Availability vs. Cost Efficiency

(1)

STOCKHOLM SVERIGE 2020,

Availability vs. Cost Efficiency

A Case Study Taking on an Integrated Approach to Spare Part Distribution in the High-Tech

Industry

EMMA BOSTRÖM JULIA LUNDELL

KTH

SCHOOL OF INDUSTRIAL ENGINEERING AND MANAGEMENT

(2)

(3)

Availability vs. Cost Efficiency

A Case Study Taking on an Integrated Approach to Spare Part Distribution in the High-Tech Industry

by

Emma Boström Julia Lundell

Master of Science Thesis TRITA-ITM-EX 2020:238 KTH Industrial Engineering and Management

Industrial Management SE-100 44 STOCKHOLM

(4)

Tillgänglighet kontra kostnadseffektivitet

En fallstudie om strategisk integrering av

reservdelsdistribution inom högteknologisk industri

Emma Boström Julia Lundell

Examensarbete TRITA-ITM-EX 2020:238 KTH Industriell teknik och management

Industriell ekonomi och organisation SE-100 44 STOCKHOLM

(5)

Availability vs. Cost Efficiency

A Case Study Taking on an Integrated Approach to Spare Part Distribution in the High-Tech

Industry

Emma Boström Julia Lundell

Approved

2020-06-02

Examiner

Jannis Angelis

Supervisor

Luca Urciuoli

Commissioner

Mycronic AB

Contact person

Tove Suntjens

Abstract

Finding the proper balance between availability and cost efficiency is an important concern within spare part management. Spare part suppliers need to respond quickly to customer demand as a stock-out can have severe consequences for both the customer and the supplier. It is critical to identify what items to keep in stock and where to allocate the inventory to avoid stock-outs.

This case study was performed at a large high-tech company producing manufacturing equipment to be used in the electronics industry. The aim was to lower the stock-levels of spare parts while not impairing the availability by combining item classification, demand forecasting, and distribution network optimization. A decision diagram for classifying spare parts was constructed using the analytical hierarchy process. Twenty items were classified using the diagram, and the demand for them was forecasted using the Syntetos Boylan Approximation- method. The shipping cost for spare parts within one region was minimized using a linear optimization model.

The analysis showed that equipment criticality, annual usage value, and installed base are critical when classing spare parts. Instead of using five distribution centers in the European region, it was discovered that the shipping costs would decrease if only three warehouses made up the distribution network. The spare parts investigated appeared to follow the typical characteristics for spare parts, showing a low and irregular demand. Hence, demand forecasting seemed to be unnecessary, considering the difficulties in getting satisfactory results. Instead of combining the results from classification, forecasting, and inventory allocation, we suggest that the processes affecting stocking decisions should cooperate and work towards a common objective, namely to satisfy the customer demand in a cost-efficient way. Thus, widening the meaning of taking on an integrated approach to spare part management.

Key-words: Spare part management, Inventory, Classification, Demand Forecasting, Distribution Network, Integrated approach, Aftermarket supply

(6)

Examensarbete TRITA-ITM-EX 2020:238

Tillgänglighet kontra kostnadseffektivitet En fallstudie om strategisk integrering av reservdelsdistribution inom högteknologisk

industri

Emma Boström Julia Lundell

Godkänt

2020-06-02

Examinator

Jannis Angelis

Handledare

Luca Urciuoli

Uppdragsgivare

Mycronic AB

Kontaktperson

Tove Suntjens

Sammanfattning

Inom hanteringen av reservdelar är det en stor utmaning att hitta rätt avvägning mellan tillgänglighet och kostnadseffektivitet. Leverantörer av reservdelar måste snabbt kunna möta kundefterfrågan eftersom uteblivna leveranser av kritiska reservdelar kan få allvarliga konsekvenser för både kund och leverantör. Vilka artiklar som ska lager-hållas och var de ska lagerhållas är avgörande beslut för att undvika att artiklar rest-noteras. I den här fallstudien, som utfördes på ett stort teknikföretag som tillverkarproduktionsutrustning till elektronikindustrin, var syftet att sänka lagernivåerna av reservdelar utan att göra avkall på tillgängligheten. Detta genom att kombineragruppering av artiklar, beräkning av kommande efterfrågan och optimering av distributionsnätverket. För att klassificera artiklar i grupper med liknande egenskaper skapades ett schematiskt beslutsdiagram med hjälp av metoden AHP. Tjugo artiklar ur sortimentet valdes ut som beslutsdiagrammet testades på. För samma tjugo artiklar gjordes prognoser för den kommande efterfrågan med metoden Syntetos-Boylan-Approximation.

Distributionsnätverket i den europeiska regionen optimerades medavseende på fraktkostnad genom att applicera en linjär optimeringsmodell. Hur kritisk en reservdel är för den relaterade maskinens funktionalitet, reservdelensårliga förbrukningsvärde och den geografiska placeringen av installerade maskinervisade sig vara kritiska för att kunna klassificera artiklarna effektivt.

Analysen av distributionsnätverket i Europa visade att fraktkostnaderna kan minskas om nätverket utgjordes av tre lager istället för fem som det gör i dagsläget. De tjugo undersökta reservdelarna uppvisade de typiska egenskaperna för reservdelar som har rapporterats i litteraturen som låg och oregelbunden efterfrågan. Att sätta prognoser på efterfrågan verkar obefogat med tanke på komplexiteten i beräkningarna och att de ger få tillfredsställande resultat.

Istället för att kombinera resultaten från klassificering, prognoser på efterfrågan och lageroptimering föreslår vi att alla de funktioner i ett företag som arbetar med att tillgodose kundefterfrågan bör samarbeta i högre grad och jobba mot ett gemensamt mål, nämligen att tillgodose kundernas efterfrågan på ett kostnadseffektivt sätt. Således vill vi utvidga betydelsen av att ta en integrerad strategi för reservdelshantering

Nyckelord: Reservdelshantering, Lager, Klassificering, Efterfrågeprognoser, Distributionsnätverk, Integrerad strategi, Eftermarknadsförsörjning

(7)

List of Tables

1 Summary of identified parameters in literature . . . 10

2 Aggregated list of parameters . . . 24

3 Overview of meetings and interviews for contextual understanding . . 28

4 Results of semi-structured interviews about classification criteria . . . 31

5 Decision diagram level structure . . . 32

6 Ratio scale (Tzeng & Huang, 2011, p 18) . . . 33

7 R.I. values adopted from (Tzeng & Huang, 2011, p 18). . . 34

8 Boundaries for classification . . . 35

9 Input parameters for distribution network model (Chopra & Meindl, 2016). . . 40

10 Decision variables for the model (Chopra & Meindl, 2016). . . 40

11 Calculated weights for classification criteria . . . 51

12 Items classification . . . 53

13 Overview of demand and forecasting values. . . 56

14 Forecasts for articles P001-P010. . . 57

15 Forecasts for articles P011-P020. . . 58

16 Capacity and excess capacity for distribution centers . . . 58

17 Shipping cost/unit (SEK) from central warehouse to distribution centers 59 18 Demand for and shipping costs to customers sorted per country . . . 60

19 Shipping quantity result from distribution model (in units) . . . 61

20 Proposed stocking policy . . . 62 21 Pairwise parameter comparison answers . . . III 22 Geometric mean value of parameter pairwise comparison . . . IV 23 AHP-matrix W with five parameters . . . V

(10)

24 AHP-matrix W after excluding two parameters . . . V 25 C.I. and C.R. calculations . . . V 26 Calculating vector r . . . VI 27 Forecast of demand for article P001-P010. . . VIII 28 Forecast of demand for articles P011-P020. . . IX

(11)

List of Figures

1 Challenges within spare part management . . . 3

2 Overview of chapter structure. . . 5

3 Decision hierarchy for spare part management proposed by Gajpal et al. (1994). . . 11

4 Size of installed base and spare part demand throughout the product life cycle (Dekker et al., 2013). . . 13

5 Multi-echelon network structures (Eruguz et al., 2016). . . 21

6 Typical multi-echelon structure for distribution of spare parts, adapted from van Houtum and Kranenburg (2015) . . . 22

7 Example of lateral transshipments. . . 22

8 Overview of research design. . . 26

9 Overview of classification . . . 29

10 Overview of decision scheme. . . 37

11 Current stocking policy . . . 47

12 Case company’s distribution network . . . 48

13 Percentage size of inventory volume for each category. . . 50

14 Percentage size of total inventory value for each category. . . 51

15 Parameter comparison answers . . . 52

16 Time series diagram for the articles in demand group 1 and 2. . . 54

17 Time series diagram for the articles in demand group 1 and 2, except for article P001. . . 54

18 Time series diagram for the articles in demand group 3 and 4. . . 55

19 Time series diagram for the articles in demand group 5,6 and 7. . . . 55

(12)

Acknowledgement

Firstly, we would like to thank Mycronic AB for giving us this opportunity. We would like to especially thank Tove Suntjens and Priscila Albuquerque Souza for always taking their time to help and guide us. Their support has been of great value.

We would also wish to thank everyone at Mycronic, who has helped us during this study and participated in our interviews, sharing their knowledge and expertise with us. Their contributions have been central to this thesis.

Secondly, we would like to express our gratitude to our supervisor Luca Urciuoli, Associated Professor at KTH Royal Institute of Technology, who has provided us with endless guidance, feedback, and support throughout this thesis.

Lastly, we would like to thank each other. For the hard work, the late nights, and all the great moments in between. It has been a pleasure to write this thesis together.

Emma Boström and Julia Lundell, June 2020

(13)

1 Introduction

In this introducing chapter, a short background to the characteristics of the aftermarket and especially the supply of spare parts i provided. The reader is introduced to the main challenges in spare part management along with the problem statement and purpose of this study. A short description of the case company at which the studied is conducted at is provided. Lastly, the research questions of the study are presented and their delimitation.

1.1 Background

During the last decades the focus has shifted from offering products to offering solu- tions, and in this new era providing spare parts and service have become an essential source of profit (Cohen et al., 2006). According to Cohen et al. (2006), the aftermarket has, in various industries, grown and become several times larger than the first product business. However, they note that only companies with an efficient spare part and service management will be able to make a profit from it. They even state a direct correlation between a company’s quality in aftermarket offerings and its stock price has been found in studies. Thus, providing excellent aftermarket service is not only about charging customers, but they also state that it is a valuable source of understanding for the customers’ businesses.

The main concern for service maintenance is the availability of correct spare parts (Huiskonen, 2001). As a supplier of spare parts, this makes the market complex due to customers’ high expectations on quick response (Murthy et al., 2004), smooth deliveries and product availability regarding a broad range of parts (Cohen et al., 2006). To meet the high market expectations, suppliers are required to handle the distribution of spare parts fast but cost-eﬀective. That means they need to find the proper balance between stocking cost and availability, i.e., finding the optimal solution on what spare parts to stock, in what quantity, and when to restock the inventory (Hu et al., 2018; Turrini & Meissner, 2019). It is a diﬃcult task due to the characteristics of spare parts; they tend to be expensive and so the associated holding costs can be very high (Turrini & Meissner, 2019; Huiskonen, 2001).Further, the demand can often be intermittent (periods with zero demand) or erratic (vary in size) (Hu et al., 2018; Boylan & Syntetos, 2010).

A useful tool to facilitate challenging stocking decisions is to classify spare parts in categories. In practice, spare parts are most often classified depending on their criticality and cost and diﬀerent stocking policies are applied to the diﬀerent classification categories depending on the category characteristics (Bacchetti & Saccani, 2012). Demand forecasts can also serve as a useful tool to find proper stocking policies (Bacchetti & Saccani, 2012). Combining classification, forecasting, and inventory management is known as taking an integrated approach towards spare part manage-

(14)

ment (Bacchetti & Saccani, 2012; Hu et al., 2018).

A suiting stocking policy that specifies at which facilities it is necessary to keep which spare parts and at what quantity, is necessary for achieving a high service level in the distribution of spare parts (Botter & Fortuin, 2000). Nozick and Turnquist (2001) address the integrated analysis of distribution center location and product- specific inventory policies, known as inventory allocation, to reduce cost and improve service level. In a two-echelon inventory system, the optimal solution is to divide products into two categories; one to be stocked locally and one to be stocked centrally depending on demand and required service level (Nozick & Turnquist, 2001). A general approach to ensure that existing resources are optimally used, i.e., to address the questions of where to allocate inventory to achieve the best service level (Whybark

& Yang, 1996). Frequency of shipments, inventory levels, and demand are three key- factors to include in such work (Whybark & Yang, 1996).

1.2 Problem statement

The main challenge in spare part management is to find a cost-eﬃcient way of handling inventory, namely, to determine the suitable balance between inventory cost and availability (Turrini & Meissner, 2019). It is a diﬃcult task for several reasons (see figure 1). The general assortment of spare parts is diverse and expensive, and it is complex to forecast the demand (Hu et al., 2018; Huiskonen, 2001). When demand occurs, there is a need for fast response, and a stock-out can have severe consequences for both the supplier and the customer (Hu et al., 2018; Huiskonen, 2001). In the case of an OEM providing production machines, a stock-out could cause unacceptable operational or safety conditions at the customer site (Molenaers et al., 2012). Besides, in a turbulent world, there is also a potential risk that contracted suppliers cannot deliver due to external factors. Hence, a spare part supplier must avoid stock-outs;

the question is to what cost.

It is critical to identify what items should be kept in stock and where the inventory should be located to avoid stock outs. To ensure a quick response time to local customer demand, OEMs often use distributor storage (Chopra & Meindl, 2016).

Hence, for companies with multiple distribution centers, the challenge of finding the proper inventory balance is extended to include multiple units, thus increasing the complexity of the problem.

1.3 Research purpose

Spare part management lack research taking on an integrated approach in which spare part classification, demand forecasting and inventory management are linked together (Bacchetti & Saccani, 2012; Duchessi et al., 1988; Turrini & Meissner, 2019). Instead, the three topics have been investigated separately and frameworks discussing the influences between them have been neglected in research (Hu et al., 2018; Bacchetti &

(15)

Saccani, 2012). In this study, the purpose is to improve the availability of spare parts and the cost-eﬃciency by taking on an integrated approach to spare part management.

To increase the availability the right parts need to be stocked in the right quantity at the right time and place, namely, to have the right inventory allocation. Hence, improving the inventory allocation requires a good understanding of the classification process and the demand forecasting.

Figure 1: Challenges within spare part management

1.4 Case company

The company studied in this case study is Mycronic AB, a global supplier of production equipment for display manufacturing and electronics. The company has more than 3,000 customers globally using Mycronic-products in their production and manufacturing. The customers’ performance and eﬃciency are dependent upon a reliably high service-level from Mycronic, and when needed, a quick supply for spare parts.

Currently, Mycronic handles the distribution of products and spare parts with a two- echelon distribution network. The central warehouse is located in Sweden, along with the primary production site and the headquarter. The central warehouse supplies ten distribution centers located in China, France, Germany, Japan, the Netherlands, Singapore, South Korea, the UK and the US, which in turn supply customers in their region. The central warehouse also ships directly to some customers, either standard orders to customers in the region or urgent orders to customers all over the world.

The product range is divided into four business areas. Two of them, called High- Flex and Pattern Generators, are actively managed in Sweden. They diﬀer significantly

(16)

in sold volumes. High-Flex products and spare parts count for a considerably higher annual demand than Pattern Generators. Also, the product range and spare part assortment is broader for High-Flex. Spare parts related to products in both High- Flex and Pattern Generators are purchased from 170 external suppliers and delivered to the central warehouse.

1.5 Research question

To fulfill the research purpose the main research question (MRQ) is:

MRQ: What is a suitable spare part inventory allocation for the case company?

To answer the MRQ, one needs to understand what spare parts should be stocked and how many need to be stocked to ensure availability. Moreover, with a global distribution network there is also the question of where spare parts should be stored.

Hence, the following sub research questions (SRQ) will be answered:

SRQ1: What is important to consider in stocking decisions for spare parts?

SRQ2: What is the future, forecasted demand for spare parts?

SRQ3: Where should spare parts be stocked to minimize shipping costs?

1.6 Delimitation

It was decided to delimit the study to one business area, namely the High Flex assortment. The High-Flex assortment was chosen due to it being the business area with the largest volume of spare parts. The larger volume and more diverse assortment were considered to be of most interest and relevance for the study. Further, the study was delimited to only consider non-repairable items. The intention of the study was to focus on inventory allocation and distribution, hence, repairable items were excluded since studying these items require more focus on service aspects than distribution aspects. Since the assortment of spare parts changes over time with new machines being introduced and older machines reaching end-of-life, the study focus on the last three years assortment. Lastly, as suggested by Huiskonen (2001) spare parts showing a stable demand was excluded in the study since they can be managed like standard parts with standard methods.

(17)

2 Theory of spare part management and distribution

This chapter introduces literature and previous research on spare part management and spare part distribution. First, a general introduction to inventory management is presented. Second, the specific topic of spare part management and its challenges is discussed. The objective and commonly suggested methods for classification of spare parts are described, followed by the characteristics of spare part-demand and an overview of the challenges with forecasting the demand for spare parts and diﬀerent forecasting methods. Third, the distribution process of spare parts and possible distribution network structures are discussed.

2.1 Inventory management

Inventory traditionally refers to stocked goods that will be used in production, sold to customers or used in maintenance activities according to Shenoy and Rosas (2018).

Shenoy and Rosas (2018) conclude four aspects influence inventories; demand, re- plenishment lead time, inventory levels, and item lifetime. They say that inventory management is a challenging task since the demand and lead time might vary, the inventory level might be unknown, and the item lifetime is not always indefinite. Since the characteristics diﬀer from item to item, they suggest that inventory decisions are also needed to oﬀer the right item quantity at the right time(Shenoy & Rosas, 2018).

The first step of inventory management is to decide which articles should be stocked

Inventory management

Spare part management

Spare part demand Classification

of spare parts Demand

forecasting

Distribution Network

Figure 2: Overview of chapter structure.

(18)

where and with what availability, and after that, inventory scheduling can be done in detail (Gudehus & Kotzab, 2012).

Inventory cost

The economic cost for inventories is generally assumed to be the sum of inventory holding costs, ordering costs, and penalty costs for stock-outs (Hu et al., 2018). Penalty cost, also known as shortage cost, is defined as the cost occurring when an organiza- tion is unable to meet demand (Shenoy & Rosas, 2018). That may be the cost of a lost profit or the additional cost related to meet the demand by placing an extra order or obtaining the item from a diﬀerent supplier (Shenoy & Rosas, 2018). The price of a spare part is usually lower than the shortage cost of the same spare part (Teixeira et al., 2018). To decrease the risk of shortage costs due to stock outs, organizations may decide to always have some items in inventory known as safety stock (Shenoy

& Rosas, 2018; Gudehus & Kotzab, 2012). However, inventory means tied capital, and therefore management often requests lower inventory and higher turnover rate (Gudehus & Kotzab, 2012).

The inventory turnover rate is aﬀected negatively by demand uncertainty (Hançer- lioğulları et al., 2016), which is typical for spare parts. Also, the selection of spare parts is often large and heterogeneous, which makes inventory turnover a misleading indicator (Gudehus & Kotzab, 2012). Causes of poor inventory management are, for example, insuﬃcient decision of what items to keep in stock, deficient or non-existent forecasting, and shared responsibility for inventory decisions (Gudehus & Kotzab, 2012).

The eﬀect of demand forecasts on inventory management

Creating a system with high availability and low inventory is the main challenge for inventory management (Hu et al., 2018). Because shortage costs for spare parts usually are high, demand forecasts are essential to set an optimal level of spare part safety stock (Zhu et al., 2020). However, achieving perfect stock control is not possible in practice as long as inventory models rely on completely accurate demand forecasts (Prak & Teunter, 2019). Demand forecasts are only probability distributions and will therefore always hold some level of uncertainty (Prak & Teunter, 2019; Cohen et al., 2006). Any inventory model that ignores the forecast uncertainties results in non-optimal safety stocks, causing stock-outs, high costs, and impaired service level (Prak & Teunter, 2019). Cohen et al. (2006) argue that demand forecasts of spare parts should, unlike standard products, be used to mitigate the risk of stock-outs, and not as an exact reality of demand.

Insuﬃcient demand forecasting leads to an increased risk for the inventory to become obsolete (Teunter et al., 2011). Obsolescence is challenging to handle for slow-moving items (Kennedy et al., 2002). Cobbaert and Van Oudheusden (1996) examined dif-

(19)

ferent scenarios with obsolescence risk and inventory costs, and found that ignoring an obsolescence risk at 20% may increase the average cost with 15%. By comparison, Molenaers et al. (2012) found that 54% of the stocked parts in a large petrochem- ical company had not moved the last five years. Another example, in Sweden, a company which had items in stock that had not shown any demand in the last ten years (Syntetos et al., 2009). A useful method to identify obsolete parts is the FSN analysis (Cavalieri et al., 2008), dividing items based on their moving rate (Shenoy

& Rosas, 2018; Teixeira et al., 2018). The FSN method consists of the three categories Fast-moving (F), Slow-moving (S), and Non-moving (N) (Teixeira et al., 2018).

Non-moving items might already have become obsolete, and slow-moving items risk becoming obsolete if not appropriately managed.

2.2 Spare part management

The dominant objective in spare part management literature is to maximize availability while minimizing the economic cost (Hu et al., 2018). Availability refers to the percentage of time a machine is not operational due to a lack of spare parts (Rusten- burg et al., 2001). Spare parts are closely related to maintenance, and because of that, high responsiveness from the spare part supplier is required to avoid downtime cost for the customers (Hu et al., 2018; Cohen et al., 2006; Huiskonen, 2001). Thus, eﬃcient handling of spare parts is essential for providing customer value.

Characteristics of spare part management

Spare part management diﬀers from managing standard parts in several ways and is a notably tricky task to handle eﬃciently for several reasons (Cohen et al., 2006; Altekin et al., 2017; Huiskonen, 2001; Molenaers et al., 2012). Spare part demand emerges from maintenance activities, either predictive or corrective maintenance (Turrini &

Meissner, 2019; Hu et al., 2018). When repairing a machine, the option of repairing or replacing the faulty part aﬀects the stock level, resulting in irregular demand for spare parts (Turrini & Meissner, 2019; Hu et al., 2018; Kennedy et al., 2002). This makes the forecasting diﬃcult (Hu et al., 2018; Rego & Mesquita, 2015), and to avoid stock-outs of critical parts, an amount of safety stock is required (Huiskonen, 2001).

However, calculating the shortage cost for spare parts is diﬃcult due to the need to include production loss (Kennedy et al., 2002). Thus, keeping a reasonable level of inventory is diﬃcult, partly because of the irregular demand and partly because of the large number and great variety of spare parts to manage (Hu et al., 2018; Cohen et al., 2006; Huiskonen, 2001). Besides, spare parts are usually expensive, and so holding costs associated with spare part inventories can be high (Turrini & Meissner, 2019; Huiskonen, 2001).

There is also a risk for the stock to become obsolete as machines reaches their end-of- life and there are still spare parts in stock (Kennedy et al., 2002; Cohen et al., 2006;

Hu et al., 2018). Thus the management of spare parts is further complicated, and as

(20)

a result, decisions regarding which spare parts to keep in stock and at what quantity are in practice often based on intuition rather than on an accurate decision analysis (Rustenburg et al., 2001). OEMs, in general, have a turnover rate for spare parts of one to two times a year, evidence for the potential of improvements (Cohen et al., 2006).

Although scholars acknowledge the close interrelations between the classification of parts, demand forecasting and inventory management, few have taken on an integrated approach combining the three of them in spare part management (Bacchetti

& Saccani, 2012; Duchessi et al., 1988; Turrini & Meissner, 2019). Decisions regarding inventory policies can and should be adopted based on an item’s classification category and forecasted demand to achieve eﬃcient spare part management (Bacchetti &

Saccani, 2012).

2.3 Classification of spare parts

Classification of parts and products into categories can be carried out for diﬀerent purposes. For spare part management, classification is an essential part of both inventory control and demand forecasting (Hu et al., 2018; Cavalieri et al., 2008).

Eﬃcient spare part management requires inventory management to be adapted to the specific characteristics of individual parts. Deciding on a suitable stocking strategy for each spare part is a diﬃcult task due to usually large and heterogenous assortment of spare parts, and so classification is a useful approach to group parts with similar characteristics into categories (Hu et al., 2018; Teixeira et al., 2018; Cavalieri et al., 2008). Several approaches to group spare parts have been suggested in the literature and used in practice. Most often, it is based on criticality (Bacchetti & Saccani, 2012; Hu et al., 2018), cost (Bacchetti & Saccani, 2012), or lead time (Hu et al., 2018). Teixeira et al. (2018) suggest that classification should include both logistics and maintenance aspects.

ABC Classification

ABC classification is the most common method used in practice (Shenoy & Rosas, 2018; Huiskonen, 2001) and proposed in theory (Bacchetti & Saccani, 2012). It has become popular due to its simplicity to understand and implement (Bacchetti &

Saccani, 2012; Huiskonen, 2001). It is based on the Pareto principle and formerly classified items into three categories (A, B, and C) depending on their value (Hu et al., 2018; Shenoy & Rosas, 2018). The most valuable items, usually items accounting for 80% of inventory value but only 20% of the quantity, are classified as A-items (Shenoy & Rosas, 2018). C-classified items are the items with the least value, around 5% of inventory, but the most substantial quantity, usually 50% (Shenoy & Rosas, 2018). The B-classified items account for the remaining 15% value and 30% of the quantity. The highest control level is recommended for class A, while simple control is enough for class C (Shenoy & Rosas, 2018).

(21)

The ABC-method has been modified in multiple ways by both scholars and practitioners to better suit spare part management (Bacchetti & Saccani, 2012). A simple modification is to use a diﬀerent guiding criterion than value, for instance, criticality (Hu et al., 2018; Bacchetti & Saccani, 2012) or lead time (Hu et al., 2018). While item value and lead-time are distinct values, criticality can be measured in multiple ways. It is often related to the impact of a machine breakdown (e.g. Teixeira et al., 2018; Stoll et al., 2015; Flores et al., 1992) but also the possibility to substitute a failed item with a similar if the assigned spare part is not available (Flores & Why- bark, 1987), the failure rate of an item (Stoll et al., 2015; Duchessi et al., 1988) or the time it takes to replace a failed item (Stoll et al., 2015; Porras & Dekker, 2008;

Duchessi et al., 1988).

Criticality based classification

Huiskonen (2001) diﬀerentiates between control criticality and process criticality. He relates control criticality to the possibilities to control the situation of a failed item and process criticality to the consequences caused by a broken item. Control criticality includes measures like lead time, the number of potential suppliers of a spare part, and the predictability of failure (Huiskonen, 2001). A practical approach for assessing process criticality is to determine the impact of a broken item to one of the following three degrees: 1) the failure needs to be fixed immediately, 2) the failure can be fixed temporarily until the correct spare part is available, 3) the machine can function without replacing the failed item (Huiskonen, 2001). A similar approach is the VED-classification that categorizes items as vital, essential, or desirable based on a qualitative criticality analysis by experts (Shenoy & Rosas, 2018; Cavalieri et al., 2008). As the VED approach is based on subjective judgment (Bacchetti & Saccani, 2012), the accuracy of the classification is dependent on the manager conducting the classification (Cavalieri et al., 2008). To overcome the drawback of subjective judgment, VED can be combined with other systematic approaches (Cavalieri et al., 2008).

A simple VED classification is to assign items to their class based on production loss (Shenoy & Rosas, 2018). For instance, spare parts that are required for a machine to function are classified as Vital items, and spare parts that are not necessary for a machine to function are considered as Desirable items. Spare parts that are not as critical as Vital items, but are not considered Desirable, are classified as Essential items. When classifying spare parts based on criticality, Huiskonen (2001) suggests to only use the first and second degree of process criticality (i.e., vital and essential items), since the third degree (i.e., desirable items) refers to non-critical parts that therefore can be managed like standard parts.

(22)

Multicriteria classification

The complex characteristics of spare part management often requires multiple criteria to be considered in the spare part classification (Roda et al., 2014; Huiskonen, 2001).

Beyond using other guiding criteria than item value, like lead time or criticality, scholars have suggested diﬀerent models that can extend ABC classification to include more than a single criterion (Bacchetti & Saccani, 2012). An overview of suggested criteria to include in such models are presented in table 1.

Multicriteria classification models are among several matrix models, weighted linear optimization, and artificial neural networks (Hu et al., 2018; Bacchetti & Saccani, 2012). However, most multiple criteria item classification models have not been mentioned in more than one paper and also lack empirical study (Hu et al., 2018). An exception is the analytical hierarchy process (AHP), proposed by e.g., Gajpal et al.

(1994) and Molenaers et al. (2012).

AHP is a systematic method used for multi-attribute decision-making, that pairwise compares alternatives to assign them relative weights (Tzeng & Huang, 2011; Gajpal et al., 1994). Decision making is often made intuitively for single-criterion problems, but for multicriteria problems, decision making is more diﬃcult (Tzeng & Huang, 2011). AHP has been widely used in corporate planning and analysis of benefit versus cost (Tzeng & Huang, 2011). The method structures the decision hierarchical

Table 1: Summary of identified parameters in literature

Williams(1984) Gajpaletal.(1994) FloresandWhybark(1987) Ramanathan(2006) PartoviandBurton(1993) PerssonandSaccani(2009) Huiskonen(2001) Duchessietal.(1988) PorrasandDekker(2008) Bragliaetal.(2004) Molenaersetal.(2012) Syntetosetal.(2009) Babaietal.(2015) Jounietal.(2011) Huetal.(2018) Bacchettietal.(2013) Baykasoğluetal.(2016) Stolletal.(2015) Teixeiraetal.(2018)

Criticality X X X X X X X X X X X X X

Impact of failure X X X X X X X X

Downtime cost X X

Safety X

Political consequences X

Control criticality X X

Lead time X X X X X X X X X X X X

Specificity X X X X

Part cost X X X X X X X X X X

Annual usage value X X X X X X X

Probability of item failure X X X X X

Availability X X X X X X

Demand size variation X X X X X X

Demand frequency X X X X X X X

Life-cycle phase X X

Obsolescence risk X X

Number of installed units X

(23)

Figure 3: Decision hierarchy for spare part management proposed by Gajpal et al.

(1994).

with the overall objective at the top, the criteria aﬀecting the objective in the middle, and the alternatives for diﬀerent criteria in the bottom (Gajpal et al., 1994). AHP consists of three steps: 1) defining the decision criteria, 2) weighting the criteria, and 3) calculate the priority of the criteria (Bevilacqua & Braglia, 2000).

Gajpal et al. (1994) were the first to apply AHP for spare part management and did so by combining it with a VED analysis. They suggested that the criticality of a spare part is dependent on the level of production loss, the lead time and the spare part type (standard or non-standard). An overview of the approach taken by Gajpal et al. (1994) can be seen in figure 3. Later, Braglia et al. (2004) constructed a decision diagram for spare part classification based on VED analysis and AHP considering criticality and logistics parameters. Molenaers et al. (2012) then used the decision diagram model by Braglia et al. (2004) together with the criticality definition proposed by Huiskonen (2001), namely to distinguish between control criticality and process criticality.

(24)

2.4 Spare part demand

Demand can be classified as deterministic, constant, varying, or irregular (Shenoy &

Rosas, 2018). When the demand is irregular, demand forecasting is particularly diﬃ- cult because the demand changes with unknown variations between periods (Shenoy

& Rosas, 2018; Chopra & Meindl, 2016). Irregular demand can be further categorized into sub-classes. If the demand infrequently occurs, i.e., there are extended periods without any demand at all, the demand is said to be intermittent (Turrini & Meissner, 2019; Cavalieri et al., 2008). If the demand, when it occurs, varies greatly in size, it is called erratic (Turrini & Meissner, 2019; Cavalieri et al., 2008). If demand is both intermittent and erratic, it is called lumpy (Turrini & Meissner, 2019; Cavalieri et al., 2008). Another term frequently used in literature is slow-moving demand. Items with low average demand, irrespective of it being intermittent, erratic, or lumpy, are called slow-moving (Boylan et al., 2008; Syntetos et al., 2005). Although irregular demand, all types of it, is challenging to forecast, it might be possible to find a probability distribution when analyzing historical data (Shenoy & Rosas, 2018).

Due to the demand for spare parts often being intermittent or lumpy, the demand forecasting is made diﬃcult (Hu et al., 2018; Rego & Mesquita, 2015). Reasons behind the sporadic demand is that the need for spare parts are eminently related to installed machines (Cavalieri et al., 2008), maintenance for installed machines (Turrini & Meissner, 2019; Bacchetti & Saccani, 2012) and machine failure (Turrini &

Meissner, 2019; Hu et al., 2018). Thus the demand for spare parts is related to the size and status of the installed machine base (Van der Auweraer et al., 2019). Maintenance could be either preventive or corrective, i.e., the aim is to prevent machine failure or to correct an already failed machine (Molenaers et al., 2012). As it is impossible to anticipate a machine breakdown, the demand for spare parts is highly unpredictable.

The life cycle phase of products has been highlighted as an essential aspect to consider in spare part management, as diﬀerent actions are needed in the diﬀerent phases of the life cycle (Hu et al., 2018; Dekker et al., 2013). Dekker et al. (2013) highlighted this from a demand forecasting perspective. They described that the demand for spare parts would follow the demand for the initially sold product, but with a slight delay, as shown in figure 4. While most products are sold as the product moves from the initial phase to the mature phase, the demand for spare parts will peak during the mature phase as the products are being used, and deterioration occurs. As products reach end-of-life, the initial product might no longer be produced, but the demand for spare parts remains, and service periods may still be ongoing (Dekker et al., 2013).

For this phase, which can go on for decades, it becomes critical to make a plan for the demand as the production of spare parts might conclude (Dekker et al., 2013).

Hu et al. (2018) also took on the equipment life-cycle phase perspective for spare part management. They discuss equipment life cycle processes and disciplines from operational research, which are needed to support the spare part management at each phase, in which demand forecasting is an essential part together with multicriteria

(25)

Figure 4: Size of installed base and spare part demand throughout the product life cycle (Dekker et al., 2013).

classification, simulation, and optimization. Hu et al. (2018) identified four phases, namely pre-life, initial procurement, regular operation, and finally, end-of-life. They highlighted the need to adapt the demand forecasting depending on which phase one is in, similar to the findings of Dekker et al. (2013). Both studies also described the importance of looking into the use of returned parts along with remanufacturing and refurbishment to fulfill the last demands during the end-of-life phase.

2.5 Demand forecasting for spare parts

Demand forecasting is a fundamental tool for supply chain decisions (Chopra &

Meindl, 2016). Standard forecasting methods, developed for deterministic, constant or varying demand, tend to cope poorly when periods with zero demand occurs (van Wingerden et al., 2014). However, several modifications of standard methods and newly developed methods have been suggested for the demand forecasting of spare parts (Hu et al., 2018). In their review, Hu et al. (2018) categorize forecasting methods of spare parts into three diﬀerent groups: judgmentally, reliability, and time-series based.

Judgmentally based forecasting

Judgmentally based forecasting refers to when demand planners at companies make adjustments to initial forecasts made from a computerized forecasting system, using a quantitative method (Fildes et al., 2009). The adjustments can be used to inte- grate available qualitative information, which is not considered in the quantitative forecast e.g., regarding internal or external changes in the environment (Hu et al., 2018). It is of frequent occurrence at companies (Hu et al., 2018; Fildes et al., 2009;

(26)

Syntetos et al., 2009), and is often done to incorporate the managers’ experiences on the demand of spare parts into the forecast (Hu et al., 2018). Judgmentally based forecasts can also be applied in cases when there is little or no data available regarding a part/product, such as when it is in the early stages of its life-cycle (Hu et al., 2018). Then, forecasts can be exclusively judgmental, as the application of quantitative methods often require historical data that does not exist yet (Hu et al., 2018).

The accuracy and eﬀects of judgmental adjustments to quantitative forecasts have been questioned. Even though it is common practice, the academic literature on the topic is very sparse (Syntetos et al., 2016). Fildes et al. (2009) found that judgmental adjustments in general increase the accuracy of the forecast. This was especially true for significant adjustments, while smaller adjustments more often harm the accuracy (Fildes et al., 2009). They also identified a general bias towards positive adjustments and optimism, and that these positive adjustments were less likely to increase the accuracy than negative adjustments (Fildes et al., 2009). A parallel study by Syntetos et al. (2009) also found evidence that judgmental adjustment could be useful for forecasting of demand, yet in the context of intermittent demand. They also found that negative adjustments were more likely to increase the accuracy than positive ones (Syntetos et al., 2009). They also recognized a lack of learning eﬀects regarding the adjustments; they did not tend to become better over time (Syntetos et al., 2009).

In contrast to the findings by Fildes et al. (2009), Franses and Legerstee (2010) found that judgmentally based forecasts are at best equally good, but more often they perform worse than the model-based forecasts. However, Hu et al. (2018) express the need of more empirical evidence on the subject, especially in the context of slow- moving items, such as for spare parts, since the studies by Fildes et al. (2009) and Syntetos et al. (2009) are both performed at the same company.

Reliability based forecasting

Van der Auweraer et al. (2019, p 181) state that the majority of forecasting techniques for spare parts do not take the "underlying demand-generating factors" into account.

Instead, the forecasting techniques have been focusing on extrapolation methods using historical data, which is considered to be a significant disadvantage (Van der Auweraer et al., 2019; Dekker et al., 2013). In contrast to this, reliability-based forecasting has been presented. In reliability-based forecasting, the forecast depends on the product’s reliability and maintenance characteristics (Van der Auweraer et al., 2019; Hu et al., 2018). Ghodrati and Kumar (2005) state that it is the optimal way to prevent unplanned downtime, especially for manufacturers that develop support packages. As with judgmentally based forecasting, the model can be applied without historical data of the demand; however, it does require the reliability and maintenance variables (Hu et al., 2018).

Van der Auweraer et al. (2019) took on a perspective of installed base information

(27)

in forecasting. They identified three different main drivers for spare part demand, the first one being the size of the installed base along with its status and the status of the spare part. The second identified driver was the maintenance policy (Van der Auweraer et al., 2019). Maintenance policies influence the demand for spare parts as they determine when parts are to be replaced, either if it is corrective or preventive maintenance. This was also investigated by Ghobbar and Friend (2003), who found that the inclusion of the primary maintenance process (PMP) had a significant effect on the forecasting. They looked into three different PMP’s namely hard-time, on- condition, and condition monitoring, the first two being preventive and the third corrective. They found that the most crucial PMP for forecasting was the hard-time PMP, in which the replacement rate of a part is based on the deterioration rate in terms of system utilization.

Another factor that has shown a significant impact on forecasting performance is the operating environment (Ghodrati et al., 2013; Barabadi et al., 2012). This was identified as the third main driver for demand for spare parts (Van der Auweraer et al., 2019). Depending on the diﬀerences in the operational environment, for example, in terms of humidity and temperature dust, there will be diﬀerences in system behaviors (Ghodrati & Kumar, 2005). This is important as it has a significant impact on reliability characteristics, such as failure rate (Ghodrati et al., 2013). Barabadi et al.

(2012), Ghodrati et al. (2013), and Ghodrati and Kumar (2005) all present models which take the operational environment into account.

As the need for spare parts often stems from the breakdown of parts or maintenance processes, the inclusion of such aspects in the forecast seems very beneficial. Yet, time-series forecasting methods remains among the most commonly used techniques.

This could be due to the added complexity which the inclusion of reliability and maintenance characteristics entail. As the adoption of special forecasting techniques for spare parts is already low (Bacchetti & Saccani, 2012), too complex models might be even more challenging to implement.

One could also question the applicability of reliability-based forecasting for OEM’s.

The authors mentioned above have all had the perspective of maintenance providers, often within an operating company. Data that they have used, such as operating environment and maintenance processes, might not be available for the OEM. Dekker et al. (2013) found that setting up forecasting based on installed information data is very challenging. Managing the data has proven difficult due to the large sizes of the installed base, the wide variety of products, and because data becomes obsolete and scattered across systems (Dekker et al., 2013). Also, if the data was available to OEM’s, the operating environments and maintenance processes among different customers would likely differ among the installed base, hence increasing the complexity of the forecasting method even further as multiple conditions would need to be included.

(28)

Time-series based forecasting

Time-series based forecasting assumes that future demand can be predicted based on historical demand (Hu et al., 2018). The method is mostly used in practice when extensive historical data can be obtained (Hu et al., 2018). However, historical data is usually limited for slow-moving items like spare parts (Cavalieri et al., 2008). Time- series based forecasting stems from traditional forecasting methods of items with a regular demand, considered to be the standard methods of demand forecasting (Hu et al., 2018). To better suit spare part management, time-series based methods have been adapted to better cope with demand intermittence (Hu et al., 2018).

Hu et al. (2018) identify three approaches to time-series based forecasting: bootstrapping, neural networks, and parametric methods. Literature on the two former is scarce, while parametric methods have been suggested more widely (Hu et al., 2018).

Parametric methods require an assumption on the demand distribution (Cavalieri et al., 2008). That assumption can be avoided by using bootstrapping (Cavalieri et al., 2008); however, it is arguable if using bootstrapping instead of a parametric method is worth the added complexity (Hu et al., 2018). Regarding neural networks, Bacchetti and Saccani (2012) report that they tend to perform well unless there is a significant decrease in the average demand. Further, they conclude that there is still no unanimous among researchers about the best time-series based forecasting method for spare parts and, especially, the proposed methods often lack concern about the imple- mentation of the method in practice. Bacchetti and Saccani (2012) found traditional time-series based forecasting to be the most common method among practitioners when they studied ten industrial companies.

In the category of parametric time-series based forecasting falls the Croston method, a seminal work for intermittent demand forecasting (Bacchetti & Saccani, 2012).

Croston (1972) suggested exponential smoothing to be used for the forecasting of intermittent demand, but to do so by handling demand frequency and demand size separately. The Croston method is widely used in practice and implemented in ERP systems like SAP (Teunter et al., 2011). Syntetos and Boylan (2001 cited in Synte- tos et al., 2005) found Croston’s method to be biased. They proposed an improved method, known as the Syntetos Boylan Approximation or SBA, including a deflating factor (Syntetos et al., 2005). Teunter, Syntetos, and Babai (2011) then further developed the Croston’s method, known as the TSB method. While Croston’s method updates the demand interval, the TSB method instead updates the demand probability (Teunter et al., 2011). As the demand probability is the inverse of the demand interval, it seems to be a small update; however, the demand interval cannot be up- dated if the demand is zero, which is typical for intermittent demand (Teunter et al., 2011).

van Wingerden et al. (2014) compared diﬀerent forecasting methods using spare part data from three industrial companies. They found the SBA to perform best on most parts, especially compared to Croston’s method and TSB. An exception was

(29)

parts with high infrequency of demand but low variance in demand size for which a bootstrapping method, called the empirical method, performed best (van Wingerden et al., 2014). Spare parts of this character are usually more expensive than the average spare part since if the demand for a costly part occurred regularly, an investigation and probably a redesign of that part would be done (van Wingerden et al., 2014).

Syntetos et al. (2015) also found simple parametric methods to perform equally good or better than bootstrapping for intermittent demand.

Statistical distributions

When applying parametric forecasting methods, it is central to find a fitting distribution for the demand (Turrini & Meissner, 2019). For items with regular demand, the normal distribution often proves to be adequate, but for items with intermittent demand (such as spare parts) it is insuﬃcient (Syntetos et al., 2011) Instead, the most fitting distribution depends on parameters related to the variation demand size, its squared coeﬃcient, and the mean inter-demand length (Turrini & Meissner, 2019).

In a recent study, Turrini and Meissner (2019) analyzed what distributions best would fit the demand for spare parts. They used eight diﬀerent distributions and compared them with two empirical datasets and found that the stuttering Poisson and negative binomial distribution performed best throughout all of their demand fitting tests (Turrini & Meissner, 2019). Similar results were also found by Syntetos et al. (2011), but they also found (regular) Poisson to yield a good fit to demand data. Mean- while, Lengu et al. (2014) only investigated diﬀerent compound Poisson distributions, namely Poisson-Pascal, Poisson-Poisson, Poisson-Log Series, and Poisson-Geometric, and all of them were deemed to provide a good fit for their spare part demands.

However, Turrini and Meissner (2019) also found that while the mentioned distributions might fit the empirical data the best, they may not yield the best result when applied in inventory management. For instance, when applied together with diﬀerent ordering policies, the negative binomial distribution achieved a level of service closest to the targeted level, but by having a much higher holding cost (Turrini & Meissner, 2019). Meanwhile, distributions with lesser fit to demand, such as the normal and gamma distribution, performed surprisingly well in those applications, which implies a need for further research (Turrini & Meissner, 2019). Lengu et al. (2014) shared similar conclusions regarding their research on demand fitting, that their work still needs to be tested in terms of its eﬀectiveness on stock control. They also high- light the need for a hierarchical list with criteria that should govern the selection of distributions for demand modeling (Lengu et al., 2014). However, if models are to be applied in practical settings, Lengu et al. (2014) state that probability distributions need to be easy to implement in Microsoft Excel, thus; distributions with many parameters should be avoided.

(30)

2.6 Distribution of spare parts

The distribution process is the movement and storage of the product between facilities along the supply chain, from the supplier to the customer (Chopra & Meindl, 2016). This stands in contrast to the supply chain itself, which includes all of the parties and functions involved in fulfilling the customer request (Chopra & Meindl, 2016). Instead, when discussing the distribution network, the primary focus is on the movements from the manufacturer to the customer or demand points, the end of the supply chain network (Govindan et al., 2017; Ambrosino & Grazia Scutellà, 2005). As the distribution has a direct impact on customer value and supply chain cost, it can be considered a crucial contributor to company profitability (Chopra & Meindl, 2016;

Mangiaracina et al., 2015). Tabrizi and Razmi (2013, p 1) even state that "A capable DND [distribution network design] can guarantee the success of the entire network performance." Due to the role of distribution for the overall cost and profitability, the distribution network design is of utmost importance (Sadeghi & Nookabadi, 2017).

Distribution network design

Distribution network design is about determining the proper distribution structure in terms of echelons, the amount of facilities to use for production and storage, their capacities, and locations while minimizing the overall cost (Ambrosino & Grazia Scutellà, 2005). However, this is a very complex task as there are many design choices and a vast amount of possible network configurations (Mangiaracina et al., 2015). Therefore, distribution network design is often divided into diﬀerent phases (Chopra & Meindl, 2016; Mangiaracina et al., 2015). The first phase can be described as broad and often includes decisions regarding the numbers of echelons and the roles of each stage, identifying possible structures (Mangiaracina et al., 2015; Chopra &

Meindl, 2016). The design choices become more specific in the second stage, using the scenarios developed in the initial step to determine the specific locations of facilities, their capacities, demand allocation, etc. (Chopra & Meindl, 2016; Mangiaracina et al., 2015). The second stage often entails a more quantitative approach, where a mathematical model is applied to minimize, for example, the cost of transportation (Mangiaracina et al., 2015).

When analyzing different distribution network designs, Chopra and Meindl (2016) explain that structures should be evaluated along two dimensions, namely of what value the design provides for the customer versus the cost of meeting the customer’s requirements and needs. They specified this into a set of service and cost factors which should be used as primary measures in the evaluation of different distribution network design. Six factors were identified that affect the customer value and depend on the distribution network design; response time, product variety, product availability, customer experience, time to market, order visibility and returnability (Chopra &

Meindl, 2016). The identified cost factors were inventory, transportation, facilities, handling, and information (Chopra & Meindl, 2016).

(31)

Chopra and Meindl (2016) presented six distinct design options for distribution networks, options which differ in the amount and types of locations the product passes through from manufacturer to customer, and in the way the products are delivered to customers. Their first presented option was manufacturer storage with direct shipping, which was followed by the addition of an in-transit merge. They then add distributor storage as an option and include different kinds of delivery options, and finally, retailer storage. The different design options are then evaluated according to the identified cost and service factors(Chopra & Meindl, 2016). Depending on which factors have been identified as relevant to a company, these design options can serve as guidelines on how the distribution network should be set up.

The conceptual framework presented by Chopra and Meindl (2016) is useful for the first phase of the distribution network design, as it provides useful measures that could guide the design in terms of the number of layers etc. Nevertheless, it does not help in specifying the exact number of facilities or locations and is thus not as helpful for the second phase of distribution network design. To determine those specifics, as well as transportation and inventory policies within the distribution network, literature mainly proposes two kinds of methods, namely analytical and dynamic simulations (Timperio et al., 2019). Analytical approaches involve modeling the decision problems with quantitative methods by using techniques such as multicriteria decision- making, network optimization, and mathematical modeling (Timperio et al., 2019).

Whereas dynamic simulations refer to computer simulations, in which supply chain configurations can be evaluated in a virtual environment (Timperio et al., 2019).

Mangiaracina et al. (2015) showed that previous literature on distribution network design, mostly presents analytical methods. The most common approach is to find an optimal solution by applying a single objective function with the aim of cost minimization (Timperio et al., 2019; Mangiaracina et al., 2015). The costs to be minimized refers to the total distribution cost, which stems from the cost factors such as transportation, inventory, and facility locations (Mangiaracina et al., 2015).

However,Mangiaracina et al. (2015) stressed the drawbacks of a single objective function with the goal of cost minimization, the most prominent being that it does not consider other objectives of interest and the conflicts between them. When the objective is to minimize costs, they exemplified that it can often urge the centralization of inventories, not taking into account how it could impair the service level. To surpass this, single-objective functions can be combined into a multi-objective function (Mangiaracina et al., 2015); for instance, by expressing the functions as constraints (Savic, 2002). In multiple objective models, the general approach was to find the best settlement between cost and service level, thus evaluating both factors, and multiple options can be proposed (Mangiaracina et al., 2015). However, the abilities of network optimization methods to model real-life complexity is still minimal due to their demand for computation time (Timperio et al., 2019).

When discussing mathematical models to solve decision problems in distribution network design, some general drawbacks have been presented. Drawbacks are that they

(32)

often include multiple assumptions, do not take enough factors into account, and are lacking qualitative input (Mangiaracina et al., 2015). To surpass these issues, Tim- perio et al. (2019) suggest an integration of multiple techniques, such as optimization together multicriteria decision-making in a multi-stage approach, thus including both optimization and qualitative variables. Also, as mentioned previously, there are high demands on the responsiveness of spare part distributors (Murthy et al., 2004). Thus, it might prove of great importance when applying distribution network design models for spare part distributors. However, in optimization models, these aspects are seldom the primary concern.

Multi-echelon structures

As mentioned previously, one of the critical decisions when designing the distribution network is regarding the number of echelons between the manufacturer and the customer (Chopra & Meindl, 2016). Every layer in the system can be seen as a process, such as manufacturing, procurement, or transportation of products (Eruguz et al., 2016). With global markets, the use of multi-echelon structures has become common for distribution (Tsao et al., 2016). Companies that distribute over large geographical areas often employ a multi-echelon structure by having central warehouses located close to a production facility and local warehouses closer to the customer (Axsäter, 2015)). However, there are multiple types of multi-echelon systems. Depending on their network structure, Eruguz et al. (2016), identified five categories (see figure 5), namely serial (a), assembly (b), distribution (c), general acyclic (d) and general cyclic (e).

For OEM’s, it is common to employ a multi-echelon structure such as displayed in figure 5, for the distribution of spare parts (van Houtum & Kranenburg, 2015). With such a distribution structure, the local distribution centers make it possible to keep a high service level at the local markets (Axsäter, 2015). Another benefit with this kind of distribution structure is that it allows for pooling of stock, i.e., that the central stock can be used distributed to any of the local warehouses (Axsäter, 2015)).

The pooling eﬀect is especially significant for the distribution of spare parts as they display uncertain demand and safety stocks are needed to avoid stock-outs, the more the demand can be grouped together (pooling) the lesser the safety stock needs to be (van Houtum & Kranenburg, 2015).

In multi-echelon systems, facilities at each stage can be considered as locations for stock holding and thus complicating the act of inventory optimization further (Eruguz et al., 2016). It becomes a complex task as inventory optimization in multi-echelon systems aim to minimize the total cost of holding inventory across the supply chain while maintaining the targeted service level, and the optimization models have to consider how inventory decision at each echelon could aﬀect the other while including the often non-linear functions for service level (Eruguz et al., 2016). As multi-echelon systems are so complex, the optimization becomes a computational challenge, often making simplifications necessary.

(33)

Figure 5: Multi-echelon network structures (Eruguz et al., 2016).

Lateral Transshipments

In multi-echelon distribution networks, there is also the possibility to move products between facilities on the same echelon, making so-called lateral transshipments (Patriarca et al., 2016; Paterson et al., 2011) as shown in figure 7. This can prove especially useful when there are local shortages, and another local distribution center is closer than the central warehouse, thus being able to supply the shortage quicker (Patriarca et al., 2016). Lateral transshipments are also a mean to enable pooling of stock, as several distributors can use local stock on the same echelon (Axsäter, 2015). Allowing lateral transshipments can increase the flexibility of the distribution structure while reducing inventory stock and cost (Zhong et al., 2018; Patriarca et al., 2016).

Paterson et al. (2011) identified two types of lateral transshipments: proactive and reactive. They refer to proactive transshipments as planned, predetermined activities. Proactive lateral transhipments are used to redistribute stock between local stocking points regularly (Patriarca et al., 2016). By planning and organizing the lateral transshipments in advance, there are possibilities to keep the handling costs low, according to Paterson et al. (2011). The opposite, reactive lateral transshipments, they refer to as a mean to respond to local stock-outs (as described earlier in this section). They suggest reactive lateral transhipments are especially suitable in

(34)

Figure 6: Typical multi-echelon structure for distribution of spare parts, adapted from van Houtum and Kranenburg (2015)

contexts where holding costs and costs of stock-outs are high, such as for spare part distribution.

Models that take lateral transshipments into account are often challenging to use and handle (Axsäter, 2015). Axsäter (1990) created a two-echelon inventory model that supports emergency lateral transshipments for repairable items. Several other models have then been proposed and a majority of them are based on that model (van Houtum & Kranenburg, 2015). Models that takes both lateral transshipments and

Figure 7: Example of lateral transshipments.

Availability vs. Cost Efficiency

Availability vs. Cost Efficiency

A Case Study Taking on an Integrated Approach to Spare Part Distribution in the High-Tech

Industry

EMMA BOSTRÖM JULIA LUNDELL

Availability vs. Cost Efficiency

A Case Study Taking on an Integrated Approach to Spare Part Distribution in the High-Tech Industry

by

Emma Boström Julia Lundell

Tillgänglighet kontra kostnadseffektivitet

En fallstudie om strategisk integrering av

reservdelsdistribution inom högteknologisk industri

Emma Boström Julia Lundell

Contents

List of Tables

List of Figures

Acknowledgement

1 Introduction

1.1 Background

1.2 Problem statement

1.3 Research purpose

1.4 Case company

1.5 Research question

1.6 Delimitation

2 Theory of spare part management and distribution

2.1 Inventory management

2.2 Spare part management

2.3 Classification of spare parts

2.4 Spare part demand

2.5 Demand forecasting for spare parts

2.6 Distribution of spare parts