)