Applied Energy 73 (2002) 2546 www.elsevier.

com/locate/apenergy

Parameter selection for diagnosing a gas-turbines performance-deterioration

S.O.T. Ogaji*, S. Sampath, R. Singh, S.D. Probert
School of Engineering, Craneld University, Bedfordshire MK 43 OAL, UK

Abstract The ability to assess faults in a system, while it is operating, requires an appropriate set of measurements. Engine availability can be increased if the faults can be detected, isolated and assessed, so enabling an optimised shutdown of the plant for maintenance to ensue. Depending on the engine-power-setting parameter, the measurements required to diagnose the faults along the gas path of a gas-turbine vary. This study used a non-linear gas-path analysis (NLGPA) model to predict the required instrumentation set, which can be optimised with respect to the number and type of sensors and their locations for the considered engine-faults. A thermodynamic model of the behaviour of a 2-shaft engine is used as a case study. Redundancy in the sensor set is shown to be unnecessary. # 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Diagnostics; Gas path; Gas turbine

1. The problem When in use, the components of a gas-turbine deteriorate: usually, the instrumentation on the engine is not optimised for the detection of the associated faults. Because it may not be obvious which combination of sensors can detect which kind of fault, methods such as gas-path analysis (GPA) have been devised [1,2]. One approach to obtaining an appropriate sensor set is by deliberately implanting defects in the gas-turbine components and observing the simultaneous eects on the measured performance parameters, and thence seeking the optimal combination of dependent variables that can describe such faults. This combination will facilitate the choice of types and locations of the sensors required for the diagnosis of the faults along the gas-path. This present study will use a non-linear approach, which is an oshoot of the LGPA introduced by Urban [1], to select an instrumentation set that can detect faults in the
* Corresponding author. 0306-2619/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved. PII: S0306-2619(02)00042-9

26

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

Nomenclature A1!A12, B1!B30, C1!C30, D1!D25 C CC COT CT DOD e EDI EGT EI ETAC ETAT F( ) FCM FI FOD GPA HOTs ICM ( ) ISO LGPA m N NDMC NDMT NLGPA n OEM P PT RMS SHP T TBO v w

Parameter combinations obtained for the various case studies considered Compressor Combustion chamber Combustors outlet-temperature Core/compressor turbine Damage due to domestic objects Dierence between implanted and detected independent parameter changes Engine-deterioration index Exhaust-gas temperature Erosion index Compressors isentropic-eciency Turbines isentropic-eciency Inuence coecient matrix; represents a functional relationship Fault-coecient matrix Fouling index Damage due to foreign objects Gas-path analysis Higher-order terms Inuence coecient matrix International Standards Organisation Linear gas-path analysis Number of dependent parameters Spool speed Flow capacity of compressor Flow capacity of turbine Non-linear gas-path analysis Number of independent parameters Original equipment-manufacturer Pressure Power turbine Root mean square Shafts horse-power Temperature Time between overhauls Measurement noise Small change applied to a variable

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

27

WFE X Y Z Subscripts 1 2 c t Glossary Aeroderivative: Down-time:

Fuel-ow rate (1m) vector of independent variables (1n) vector of dependent variables Eciency Flow capacity High-pressure or gas generator shaft Low-pressure shaft Compressor Turbine

An aeroengine adapted for stationary applications. The length of time taken to get an engine from a mal- or non-functional state to a functional state. Fault ratio: The change in ow capacity per unit drop in eciency. Forced outage: The sudden breakdown of an engine in operation Gas generator: The section of the gas-turbine used to provide gas at high pressure for expansion in either a power turbine producing torque or in a nozzle producing thrust, Gas path: The enclosure through which the working uid ows. Power-setting parameter: A parameter used to control the operation of the engine.

compressor and turbines of a single-spool gas-turbine. Its application to a two-spool gas-turbine has been discussed by Ogaji and Singh [3]. Consideration of the combustor is omitted because its eciency normally remains relatively invariant with time [4,5], i.e. its deterioration does not cause signicant changes in the engines performance. The eects of using a non-optimised instrumentation set are also assessed along with the benets resulting from a wiser choice of instrumentation.

2. Fault characteristics Several faults in the components along the gas path of a gas-turbine are commonplace. The principal ones that aect the performance of the engine are listed below.  Fouling is the accumulation of deposits on the blade surfaces, so causing an increase in surface roughness, changes in the aerofoil shapes and/or aerofoil inlet-angles and a narrowing of the aerofoil-throat aperture [4,6]. It is one of the commonest causes of performance reduction encountered by users of

28

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

gas-turbines [7,8] and can account for more than 70% of the performance loss during operation [4]. Fouling primarily results in mass ow and the compressors delivery-pressure (CDP) reductions, and ultimately in a fall in power and a rise in heat rate [4,8,9] with a slight change in the compressors eciency [10]. Fouling has ensued in up to 50% of the compressor stages [9]. Fouled or worn turbine-seals result in a loss of eciency and an increase in an engines exhaust-temperature [4,11]. There is, however, a limit to fouling, and this is determined by the aerodynamic forces that prevent further depositions on the blades [6]. Performance deterioration due to fouling is for the most part recoverable by cleaning (i.e. washing) and the recommendation is that cleaning is needed when the mass ow decreases by approximately 2.5% [4]. Tip clearance inuences both eciency and ow capacity: a 0.8% reduction in tip clearance, for an axial compressor, gave a 3% reduction in ow and a 2% reduction in eciency [12,13]. The eciency is more sensitive to tip clearance than fouling. Surfaces exposed to particle impacts become eroded and hence rougher. As a result, there are changes in aerofoil proles and throat openings, with increases in blade and seal clearances and hence decreases in the gas-turbines performance. In the compressor, the erosion of the blades leads to losses of the compressors delivery pressure and mass-ow rate [8,14]. A further eect of erosion is to make the front stages of an HPC and a LPC, for a multiplecomponent gas-turbine, prone to surges. Erosion has been shown to aect the rear stages of a compressor (due to the higher pressure in this region) more signicantly than the initial stages [6]. Erosion of turbine nozzles or blades increases the turbines ow and reduces its eciency, and hence the power output. When losses of material from the ow-path components are caused by the chemical reactions with the contaminants that enter the gas-turbine with the inlet air, fuel injected, water or steam, the process is called corrosion [4]. This is experienced more at the hot end, with the presence of elements such as vanadium, sodium and lead enhancing the corrosion of a turbines aerofoils. The eect is a reduction of the engines performance. Object damage is the result of a body striking the internal surfaces of the gaspath components of the gas turbine. The origin of such particles could be via the inlet section with the working uid [i.e. foreign object damage (FOD)] or particles from the engine itself breaking o and being carried downstream [i.e. domestic object damage (DOD)]. Here again, the eect is a deterioration of the engines performance. The observed fault signature, with respect to its eect on performance, is sometimes identical to that of fouling.

Gas-path faults usually have distinct features or signatures, but some faults may have identical signatures and would require other engine-monitoring techniques, such as vibration analyses, to distinguish them.

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

29

In the current investigation therefore, fouling and erosion faults in the compressor(s) and turbines respectively were simulated by perturbing the independent variables as follows. Compressor fouling: c by 1%, and c by 4% Turbine erosion: t by 1%, and t by 2% The fault percentages were also increased to higher values. Random selections from the presented optimal sets were tested with dierent fault ratios from those given above: the results were satisfactory. 3. Gas-path analysis This has, as its aim, the detection, isolation and quantication of some of the gaspath faults that have observable eects on the measurable (i.e. dependent) variables. This implies that those faults, such as blade cracks that cannot be implicitly detected via these measurable variables, as well as those failures that occur suddenly, such as fracture, cannot be identied by gas-path analyses. The primary use of dependent parameters to isolate gas-path faults presumes that primary performance parameters or independent variables, such as component ow capacities and eciencies, are not directly measurable. They are thermodynamically correlated with the dependent variables, such that changes in the latter are induced by changes in the former, with the possibility of accurately identifying the faulty component(s) hinging on the choice of measurements taken. In an arbitrary gas-turbine conguration, the dependent variable, Y, is a function of the component vector X and the input vector P, i.e., Y F P; X 1

The input vector, in this case, denes the operating conditions, which are functions of the environmental variables (i.e. ambient pressure and temperature) and the power-setting parameter. The components-independent parameter vector contains characteristic performance parameters, such as eciencies and ow capacities. A linearization of Eq. (1) with respect to X gives: Y ICM P X 2 where X implies a small percentage change in X. For a single operating-condition, P=constant. If noise is considered, Eq. (2) can be written as: Y ICM X v 3

30

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

Eq. (3) is the classical formulation of the GPA. The theory behind this concept is shown in Fig. 1. The original concept of GPA, developed by Urban [11], has now been termed [5,15] the linear gas-path analysis (LGPA). The conceptual framework (Fig. 1) is such that it creates a cause-and-eect link between the faults, the performance parameters, and the measured variables. The presence of a primary gas-path fault induces a change in the component characteristic that shows up as a deviation of the measurable parameters from the baseline conditions. This forward pass can be used to generate an ICM by solving a set of dierential equations that thermodynamically relates the dependent to the independent variables or, by using a rather approximate and less cumbersome approach, that involves perturbing the independent variables one after the other and noting their eects on some specied dependent variables. This ICM is then inverted to obtain a fault coecient matrix (FCM), so that changes in the independent variables are obtained from a linear summation of the product of the corresponding coecients in the FCM, which is assumed constant for a given operating condition, and the change in the dependent variable. LGPA has been found to be limited both in its ability to handle changes or degradations beyond the soft (low-fault) range, as well as in the level of error introduced by the assumption of linearity [5]. It is, however, possible to obtain a numerical solution to the non-linear set dened by Eq. (1) through successive iterations using LGPA and the NewtonRaphson technique, until an exact solution is obtained. This is called the non-linear approach to gas-path analysis, or NLGPA [5]. What follows, presents a brief overview of NLGPA; the reader is referred to Escher and Singh [5] for more information. Instrument problems and sources of data noise are not considered in this analysis.

Fig. 1. GPA diagnostic approach.

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

31

4. Non-linear gas-path analysis If we assume a small change occurs in X, then the corresponding change in Y would, from Eq. (1), at a xed operating point, give FX X Y Y FX Y 4

Given a small change w for a single-variable function Z(a), the Taylor series expansion would be Za w Za wZ0 a Thus Eq. (4) can be written as FX X FX JX HOTs 6 w2 00 w3 Z a Z000 a . . . 2! 3! 5

where the Jacobian notation stands for the rst derivative in the Taylor series expansion of the matrix FX X, that is 2 @f X @f X 1 1 6 @X1 @X2 6 6 6 @f2 X @f2 X 6 6 @X2 J 6 @X1 . 6 . . 6 . . 6 . 6 4 @f X @f X n n @X1 @X2 3 X1 6 X2 7 and X 6 . 7 4 . 5 . Xm For small changes in X, we can neglect HOTs, and thus Eq. (6) becomes FX X FX JX From Eqs. (4) and (7), we have Y JX The solution is made easier if the Jacobian is inverted [16], i.e. J 1 Y X 9 7 2 @f1 X 3 @Xm 7 7 7 @f2 X 7 7 @Xm 7 7 . 7 . . 7 . . 7 . 7 @fn X 5 @Xm

32

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

X is the correction to the independent variable and this is added to the solution vector, viz XNew XOld X 10

Each execution of the process described by Eq. (10) creates a new baseline to which further corrections are made until a dened convergence criterion, that minimises the error (i.e. the dierence between measurement and prediction), is achieved, or the process stopped after a given number of iterations. For each linear GPA calculation, an appropriate baseline is required. In the rst iteration, a measured baseline is used and subsequent iterations employ a calculated baseline derived from the implanted faults that are detected in the previous iteration. It is necessary to note that the rst solution vector is the LGPA. Fig. 2 illustrates this process.

5. Performance data for an engine suering deterioration The option of introducing physical faults to such a complex engine as a gas-turbine by stripping and re-coupling components may not be a worthy one because this is not only costly, time consuming and laborious, but has the capability of introducing initially non-existent faults, which can ultimately, produce erroneous data. A computer-simulation model is a possible alternative. Such models are based principally on matching of the various components of the gas-turbine to ensure ow, power and speed compatibility, while allowing for all possibilities of losses and

Fig. 2. Simplied illustration of the non-linear gas-path analysis method.

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

33

bypasses. Robust simulation programs possess generic features that permit the setting up of dierent engine congurations. In the present study, an NLGPA simulation program is used to both model mathematically the engines behaviour and perform the instrumentation assessment. Varying the independent-variable scaling factors was used as the means of implanting faults. The simulation program was developed with LGPA options. Advantages of the NLGPA technique over the LGPA are highlighted subsequently. It should be noted that the NLGPA simulation program used has been validated against several dierent gas-turbine congurations and found to be robust it produces accurate predictions [5,17].

6. Implications of the use of sub-optimal instrumentation A major consequence of employing a non-optimal instrumentation set for engine monitoring could be the non-detection of the setting in of creep damage, and hence a reduction of engine availability and increased maintenance cost. Besides life failures due to poor design or manufacture, creep and fatigue are two basic mechanisms that limit the lives of gas-turbine engines [18]. The strengths of materials, and particularly metals, decrease upon increasing their temperature: prolonged exposure to high temperatures causes a progressive deformation of grain boundaries and other metallurgical instabilities. This thermal fatigue passes through three stages before fracture (i.e. failure) occurs. Primary creep is usually interpreted as the adaptation stage; secondary creep is the useful-life period; while tertiary creep corresponds to decreasing resistance due to advanced age. Microscopically, creep failures, at higher temperatures, result more from intergranular cracking, whilst at lower temperatures, they are from trans-granular cracking. The relationship between the power output and deterioration on a turbines creep-life is shown in Fig. 3. The time taken for failure to occur in creep is also dependent on the stress and other environmental conditions (e.g. corrosion). Original equipment manufacturers (OEMs) normally quote the engines life based on 100% power output under ISO conditions. For an aeroderivative gas-turbine, this could typically correspond to 20,000 hours, and longer for an industrial gas-turbine. Most engines rarely operate at full power, and so, at 90% power, such an aeroderivative gas-turbine with no deterioration will have a creep life of about 5 times that at 100% power, that is 100,000 red hours (Fig. 3). These values would be lower for an engine suering compressor faults, such as fouling. Engine deterioration generally results in a reduced power output and a higher specic fuel-consumption for the same spool speed and the turbines entry-temperature. To attain the normal power output, engines have to be run at higher spool speeds and/or higher turbine-entry temperatures. This results in greater rates of fuel consumption and a shorter life (due to increased creep damage). Berenblut and Masom [18] noted that an increase in hot-section temperature of $ 15  C would cause a two-fold reduction of blade life. There is obviously a critical temperature beyond which creep failure can result from plastic deformation. Creep, in addition

34

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

to oxidation, determines the time that should elapse between overhauls (TBO) and turbine life-usage manifests itself by creep damage. The availability of an engine is directly proportional to the time taken to return it from an undesirable state to functionality. Singh [17] demonstrated this for the case of a gas-turbine (Fig. 4). Because of competitive pressures, several design advances have been applied continually to various engine components. The high cost of the gas-path components, coupled with the low likelihood that they will be required means that they are often not held as spares. However, in the case of damage, the manufacture of replacement components can take several weeks. This is an undesirable situation, especially when viewed from the perspective of cost of the resulting

Fig. 3. Eects of power output and performance deterioration on a turbines creep life.

Fig. 4. System ranking by reliability and availability.

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

35

downtime. The frequency of this occurring can be reduced by the use of an appropriate instrumentation set to monitor the performance of the gas-turbine.

7. Optimal instrumentation-selection It is sometimes possible to make intuitive guesses about dependent variables that would be relevant in detecting particular types of gas-path faults, but using an analytical approach based on techniques, such as GPA, is usually preferable. Also, as stated earlier, gas-path faults are detectable as long as there are discernible patterns in the dependent variables, and by extension, in the independent variables. Compressor fouling and turbine erosion have high probabilities of occurring in a gas-turbine. The power turbine, by virtue of its position in the gas stream, coupled with its physical design is much less likely to experience gas-path damage than other components upstream. In this study, we have considered the power turbine on an equal basis with other components in the analysis of faulty components, because any appropriate choice of instrumentation may be shown to operate under various fault levels. The minimum requirement to determine a set of n unknowns in simultaneousequation analysis is to use a set of n equations. The task is then to seek the n (i.e. the optimal number of sensors) set of dependent variables, optimised by types and locations, that would eectively estimate the changes in the set of n independent variables, which describe the signatures of the desired faults. In other words, the aim is to obtain a non-redundant set of sensors for a given fault-scenario. The approach requires a prior knowledge of the independent variables that would be aected by the presence of any particular fault or fault combinations, and the signatures imposed by these faults on the performance parameters. For given levels of each fault, an attempt is then made to obtain the most appropriate instrumentation that can eectively diagnose the fault(s). As noted by Urban [1], a poor choice of instrumentation rst becomes obvious with the presence of large coecients in the inverted matrix, or FCM, which is brought to the fore by the occurrence of nonconvergence of the iterative process or by large RMS error values. The processes applied in choosing the instrumentation, and its location for a given set of performance parameters are:  Devising a model of the behaviour of the engine to be used  Evaluating the degree of relevance of each possible measurement to a given change in each independent variable  Determining the possible combinations of instruments that would accurately quantify changes in a given set of independent variables

What follows, elaborates the application of these processes to the current study. An engine that compares favourably with its real operational counterpart was thermodynamically modelled with the NLGPA program mentioned earlier. Table 1

36

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

shows how predictions compared with published (from the Gas-Turbine World Handbook, [19]) values for the aeroderivative Rolls Royce Avon engine, schematically represented by Fig. 5. This engine has been chosen to represent the class of two-shaft stationary gas-turbines for this study. The small dierences in values between predictions from the mathematical model and the corresponding actual engine parameters obtained from the literature indicate that a suciently accurate model was being used for this study. The developed model was run on the NLGPA program to obtain an order of sensitivity for each individual measurement to each independent variable, when perturbed by a given amount through varying its scaling factor. What is implied here is that each of the independent variables was changed by some amount to simulate a fault, and all the possible measurements along the engines gas-path applied, one at a time, to determine the degree of their eectiveness in quantifying the implanted fault. The results in Table 2 show those parameters that respond most to a given independent variable change to within an error not greater than an NLGPA RMS value of 0.7 using a SHP power-setting parameter. It should be noted that this RMS value, was subjectively chosen to specify the minimum level of accuracy, required from any sensor, to t as a possible candidate in the choice of sensors to quantify changes in the considered independent variable. The engines
Table 1 Comparison between predictions using the thermo-mathematical model and published values for the RR AVON gas-turbine Parameter Mass ow Pressure ratio Shaft power Thermal eciency EGT Units kg/s kW %  C Published 77.1 8.8 14580 28.2 442 Prediction 76 8.8 14500 29.6 434 Dierence 1.1 0 80 1.4 8

Fig. 5. Schematic diagram of a 2-shaft gas-turbine showing the measurement-station location numbering along the gas path.

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

37

power-setting parameters used in this study are the combustors outlet-temperature (COT) and the shafts horse-power (SHP). This nal step required seeking various combinations of these parameter measurements that respond to faults implanted on components with an NLGPA RMS of less than 0.5. The RMS value indicated is subjective and species the minimum level of accuracy, required for any combination of sensors to t as possible candidates in the choice of sensors to estimate changes in the implanted faults. The aim is to ensure that, at least, each sensor present in the combination represents one of the independent variables whose implanted fault we seek to quantify. This approach proved useful in helping avoid situations where randomly-selected sensors do not completely represent all possible independent variables, which would be aected by a fault. Components were initially grouped into likes, that is, all compressors were put into one group and similarly all turbines were put into a separate group, with various levels of fault index applied to determine suitable instrumentation sets. An FI of unity indicates a drop in eciency of 1% with a

Table 2 Order of sensitivity of dependent to independent parameter changes using an NLGPA program, and SHP as the power-setting parameter Power-setting parameter: SHP. Engine: RR AVON (2-SHAFT) Independent ! ! ! ! ! Direction of decreasing observability by dependent parameters parameter ETAC1 NDMC1 ETAT1 NDMT1 ETAT2 NDMT2 T3 N1 N1 P6 WFE T8 T6 P9 T9 P3 T3 P6 T8 T6 WFE P3 N1 T9 T8 T9 P6 P6 WFE T8 N1 P3 P3 T3 T6 WFE P6 P9 T9 N1 T3 P8 T8 P8 T6 P8

N1

Table 3 Order of sensitivity of dependent to independent parameter changes using an NLGPA program, and COT as the power-setting parameter POWER SETTING PARAMETER: COT. ENGINE: RR AVON (2-SHAFT) Independent ! ! ! ! ! Direction of decreasing observability by dependent parameters parameter ETAC1 NDMC1 ETAT1 NDMT1 ETAT2 NDMT2 SHP N1 T9 SHP SHP SHP N1 SHP SHP P6 P8 WFE P8 T3 P3 N1 P3 WFE P8 P8 WFE P6 P6 WFE WFE T9 N1 P3 P3 T9 P3 T3 T9 P6 N1 P6 P9 T8 N1 T3 T8 T8 T3 T8 P9 T3 P8

T8 T9

38

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

simultaneous drop in ow capacity of 4% to simulate compressor fouling for each compressor of the engine. An EI of unity represents a drop in eciency of 1% and an increase in ow capacity of 2% to simulate turbine erosion for each turbine of the engine. The engine was then considered as a whole under a multi-component faultscenario and a search was made for suitable instrumentation sets that would enable an eective diagnosis to be achieved. Sample results from this step are presented in Tables 47 for the 2-shaft engine considered. Also presented are the RMS values obtained when an LGPA program and the NLGPA program are used. The relation shown by Eq. (11) denes the GPA RMS sensitivity used. Its units depend on the specic parameter in use, but, in this study, the independent parameter errors are expressed as percentages. v u P 2 un u ej tj1 RMS value 11 n

Table 4 Possible instrumentation sets for isolating compressor faults in a 2-shaft engine Case study: A Power-setting parameter: SHP Dependent value 1 N1 P3 T4 P6 T6 WFE P8 T8 P9 T9 SHP X X X X X X X X X 2 X 3 X 4 X 5 X 6 X X Engine: RR AVON (2SHAFT) Fault level: FI=1% 7 X 8 9 10 11 12 X 13 X X X X X X X X X X

X X

Independent value ETAC1 X NDMC1 X ETAT1 NDMT1 ETATPT NDMPT RMS value LGPA NLGPA

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

0.767 0.652 0.629 0.621 0.724 0.730 0.575 2.213 2.214 2.122 2.086 0.978 0.678 0.108 0.097 0.123 0.124 0.219 0.214 0.108 0.091 0.086 0.197 0.118 1.523 0.053

Table 5 Possible instrumentation sets for isolating turbine faults in a 2-shaft engine
INDEPENDENT Case study: B Power-setting parameter: SHP 1 X 2 3 X X 4 X X 5 X 6 X 7 X 8 X 9 X 10 11 X X 12 X X 13 X 14 X Engine RR AVON (2SHAFT) Fault level: EI=1% 15 16 X 17 18 19 X X 20 X X 21 X X 22 X X 23 X 24 X 25 X 26 X 27 X 28 X 29 X X 30 X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Independent variables X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X RMS 0.628 0.068 0.630 0.736 0.233 0.108 0.862 0.068 0.127 0.161 0.939 0.897 0.082 0.061 Values 0.160 0.143 0.401 0.424 0.083 0.171 0.126 0.143 0.140 0.140 0.195 0.313 0.142 0.145 N1 P3 T3 P6 T6 WFE P8 T8 P9 T9 SHP ETAC NDMC1 ETAT NDMT1 ETATPT NDMPT LGPA NLGPA

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

X X X X

X X X X X

X X

X X X X

X X X X X X

X X X X X X X X

X X X X X X X X X X X X X X X X

X X

X X X X X

X X X

X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X

0.060 1.158 0.059 1.185 0.083 0.159 0.253 0.114 0.110 0.323 0.533 0.200 0.217 0.060 9.117 0.076

0.146 0.278 0.385 0.218 0.142 0.143 0.288 0.229 0.290 0.226 0.133 0.138 0.138 0.145 1.234 0.052

39

40

Table 6 Possible instrumentation sets for isolating combined compressor and turbine faults in a 2-shaft engine with SHP as the power setting parameter
DEPENDENT Case study: C Power-setting parameter: SHP 1 X X X X 2 X X X X 3 X X X 4 X X X 5 X X X 6 X X X 7 X X 8 X X 9 X X 10 X X 11 X X 12 X X 13 X X X X Engine RR AVON (2SHAFT) Fault level: FI=1%EI=1% (EDI=1%) 14 X X 15 X X X 16 X X 17 X X 18 X X 19 X X 20 X X 21 X X X 22 X X X 23 X X X 24 X X 25 X X X 26 X X X 27 X X 28 X X 29 X X 30 X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Independent variables X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X RMS 0.146 0.427 0.320 0.344 0.617 0.622 0.554 0.573 0.406 0.408 0.412 0.417 0.409 Values 0.014 0.014 0.019 0.015 0.075 0.042 0.086 0.076 0.013 0.013 0.011 0.014 0.015 N1 P3 T3 P6 T6 WFE P8 T8 P9 T9 SHP ETAC NDMC1 ETAT NDMT1 ETATPT NDMPT LGPA NLGPA

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

X X X X X

X X X

X X X X

X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X

X X X X X X X X X

X X X X X

X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

0.406 0.635 0.594 0.602 0.617 0.597 0.605 0.622 0.409 0.407 0.608 0.418 0.416 0.593 0.601 7.913 0.140

0.013 0.251 0.087 0.077 0.100 0.086 0.086 0.103 0.014 0.013 0.095 0.016 0.013 0.073 0.083 NC 0.012

Table 7 Possible instrumentation sets for indicating and isolating combined compressor and turbine faults in a 2-shaft engine with COT as the power-setting parameter DEPENDENT N1 P3 T3 P6 T6 WFE P8 T8 P9 T9 SHP ETAC NDMC1 ETAT NDMT1 ETATPT NDMPT LGPA NLGPA Case study: D Power-setting parameter: COT (T6) 1 X X 2 X X X 3 X X 4 X X X 5 X X X 6 X X X 7 X X 8 X X X 9 X X 10 X X X 11 X X X Engine RR AVON (2SHAFT) Fault level: FI=1%EI=1% (EDI=1%) 12 X X X X 13 X X X X 14 X X X X 15 X X X 16 X X X 17 X X X 18 X X 19 X X 20 X X 21 X X X 22 X X 23 X X X 24 X X X X 25 X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Independent variables X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X RMS 0.527 0.549 0.547 0.548 0.553 0.654 0.552 0.553 0.552 0.549 0.553 Values 0.009 0.004 0.007 0.007 0.007 0.238 0.003 0.007 0.007 0.007 0.006 S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

X X X X X X X X X X

X X X X X X X X X X X X X

X X X X X X X X X X X X X X X

X X X

X X

X X X X X X X X

X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

0.553 0.551 0.548 0.505 0.549 0.546 0.508 0.511 0.528 0.511 5.305 0.509 7.865 0.553

0.006 0.004 0.007 0.005 0.005 0.006 0.007 0.007 0.007 0.008 0.013 0.008 7.603 0.012

41

42

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

8. Discussion Tables 47 show some results obtained when using NLGPA in an optimal instrumentation-set selection for a 2-shaft engine, with a SHP as the power-setting parameter in the rst three cases and the COT in the last case. Except for the last two columns, all the presented combinations are considered to be optima because they are able to detect, isolate and quantify the implanted faults to a high degree of accuracy, as indicated by the RMS values. Generally, the lower the RMS value, the less the smearing eect and the better the capability of the instrumentation set employed to detect changes in the independent variables. Again, it needs to be emphasised that the optimal instrumentation set, as used in the current analysis, involves the appropriate number, eective combination, and the optimal locations of sensors on the gas-turbine, so enabling the detection and isolation of single, dual and simultaneous multiple component-faults. In Tables 2 and 3, the order of relevance of measurements to independent-variable changes is shown for the two power-setting parameters used. This highlights the requirements of some key measurements, such as shaft speed and fuel ow in fault diagnosis, as well as the pre-eminence of pressure and temperature measurements in ow and eciency determinations when SHP is used as the power-setting parameter (Table 2). The scenario is dierent when COT is the power-setting parameter as SHP turns out to be one of the most signicant parameters for diagnostics, while temperature measurement plays a slightly lesser role except in the eciency of the rst turbine. In Table 4 or case study A, implanted compressor faults for a 2-shaft engine were used to obtain suitable instrumentation-sets for the compressor-fault diagnosis. Because the engine was taken to be of xed design, it was assumed that faults would aect the eciencies and ow capacities of the components, thus requiring at least two measurements to diagnose fault levels. The study initially used an FI of 1%, which represents simultaneous drops of 1% eciency and 4% ow capacity for each compressor of the engine. Eleven possible combinations (A1 ! A11) of such sensors were identied. The combination in A12 shows a choice of instrumentation set that is decient in identifying compressor faults, though each of the sensors involved can be at least assigned to one of the independent variables. A13 shows the use of a redundant set involving all possible sensors (10 in this case) on the engine. A careful ` consideration of A13s RMS value, vis-a-vis those of A1 ! A11, shows that not much is gained in terms of accuracy of predictions from over-instrumenting an engine. In fact, there is the danger of running at higher operational costs. The point being made here is that both a representation of each independent variable by a sensor and a careful choice of the participating sensors in the set are required for a comprehensive fault-diagnosis. Using a few sensors that are optimally selected and located could be as good as over-instrumenting the engine, but would obviously be cheaper than the latter in terms of installation and maintenance costs. Finally, some combinations such as A9, may appear unrealistic to the reader at rst glance, but our simulation shows them to be eective in quantifying compressor gas-path faults. In Table 5 or case study B, the scenario of fault diagnoses for all the turbines in the engine is presented. Initially, an EI of 1%, representing a simultaneous drop of

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

43

1% eciency and an increase of 2% ow capacity for all turbines on the engine was used. This level was gradually increased to a 4% drop in eciency, which is matched by a corresponding increase of 8% in ow capacity simultaneously for all the turbines. As in case study A, combinations B1 ! B28 show dierent sets of instrumentation that can be used to simultaneously diagnose faults in all the turbines (two in this case) present on the 2-shaft engine. B29 shows a possible combination of sensors that is not diagnostically useful for turbine-fault isolation: the reasons for this are as given in case study A. Again, the RMS value for B30 obtained from a redundant set does not justify over-instrumenting an engine. In case study C (Table 6), a combined fault scenario involving the compressor and all turbines on the engine was analysed. An EDI of 1%, indicating a combination of FI=1% and EI=1%, was used initially and incremented. As before, C1!C28 are dierent instrumentation sets that can be used in fault diagnoses for faults that aect both the compressor and turbines simultaneously, as well as aect individual components. C29 indicates a possible combination that is not optimised for such tasks; this being portrayed by the non-convergence of the NLGPA and the high value of the LGPA. C30 shows the set involving redundant instrumentation. In case study D (Table 7), the same scenario is presented as in case study C, but this time with the COT as the engines power-setting parameter. One major point being highlighted in this study is the improvement obtained by using NLGPA rather than LGPA. The successive iteration, to an appropriate convergence using LGPA as the rst solution, provides a better approach to dealing with the non-linear problem of parameter inter-relationships. Fig. 6, which is the bar-chart format of the RMS values of Table 6, shows some relative benets of the NLGPA approach. The degree of estimation accuracy involved when NLGPA is used is higher than that obtainable from LGPA, and for combinations that were

Fig. 6. Comparison of NLGPA and LGPA RMS values for various instrumentation choices.

44

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

deemed inappropriate such as C29. LGPA provides an answer, but with a large RMS error, while the NLGPA situation simply did not converge. It should be noted that the RMS value is indirectly proportional to the degree of optimality of the given set. In the 28 optimal combinations considered for the 2-shaft engine under multiple fault scenarios, with SHP as the power-setting parameter and twenty-three optimal combinations obtained with COT as the power-setting parameter, the usage frequencies of the various sensors are given in Figs. 7 and 8. It is worthwhile to recall

Fig. 7. Usage frequency of sensors for optimal instrumentation-set selection for a 2-shaft engine with SHP as the power-setting parameter and faults implanted on all compressors and turbines.

Fig. 8. Usage frequency of sensors for optimal instrumentation-set selection for a 2-shaft engine with COT as the power-setting parameter and faults implanted on all compressors and turbines.

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

45

that, a rigorous approach was adopted to determine the selection and location of the sensors on the engine. This is because a poorly positioned or an inappropriate sensor in a given set could result in high RMS values, which indicate an unwise choice. This being the case, the sensors were randomly removed and others added into the set with the program re-run until a better combination was obtained. From Figs. 7 and 8, it is shown that some sensors, such as N1, P3, T3, P6, P8, WFE and T9, are crucial to a proper fault-diagnosis irrespective of the control parameter chosen. This assertion is made, based on the grounds that for all the optimal instrumentation-sets generated, these sensors were found repeatedly useful in the diagnostic process. On the other hand, some sensors, such as T6, T8, P9 and SHP, are power-setting parameter-dependent. This is obvious for the T6 and SHP sensors. The authors have considered various case studies and various possible combinations. This was motivated by the cost of installation and maintenance for each sensor varying according to sensor type and its location on the gas-turbine: therefore it was deemed wise to provide a wide range of possible combinations from which to choose. In real applications, especially for industrial gas-turbines, the available instrumentation set installed by the OEM could be well below that recommended by this study. Nevertheless, these provided sets may still be sucient for isolating most engine faults, but in the case of simultaneous multiple-component faults, they could prove ineective.

9. Conclusions The gas-turbine user or manager will benet from reductions in maintenance frequency and overhaul costs provided by a more exact knowledge of the engines status. This can be achieved with appropriate instrumentation that indicates likely faults. Furthermore, an optimal instrumentation set would allow for the simultaneous diagnosis of multiple faults, as the possibility of their occurrence exists. Knowledge of optimal sensor-sets should be indispensable to diagnostic and commissioning engineers. Because some faults result in identical signatures, diagnostic reports are normally presented as involving any of the possible faults with that feature. That is, for identied change(s) in the independent variables, the faults associated with such signatures are outlined for closer consideration in order to isolate the actual fault, possibly using other health-monitoring techniques. These techniques, such as vibration and oil analyses, complement the GPA. If the level of errors, involved in the estimation of the fault-aected independent parameters, is not minimised, the probability of accurately determining the signatures created by such faults would be low. The use of the NLGPA approach in instrumentation-set selection provides a signicant improvement when compared with the LGPA method, because it addresses the non-linear nature of the problem. Successive applications of LGPA leads either to a divergence from or a convergence to an exact solution depending on the choice of instrumentation used.

46

S.O.T. Ogaji et al. / Applied Energy 73 (2002) 2546

Over-instrumenting the engine does not necessarily provide a better diagnosis, but rather, incurs the increase in cost arising from the installation and maintenance of such redundant instruments.

References
[1] Urban LA. Parameter selection for multiple-fault diagnostics of gas-turbine engines. ASME Journal of Engineering for Power 1975:22530. [2] Stamatis A, Mathioudakis K, Papailiou KD. Optimal measurement and health index selection of gas-turbine performance status and fault diagnosis. ASME Journal of Engineering for Gas-turbines and Power 1992;114:20916. [3] Ogaji SOT, Singh R. Study of the optimisation of measurement sets for gas-path fault diagnosis in gas-turbines. Proceedings of ASME Turbo Expo 2002, 36 June, Amsterdam, The Netherlands. No. GT-200230050; 2002. [4] Diakunchak IS. Performance deteriorations in industrial gas-turbines. ASME Journal of Engineering for Gas-turbines and Power 1992;114:1618. [5] Escher PC, Singh R. An object-oriented diagnostics computer-program suitable for industrial gasturbines. United 21st International Congress on Combustion Engines (CIMAC), Interlaken Switzerland 1518 May, 1995. [6] Zaita AV, Bulet G, Karlsons G. Performance deterioration modelling in aircraft gas-turbine engines. ASME Journal of Engineering for Gas-turbines and Power 1998;120:3449. [7] Upton AWJ. Axial-ow compressor and turbine-blade fouling: some causes, eects and cleaning methods. First Symposium on Gas-turbine Operation and Maintenance; National Research Council of Canada, 1974. [8] Lakshminarasimha AN, Boyce MP, Meher-Homji CB. Modelling and analysis of gas-turbine performance deterioration. ASME Journal of Engineering for Gas-turbines and Power 1994;116:4652. [9] Aker GF, Saravanamuttoo HIH. Predicting gas-turbine performance behaviour due to compressor fouling using computer-simulation techniques. ASME Journal of Engineering for Gas-turbines and Power 1989;111:34350. [10] Agrawal RK, MacIsaac BD, Saravanamuttoo HIH. An analysis procedure for validation of on-site performance measurements of gas-turbines. ASME Journal of Engineering for Power, Paper No. 78GT-152, 1978. [11] Urban LA. Gas-path analysis applied to turbine-engine condition monitoring. AIAA/SAE 8th Joint Propulsion Specialist Conference. AIAA No.721082; 1972. [12] MacLeod JD. Computer modelling of implanted component faults in a gas-turbine engine. AIAA/ SAE/ASME/ASEE 27th Joint Propulsion Conference; Sacramento, CA, 2426 June, 1991. [13] MacIsaac BD. Engine performance and health monitoring models using steady-state and transient prediction methods. Advisory Group for Aerospace Research and Development AGARD-LS-183 1992:9.19.21. [14] Batcho PF, Moller JC, Padova C, Dunn MG. Interpretation of gas-turbine response due to dust ingestion. Journal of Engineering for Gas-turbines and Power 1987;109:34452. [15] Grewal MS. Gas-turbine engine performance deterioration modelling and analysis. PhD dissertation, SOE, Craneld University; 1988. [16] Donaghy MJ. Gas-path analysisfault diagnosis using DETEM. MSc thesis, SOE, Craneld University; 1991. [17] Singh R. Managing gas-turbine availability, performance and life usage via advanced diagnostics. 44th Gas-turbine Users Association Annual Conference, Dubai, UAE; 1999. [18] Berenblut BJ, Masom RA. Radiation pyrometry for gas-turbine enginesan introduction. The British Journal of Non-Destructve Testing 1982;24(5):2689. [19] Gas-turbine World Handbook, Vol. 19A. Requot Publication; 19981999.