Opportunstc Mantenance n Araft usng Relevant Condton Parameter based Approach Hartha Saranga * Indan Insttute of Management Calcutta P O Box 16757, Alpore (Post) Calcutta 700 027, INDIA Emal:hartha@mcal.ac.n 1 Abstract Opportunstc mantenance has been an essental part of all systems-mantenance, mplctly f not explctly. However, as systems become more expensve and complex, the decsons nvolvng opportunstc mantenance actvtes become more complcated. In ths paper, we dscuss a few ssues that arse whle carryng out the opportunstc mantenance, and try to resolve them wth the help of a popular optmzaton technque called Genetc Algorthms. We also present a few results concernng relevant condton parameter based mantenance, as t has a hgh potental to be an opportunstc mantenance n complex systems. A systematc methodology s desgned to enable the mantenance ew n decdng whch tems to be mantaned when an opportunty arose. The cost of premature replacement s compared wth the cost of down tme, n the optmzaton process usng Genetc Algorthms. Keywords: Opportunstc Mantenance, Condton based Mantenance, Relevant Condton Parameters, Condton Montorng, and Genetc Algorthms 1. Introducton Opportunstc mantenance plays a ucal role n systems that are mantaned under both tme-based as well as condton-based mantenance polces. In opportunstc mantenance, when a system or module s grounded for correctve or preventve mantenance, that opportunty s utlzed to do mantenance on other parts of the module, whch are found to be damaged or have started to deterorate. On one hand, ths mproves the safety and relablty of the system, and on the other hand t reduces the downtme by avodng un-scheduled mantenance. Ths n turn reduces the cost of mantenance, loss of revenue due to extra groundngs, customer goodwll and so on. But, 1 * Correspondng author. Tel.:+91-033-467-9189, Extn.544(O),285(R); Fax:+91-033-467-8307. E-mal adress:hartha@mcal.ac.n
what s the teron used to decde whch parts to be consdered for opportunstc mantenance and how far does one go? There s the engneerng maxm, whch says that, unless t s broken, don t fx t! Therefore, there are numerous factors that need to be consdered before gong ahead wth opportunstc mantenance of a partcular part. A detaled despton of varous stuatons that demand an n-depth analyss of advantages and dsadvantages of carryng out the opportunstc mantenance s gven below. Mantenance cost has always been the most unappealng and at the same tme the most unavodable cost of a systems Lfe Cycle Cost (LCC). Especally n an arlne ndustry, mantenance of an araft, commercal or defense, contrbutes a substantal amount of expendture to the LCC. For example, the routne mantenance program of an araft tself s very extensve, and conssts of mantenance checks lke preflght before each flght, A and B checks, whch are performed on a regular bass, once n every 150 and 750 flght hours respectvely. In addton to these checks, major nspecton and recondtonng are done through C and D checks, whch come under heavy mantenance, every 3,000 and 20,000 flght hours respectvely. The annual demand for heavy mantenance of a narrow body araft s 11,000 labor hours and for a wde body araft, t s 17,150 labor hours approxmately [3]. In spte of ths major nspecton and mantenance schedules, araft stll requres condton-based mantenance for most part of the engne and correctve mantenance n some cases. Although, most part of the engne comes under condton-based mantenance, n practce, t s very dffcult to estmate the condton of the tems wth boroscopes and ntrascopes [1], as only a part of the surface s vsble and that too at a very oblque angel. Thus, most of ths oncondton mantenance s carred out as an opportunstc mantenance, when the engne or any module of the engne s strpped down to part level durng the shop vsts and performance restoratons. In such stuatons, f some safety sgnfcant tems seem to have been damaged, then they are obvously repared or replaced. But the dffculty arses, when some potental falures of non-safety sgnfcant tems are detected. These falures, f and when they occur, would only cause small nease n vbraton, reducton n thrust or specfc fuel consumpton etc., whch may result n engne beng run hotter to acheve the requred performance. However, f there was no actual falure and these
secondary damages do not measure up to the cost of the remanng lfe of the component, the mantenance ew s n an ambguty as to carry out a premature replacement or not. Another mportant factor that puts the mantenance ew n a dlemma durng opportunstc mantenance s features lke hard lfe (hard tme) and soft lfe (soft tme). Hard lfe s the age of the component at or by whch the component has to be replaced. For example, at least 10% of parts n an araft engne consst of Lfe Lmted Parts (LLPs), whch are replaced at pre-determned tme ntervals, rrespectve of ther condton. Safety tcal parts lke dscs and shafts are gven hard tmes and come under LLPs, as they can cause the loss of araft f they burst. These parts are usually very expensve and any wastage of lfe remanng n a LLP wll sgnfcantly add to the mantenance costs. But at the same tme, an unscheduled groundng of the araft or removal of the engne, just to replace a few LLP's would cost the operators huge sums of money. A trade off would be to replace the LLP's durng the regular shop vsts for performance restoratons or full overhaul [4]. However, snce all the LLP's may not have same "hard tmes", t s always not feasble to wat untl the complete useful lfe of an LLP s exhausted and hence at least few LLP s are replaced prematurely durng an opportunstc mantenance. On the other hand, soft tmes are lfe thresholds for certan hot components of an engne, and usually determne the absolute lmts for a performance restoraton and full overhaul. Soft tmes are n fact the bass of on-condton engne mantenance. Whle some components may have reached ther lfe thresholds, others may not and may also be n good workng condton, so do not have to be repared or replaced. Now the thousanddollar queston would be, how to decde whether to replace them now or not to avod another engne removal n near future, when the soft tmes would have reached ther thresholds. Another mportant aspect, whch adds to the complexty of opportunstc mantenance, s the level of workscope requred by the arlne operators. Dependng on ther requrements, the arlne may wsh to have the engne overhauled ether so that t remans on the wng for maxmum tme possble, or just the mnmum work done to keep the engne servceable for a shorter perod at lower cost. Ths later practce s typcal when an arlne expects to keep the araft or engne n servce for only a short tme or t can acqure used engnes on the market for a prce lower than the cost of overhaul [3].
Condton based mantenance s replacng tme based mantenance n modern araft, especally n case of mechancal components, whch undergo degradaton falure mechansms lke ack growth, corroson etc. As mentoned earler, 90% of the parts n araft engne undergo condton-based mantenance, whch s carred out durng scheduled shop vsts. Now the queston arses, when some expensve parts are found to be deteroratng, and there may be a potental falure n future. These parts are not always necessarly repared or replaced, mmedately after the deteroraton s detected. For example, fracture tcal hardware such as rotors and dsks are assgned servce lmts for ack growth, where acks are allowed to grow to a specfed sze before the hardware tem s repared or replaced [12]. There fore, tems that fall under ths category also need to be analyzed for cost of premature replacement aganst the cost of down tme n future. Also, there s a possblty that there may be enough components wth more or less equal useful lves left, whch wll collectvely compensate the cost of down tme for a future groundng. Ths s another factor that f optmzed would result n a sgnfcant savng to the arlne operators. Thus, n the present paper, we manly concentrate on two key ssues. Frstly, as mentoned n the prevous paragraph, condton-based mantenance s slowly but steadly takng over the tradtonal mantenance strateges lke tme-based mantenance (hard lfe). Hence, there s a need to develop varous tools and technques that not only montor the condton of varous components and subsystems, but also enable the mantenance ew to predct varous relablty characterstcs. Presently, many modern araft engnes are equpped wth Health and Usage Montorng Systems (HUMS) to montor varous mantenance sgnfcant tems. Strateges lke Relevant Condton Parameter (RCP) based mantenance [8&9], whch were developed manly to model the degradaton falure mechansms, can be ntegrated nto on-condton mantenance to predct the falures dependng on the orgnal condton of the components. RCP-based strategy s an effectve way of condton based mantenance, whch not only reduces the number of nspectons and hence downtme [10], but also lends tself to opportunstc mantenance gracefully. The latter part can be attrbuted to the unque features of RCP-approach, lke pror nformaton and lead-tme avalable to the mantenance ew regardng the emnent and potental falures of the components under observaton.
Secondly, we develop a procedure to decde what components to be ncluded n an opportunstc mantenance actvty, usng a popular optmzaton technque called Genetc Algorthms. In the followng secton we gve a bref despton of RCP-based mantenance. 2. Despton of RCP based Mantenance RCP-based mantenance s carred out n two parts. The frst part s a systematc approach, wth each step addressng a sequenced set of questons for each ndvdual tem of the system. The answers to these questons lead to the type of mantenance strategy that s most sutable to the correspondng tem. The second part s a mathematcal model, whch s the mplementaton of RCP-based approach to mantenance plannng. These are the four steps nvolved n the frst part [2]: Identfcaton of the mantenance sgnfcant tems (SSI s) Determnaton of all condton parameters Identfcaton of Relevant Condton Predctors (RCP s) Selecton of condton montorng technques Once the above steps are carred out, the next step s the mplementaton of RCPbased mantenance. The fundamental dea behnd RCP-based approach s to ntegrate relablty of an tem or a system nto mantenance plannng and to be able to predct the relablty of the system wth the help of condton montorng devses. In order to acheve ths, an RCP for each mantenance sgnfcant tem, whch s at a rsk of degradng, s beng montored wth a sutable condton montorng technque. For an tem to be able to functon satsfactorly ts RCP should le between certan presbed lmts, denoted by RCP n and RCP lm set by the manufacturers. Once the numercal value of RCP osses these lmts, the tem s qualfed as a falure. Therefore, n RCP-based approach, the relablty of an tem or system at tme t s defned as the probablty that the RCP lyng between the presbed lmts RCP n and RCP lm at tme t, whch can be gven by the followng expresson [7]. R(t) = P(RCP n < RCP(t) < RCP lm ) (1)
Dependng on the nature of these lmts, RCP may be dvded nto the followng four categores. 1. Fxed RCP n and fxed RCP lm 2. Fxed RCP n and dstrbuted RCP lm 3. Dstrbuted RCP n and Fxed RCP lm 4. Dstrbuted RCP n and dstrbuted RCP lm The relablty functons for each of the above mentoned categores and the cases nvolved wthn them are derved n Knezevc (1987), El-Haram (1995) and Saranga (2000) [8,9,2&10]. Once the relablty functons have been obtaned for each mantenance sgnfcant tem, dependng on the nature of the lmts, the next step s to plan the nspecton ntervals. In RCP-based approach, requred relablty s consdered as optmzaton teron to mantan each sgnfcant tem. Therefore, as soon as the relablty of an tem reduces to ts requred level of relablty, mantenance actons needs to be carred out to restore the tem to ts orgnal condton. In order to acheve ths, the 1 tme to the frst examnaton of each tem, denoted by T s calculated usng the followng expresson [Knezevc (1987b)], 1 n 1 lm R ( T ) = P( RCP < RCP ( T ) < RCP ) = R r (2) Where, r R s the requred relablty level for th tem and 1 ( 1 ) R T s the relablty of tem, at tme T. Now, we measure the relablty of each mantenance sgnfcant tem at tme to the frst examnaton, T to see f the measured value 1 M, s less than the RCP tcal value RCP, where RCP may be obtaned from the manufacturers. For all the tems, whose M RCP < RCP, the tem s allowed to operate untl the next tme to the examnaton. And for all the tems, whose M RCP RCP, requred mantenance actons be carred out to restore the tems to ther orgnal condton. The next subsequent tmes to the examnatons depend on the dfference between the measured value of RCP at the prevous tme to the examnaton and tcal value of RCP, and can be obtaned usng the expresson [2],
Where j T RCP f ( j 1) j ( j 1) M RCP ( T T ) M RCP T RCP ( ) j T ( τ ) dτ = f ( c) dc (3) j 1 s tme to the j th examnaton, T s tme to the (j-1) f M ( τ ) RCP ( t) s the probablty densty functon of M RCP at tme t and s the probablty densty functon of RCP at tme T. j th f examnaton, j RCP ( T ) Thus, for all those tems, whch dd not need mantenance at the tme of ( j 1) T examnaton, j=2,3 the tme to the next examnaton wll be T, where one of j ( c) the above two decsons are made dependng on ther measured value ( j 1) RCP T M ( ). For the tems that have undergone mantenance at ( j 1) T, the tme to the next examnaton wll be 1 T, as they are treated as good as new and the entre process starts all over agan. An mportant pont to note here related to opportunstc mantenance s that, although we know that the condton of RCP s tcal once t exceeds RCP, we stll do not know exactly how much tme t takes for RCP to reach RCP lm.e., to fal, from RCP. And hence we do not know the tme avalable for the mantenance ew n order to decde whether to ground t mmedately or to wat untl the system s avalable for mantenance n the mmedate future. In such stuatons, t s helpful to know the Resdual Lfe of the tem, whch s the mean remanng lfetme of the tem that has survved up to tme T0 and can be obtaned from, 1 MTTF ( T0 ) = R( t' ) dt' (4) R( T ) 0 T 0 = R( T 1 0 0 ( MTTF ) T R( t) dt) 0 Where t ' = t + T 0. Thus, once the tem reaches RCP, we can calculate the resdual lfe of the tem n order to know as to how long the tem wll survve.
Thus the queston that arses here, wth regard to opportunstc mantenance s that whether the nspectons or the mantenance actvtes can be carred out as a part of scheduled mantenance whenever there s an opportunty. The lead tme and the pror nformaton avalable to the mantenance ew regardng the nature of the potental or emnent falures enables them to plan the requred preventve or correctve mantenance actvtes accordng to ther convenence. But agan, the comparson between the cost of groundng and the cost of premature replacement needs to be made even n ths case. 3. Genetc Algorthms In order to ncorporate all the above-mentoned factors and lke nto the decson makng process of opportunstc mantenance, one needs to come up wth an effcent optmzaton tool that s robust enough. One such rapdly expandng optmzaton tool, recently beng used for opportunstc mantenance strateges s Genetc Algorthms (GAs). Genetc algorthms are a subclass of Evoluton Programs (EPs), whch mtate natural selecton process n searchng for an optmum soluton for an objectve functon called the ftness functon. GAs were frst proposed by Holland (1975)[7] and were further developed by hs student, Goldberg and others n the 1980s. The GAs dffers from most optmzaton technques, due to ther nature of searchng a populaton of solutons, rather than a sngle soluton. Savc et al (1995)[11&12] used genetc algorthms for optmum group replacement problem durng opportunstc mantenance. Followng key features of Genetc Algorthms [6] lend themselves to use GAs as an optmzaton tool n the current context. Optmzes hghly complex cost functons Optmzes wth contnuous or dsete parameters Smultaneously searches from a wde samplng of cost surface Deals wth a large number of parameters Provdes a lst of optmum parameters, not just a sngle soluton Works wth numercally generated data, expermental data or analytcal functons
Snce, whle talkng about araft mantenance, we are dealng wth thousands of unque parts, whch nvolve numerous factors to be consdered to be ncluded n the opportunstc mantenance, GAs, beng the best technque to model hghly complex cost functons, wll sute the present stuaton very well. The objectve here s to be able to decde, whether the tem under consderaton should be replaced under opportunstc mantenance or not, purely dependng on the bass of whether t s cost-effectve to replace t now or not. Those tems, whose hard or soft lves have been expred, are replaced wthout ambguty. Queston arses, only for tems wth useful lfe left n them, whch most of the tme are expensve, as otherwse are replaced prematurely whenever an opportunty arses, nstead of groundng the system. The choce of replacng t now should be weghed between the cost of premature replacement and the cost of groundng the system n future, purely for the purpose of replacement. Thus, wth each tem under queston, we attach a cost or ftness functon, and use Genetc Algorthms, to decde whch ones to be replaced. We formulate the ftness functon as follows: Where, Z, j Maxmze Z D * C j * C ] (5) j[ d 1 If the tem s replaced j hours before the scheduled tme = 0 Otherwse D = Downtme due to replacement C d = Cost of Downtme per hour C = Cost per hour for tem At ths pont, t s mportant to note that, we have only consdered the smplest case, just as an example, and the ftness functon may be extended to ncorporate varous factors that play a ucal role n decdng whether an tem should be replaced prematurely or not. In the present paper, we dscuss a step-by-step procedure to decde whether a partcular component should be ncluded n the opportunstc mantenance and suggest a method to decde when GAs can be used. To start wth, we need to dentfy the ponts at whch opportunstc mantenance can be performed. In general, for any complex system,
the below mentoned tmngs provde an approprate means to carryout opportunstc mantenance. 1. Scheduled mantenance ntervals: these are predetermned tmes, at whch scheduled tasks lke overhaulng, checks lke A, B, C and D etc. are carred out. 2. Correctve mantenance: when a module s grounded to repar or replace a faled tem. 3. Progressve nspecton: f an nspecton or examnaton of a condton montorng devse needs to be carred out. Once the opportunty arses, the next problem s to dentfy whch components to be consdered for repar or replacement amongst the components that are accessble and are at a rsk of potental falure. At ths pont, one needs to dvde the potental falures nto age related and non-age related, and focuses only on age related falures. And also, we concentrate only on mantenance sgnfcant tems, as the falure of non-sgnfcant tems should not affect the system performance drectly. The followng flow chart dagram wll be of great help n desbng the procedure. Notaton: MSI --- Mantenance Sgnfcant Item TBM --- Tme Based Mantenance RCPBM --- Relevant Condton Parameter Based Mantenance HT --- Hard Tme ST --- Soft Tme t --- tme of opportunstc mantenance t 1 --- tme of next scheduled mantenance RCP ---Relevant Condton Parameter RCP - - - Crtcal Value of RCP, for tem RCP (t) --- Value of RCP for tem at tme t GA --- Genetc Algorthm
MSI s under consderaton Items under Scheduled Mantenance or TBM Items under Condton based Mantenance or RCPBM Hard lfe Soft lfe RCP ( t) RCP RCP ( t) < RCP HT= t HT>t ST= t ST > t Replace Replace Replace 1 1 T t 1 1 T < t HT>t 1 HT t 1 ST t 1 ST > t 1 Defer replacement to next scheduled mantenance Use GAs to decde on opportunstc mantenance Defer replacement to next scheduled mantenance Use GAs to decde on opportunstc mantenance Fgure 1. Flow Chart Despton of the procedure to decde on opportunstc mantenance As one can see from the flow chart, the MSI's accessble for opportunstc mantenance have been dvded between tme-based mantenance and condton-based mantenance, dependng on how they are beng mantaned. For all the tems that fall under TBM, we have to further dvde them nto hard lfe and soft lfe, dependng on what tmes are allotted to them. At the tme of opportunstc mantenance, all those tems, whose hard lfe or soft lfe has elapsed, would any way be replaced. For those tems, whose hard lfe or soft lfe s remanng, dependng on how much lfe s remanng, the decson to defer the replacement to the next scheduled mantenance, t 1 or to replace
now has to be taken. If the lfe remanng s more than t 1, then obvously, the replacement s deferred to t 1. If the lfe remanng s less than t 1, then the decson of whether to replace now, or ground the system when t s requred, s taken usng Genetc Algorthms. Smlarly, for tems under condton-based mantenance, especally, n case of relevant condton parameter based mantenance, the decson to replace now or not, s taken dependng on the value of RCP. If the value of RCP for tem, at the tme of opportunstc mantenance, t s equal to the tcal value, then the tem s replaced. If RCP (t) s less than the tcal value, and f the tme to the next examnaton, T 1 s greater than the tme to the next scheduled mantenance, t 1, then the replacement can be deferred to t 1. And f the tme to the next examnaton, T 1 s less than the tme to the next scheduled mantenance, t 1, then Genetc Algorthms s used to decde whether to replace now or to ground the system whenever the requrement arses. 4. Conclusons and Lmtatons The sgnfcance of opportunstc mantenance n felds lke araft engne mantenance, where complex systems are nvolved has been emphaszed. There are numerous factors to be consdered, whle decdng on whether a partcular component should be replaced or repared when an opportunty arses. An effectve condton-based mantenance, called relevant condton parameter based mantenance, has a great potental to be ntegrated nto araft engne mantenance and adopts gracefully for opportunstc mantenance wth ts unque features. Genetc Algorthms s ntroduced, as an optmzaton tool, to compare the cost of premature replacement wth the cost of downtme f grounded for the sole purpose of replacement. Ths paper was an ntal attempt to use Genetc Algorthms for opportunstc mantenance n complex systems. Much work needs to be done n ths area, and numerous other relevant factors needs to be ncorporated for a comprehensve study. A case study nvolvng a practcal applcaton of Genetc Algorthms to real lfe data, would brng about the practcal dffcultes nvolved, but was not carred out here. Ths was just a prelmnary research done to ntegrate the two areas of opportunstc mantenance and
relevant condton parameter based mantenance and to explore the scope of Genetc Algorthms as an optmzaton tool n ths respect. References 1. Dnesh Kumar, U., Crocker, J., Knezevc, J. and El-Haram, M. Relablty, Mantenance and Logstc Support A Lfe Cycle Approach, Kluwer Academc Publshers, 2000, pp. 235-243. 2. El-Haram, M. A. Integrated Approach to Condton-Based Relablty Assessment and Mantenance Plannng. PhD Thess, Unversty of Exeter, 1995. 3. Engne Shop Vst Patters, Engne Mantenance, Araft Economcs, No. 35, January/February 1998. 4. Extractng the maxmum for engne overhaul Araft Economcs, No. 19, pp.35-40, 1995. 5. Goldberg, D.E., Genetc Algorthms n Search, Optmsaton and Machne learnng, Addson-Wesley, Readng, MA, 1989. 6. Haupt, R.L., and Haupt, S.E., Practcal Genetc Algorthms, John Wley & Sons, pp.17-18, 1998. 7. Holland, J.H., Adaptaton n Natural and Artfcal Systems, MIT Press, Cambrdge, MA, 1975. 8. Knezevc, J. Condton Parameter Based Approach to Calculaton of Relablty Characterstcs. Relablty engneerng, Vol. 19, No. 1, pp 29-39, 1987a. 9. Knezevc, J. Requred Relablty level as the optmsaton teron. Mantenance Management Internatonal, Elsever, Vol. 6, No. 4, pp249-256, 1987b. 10. Saranga, H Relevant Condton Parameter based Approach to Relablty and Mantenance. Ph.D Thess, Unversty of Exeter, 2000. 11. Savc, D.A., Walters G.A., and Knezevc, J. Optmal Opportunstc Mantenance Polcy usng Genetc Algorthms, 1: formulaton, Journal of Qualty n Mantenance Engneerng, Vol. 1, No.2, (1995) pp. 34-49. 12. Savc, D.A., Walters G.A., and Knezevc, J Optmal, opportunstc mantenance polcy usng genetc algorthms, 2: analyss, Journal of Qualty n Mantenance Engneerng, Vol. 1, No.3, (1995) pp. 25-34.
13. The Integrated Supportablty Analyss and Cost System (ISACS 2000) Lfe Cycle Cost Methodology.