8t European Worksop On Structural Healt Monitoring (EWSHM 2016), 5-8 July 2016, Spain, Bilbao www.ndt.net/app.ewshm2016 Optimization of maintenance strategies and ROI analysis of CMS troug RAM-LCC analysis. A wind energy sector case study. More info about tis article: ttp://www.ndt.net/?id=19851 Asier ERGUIDO 1, Eduardo CASTELLANO 2, Juan Fº GÓMEZ 3, Adolfo C. MÁRQUEZ 4 1 IK4-Ikerlan Tecnology Researc Centre, Operations & Maintenance Area. Pº J.M. Arizmendiarrieta, 2. 20500 Arrasate-Mondragón. aerguido@kerlan.es 2 IK4-Ikerlan Tecnology Researc Centre, Operations & Maintenance Area. Pº J.M. Arizmendiarrieta, 2. 20500 Arrasate-Mondragón. ecastellano@kerlan.es 3 Dpt Industrial Management I, US ( University of Seville) Scool of Engineering 41092 - Seville,SPAIN juan.gómez@iies.es 4 Dpt Industrial Management I, US ( University of Seville) Scool of Engineering 41092 - Seville (SPAIN adolfo@etsi.us.es Key words: Wind Farm, opportunistic maintenance model, CMS, LCC, agent based simulation. Abstract One of te main sources of costs witin assets life cycle (LCC) are Operation and Maintenance (O&M) costs, wic appear during te wole assets life and in some cases tey can represent as muc as ten times te capital investment costs (i.e. CAPEX). Despite tis fact, in general terms, few sectors consider tem a key factor witin teir business strategies. Hence, te advantage of reducing tese costs and improving economic results is rarely taken. In order to reduce O&M costs, following te correct maintenance strategy plays a very important role, as it determines wen an asset sould be maintained or replaced in order to reduce long term costs. Particularly, in te case of te wind energy sector, te importance of coosing te correct maintenance strategy is a key factor due to te ig O&M costs, principally caused by logistics and wind turbines operational beavior caracteristics. In te proposed researc, a RAM-LCC (Reliability, Availability and Maintainability) analysis as been developed in order to decide wic maintenance strategy sould be followed for reducing long term costs. In fact, te model proposed gives te advantage of identifying adequate opportunistic maintenance strategies considering at te same time condition based maintenance (CBM) strategies. Furtermore it allows evaluating te performance of installed Condition Monitoring Systems (CMS), according to ROI analysis. For tis researc, wind turbines reliability, availability and maintainability costs ave been modeled and calculated based on real field data. 1 INTRODUCTION Te impact of fossil energies on te environment as been fostering te development of renewable energies during te last years. Among all te renewable energy tecnologies, tis evolvement as been especially notorious in wind energy sector, on wic te main World Powers are making great investments. Indeed, during 2014 te annual installation of wind energy crossed te 50 GW for te first time, almost reacing 369.6 GW of installed capacity [1].
Tis growt as come along wit new callenges to deal wit. Among tese callenges one tat acquires ig relevance is te asset management, since costs of Operations and Maintenance (O&M) are considerable. In fact, O&M can represent between 32% or 12-30% of total LCC in offsore or in onsore Wind Farms (WF) respectively [2,3]; mainly due to reliability, availability and logistics costs [4,5]. Terefore, reducing maintenance costs by optimizing maintenance strategies is crucial for reducing te LCC and maximizing te benefits of te WF. Currently corrective maintenance (CM) and time-based minor preventive maintenance (PM) are te most applied maintenance strategies [6]. In addition to tem, Condition Based Maintenance (CBM) strategies ave also gained importance lately [7], taking advantage of te several Condition Monitoring Systems (CMS) installed in te wind turbines (WT) [8]. In CBM strategies te ealt condition of te critical systems is continuously monitored and compared to teir normal beavior, in order to find te adequate moment to preventively maintain tem [9]. Altoug CBM strategies require an important investment on initial pases of te asset management, tey ave been proved to be cost-effective [10]. Te maintenance strategies mentioned above are mainly applied as single-component maintenance policies in te wind energy sector, wereas WFs are multi-component systems, composed by several WTs tat ave dependencies among tem, wic are generally classified as [11]: a) Structural: maintaining a system implies undertaking maintenance activities in oter systems. b) Stocastic: te failure risk of two systems is not independent. c) Economic: performing maintenance activities in different systems simultaneously ave different economic consequences tan doing it individually. Accordingly, wen suc dependences exist, opportunistic maintenance policies ave been demonstrated to be suitable in different sectors [12]. In fact, it as also been demonstrated tat WTs witin a WF present economic dependencies [13] and tat opportunistic maintenance policies are suitable for te sector [6]. Te main objective of tis researc is to take advantage of te application of opportunistic maintenance policies togeter wit CBM strategies in order to minimize te LCC of te WF. At te same time CMS economical performance is evaluated according to ROI analysis. Wit tis purpose an opportunistic maintenance model as been matematically represented. Due to te difficult analytical andling of te stocastic processes tat presents te sector [13], simulation tecniques ave been used to evaluate te different maintenance strategies tat can be adopted, according to bot te LCC of te WF and te ROI of CMS. Te simulation as been fed wit real operational and reliability data provided by a leading company in te sector. 2 STATE OF THE ART 2.1 Opportunistic maintenance in te wind energy sector Opportunistic maintenance policies ave been successfully implemented in several sectors [12], taking advantage of te economic, stocastic or structural dependencies tat present multi-component systems. Particularly, opportunistic maintenance policies try to take advantage of sut downs or failures in systems in order to perform PM in oter running systems, before any failure occurs to tem and minimizing teir risk to fail. Te decision of preventively maintaining te systems wen implementing opportunistic maintenance policies usually relies on te probability distribution of components residual 2
life and teir maintenance costs [14]. Accordingly, several strategies ave been traditionally followed to consider tose two factors and find te optimal moment for undertaking PM: age limit tresolds, combined failure distribution of systems or accumulated operated periods of systems. [12]. Traditionally opportunistic maintenance policies ave not been implemented in wind energy sector [15]. However, more recently, some autors ave demonstrated te suitability of tese maintenance policies in te sector following varied strategies. In Besnard et al [16], an opportunistic maintenance policy based on te weater conditions is presented. In tis researc PM is planned wen failures ave appened in oter WTs and weater conditions are favourable. In Tian et al [15] opportunistic maintenance strategies are combined wit CBM systems. In teir researc, te autors are able to calculate te remaining life of te systems according to te operational data obtained from CMS, identifying te optimal periods to opportunistically perform PM. In Ding et al [17] bot perfect and imperfect maintenance levels are considered, wic restore te maintained systems to different extent [18]. Perfect repairs leave te component as good as new, so tey are usually equated to system replacements. However, imperfect repairs leave te components in a state better tan just before te failure occurrence, but worse tan new. For te decision making process of wic maintenance activity sould be performed in eac system, te autors establis different age tresolds. Depending on wic tresold is surpassed by te systems, tey undergo perfect, imperfect or no maintenance. Te autors extended teir researc in [6], considering different age tresolds for running and failed systems. Sarker and Faiz [19] propose in teir researc a multi-level preventive opportunistic maintenance strategy. In tis researc, te adopted maintenance strategy is based on reliability tresolds. Te autors establis an upper reliability tresold, above of wic a system sould not undergo PM, and a lower limit, below of wic te system sould undergo perfect maintenance. Between tose two tresolds, several tresolds are establised, making groups of imperfect PM wit a restoration factor assigned to eac group. Wen a failure appens, it is analyzed in wic group is eac system and wen needed, tey undergo te according imperfect PM. Atasgar and Abdollazade [20] searc for optimal maintenance strategies following a double objective: minimization of maintenance costs and minimization of loss of production. Furtermore, in order to minimize loss of production, te autors consider te option of installing redundant WTs in a WF. Wit tis double objective, tey establis different imperfect PM levels for te systems and tey consider bot preventive and corrective dispatc of maintenance teams. Finally, in Abdollazade et al. [13] different reliability tresolds are determined for optimizing te decision making about wic imperfect maintenance sould be performed in te critical systems of a WT. Furtermore, a sensitivity analysis is performed for te most relevant factors tat impact on te maintenance strategies: number of maintenance teams, time to repair, lead time of maintenance teams to te WF, etc. 2.2 Life Cycle Cost Analysis (LCCA) Te LCCA costing tecnique aims to optimize te decision making troug te all life pases of an asset, from te design conception and development, to te retirement and disposal [21]. Te most common classification of costs during te wole life of a product is 3
[22,23]: a) Capex (Capital Expenditure). Costs associated to researc and development, design, acquisitions and installation. In te particular case of wind energy sector, tese costs comprise material and labour costs for WTs, civil work and foundation, transportation, grid connection, etc. b) Opex (Operational Expenditure). Costs principally associated to operate and maintain te plant. As stated, te main focus of tis researc is to implement an adequate opportunistic maintenance strategy, wic will ave a direct impact on systems reliability and, consequently, on te minimization of Opex [24]. For te particular case of wind energy sector, te main Opex costs to consider in a LCCA are te following [10]: a) Corrective maintenance costs: travel and access, uman work, tools and materials. b) Preventive maintenance costs: costs associated to sceduled maintenance (travel and access, uman work, tools and materials, etc.) and to CBM (false alarms, updating of systems, online and offline diagnostics, etc.). c) Investment costs in CMS. d) Spare Parts. e) Retirement costs. f) Rate of interest. According to te reviewed LCC analysis for wind energy sector in te literature, it is remarkable te matematical models proposed by Castro Santos [25] for estimating te costs of an offsore WF. In Nilsson and Bertling [25] it is analyzed te aggregated value of including CMS in WF s, analyzing teir impact on maintenance strategies and te according economical consequence in LCC. Tis study was extended in Puglia et al [26] considering also te effect of ageing in systems reliability. Finally, in [10] te added value of CMS is evaluated during te life cycle of te assets. In tis study te autors consider te performance of CMS according to 1) detectability or te probability of detecting a certain failure and 2) efficiency or te time window in wic te failure is detected. 3 OPPORTUNISTIC MAINTENANCE MODEL Te aim of te researc performed is to identify te benefits of combining CBM and opportunistic maintenance on finding cost-effective maintenance strategies for te wole LCC of WFs. At te same time, te ROI of te CMS systems is analyzed, identifying wic are te most suitable CMS to be installed in te WT according to teir caracteristics [10]: detectability and efficiency. Wit tis purpose, an opportunistic maintenance model is proposed, wic will elp establis te maintenance strategies tat sould be followed. Tese maintenance strategies will be defined by te called reliability tresolds, wic will determine weter a system sould undergo a determined imperfect PM during a period or not. Te different imperfect PM activities tat can be performed to a system will ave different restoration effect (q), diminising te virtual age of te system to different extent, according to te Generalized Renewal Process (GRP) metod [13]. 3.1 Problem definition Te WF consists of H WTs composed by I critical systems, tat can undergo K different CM or J different PM. Eac PM will be performed according to te tresolds establised in te maintenance strategy for eac system (minr ij ). Every maintenance activity as a 4
restoration effect associated to te system on wic it is performed, defined by q, wic can be approximated according to te routine followed in te maintenance [6]. Likewise, eac maintenance activity involves some maintenance costs (see Table 1): dispatc of maintenance teams to te WF, working ours of maintenance teams, materials and tools, loss of production, penalty costs, CMS costs, etc. Te performance of te CMS is defined according to teir efficiency and detectability of failures, following te metodology proposed in [10]. Te detectability (γ) represents te ability of a CMS to detect a failure and te efficiency (η) represents te spot on te P-F curve were te failure is detected. Terefore te performance of a CMS is modeled by a single point (γ and η), establised by expert knowledge (see Figure 1) (for furter information te reader is addressed to [10]). In te proposed approac, for te detected percentage of failures (γ) te PM is compulsorily performed at point (η), and if tey are not detected, te systems are maintained according to te opportunistic maintenance establised. Figure 1(a) Representation of parameter η on te P-F curve and (b) relation between efficiency η and detectability γ [10]. Witout loss of generality, some assumptions ave been made for te model development: 1) degradation process of te systems are independent from eac oter, 2) increasing failure rate (IFR) is considered, 3) installed WTs and te systems are te same model, wit identical reliability distributions, 4) PM is assumed to be less resource consuming tan CM due to a reduction of damage because of failure, minor time to repair, better planning of interventions, etc. 5) maintenance teams are composed by two workers, 6) WF operators make decisions in discrete time and frequently, 7) k=1 are minor failures tat do not require WF visit nor materials for reparation and 8) no stock or uman resources capacity constraints are considered. 5
3.2 Model formulation Table 1: Nomenclature used on te problem modelling. In tis section it is represented te matematical formulation of te proposed opportunistic maintenance model 1, wic afterwards, is strictly represented in te simulation. As te main objective of te proposed opportunistic maintenance model is to minimize LCC associated to maintenance, on te objective function (Eq.1) costs of CM and PM are considered, attending to working ours, materials and opportunity costs, as well as dispatc and CMS costs. Due to te long term caracteristic of LCCA, costs sould be accordingly updated. 1 2 + + ( + ) + 2 + + +, (1 + ) +1 +, + + + Eac time a failure appens, a maintenance team is correctively dispatced ( = 1) to te WF in order to perform te required CM (Eq.2). (1) = 1 1 1 0 (2) In order to preventively dispatc a maintenance team to te WF (w t = 1), several conditions ave to be met: te minimum reliability required for te WF is not reaced (Eq.3), tere is not anoter dispatc on te same period (Eq.6) and te total number of preventive maintenance dispatces on te past P periods do not exceed a certain number (Eq.4). Furtermore, a maintenance team can also be preventively dispatced due to an alarm of CMS (u t = 1) (Eq. 5). Te total number of maintenance dispatces is set by Eq.7. 1 Notation and binary variables used on te formulation are specified on Table 1 6
Table 2: Binary variables used on te problem modelling. ( ),, (3) = = 1 = + < 1 1 (5) 0 (4) + + 1 (6) = ( + + ) (7) In te presented opportunistic maintenance model, if a maintenance team as been dispatced to te WF PM is allowed to be performed. However, imperfect PM activities are only performed in a system wen its reliability is between te tresolds establised for a particular opportunistic strategy. = 1 (+1) 1 ( + + ) = 1 0,,, (8) Finally, after bot PM or CM, te virtual age of te system is updated, according to te restoration factor of te preventive activity performed (Eq.9) = ( 1 + 1) (1 ),, (9) 7
3.3 Simulation Process Te simulation as been developed in an agent based simulation environment, due to its appropriate adaptation to engineering problems tat involve multi-agent systems [27]. Te simulation process is composed by 6 different steps, at wic different decisions are made or different actions are performed, according to te matematical model explained in te previous section: Step 1. Te simulation is initialized; establising wic maintenance strategy sall be followed and accordingly setting all te parameters needed for te simulation process (see Table 1). Step 2. Required CM is performed in failed systems and virtual age and reliability of te systems is updated. If no failure appens, te decision of preventively dispatcing a maintenance team is made according to te maintenance strategy followed. Step 3. PM decision is made, according to te maintenance strategy followed. Virtual age and reliability of maintained systems is updated. Step 4. Total costs are updated. Wile te iteration period does not correspond to maximum iteration period, steps 2, 3 and 4 are repeated. Step 5. Total LCC are calculated and te adopted strategy is evaluated. 4 COMPUTATIONAL RESULTS In tis section some of te computational results obtained troug te simulation process. Te simulation as been fed wit real operational and maintenance data, so a representative environment of a real WF as been acieved. However, due to confidentiality issues, no reliability data can be provided. Te virtual WF under study is composed by 30 WT. For eac WT te four most critical systems identified, bot according to teir failure frequency and severity are te blades, gearbox, pitc system and yaw system. In order to contrast te suitability of te proposed opportunistic maintenance model and te CBM, 3 different scenarios ave been developed, in wic 3 different maintenance strategies are performed: a) Scenario 1. Only CM and minor PM is applied, as currently done in te industry [6]. b) Scenario 2. CM and imperfect PM is performed, according to te opportunistic maintenance model proposed in section 3 but witout using CMS or implementing CBM strategies. c) Scenario 3. CM and imperfect PM is performed, according to te opportunistic maintenance model proposed in section 3. In tis scenario CMS are used for Gearbox and Blades, two of te systems on wic te application of CBM strategies is more extended. For te Gearbox oil analysis and vibration measurements [10] are usually performed, wereas for Blades damage vibration of te main saft is analyzed [28] 2. In eac scenario, for te maintenance costs and te CMS costs evaluation [29, 10, 30] ave been respectively addressed (see Table 3). Furtermore, for te comparison made between te maintenance policies, in eac case te ig quality maintenance strategies ave been searced. Te results ave been obtained identifying adequeate values of te reliability tresolds presented in section 3 (, ). Tese results ave been found troug te OptQuest Engine, a commercial optimization software developed by Fred Glover in OptTek Systems Inc. (Opttec Systems Inc. 2000), wic as been proved to be robust and efficient on 2 In bot cases a PF interval of 30 days as been considered, being conservative, as most of te failures of te monitored systems can be detected wit more tan 2 monts time interval prior to teir occurrence [31]. 8
finding ig quality solutions. System Investement Expenditure ( ) Operation Expenditure ( ) Detectability (%) Efficiency (%) Gearbox 22000 2100 0.7 0.95 Blades 35152 2100 0.9 0.95 Table 3: CMS parameters used as Simulation Input Data. In Figure 2 te performance of te different strategies is analyzed according to LCC evolution for te estimated life cycle of te WTs (20 years). In Figure 2 it is sown tat wereas initially a major investment is needed in order to be able to implement condition based maintenance strategies (strategy 3), tis investment is likely to be recovered in te first alf of te life cycle of te WTs. Figure 2 LCC evolution following te proposed strategies Figure 3 LCC associated to gearbox system before and after implementing CMS On te oter and, Figure 3 focuses on te impact of te CMS witin te gearbox O&M, according to material, CMS and manpower work costs. If ROI is calculated for te CMS installed in te gearbox and te blades system (see Eq.10) according to te revenue obtained 9
by following strategy 3 instead of 2, ROI of te gearbox is set to 161% and blades to 111%. Altoug tecnical performance of te CMS implemented in te blades is better tan te one in te gearbox, CMS of te gearbox as a better economical performance, mainly associated to lower cost of investment of CMS and te iger failure rate and severity of gearbox failures compared to blades failures. = 2,3 100 (10) 5 CONCLUSIONS On tis researc it is proposed a matematical model tat allows estimating LCC and facilitates te decision making process tat determines wic maintenance strategies sould be applied. Particularly, te proposed model takes advantage of te economic dependencies tat ave te WTs witin a WF by applying opportunistic maintenance policies. Furtermore, te model also allows estimating te economical performance of te CMS implemented in te systems of te WT wen opportunistic maintenance policies are followed. After obtaining te computational results for four of te most critical systems of te WT, based on real operational and reliability data of WFs, it is sown tat opportunistic maintenance strategies are adequate for wind energy sector. Moreover, it is also proven tat CMS systems implemented bot in gearbox and blades are economically effective and tat tey can be a successful complement of te opportunistic maintenance strategies. Furter researc lines will focus on applying capacity constraints to te model, suc as uman resources capacity or material stock constraints, in order to represent more realistic scenarios. At te same time, furter researc will be done in te utilisation of CMS togeter wit opportunistic maintenance policies, in order to be able to facilitate decision making about te correct CMS to be installed in te several systems witin te WF. REFERENCES [1] Anon., Global status of wind power in 2014, Global wind energy Council, 2015. [Online]. Available: ttp://www.gwec.net/wp-content/uploads/2015/03/gwec_global_wind_2014_report_lr [2] J. Kaldellis and M. Kapsali, _Sifting towards offsore wind energy-recent activity and future development, Energy Policy, 53, pp. 136-148, 2013. [3] E. Byon, Wind turbine operations and maintenance: a tractable approximation of dynamic decision making, IIE Transactions, 45, no. 11, 1188-1201, 2013. [4] P. J. Tavner, J. Xiang and F. Spinato, Reliability analysis for wind turbines. Wind energy, 10, 1-18, 2007. [5] Safiee, M, Maintenance logistics organization for offsore wind energy: Current progress and future perspectives, Renewable Energy, 77, 182-193, 2015. [6] F. Ding and Z. Tian, Opportunistic maintenance for wind farms considering multi-level imperfect maintenance tresolds, Renewable Energy, 45, pp. 175-182, 2012. [7] Y. Amirat, M. Benbouzid, E. Al-Amar, B. Bensaker, and S. Turri, A brief status on condition monitoring and fault diagnosis in wind energy conversion systems, Renewable and Sustainable Energy Reviews, 13, 2629-2636, 2009. [8] W. Yang, P. J. Tavner, C. J. Crabtree, Y. Feng, and Y. Qiu, Wind turbine condition monitoring: tecnical and commercial callenges, Wind Energy, 17, 673-693, 2014. [9] M. C. Garcia, M. A. Sanz-Bobi, and J. del Pico, Simap: Intelligent system for predictive maintenance, Computers in Industry, 57, 552-568, 2006. [10] A. V. Horenbeek, J. V. Ostaeyen, J. R. Du_ou, and L. Pintelon, Quantifying te added value of an imperfectly performing condition monitoring system_application to a wind turbine gearbox, Reliability Engineering & System Safety, 111, 45-57, 2013. 10
[11] R. Nicolai and R. Dekker, _A review of multi-component maintenance models, Proceedings of European Safety and Reliability Conference, 2007, 289-296. [12] H. Wang, A survey of maintenance policies of deteriorating systems, European Journal of Operational Researc, 139, 469-489, 2002. [13] H. Abdollazade, K. Atasgar, and M. Abbasi, Multiobjective opportunistic maintenance optimization of a wind farm considering limited number of maintenance groups, Renewable Energy, 88, 247-261, 2016. [14] D.A. Savic, G.A. Walters and J. Knezevic, Optimal, opportunistic maintenance policy using genetic algoritms, 2: analysis, Journal of Quality in Maintenance Engineering, 1, 25-34, 1995. [15] Z. Tian, T. Jin, B. Wu, and F. Ding, Condition based maintenance optimization for wind power generation systems under continuous monitoring, Renewable Energy, 36, 1502-1509, 2011. [16] F. Besnard, M. Patrikssont, A. B. Strombergt, A. Wojciecowskit, and L. Bertling, An optimization framework for opportunistic maintenance of offsore wind power system, in IEEE Bucarest PowerTec. IEEE Institute of Electrical & Electronics Engineers, 2009. [17] H. Pam and H. Wang, Imperfect maintenance, European Journal of Operational Researc, 94, 425-438, 1996. [18] B. R. Sarker and T. I. Faiz, Minimizing maintenance cost for offsore wind turbines following multi-level opportunistic preventive strategy, Renewable Energy, 85, 104-113, 2016. [19] K. Atasgar and H. Abdollazade, Reliability optimization of wind farms considering redundancy and opportunistic maintenance strategy, Energy Conversion and Management, 112, 445-458, 2016. [20] Kirk, S. J., & DellÍsola, A. Life cycle costing for design professionals. McGraw-Hill. ISBN 0070348049 (1995). [21] I. Blanco, Te economics of wind energy, Renewable and Sustainable Energy Reviews, 1372-1382, 2009. [22] A. C.Levitt, W. Kempton, A. P. Smit, W.Musial, and J.Firestone, Pricing offsore wind power. Energy Policy, 39, 6408-6421, 2011. [23] D. G. Woodward, Life cycle costing-teory, information acquisition and application, Internattonal journal of project management, 15, 335-344, 1997. [24] L. Castro Santos, Metodología para la evaluación económica de parques eólicos offsore flotantes a través del análisis del coste de las fases de su ciclo de vida, PD tesis (2013). [25] J. Nilsson and L. Bertling, Maintenance management of wind power systems using condition monitoring systems- Life cycle cost analysis for two case studies, IEEE Transactions on Energy Conversion, 22, 223-229, 2007. [26] G. Puglia, P. Bangalore, L. Bertling and L Tjenberg, Cost efficient maintenance strategies for wind power systems using LCC. International conference on probabilistic metods applied to power systems, 1-6, 2014. [27] M. Niazi and A. Hussain, Agent-based computing from multi-agent systems to agent-based models: a visual survey, Scientometrics, 89, 479-499, 2011. [28] A. Takeuci, T. Haseba, H. Ikeda, Application of Condition Monitoring System for Wind Turbines. NTN Tecnical review, 80, 15-18, 2012. [29] M. Martin-Tretton, M. Rea, M. Drunsic, and M. Keim, Data Collection for Current U.S. Wind Energy Projects: Component Costs, Financing, Operations, and Maintenance, DNV Renewables, 2011 [30] F.P. García and B. Lev. Advanced Business Analytics. Capter 2 Economic Viability Analytics for Wind Energy Maintenance Management Springer ISBN: 9783319114156 (2015). [31] D. Coronado and K. Fiscer. Condition monitoring of wind turbines: state of te art, user experience and recommendations, Project number: VGB-Nr.383, 2015. 11