You are on page 1of 5

An Optimal Maintenance Policy for Compressor of a Gas Turbine Power Plant

P. Naga Srinivasa Rao V. N. Achutha Naikan


Reliability Engineering Centre, Indian Institute of Technology, Kharagpur 721 302, West Bengal, India

This paper proposes a condition based maintenance policy for compressors of industrial gas turbine. Compressor blade fouling contributes a major performance loss in the operation of gas turbine. Water washing is usually done for recovery of the blade fouling problem. There exist two different washing methods, namely, online and ofine washings. Many researchers suggested that performing a combined program of regular online washing plus periodic ofine washing would give fruitful results with respect to economy. However, such studies are of empirical nature or have considered only deterministic treatment. Considering the rate of fouling as discrete state random process, we propose a condition based maintenance policy with periodic online washing and inspection directed ofine washing. According to this policy, the compressor undergoes regular online washes for every 1 / m operating hours, and also undergoes inspections at constant rate I. If the observed condition at an inspection is worse than threshold deterioration state, then perform ofine washing. Otherwise, continue with online washing. The proposed algorithm gives optimum schedules for both online washing and inspections considering minimization of total cost per operating hour as objective. It also gives optimum threshold deterioration level for performing ofine washing. A comparison of the results for a hypothetical gas turbine compressor is presented as illustration. DOI: 10.1115/1.2795762 Keywords: gas turbine fouling, preventive maintenance, condition based maintenance, online washing, ofine washing

Introduction
With the growing interest in life cycle costs for heavy-duty gas turbines, equipment operators are investigating the trade-off between performance improvements and associated maintenance costs. One of the key factors leading to performance losses during the gas turbine power plant operation is compressor fouling. The performance degradation attributed to compressor fouling is mainly due to deposits formed on the compressor blades by particles carried in by the air. Despite the use of advanced ltering methods and lter maintenance, some of the particles are not large enough typically, a few microns diameter to be blocked by these inlet lters. Therefore, the ingestion of substances that can cause fouling cannot be completely suppressed. Depending on the environment, these particles may range from dust and soot particles to water droplets or even insects, causing an increase in surface roughness, changes in the aerofoil shapes and/or aerofoil inlet angles, and a narrowing of the aerofoil-throat aperture. Fouling also takes place when the plant is operating in a sandy or dusty environment, which severely erodes compressor blades. The rate of fouling further increased when the operating environment includes the presence of alkali or similar minerals, which adhere to compressor blades when sulfur is present in the environment. The water quality is also a signicant parameter, which directly inuences the fouling rate 1 . In general, plant operating environment contains a combination of the above mentioned conditions. Compressor blade fouling reduces the ow capacity, and thereby the pressure ratio of the unit 25 . This results in an overall loss in power output and efciency of a gas turbine. It is one of the commonest causes of
Submitted to ASME for publication in the JOURNAL OF ENGINEERING FOR GAS TURPOWER. Manuscript received April 1, 2006; nal manuscript received August 2, 2007; published online January 22, 2008. Review conducted by Dilip R. Ballal.

performance reduction encountered by the gas turbines 4,6 and can account for 7085% of the performance loss during operation 7,8 . A deterministic cost estimate is given by Diakunchak 2 . In extreme cases, fouling may also result in surge problems. The fouling rate depends largely on the following ve factors: site location surrounding environment layout of the air intake system atmospheric parameters plant maintenance

BINES AND

The rst four factors belong to plant design that management makes decisions on before constructing the plant. Once the plant constructed, the rst four factors cannot be inuenced during the operation; the plant maintenance is the critical one for preventing extra costs resulting from degraded plant performance. Fortunately, compressor fouling is a recoverable deterioration in that periodic cleaning of the compressor can minimize it. Various methods have been used in the past to clean fouled compressors. At times when heavy-duty gas turbines did not yet possess highly sophisticated cooling schemes and coated compressor blades, cleaning was achieved by abrasion with the injection of solid compounds such as nutshells or rice husks. This had to be replaced by wet cleaning methods water or solvent based to protect modern coatings and to keep state-of-the-art cooling systems from blockages. The most effective wet cleaning process is the crank soak or ofine wash. For this, the unit has to be shut down and cooled off in order to assure that the cleaning agent reaches all compressor stages and does not evaporate. The cleaning agent is injected into the compressor with the turbine turning at low speed. After a soaking time, the compressor is rinsed with water, which must be drained from the engine. Before the unit can be operated again commercially, it has to be dried. Thus, ofine washing reduces the availability of a unit. MARCH 2008, Vol. 130 / 021801-1

Journal of Engineering for Gas Turbines and Power Copyright 2008 by ASME

Downloaded 11 Jul 2011 to 80.191.123.192. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 1 Effects of washing and overhaul on compressor efciency

The present research is aimed i to propose a method of mathematical modeling of the compressor online and ofine washings with considering the random nature of the fouling and ii to propose an optimum condition based preventive maintenance policy for the compressor of a gas turbine power plant. Many researchers Meher-Homji and Bromley 10 and Stalder 11 proved that the fouling rate of gas turbine compressor is random and follows an exponential law. Therefore, nding of the optimum schedules for i mean time between successive on line washes, ii mean time between successive inspections by considering the fouling rate of compressors as exponential, and also iii threshold deterioration level for performing ofine washing is considered as objectives of the proposed research. Here, the purpose of the inspections is to know the condition of the compressor in order to make a decision regarding whether or not to perform an ofine wash.

With a growing number of gas turbines being used in combined cycle or combined heat and power applications, there was a need for the development of online washing systems that would provide acceptable performance improvement without shutdown and that would extend the operating period until the next ofine washing. These systems are now state of the art in modern heavy-duty gas turbines. In an online wash, demineralized water or a mixture of demineralized water and detergent is injected into the compressor while the gas turbine is running. The uid droplets impact and thoroughly wet the blades and inner surface of the compressor at high speeds to loosen and partially remove deposits. However, high performance recovery can only be achieved by an ofine wash. Therefore, to get the benets of both higher availability and increased degradation recovery, it is required to perform a combined maintenance policy with online and ofine water washes. Proper combination of online and ofine washing schedules can give optimum benets by minimizing the life cycle cost of the plant. The basic objective of ofine cleaning is to clean a dirty compressor and to restore the power and efciency to the maximum extent possible. The primary objectives of online washing are the following: a b c to maintain the cleanliness of a compressor after ofine washing to maintain power and efciency by minimizing ongoing losses, and to extend the period between ofine washes by minimizing the buildup of deposits in the compressor, thereby reducing the ongoing incremental power losses.

Denitions
Failure Type i. Deterioration: A process where the important parameters of a system gradually worsen. If left unattended, the process leads to deterioration failure for example, wear, dimensional changes with time, effect of contamination, and deleterious environmental factors, e.g., temperature, voltage, and radiation are relevant examples and sources of component degradation . For most of the systems, it is possible to measure the amount of deterioration incurred up to time t either by directly measuring the above mentioned parameters or any other dependent parameters such as vibration, power, performance, etc. The deterioration failure may or may not be a physical failure of the system. It is a situation where the equipment performance is unacceptable due to economy or statutory environmental regulations, or any other . In this paper, we are using the term deterioration failure to represent the situation where the compressor performance is unacceptable. ii. Random failure: A failure whose occurrence is unpredictable except in a probabilistic or statistical sense for example, failure of an automobile to start, occurrence of fatigue crack on the compressor blade surface . Maintenance Type i. Inspection: Inspections at specied schedules are performed to know the operating condition of the system. It involves direct measurement of some related parameters of the system, which directly represent the current operation condition of the system. ii. Preventive maintenance PM : A type of restoration or maintenance activity wherein an unfailed system, at preselected intervals, has its accumulated deterioration reduced or eliminated. iii. Minimal PM: Maintenance of limited effort and effect. For deterioration modeled as discrete stage process, minimal PM activity restores the system to the previous deterioration stage or to deterioration Stage 1, whichever is worse. iv. Major PM: Maintenance wherein a system is restored to as good as new condition. v. General PM: It restores the system to the previous deterioration stages or to deterioration Stage 1, whichever is worse. vi. Repair: A restoration wherein a failed system is returned to a working condition. vii. Minimal repair: Repair wherein the system is returned to that operable state in which it was operating just before the failure. viii. Perfect repair/overhaul/replacement: A restoration action wherein the failed system is returned to as good as new condition. Transactions of the ASME

Ofine washing requires shutdowns of the plant and cooling of the components; therefore, it requires more time and money, and it also reduces the throughput of the plant. On the other hand, outages or shutdowns are not required for online cleaning. The effect of the various maintenance tasks washing and overhaul on gas turbine engine efciency is shown in Fig. 1.

Statement of the Problem


With a growing presence of gas turbine technologies and competition among the industries, a stronger focus is being placed on trade-off analysis between performance optimization and operational and maintenance O&M costs. As a result, cost/benet evaluation of performance recovery methods has been at the forefront of these efforts. However, optimization of both compressor ofine crank and online washing intervals from the standpoints of operating cost and proactive maintenance is of primary interest to the organizations and the focus of this paper. Although the benets of compressor washing are undoubted, the frequency of online and ofine washes and the type of detergents to be used if at all are widely debated issues. It is essential to develop maintenance schedules based on the deterioration characteristics of the plant in order to balance the maintenance costs with lost revenue and extra operating costs 9 . 021801-2 / Vol. 130, MARCH 2008

Downloaded 11 Jul 2011 to 80.191.123.192. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 2 State-transition diagram for periodic online washing and condition based ofine washing policy

Assumptions
The following assumptions are made. 1. The compressor encounters two types of shutdowns. The rst one is due to unacceptable level of deterioration, and the other is due to random failure of the compressors parts for example, blade fracture . 2. Compressor deterioration is modeled as discrete state process with k acceptable operating stages and one unacceptable failed stage. The compressor has a deterioration failure unacceptable level of deterioration immediately following the completion of k stages of deterioration. 3. The duration of each deterioration state has an exponential distribution with rate k d. That gives a mean life of 1 / d between successive overhaul shutdowns. 4. Following deterioration failure unacceptable level of deterioration , the compressor is overhauled, i.e., restored to as good as new. The overhaul duration is exponentially distributed. 5. Random failure occurs at an increasing rate or intensity, depending on the deterioration stage of the compressor. Minimal repairs are performed. The minimal repair duration is exponentially distributed. 6. The time between successive online washes is exponentially distributed; online washing makes the device 1 deterioration states younger. That is, if the deterioration stage before an online washing is nth, then the deterioration stage after the online washing is n 1 th. 7. The compressor undergoes inspection at constant rate based on the observed deterioration stage Di of the compressor; an ofine wash can take place if i k1 , which makes the compressor 2 deterioration stages younger, or do nothing if i k1 . 8. Mean times to perform online and ofine washings are exponentially distributed. 9. There are no transitions between the failure and maintenance states.

tion stage and is under ofine washing. RFi is the state where the device is in the ith deterioration stage and is under repair after a random failure. In this model, the compressor undergoes periodic online washings with constant rate m with a mean duration of 1 / pm that makes the compressor 1 deterioration stages younger. The compressor is inspected after a random period that is exponentially distributed with mean 1 / I. If the device is found in deterioration stage i such that i k1, then ofine washing of the compressor takes place that restores the compressor to i 2 deterioration stage with a mean duration of 1 / pM. When the device is in the random failure state, minimal repair takes place that brings the device to just before failure condition in operating mode with a mean duration 1 / DF. When the device is in deterioration failure DF state, complete overhaul is carried out to bring the device to as good as new state with a mean duration of 1 / DF. To increase the clarity of the gure, the random failure states of the compressor are not shown in Fig. 2.

Steady-State Solution
Using the Markov approach associated with Fig. 2, we rst obtain the ChapmanKolmogorov equations Misra 12 . These are then converted into a set of linear equations also called system balance equations representing the steady-state conditions of the process in Fig. 2. Their solution yields the long-term state probabilities. The state-transition equations for the proposed maintenance policy depend on the real values of the number of deterioration stages k, limit for starting ofine washings k1 , recovered deterioration states by performing online washing 1 , and that by ofine washing 2 . The state-transition equations of the system depend on the following three important quantities: k 1 k 2 k1 2

Values of these quantities depend on the values of the system parameters k, k1, 1, and 2 . Based on the possible ranges of values of these quantities, we developed different sets of statetransition equations. Any one of them will execute depending on the above parameters. A recursive solution algorithm and software code in C has been developed for solving these linear equations for steady-state probabilities of the different states of the system.

Optimization
Optimal inspection and maintenance schedule can be obtained by minimizing the total cost. The cost components include maintenance, repair, and out-of-service costs. A simple and more appropriate model for the cost analysis is proposed. Let CI, Cpm, CpM, CRF, CDF, Cdt, and Cop be, respectively, the unit cost components for various outages for inspections, online washings, offline washings, repairs after random and DF, down time, and operation. We rst calculate respective cost for a deterioration replacement cycle CT . CT is the expected time between successive deterioration replacements. From these costs, cost per operating hour or cost per unit operating time is calculated for decision-making purpose. If CT is the expected time between successive deterioration replacements, then for a CT, there is one deterioration replacement. Therefore, CTPDF = 1/
DF

Model
Continuous time Markov chain CTMC model for the proposed condition based preventive maintenance policy of a gas turbine compressor is shown in Fig. 2. According to this policy, the compressor undergoes periodic online washings, and inspection directed ofine washings. States D1 to Dk are the k operating states in increasing order of deterioration. D1 is the best operating state and Dk is the worst. Di represents the state in which the compressor is in the ith deterioration stage and is in working condition. Ii represents the state in which the device is in the ith deterioration stage and under inspection. Pmi is the state where the device is in the ith deterioration stage and is under online washing. PMi is the state where the device is in the ith deterioraJournal of Engineering for Gas Turbines and Power

CT = 1/

DF PDF

MARCH 2008, Vol. 130 / 021801-3

Downloaded 11 Jul 2011 to 80.191.123.192. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Cost Model
Expected costs per CT for DF, inspections, preventive maintenance, random failures, down time, and operating costs are calculated as follows: i. Expected cost of DF during CT Since for a DF cycle CT there exists only one expected replacement due to DF, therefore, E TCDF = CDF ii. Expected cost for performing inspections during CT,
k

E TCpM =
i=1

P Mi

pMCpMCT

v.

Expected cost for random failures during CT,


k

E TCRF =
i=1

P fi

CRFCT

vi.

Expected cost of downtime during CT,


k

E TCdt =
i=1

PIi + Pmi + P Mi + P fi + PDF CdtCT 7

E TCI =
i=1

PIi

IC piCT

3 vii. Expected operating cost during CT,


k

iii. Expected cost for performing online washings during CT,


k

E TCop = Pmi
pmCpmCT

Pi + P pmi
i=1

CopCT

E TCpm =
i=1

iv.

Expected cost for performing ofine washings during CT,

Using i vii , the expected cost per operating hour is expressed as

E TC =

E TCDF + E TCI + E TCpm + E TCpM + E TCRF + E TCop + E TCdt CT

Optimization Procedure
The optimization procedures are as follows. i. Depending on the accuracy required and/or deterioration replacement rate, choose an appropriate unit for a b c ii. iii. iv. v. number of deterioration states k time between inspections 1 / I time between online washings 1 /

Set k1 = 1 Set 1 / m equal to 1 Set 1 / I equal to 1 Calculate different deterioration state probabilities and availability. The complete procedure for evaluation of these quantities is available in Ref. 13 . vi. Using Eqs. 1 9 , nd the expected total costs for various cost components, and cost per operating unit time vii. Increase 1 / I by 1, and repeat the calculations from step v. until 1 / I reaches a maximum 1 / d viii. Select the PM interval time 1 / I that minimizes the cost per operating hour, and note the corresponding k1, 1 / m , and 1 / I values ix. Increase 1 / m by 1 and repeat steps from iv to x x. Increase k1 by 1 and repeat steps from iii to xi xi. By minimizing the total costs, select a policy from all the selected policies in step x

Illustration
Let us consider a 30 MW gas turbine power plant. The output of the plant drops to 27 MW 10% deterioration in power output after an average operation of 500 h if it is not washed. At this stage, the performance of the turbine is not economical and also the condition is not safe. Therefore, this state is treated as DF stage; a complete overhaul is required to make the system like new, which takes more downtime and money. Currently, washes 021801-4 / Vol. 130, MARCH 2008

are performed on a preventative schedule of 50 h for online washing and 500 h for ofine washing. This maintenance policy is adopted by neither engineering assessment of conditional need nor optimal time to perform. In addition to the loss of availability and maintenance time incurred, unnecessary washes generate an environmental impact with the used detergent. Clearly, operating with a module that assesses condition and predicts the time to wash more appropriately would benet the organization and environment. In addition to DF, the system also undergoes random failures whose intensity increases with deterioration. After every random failure, only minimal repair is carried out. Therefore, the system condition after repair is the same as that of before failure condition. Each ofine wash requires a mean downtime of 50 h and average cost of $3000. Online wash takes 15 min, requires no downtime, and incurs average cost of $2000. Take 1 = 1 MW, 2 = 2 MW. The management wants to decide the optimum schedules for online and ofine washes, which reduces total cost by minimizing unnecessary washes. To get more accurate results, we model the total deterioration 3 MW performance reduction with large number of deterioration stages 60 deterioration stages each represents 50 kW deterioration. Therefore, each online washing and ofine washing makes the plant, respectively, 20 and 40 deterioration states younger. The following parameters are deducted from the above data. Let 1 / d = 500 h, 1 / pm = 0.25 h, and 1 / DF = 50 h. The following data are also considered. Random failures of the device have a Weibull intensity with scale parameter = 500, shape parameter = 3, online washing cost Cpm = $ 2000, minimal repair cost at random failure CRF = $ 2000, and ofine washing cost CDF = $ 3000. The mean time to minimal repair at random failure is 25 h. As the deterioration of the compressor due to blade fouling increases, the operating cost of the power plant also increases. It includes fuel and other consumables; its value depends on the deterioration stage of the plant. Now, a days energy prices fuel and electricity change rapidly. This will be taken care by systematically collecting the data on Transactions of the ASME

Downloaded 11 Jul 2011 to 80.191.123.192. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 1

Solution for the illustration Time between Total cost per Savings per ofine washes Availability operating hour year $ 500 5100 2213 0.906883 0.984359 0.972896 1079 1065 1062 0 120,722 144,884

Policy Present online wash for 50 h, and ofine wash for 500 h of operation Time based optimal PM policy Condition based PM policy

Time between Time between inspections online washes 1/ I 1/ m 50 60 69 15

k1 30

changes in price of these commodities, and nding a relation with time. We can make a forecast for the next year either on weekly or monthly basis . In the example, we have considered a linear relationship between operating cost and level of deterioration of the compressor. If we consider the varying price, then this function will have a nonlinear nature. This requires collection of data on uctuation of the fuel price on a monthly or yearly basis, and preparing and updating the washing schedules. In this illustration, we assume that the operating cost at base load operating condition for state D1 is $1000/ h. It uniformly increases with deterioration, and at worst operating condition D60 , it is $1200/ h. Solution for the Illustration. For this problem, if we consider availability maximization as objective, we require to perform continuous online washes of the compressor. However, as mentioned earlier, unnecessary washes require large resources. It will also generate an environmental impact due to the large quantity of detergent used. So, total cost minimization is the better objective for this problem. Following the different steps of the optimization procedure as discussed earlier, we get the optimal solution for total cost minimization. This is shown in Table 1. By the analysis of the results in Table 1, the following observations are made. i It is estimated that by applying the proposed model, we can reduce the annual operating cost of the plant up to a sum of $148,920. ii Signicant improvement in availability is also achieved with the application of the proposed model from 90.7% to 97.3% . This increased availability improves the throughput of the plant, and thereby protability. iii Reduced environmental pollution is also achieved with the proposed model.

line washing and inspection directed ofine washing. Proposed policy improves the availability of the plant by optimizing online and ofine washing frequencies. Considerable amount of savings in operating cost, considerable gain in availability, and thereby protability are also possible with the application of the proposed model.

References
1 Boyce, M. P., and Gonzales, F., 2005, A Study of On-Line and Off-Line Turbine Washing to Optimize the Operation of a Gas Turbine, ASME Turbo Expo 2005, Reno, NV, June, ASME Paper No. GT2005-69126. 2 Diakunchak, I. S., 1992, Performance Deteriorations in Industrial GasTurbines, ASME J. Eng. Gas Turbines Power, 114, pp. 161168. 3 Zaita, A. V., Bulet, G., and Karlsons, G., 1998, Performance Deterioration Modelling in Aircraft Gas-Turbine Engines, ASME J. Eng. Gas Turbines Power, 120, pp. 344349. 4 Lakshminarasimha, A. N., Boyce, M. P., and Meher-Homji, C. B., 1994, Modelling and Analysis of Gas-Turbine Performance Deterioration, ASME J. Eng. Gas Turbines Power, 116, pp. 4652. 5 Aker, G. F., and Saravanamuttoo, H. I. H., 1989, Predicting Gas-Turbine Performance Behaviour Due to Compressor Fouling Using ComputerSimulation Techniques, ASME J. Eng. Gas Turbines Power, 111, pp. 343 350. 6 Upton, A. W. J., 1974, Axial-Flow Compressor and Turbine-Blade Fouling: Some Causes, Effects and Cleaning Methods, First Symposium on GasTurbine Operation and Maintenance, National Research Council of Canada. 7 Christoph, P. L., Christoph, S., and Lutz, D., 2004, Performance Benets Using Siemens Advanced Compressor Cleaning System, ASME J. Eng. Gas Turbines Power, 126, pp. 763769. 8 Ogaji, S. O. T., Sampath, S., Singh, R., and Probert, S. D., 2002, Parameter Selection for Diagnosing a Gas-Turbines Performance-Deterioration, Appl. Energy, 73, pp. 2546. 9 Tarabrin, A. P., Schurovsky, V. A., Bodrov, A. I., and Stalder, J. P., 1998, An Analysis of Axial Compressor Fouling and a Blade Cleaning Method, ASME J. Turbomach., 120, pp. 256261. 10 Meher-Homji, C. B., and Bromley, A. F., 2004, Gas Turbine Axial Compressor Fouling and Washing, 33rd Turbomachinery Symposium, Turbomachinery Laboratory, Texas A&M University, Houston, TX, Sept., pp. 163192. 11 Stalder, J. P., 2001, Gas Turbine Compressor Washing State of the Art: Field Experiences, ASME J. Eng. Gas Turbines Power, 123, pp. 363370. 12 Misra, K. B., Reliability Analysis and Prediction: A Methodology Oriented Treatment, Elsevier Science, Amsterdam, 1992. 13 Rao, P. N. S., and Naikan, V. N. A., 2006, Generalized Condition-Based Preventive Maintenance Policy for Markov Deteriorating Systems, Int. J. Performability Engineering, 2 2 , pp. 175189.

Conclusions
Considering the rate of fouling as discrete state continuous time random process, we propose a condition based maintenance policy for compressors of industrial gas turbine with periodic on-

Journal of Engineering for Gas Turbines and Power

MARCH 2008, Vol. 130 / 021801-5

Downloaded 11 Jul 2011 to 80.191.123.192. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

You might also like