You are on page 1of 12

European Journal of Operational Research 181 (2007) 704715 www.elsevier.

com/locate/ejor

Production, Manufacturing and Logistics

An assessment of the eect of mass customization on suppliers inventory levels in a JIT supply chain
Henry Aigbedo
*
Department of Decision and Information Sciences, School of Business Administration, Oakland University, Rochester, MI 48309-4401, USA Received 8 December 2003; accepted 5 June 2006 Available online 20 September 2006

Abstract In some industries, mass customization requires a supplier to provide an Original Equipment Manufacturer (OEM) with a wide range of variants of a given part. We consider an OEM-parts suppliers system for an automotive supply chain where parts are delivered to the assembly line several times a day in a just-in-time environment. Simulating varying assembly schedule and parts delivery schemes, we assess the eect of mass customization on the level of inventory the supplier needs for each variant in order to prevent stockouts. We nd, among other things, that as the level of mass customization increases, there tends to be an increase in the level of inventory the supplier needs to maintain for each part variant in order to prevent stockouts. Theoretical support is provided for the phenomenon. The presented framework is also useful for evaluating the levels of mass customization that will enable the manufacturer meet customers requirements in a cost eective manner. Furthermore, the study conrms the superiority, in terms of inventory levels, of the minmax over the min sum optimization framework. 2006 Elsevier B.V. All rights reserved.
Keywords: Supply chain management; Manufacturing; Mass customization; Just-in-Time assembly systems; Automotive

1. Introduction Manufacturing and service rms have always adopted strategies that they consider to be vital to success. For example, when Henry Ford introduced the moving assembly line, product standardization was considered critical. This facilitated mass production, which in turn boosted productivity and led to reduced unit costs and better wages for workers. A major characteristic of the automotive indus*

Tel.: +1 248 370 4959; fax: +1 248 370 4275. E-mail address: haigbedo@oakland.edu

try then was vertical integration: all parts required for making cars were manufactured in-house. That period, sometimes referred to as the product-out phase for the automobile, was essentially a producers market: demand was high, competition was low, and all items produced were sold (Harrison, 1992). Over several years, the technologies and processes used for automobile manufacture have become relatively standard throughout the industry. Furthermore, the market has in some respect become saturated, leading to the market-in phase. As is typical with many types of product in this

0377-2217/$ - see front matter 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ejor.2006.06.037

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

705

phase, growth in demand is stimulated by innovations that provide distinctive characteristics, while attempting to contain costs. Automobile rms are responding to this challenge through constant innovations in their products and processes. Quite often, though, due to competitive benchmarking that is quite pervasive in the industry, the edge maintained over other rms tends to be very short-lived, thus leading to a stream of sustained innovations and improvements. This state of aairs has created a need for signicant investments in product development. Firms in the automotive industry are also adopting other strategies that have been successful in other industries. Although the automotive industry is not the most signicant industry in terms of revenue generation in many industrialized economies, it certainly has a very signicant impact on many economies, including those of the United States, Japan, France, and Germany. Most of its impact relates to the fact that it supports many non-manufacturing sectors such as logistics/transportation, software development, and consulting. Our focus in this paper is on mass customization in a just-in-time automotive supply chain. In particular, we seek to examine how mass customization impacts the amount of inventory that needs to be carried by rms that supply many part variants to an assembly line in a just-in-time manufacturing environment. Furthermore, we consider a system where parts are supplied by using a constant order cycle, non-constant order quantity framework. This means that a given part supplier delivers parts to the assembly plant a specied number of times a day as agreed upon with the Original Equipment Manufacturer (OEM). Several issues underscore the importance of this type of research: In view of the repetitive nature of discrete manufacturing in the automotive industry, many automotive rms around the world; including Europe, Japan, and North America; have introduced just-in-time manufacturing principles to their operations. Customers have become more demanding in terms of their expectations from manufacturers. Schwartz (1999) succinctly describes this phenomenon by observing that customers now have the prominent role as they say to the companies You make what we buy as opposed to the scenario in which the manufacturer used to be prominent and would say to the customer, You buy what we make.

The Internet-revolution has produced a more empowered customer, making it easier for him/ her to clearly indicate his/her preferences on several items that make up the product. Firms are becoming more and more aware of the importance of building synergies with their partners in the supply chain in order to reap benets that would otherwise have been dicult to realize by acting alone. Some of the issues outlined above have generated interest in the research community, leading to several published works on just-in-time supply chains. The remaining portion of this article is organized as follows: Section 2 provides some discussion on mass customization and its application in the automotive industry. In Section 3, we proceed to review some research relating to sequence scheduling for just-in-time supply chains. Sections 4 and 5 respectively present the mathematical equations for the study and describes the simulation study. Results and discussions of the simulation study follow in Section 6. Finally, we conclude and discuss extensions in Section 7. 2. Mass customization as a competitive strategy The mass production strategy works well if there is no signicant need for product dierentiation among customers. Mass customization (MC), a strategy that recognizes this need, makes use of mass production techniques to quickly assemble goods and services that are uniquely tailored to the needs of individual customers at prices comparable to mass-produced goods and services (e.g., Pine, 1999). Providing customers with a wide variety of any given product or service has been a characteristic of some manufacturing and service industries for slightly over a decade now. In the service sector, for example, some restaurants compete on the basis of their ability to provide one-of-a-kind meals to their patrons while still attempting to meet cost and timely delivery requirements. In recent times, in a bid to reduce costs of textbooks for students, many publishers now oer instructors and their students the opportunity to customize for their needs, texts that contain case teaching materials, multi-media materials, and selected chapters from various standard textbooks. Apparently, this growing trend in the use of mass customization seems to

706

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

be unabated in spite of the many challenges it poses to rms and their management. Mass customization has particularly been an important phenomenon in some manufacturing industries such as the apparel, electronics, and computer industries, where customers decide the exact features that make up their products. Benetton Company, a manufacturer of clothing, was a pioneer of this concept in the early 1980s in the apparel industry (Andel, 2002). Dell Computer Corporation, a pioneer of this strategy in the computer industry, produces computers to meet the customers requirements for items such as Random Access Memory (RAM), monitor type, keyboard type, zip drive type, audio-player, etc. Dells significant growth over the past 20 years to become one of the most successful business ventures has also been greatly facilitated by the Internet. Two other classic examples of mass customization discussed in the literature include: National Bicycles Industrial Company of Japan (NBIC)s use of mass customization to provide personalized bicycles to its customers (Kotha, 1996); and Hewlett Packards use of mass customization to manufacture and package its Deskjet Printer meant for markets in various countries (Feitzinger and Lee, 1997). Alford et al. (2000) present a framework for customization in the automotive industry that hinges on where the customization starts in the value chain. They dene Core, Optional and Form customization for the cases where customization commences in the design, manufacturing, and distribution stage, respectively. Following Dells success, Ford Motor Company and other automotive companies such as General Motors and BMW are already experimenting on pilot programs that will allow consumers to customize their products while at the same time reducing the lead times for delivering the products to them. For example, Ford Motor Company set up a task force to consider the issue for its supply chain (Austin, 2001). One of the problems identied by the study group relates to the complexity of the automobile as compared to the computer. Forrester Research estimates that build-to-order sales will account for about 20% of all new car sales by the year 2010 (Mello, 2001). Several articles examine various aspects of mass customization (e.g., Gilmore and Pine, 1997; Salvador et al., 2002). While mass customization is a very useful strategy from the rms as well as the customers point of view, it is not viable in all environments. Even

for environments where it can be implemented, managers need to deal with several challenges, including potential proliferation of input parts and components used for production (e.g., Alford et al., 2000; Zipkin, 2001). 3. Review of related research Research relating to automotive assembly systems have been conducted for well over three decades. One of the earliest works was that of Thomopoulos (1967), which addressed the issue of task assignment and balancing among various workstations as well as sequencing multiple product models manufactured on a paced assembly line. The procedure essentially seeks to optimally utilize assembly line operators, by taking into consideration dierent workstation losses, which include idleness and utility work. With the advent of the just-in-time paradigm whose aim is to reduce wastes in manufacturing, inventory reduction became a very important issue considered in determining suitable assembly line sequences. Toyota Motor Corporation pioneered this framework by proposing the Goal Chasing heuristic (Monden, 1983). Considering the total number of product units to be scheduled per day or production shift as the planning horizon, this heuristic seeks to minimize the variance between the actual number of units of part required by the assembly line and the average demand rate on a product-unit-by-product-unit basis. It penalizes shortages and overages thus avoiding excess inventory as well as attempting to prevent stockouts. Toyota has also proposed other variants of this basic procedure, including the goal coordinating method, which incorporates spacing constraints that are meant to address the workload issue for the assembly line (Monden, 1998). There have been several research papers that deal with this inventory related issue from dierent perspectives. Miltenburg (1989) proposed a mathematical programming formulation for the case of smoothing product launch rates and proposed optimal and heuristic solution procedures. Unlike Toyotas framework above which smoothes parts usage directly this latter formulation is considered to smooth parts usage if products require approximately equal number and mix of parts. Kubiak (1993) refers to the problems described by Miltenburgs and Toyotas formulation as the product rate variation smoothing (PRV) and output rate variation smoothing (ORV) problems, respectively.

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

707

Broadly speaking, therefore, published research relating to inventory minimization in just-in-time supply chains have basically been within these two classes. Groein et al. (1989) propose a nal assembly sequencing procedure for smoothing parts from just-in-time feeder shops. Special features of their procedure included the use of a lexicographic min max objective function that dynamically gives priority to parts with the largest variability as well as using local search and pairwise interchanges. Inman and Buln (1991) proposed a very ecient heuristic sequencing procedure for PRV using due-dates derived from the quantity of each product type, and they report that it outperforms the procedures in Miltenburg (1989). Miltenburg et al. (1990) propose a dynamic programming-based procedure for optimally sequencing the assembly line to simultaneously smooth product usage rates and assembly workload. Kubiak and Sethi (1991) provide an assignment formulation for PRV that allows problems of virtually any size to be solved to optimum. Steiner and Yeomans (1993) observe that earlier formulations in the literature, which were based on minimizing the sum of squared deviations produce smooth schedules on the average but that this does not preclude having large deviations at some points in the planning horizon. They therefore propose a minmax type objective function for the problem and provide a graphtheoretic procedure for its optimal solution. Some other articles that address the PRV problem include Ding and Cheng (1993), Ng and Mak (1994), Kubiak and Sethi (1994), and Aigbedo (2000). In addition to Toyotas formulation and heuristic algorithms (Monden, 1983) several articles have addressed the ORV problem. For example, Miltenburg and Sinnamon (1989) formulate the multi-level problem that comprises parts at dierent levels and they provide heuristic solutions. Sumichrast et al. (1992) propose the time-spread method and also present results of a comparative analysis of assembly line sequencing procedures. They conrm that the two-step heuristic, which considers deviation over two-steps in determining a product to schedule at a given scheduling stage, outperforms the one-step procedure. Pleshberger and Hitomi (1993) propose a scheduling procedure for a JIT nal assembly line with the goal of ensuring that costly shortages do not occur. The procedure considers factors such as the deviation of each part from its ideal consumption path, the time remaining before the part will be

replenished, and the monetary value of the parts. Bautista et al. (1996) provide some insights into the ORV problem by examining the forms of the sequences and several mechanisms such as symmetry and rate-preservation that could be exploited for solving it. They also propose a bounded dynamic programming procedure for its optimal solution. Kubiak et al. (1997) formulate the weighted min sum and minmax multi-level scheduling problems and develop a dynamic programming procedure for its optimum solution. They show that the minmax multi-level problem is NP-hard and also that the number of DP states may grow exponentially in the number of product types even though it is polynomial in the total number of product units to be produced over the planning horizon. Other articles that examine the ORV problem include Sumichrast and Clayton (1996), Aigbedo and Monden (1997), and Duplaga and Bragg (1998). Although there have been a number of articles addressing inventory of parts used in a just-in-time manufacturing environment, none has examined the eect which mass customization might have on such a system. In particular, our framework takes into consideration the fact that due to geographical distance between most parts suppliers and the assembly plants, items are supplied in batches, using a constant order cycle, non-constant order quantity method. In other words, the Original Equipment Manufacturer (OEM), that is, the automaker, agrees with the parts supplier, how often parts are to be delivered to the assembly line per day or production shift. 4. Mathematical equations for the study 4.1. Sequence scheduling formulations Various mathematical models have been developed in the literature to determine deviations from level schedules in a just-in-time environment. Our study described in the next section uses a variant of the minsum and minmax formulations provided in Kubiak et al. (1997) for optimum sequences and Toyotas equations for our large scale experimentation. In each case we consider deviations at only one level, the sub-assembly (parts) level. The minsum DP recursive equation of deviation for a DP state Y is cY MinimumcY vi lY 8 feasible i; 1

708

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

where lY
b X j1

kT j =P

X
Yj

!2 dij ; 2 8 i; 3 4

c/ cYy 1 ; y 2 ; . . . ; y a 0 and y i 0 a X jYj y i k;


i1

of parts that needs to be constantly maintained to prevent stockout from occurring. Next, we explain the mathematical form of this metric. Let SF be the supply frequency. Then the average production rate for part j is given by Pj Tj : SF 8

Yj dij is the total number of part type j required to make all product units represented by the DP state, Y. And vi is a unit vector representing the product i launched at a given stage. Tj is the total number of units of part type j required to manufacture all P units of products scheduled for production over the planning horizon. The minmax DP recursive equation of deviation for a DP state Y is

cY minfmaxfcY vi ; ksYkgg 8 feasible i;


i

5 where ) ( X ksYk max kT j =P dij : j Yj

The maximum inventory of part j that needs to be constantly maintained to prevent a stockout condition can be expressed in terms of the dierence between the actual requirement for part j at the assembly line and the average production of part j at the suppliers plant. Let Fjt be the cumulative number of units of part j required at the assembly plant until supply epoch t (1, 2, . . . , SF). Then the maximum inventory of the said part that should be maintained constantly to prevent a stockout, expressed as a percentage of the total number of units of that part required for the day or production shift would be   100 I max max F jt tPj ; j 1; 2; . . . ; b ; j T j t1;2;...SF 9 where b* is the total number of parts variants that are used to meet all customer orders for the day or production shift in question. (b is the total number of parts variants oered by all parts suppliers) and dxe is the smallest integer greater than or equal to x. Expressing this parts shortage as a percentage facilitates aggregation and assessment of dierent parts variants with diering daily (or per shift) requirements. We recognize that since a customer only orders 1 unit (or 1 set) of part variant from each part class, the total daily requirement for a given part variant, Tj, will generally decrease proportionately when number of available variants of that part class increases. We account for this phenomenon in the following expression by normalizing with respect to our base case. Thus, I max Adj j I max j ; L j 1; 2; . . . ; b ; 10

The classical Toyota equation is used and the deviation for a product i at a scheduling stage k is v u b uX V ki t kT j =P C j;k1 dij 2 ; 7
j1

where Cjk is the cumulative number of part type j used until stage k. 4.2. Metrics for evaluating parts supplier inventory We adopt metrics that are particularly suited to the constant order cycle, non-constant order quantity framework that exists between automotive OEMs and their parts suppliers. A number of metrics can be used for evaluating this scenario. For example, Monden (1998) suggests coecient of variation in parts requirement across time windows. OEMs would not like to have stockout condition occurring for any part variant that is used for assembling the cars. Such stockouts are very expensive because of lost production time or cost of premium shipment of parts to the assembly line. In line with this reasoning, in this study, we use a metric that relates to the maximum amount of inventory

where the value L equals 1.0, 1.5, 2, 2.5, and 3.0 are for the base case (0%), 50%, 100%, 150%, and 200% customization, respectively. Therefore, the adjusted average maximum percentage inventory that needs to be constantly maintained to prevent stockout from occurring,

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715


b 1 X max I Adj: b j1 j

709

AAMPI

11

Also, we track the part variant exhibiting the worstcase shortage by using the adjusted maximum percentage inventory of part to prevent stockouts AMPIP max I max Adj: j
j1;2;...;b

12

5. Simulation study for mass customization in automotive assembly environment Most automotive companies now operate under a just-in-time system. Typically, parts suppliers are co-located in the neighborhood of the OEM and are required to deliver parts, as needed, several times in a day. A constant order cycle, non-constant order quantity framework is typically adopted for the delivery of outsourced parts as well as in-house parts from geographically dispersed OEM-owned plants. Monden (1998, pp. 59, 253) describes how Toyota Motor Corporation adopts this framework with its suppliers. Japan Glass Sheet Company delivers parts 10 times and 16 times to Toyotas Motomachi and Tsutsumi plant, respectively. Discussions between the author and two managers of tier-one automotive parts suppliers in North America suggest that similar practice is adopted. We know that increased customization will generally lead to proliferation of variants of a given part. This study examines the implication of increased customization on the operations of suppliers for parts used for assembly at an automotive OEMs plant within the environment described above. In particular, we investigate the extent to which customization impacts the average inventory of each variant that should be maintained to meet the OEMs needs. We adopt computer simulation methodology using data that are representative of actual assembly line and supplier systems. We chose to simulate the system rather than use data for a particular automobile, because of the widely varying possibilities that can exist for dierent companies. Although we do not test all possible instances, we believe that those tested will provide useful insights. 5.1. Description of experiment and parameter settings The automobile consists of many parts and subassemblies: including engines, transmission systems, frames, front-axles, rear-axles, steering assemblies,

body types, grades (series), wheels, doors, air-conditioners, seats, etc. There are over 20,000 parts and sub-assemblies used in an automobile, comprising of about 20 major sub-assemblies, each of which can be of dierent types (e.g., Monden, 1998, pp. 264). For example, a customer might want a V-6 engine of 2.0-l capacity, while another customer might want an in-line engine of 3.0-l capacity. Thus, many possible variants of engines can be provided to customers by the automaker. The same scenario applies to many of the other sub-assemblies. Parts that are common across all product models being assembled need not be considered under a sequence-scheduling framework because their use is inherently uniform over the entire scheduling horizon. However, usage of certain parts variants require levelling in view of the variable customer demand for these items. Levelling the usage of parts is one of the key objectives in a just-in-time environment because it reduces inventory levels and facilitates eective use of resources in supplying processes. In line with industry practice, we dene the Bill of Materials with respect to parts classes, where each part class comprises of part variants. For example, vehicle engines would be a part class, whereas 2.0 l, V-6 engine and 3.0 l, in-line engine are two part variants within this part class. A customer order comprises of choice of one part variant from each of the available parts classes. This leads to a wide range of possibilities of outputs, each potentially unique and dierent for each customer. Fig. 1 provides schematic representation of a typical customer order. The level of customization is dened relative to a base structure comprising of a set number of parts classes and variants. For example, if there are three part classes with 2, 3, and 5 parts variants respectively, then 100% customization leads to 4, 6, and 10 parts variants within these parts classes. Thus the total number of parts increase from 10 to 20. Notice that this percentage is uniformly applied to all 10 parts classes. Although increase in parts variants need not be the same for all parts classes, we have adopted this uniform framework to enable us better assess the impact of various levels of mass customization. Another factor in the experimentation is the supply frequency or the number of times per day or production shift that a parts supplier delivers parts to the assembly plant. We basically assumed that all parts suppliers deliver parts the same number of times per day or shift. This was to enable us assess

710
Part Class Engines Drive Train

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

Steering

Seats

. Frames .
F1 F2

E1 E2 E3 E4

D1 D2 D3 S1 S2S3 S4 S5 A1 A2 A3

Part variant
Customer As order Customer Bs order

1990; Kubiak et al., 1997) to determine optimum sequences and Toyotas Goal Chasing Method (Monden, 1983, 1998) for the heuristic sequences. Pilot studies were also run with the 2-step heuristic sequences alluded to earlier. Details of the data follow below. The computer programs were coded in Fortran and C and computational results were derived from a SUN Engineering workstation. (The storage capacity of the computer system available limited the size of optimum instances that could be tested.) 5.1.2.1. Optimum problem instances. Three dierent production plans were adopted comprising of 12, 18, and 24 product types and with the same corresponding quantity produced over the planning horizon. There were ve parts classes with a base structure of 5, 2, 3, 4, and 6 part variants, leading to a total of 20 part types. Two levels of customization were considered: 0% customization which is same as the base structure and 100% customization. Thus there are 40 parts at 100% customization. Considering the quantity in the production plan, supply frequencies of 2 and 3 were basically used. For theoretical purposes, supply frequencies of 6 as well as the total products manufactured in each case; 12, 18, and 24; were also considered. Ninety problem instances were solved for each production plan and for each level of customization. Sequences from these instances were used to determine pertinent values for the various supply frequency levels. This led to a total of 90 3 production levels 2 levels of customization 4 supply frequency levels = 2160 problem instances. We applied both the minsum and minmax dynamic programming optimizing procedures described earlier to the above problem instances. 5.1.2.2. Heuristic problem instances. Here, we simulate an assembly system with 10 parts classes. We consider the number of variants for any given part class to be from the set {4, 6, 8, 10, and 12} variants. The actual parts composition is randomly generated from the above set based on a uniform distribution. Two basic parts variants composition (PVC) structures were used: PVC structure 1 4; 12; 4; 4; 8; 6; 4; 10; 6; 10 Total of 68 parts variants: PVC structure 2 8; 6; 8; 12; 6; 12; 10; 4; 10; 4 Total of 80 parts variants:

Fig. 1. Relationship among parts classes, parts variants, and customer orders for an automotive supply chain.

the impact of supply frequency. Small pilot studies in which we relaxed this assumption showed no signicant dierence in the general pattern of results. 5.1.1. Generation of the bill of materials The part that is selected among the dierent variants by a given customer (for a given product) is determined from randomly generated values obtained from a Uniform, U(0, 1) distribution. Also, we have assumed that each variant has the same probability of demand as any other variant in a given part class. We consider this reasonable because this will likely be the case in the absence of information about relative demand distributions for the variants in a given part class. Each part class has its distinct random number stream, illustrating the fact that demand for a given variant from one part class is independent of demand for another variant from another part class. However, demand among the variants in a given part class are interdependent, since for one product, only one unit (or cluster of unit as in the case of four tires making one set) can be requested by a customer. More specically, for a part class that has four variants, each variant has a probability of 25% of being selected by a customer. It is important to note that the same random number streams are used for dierent levels of customization. The essence of this design is to enable us assess the impact of customization or providing the same set of customers with various levels of options (variants) for a given part type. 5.1.2. Problem instances We conducted experimental analysis using two sets of data, one for smaller problem instances to produce optimum sequences and the other was for very large scale instances using heuristics. We used dynamic programming (e.g., Miltenburg et al.,

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

711

The above PVC structures were considered as the base structures. The following levels of mass customization; 50%, 100%, 150%, and 200% were applied to these base PVC structures. For example, 50% customization for PVC structure 1 leads to the following PVC structure, {6, 18, 6, 6, 12, 9, 6, 15, 9, 15}, resulting in a total of 102 parts variants for all parts classes. For each of these PVC structures, we consider four levels of demand for cars per day or production shift (400, 600, 800, and 1000 units). There are over 88 million and 530 million distinct possible customer orders to the OEM for PVC structure 1 and PVC structure 2, respectively. One hundred instances of daily customer orders were randomly generated for each of the scenarios indicated above. There were three supply frequency levels; 2, 5, and 10. Thus, a full factorial experimentation was undertaken, leading to 12,000 total instances (2 PVC structures 4 daily/shift demand 5 levels of customization (including the base structures) 3 supply frequencies 100 cases). 6. Results and discussion First, we examine the results for the optimum sequences based on the smaller problem instances. Table 1 shows the supply frequency amplication ratios [AAMPI (SF = 3)/AAMPI (SF = 2)] under various parameters and using the minsum and minmax objective functions. Except for very few cases, the ratios are greater than 1. This suggests that increasing supply frequency does amplify the inventory level. We also found this phenomenon for the ratios based on the other supply frequencies used in the study.

The customization inventory ratios for 100% mass customization relative to the base case (0%) mass customization [AAMPI(100%)/AAMPI (0%)], including the percentage of cases for which the ratios are greater than 1 are shown in Table 2. The values depended on the specic problem instances. In some cases the value was greater than 1 whereas in others, it was less. We only encountered one instance when it was equal to 1. A value greater than 1 suggests that customization leads to increase in level of inventory required to prevent stockouts. Although the values for specic instances were not always greater than 1, on average, they were greater than or close to 1 in most cases. Following discussions in Steiner and Yeomans (1993) and Kubiak et al. (1997) about the relative advantage of the minmax formulation over the minsum formulation with respect to inventory levels at the scheduling stages, we computed the ratios of the values for the minmax/minsum functions and these results are summarized in Table 3. Ratios that are less than 1 suggest that the minmax function leads to lower inventory levels. The results show that, even though the values for the respective problem instances were sometimes greater than 1, on average, they were less than 1 in most cases. Now let us turn to the results for the large problem instances. As expected, while at some supply epochs there was surplus inventory relative to assembly line requirements, at others there were shortages. For over 95% of the instances tested, there was at least one part for which, at some epoch, the cumulative production level at the suppliers plant exactly matched the cumulative amount needed at the assembly line. We also noticed for

Table 1 Supply frequency amplication ratios for optimum solution cases supply frequency 3 relative to supply frequency 2 Case a = 12, a = 12, a = 12, a = 12, a = 18, a = 18, a = 18, a = 18, a = 24, a = 24, a = 24, a = 24, Base Base 100% 100% Base Base 100% 100% Base Base 100% 100% Objective + * + * + * + * + * + * Average 2.6097 2.3265 1.9275 1.7369 2.3799 2.2851 2.1272 1.8743 2.1950 1.9662 2.3568 2.0761 Minimum 1.4171 0.8511 1.2500 1.0600 1.0493 0.8094 1.5166 1.2529 0.7253 0.7023 1.2387 1.1820 Maximum 14.6250 13.1250 3.2769 2.7000 6.1176 7.0216 3.2350 2.9521 14.0553 6.7353 5.1651 3.6946

Customization Customization

Customization Customization

Customization Customization

a, number of product types; Base, 0% customization; +, MinSum; *, MinMax.

712

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715 Table 3 Minmax/minsum ratios for optimum solution cases 100% customization level Case a = 12, SF = 2 a = 12, SF = 3 a = 12, SF = 6 a = 12, SF = 12 a = 18, SF = 2 a = 18, SF = 3 a = 18, SF = 6 a = 18, SF = 18 a = 24, SF = 2 a = 24, SF = 3 a = 24, SF = 6 a = 24, SF = 24 Average 1.0527 0.9440 0.9507 0.9752 1.0668 0.9266 0.9106 0.9670 1.0871 0.9398 0.9392 0.9914 Minimum 0.7895 0.7119 0.8039 0.8438 0.5412 0.6550 0.6792 0.8411 0.5128 0.6329 0.6807 0.8627 Maximum 1.3929 1.1026 1.1203 1.0659 1.5976 1.1822 1.0570 1.0620 3.0438 1.4132 1.1138 1.1256 % < 1.0 28.9 68.9 76.7 53.3 31.1 75.6 91.1 67.7 46.7 63.3 80.0 63.3 % = 1.0 18.9 8.9 5.6 30.0 8.9 0.0 0.0 25.6 0.0 0.0 0.0 1.1

Table 2 Customization inventory ratios for optimum solution cases 100% customization relative to base case Case a = 12, SF = 2 a = 12, SF = 2 a = 12, SF = 3 a = 12, SF = 3 a = 18, SF = 2 a = 18, SF = 2 a = 18, SF = 3 a = 18, SF = 3 a = 24, SF = 2 a = 24, SF = 2 a = 24, SF = 3 a = 24, SF = 3 Objective + * + * + * + * + * + * Average 1.1592 1.1718 0.8411 0.8595 1.2741 1.3137 1.1604 1.0958 0.9555 0.9377 0.9514 0.9336 Minimum 0.4424 0.4597 0.5003 0.4488 0.4725 0.4915 0.4845 0.5260 0.1330 0.1186 0.3143 0.1974 Maximum 8.1563 8.0156 1.3849 1.4397 3.7966 2.9888 2.0478 1.7179 6.6803 4.6745 2.3605 2.1766 % > 1.0 50.0 45.6 20.0 21.1 61.1 67.8 64.4 56.7 32.2 33.3 43.3 42.2

a, number of product types; SF, supply frequency; +, MinSum; *, MinMax.

a, number of product types; SF, supply frequency.

each problem instance tested, that all customer orders comprising the complete set of all variants from all parts classes were distinct. This was expected, considering the very large number of possibilities. The results indicate some variation in the amounts of shortages for any given part variant at the various supply epochs. For example, Table 4 shows averages of I max (Eq. (9)) for the variants j within each of the parts classes under a given set of scheduling parameters (number of product types/demand level = 1000 units, PVC structure 1 at 100% customization level) when supply frequencies are 2, 5, and 10. We observe that these averages can vary widely among parts classes. This implies that the values dier among parts suppliers, even

though each one of them experiences uniform demand for the variants it supplies to the assembly line. Figs. 2 and 3 show prole plots of the adjusted average maximum percentage inventory to avoid stockouts (AAMPI) for PVC structures 1 and 2, respectively when the supply frequency is 5. This result for each PVC structure represents 2000 instances of the 12,000 total instances tested. We see a steady increase in the average level of inventory that needs to be maintained to be prevent part stockout as the level of mass customization increases. The right hand side of the gures show corresponding plots of the mean values of AAMPI under the three supply frequencies tested. Here again, we see a steady increase in the amount of

Table 4 Sample results of average maximum percentage inventory for the parts classes SF Part class 1 2 5 10 0.2998 0.6710 0.9923 2 1.0856 2.4095 3.1962 3 0.1985 0.7194 0.8930 4 0.1931 0.5926 0.6967 5 0.4595 1.5023 2.1629 6 0.4147 1.1368 1.3327 7 0.2945 0.6862 0.7895 8 0.4053 1.6847 2.0602 9 0.3124 0.9962 1.3197 10 0.5793 1.8375 2.2092

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715


Profiles for PVC structure 1 (Supply Frequency = 5)
3 2.5 2 1.5 1 0.5 0 0% 50% 100% 150 % 200% 250%

713

Mean AAMPI values (PVC structure 1)


2 AAMPI (%) % 1.5 1 0.5 0 0% 50% 100% 150% 200%

Maximum 75th Percentile Mean Median 25th Percentile Minimum

AAMPI (%) %

Supply Freq. = 2 Supply Freq. = 5 Supply Freq. = 10

Level of Customization

Level of Customization

Fig. 2. Results for PVC structure 1.

Profiles for PVC structure 2 (Supply Frequency = 5)


3 2.5 2 1.5 1 0.5 0 0% 50% 100% 150% 200% 250%

Mean AAMPI values (PVC structure 2)


2.5 AAMPI (%) % 2 1.5 1 0.5 0 0% 50% 100% 150% 200% Supply F req. = 2 Supply F req. = 5 Supply F req . = 10

Maximum 75th Percentile Mean Median 25th Percentile Minimum

AAMPI (%) %

Level of Customization

Level of Customization

Fig. 3. Results for PVC structure 2.

inventory as the level of customization increases under all supply frequencies. Also, we observe that the amount of inventory requirement tends to increase with increase in supply frequency. Furthermore, the worst-case metric for part inventory level (AMPIP) portrays similar patterns as shown in Table 5. Tests carried out under various other parameter conditions, including the number of parts classes, as well as by using the two-step heuristic yielded similar pattern of results. Before proceeding with the discussion, we state the following observation from an earlier published work. 6.1. Observation (Aigbedo, 2000) Levelling would tend to be more dicult to achieve for products with comparatively smaller number of units in the production plan (smaller
Table 5 Average values of AMPIP for all 12,000 instances Level of customization (%) PVC structure 1 SF = 2 Base-case 50 100 150 200 2.4563 3.0444 3.3439 3.8846 4.3393 SF = 5 3.6438 4.4740 5.0509 5.5856 6.1817

product ratios) as they tend to experience larger deviations from level schedule. Aigbedo (2000) derived a lower bound for the contribution of products to the overall deviation from level schedule for the PRV problem. This led to an expression based on which the paper makes the above observation. Following the exact same procedure in that paper, from the lower bound on the optimum solution of the parts usage smoothing problem, it is easy to arrive at a corresponding observation with respect to products that make use of a given part type. There is therefore theoretical support for the mass customization phenomenon observed in the experimentation. In other words, the variation from level schedule for parts can be shown to be larger as the number of product units that use that part decreases. Since mass customization tends to reduce the number of units of products making use of a

PVC structure 2 SF = 10 4.1554 4.9798 5.5518 6.1643 6.8211 SF = 2 2.8502 3.4955 4.0370 4.4150 4.9975 SF = 5 4.1911 4.8590 5.4780 6.1722 6.7166 SF = 10 4.6473 5.3687 5.9469 6.7137 7.3734

714

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715

given part type, this phenomenon would tend to occur. Thus there would be tendency towards higher level of inventory requirement to prevent stockouts with increase in level of mass customization. It is pertinent to note that normalization in Eq. (10) only accounts for this dierence from the perspective of our denition of inventory as a percentage of daily (per shift) usage. It does not account for the inherent dierences as the optimizing or heuristic algorithm attempts to smooth usage of each part variant. As noted above the optimum sequences revealed the same general patterns of result as the heuristic sequences. The fact that the results for the optimum sequences were not perfect in portraying some of the expected characteristics may not be unconnected with scale eects for the parameters due to the smaller problem sizes. The evaluation metrics used in this study dier from the conventional objective functions for the minmax and minsum formulation in two important ways. First, it examines what happens to the deviations from ideal at the supply epochs rather than at stage-by-stage or on product unit-by-product unit basis. Second, it increments the determined inventory values to the next higher integers to avoid fractional inventory. These dierences may explain why results for the minmax framework were sometimes higher (only for comparatively fewer instances) than the minsum framework, dierent from the observation by Steiner and Yeomans (1993) and Kubiak et al. (1997). The increase in inventory requirement with increasing supply frequency appears to be related to increased constraint arising from integer requirements for parts to be supplied to the assembly line as supply frequency increases. 7. Conclusion Many automotive companies operate a just-intime system in which parts suppliers deliver parts several times a day to the assembly line. Furthermore, some are applying mass customization based on framework that is similar to those adopted by rms such as Dell Computer Corporation. Our study focused on mass customization as it aects inventory of parts variants supplied to an automotive OEMs assembly line using the supplier delivery system described above. Through simulation of this system under various assembly scheduling parameters, the study nds, among other things, that in addition to the increase in number of parts variants

that results from mass customization, there tends to be increase in the average amount of inventory of the parts variants needed to be held constantly to prevent stockouts from occurring. This result was readily linked to a theoretically derived lower bound on the optimum contribution of parts to an objective function value where minimizing inventory levels is sought. Furthermore, the presented framework can be adapted in helping management plan the appropriate level of mass customization that is desirable to meet customer requirements, while ensuring that supply chain costs are not excessive. The study uses an inventory evaluation metric suited to the multiple part delivery per day system, dierent from the conventional metrics based on stage-by-stage type objective functions developed in the literature. The study conrms with this metric, earlier suggestion in the literature that the minmax optimization framework is superior to the minsum optimization framework in terms of minimizing inventory levels. In the paper it is assumed that the distribution of demand for parts variants within each part class is uniform. This is reasonable when information about distributions is not yet available. The presented framework can be extended to investigate the eect of various parts distribution on the required level of inventory. Firms can aggregate part variant demand information over a period of time to enable them determine the appropriate level of mass customiza` tion vis a vis the desired amount of inventory levels and costs. References
Aigbedo, H., 2000. Some structural properties for the just-in-time level schedule problem. Production Planning and Control 11, 357362. Aigbedo, H., Monden, Y., 1997. A parametric procedure for multicriterion sequence scheduling for just-in-time mixedmodel assembly lines. International Journal of Production Research 35, 25432564. Alford, D., Sackett, P., Nelder, G., 2000. Mass customization An automotive perspective. International Journal of Production Economics 65, 99110. Andel, T., 2002. From common to custom: The case for make-toorder. Material Handling Management 57 (12), 2431. Austin, R., 2001. Ford Motor Company: Supply chain strategy. Harvard Business School Case, 9-699-198. Bautista, J., Companys, R., Corominas, A., 1996. Heuristics and exact algorithms for solving the Monden problem. European Journal of Operational Research 88, 101113. Ding, F., Cheng, L., 1993. A simple sequencing algorithm for mixed-model assembly lines in just-in-time production systems. Operations Research Letters 13, 2736.

H. Aigbedo / European Journal of Operational Research 181 (2007) 704715 Duplaga, E., Bragg, D., 1998. Mixed-model assembly line sequencing heuristics for smoothing component parts usage: A comparative analysis. International Journal of Production Research 36, 22092224. Feitzinger, E., Lee, H., 1997. Mass customization at Hewlett Packard: The power of postponement. Harvard Business Review 75 (1), 116121. Gilmore, J., Pine, B., 1997. The four faces of mass customization. Harvard Business Review 75 (1), 91101. Groein, H., Luss, H., Rosenwein, M., Wahls, E., 1989. Final assembly line sequencing for just-in-time manufacturing. International Journal of Production Research 27, 199213. Harrison, A., 1992. Just-in-Time Manufacturing in Perspective. Prentice Hall, New Jersey. Inman, R., Buln, R., 1991. Sequencing JIT mixed-model assembly lines. Management Science 37, 901904. Kotha, S., 1996. From mass production to mass customization: The case of the National Industrial Bicycle Company of Japan. European Journal of Management 14 (5), 442450. Kubiak, W., 1993. Minimizing variation of production rates in JIT systems: A survey. European Journal of Operational Research 66, 259271. Kubiak, W., Sethi, S., 1991. A note on level schedules for mixedmodel assembly lines in just-in-time production systems. Management Science 37, 121122. Kubiak, W., Sethi, S., 1994. Optimal just-in-time schedules for exible transfer lines. International Journal of Flexible Manufacturing Systems 6, 137154. Kubiak, W., Steiner, G., Yeomans, S., 1997. Optimal level schedules for mixed-model, multi-level just-in-time assembly systems. Annals of Operations Research 69, 241259. Mello, A., 2001, Mass customization wont come easy. Tech Update, December. Available from: <http://techupdate. znet.com>. Miltenburg, J., 1989. Level schedules for just-in-time production systems. Management Science 35, 192207. Miltenburg, J., Sinnamon, G., 1989. Scheduling mixed-model multi-level just-in-time production systems. International Journal of Production Research 27, 14871509.

715

Miltenburg, J., Steiner, S., Yeomans, S., 1990. A dynamic programming algorithm for scheduling mixed-model just-intime production systems. Mathematics and Computer Modeling 13, 5766. Monden, Y., 1983. Toyota Production System: A Practical Approach to Production Management. Industrial Engineering and Management Press. Monden, Y., 1998. Toyota Production System: An Integrated Approach to Just-in-Time. Engineering and Management Press. Ng, W., Mak, K., 1994. A branch and bound algorithm for scheduling mixed-model assembly lines. International Journal of Production Economics 33, 169183. Pine, B., 1999. Mass Customization: The New Frontier in Business Competition. Harvard Business School Press, Cambridge, MA. Pleshberger, T., Hitomi, K., 1993. Flexible nal assembly sequencing method for JIT manufacturing environment. International Journal of Production Research 31, 11891199. Salvador, S., Forza, C., Rungthusanatham, M., 2002. How to Mass Customize: Product Architectures, Sourcing Congurations. Business Horizons 45 (4), 6169. Schwartz, E., 1999. Buid-to-order drives change. Infoworld 21 (39), 12. Steiner, G., Yeomans, S., 1993. Level schedules for mixed-model, just-in-time processes. Management Science 39, 728735. Sumichrast, R., Clayton, E., 1996. Evaluating sequences for paced, mixed-model assembly lines with JIT component fabrication. International Journal of Production Research 34, 31253143. Sumichrast, R., Russell, R., Taylor, B., 1992. A comparative analysis of sequencing procedures for mixed-model assembly lines in a just-in-time production system. International Journal of Production Research 30, 199214. Thomopoulos, N., 1967. Line balancing-sequencing for mixedmodel assembly. Management Science 14, B59B75. Zipkin, P., 2001. The limits of mass customization. Sloan Management Review 42 (3), 8187.

You might also like