applied

sciences
Article
Research on Model-Based Fault Diagnosis for a Gas
Turbine Based on Transient Performance
Detang Zeng 1,2, *, Dengji Zhou 3,4, *, Chunqing Tan 1,2 and Baoyang Jiang 4
1 Institute of Engineering Thermophysics, Chinese Academy of Science, Beijing 100190, China; tan@iet.cn
2 University of Chinese Academy of Sciences, Beijing 100049, China
3 The Key Laboratory of Power Machinery and Engineering of Education Ministry,
Shanghai Jiao Tong University, Shanghai 200240, China
4 Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109, USA;
byjiang@umich.edu
* Correspondence: zengdetang@iet.cn (D.Z.); zhoudj@sjtu.edu.cn (D.Z.)

Received: 7 November 2017; Accepted: 26 December 2017; Published: 22 January 2018

Abstract: It is essential to monitor and to diagnose faults in rotating machinery with a high
thrust–weight ratio and complex structure for a variety of industrial applications, for which reliable
signal measurements are required. However, the measured values consist of the true values of the
parameters, the inertia of measurements, random errors and systematic errors. Such signals cannot
reflect the true performance state and the health state of rotating machinery accurately. High-quality,
steady-state measurements are necessary for most current diagnostic methods. Unfortunately, it is
hard to obtain these kinds of measurements for most rotating machinery. Diagnosis based on
transient performance is a useful tool that can potentially solve this problem. A model-based
fault diagnosis method for gas turbines based on transient performance is proposed in this paper.
The fault diagnosis consists of a dynamic simulation model, a diagnostic scheme, and an optimization
algorithm. A high-accuracy, nonlinear, dynamic gas turbine model using a modular modeling method
is presented that involves thermophysical properties, a component characteristic chart, and system
inertial. The startup process is simulated using this model. The consistency between the simulation
results and the field operation data shows the validity of the model and the advantages of transient
accumulated deviation. In addition, a diagnostic scheme is designed to fulfill this process. Finally,
cuckoo search is selected to solve the optimization problem in fault diagnosis. Comparative diagnostic
results for a gas turbine before and after washing indicate the improved effectiveness and accuracy of
the proposed method of using data from transient processes, compared with traditional methods
using data from the steady state.

Keywords: fault diagnosis; gas turbine; transient performance; cuckoo search; modeling and
simulation; deterioration

1. Introduction
With the deepening of the industrialization process, the revolution of industry 4.0 is emerging in
many industry areas [1]. A variety of intelligent equipment and intelligent equipment management
technologies have been created, so as to redefine the interaction between humans and machines [2].
With the increased complexity of equipment, fault diagnosis has become a key technology to ensure
the safety and efficiency of the production process [3]. Smart fault diagnostic technologies will be
essential in the age of industry 4.0. With the development of sensing and information technology,
massive multi-source data can be obtained automatically from equipment, which will become the
foundation of developing advanced fault diagnosis methods [4].

Appl. Sci. 2018, 8, 148; doi:10.3390/app8010148 www.mdpi.com/journal/applsci

Appl. Sci. 2018, 8, 148 2 of 14

Gas turbines are widely used in the defense and energy industry. It is essential to perform fault
diagnosis on a regular basis to maintain the reliability of gas turbine. However, fault diagnosis is
generally a challenging task due to the structural complexity, which involves various fault modes [5].
In traditional diagnostic processes [6,7], measurement data represent the steady-state values, while the
trending patterns in the transient phase are neglected. However, it is hard to determine steady-state
values in the transient phase. For example, the time for military aircrafts fighting in unsteady
conditions, deviating from steady states, accounts for 70% of the total fight time [8]. It is not feasible to
obtain the steady-state data for a military aircraft. Additionally, the effects appear to be remarkable
in transient processes compared to the steady-state condition for some kinds of fault modes [9,10].
Compared with the study of fault diagnosis with steady-state data, this lacks the study of fault
diagnosis with transient process data. Therefore, the objective of this thesis is to study the fault
diagnosis of gas turbines based on transient process data.
Many gas turbine performance-analysis-based diagnostic technologies have been developed
since Urban [11] introduced the first gas path analysis method in 1967. Gas turbine engine
fault diagnosis during transient processes was first analyzed by Merrington [9]. Li developed
a non-linear-model-based diagnostic method, combined with a genetic algorithm, and applied it
to a model gas turbine engine to diagnose engine faults by using the accumulated deviation obtained
from transient measuring data [8]. Naderi proposed a data-driven fault diagnosis and estimation
scheme. Fault detection, isolation and, estimations filters was developed using system Input/Output
(I/O) data in operation [12]. Meher-Homji [13] provided an overview of the use of both performance
and mechanical transient analysis as a means to detect gas turbine problems, to express the importance
of transient analysis. Henry [14] utilized engine data acquired during takeoff to trend the performance
of a modern turbofan engine. Analytical redundancy methods have been applied to gas turbine
engine transient data with a view to extracting the desired fault information [15]. Simani and Fantuzzi
presented a model-based procedure exploiting analytical redundancy for the detection and isolation of
faults on a gas turbine process, which integrated linear model identification and output estimation [16].
The system was used in a single-shaft industrial gas turbine plant to experimentally verify the
robustness of the solution obtained [17]. Lu proposed a data hierarchical fusion method, using
improved, weighted Dempster–Shaffer evidence theory, to integrate data-driven and physics-based
models for the gas-path fault diagnosis of engines [18].
Dynamic, process-based gas turbine fault diagnosis requires special parameters, which can
describe the dynamic process of gas turbines. Due to the accumulative effects of the transient process,
it will be more sensitive to choosing transient process data than steady-state data for fault diagnosis.
However, current research results cannot support the fault diagnostic technology for a gas turbine
based on transient process data. The following issues still remain unsolved:

• The accuracy of the algorithm should be increased;

• The diagnostic result based on steady-state data and dynamic data should be compared;
• The diagnostic method based on transient process data should be used to analyze field data.

In this paper, a novel gas-path fault diagnostic model is proposed, by engine transient performance
analysis and a system dynamic model. Similar to the conventional gas-path fault diagnostic method,
the proposed method has the same basic idea that a surrogate model stands for the real engine under
different working conditions and health conditions. However, due to the change of input parameters
from steady-state to transient process, the three key parts of the diagnostic model, which are calculation
flowchart, simulation model, and algorithm, should be modified. Integrated deviations between
measurements and simulation results are chosen as the fitness function in the diagnostic problem.
Then this problem will be solved by the proposed calculation flowchart and heuristic algorithm.
Appl. Sci. 2018, 8, 148 3 of 14

2. Methodology
To promote diagnosis based on the transient process data from theoretical study to commercial
application, a novel diagnostic method based on a dynamic simulation model and Cuckoo
Appl. Sci. 2018, 8, x FOR PEER REVIEW
search
3 of 14
algorithm
Appl. Sci.is2018,
proposed. A comparative
8, x FOR PEER REVIEW study to steady-state analysis and application to analyze 3 of 14 field
data algorithm
are conducted based A
is proposed. oncomparative
this novel study
method.
to steady-state analysis and application to analyze field
data are conducted
algorithm based
is proposed. on this novelstudy
A comparative method.
to steady-state analysis and application to analyze field
2.1. Modeling of Gas Turbine
data are conducted based on this novel method.
2.1. Modeling of Gas Turbine
The dynamic system of a gas turbine plant consists of working fluid and rotary parts. The system
2.1. Modeling of Gas Turbine
behavior Thecan dynamic
be derived system of a gas turbine
by conservation plant
laws and consists of working
the equations of fluid
motion and[19].
rotaryToparts. The the
analyze system
dynamic
behavior can be derived
The dynamic system ofbya conservation lawsconsists
gas turbine plant and the equationsfluid
of working of motion [19].parts.
and rotary To analyze the
The system
characteristics of a gas turbine, unsteady three-dimensional calculations can be used. However,
dynamic
behavior cancharacteristics
be derivedofbya conservation
gas turbine, laws
unsteady three-dimensional
and the equations of motioncalculations
[19]. Tocan be used.
analyze the
applying the unsteady three-dimensional simulation requires vast computational resources. It has
However, applying theofunsteady
dynamic characteristics three-dimensional
a gas turbine, simulation requires
unsteady three-dimensional vast computational
calculations can be used.
beenresources.
proved that
However, hasone-dimensional
Itapplying
been proved simulation
that one-dimensional
the unsteady gives sufficiently
simulation
three-dimensional givesaccurate
simulation requiresresults
sufficiently [20].
accurate In this
results
vast computational [20].study,
the modular
In
resources. modeling
this study, thebeen
It has method
modular is applied
modeling
proved that methodtoisbuild
one-dimensional appliedthe model.
to build
simulation the A
gives two-shaft
model. industrial
A two-shaft
sufficiently accurate gas[20].
industrial
results turbine,
gas
which In consists
turbine, which
this study, ofthe
one compressor,
consists
modularof one onemethod
combustor,
compressor,
modeling one one to
high
combustor,
is applied pressure
one
build high turbine,
pressure
the model. and and
turbine,
A two-shaft one power
one turbine,
power
industrial gas
is cited as anwhich
turbine, isexample forexample
citedconsists
as an gas
of turbine
one for gasmodelling
turbine
compressor, and fault
onemodelling
combustor, anddiagnosis,
onefault as shown
highdiagnosis,
pressure in Figure
as shown
turbine, andin one1.power
Figure 1.
turbine, is cited as an example for gas turbine modelling and fault diagnosis, as shown in Figure 1.

Figure1.1.Gas
Figure Gasturbine
turbine configuration
configurationofof
this case
this study.
case study.
Figure 1. Gas turbine configuration of this case study.
Based on the principles of the modular modeling method, the entire plant is divided into four
Based
parts: on the
Based principles
compressor,
on the ofofthe
combustor,
principles themodular
turbine modeling
(high pressure
modular method,
modelingturbine and
method, the
power
the entire plant
turbine)
entire plant isrotating
isand divided
divided into four
intoshaft.
four
parts:The
compressor,
parts:input combustor,
and output
compressor, combustor, turbine
parametersturbine (high
of each pressure
module
(high need
pressure turbine and
to beand
turbine power
defined.
powerThe turbine)
modeling
turbine) and rotating
and flowchart and shaft.
rotating shaft.
The input
modular and output
The inputinputs parameters
and outputs
and output of
are shown
parameters each module
module2.need to be defined. The modeling flowchart and and
in Figure
of each need to be defined. The modeling flowchart
modular
modularinputs andand
inputs outputs are
outputs areshown
shownin inFigure 2.
Figure 2.
Qf
Qf
Qt
Turbine
Qt
Combustor
Turbine
Combustor
QC,T2,P2 Qp
QC,T2,P2 Qp
T1,P1,n1
P3,T3,n1 P34,T34,n2
T1,P1,n1 Compressor Turbine P4,T4
P3,T3,n1 P34,T34,n2 Power Turbine
Compressor Turbine PP4p,T4
Power Turbine
Pp
Pc Pt
Pc n1 Pt
Rotating
n1 Shaft
Rotating Shaft
Figure 2. Modeling flowchart based on the modular modelling method for gas turbine.
Figure
Figure 2. Modeling
2. Modeling flowchartbased
flowchart based on
on the
themodular
modularmodelling
modelling method
methodfor gas
for turbine.
gas turbine.
Compressor and Turbine. The compressor and two turbines (high pressure turbine and power
turbine) are modelled
Compressor using theThe
and Turbine. characteristic
compressormap andthat
twodescribes the correlation
turbines (high between
pressure turbine and pressure
power
Compressor
ratio,
turbine)mass
areflow and
modelled Turbine.
rate, rational
using the The compressor
speed, andthat
and efficiency.
characteristic map twodescribes
For turbinescompressor
example, (high
the pressure
speed
correlation turbine and power
and isentropic
between pressure
turbine)
ratio,are
efficiencymodelled
masscan
flow using
berate, the characteristic
determined
rational using the
speed, map thatFor
andcharacteristic
efficiency. describes
map. the
Since
example, correlation
the between
design pressure
compressor speed pressure
andratio of the ratio,
isentropic
masscompressor
flow rate,can
efficiency rational
is be
known, speed,
determined and
and the efficiency.
mass
using flow
the For of
rate example,
the map.
characteristic compressor
air is determined
Since speed and
frompressure
the design isentropic
the plant ofefficiency
operation
ratio the
requirement,
compressor
can be determined ismainly
known,
using referring
and thetomass
power
the characteristic output,
flow the
rate of
map. compressor
the
Since air
the speedpressure
is design
determined andfrom
isentropic efficiency
the plant
ratio of can
operation
the compressor
be read off themainly
requirement, characteristic
referring map.
to power output, the compressor speed and isentropic efficiency can
be read off the characteristic map.
Appl. Sci. 2018, 8, 148 4 of 14

is known, and the mass flow rate of the air is determined from the plant operation requirement,
mainly referring to power output, the compressor speed and isentropic efficiency can be read off the
characteristic map.
The following lists their performances:
 √ 
Q Tin  p 
, η = f π, n/ Tin (1)
pin

where π, n and η stand for pressure ratio, rotational speed and component efficiency, and Tin is the
inlet temperature. From steady-state characteristics map generated by the field operating data [21],
compressor mass flow Gc and component efficiency η can be calculated by Equation (1), which can be
used to calculate the power required and the temperature of the discharge air.
 R

cp
π − 1
T2 = 1 +  T1 (2)
ηc

Pc = Gc C p ( T2 − T1 ) (3)

where Cp is the heat capacity and R is the gas constant.

The degradation of gas path components can be mathematically determined by scaled component
maps. It is assumed that the degraded maps of the compressor, combustor, and turbine are down-scaled
versions of the corresponding original maps, due to the fact that the geometries do not change
significantly. More details on the definitions of the degradation parameters of gas turbine components
can be found in reference [22,23].
Combustor. The combustor model involves the prediction of the dynamic response of pressure
and temperature inside the combustor. The pressure is determined by the combustor model, while the
flow rate is determined by the compressor and turbine model using the characteristics maps. Therefore,
the state equation of pressure can be written as:

dp R g Tg  
= G f + Gc − Gt (4)
dt V
where the subscripts f, c and t stand for the fuel, compressor and turbine, respectively. V is the volume
of the combustor, t is the time.
The turbine inlet temperature is the most important parameter for the gas turbine, which is also
calculated in the combustor model. The state equation of the turbine inlet temperature can be written
as follows:    
dTg G f h f + HV + Gc hc − Gt ht − (ht − Rt T3 ) G f + Gc − Gt
= (5)
dt ρ g Vc p,g
where HV stands for the heating value of the fuel, h is the enthalpy, cp,g is heat capacity, and ρg is the
gas density in the combustor.
The turbine inlet pressure and temperature are both calculated through the state equations.
The results are sent to the compressor and turbine model to obtain the pressure ratio for characteristics
maps and to calculate the output power.
Rotating Shaft. A rotor is used to connect the compressor and load to the turbine. The mode is
described by the following equation.

dn 900
= ( Pt − Pc ) (6)
dt nIπ 2
where I is the moment of inertia, Pt is input power from turbine, n is the rotational speed, and Pc is the
output power to drive the compressor.
Appl. Sci. 2018, 8, 148 5 of 14

The fuel flow rate is calculated by a proportional–integral–derivative (PID) controller. The input
of this controller is the deviation between the set rotational speed and the measuring rotational speed.
AfterAppl. Sci. 2018, 8, x FOR PEER REVIEW
parameter 5 of 14
tuning of the PID controller, this dynamic model can simulate the measurement
parameters of this gas turbine, setting the power condition and ambient condition. The Block diagram
parameters of this gas turbine, setting the power condition and ambient condition. The Block diagram
for the
for speed
the speed control is shown
control is shownininFigure
Figure 3.
3.

Figure3.
Figure 3. Block
Block diagram
diagramfor
forthe speed
the control.
speed control.

2.2. Cuckoo Search Algorithm

2.2. Cuckoo Search Algorithm
Cuckoo search (CS) was put forward on the basis of swarm intelligence technology by British
Cuckoo
scholar search
X.S. Yang (CS)
and S. wasDebput forward
in 2009 [24]. on
A CSthealgorithm
basis of simulates
swarm intelligence technology
the brood parasitism byofBritish
habit
scholar X.S.species
certain Yang and S. Debtoinsolve
of cuckoo 2009 the
[24].optimization
A CS algorithm
problem simulates theThe
effectively. brood parasitism
structure of a CShabit of
certain speciesisof
algorithm cuckoo
simple. to solve
It has the optimization
a few control parametersproblem effectively.
and a strong ability to The
jumpstructure ofextremum.
out of local a CS algorithm
It shows
is simple. thata few
It has in many
control cases, a CS algorithm
parameters shows ability
and a strong better performance
to jump out of compared to a genetic
local extremum. It shows
algorithm (GA), artificial bee colony (ABC) algorithm, particle swarm optimization
that in many cases, a CS algorithm shows better performance compared to a genetic algorithm (PSO) algorithm (GA),
and some
artificial other typical
bee colony (ABC) swarm intelligence
algorithm, algorithm
particle swarm[25,26]. In this paper,
optimization the CSalgorithm
(PSO) algorithm was
andchosen
some other
to solve this problem, as it features high computational efficiency and accurate optimization results.
typical swarm intelligence algorithm [25,26]. In this paper, the CS algorithm was chosen to solve this
In nature, the way that cuckoos look for a suitable location of bird’s nests for their own eggs is
problem, as it features high computational efficiency and accurate optimization results.
random. In order to simulate the way that cuckoos find a nest, it is necessary to set the following
In nature,
three the way that cuckoos look for a suitable location of bird’s nests for their own eggs is
ideal states:
random. In order to simulate the way that cuckoos find a nest, it is necessary to set the following three
1. A cuckoo lays one egg at a time, and selects a bird’s nest to hatch it randomly;
ideal states:
2. In a randomly selected group of bird’s nests, the best bird’s nest will be retained to the next
1. generation.
A cuckoo lays one egg at a time, and selects a bird’s nest to hatch it randomly;
3. The number n of available bird’s nests is fixed and the probability that an owner of a bird’s nest
2. In a randomly selected group of bird’s nests, the best bird’s nest will be retained to the
can find an exotic birds’ eggs is Pa ∈ [0, 1]. Based on the three ideal states, the updating formula
next generation.
of path and location is as follows, when the cuckoo finds a nest:
3. The number n of available bird’s nests is fixed and the probability that an owner of a bird’s nest
can find an exotic birds’ eggs X is 1 i ∈
t Pa X[0, t1].
i Based
  L  on
 , ithe
 1,three
2, , nideal states, the updating
(7) formula
of path and location is as follows, when the cuckoo finds a nest:
where X(t)i is the next location number in the cuckoo generation t, α represents the step control
variable and L(λ) represents the Levy random search path. Levy flight is a random movement
process, the step of its flight + 1)i = obeys
X (t distance + α ×distribution.
X (t)i Levy L(λ), i = 1, · · · , n formula is used to
2, following
The (7)
produce a Levy random number:
where X(t)i is the next location number in the cuckoo generation t, α represents the step control
 u
variable and L(λ) represents the Levy L   random
 1  1search
   3 path. Levy flight is a random movement
(8)
process, the step of its flight distance obeysv Levy distribution. The following formula is used to
produce a Levy random number:
u and ν obey standard normal distribution and λ is the distribution factor, In this paper, it equals
1.5 [27]. The formula of φ is as follows: φ×u
L(λ) = 1 (1 < λ < 3) (8)
1

|v| λ
 


  1     sin     
  2  (9)
   1 
   1    
  2     2  
2

 
Appl. Sci. 2018, 8, 148 6 of 14

u and ν obey standard normal distribution and λ is the distribution factor, In this paper, it equals
1.5 [27]. The formula of ϕ is as follows:

  1
 (1 + λ) × sin π × λ2  λ
φ= h  λ −1
i (9)
1+ λ

2 × λ × 2 2 

Appl. Sci. 2018, 8, x FOR PEER REVIEW 6 of 14

2.3. Diagnostic System
2.3. Diagnostic System
The diagnostic scheme was designed as in Figure 4. The input of the real gas turbine is the
The diagnostic scheme was designed as in Figure 4. The input of the real gas turbine is the
working condition, and the input of the simulation model is the working condition and health condition
working condition, and the input of the simulation model is the working condition and health
represented by the health parameters. The health condition consists of two parts, the degradation
condition represented by the health parameters. The health condition consists of two parts, the
of component working
degradation ability working
of component and inertial coefficients.
ability and inertialAcoefficients.
detailed description of the degradation
A detailed description of the is
shown in reference
degradation [23]. in reference [23].
is shown

Figure 4. Scheme of fault diagnosis based on transient process measuring data.

Figure 4. Scheme of fault diagnosis based on transient process measuring data.
The diagnostic process can be divided into three steps.
The diagnostic process can be divided into three steps.
1. Model simulation
1. Model
At simulation
the beginning of the calculation, we set the gas turbine faults in the gas turbine and obtained
the measurable parameters after the operation of the gas turbine. By inputting the same input
At the beginning
parameters as the of the calculation,
measurable we set
parameters of the
the gas
faultturbine
model faults in the gas
and inserting theturbine and obtained
health initial values the
measurable parameters
diagnosed after themodel,
by the calculation operation of the gas turbine.
the measurable Byof
parameters inputting
the fault the same
model caninput parameters
be estimated
as thethrough
measurable parameters
the calculation of the fault model and inserting the health initial values diagnosed by
model.
the calculation model, the measurable parameters of the fault model can be estimated through the
2. Objective function calculation
calculation model.
The estimated parameters given in the initial calculation are generally incorrect, therefore the
2. Objective exists
deviation between
function the model and the measurable parameters in the actual gas turbine. The
calculation
deviation can be determined by the objective function in a certain calculation way. The object function
of this
The paper is defined
estimated as:
parameters given in the initial calculation are generally incorrect, therefore
the deviation exists between the model and thet measurable parameters in the actual gas turbine.
j

fobj   function
The deviation can be determined by the objective Zt  Zt (10)object
in a certain calculation way. The
t t
function of this paper is defined as: i

tj
3. Optimization of the estimation parameters Z
f = Z −Z
is small, thet estimate
If the value of the objective functionobj t can recognize and isolate the fault to
(10)
achieve the goal of the fault diagnosis. However,t=ti when the deviation does not achieve the desired
accuracy, the estimated initial value needs to be corrected through the optimization algorithm to
3. Optimization of the
return to the first estimation
process parameters
calculation. Such a process takes place recursively until the objective
function reaches an acceptable accuracy.
If the value of the objective function is small, the estimate can recognize and isolate the fault to
achieve the Study
3. Case goal of the fault diagnosis. However, when the deviation does not achieve the desired
accuracy, the estimated initial value needs to be corrected through the optimization algorithm to return
A single spool and free power turbine industrial gas turbine is cited as an example for fault
to thediagnosis
first process calculation. Such a process takes place recursively until the objective function
with transient data. The basic performance parameters are shown as in Table 1.
reaches an acceptable accuracy.
Table 1. Base load parameters of gas turbine.

Parameter Value
Pressure ratio 24.1
Exhaust mass flow 84.31 kg/s
Power output 31.4 MW
Efficiency of compressor 85%
Appl. Sci. 2018, 8, 148 7 of 14

3. Case Study
A single spool and free power turbine industrial gas turbine is cited as an example for fault
diagnosis with transient data. The basic performance parameters are shown as in Table 1.

Table 1. Base load parameters of gas turbine.

Parameter Value
Pressure ratio 24.1
Exhaust mass flow 84.31 kg/s
Power output 31.4 MW
Efficiency of compressor 85%

3.1. Overall Performance Test Rig

Appl. Sci. 2018, 8, x FOR PEER REVIEW 7 of 14
An overall performance test rig was established to acquire transient process measuring data,
3.1. 5.
see Figure Overall
The Performance
data for the Test Rig
instrument and control system of the gas turbine, through the isolated
safety barrier,
Anis sent to
overall the signaltest
performance acquisition board, where
rig was established data
to acquire acquisition,
transient processdata pre-processing,
measuring data, see A-D
signal conversion
Figure 5. Theand datadata output
for the are being
instrument completed.
and control system ofThe
thecalculative
gas turbine, process
through the is finished
isolated in the
portablesafety
decay barrier, is sent to
controller ofthe
thesignal acquisition
board. board, where
The processed data acquisition,
digital signal willdata pre-processing,
be published A- level I
to the
D signal conversion and data output are being completed. The calculative process is finished in the
network through wireless access point (WAP) or access point (AP) web. The data management system
portable decay controller of the board. The processed digital signal will be published to the level I
of the portable condition monitoring system is in charge of data transformation and transmission from
network through wireless access point (WAP) or access point (AP) web. The data management
the level
system of thetoportable
I network level II.condition
The interface management
monitoring system is insystem
chargeofofthe remote
data data center,
transformation and obtains
data from the internet,
transmission fromsends data
the level to main
I network database
to level II. The via the level
interface III network
management system(remote data data
of the remote center local
center, obtains
area network (LAN)) data from the internet,
according sends data
to the defined to main database
interface via the
rules, and level the
stores III network
test data(remote
chronically.
data center local area network (LAN)) according to the defined interface
The remote data management system is mainly responsible for human–computer interactions rules, and stores the test data for
chronically. The remote data management system is mainly responsible for human–computer
a remote data center, with a level III network and interface management server.
interactions for a remote data center, with a level III network and interface management server.

A Micro WAP
Level II Network

V Controller /AP
Interface
Instrument & T Management
P Low Speed DA
Control Portable
System System
D
High Speed DA
Level I Network

D Main
Database
Level III Network

Database
Remote
System

Plant DCS Remote

Figure 5. Gas turbine overall performance test rig.
Figure 5. Gas turbine overall performance test rig.
The data for the model validation and the case study is acquired from the instrument and control
system based on the overall performance test rig. However, this is sometimes forbidden by users
considering operation safety. Thus another channel is designed for data acquisition. The portable
data management system is linked to the plant database via a level I network directly to get the
historical data for this selection. The measuring parameters and performance parameters related to
this case are listed in Table 2.
Appl. Sci. 2018, 8, 148 8 of 14

The data for the model validation and the case study is acquired from the instrument and control
system based on the overall performance test rig. However, this is sometimes forbidden by users
considering operation safety. Thus another channel is designed for data acquisition. The portable data
management system is linked to the plant database via a level I network directly to get the historical
data for this selection. The measuring parameters and performance parameters related to this case are
listed in Table 2.
Appl. Sci. 2018, 8, x FOR PEER REVIEW Table 2. Parameter list for this case. 8 of 14

Table 2. Parameter list for this case.

Measurement Parameter Symbol
Measurement Parameter
Rotational Speed Symbol
n1
Rotational Speed
Discharge Temperature of Compressor T2n1
DischargeTemperature
Discharge Pressure of Compressor
of Compressor P2T2
Discharge Temperature
Discharge of High
Pressure Pressure Turbine
of Compressor T34P2
Discharge pressure of High Pressure Turbine P34
Discharge Temperature of High Pressure Turbine T34
Compressor efficiency degradation DEC
Discharge pressure
Compressor flowof rate
High Pressure Turbine
degradation P34
DGC
High Compressor efficiency
pressure turbine degradation
efficiency degradation DEC
DET
High Compressor flowflow
pressure turbine raterate
degradation
degradation DGC
DGT
Power
High Turbine
pressure efficiency
turbine degradation
efficiency degradation DEP
DET
HighPower Turbine
pressure flow flow
turbine rate degradation
rate degradation DGP
DGT
Power Turbine efficiency degradation DEP
Power Turbine flow rate degradation
3.2. Simulation Model Validation DGP

The
3.2.dynamic
Simulationsimulation model is validated in this section. The start-up process is selected to
Model Validation
show the maximum error of this
The dynamic simulation modelmodel. All measurement
is validated parameters
in this section. simulated
The start-up process isby this model
selected to and
tested by the overall performance test rig are compared, including high-pressure
show the maximum error of this model. All measurement parameters simulated by this model and turbine rotational
speed ntested
1 , andbyhigh-pressure
the overall performance test rig are compared,
turbine discharge pressure. including high-pressure turbine rotational
speed n 1, and high-pressure turbine discharge pressure.
The P34 and compressor discharge temperature T2 , standing for the minimum, moderate and
The P34 and compressor discharge temperature T2, standing for the minimum, moderate and
maximum simulation error, are cited as three examples. The comparative results are shown in
maximum simulation error, are cited as three examples. The comparative results are shown in
FiguresFigures
6–8. The mean relative error of rotational speed is 0.5%, and the mean relative error of
6–8. The mean relative error of rotational speed is 0.5%, and the mean relative error of
compressor discharge
compressor discharge temperature
temperatureisis4.6%.
4.6%. InIn conclusion,
conclusion, thisthis model
model can predict
can predict the trendtheoftrend of
measurement
measurement parameters, and
parameters, thethesimulation
and errorisisunder
simulation error under 4.6%.
4.6%. It proves
It proves thatmodel
that this this can
model can
supportsupport fault diagnosis
fault diagnosis withwith transientdata.
transient data.

Figure Figure 6. Comparison

6. Comparison of simulation
of simulation results
results andand measuring
measuring datafor
data forhigh-pressure
high-pressure turbine
turbinerotational
rotational speed.
speed.
Appl. Sci. 2018, 8, 148 9 of 14
Appl. Sci. 2018, 8, x FOR PEER REVIEW 9 of 14
Appl. Sci. 2018, 8, x FOR PEER REVIEW 9 of 14

Figure 7. Comparison of simulation results and measuring data for high-pressure turbine discharge
Figure 7. Comparison of simulation results and measuring data for high-pressure turbine discharge
Figure 7. Comparison of simulation results and measuring data for high-pressure turbine discharge pressure.
pressure.
pressure.

Figure 8. Comparison
Figure of simulation
8. Comparison of simulationresults
resultsand
and measuring datafor
measuring data for compressor
compressor discharge
discharge temperature.
temperature.
Figure 8. Comparison of simulation results and measuring data for compressor discharge temperature.
3.3. Simulation with Component Degradation
3.3. Simulation withwith
3.3. Simulation Component Degradation
Component Degradation
Compressor fouling is implanted in this section, to show the effect of gas path faults on the
Compressor
Compressor fouling
fouling is isimplanted
implanted in in this section,totoshow
this section, showthethe effect
effect of gas
of gas pathpath faults
faults on the
on the
monitoring parameters. As the rule base of gas-path fault, flow rate degradation due to compressor
monitoring
monitoring parameters. As the rule base of gas-path fault, flow rate degradation due to compressor
fouling parameters.
is three timesAsthe theefficiency
rule basedegradation
of gas-path[28].
fault,
The flow rate degradation
start-up due toand
process is studied, compressor
the
fouling
fouling is is three
three timestimesthe the efficiencydegradation
efficiency degradation [28].
[28]. TheThe start-up
start-up process is studied,
process is and the
studied, and the
degradation of the compressor flow rate is assumed to be 3%, the degradation of compressor
degradation
degradation of compressor
ofisthe the compressor flow flow
rate rate
is is assumed
assumed beto
to of bethe3%,degradation
the degradation of compressor
efficiency assumed to be 1%. The simulation result a3%,
healthy of compressor
engine and deteriorated efficiency
engine is
efficiency is assumed to be 1%. The simulation result of a healthy engine and deteriorated engine is
is assumed
listed into be 1%.
Figures The10.
9 and simulation result of a healthy engine and deteriorated engine is listed in
listed in Figures 9 and 10.
Figures 9 Itand is obvious
10. that the deviation between the monitoring parameters for healthy and deteriorated
It is obvious that the deviation between the monitoring parameters for healthy and deteriorated
engines
It is is small.
obvious thatFor instance,
the deviation this deviation
between of compressor
the monitoring discharge temperature
parameters is just and
for healthy 2–4 °C under
deteriorated
engines is small. For instance, this deviation of compressor discharge temperature is just 2–4 °C under
100%
engines working
is small. Forconditions. It
instance, Itthis is easy to
deviation be annihilated by sensor noise. However, in the transient
2–4 ◦ C under
100% working conditions. is easy to be of compressor
annihilated by discharge
sensor noise. temperature
However, in is just
the transient
process, this deviation can be integrated. Figures 11 and 12 show the integrating result. Thus, the
100%process,
working conditions.
this deviation It is be
can easy to be annihilated
integrated. Figures 11byand
sensor noise.theHowever,
12 show in the
integrating transient
result. process,
Thus, the
integrated deviation is selected as the object function for the cuckoo search algorithm.
this deviation
integrated can be integrated.
deviation is selected Figures 11 and
as the object 12 show
function for thethe integrating
cuckoo result. Thus, the integrated
search algorithm.
deviation is selected as the object function for the cuckoo search algorithm.
Appl. Sci. 2018, 8, 148 10 of 14
Appl. Sci.
Appl. Sci. 2018,
2018, 8,
8, xx FOR
FOR PEER
PEER REVIEW
REVIEW 10 of
10 of 14
14

Appl. Sci. 2018, 8, x FOR PEER REVIEW 10 of 14

Figure
Figure
Figure 9. Simulation
9. Simulation
9. Simulation resultofof
result
result ofthe
thehigh-pressure
the high-pressure turbine
high-pressure turbine discharge
turbinedischarge pressure.
discharge pressure.
pressure.

Figure 9. Simulation result of the high-pressure turbine discharge pressure.

Figure 10. Simulation result of the compressor discharge temperature.

Figure
Figure 10. 10. Simulationresult
Simulation resultof
ofthe
the compressor
compressor discharge
dischargetemperature.
temperature.
Figure 10. Simulation result of the compressor discharge temperature.

Figure 11.
Figure 11. Integrating
Integrating result
result of
of the
the high-pressure
high-pressure turbine
turbine discharge
discharge pressure.
pressure.

Figure 11. Integrating result of the high-pressure turbine discharge pressure.

Figure 11. Integrating result of the high-pressure turbine discharge pressure.
Appl.Appl.
Sci. Sci.
2018, 8, 8,
2018, 148x FOR PEER REVIEW 11 11
of 14 11 of 14
Appl. Sci. 2018, 8, x FOR PEER REVIEW of 14

Figure 12.12.
Figure Integrating result
Integrating of of
result thethe
compressor discharge
compressor temperature.
discharge temperature.
Figure 12. Integrating result of the compressor discharge temperature.
3.4. Comparative
3.4. Study
Comparative Study
3.4. Comparative
The Study results based on the steady-state data and transient data are compared in this
diagnostic
The diagnostic results based on the steady-state data and transient data are compared in this
section.
The
section.UpUp to to
diagnosticnow,
now,extensive
results
extensive research
based on the
research has
hasbeen conducted
steady-state
been conducted to to
data investigate
and thethe
transient
investigate effects
data
effectsof physical
are
of compared
physical faults in this
faults
on
section. the
on Up performance
the performance of
to now, extensive gas turbines.
of gas research From
turbines. From the experimental
the experimental
has been conducted to results, the
results, theoretical
the theoretical
investigate value
the effects valueof the
of of ratio
the
physicalratiofaults
of thethecompressor flow rate
ratedegradation over thetheefficiency degradation is isthree [28–30]. This
on theofperformance
compressor of gasflowturbines. degradation
From the over experimental efficiency degradation
results, the theoretical three [28–30].
value of theThisratio of
theoretical
theoretical value
valuewaswaschosen
chosen as the
as thecriterion
criterion for the
for comparison
the comparison of different
of diagnostic
different diagnostic methods.
methods.
the compressor
For
flow rate
comparison, two
degradation
pieces of of
over thestable
relatively
efficiency degradation
steady-state data
isselected
areare
three [28–30].
forfor
thethe
This kind
same
theoretical
For comparison, two pieces relatively stable steady-state data selected same kind
value was
of of
fault chosen
diagnosis, as the
asas criterion
in in
Figure for
13.13.
Theythe comparison
areare
thethe
input of
forfordifferent
thethe diagnostic
steady-state diagnosis.methods.
The input forfor
thethe
fault diagnosis, Figure They input steady-state diagnosis. The input
For comparison,
transient process is twotwo
the pieces of relatively
complete pieces of stable
data steady-state
before and data
after are selected for the same kind of
washing.
transient process is the two complete pieces of data before and after washing.
fault diagnosis, as in Figure 13. They are the input for the steady-state diagnosis. The input for the
transient process is the two complete pieces of data before and after washing.

Figure
Figure 13.13.
13.
Figure Selected
Selected working
Selectedworkingcondition
working forfor
condition
condition diagnosis.
for diagnosis.
diagnosis.

Component degradation
Component degradationis is
thethe
relative value
relative of of
value thethe
clean component.
clean component. The
Theimprovement
improvement value
value
ofComponent
component degradation
degradation is is the relative
selected for thisvalue of the clean
comparative study. component.
The The
diagnostic improvement
result of transient
of component degradation is selected for this comparative study. The diagnostic result of transient
value
of component degradation is selected for this comparative study. The diagnostic result of transient
process analysis method is better than using steady state measurements, according to the result
comparison, see Table 3. For convenience, diagnosis based on the steady-state is called Method A,
and diagnosis based on the transient process is called Method B below.
Appl. Sci. 2018, 8, 148 12 of 14

• The improvement of the compressor flow rate degradation is 13 times the value of its efficiency
degradation based on Method A. It does not meet the theoretical value of three. The result of
Method B is 1.96, closer to the theoretical value.
• These ratios for high pressure turbines are 3.86 and 5.33. It seems that Method A is more accurate
than Method B for high-pressure turbines. Thus, the health condition of high-pressure turbines
can be monitored by both Method A and Method B. However, washing decisions are usually
made by compressor degradation. This means that Method B is more suitable for supporting
maintenance scheduling.
• There is not any maintenance work being carried out for power turbines in this process. Therefore,
the improvement of both its flow rate and efficiency should be zero, theoretically. This proves
that Method B is more accurate for power turbine diagnosis.

Table 3. Comparison result of diagnostic results based on steady-state analysis and transient process analysis (%).

Time Process DGC DEC DGC /DEC DGT DET DGT /DET DGP DEP
Steady-state 5.79 1.70 - 4.85 2.03 - 0.88 1.53
Before Washing
Transient process 7.83 4.62 - 5.71 2.10 - 3.28 1.89
Steady-state 2.02 1.48 - 2.03 1.30 - −0.17 1.07
After Washing
Transient process 1.54 1.41 - 2.56 1.42 - 2.95 1.76
Steady-state 3.77 0.29 13.0 2.82 0.73 3.86 1.05 0.44
Improvement
Transient process 6.29 3.21 1.96 3.15 0.57 5.53 0.33 0.13

4. Conclusions
In this paper, a gas turbine dynamic model was introduced to simulate its nonlinear dynamic
behavior. The model was developed in the environment of Matlab/Simulink(MATLAB 2013a,
The Mathworks, Inc., Natick, MA, USA). The start-up process was simulated via this model.
The comparison of the simulation results to the field operating data showed the validity of the
model and the advantage of using transient accumulated deviations. Some conclusions have been
obtained as follows:

1. A new, non-linear, model-based diagnostic method, using gas turbine transient measurements
and a cuckoo search (CS) algorithm, was tested to diagnose a gas turbine before and after
a washing process.
2. Diagnosis with transient measurements is more relevant than diagnosis with steady-state
measurements, when gas turbine faults contribute little to performance deviation in steady-state
conditions or gas turbine output fluctuates greatly.
3. Gas turbine component fault diagnosis using transient data can be more effective than using
steady state data, owing to magnifying fault signatures and extending the tracking time to
eliminate variable uncertainties.

Acknowledgments: This study was co-supported by the National Natural Science Foundation of China
(No. 51676182) and the National Natural Science Foundation of China (No. 51706132).
Author Contributions: De-tang Zeng and Dengji Zhou have proposed a research method based on fault diagnosis
for transient performance of gas turbine, the simulation and experiment data are designed, analyzed and processed,
and the manuscript is written. Chun-qing Tan has offered help for the simulations. Baoyang Jiang has proposed
the resiving suggestions of manuscript.
Conflicts of Interest: The authors declare no conflict of interest.
Appl. Sci. 2018, 8, 148 13 of 14

Nomenclature
π component pressure ratio
η component efficiency
T1 compressor inlet air temperature
P1 compressor inlet air pressure
n1 rotational speed of high pressure turbine
Qc compressor inlet air mass flow
T2 compressor discharge air temperature
P2 compressor discharge air pressure
Qf fuel gas flow
T3 high pressure turbine inlet temperature
P3 high pressure turbine inlet pressure
T34 high pressure turbine discharge temperature
P34 high pressure turbine discharge pressure
Qt high pressure turbine mass flow
T4 power turbine discharge temperature
P4 power turbine discharge pressure
Qp power turbine mass flow
Pc power consumption of compressor
Pt power generation of high pressure turbine
Pp power generation of power turbine

References
1. Kagermann, H.; Helbig, J.; Hellinger, A.; Wahlster, W. Recommendations for Implementing the Strategic
Initiative INDUSTRIE 4.0: Securing the Future of German Manufacturing Industry. Final Report of the
Industrie 4.0 Working Group. Available online: http://forschungsunion.de/pdf/industrie_4_0_final_report.
pdf (accessed on 16 January 2018).
2. Brynjolfsson, E.; McAfee, A. Race against the Machine: How the Digital Revolution is Accelerating Innovation,
Driving Productivity, and Irreversibly Transforming Employment and the Economy; Digital Frontier Press:
Lexington, MA, USA, 31 January 2012.
3. Isermann, R.; Ballé, P. Trends in the application of model-based fault detection and diagnosis of technical
processes. Control Eng. Pract. 1997, 5, 709–719. [CrossRef]
4. Zhang, J. Multi-source remote sensing data fusion: Status and trends. Int. J. Image Data Fusion 2010, 1, 5–24.
[CrossRef]
5. Lu, F.; Zheng, W.; Huang, J.; Feng, M. Life cycle performance estimation and in-flight health monitoring for
gas turbine engine. J. Dyn. Syst. Meas. Control 2016, 138, 091009. [CrossRef]
6. Li, Y.G. Performance-analysis-based gas turbine diagnostics: A review. Proc. Inst. Mech. Eng. Part A J.
Power Energy 2002, 216, 363–377. [CrossRef]
7. Sampath, S.; Gulati, A.; Singh, R. Fault diagnostics using genetic algorithm for advanced cycle gas turbine.
In Proceedings of the ASME Turbo Expo 2002: Power for Land, Sea, and Air, Amsterdam, The Netherlands,
3–6 June 2002; American Society of Mechanical Engineers: New York, NY, USA, 2002; pp. 19–27.
8. Li, Y.G. A gas turbine diagnostic approach with transient measurements. Proc. Inst. Mech. Eng. Part A J.
Power Energy 2003, 217, 169–177. [CrossRef]
9. Merrington, G.L. Fault Diagnosis of Gas Turbine Engine from Transient Data. J. Eng. Gas Turbine Power 1989,
111, 237–243. [CrossRef]
10. Merrington, G.L. Identification of dynamic characteristics for fault isolation purposes in a gas turbine using
closed-loop measurements. In Proceedings of the Advisory Group for Aerospace Research and Development
(AGARD) Conference, Quebec, QC, Canada, 30 May–3 June 1988; Volume 36, pp. 1–13.
11. Urban, L.A. Gas Turbine Engine Parameter Interrelationships; Hamilton Standard Division of United
Aircraft Corporation: Hartford, Connecticut, 1969.
Appl. Sci. 2018, 8, 148 14 of 14

12. Naderi, E.; Khorasani, K. Data-driven fault detection, isolation and estimation of aircraft gas turbine engine
actuator and sensors. In Proceedings of the IEEE 30th Canadian Conference on Electrical and Computer
Engineering (CCECE), Windsor, ON, Canada, 30 April–3 May 2017; pp. 1–6.
13. Meher-Homji, C.B.; Bhargarva, R. Condition Monitoring and Diagnostic Aspects of Gas Turbine Transient
Response. Int. J. Turbo Jet-Engines 1994, 11, 99–111. [CrossRef]
14. Henry, J.R. CF-18/F404 transient performance trending. In Proceedings of the Advisory Group for Aerospace
Research and Development (AGARD) Conference, Quebec, QC, Canada, 30 May–3 June 1988; Volume 37,
pp. 1–13.
15. Merrington, G.; Kwon, O.K.; Goodwin, G.; Carlsson, B. Fault detection and diagnosis in gas turbines. J. Eng.
Gas Turbines Power 1991, 113, 276–282. [CrossRef]
16. Simani, S.; Fantuzzi, C. Dynamic system identification and model-based fault diagnosis of an industrial gas
turbine prototype. Mechatronics 2006, 16, 341–363. [CrossRef]
17. Simani, S.; Patton, R.J. Fault diagnosis of an industrial gas turbine prototype using a system identification
approach. Control Eng. Pract. 2008, 16, 769–786. [CrossRef]
18. Lu, F.; Jiang, C.; Huang, J.; Wang, Y.; You, C. A novel data hierarchical fusion method for gas turbine engine
performance fault diagnosis. Energies 2016, 9, 828. [CrossRef]
19. Kim, J.H.; Song, T.W.; Kim, T.S.; Ro, S.T. Model development and simulation of transient behavior of heavy
duty gas turbines. J. Eng. Gas Turbines Power 2001, 123, 589–594. [CrossRef]
20. Schobeiri, M.T.; Attia, M.; Lippke, C. GETRAN: A generic, modularly structured computer code for
simulation of dynamic behavior of aero-and power generation gas turbine engines. J. Eng. Gas Turbines Power
1994, 116, 483–494. [CrossRef]
21. Muir, D.E.; Saravanamuttoo, H.I.H.; Marshall, D.J. Health monitoring of variable geometry gas turbines for
the Canadian Navy. J. Eng. Gas Turbines Power 1989, 111, 244–250. [CrossRef]
22. Blinstrub, J.; Li, Y.G.; Newby, M.; Zhou, Q.; Stigant, G.; Pilidis, P.; Hönen, H. Application of gas path
analysis to compressor diagnosis of an industrial gas turbine using field data. In Proceedings of the ASME
Turbo Expo 2014: Turbine Technical Conference and Exposition, Düsseldorf, Germany, 16–20 June 2014;
American Society of Mechanical Engineers: New York, NY, USA, 2014; p. V03AT07A008.
23. Zhou, D.; Zhang, H.; Weng, S. A new gas path fault diagnostic method of gas turbine based on support
vector machine. J. Eng. Gas Turbines Power 2015, 137, 102605. [CrossRef]
24. Yang, X.; Deb, S. Cuckoo search via L’evy flights. In Proceedings of the IEEE World Congress on Nature &
Biologically Inspired Computing. Piscataway, Coimbatore, India, 9–11 December 2009; pp. 210–214.
25. Civicioglu, P.; Besdok, E. A conceptual comparison of the Cuckoo-search, particle swarm optimization,
differential evolution and artificial bee colony algorithms. Artif. Intell. Rev. 2011, 39, 315–346. [CrossRef]
26. Yang, X.S.; Deb, S. Engineering optimization by cuckoo search. Int. J. Math. Model. Numer. Optim. 2010,
1, 330–343.
27. Nguyen, T.T.; Vo, D.N.; Dinh, B.H. Cuckoo search algorithm for combined heat and power economic dispatch.
Int. J. Electr. Power Energy Syst. 2016, 81, 204–214. [CrossRef]
28. Diakunchak, I.S. Performance deterioration in industrial gas turbines. J. Eng. Gas Turbines Power 1992,
114, 161–168. [CrossRef]
29. Joly, R.B.; Ogaji, S.O.T.; Singh, R.; Probert, S.D. Gas-turbine diagnostics using artificial neural-networks for
a high bypass ratio military turbofan engine. Appl. Energy 2004, 78, 397–418. [CrossRef]
30. Mohammadi, E.; Montazeri-Gh, M. Simulation of full and part-load performance deterioration of industrial
two-shaft gas turbine. J. Eng. Gas Turbines Power 2014, 136, 092602. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).