“The application of spare parts pooling within the oﬀshore wind industry”

(1)

“The application of spare parts pooling within

the offshore wind industry”

Mark Eggo Alfred Dijkhuizen (S2774925/B19057371)

A master thesis presented for the degree of MSc. Dual Degree Operations & Supply Chain Management

Faculty of Economics and Business University of Groningen Newcastle University Business School

Supervisor: Dr. O.A. Kilic Co-assessor: Dr. Hu Country: The Netherlands

(2)

Abstract

Globally, the fossil based energy sector is decreasing as a consequence of the recognised impact of pollution and depletion of reserves. Meanwhile, the renewable energy industry is on the rise, of which the offshore wind sector is an important contributor due to its growth potential. However, this rapid growth is accompanied with emerging challenges such as rising operations and maintenance costs, which must be lowered before the offshore wind industry is able to reach its full potential. To overcome the difficulties faced by the industry, an opportunity for collaboration of companies within that industry arises. Consequently, this thesis investigates the possibilities of pooling major spare parts of offshore wind turbines. Through the use of simulation this thesis reveals insights on the impact that pooling can have on cost savings and availability under different criteria. It is concluded that the greatest saving potential of applying pooling as a way to manage inventory lies in the minimization of costs rather than aiming to increase availability.

Keywords: Spare parts, pooling, offshore wind, wind turbine, energy, maintenance, operations and mainte-nance, offshore wind farm, simulation

Acknowledgements

(3)

List of Figures

2.1 Different pooling strategies. Figure obtained from Kilpi et al. (2009, p. 362) . . . 9

3.1 Overview of the components of the quantitative model . . . 11

4.1 Wind turbine spare parts. Figure obtained from Tian et al. (2011, p. 1506). . . 13

4.2 Spread of failure behaviour with a CV of 4.7 . . . 17

4.3 Schematic overview of the the spare parts delivery process - Stock replenishment . . . 17

4.4 Schematic overview of the the spare parts delivery process - Inbound transportation . . . 18

4.5 Power curve and daily electricity output . . . 18

4.6 Offshore wind farm operations and maintenance process diagram (System Dynamics) . . . 19

4.7 Visual representation of servicing from own warehouse . . . 21

4.8 Visual representation of servicing from centralized warehouse . . . 21

5.1 Base-case scenario, costs and responsiveness . . . 22

5.2 The responsiveness and costs with the inbound transportation lead time as subject of analysis . . 25

5.3 MTTR and TBA as a function of different stock replenishment lead times . . . 26

5.4 Changing costs as a function of different Stock Replenishment Lead Times . . . 26

5.5 Comparison of costs scenario 1 against scenario 7 . . . 27

5.6 Sensitivity analysis for main model parameters of the base-case scenario (No Pooling policy (NP-policy)) . . . 28

5.7 Sensitivity analysis for main model parameters of the base-case scenario (Pooling policy (P-policy)) 29

List of Tables

4.1 Failure rates and probabilities per spare part. Data obtained from Nguyen et al. (2018, p. 91) . 13 4.2 Labels, corresponding wind speeds and failure rates Data obtained from Reder et al. (2018, p.561) 15 4.3 Model parameters including values for the sensitivity analysis . . . 16

5.1 Results of the base-case scenario . . . 22

5.2 Savings per spare part when pooled individually for the base-case scenario . . . 23

5.3 Model scenarios . . . 23

5.4 Results of scenario 1 - 3 with inbound transportation lead time as variable . . . 24

5.5 Results of scenario- 1 and 4 - 6 with the SRLT as subject of analysis . . . 25

5.6 Results of scenario- 1 and 7 with wind speed as the subject of analysis . . . 27

5.7 Results of the sensitivity analysis for the variation in failure rate . . . 28

5.8 Results of the sensitivity analysis for the holding cost per day . . . 28

5.9 Results of the sensitivity analysis for the service level . . . 29

5.10 Results of the sensitivity analysis for the holding cost per day . . . 30

(5)

Abbreviations

CM Corrective Maintenance. COD Costs Of Downtime. COE Cost Of Energy. CV Coefficient of Variaton. GBP British Pounds.

ITLT Inbound Transportation Lead Time. KPI’s Key Performance Indicators. MTTR Mean Time To Repair. Mwh Megawatt hour.

NP-policy No Pooling policy. O&M Operations and Maintenance. OEM Original Equipment Manufacturer. OWF Offshore Windfarm.

OWF’s Offshore Wind Farms. OWI Offshore Wind Industry. OWT Offshore Wind Turbine. OWT’s Offshore Wind Turbines. P-policy Pooling policy.

PM Preventive Maintenance.

SRLT Stock Replenishment Lead Time. TBA Time-based Availability.

(6)

1 Introduction

As a consequence of the depletion of fossil fuels and their acknowledged impact on our climate combined with the rapid growth of energy demand for future years, the need for renewable energy is rising, while there is observed a decrease in the non-renewable sector (Wangler, 2013; REN21, 2015; Manz, 2017). This transition in the energy sector is reinforced by government initiatives such as e.g. the Paris climate agreement which leads to an even greater pressure to satisfy energy demands in a climate friendly manner (UNFCCC, 2015).

The Offshore Wind Industry (OWI) is considered a viable option to meet these rising demands as offshore areas are generally windier than onshore regions. This enables Offshore Wind Turbines (OWT’s) to generate more electricity. Additionally, offshore installations allow for larger Wind Farms (WF’s), easier transport and when installed far enough from shore, the visual impact can be nearly eliminated (REN21, 2015; Shafiee, 2015b). This makes the OWI an attractive opportunity for investors. An investment of €19 billion in the European wind energy sector in 2019 resulted in a 25% growth of the OWI that year (WindEurope, 2019). However, the rapid growth coupled with the OWI-potential are offset by several disadvantages such as extra required capital investments, limited access for Operations and Maintenance (O&M) and long- and expensive down times (Shafiee, 2015a). According to Carroll et al. (2015), an Offshore Wind Turbine (OWT) faces approximately eight failures per year of which one concerns a major spare part failure resulting in an operational availability somewhere around 70%, which is 30% less than for onshore Wind Turbines (WT’s) (Bilgili et al., 2011; Shafiee, 2015a). Coincidentally, this also results in 30% of the total costs off OWT’s attributable to O&M, making it a critical area to improve profitability (Blanco, 2009; Rodrigues et al., 2015). To reduce the consequences of equipment downtime and thus the high O&M-costs, spare parts are held. However, it has been observed that the cost of spare parts takes a large share of the costs of machinery (Hu et al., 2018). As illustrated by Gallagher et al.(2005), machinery with a lifetime of 30 years consumes spare parts amounting to as much as 2,5% of the purchase price each year. However, the unavailability of spare parts, for maintenance, will result in considerable financial loss. Therefore, spare parts management is an important factor in reaching the desired availability at a minimum cost (Hu et al., 2018).

Due to its importance, spare parts management is widely addressed within literature (Cheng and Prabhu, 2012; Dahane et al., 2017) where research has shown, that decentralized spare parts control is sub-optimal in terms of long-term average costs (Paterson et al., 2011). This has also been noticed in the study of J¨ager-Roschko et al.(2019). In their study, the authors developed an agent-based model for the O&M of Offshore Wind Farms (OWF’s) to assess if storing OWT spare parts in a centralized warehouse can be advantageous. They found that pooling major spare parts is beneficial regarding both availability and costs, in particular through maximizing the availability rather than reducing costs. Yet, the authors suggested to further investigate whether the possibilities of using pooling strategies for OWT spare parts as a way to manage inventory could be beneficial under different criteria. In particular, a change in both the failure rate and storage costs might leads to interesting results. This research describes the results of a study into these possibilities. Spare parts pooling can be defined as: “an inter-‘company’ collaboration where the cooperating companies share their inventories, it is an effective policy to improve a company’s logistical performance without requiring any additional cost” (Wong et al., 2005, p. 207). Pooling can result in significantly reduced inventory costs if spares per party involved do not differ considerably (Schlicher et al., 2017). This is usually the case if parts have not been produced more than two years apart (Lindqvist and Lundin, 2010). Arising economies of scale as a result of these pooling techniques, can lead to a reduction in costs and to a reduced demand variability compared to acting alone (Kilpi et al., 2009). This improves the firms’ ability to efficiently match supply and demand, allowing to reduce safety stock increasing profit as a result (Swinney, 2012).

Although this sounds promising, as a consequence of the added financial- and operational complexity, pooling is already difficult to implement in general let alone in an offshore setting. (F. Karsten and Basten, 2014). Due to the remote location of OWF’s, far distances need to be bridged resulting in high and costly complexities compared to onshore WF’s (Irawan et al., 2017). Additionally, these trips are regularly affected by a rough sea climate which causes additional delays resulting in increased OWT down times (Astariz and Iglesias, 2015). These remarkable conditions pose difficulties for the implementation of spare parts pooling.

While recognizing the difficulties that need to be overcome, this thesis aims to evaluate the possibilities of pooling spare parts of OWT’s as a way to reduce the high O&M-costs in the OWI, which built the ground for this thesis. This results in the following research question:

(7)

Following the insights of the gap identified by J¨ager-Roschko et al. (2019), this research contributes by inves-tigating pooling spare parts within the OWI under different criteria. As suggested by the authors, our work differentiates inter alia on turbine failure rate and holding costs. By assessing the effect of pooling on the per-formance of an Offshore Windfarm (OWF) under a variety of criteria, this thesis gives valuable insights in the trade-off between service and costs to both researchers and practitioners. Grasping the impact of implement-ing poolimplement-ing spare parts aids for more informed decision makimplement-ing regardimplement-ing the handlimplement-ing of spare parts within the OWI. This is where the practical relevance of this thesis stems from. To answer the research question, a quantitative model is developed. This model simulates the O&M-environment of an OWF and investigates the impact of pooling on the O&M-performance of an OWF under multiple configurations. Tis thesis revealed that utilizing the pooling of spare parts as a way to manage inventory should focus on cost minimization rather than increasing availability.

The remainder of this thesis is structured as follows. First, in Chapter 2 concerning the theoretical background, the existing literature on O&M of an OWF as well as on pooling spare parts will be discussed. Thereafter, in Chapter 3 the methodological approach of this research is explained and the model components are briefly introduced. Chapter 4 outlines the developed quantitative model, explains its components and briefly introduces the scenarios. In Chapter 5 the scenarios are specified and the numerical results of the model are presented. Then, the results, their implications, limitations and suggestions for future research are discussed and compared with existing literature in Chapter 6. Finally, in Chapter 7 the concluding remarks are provided.

2 Theoretical background

In the first section the O&M of an OWF will be discussed, then there will be elaborated on the different maintenance strategies where after a section is dedicated to the inventory management of OWT spare parts. This chapter closes with discussing different forms of pooling and their properties, coincidentally with how the term is interpreted for the purposes of this thesis.

2.1 Operations and Maintenance of offshore wind farms

To understand how the OWI can benefit from pooling spare parts to enhance their O&M, it is evident to understand what the O&M of an OWF entails. The O&M-phase of an OWF will last for approximately 20 years which makes it the longest phase in the life cycle of an OWT (Snyder and Kaiser, 2009). To ensure that throughout this phase the OWT is generating as much energy as possible, management and maintenance is required referred to as O&M (Shafiee, 2015b). The costs linked to these activities can be divided in two categories, the operational expenses and the expenses associated with maintenance. The operational expenses refer to the cost of daily operations such as rent- or leasing payments, monitoring turbine operation and scheduling site personnel. Maintenance costs on the other hand, are the costs of conducting maintenance, that is, aiming to maximize the OWTavailability while keeping the costs associated with failures as low as possible. These costs can in turn be broken into two types, indirect- and direct maintenance cost. Indirect costs are the costs linked to activities which are a prerequisite to conduct the actual repair, these costs may be either fixed, such as port fees and weather forecasting or variables such as onshore maintenance support and the number of vessels required for conducting maintenance. Direct costs are associated with conducting repairs such as performing maintenance, transportation and the costs of spare parts (Shafiee et al., 2016). The latter is as explained the focus of this thesis.

(8)

When a P-policy is applied, the overall spare part demand will increase meaning that more spares should be kept in inventory. This results in higher overall holding costs but now, these costs can be divided among members in the pool. Additionally, when spare parts are pooled, the uncertainty of demand is pooled as well due to economies of scale (Kilpi et al., 2009). This results in a decrease in uncertainty and it becomes therefore possible to reduce the overall safety stock, which saves on costs (Swinney, 2012). So, the same availability on the technical system, i.e. OWT, can be obtained against a lower cost (Muhaxheri, 2010). In that regard, the costs associated with holding spares become lower compared to deploying a NP-policy.

Due to its relevance, academic literature has extensively studied optimization of the O&M-phase of OWF’s. The early works were mainly explanatory trying to understand the OWF-environment such as the article of Van Bussel et al. (2001) and Rademakers et al. (2003). Initially, OWF’s mostly served as a proof of concept, located in shallow waters and close to shore leading to low investment costs (Rodrigues et al., 2015). When the concept proofed to be working, commercial projects emerged and the area of study shifted. Hofmann (2011), presented a review of decision support models for O&M-planning. He found that new cost-efficient technical solutions to optimize maintenance was one of the main research areas in the offshore wind sector by that time. As an attempt to cover the intermediate development, Seyr and Muskulus (2019) also conducted a review about decision support models and found a considerable amount of decision models dedicated to the O&M-phase in the offshore sector. These models can be used to assess the influence of different factors on maintenance activities and several of them also included spare parts as a factor of influence. More recently, studies have become more data-driven and focused on the automation of processes (Stock-Williams and Swamy, 2019; Pandit et al., 2020), determining the ideal configurations of maintenance operations (Zhou and Yin, 2019; Shafiee and Sørensen, 2019) and to some extent, sharing resources to minimize costs (Broek et al., 2019; J¨ager-Roschko et al., 2019) which can be seen as a more exploratory approach in this field of research. In accordance with the latter this thesis also follows a more exploratory route in the form of sharing spare parts to improve the O&M-phase by reducing holding costs. However, to do this in an economical way, optimal stock levels should be set. As the required stock level is often dependent upon the applied maintenance policy (Ilgin and Tunali, 2007), the next section elaborates.

2.2 Maintenance policies

Now that the activities and costs of the O&M of OWF’s are explained, the overarching cost elements of the O&M-phase should be identified. This understanding is a prerequisite to reduce O&M-costs for wind energy projects (Walford, 2006).

Following Walford (2006), these costs can be broken into three broad categories: Preventive Maintenance (PM), Corrective Maintenance (CM) and operations. PM and CM concern the actual repairs to the OWT’s while operations comprises activities associated with day-to-day activities required to perform maintenance. These elements are relevant for the purposes of this thesis since they can affect- and can be affected by executing a P-policy. Additionally, they affect the overarching goal of cost reduction because where 30% of the total COE are attributable to O&M, maintenance alone represents 10% to 15% of the total COE (Dahane et al., 2017). Taking this into account, the maintenance of OWT’s represents an active research area on several issues. PM is used to execute maintenance just before failure to prevent or delay a breakdown and thus increase availability (Nilsson and Bertling, 2007). The advantage of PM is that it reduces the change on complete malfunctioning resulting in lower downtime and maintenance costs compared to CM. Besides, down times are usually shorter and damage to secondary parts can thus be reduced. However, PM also has its limitations. When a failure occurs it is impossible to determine which exact part requires attention, this could result in not having spare parts or proper tools directly available. Moreover, due to the higher frequency of trips of a preventive nature, this maintenance strategy is costly and labor intensive (Jonker, 2017).

(9)

Where PM is a proactive approach as it prevents failures, CM is a reactive approach. When a system failure occurs maintenance is carried out to restore the system performance (Nilsson and Bertling, 2007). Advantages are that maintenance trips are kept to a minimum and implementing and planning CM is easy. Yet, CM also has its drawbacks. It is associated with longer down times and potential higher maintenance costs. Another disadvantage is the prerequisite for having crew- and spare parts always available just as the ability for transport. Furthermore, there is an increased risk of possible damage to secondary parts and CM is always linked to lost production time as the OWT does not generate any electricity during scheduling, anticipating weather windows and repairing. The combination of these make this maintenance strategy expensive (M´arquez et al., 2012; Mobley, 2002). Finally, a major contributor to these costs is the required availability of spare parts as large and in particular costly inventories are required to minimize downtime. Especially during the winter season the waiting times as a consequence of weather conditions under which the OWF becomes inaccessible for maintenance, can result in high inventory costs (Van Bussel and Bierbooms, 2003). In an effort to mitigate this, a P-policy could be of interest as a way to manage inventory.

As indicated earlier, operations are required to perform maintenance and therefore important for this thesis. The most note-worthy operational activities are crew management, monitoring turbines and spare parts management. These are all stand-alone activities which can be performed separately, but are of influence of each other (Walford, 2006). However, the exact organisation and content of these activities largely depends upon the chosen maintenance strategy, which on its turn impacts the used methods for maintenance scheduling (Besnard et al., 2012). Maintenance scheduling can be defined as planning the maximum of maintenance activities with the available resources (Kumar, 2000). As discussed above, only CM will be performed which simplifies the scheduling of maintenance as a repair will be initiated when a OWT breakdown occurs. However, as already mentioned, performing maintenance is restricted by the weather as harsh weather conditions result in a limited accessibility of the OWF (Besnard et al., 2012). To limit the impact of uncontrollable weather delays on performing maintenance, it is evident that the controllable operational activities such as e.g. the inventory management of spare parts are organized optimally. Subsequently, the next section elaborates on managing spare parts.

2.3 Inventory management of spare parts

From out the spare parts management domain three main themes can be identified to improve the O&M of OWF’s, processing, inventory management and demand forecasting (Schuh et al., 2015). Data pre-processing aims to reduce complexity by grouping spare parts with similar properties based on their demand or nature (Boylan and Syntetos, 2010). Inventory management is defined as: “the continuing process of planning, organizing and controlling inventory that aims at minimizing the investment in inventory while balancing supply and demand” (Singh and Verma, 2018, p.3868). Lastly, demand forecasting uses prediction techniques to optimally set inventory levels (Schuh et al., 2015). In this thesis, we focus on the inventory management of spare parts to improve the O&M of OWF’s.

Spare parts are used to replace worn-out and defective units, which are unable to fulfill their purpose. Forecasting and handling the amount of spare parts required in the future is denoted as spare parts management (Schuh et al., 2015). To attain economic operation, a high service level and low inventory costs are desired (Schuh et al., 2015). The service level reflects the amount of demand which can be directly satisfied from stock (Chen and Krass, 2001) and is viewed as a constraint rather than an objective for the purposes of our research. Specifically, the amount of stock in inventory is a function of the desired service level, demand uncertainty and Stock Replenishment Lead Time (SRLT) (Chopra et al., 2004). SRLT entails the time associated with getting the spare parts from the Original Equipment Manufacturer (OEM) to the warehouse, a short stock replenishment time allows reducing safety stock and thus saves on costs (Chandra and Grabis, 2008; Jacobs and Chase, 2012). To obtain a high service level, generally high stock levels need to be maintained resulting in high spare parts availability. However, high inventory levels are difficult to reconcile with low inventory costs resulting in a natural trade-off between inventory cost and service (Schuh et al., 2015; Schlicher et al., 2016). It is therefore evident to find an optimal middle ground in setting the most economic inventory level so that costs can be minimized, which is the main objective of inventory management (Singh and Verma, 2018).

(10)

Another work conducted by Boudhar et al. (2017) studied the possibility to use new or re-manufactured spare parts. In doing so, the authors proposed a heuristic to determine the spare part quality needed in the next replacement, depending on the level of degradation of operating spare parts. In a more recent work, Zhang et al. (2019) suggested an opportunistic maintenance- and (S,s) spare parts management policy for OWT’s considering stochastic weather conditions. The authors demonstrate that by applying these strategies, a significant cost reduction of more than 10% can be realized. Finally, as can we can recall from the introduction, J¨ager-Roschko et al. (2019) analyzed if using a centralized warehouse to store spare parts (i.e., pooling) of OWF’s could be beneficial. In their study they found that using a centralized warehouse yields larger profits for major spare parts as a consequence of a higher availability and a decrease in inventory costs.

However, within the vast majority of research one recurrent theme can be recognized: ‘how to address the uncertainty in demand to optimally manage the trade-off between service and costs?’. In this thesis, we address this trade-off from an inventory management perspective by applying pooling to control inventory. As in line with the suggestion made by J¨ager-Roschko et al. (2019), we assess the influence of pooling under different criteria. As mentioned, through pooling spare parts the uncertainty in demand decreases as a consequence of economies of scale enabling to maintain lower safety stock which saves costs (Kilpi et al., 2009; Swinney, 2012). As a result of the combination of these factors, it is interesting to further investigate the possibility of pooling spare parts and assess how the criteria as used in this thesis affect the O&M-performance of an OWF. The following section outlines different pooling techniques and identifies how pooling is referred to as for the purposes of this thesis. Additionally, the differences between the study of J¨ager-Roschko et al. (2019) and our work are emphasized.

2.4 Pooling policies

According to Muhaxheri (2010), there can be distinguished four different strategies to pool spare parts. Solo strategy, Ad hoc cooperation, cooperative pooling and commercial pooling. Within operations literature, these types of sharing can be identified as a type of horizontal collaboration (Broek et al., 2019).

The most traditional strategy is to perform the availability service in-house so that the company provides spare parts service within the company scope, this is called solo strategy.

Ad hoc cooperation entails a loose form of cooperation by providing a loan unit against a fee when another company needs it. If the relationship and trust between parties gets stronger, there is an opportunity of utilizing economies of scale without full pooling. In this way, both parties can lower their base stocks and rely on loans from the other party without any contractual difficulties.

One step further is cooperative pooling. Here, two or more industry operators with commonality in spare parts can formally agree at a set of rules to share their spare inventories. All members involved are equal and they share their spare parts between each other according to mutual agreement embedded in general clauses. When a member uses a spare unit out of the pool, it becomes responsible for delivering the failed unit back to the pool after it has been repaired. The used scope can vary from ad hoc cooperation with loose loan arrangements to relatively tight cooperation.

(11)

contrary, origin-and-destination carriers connect a limited number of destinations to each other by offering non-stop services between them; i.e. there are numerous bases, each of them with relatively low operative volume.

From the viewpoint of availability service, the most important factors in the network complexity are the number of aircraft bases and the distribution of the operational volume between them. The number of destinations in the network is less signiﬁcant, as the majority of component support needs materialize in the bases.

Although the aircraft manufacturers are consolidating and there are fewer competitors than before, the variety of aircraft types available has been constantly increasing and also utilized by the airlines (Kilpi, 2007, p. 83). Because most of the components are suitable for one ﬂeet or family of aircraft, every aircraft family practically requires its own supply of spare components.

In addition to the economies of scale and the required official authority certificate, there are few barriers of entry into the component availability service market, whose main function is to bring together business volume. Since business volume means the demand for spare units, the intuitive way of accumulating it is to pool inventories together and use the pool to satisfy the demand from several aircraft fleets. There are two types of these pools: commercial pools and cooperative pools.

In commercial pooling there is one service provider and several customers that buy availability services from the service provider. In cooperative pooling there are several equal members that share their spare units between each other according to a mutual agreement, whose scope can vary from an ad hoc cooperation with loose loan arrangements to relatively tight cooperation. Thus, in a cooperative pool, every member has connections to all other members in that pool, and in a commercial pool every member has a connection only to the service provider and the service provider has a connection to all the members. In a cooperative pool with n members, there are n*(n1) connections in total, while in a commercial pool of n members, there are 2*n connections, respectively.

An average airline operator sources all the mainte-nance of its ﬂeet from its own MRO department that subcontracts parts of it to external service providers. It is common to include the availability service of aircraft components within a general component support agreement.

2.3. The cost of availability service

Carter and Monczka (1978, p. 28) identify three cost elements in MRO inventory pooling as follows: inventory holding, ordering, and back-order costs. In order to bring out the differences between the cooperative strategies, ordering costs need to be further divided into handling and transfer costs. Similarly, back-order costs have to be divided into loan-in and wait costs. In addition, interface costs represent the annual ﬁxed costs of maintaining relationships between the cooperating parties.

Inventory holding cost is the prevailing item including the cost of capital and all storage costs. Regardless of the size of spare components inventory, there will be shortages that are usually solved by borrowing the required spare units from external loan provider leading to loan-in costs.

The handling costs cover the on-site per transaction costs that arise when a spare unit is needed. This includes picking and transportation costs within one base. The handling costs are the same should the unit come from the local stock or from a remote site. In the latter case, the handling costs include identifying the unit in the goods receiving area instead of picking it from the stock.

The transfer costs incur when the unit is not stored in the customer’s home base or when units need to be transferred between sites for stock balancing. These costs cover all the per transaction costs of the transfer.

In those cases when the spare unit is not available at the home base, there is a risk of flight delays. Sometimes it is possible to avoid or at least postpone the flight delay by utilizing the scheduled downtime. This arrangement involves some costs. Wait costs include both arrangement costs and flight delay costs.

The interface costs are the annual transaction costs between two parties involved in the availability service. These costs are proportional to the interface complexity between the two parties and include the annual costs of maintaining the relationship as well as the amortization of the initial negotiation costs.

3. Cooperative strategies

There are several cooperative strategies for component availability services which airlines can practice. Fig. 1 illustrates a strategy framework that accounts for the number of participants involved and the tightness of the contractual integration. The most traditional strategy is to perform the availability service in-house so that the airline provides the service only for its own ﬂeet. This is called solo strategy.

Neighbor airlines that have some ﬂeet commonality easily drift into a loose form of cooperation by providing a loan unit against a fee when there is an aircraft on ground needing it. If the relationship and trust between parties

ARTICLE IN PRESS

Contractual Integration Number of Participants ₁ 2 3 >3 Tight None Ad Hoc Cooperation Solo Loose Cooperative Pooling Commercial Pooling Partnership based 1 2 3 >3 Market based

Fig. 1. Framework of cooperative strategies. J. Kilpi et al. / Int. J. Production Economics 117 (2009) 360–370

362

Figure 2.1: Different pooling strategies. Figure obtained from Kilpi et al. (2009, p. 362)

As a result of the great potential of spare parts pooling, it is extensively studied within operations management literature (Benjaafar et al., 2001; Swinney, 2012). In recent years, it has shown its potential in several industries such as the airline industry (Kilpi and Veps¨al¨ainen, 2004) and the semi conductor industry where in the study of Kranenburg and Van Houtum (2009) substantial savings of 35% were realized by applying pooling for spare parts. Within the electricity industry pooling also proofed to be beneficial. An example is given by Kukreja et al.(2001) on independently managed power-generating plants of a large energy company. They found that annual savings of 67% could be obtainable if pooling was applied here.

While it is known that generally horizontal collaborations may lead to compelling cost reductions accompanied by several other benefits (Soosay and Hyland, 2015; Basso et al., 2019), literature on resource sharing for O&M in offshore wind settings has received scant attention (Shafiee, 2015b; Broek et al., 2019). Nonetheless, there are scholars who investigated the influence of pooling on performance within the OWI. Irawan et al. (2017) examine the cost-saving potential of sharing transfer vessels and technicians among OWF’s served by the same harbour. A study which solely focuses on the sharing of technicians but served from different harbours is performed by Schrotenboer et al. (2018). Two studies focusing on sharing jack-up vessels in the are performed by Beinke et al.(2017) and Broek et al. (2019). Where the former investigated the benefits of resource sharing during the installation phase of offshore wind parks, and the latter during its operational phase. However, all in all existing literature remains limited. This may be explainable through the exploratory nature of sharing within the OWI combined with the offshore aspect of the industry. This, as operating in an offshore setting goes accompanied with some industry-specific difficulties on which will be elaborated below.

2.4.1 Industry-specific difficulties of pooling

(12)

This, as harsh weather conditions result in an increase in spare parts demand. According to Carroll et al. (2015), there exists a clear relationship between the amount of turbine failures and wind speed. This is underlined by the findings of Tavner et al. (2006) and Wilson and McMillan (2014). They argue that there is a trend where turbines sited in areas with higher wind speeds, tend to experience higher failure rates. Lastly and as for all industries, the utilization of spare parts is highly related to maintenance since a spare is required when maintenance is conducted. This dependence is more challenging within an offshore setting due to the aforementioned weather constraint, which leads to demand arriving in stochastic batches (Attar et al., 2016).

Despite these difficulties, the study of J¨ager-Roschko et al. (2019) investigated whether pooling spare parts of OWT’s by using a centralized warehouse is beneficial. In doing so, the authors assessed the potential of pooling for small, medium and large spare parts. As mentioned, they have found that solely major spare parts are beneficial for pooling purposes while admitting that due to poor availability of data, their results are limited. Accordingly, the authors suggest to assess the influence of pooling spare parts on the performance of the O&M of an OWF while using different criteria. In particular, a change in turbine failure rate and holding costs seems to be fitting. This thesis complements their work as our aim is to investigate the possibilities of solely pooling major spare parts of OWT’s while using different criteria for inter alia the turbine failure rate, the holding costs but also for other input data. Moreover, as became apparent in this section, lead times and the influence of wind speeds are also atypical concerning pooling within the OWI. Accordingly, as an extension to the suggestion made by J¨ager-Roschko et al. (2019) we are going to assess how the different lead times and high wind speeds are influencing the application of pooling on the performance of an offshore wind farm.

3 Methodology

3.1 Research Method

As becomes apparent from the research question, its aim concerns the identification of the relationship between applying spare parts pooling as a policy to manage inventory and the performance of an OWF. Hereby aiming to identify- and compare links between control variables (i.e., inputs), and performance variables (i.e., outputs). Accordingly, a quantitative research approach should therefore be applied as establishing links between control-and performance variables is one of its strengths (Bertrcontrol-and control-and Fransoo, 2016). Specifically, simulation is used. Simulation can be defined as:“the process of designing a model of a real or imagined system and conducting experiments with that model” (Smith and Chief Scientist, 1999, p. 2). In doing so, simulation is able to to cope with a high numerical complexity (Bertrand and Fransoo, 2002). The latter being an important reason that simulation lends itself well for the purposes of this thesis. This, as due to the numerous factors of influence coupled with the dynamic offshore environment, the O&M of the OWI faces a high degree of complexity which is tricky to analyze analytically (Jacobsson and Karltorp, 2012; Schwanitz and Wierling, 2016). By using simulation this saves on costs and simultaneously allows for greater control than for example experimental research (Budnick et al., 1977). Simulations thus function as ’what-if’ tools and allow us to evaluate a variety of alternatives (Gass, 1977; Robinson, 2014) which is desired here, as we want to assess the influence of pooling on the O&M-performance of an OWF under different conditions. So, as a consequence of the numerical complexity, high real-life setup costs and the flexibility to run a variety of experiments, simulation is highly suitable for this thesis. Yet, simulation also has drawbacks. It is labour intensive and may result in issues with external validation of the results as a simulation remains a simplified version of reality (Robinson, 2014). To counter these pitfalls, the assumptions made in this research are in line with assumptions made in preceding research as much as possible and the inputs are tested by means of a sensitivity analysis.

(13)

3.2 Model components

In Figure 3.1 the components of the quantitative model are depicted. The model is made up of 3 layers. The known information that enters the simulation (Data), their inter-relatedness (System Dynamics) and the Key Performance Indicators (KPI’s) (outputs) used to assess the effects of pooling under various conditions (Robinson, 2014).

The data can be divided into five components. The deterioration behaviour refers to the deterioration path an OWT follows and the spare parts failure behaviour. Each OWT follows an independent deterioration path defined by parameters which are outlined in Section 4.5. In the warehouse the stock levels are determined which are constrained by the service level. The warehouse can follow a NP-policy or P-policy and the amount of time associated with the SRLT is incorporated here. The OWF-environment refers to the properties of the simulated OWF such as the amount of OWT’s, their electricity output and the ITLT. The weather determines if maintenance can be performed based on wind speeds and wave height. Lastly, the costs refer to all costs such as holding costs and costs associated with for example maintenance. The system dynamics are the key state variables that define the behavior of the system and relate all variables to another (Dooley, 2002). Finally, to measure the performance of the different model-configurations we need to have measurable output. Accordingly, we use several KPI’s to quantify the output data. The KPI’s are reflecting on the responsiveness via the Time-based Availability (TBA) and Mean Time To Repair (MTTR) and the costs via the Costs Of Downtime (COD), the inventory cost and the cost of maintenance.

As can be seen in Figure 3.1, each component has an influence on the components in subsequent layers. To understand the layers and their components properly, first the quantitative model will be explained after which the input data, system dynamics and KPI’s will be explained in detail per layer.

K P I’s D at a Sy st em D yn am ic s Responsiveness Costs Deterioration Behaviour Warehouse Offshore Wind Farm Environment Weather Costs Failure Occurence Resource Availability Operations and Maintenance Weather Availability Time-Based Availability Mean Time to

Repair Cost of Delay Inventory Cost

Cost of Maintenance

(14)

4 Quantitative Model

This chapter elaborates on the developed quantitative model used to assess the research question. We assume an OWF-environment in which spare part failures occur independent of each other and when occurring, they lead to a complete system malfunction, meaning that the OWT can only be maintained in its entirety.

4.1 Deterioration behaviour

Deterioration can be modelled in various ways but as it is uncertain over time, ideally it should be represented as a stochastic process (Van Noortwijk and Pandey, 2004). With respect to OWT’s, examples where a stochastic process is successfully applied as a way of modelling deterioration can be found in Hameed and Vatn (2011), Shafiee and Finkelstein (2015), and Florian and Sørensen (2017).

To model deterioration, we use the stationary gamma process. The gamma process was introduced in the area of reliability by Abdel-Hameed (1975) and is suitable to model gradual damage monotonically accumulating over time, such as wear, fatigue and corrosion which are also common issues for OWT’s (Van Noortwijk and Pandey, 2004). It is a stochastic process with independent, non-negative increments having a gamma distribution with an identical scale parameter (Noortwijk, 2009). A contemporary study where the gamma process was utilized to model the deterioration for wind turbines can be found in H. Chen et al. (2017). To simulate the gamma process, we make use of gamma increment sampling after the example of Noortwijk et al. (2007) and Hoekzema (2020). The deterioration increases with age, whereby step sizes in time are denoted as ∆t, using random increments from the gamma distribution. Specifically, the longer the OWT is operative, the higher the likelihood it fails (Lo et al., 2007). We use the definition as stated in Equation 4.1 for the density function f of the gamma distribution with shape parameter α > 0 and scale parameter β > 0.

f(t; α, β) = β

α_tα−1_e−β t

γ(α) , (4.1)

with the following properties: 1. t(0) = 0 with probability 1 2. X(t) has independent increments

3. X(t) is a jump process with infinite jumps in any time interval

To determine the number of failures per time unit, the average daily deterioration increment, given by µ = α β,

is used. When we divide the failure threshold by µ, the average amount of time for a OWT to fail is calculated. To determine the variability one could use the standard deviation given by σ = p(α) · β. However, as the Coefficient of Variaton (CV), given by CV = σ

µ, is a more general measure, in this thesis we use the CV.

4.1.1 Modelled spare parts and their failure behaviour

If a company intends to pool its spares, two prerequisites need to be met: (1) the need of having at least one other company using similar technical systems within a certain geographical area; and, (2) the value of the spares a company chooses to pool must constitute the main part of the total value for all spare parts in the system (Muhaxheri, 2010).

(15)

Figure 4.1: Wind turbine spare parts. Figure obtained from Tian et al. (2011, p. 1506).

As four spares are considered, it must be identified which spare has failed when a failure occurs. Following the example of Nguyen and Chou (2018), this is determined based on the failure rate per year expressed as a probability, which is then converted in a weighted probability of the likelihood of a certain spare part failure. The failure rates are also adopted from Nguyen and Chou (2018) and shown in Table 4.1 together with their corresponding probabilities.

Spare part Failure type Failure rate/year Probability

Bearing 1 0.6001 29,58%

Gearbox 2 0.5076 25,02%

Generator 3 0.7897 38,93%

Rotor system 4 0.1307 6,44%

Table 4.1: Failure rates and probabilities per spare part. Data obtained from Nguyen et al. (2018, p. 91)

4.2 Resource availability modelling

To control inventory, a variety of inventory control policies can be used. For an extensive overview of different inventory policies used for controlling inventory, we refer to Do Rego and Mesquita (2011). In determining which inventory control policy to use for this study, we need to look at the nature of the demand for spare parts. As we can recall from Section 2.3, there exists a relatively low and intermittent demand, occurring in batches. Additionally, the order replenishment costs are small enough compared to the cost of spares to be neglected (Zhang et al., 2019).

Under these conditions, the base-stock policy (S 1, S) is a sensible policy, and thus used in this thesis (Kouki and Larsen, 2020). The base-stock policy works as follows: A replenishment order is initiated whenever the inventory position drops under a certain base-stock level ’S’. The size of the order should increase the inventory position up to the base-stock level, keeping the inventory position always constant and equal to ’S’ (Do Rego and Mesquita, 2011; Attar et al., 2016). Although the base-stock policy may be not optimal regarding lost sales as exemplified by Hill (2007), it is generally considered a good policy for modelling spare parts in inventory systems (Kouki and Larsen, 2020).

4.3 Operations and Maintenance

4.3.1 Response time modelling

The response time is the time between failure detection and the performed maintenance action as a response (Gonzalez et al., 2017). This time is modelled as the inbound transportation time plus the time as a consequence of delays (Monbet et al., 2007; Broek et al., 2019). The response time is dependent upon the spare part availability and the weather. Equation 4.2 below represents the response time. For this thesis it is assumed that the ITLT is the sum of the jack-up vessel mobilisation time, planning time and the on- and offshore transportation time. A jack-up vessel is used for maintaining the OWT’s as this type of vessel is required for major spare parts. It is a large vessel using extendable legs to elevate the hull above the sea surface and is equipped with a crane competent to lift heavy spares (St˚alhane et al., 2017).

(16)

4.3.2 Repair modelling

If the deterioration level exceeds the failure level (L) and the response time has passed, CM is initiated. It is assumed that a CM action will bring the system to a state which is as-good-as-new, assuming perfect maintenance. Moreover, as a simplification, the time to perform a maintenance action is assumed to be within one day. Specifically, when a maintenance action is performed, the deterioration level directly goes back to 0 (zero) indicating an as-good-as-new state (Rinaldi et al., 2017; Sahnoun et al., 2019). Further, it is assumed that when a day with suitable weather occurs, all accumulated repairs who could not be initiated due to previous weather conditions or inventory deficits will be initiated as long as the required spare parts are in stock, assuming infinite repair capacity (Fritzsche, 2012). Lastly, condition monitoring equipment is assumed to be perfect, therefore obtained deterioration levels are considered true (Byon et al., 2010; Yildirim et al., 2017; Canizo et al., 2017).

4.3.3 Electricity output

For the electricity output, we use wind turbine power curve modelling after the example of Lydia et al. (2014). In generating energy, an OWT follows a specific power curve existing of four segments divided by the cut-in, maximum production and cut-out wind speeds. The cut-in speed (Vcut−in) is the minimum required wind

speed for the OWT to generate energy. The wind speed whereby the OWT is generating energy at full capacity (Vmax) is referred to as the maximum production speed. Lastly, the cut-out speed (Vcut−out) represents the

maximum wind speed at which the OWT is still able to generate electricity. When the wind speed surpasses this threshold, the OWT is shut down as a safety measure to prevent damage. The electricity generated at a given wind speed is denoted as Pv and can be obtained by Equation 4.3, where µ represents the nameplate

capacity of the OWT.

Pv=              0 · µ for v ≤ vcut−in 0.15 cos µ + v − vcut−in vmax− vcut−in π+ 1

· µ for vcut−in≤ v ≤ vmax

1 · µ for vmax≤ v ≤vcut−out

0 · µ for v ≥ vcut−out

(4.3)

4.4 Weather modelling

Any OWT maintenance action is subject to weather conditions (Zhang et al., 2019). Due to its undeniable influence, the weather is included in the model. The state-of-the-art in weather window estimation today is limited to the use of simple ocean parameters, such as wind speed significant wave height (Gintautas et al., 2016). Despite weather conditions may vary from day to day, a clear seasonality can be recognized over time enabling it to be modelled based on historical data (Seyr and Muskulus, 2019). Several sophisticated weather models can be developed when the required data is available. Hagen et al. (2013) showcase a multivariate Markov model to generate sea state time series. They assumed the wave height, wave direction, wave period, wind speed and wind direction to be of influence on the sea state. To determine the seasonal variability, two ways are outlined. The first way is by using monthly models assuming piece-wise stationarity. The second model uses a transformation in data to deal with the seasonal variation. Another example is the study of Martins et al.(2015), where the authors also made use of a probabilistic model.

However, to incorporate the weather in our model, we follow the example of Santos et al. (2015). For each day a random number between 0 (zero) and 1 is generated out of a uniform distribution. The generated number is then set against the probability of suitable weather for the season observed. If the generated random number turns out to be greater then the probability, then it is assumed the weather is not suitable to perform maintenance, if it turns out to be less the weather is assumed suitable. This method adequately captures the variation in weather (Santos et al., 2015). The weather is considered suitable when the operational limits of the jack-up vessel of 10 meters per second for the wind speed and 2 meters for the wave height are not exceeded, these capabilities are in line with the study of Dinwoodie et al. (2015).

4.4.1 Influence wind speed on WT-failures

(17)

To assign labels to the input parameter wind speed, we adopted the labels used by Reder et al. (2018). In their study, the authors distinguished four categories of wind speeds; calm, low, high and stormy. We translated these labels to calm, normal, high and stormy where a calm, normal and stormy wind speed pose no additional damage to the OWT and a high wind speed does cause additional damage. In the model, for high levels of wind speed the increments are picked from a different distribution which is subordinate to a higher annual failure rate and thus causes the OWT to deteriorate faster. In their study, Carroll et al. (2015) found that there is a 33% increase in annual failure rates due to wind speeds alone. Let us assume that only for high wind speeds this applies as for stormy wind speeds the OWT’s will be shut down and thus follow their normal failure behaviour. The labels, their corresponding wind speeds and the annual failure rate for the different wind speeds are depicted in Table 4.2.

Label Wind speed [m/s] average annual failure rate

Calm >3 1

Normal 3 -13 1

High 14 - 24 1.33

Stormy ≥25 1

(18)

4.5 Input parameters

In this section, the input data of the simulation model are discussed. In Table 4.3 all inputs are outlined arranged by data component. For the warehouse and the OWF-environment component there is differentiated between a NP-policy and a P-policy for the lead times. This, as there are two configurations for the base-case scenario, one where pooling spare parts is applied and one without.

Value

Data Component Parameter Sensitivity

Analysis Base-case Scenario Sensitivity Analysis Deterioration behaviour Failure level 1000

Failures per year 0.5 1 1.5

Average deterioration per day [wind influence] ≈ 3.03

Average deterioration per day [no wind influence] ≈ 4.03

Coefficient of variation 4.7

Warehouse Service level 91% 95% 99%

SRLT NP-policy [days] 7

SRLT P-policy [days] 6

Offshore wind farm environment

ITLT NP-policy [days] 5

ITLT P-policy [days] 6

Amount of wind turbines 20

Wind turbine nameplate capacity [MWh] 1.8 3.6 5.4

Cut-in wind speed [m/s] 3

Max. production wind speed [m/s] 14

Cut-out wind speed [m/s] 25

Weather

Probability suitable weather Spring 66%

Probability suitable weather Summer 72%

Probability suitable weather Autumn 41%

Probability suitable weather Winter 33%

Costs

Cost of energy [€/MWh] 50 100 150

Holding cost Bearing [€/day] 300 600 900

Holding cost Gearbox [€/day] 760 1520 2280

Holding cost generator [€/day] 500 1000 1500

Holding cost rotor system [€/day] 560 1120 1680

Ordering cost [€/order] 400

Cost of maintenance [€/day] 21,069 42,138 63,207

Settings

Step size ∆t [days] 1

Simulation length ∆t [years] 10

Number of runs 15

Table 4.3: Model parameters including values for the sensitivity analysis

4.5.1 Deterioration behaviour

(19)

To the best of our knowledge, no studies elaborate on certain CV’s to adequately describe the deterioration process of an OWT in practice, the CV is arbitrarily set to 4.7 whereas we found that this CV obtains the desired spread in deterioration behaviour. In Figure 4.2 the spread in OWT-failure behaviour is depicted. Taken over 10 years, this results in a random failure process with an average of one failure a year for each OWT.

8% 22% 38% 23% 9% 0% 5% 10% 15% 20% 25% 30% 35% 40% 115-208 209-302 303-396 397-490 491-584 Pe rc en ta ge o f f ai lu re s Time [days]

Figure 4.2: Spread of failure behaviour with a CV of 4.7

4.5.2 Warehouse

As stated in Section 2.3, the service level acts as a constraint on which the stock levels are based. It has been found that a 95% service level is considered appropriate for intermittent demand (Dunsmuir and R. Snyder, 1989; Strijbosch et al., 2000; Syntetos et al., 2009; Teunter and Duncan, 2009). It is assumed that for a P-policy the members in the pool have the same target service level (Kilpi and Veps¨al¨ainen, 2004). Accordingly, the base stock levels are set in that way that at a minimum, 95% of the demand can directly be satisfied from stock. According to Silver et al. (1998), estimating the demand distribution based on historical demand is the best known basis for choosing stock parameters. So, an empirical demand distribution based on demand data for a period of 10 years is generated for each spare part. This is done for the NP-policy and the P-policy as well as for the scenarios where high wind speed accelerates deterioration. Also, as became apparent in Section 2.3, the stock level is dependent upon the SRLT, a longer SRLT implies a higher base-stock level (Chopra et al., 2004). Subsequently, the same procedure is executed for the different SRLT’s. In Appendix- A and B the different inventory levels required to achieve the 95% service level for the NP-policy and the P-policy are given.

For the stock replenishment lead time, we assume it to be independent of the specific spare part and use the lead time for the bearing and the generator for all spare parts, assuming equality in SRLT. Following the lead of multiple scholars, we set the stock replenishment lead time to 7 days for a NP-policy and for the P-policy we assume a lead time of 6 days (Lindqvist and Lundin, 2010; Dewan, 2014). This difference stems from the fact that the centralized warehouse required to perform a P-policy has a better negotiation position due to scale advantages resulting in a lower SRLT. In Figure 4.3, a schematic overview of the spare parts delivery process is depicted, where the SRLT is highlighted with the green arrow.

W WF

W Warehouse

WF Wind Farm

Inbound transport lead time OEM

Stock replenishment lead time

OEM Original Equipment

Manufacturer

Figure 4.3: Schematic overview of the the spare parts delivery process - Stock replenishment

4.5.3 Offshore wind farm environment

(20)

This, as for a P-policy, the warehouse has to be located somewhere in the middle of the OWF’s it serves resulting in a sub-optimal location. When one warehouse serves only one OWF as for the NP-policy, it can be positioned as close to the OWF as possible saving inbound transportation time. In Figure 4.4, an overview of the spare parts delivery process is depicted, where this time the inbound lead time is highlighted with the green arrow.

W WF

W Warehouse

WF Wind Farm

Inbound transport lead time

OEM

Stock replenishment lead time

OEM Original Equipment

Manufacturer

Figure 4.4: Schematic overview of the the spare parts delivery process - Inbound transportation

Within the model, we assume an OWF consisting of 20 OWT’s. Their nameplate capacity is set to 3.6 Megawatt hour (Mwh) following the example of Koukoura et al. (2016). To determine the electricity output per wind speed, we followed Broek et al. (2019). Accordingly, the cut-in wind speed is assumed to be 3 m/s, with a maximum production speed of 14 m/s and a cut-out speed of 25 m/s. These settings result in the power curve depicted Figure 4.5a. There are three levels of electricity output; no electricity output, an output of 12,96 MW/day and an output 86,4 MW/day. This is a result of the four considered categories of wind speed (low, normal, high, stormy), where no output is generated for calm- and stormy wind speed conditions, an output of 12,96 MW is generated for normal wind speeds and an output of 86,4 MW for high wind speeds. The output levels are set in that way so that they accurately reflect reality. According to EWEA (2019), an average OWT of 3,6 MW generates 13.248 MW a year. This results in an average daily electricity output of 13.248

365 ≈ 36, 3

MW. In Figure 4.5b, this output level is indicated with the orange line. The variation in daily electricity output in MW is between 34,9 and 37,36. Hereby, the vast majority of daily electricity output lays between 36,14 and 36,54 MW. Considering that the average daily electricity output is 36,3 MW, this reflects reality in a credible way. 0 10 20 30 40 50 60 70 80 90 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 Po w er o ut pu t [M W ] Wind speed [m/s] Electricity generated [MW] Cut-out speed Max. prod. speed

Cut-in speed

(a) Wind turbine power curve [3,6 MW]

5% 10% 15% 42% 13% 15% 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 34,9 - 35,3 35,31 - 35,71 35,72 - 36,13 36,14 - 36,54 36,55 - 36,95 36,96 - 37,36 Am ou nt o f e le ct ric ity o ut pu t [in % ] Electricity [in MW]

(b) Distribution of daily electricity output [MW]

Figure 4.5: Power curve and daily electricity output

4.5.4 Weather

(21)

4.5.5 Costs

The COE, are assumed to be€100/Mwh after the example of Broek et al. (2019). To assign the daily holding costs to each spare part, we follow the example of Zhang et al. (2019) and use a holding cost rate of 0.01. By multiplying this rate by the purchase cost of the respective part we find the holding cost. The purchase costs for the spare parts are adopted from Tian et al. (2011) and are: €60,000 for the bearing; €152,000 for the gearbox; €100,000 for the generator and €112,000 for the rotor system. The resulting holding costs are given in Table 4.3. In the study of J¨ager-Roschko et al. (2019), the costs for a major spare part are set to€106,500 and the holding costs are calculated by multiplying the spare parts cost with the holding cost rate of 28.7%, the average stock level and the duration of the simulation. This difference in the holding cost rate is due to the fact that J¨ager-Roschko et al. (2019) include a service fee for the service provider of the centralized warehouse. In this research, we do not incorporate such a fee as we are solely assessing the effects of pooling spare parts. The order replenishment costs are set to €400 per order assumed to be independent of order size- and OEM distance (Volkmar, 2014). The order cost are included as they are part of the running cost of operations but they are not influenced by- or influencing the chosen inventory policy (NP-policy or P-policy) in any fashion. In this research, the maintenance cost represent an overarching cost item which has any maintenance-related cost embedded in it. When an OWT fails and spare parts are available just as there is a suitable weather window, the inbound lead time starts to run. Maintenance cost are incurred during this whole period of downtime. We assume these cost to be per day and they can be sub-divided in on- and offshore operation costs which include all fixed costs, operating cost, CM costs and installation costs. For onshore operation, we make use of the costs used by Shafiee and Dinmohammadi (2014). In their study, the onshore operation cost per kilometer are €500 at a speed of 28.8 minutes per km. Given that an average working day consists of 8 hours, this results in approximately 16 km travelled a day. This results in a daily cost of 16 · 500 =€8000. For offshore operation we follow the insights of Dalgic et al. (2013), resulting in a cost of 67.800 for operating a jack-up vessel originating from 2012 per day. To convert these costs from British Pounds (GBP) to euros, we assume 1 GBP = 1,125 Eur (Broek et al., 2019), which results in daily operating cost of€76.275 for a jack-up vessel. To come to the daily maintenance cost lets assume that it is represented by the average of these two costs resulting in a maintenance cost of 76.275+8000

2 ≈€42,138 a day. Since this is a rough estimation of the actual costs they are subjected to

a sensitivity analysis.

Finally, the simulation length is set to 10 years with a step size (∆t) of one day and the number of runs 15. We found that this number is sufficiently large to get stable results for both the KPI’s assessing the responsiveness and the costs.

4.6 System Dynamics

The model content consists of the dynamic components that are represented in the model and their intercon-nections (Robinson, 2008) which can be split into two overarching dimensions (Robinson, 1994):

• the scope of the model: the model boundary or the breadth of the real system that is to be included in the model

• The level of detail: the detail to be included for each component within the models’ scope

Specifically, this part of the model reflects the O&M of an OWF and determines how the inputs are treated. In Figure 4.6 a logic flow diagram is presented containing the model scope and visualizing its level of detail.

Weather window suitable? Spare parts failure type available? Failure Failure type 1? Failure type 2? Failure type 3? Failure type 4? Planning time Corrective maintenance initiated Corrective maintenance performed

(22)

In the simulation, each day the state of the OWT’s is assessed. An OWT can be either in active- or inactive state. If an OWT is in active state, then no action will be taken. If its in inactive state, the process depicted in Figure 4.6 is initiated. First there will be assessed which type of failure occurred. Each failure type represents a certain spare part failure, as described in Section 4.1.1. Then, there will be checked if the weather is suitable to conduct maintenance, where after there will be examined whether the spare part required for the repair is available. In the model, this is checking the failed spare against the simulated inventory. The warehouses serves as a base of operations for the OWF’s. From these locations, spare parts will be transported to the concerning OWF. The OWF’s considered in this thesis are hypothetical, derived from an average OWF. Yet, as a consequence of how we developed the model, as mentioned we assumed to have 20 OWT’s instead of 100 which is typical for an OWF (WindEurope, 2019). If the spare part is unavailable or the weather is unsuitable, a delay occurs and the repair will be postponed until the spare part arrives and the weather is suitable. Since only key spares are considered, the OWT is in failed state during this whole time. If the respective spare part is available and the weather allows for maintenance, the repair will be concluded by using a jack-up vessel and the OWT returns to an active state. As can be inferred from the process described above, all system dynamics are related to each other. If one component fails to deliver, this affects the rest of the process.

4.7 Key Performance Indicators

As can be recalled from the theoretical framework, we want to investigate both the trade-off between inven-tory costs and service, and between inveninven-tory costs and lead time costs. Thus, on the one hand we want to determine the OWF-responsiveness and on the other hand we want to reflect on costs. Five different KPI’s are considered which are relevant for O&M within the OWI (Gonzalez et al., 2017). As mentioned, to quantify the responsiveness, we use the TBA and the MTTR, to determine the costs we use the COD, inventory cost and the cost of maintenance.

The TBA represents the time that an OWT is able to generate electricity divided by the total period of time (Gonzalez et al., 2017). The TBA can be seen as an indication of the performance of maintenance since a low TBA typically indicates poor maintenance (Dinwoodie et al., 2015). In equation 4.4 the mathematical expression for the TBA is provided.

TBA = Total Available Time

Total Time (4.4)

The MTTR is the average time in which a OWT can return to its functional state and should be minimized as much as possible (Gonzalez et al., 2017). The MTTR directly mirrors the speed of a repair and is thus an essential measure of the efficiency of the OWI-supply chain (J¨ager-Roschko et al., 2019). In equation 4.5 the mathematical formulation for the MTTR is given.

MTTR = Total Down Time

Number of Failures (4.5)

The COD reflects upon the loss in revenue as a result from turbine downtime (Dinwoodie et al., 2015). Also for this measure applies that delays will lead to a higher COD which makes this KPI a reflection of the service quality expressed in costs. To incorporate the costs associated with downtime in the model, we follow the example of Broek et al. (2019) by modelling downtime costs as the loss of electricity production during downtime multiplied by the COE. The calculation of the COD is depicted in equation 4.6.

COD= (Turbine Capacity - Actual energy produced) · Cost of Energy

Total Time (4.6)

As already mentioned, the cost of inventory refer to the costs associated with ordering- and holding spare parts. As we can recall from Section 4.5.5, the ordering cost are not related to the deployed inventory policy. The inventory cost can be calculated by equation 4.7.

(23)

The last output are the costs associated with conducting maintenance which can be obtained through equation 4.8.

Cost of Maintenance = #Days Inbound Transportation Time · Daily Maintenance Cost (4.8)

4.8 Scenarios

This section outlines the different scenarios. The base-case scenario consists of two configurations regarding their spare parts management. One configuration uses a NP-policy while for the other a P-policy is applied. Both configurations of the base scenario are described below.

Configuration 1: NP-policy - Current practice

The first configuration most trustworthy reflects the current practice of servicing OWF’s. Here, the spare parts are delivered from one warehouse to exactly one OWF. A visual representation is depicted in Figure 4.7.

W

a

WF

a

W

b

WF

b

W Warehouse (a), (b) or centralized (c)

WF Wind Farm (a) or (b)

Supply and service transfer

Figure 4.7: Visual representation of servicing from own warehouse

Configuration 2: P-policy

Under this configuration the use of a centralized warehouse among multiple OWF’s is introduced. The visual-ization is shown in Figure 4.8 below.

W

c

WF

a

WF

b

W Warehouse (a), (b) or centralized (c)

WF Wind Farm (a) or (b)

Supply and service transfer

Figure 4.8: Visual representation of servicing from centralized warehouse

“The application of spare parts pooling within the oﬀshore wind industry”