decision support systems - university of...

Decision Support Systems 52 (2012) 790–801

Contents lists available at SciVerse ScienceDirect

Decision Support Systems

j ourna l homepage: www.e lsev ie r .com/ locate /dss

A decision support system for integrating manufacturing and product design into thereconfiguration of the supply chain networks

Yohanes Kristianto a,⁎, Angappa Gunasekaran b, Petri Helo a, Maqsood Sandhu c

a Department of Production, University of Vaasa, P.O. Box 700, FI-65101 Vaasa, Finlandb Department of Decision and Information Sciences, Charlton College of Business, University of Massachusetts – Dartmouth, 285 Old Westport Road, North Dartmouth, MA 02747-2300, USAc Faculty of Business and Economics, United Arab Emirates University, P.O. Box 17555, Al-Ain, United Arab Emirates

⁎ Corresponding author.E-mail address: [email protected] (Y. K

0167-9236/$ – see front matter © 2011 Elsevier B.V. Alldoi:10.1016/j.dss.2011.11.014

a b s t r a c t
a r t i c l e i n f o
Available online 23 November 2011

Keywords:AssemblyDecision Support SystemsInventory allocationProduct designSupply chain

A supply chain needs to meet its customers' requirements (CRs) in terms of delivery lead times, total costsand product quality. The objective of this article is to improve the level of integration in all aspects of supplychain reconfiguration, such as the inventory allocation and manufacturing process involved, by incorporatingmanufacturing and product design into logistic design. The effect of uncertain customer demand, productionand supply lead times are studied. An optimum supply chain network is configured by combining optimiza-tion at the strategic and tactical level. A system dynamic based computer simulation model is used to validatethe operations of the supply chain. The performance of the system is measured in terms of backorders andinventory level. The results and analysis indicate that fewer stockholding points and a shorter review periodof demand can improve performance in this respect. In addition, a proposal for improving the performance ofsupply chain in terms of lower safety stocks is presented. Finally, management decision-making is discussed,among other concluding remarks.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

This paper examines the optimal supply chain configuration forcustomized products. Our intent is to develop a Decision Support System(DSS) for integrating manufacturing and product design into the designof the logistic process, since the supply chain must be reconfiguredbefore determining the product design, manufacturing technologiesand vendors. We use assembly to represent manufacturing, since it isinherently integrative and should be composed to meet the FunctionalRequirements (FRs) of the product. The result is that the supply chaincan be reconfigured with full knowledge of the way in which theproduct is supposed to work. Supply chain reconfiguration involvescreating a suitable assembly sequence, identifying subassemblies,integrating inventory control, and designing supplier–buyer coordina-tion so that the performance in terms of backorders and inventory levelsis compatible with the assembly method [40].

The liaison between manufacturing and product design, and thelogistic process design should make a tradeoff between a higherunit of manufacturing cost with a more responsive supply chain or alower manufacturing cost with a less responsive supply chain.While the FRs on this point have already been determined, there are

ristianto).

rights reserved.

severalmanufacturing options available formanufacturing or assemblingthe product. The supply chain configuration chooses a manufactur-ing option in terms of make-to-stock (MTS), make-to-order (MTO)or assemble-to-order (ATO) for each stage of the supply chain, soas to achieve the product functionality at minimum manufacturingcost and with higher supply chain responsiveness.

The supply chain reconfiguration frameworks consider three areaswhich are relevant to choosing a manufacturing option for each stageof the supply chain: assembly planning, demand planning, and inven-tory allocation. A better quality of demand planning leads to optimuminventory allocation [15,17,19,28]. The optimum further assists thesupply chain with decision methods for responding to demands andimproving the supply chain delivery performance [7,8,11,33,36]. Theoptimum assembly planning provides knowledge to the supplychain so as to realize the product with minimum lead times withoutimpairing its functioning.

The frameworks of supply chain reconfiguration change the ex-pectations regarding the DSS associated with the liaison betweenmanufacturing and product design and the logistic process design,with the following challenges.

(a) Assembly redundancy: Whitney [40] mentions that a well-designed product is a predictable product. This implies thatproduct FRs analysis helps a designer to manufacture a productwith minimum manufacturing effort without impairing itsfunctioning. Thus, the product designer must consider theProduct Key Characteristics (P-KCs). P-KCs are the FRs of a

http://dx.doi.org/10.1016/j.dss.2011.11.014

mailto:[email protected]

http://dx.doi.org/10.1016/j.dss.2011.11.014

http://www.sciencedirect.com/science/journal/01679236

791Y. Kristianto et al. / Decision Support Systems 52 (2012) 790–801

product to satisfy the Customer Requirements (CRs). P-KCsneed support from the lower level of FRs to deliver CRs,which are called the Assembly Key Characteristics (A-KCs).Optimizing the assembly sequence according to the P-KCsand A-KCs hierarchy avoids redundant coupled assembly[35]. The optimum assembly sequence reduces the contentof information within the product such that it also minimizesthe information content within the product so as to reduceiterative operations in manufacturing [23]. In other words,instead of reducing the interdependency of the parts, the supplychain has to explicate the functional relationship between partsto properly deliver functionality in a product and minimize prod-uct failures. Product failures are defined as the non-conformity ofa product to the desired FRs. A lower probability of productfailure minimizes the requirement on safety stock and reducesbackorders.

(b) The imprecision of demand information: safety stock costs andbackorders can be minimized by providing credible demandinformation through the application of DSS. The quality ofdemand information affects the ordering process and controlsthe target inventory level of the buyer [3]. Some research arti-cles [9,27,37] focus on DSS development to forecast customerdemands. Yao et al. [42] suggest the application of a VendorManaged Inventory (VMI) as a DSS to minimize the inventorycosts by distributing inventory status and demand informationevenly along the supply chain. While VMI does not share theinventory status, the way in which VMI responds to demanddepends on variations in the manufacturing process. To dealwith the imprecision of demand information and manufacturingprocess variation, it is necessary to understand the implica-tions of manufacturing process variation and to characterizethe way in which the variations are hedged effectively throughVMI.

(c) Non optimum inventory allocation: The imprecision of demandinformation affects the planning of the logistic process interms of flexibility and adaptability under the limitations ofthe capacity to supply with respect to the system dynamics[5]. Most of VMI is used to generate production flexibility soas to minimize the discrepancy between demand and orderrate [38,39,42] without considering the production capacityand capability in terms of the variability of production leadtimes. In other words, instead of a decision on production andorder, the DSS needs to optimize the composition of the stock-holding points; it is possible to mix push and pull inventorysystems to satisfy customer demand at various levels of produc-tion variability [41].

To this end, a DSS, called the two-level optimization of supplychain networks, is proposed, for the purpose of integrating themanufacturing and product design into the supply chain reconfigura-tion in order to improve demand planning, inventory allocation andassembly planning. To deal with integration effectively, the concept ofthe FRs analysis of a product is put forward. On the first level, optimizedassembly sequencing is developed to allocate the safety stocks and tochoose the stockholding points. On the second level, optimized produc-tion and distribution control is developed to characterize the supplychain networks effectively. The merit of a two-level optimization ofsupply chain networks is detailed in terms of inventory levels through-out supply chains and backorders.

The rest of the paper proceeds as follows. Section 2 reviews theliterature on inventory allocation in the supply chain. Section 3 high-lights the scope and definition of the problem. Section 4 focuses on thefeatures of DSS modeling, mainly aiming at supply chain reconfigurationand safety stock allocation. Section 5 validates the supply chain model inthe previous section and discusses the DSS implications for managementdecision-making. Section 6 concludes this article.

2. Background of the research

The integration of FR analysis into the reconfiguration of supplychain networks extends the supply chain decisions by adding ‘howto manufacture’ to ‘how much to deliver’ and ‘how much to produce’.The integration requires an interplay between quality management(QM) and supply chainmanagement (SCM) to coordinate and integratethe manufacturing processes [32]; however, not much research hasbeen reported on the topic. Balachandran and Radhakrishnan [2]model the effect of supply chain coordination in terms of warranty con-tract to minimize product failures. Whitney [40] suggests that supplychains have to meet the design intent (FRs) to obtain a well-designedassembly. Dong and Whitney [9] use an axiomatic design approach tomeet FRs at a minimum degree of integrality and information content[35]. We are aware of prior research which attempts to analyze theFRs as a tactic for assembly planning in a supply chain.

Themodel whichwe develop falls within the literature on assemblyplanning in a supply chain, in particular for a strategic inventory alloca-tion of supply chain reconfiguration. In this way, our work is related tothat of Novak and Eppinger [30] and Graves and Willems [18,19]. Theformer authors [30] hypothesize that product complexity and verticalintegration are complementary and suggest greater coordinationbetween the product design engineer and the supply chain engineer.They use original empirical evidence from the auto industry and findthat simpler product architecture can be suggested for an autonomoussupply chain. Graves and Willems [18,19] develop a model for posi-tioning safety stock in a supply chain, subject to non-stationary de-mand. It is suggested that optimal safety stock allocation mayentail changing the service times, and thus the locations of safetystocks, as demand evolves over time. The authors [18,19] assumethe same demand process for a multi-echelon system consisting ofsome manufacturing sites. The sites operate with an independentbase-stock policy.

Our work differs from that of Novak and Eppinger [30], however,in that we consider axiomatic design, for which product complexityis reduced by first identifying the liaison between the parts of a productand then providing information about the interdependency of the parts[43]. Liaison identification helps a designer to convey a well-designedmanufacturing method so as to put into action the logistic planningby minimizing the interdependency of parts and to meet FRs [9]by relating the axiomatic design to assembly sequencing [43]. Notbeing able to do this leads to capacity overload and yields only80% of successful assemblies on the first try [6,14,40]. In addition,part modularization helps a logistics manager to minimize theamount of uncertain information about supply and demand withinsupply chains [12,37] and thus to reduce the level of safety stock[22]. While liaison identification has an important role in supportingsupply chain reconfiguration in terms of better assembly planning andproduction clustering, the implications for the supply chain inventorycost are ignored or at best poorly understood [25].

Our model differs from that of Graves and Willems [18,19] in thatwe provide an insight into the behavior of inventory allocation andbenefit from information sharing to minimize uncertainty about sup-ply and demand [4,19,29,31]; this being the case, we have structuredthe model to allow the greatest possible responsiveness of production–distribution and clarify where the inventory hedge should be set. Gravesand Willems [18,19] allocate the safety stock by assuming that theprocessing times are deterministic and that no capacity constraintslimit production at any stage. Furthermore, the previous contributionsminimize the safety stock placement at optimum processing time byconsidering the service level of the vertically integrated supply chain.Conversely, we consider backorder costs rather than service leveltargets in order to reflect managers' commitment to providing a100% service level. In addition, VMI is employed to an autonomoussupply chain and the service level is adjustable so as to maximize thesupply chain performance.

792 Y. Kristianto et al. / Decision Support Systems 52 (2012) 790–801

This literature review leads on to supply chain coordination in termsof manufacturing and logistics planning. The supply chain needs tosimplify the manufacturing method and minimize the content ofinformation among the processing facilities. From the manufacturingviewpoint, assembly sequencing should minimize interdependenciesbetween parts. From the logistical viewpoint, VMI provides a promisingsolution. This article covers the integration of manufacturing andlogistics planning to improve the supply chain performance in termsof backorders and inventory levels by studying the effect of a reviewperiod of demand and assembly sequencing after analyzing productFRs [13].

3. Problem scope and definition

In this section, we present the assembly and distribution operationsof a furniture product family for reconfiguring the supply chain andallocating the inventory at the lowest possible backorder and inventorylevels. The processing facilities have limited capacity for manufacture.However, the processing facilities have storage facilities. Dependingon customer orders, the production quantities at the processing fa-cilities fluctuate over time and are not necessarily pre-determined.In addition, a processing facility has the probability of failure during theproduction process. The suppliers have no additional informationregarding the buyers' inventory level. Indeed, forecasting matricesabout demand are distributed across the supply chain, which is usedby the suppliers to decide on delivery and production rates.

In considering probability of failure, we formulate the problem as aqueue model and propose a dynamic simulation method. The methoduses a GI/G/1-based inventory control which considers supply anddemand uncertainties. The production rate in the queue model isdynamic by allowing backorders. Distribution from suppliers tobuyers is considered in our model by implementing VMI to maximizethe economies of scale in transportation costs.

First, in order to illustrate the proposed DSS in supply chain recon-figuration and inventory allocation, Fig. 1 presents below a descriptionof the strategic and tactical level optimization model.

4. DSS for tactical and operations level optimization

Fig. 1 shows that the DSS is split into a two-stage optimization, astrategic and tactical level. The optimization covers:

Materile

MS-Excel DSS for strategic level optimization

GoldSimTM

DSS for tactical level optimization

Optimization of supply chain response

parameter Tw, Ti,Tq

Demand from

downstream

( )tD

Exponential smoothing ( )tF

KC-analysis and assembly sequencing

Supplyreconfi

Fig. 1. A DSS for supply chain reconfig

4.1. Strategic level optimization

The supply chain reconfiguration considers delivery lead times Lnand service level zn in the supply chain against demand uncertaintyσd(n). The sequence should reduce part interdependency. Since mostof the literature on inventory allocation considers the first two com-ponents [24,19,29], part interdependency is rarely discussed.

4.1.1. KC analysis and assembly sequencing for supply chain reconfigurationWeuse the generic product structure of an office chair as an example

[10]. There are six part frames (Fig. 2) comprising back, upholstery andarm rest as standard parts, and a seat frame, stand and support asthree distinctive modules. The FRs of the chair includes its capabilityfor making 360° rotations (FR 1) and choosing the quality of frame (FR2). Both of these FRs are supported by defining P-KC. The seat frameand the seat deliver P-KC 1 by allowing the customer to choose thematerial of the seat frame. The support and stand of the office chairdeliver P-KC 2 by allowing the customer to rotate the chair 360°.Each P-KC is supported by using A-KCs to meet the quality objectiveof the product.

In addition, A-KCs are presented in the form of a liaison diagram,as follows.

Fig. 3 shows mate joints between upholstery and seat, betweenback and under frame, and between arm rest and under frame. Matejoints (dotted lines) do not affect A-KCs deliverability, but contact joints(directed lines) do so directly. Joints between the support and underframe, between under frame and stand, between seat frame andunder frame, and between seat and under frame are represented asA-KCs 1 to 4. A-KCs 1 and 2 support the deliverability of P-KC 1 andA-KCs 3 and 4 for P-KC 2. The assembly sequence planning in Fig. 3guides the supply chain reconfiguration, where the arrows signifythe direction of product assembly.

The assembly sequencing in Fig. 3 is used to reconfigure the supplychain networks, which are represented by the design structured matrix(DSM). DSM is used for the clustering of dependencies in a matrix [43].The goal of DSM is to cluster closely related activities as shown in Fig. 4bso as to minimize the coupling between activities. The DSM squarematrix comprises rows for representing independent variables andcolumns representing dependent variables. The diagonal line isblank when it is expected that either dependent or independent var-iables are dependent upon themselves. For instance, in Fig. 4, activity

al and product inventory vel ( ) ( ) ( )tIt n,I nm

Backorders BOn(t)

Order

rate (t)nq

Production rate (t)nμ

Inventory adjustment

( )t(n)IΔ

WIP adjustment

( ) ( )tnWIPΔ

chain guration Inventory

allocationInventory

risk sharing

uration and inventory allocation.

Chair

Back Seat Underframe Arm rest

Seat Frame Upholstery Stand Support

Seat Frame A Seat Frame B Stand A Stand B Pad Wheel

Fig. 2. General product structure of a chair.


number 3 is sequentially accomplished just after activity number 1and is recognized as the second of two activities in a series. Finally,activities number 2 and number 3 have no relationship that wouldsignify simultaneous activities. It is shown in Fig. 1a that activities2 and 3 are overlapping before clustering.

The assembly sequencing can be used to answer technical challenge1 by the following proposition 1.

Proposition 1. If the number of stages within the supply chain is reducedand there is no overlapping cluster in the assembly sequence planning, thenthe product quality of the supply chain will increase. Furthermore, higherproduct quality minimizes the requirement for safety stock.

Proof. Let A1 and A2 be the failure probability (e.g. lateness, inappropri-ate quality, etc.) for materials at stages (n−2) and (n−1) respectively,their occurrence not being mutually exclusive. Thus, the probability ofcreating delivery delay due to the failure of one of the two coupledmaterials is pf(2)=A1+A2−(A1×A2) for An=1−zn to represent theservice level. Further, for N coupled materials we then have a failureprobability as much as

pf nð Þ ¼X1

x¼n−1

A n−xð Þ− A n−1ð Þ∩A n−2ð Þ∩ ::∩A 1ð Þ� �

: ð1Þ

The first component of Eq. (1) represents the probability of failuredue to one of N coupled materials. The second component representsthe joint probability for two or more components to fail together. Weuse this formulation since in a non-mutually exclusive event, materialfailure can occur altogether so that the entire product manufacturingbreaks down. Eq. (1) shows that the coupled operations should bedecoupled or be localized and manufactured integrally in the same

Back Seat Under frame Arm rest

Seat frame Upholstery Stand Support

P-KC 2

P-KC 1

A-KC 1&2

A-KC 3&4

Fig. 3. Liaison diagram of office chair.

processing site of the supply chain. Thus, it simply informs us that thebuyer will reject the material even if only one part of the material fails.

The localization can be accomplished by clustering the coupledoperations and supplying them from the same stage to minimize pf(n).Then Eq. (1) will be changed to the joint probability of failure for anon-mutually exclusive event, as follows:

pmf nð Þ ¼ max A n−1ð Þ;A n−2ð Þ; ::;A 1ð Þ� �

: ð2Þ

We can compare in Eq. (1) and Eq. (2) that the failure probabilityreduces from pf(n) to pmf(n) by as much as

pf nð Þ−pmf nð Þ ¼X1

x¼n−1

A n−xð Þ−max A n−1ð Þ;A n−2ð Þ; ::;A 1ð Þ� �

− A n−1ð Þ∩A n−2ð Þ∩ ::∩A 1ð Þ� �

:

ð3Þ

Furthermore, reducing the value of (A(n−1)∩A(n−2)∩..∩A(1))supports the risk sharing effort byminimizing the number of interactions(i.e. sharing functionality, interfaces, etc) within the module. Reducingpf(n) supports safety stock decoupling at stage (n). Finally, the safetystock reduction ΔSSn at stage n due to lower failure probability is repre-sented as follows:

ΔSSn ¼ zn pf nð Þ−pmf nð Þ� � ffiffiffiffiffi

Lnp

ð4Þ

where Ln stands for delivery lead times at stage n. In the following, wedescribe in more detail the inventory allocation and risk sharing tominimize the total inventory costs (cycle and safety stocks) afterreconfiguring the supply chain.

4.1.2. Inventory allocationIf one operates a supply chain while disregarding demand volatility

and capacity constraints, one may encounter unexpected stock-outs, asdeliveries are delayed at system bottlenecks. However, the safety stock

1 2 3 41 12 2 13 1 34 1 4

1 3 2 41 13 1 32 2 14 1 4

Fig. 4. (a). An example of a DSM matrix. (b). An example of a clustered DSM matrix.


follows a non-linear pattern and arbitrarily large for demand volatilitywhich is only slightly greater than average demand. Therefore, by char-acterizing the necessary safety stock levels as probability for stock out,we can determine the optimal safety stock placement in supply chainswith one or many capacity constraint(s).

The inventory allocation is modeled as a GI/G/1 queue model bybearing in mind the following reasons. First, consider the demand pro-cess of the upstream stage, namely the order stream (i.e., stage n) fromthe downstream stage (i.e., stage n−1). Second, the model considersthat the demand inter-arrival and processing rates are sometimes inheavy traffic (1−ε)bρn or sometimes not stationary and divergent ata certain demand level Vi at time i (i.e.,ρ≥1) for ρn ¼ λn

μn(λn and

μnrepresent demand rate and production capacity at stage n respective-ly). As a result, the waiting time in a queueWq(n) at stage n depends onthe demand inter-arrival rate standard deviation σA(n) and service ratestandard deviation σn [20]. Third, the model considers the lead time

variability in some ranges− ln 1−ρn2ð Þ

λn≤Wq nð Þ≤

λn σ2A nð Þþσ2

n

� �2 1−ρnð Þ . The variabil-

ity of lead times σA(n) signifies that the supply chain allocates the inven-tory to cover the demand volatility σV(n). Thus, within T time horizonwith review period j from time i=T− j to T, σA(n) is stated as σA nð Þ ¼

1λn− 1

λnþσV nð Þfor σV nð Þ ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPTi¼T−j

Vi−λn

j−1

vuut.

The longer review period of the demand j reduces the level of sen-sitivity about the demand changes. Lower sensitivity creates lowerdemand responsiveness atσn ¼ 1

σV nð Þþ 1

μnsince σn represents the stan-

dard deviation of production lead times, which is inappropriate for anon-stationary demand process. Lower demand responsiveness fur-ther suggests that the supply chain should invest in a higher level ofsafety stock. Thus, an analysis of demand responsiveness is requiredto allocate the inventory.

This article uses the order of Erlang distribution kn and hn to mea-sure the operation uncertainty in non-stationary demand at stage n.The reason for using the Erlang distribution in this case is that it facil-itates the calculations of a stock out that is stated by using a stock outprobability formulation of Leven and Segerstedt [26] as

Probability for stock out pso ¼ 1−∫In

0

hnkn xkn−1e−hnx

hn−1ð Þ!

dx ¼ hnkn e−hnIn

Xkn−1

i¼0

Ini 1i!hn

kn−1

ð5Þ

where In is the inventory on hand at stage n. For calculating the distri-bution parameters the following formulas can be used:

kn ¼ λn

σV nð Þ

!2

ð6Þ

hn ¼ λn

σV nð Þ2 : ð7Þ

For calculating the on hand inventory at stage n, In, the followingformulas can be used:

In ¼ μn:λn Wq nð Þ þ1λn

� �: ð8Þ

Eq. (8) signifies that In is increased at the increasing value of theσV(n)

and Wq(n). This trend signifies that either the lower σn2 or the higher μn

can help the supply chain to minimize the amount of inventory alloca-tion. This implies that safety stock can be allocated to nodes or stock-holding points with higher pso(n)>SL [26] for SL stands for customer

service level. The inventory allocation can be used to answer technicalchallenge 3 by the following proposition 2.

Proposition 2. If the level of pso(n) is higher than SL at stage n, the addi-tional safety stock, ssn, required by stage n is given by putting pso(n)−SL inthe left hand side of Eq. (5), and finding the demand standard deviationduring delivery lead times,σd(n),by iterative calculation.

Proof. Consider a case where the level of demand variation is large andthus the probability of stock outmore than the expected service level. Inthis situation, there are two possible outcomes for the realized demandprocess, i.e., the demand realization is equal to λn and a shortage doesnot occur or equal to λn+σV(n) if a shortage occurs after consideringSL. Defining pso(n)−SL to be the probability of having a shortage at theend of planned interval gives the safety stock of the realized order intervalas an iterative calculation of Eq.(5)□.

In addition to the inventory allocation problem, risk sharing isdeveloped in Section 4.1.3 to eliminate the safety stocks in somestockholding points.

4.1.3. Risk sharing through safety stock allocationSection 4.1.2 mentions that the choice of kn also influences the

demand response across the supply chain. Lower kn represents a sluggishresponse and, conversely, higher kn represents a quicker response. Thus,in providing guaranteed lead times and anticipating the forecastingerror, safety stock needs to be allocated to cover demand uncertainty byas many as

ssn ¼ zn:σd nð Þ:ffiffiffiffiffiLn

pð9Þ

we can see from Eq. (9) that σd(n) is the only factor for allocating safetystock across the supply chain at service level zn. Thuswe have Proposition3, as follows.

Proposition 3. Safety stock allocation for stockholding points can beoptimized by shifting the safety stock at downstream (stage n) to upstream(stage n−1), if σd(n−1)bσd(n) holds.

Proof. We can transfer safety stock at stage n to all other stages thathave a direct link to stage n (Φn(n−1)=1) if the safety stock after risksharing is lower than before risk sharing. Thus, we have the followingrelationship:

X1k¼n−1

Φn n−1ð Þhn−1

hnΩn n−1ð Þ > 1 ð10Þ

Ωn n−1ð Þ ¼zk:σd n−1ð Þ:

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiLn þ L n−1ð Þ

q� �zn:σd nð Þ:

ffiffiffiffiffiLn

p� � ð11Þ

where hnand hn−1are the safety stock costs at stages n and n−1respectively.

This article provides a graph to make the analysis in Eq. (11) faster.The decision can be taken simply by multiplying the holding cost ratiohn

hn−1to Ωn(n−1) in Fig. 5. The calculation is applied to all direct links be-

tween stage n and stage k for k=1 to n−1, such that Φn(n−1)=1 ifthere is a direct link, otherwise, Φn(n−1)=0.

Fig. 5 shows that higher demand uncertainty at stage (n−1) ascompared to stage (n) encourages stage (n) to decouple safety stockto stage (n−1) tominimize the total of safety inventory costs betweenthe two stages. Conversely, lower demand uncertainty at stage (n)motivates both stages to keep their own safety stocks. The supplychain can extend this relationship to many stages.

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8

Ln/L(n-1)

( )

( )1,0

1

=−nd

nd

σσ

( )

( )5,0

1

=−nd

nd

σσ

( )

( )0,1

1

=−nd

nd

σσ

( )

( )4,1

1

=−nd

nd

σσ

( )

( )0,2

1

=−nd

nd

σ

σ

Fig. 5. Holding cost ratio at different delivery lead times (Ln/L(n−1)) and demand standard deviation ratios.


4.2. Tactical level optimization

At the tactical level, we define a node as any activity point in the sup-ply chain (material receiving and inspection, work in process (WIP) orintermediate product processing and final product processing). Thus, ineach location the supply chain has at least one activity point (i.e.materialreceiving and inspection forwarehouse). Otherwise, the location has twoactivity points (i.e. material and final product processing facilities),or at most three points (i.e. material, WIP and final product processingfacilities).

For stage n, the replenishment time, LR(n), comprising productionlead time, Lp(n),and delivery lead times, Ln, is given as follows:

LR nð Þ ¼ Lp nð Þ þ Ln: ð12Þ

The delivery rate in all periods must be enough to cover the demandover the upcoming Ln.

The optimization of tactical level comprises demand forecast,production and inventory control. Demand responsiveness in terms ofinventory responses is optimized to minimize the backorders and inven-tory level.

4.2.1. Demand forecast methodWe assume that in each period, t, the observed demand D(t) from

period t is used to issue the demand forecast from period t to t+1 orF(t+1). We assume that it is possible to set an initial inventory levelIn(0) and that F(t)=λ for t≤0 and furthermore F(t)≥0.

Exponential smoothing is used to estimate the future demand. Thereason has its roots in the ARIMA-based demand process model, inwhich the forecast demand F(t) at time t and its mean value λ aredefined as follows [19]:

F 1ð Þ ¼ λþ ε tð Þ; ð13Þ

Problem coding Evaluation of chromosom

fitness Selection

Fig.6. Template for iteration

F tþ 1ð Þ ¼ 1−αað ÞF tð Þ þ αa:D tð Þ þ ε tð Þ; ð14Þ

ε tð Þ ¼ D tð Þ−F tð Þ: ð13Þ

Eq. (13) shows that the demand forecast, F (1), at time t=1dependson themean value of customer demands, λ, and random noise term ε(t)of the time series random variable which represents the forecast error.Eq. (14) shows that the future demand forecast, F(t+1), depends onthe current demand forecast, F(t) and ε(t). The value of αastands forthe smoothing constant that is obtained from time to adjust the demandresponse time Ta for Ta=1/αa. When 0bαab1, the demand process is anon-stationary process. When αa=1, Eq. (14) shows that the demandforecast is correlated to the previous demand and the demand processresembles a random walk.

4.2.2. Production and inventory controlProduction and inventory control is accomplished by adjusting the

production rates μn(t) at stage n to minimize final product inventoryIn(t) and the WIP inventory IWIP nð Þ tð Þ at stage n and time t. The ad-justment requires the supply chain to fulfill the demand from productinventory according to delivery order at stage n, qn(t) , and considersproduction capacity constraint Kn at stage n as follows:

μn tð Þ ¼ min K; F t þ 1ð Þ þ ΔIn tð Þ þ ΔWIPn tð Þ½ �; ð15Þ

qnðtÞ ¼ min I nð Þ tð Þ;Fðtþ 1Þ þ DðtÞ�qn t�Lnð Þð ÞLnTq nð Þ

!; ð16Þ

ΔIn tð Þ ¼μn tð Þ � Lp nð Þ� �

� qn tð Þ � Lnð ÞTi nð Þ

; ð17Þ

result Recombination and mutation

Iteration stop

of a genetic algorithm.

Table 1GA parameters.

Population 100Generation 200Crossover % 0.7Crossover method Elitism (Recommended)Mutation rate 0.000002Fitness scaling Linear normalize


ΔWIP nð Þ tð Þ ¼IWIP nð Þ tð Þ− μn tð Þ � Lp nð Þ

� �TW nð Þ

: ð18Þ

Inminimizing the backorders BOn(t) at stage n, the supply chainmustmeet the next day demand forecast F(t+1) by optimizing time to adjustqn(t) level Tq and time to adjust product inventory Ti and WIP inventoryTW [8,36,40]. The response parameters (Tq,Ti,TW) represent the willing-ness of the stockholding points to meet the demand changes.

4.2.3. Optimization of the response parametersThe response parameters are optimized to minimize BOn(t) for

BOn(t)=Dt−μn(t), at minimum difference between qn(t) and μn(t).Since the forecast of the demand is centralized, the optimizationmodel can be obtained by inserting Eq. (17) to Eq. (18) into Eq. (15).

However, BOn(t), IWIP nð Þ tð Þ, F(t) and εn(t) are the four independentvariables that are always changing. Here, the optimization should coverthe worst scenario, for instance, highest standard deviation of demandσd(n) (to invest in material safety stock IWIP nð Þmax ¼ zpσd nð Þ:

ffiffiffiffiffiLp

pfor

zprepresents the maximum allowable backorders and Lp representsthe lead time of the production process) and lowest BOn(t)=Dt−μn(t)(in this article, 1% backorders is taken as the customer risk α=0.01).Manipulating Eq. (15) to Eq. (18) to get BOn(t) we then have the follow-ing optimization problem:

minTa ;TW nð Þ ;Ti nð Þ ;Tq nð Þ

XNn¼1

ΔBOn ¼XNn¼1

σd nð Þ1þTa

1− LnTi nð Þ

� �− 1

Ti nð Þ

BO nð Þmax :Ln2

Tq nð Þþ IWIP nð Þmax

TW nð Þ

1−Lp nð ÞTW nð Þ−Ti nð ÞTW nð ÞTi nð Þ

� �ð19Þ

1≤Ta; Ti nð Þ; TW nð Þ; Tq nð Þ≤5 ð20Þ

Tq nð Þ≤Ti nð Þ ð21Þ

Table 2Analysis results for allocating the safety stock of the problem example (without assembly s

Stage (n) Component/Product

λn μn ρn σd(n) σAn)2

3 Product ABDEGI 276 330 0.84 132 1E−063 Product ABDEHI 275 280 0.98 203 2E−063 Product ABDFGI 400 410 0.98 282 1E−063 Product ABDFHI 300 310 0.97 203 2E−063 Product ACDEGI 310 320 0.97 210 2E−063 Product ACDEHI 200 220 0.91 111 3E−063 Product ACDFGI 500 520 0.96 328 6E−073 Product ACDFHI 800 810 0.99 620 3E−072 Back 3060 4200 0.73 1247 9E−092 Seat 3060 4000 0.77 1313 1E−082 Under frame 3060 3300 0.93 1782 1E−081 Seat frame A 1250 2000 0.63 445 4E−081 Seat frame B 1810 2000 0.91 995 4E−082 Upholstery 3060 3100 0.987 2358 2E−081 Stand A 1060 1100 0.964 702 1E−071 Stand B 2000 2010 0.995 1722 5E−081 Pad 1485 1500 0.990 1182 9E−081 Wheel 1575 1580 0.997 1420 9E−081 Armrest 3060 3100 0.99 2358 2E−08

XNn¼1

ΔBOn≥0: ð22Þ

Eq. (20) is used to constrain the limit of responsiveness. Eq. (21)signifies that the delivery response is quicker than the productionresponse. This signifies that the supplier (upstream) commits to pro-viding a 100% service level. Considering that the backorder is alwaysa positive value, Eq. (22) limits the solution in Eq. (19) to be positive.Section 4.2.4 is then used to find the optimum response parameters.

4.2.4. GA for parameter selection to give optimum value in the responseparameters

The XLBit Genetic algorithms add-in of MS-Excel spreadsheet isused to solve the non-linear supply chain dynamics in terms of inven-tory, production and distribution decisions. This free general-purposesimulator combines system dynamics with some aspects of discrete-event simulation, and embeds a meta-heuristic optimization enginewithin a Monte Carlo simulation framework, which is well-suitedfor modeling time-dependent conditions or processes. The Excelspreadsheet also provides easy access for the user to edit the mathe-matical models in Section 4.2.3.

The solution to a problem is called a chromosome, which simplyrepresents the parameters to be optimized. A GA creates an initialpopulation (a collection of chromosomes), evaluates this populationand then evolves the population through multiple generations inthe search for a good solution to the problem in hand. GA will stopthe iteration process whenever the regeneration process has reacheda specified value [16,23]. Basically, GA has the structure described inFig. 6.

Table 1 shows the GA parameters for optimizing the parametersTa, Ti(n), TW(n) and Tq(n). Table 2 exhibits the input parameters whichare used during the simulation. The different settings of lead timesare used within the supply chain to observe the effect of lead timeson the performance indicators.

4.3. Results and analysis

In analyzing the proposed inventory allocationmodel, we benchmarkthe original supply chain configuration before and after the assemblysequence planning. Part variants in Fig. 2 are used to produce eightproduct variants in the product family. The parts are coded into capitalletters, back (A), Seat frame A (B), Seat frame B (C), Upholstery (D),Stand A (E), Stand B (F), Pad (G), Wheel (H), Armrest (I). In Table 2,

equence planning).

σn2 Ln kn hn σVn In pso Inventory

allocation

1E−04 1 4.3 0.02 110 276 7.6E−07 No2E−04 3 1.8 0.01 110 275 1.1E−02 Yes8E−05 3 2.0 0.01 160 400 2.6E−03 No1E−04 3 2.2 0.01 120 300 1.7E−03 No1E−04 3 2.2 0.01 124 310 1.8E−03 No3E−04 2 3.2 0.02 80 200 5.5E−05 No5E−05 3 2.3 0.00 200 500 5.2E−04 No2E−05 4 1.7 0.00 320 800 1.2E−02 Yes1E−06 1 6.0 0.00 1224 3060 1.2E−14 No1E−06 1 5.4 0.00 1224 3060 6.0E−13 No1E−06 2 2.9 0.00 1224 3060 1.6E−06 No6E−06 1 7.9 0.01 500 1250 1.6E−15 No4E−06 2 3.3 0.00 724 1810 3.7E−07 No1E−06 4 1.7 0.00 1224 3060 4.2E−03 No3E−06 3 2.3 0.00 424 1060 2.4E−04 No3E−06 5 1.3 0.00 800 2000 5.3E−02 Yes3E−06 4 1.6 0.00 594 1485 1.3E−02 Yes3E−06 5 1.2 0.00 630 1575 1.3E−01 Yes1E−06 4 1.7 0.00 1224 3060 4.2E−03 No

Table 3Analysis results for allocating the cycle inventory of the problem example (after assembly sequence planning).

Component/product

λn μn ρn σd(n) σAn)2 σn

2 Ln kn hn σVn In pso Inventoryallocation

Product ABDEGI 276 810 0.34 66 5E−07 1E−04 0 17.3 0.06 110.4 276 1.5E−20 NoProduct ABDEHI 275 810 0.34 66 5E−07 1E−04 0 17.3 0.06 110 275 1.2E−20 NoProduct ABDFGI 400 810 0.49 120 3E−07 6E−05 1 11.0 0.03 160 400 5.8E−17 NoProduct ABDFHI 300 810 0.37 76 5E−07 9E−05 0 15.7 0.05 120 300 2.0E−19 NoProduct ACDEGI 310 810 0.38 80 4E−07 9E−05 0 15.1 0.05 124 310 8.4E−20 NoProduct ACDEHI 200 810 0.25 40 7E−07 2E−04 0 24.5 0.12 80 200 3.2E−22 NoProduct ACDFGI 500 810 0.62 176 3E−07 4E−05 1 8.0 0.02 200 500 7.1E−14 NoProduct ACDFHI 800 810 0.99 620 3E−07 2E−05 4 1.7 0.00 320 800 1.2E−02 YesBack 3060 4200 0.73 1247 9E−09 1E−06 1 6.0 0.00 1224 3060 1.2E−14 NoSeat 3060 4000 0.77 1313 1E−08 1E−06 1 5.4 0.00 1224 3060 6.0E−13 NoUnder frame 3060 3300 0.93 1782 1E−08 1E−06 2 2.9 0.00 1224 3060 1.6E−06 NoSeat Frame A 1250 4000 0.31 536 6E−08 5E−06 1 5.4 0.00 500 1250 3.2E−11 NoSeat Frame B 1810 4000 0.45 515 1E−08 3E−06 1 12.3 0.01 724 1810 2.0E−25 NoUpholstery 3060 3100 0.987 2358 2E−08 1E−06 4 1.7 0.00 1224 3060 4.2E−03 NoStand A 1060 3300 0.321 1860 4E−07 2E−06 1 1.0 0.00 424 1060 9.6E−01 YesStand B 2000 3300 0.606 1860 6E−08 2E−06 1 1.2 0.00 800 2000 2.2E−01 YesPad 1485 3300 0.450 594 4E−08 2E−06 1 6.3 0.00 594 1485 1.7E−13 NoWheel 1575 3300 0.477 630 3E−08 2E−06 1 6.3 0.00 630 1575 1.2E−13 NoArmrest 3060 3100 0.99 2358 2E−08 1E−06 4 1.7 0.00 1224 3060 4.2E−03 No


proposition 2 is applied to the decision whether or not to allocate theparts and product safety stocks to the supply chain without assemblysequence planning. The parameters of LR(n) and Ln, λn, and μn are statedin advance. The volatility is set at 40% of λn. In Table 3, the same opera-tions are applied to the supply chain configuration which follows as-sembly sequencing.

4.3.1. Assembly sequencing and the supply chain reconfigurationEach part in the product structure (Fig. 2) is represented as one

stage in the supply chain. However, the two dashed line which linkto the same upstream part share the same stage since those two linesare options for the customers. The DSM matrix (Fig. 7) shows that theseat and seat frame should be assembled together in one stage. Similarly,the stand and support should also be assembled together in the samestage. Two benefits of the assembly sequencing re that they reducethe number of workstations from 11 stages to 6 stages and reducethe interdependencies between them.

4.3.2. Inventory allocationTables 2 and 3 show that the assembly sequence planning is capable

of eliminating safety stocks in some of stockholding points at stage 3(product ABDEHI) and stage 1 (wheel, pad, stand A), and furthermoreminimizes the total safety stock costs for all the activity points within

Bac

k

Und

erfr

ame

Arm

res

t

Uph

olst

ery

Seat

Seat

fra

me

A

Seat

Fra

me

B

Stan

d A

Stan

d B

Pad

Whe

el

1 3 4 5 2 6 7 8 9 10 11Back 1 1 Underframe 3 3 Arm rest 4 4Upholstery 5 5Seat 2 2 1 1 Seat frame A 6 1 1 6Seat Frame B 7 1 1 7Stand A 8 1 8 1 1Stand B 9 1 9 1 1Pad 10 1 1 1 10Wheel 11 1 1 1 11

Fig. 7. The assembly sequence planning for supply chain configuration.

the supply chain. Tables 2 and 3 summarize pso(n) and if pso(n)>(1−SL), an additional safety stock will be required (in our example SL=0.99). Placing the safety stock iteratively according to Eq.(5) willincrease the inventory position which is only slightly greater thanaverage demand. In addition, Table 3 shows that except productACDFHI, all the rest product variants are assigned as safety stockfree locations. This implies that the supply chain can make a savingon those products and increase the agility and production capacitiestowards process standardization. Process standardization reduces thenumber of activity points within the supply chain by clustering theinterdependence parts into one location.

4.3.3. Risk sharing in terms of safety stock allocationFurthermore, Proposition 3 is used to share the risk within the

supply chain. Risk sharing reduces the total inventory costs of theproducts and their supporting parts. Fig. 4 is used to decide on risksharing by obtaining a lead time ratio between stage n and stage n−1,and a standard deviation ratio between stage n and stage n−1.

Table 4 shows the application of Fig. 5 and Eq. (5). Product ACDEHIis the only stage which can share its safety stock to its parts (back,seat, under frame, upholstery and armrest). It is shown by iterativelycalculating Eq. (5) that the holding costs are the determining factor indeciding on risk sharing. In addition, σV(n) also affects the risk sharingdecision. It is shown that the product or parts safety stocks are sharedif the demand standard deviation at stage n is lower than stage k fork=1 to n−1.

Table 4Safety stock allocation for stockholding points.

Component/Product

λn σd

(Eq. 5)σd (Tables2 and 3)

hn Safety stockcosts Eq. (5)

Safety stock costs(Table 2 and 3)

ProductACDFHI

ProductACDFHI

ProductACDEHI

800 620 620 1 2030 2030

Back 3060 420 1247 0.25 181 537Seat 3060 585 1313 0.25 265 595Underframe

3060 700 1782 0.25 431 1096

Armrest 3060 1020 2358 0.25 830 1919Safety stock allocation Share Not share

Table 5Optimum value of Ta, Ti(n), TW(n) and Tq(n).

Stage Ta Ti Tw Tq

Product 4.9 3 2.1 1.3Back 4.9 2.5 2 1.2Underframe 4.9 2.5 2 1.2Arm rest 4.9 2.5 2 1.1Upholstery 4.9 2.5 2 1.2Seat frame 4.9 3.2 1.6 0.6Underframe 4.9 3.4 1.2 0.6


4.3.4. Optimum value of response parameters Ti(n), TW(n) and Tq(n).The optimum inventory response parameters to minimize the

backorders and inventory level are presented in Table 5. The responsesare applied to the new configuration of supply chain (Fig. 7) and wehave six workstations.

5. Model validation and management decision-making

We chose products ABDEHI and ABDEGI in the simulation, since bothproducts act as stockholding (before clustering) and non-stockholdingpoints.

5.1. Model validation

The simulation validates the amount of safety stock of the twoadjacent stages in the supply chain (the final product at stage nand its direct upstream at stage n−1). The safety stock is countedas the required safety stock for products ABDEHI and ABDEGI. Inorder to validate the proposed analytical model, GoldSim simulationsoftware developed by the GoldSim Technology Group is used. Thisgeneral purpose simulator combines system dynamics with someaspects of discrete-event simulation, and embeds a dynamic simula-tion engine within a Monte Carlo simulation framework, which iswell-suited for modeling time-dependent conditions or processes.The analytical models and discrete event simulation are tested atdifferent levels of demand volatility. All other parameters in the

Table 6Inventory allocation between the analytical model and simulation for product ABDEHI.

Demand process Product ABDEHIsafety stock

Simulation Analyticalmodel

Difference

σVA=0.4; λn=200 Stage-n 200 191 9Stage k 190 235 −45


σVA=0.4; λn=275 Stage-n 259 263 −4Stage k 235 248 −13


Average difference −18

Table 7Inventory allocation between analytical model and simulation for product ABDEGI.

Demand process Product ABDEGIsafety stock


Difference

σVA=0.4; λn=200 Stage-n 0 0 0Stage k 265 263 2




Average difference 2

model follow Tables 2 to 5.Moreover, some failure rates in the assemblyplants are introduced to accommodate the effect of process quality tothe backorders.

The validation applies the ANOVA (Table 12). In Tables 6 and 7, theanalytical models of safety stock are not statistically different from thesimulation safety stocks at a significant 5% level, where average P-values are 0.843 in all stages. The results show that the analyticalmodel is capable of hedging demand variations. However, the resultsalso show that the product ABDEGI inventory allocation by simulationis slightly higher than the analytical model and the opposite featureapplies to product ABDEHI safety stock allocation. This signifies thatproduct ABDEHI has higher inventory variance than product ABDEGIbecause of a higher number of stockholding points.

In addition to safety stock analysis, Tables 8 and 9 make it clearthat the simulation has a different order rate at different demand vol-atility levels. The results imply that in non-stationary demand, the dy-namic lot sizing is important in providing 100% guarantee lead times(Tables 10 and 11). Furthermore, in non-stationary demand, a morefrequent demand sampling period is suggested, either for the simulationor the analytical model.

5.2. Management decision-making

In addition to the DSS model validation, Tables 6 and 7 illustratethat either at lower demand volatility (σVA=0.2) or higher demandvolatility (σVA=0.4), the product ABDEHI inventory allocation forthe downstream stage (n) is always higher than for the upstreamstage (n−1). This implies that the proposed analytical and simulationmodels are capable of sharing the risk in terms of safety stock allocationand improving the quality of information about demand. Tables 8 and 9summarize that the magnification of the order is eliminated withoutcreating excessive backorders (Tables 10 and 11). Thus, the proposedDSS model contributes to current DSS commercial software (i.e. SAPAPO, I2 Rhythm) [1,21,34] by optimally designing supply chainnetworks, and at the same time allocating the inventory with thefewest possible backorders.



Difference






Demand process Product ABDEGIsafety stock


Difference






Table 8Order rate between analytical model and simulation for product ABDEHI.

Demand process Product ABDEHIorder rate


Difference Demand process Product ABDEHIsafety stock


Difference

σVA=0.4; λn=200 Stage-n 220 200 20 σVA=0.2; λn=200 Stage-n 185 200 −15Stage k 198 200 −2 Stage k 185 200 −15

σVA=0.4; λn=250 Stage-n 218 250 −32 σVA=0.2; λn=250 Stage-n 238 250 −12Stage k 248 250 −2 Stage k 242 250 −8



Table 9Order rate between analytical model and simulation for product ABDEGI.

Demand process Product ABDEGIorder rate


Difference Demand process Product ABDEGIorder rate


Difference

σVA=0.4; λn=200 Stage-n 203 200 3 σVA=0.2; λn=200 Stage-n 203 200 3Stage k 195 200 −5 Stage k 205 200 5

σVA=0.4; λn=250 Stage-n 251 250 1 σVA=0.2; λn=250 Stage-n 271 250 −11Stage k 243 250 −7 Stage k 227 250 −23

σVA=0.4; λn=275 Stage-n 273 275 −2 σVA=0.2; λn=275 Stage-n 271 275 −4Stage k 277 275 2 Stage k 217 245 −28

σVA=0.4; λn=310 Stage-n 315 310 5 σVA=0.2; λn=310 Stage-n 311 310 1Stage k 310 310 0 Stage k 317 310 7


We need to be cautious about the analytical model, since it assumesthat increasing demand volatility should be responded to linearly byproviding extra safety stock. However, the simulation shows that thetrend is not linear. Furthermore, with higher demand volatility, thebackorder level of the simulation is a little higher than that of theanalytical model. We suggest a shorter sampling period of demandto reduce this difference, or using simulation to investigate thetrue pattern of the demand process. We can validate our suggestionby measuring order rates in the three stage supply chain (Tables 10and 11). Furthermore, a shorter sampling period implies that alower number of stockholding points in the supply chain make thesupply chain more autonomous. The reason is that the length of thesupply chain is reduced and the supply chain configuration becomessimpler than before. The simpler supply chain brings the opportunitytomeet the production and distribution decisions in terms of economiesof scale of transportation cost. Below, the contributions of the DSS tocurrent commercial DSS are summarized (Table 13).

6. Concluding remarks

The proposed DSS model improves supply chain efficiency throughsupply chain reconfiguration and inventory allocation [18,19]. Fur-thermore, risk sharing adds benefit to the supply chain by reducingthe safety stock investment. Proposition 1 endorses the importanceof product functional analysis and assembly sequencing to supply

Table 10Backorders between analytical model and simulation for product ABDEHI.

Demand process Product ABDEHIorder rate


Difference


σVA=0.4; λn=250 Stage-n 1 2 −1Stage k 2 2 0

σVA=0.4; λn=275 Stage-n 3 4 −1Stage k 4 4 0


chain reconfiguration [22]. Proposition 2 suggests a condition forsafety stock allocation, and Proposition 3 guides us to decide on theconditions in which the safety stock can be decoupled. The safetystock decoupling can minimize the effect of demand uncertaintyand inventory cost [4,31]. The model validation benchmarks the pro-posed analytical model against the simulation model, at the samemodel parameters. The statistical analysis for stockholding andnon-stockholding points shows that the analytical and simulationmodels are not statistically different in terms of safety stock levels,order rates and backorders.

The supply chain reconfiguration reduces safety stock distribution ata lower number of stockholding points. Furthermore, the new supplychain configuration makes the supply chain more autonomous. Thelevel of autonomy is higher when the interdependencies among stagesin the supply chain are minimized. In addition, autonomy encouragessupplier and buyer coordination. Finally, the value of the demandforecast is higher at a higher level of supplier and buyer coordination[3].

However, our model has two drawbacks: first, the smoothingconstant of the demand forecast is not related to supply and demanduncertainty. However, this relation is important in deciding the levelof responsiveness at which the safety stock can be eliminated. Thesecond drawback is the limitation of the analytical model in beingsensitive to backorders. It is shown in Tables 10 and 11 that the sim-ulation yields a variety of results on backorders. In the future, the



Difference





Table 11Backorders between analytical model and simulation for product ABDEGI.

Demand process Product ABDEGIorder rate


Difference Demand process Product ABDEGIorder rate


Difference


σVA=0.4; λn=250 Stage-n 3 2 1 σVA=0.2; λn=250 Stage-n 1 2 −1Stage k 2 2 0 Stage k 1 2 −1

σVA=0.4; λn=275 Stage-n 3 4 −1 σVA=0.2; λn=275 Stage-n 2 2 0Stage k 3 4 −1 Stage k 1 2 −1


Table 12ANOVA test of analytical model and simulation differences.

System parameter F P-value

ProductABDEHI

ProductABDEGI

ProductABDEHI

ProductABDEGI

Safety stock difference (analyticalmodel and simulation model)

0.107 0.005 0.746 0.939

Order rate (analytical modeland simulation model)

0.312 0.017 0.58 0.896

Backorders (analytical modeland simulation model)

3.253 4.171 0.081 0.612

Table 13Comparison of other operational level Supply Chain Optimization software.

SAP APO [23] I2 RHYTIM [23] The proposed DSS

DemandPlanning

1. Promotionalplanning, causalanalysis

2. life cycleconcept

3. Collaborativeforecasting

Forecasting processthrough statisticalmethods andmultiple inputsfrom differentorganization units

1. Centralizedforecast

2. Supply chainreconfiguration

3. Aligning productand supply chainreconfiguration

Available topromise(ATP)

1. Rule-based ATP2. Multi-level

multi site ATP.3. Capable to

Promise (CTP)function

User friendlyproduct catalogand productconfiguration

1. Inventoryallocation

2. Risk sharing3. Implement

APIOBPCS

DistributionPlanning

1. Transportationplanning andvehicle schedulingto multi-siteoptimization byGA and additionalheuristiccomponents

2. VMI Support3. Demand–supply

synchronization

Use transportationmodeler, optimizerand manager orderby customer serviceand financialsettlement

1. VMI application.2. The application

of APIOBPCS3. Optimal

response ondemand andsupplysynchronization


effect of demand sampling frequency must be considered. Thus, theanalytical model should optimize the demand forecast method inorder to obtain the minimum backorders.

Acknowledgments

The authors are most grateful to two anonymous reviewers fortheir constructive and helpful comments which helped to improvethe presentation of the paper considerably.

This research is supported by FUDGE Project and Fimmec undercontract agreement with the Department of Production, Universityof Vaasa, Finland.

References

[1] S. Bagchi, S. Buckley, M. Ettl, G.Y. Lin, Experience using IBM supply chain simulator,in: D.J. Medeiros, E.F. Watson, J.S. Carson, M.S. Manivannan (Eds.), Proceedings ofthe 1998 Winter Simulation Conference, 1998, pp. 1387–1394.

[2] K.R. Balachandran, S. Radhakrishnan, Quality implications of warranties in asupply chain, Management Science 51 (8) (2005) 1266–1277.

[3] S. Balan, P. Vrat, P. Kumar, Information distortion in supply chain and its mitigationusing a soft computing approach, Omega: The International Journal of ManagementScience 37 (2009) 282–299.

[4] G.P. Cachon, The allocation of inventory risk in a supply chain: push, pull, andadvance-purchase discount contracts, Management Science 50 (2) (2004)222–238.

[5] H.K. Chan, F.T.S. Chan, Comparative study of adaptability andflexibility in distributedmanufacturing supply chains, Decision Support Systems 48 (2010) 331–341.

[6] T.C.E. Cheng, C. Gao, H. Shen, Production planning and inventory allocation of asingle-product assemble-to-order system with failure-prone machines,International Journal of Production Economics 131 (2) (2011) 604–617.

[7] T.W. Chien, A. Balakrishnan, R.T. Wong, An integrated inventory allocation andvehicle routing problem, Transportation Science 23 (2) (1989) 67–76.

[8] J. Dejonckheere, S.M. Disney, M.R. Lambrecht, D.R. Towill, The impact of informationenrichmenton theBullwhip effect in supply chains: a control engineeringperspective,European Journal of Operational Research 153 (2004) 727–750.

[9] Q. Dong, D. Whitney, Designing a requirement driven product developmentprocess, 13th International Conference on Design Theory and Methodology,ASME, Pittsburgh, 2001.

[10] X. Du, J. Jiao, M.M. Tseng, Architecture of product family for mass customization,ICMIT conference, 2000, pp. 437–443.

[11] A. Federgruen, P. Zipkin, A combined vehicle routing and inventory allocationproblem, Operations Research 32 (5) (1984) 1019–1037.

[12] E. Feitzinger, H.L. Lee, Mass customization at Hewlett Packard: the power ofpostponement, Harvard Business Review, January–February 1997, pp. 116–121.

[13] G. Ferrer, Open architecture, inventory pooling and maintenance modules,International Journal of Production Economics 128 (1) (2010) 393–403.

[14] C. Gao, H. Shen, T.C.E. Cheng, Order-fulfillment performance analysis of anassemble-to-order system with unreliable machines, International Journal ofProduction Economics 126 (2) (2010) 341–349.

[15] E. Gebennini, R. Gamberini, R.Manzzini, An integrated production–distributionmodelfor the dynamic location and allocation problem with safety stock optimization,International Journal of Production Economics 122 (1) (2009) 286–304.

[16] M. Gen, R. Cheng, Genetic Algorithms and Engineering Design, John Wiley andSons Inc, New York, 1997.

[17] S.C. Graves, B.T. Tomlin, Process flexibility in supply chains, Management Science49 (7) (2003) 907–919.

[18] S.C. Graves, S.P. Willems, Optimization the supply chain configuration for newproducts, Management Science 51 (8) (2005) 1165–1180.

[19] S.C. Graves, S.P. Willems, Strategic inventory placement in supply chains: nonsta-tionary demand, Manufacturing and Service Operations Management 10 (2)(2008) 278–287.

[20] D. Gross, C.M. Harris, Fundamental of Queueing Theory, John Willey and Sons,New York, 1974.

[21] P. Helo, Y. Xiao, J.R. Jiao, A web based logistics management system for agile supplydemand network design, Journal of Manufacturing Technology Management 17 (8)(2006) 1058–1077.

[22] M. Jin, Y. Luo, S.D. Eksioglu, Integration of production sequencing and outboundlogistics in the automotive industry, International Journal of Production Economics113 (2) (2008) 766–774.

[23] R. Klein, Genetic algorithm, in: H. Stadtler, C. Kilger (Eds.), Supply ChainManagementandAdvancedPlanning:Concepts,Models, Software andCase Studies, Second Edition,Springer-Verlag, Berlin, 2002, pp. 404–410.

[24] H.L. Lee, Effective inventory and service management through product andprocess redesign, Operations Research 90 (3) (1996) 361–367.

[25] H.L. Lee, C. Billington, Managing supply chain inventories: pitfalls and opportunities,Sloan Management Review 33 (3) (1992) 65–73.

[26] E. Leven, A. Segerstedt, Inventory controlwith amodifiedCrostonprocedure andErlangdistribution, International Journal of Production Economics 90 (2004) 361–367.

[27] S. Li, B. Lin, Accessing information sharing and information quality in supply chainmanagement, Decision Support Systems 42 (2006) 1641–1656.


[28] H.Ma, R. Davidrajuh, An iterative approach for distribution chain design in agile virtualenvironment, Industrial Management and Data Systems 105 (6) (2005) 815–834.

[29] J.J. Neale, S.P. Willems, Managing inventory in supply chains with nonstationarydemand, Interfaces 39 (5) (2009) 388–399.

[30] S. Novak, S.D. Eppinger, Sourcing by design: product complexity and the supplychain, Management Science 47 (1) (2001) 189–204.

[31] G. Pishchulov, K. Richter, Inventory rationing and sharing in pre-sell distributionwith mobile communication technologies, International Journal of ProductionEconomics 121 (2) (2009) 584–600.

[32] C.J. Robinson, M.J. Maholtra, Defining the concept of supply chain qualitymanagement and its relevance to academic and industrial practice, InternationalJournal of Production Economics 96 (2005) 315–337.

[33] J. Shang, P.R. Tadikamalla, L.J. Kirsch, A decision support system for managinginventory at GlaxoSmithKline, Decision Support Systems 46 (2008) 1–13.

[34] H. Stadtler, Supply chainmanagement— an overview, in: H. Stadtler, C. Kilger (Eds.),Supply ChainManagement andAdvancedPlanning: Concepts,Models, Software andCase Studies, Second Edition, Springer, Verlag, Berlin, 2002, pp. 7–29.

[35] N.P. Suh, Axiomatic Design: Advanced and Application, Oxford University Press, NewYork, 2001.

[36] H. Topaloglu, S. Kunnumkal, Approximate dynamic programming methods for aninventory allocation under uncertainty, Naval Research Logistics (2006) 822–841Doi:10.10027nav.20164.

[37] D.R. Towill, S.M. Disney, The effect of vendor managed inventory (VMI) dynamicson the Bullwhip Effect in supply chains, International Journal of ProductionEconomics 85 (1) (2003) 199–215.

[38] P. van der Vlist, R. Kuik, B. Verheijen, Note on supply chain integration in vendor-managed inventory, Decision Support Systems 44 (2007) 360–365.

[39] W.T. Wang, H.M. Wee, H.S.J. Tsao, Revisiting the note on supply chain integration invendor-managed inventory, Decision Support Systems 48 (2010) 419–420.

[40] D.E. Whitney, Manufacturing by design, Harvard Business Review, July–August 1988,pp. 83 –91.

[41] J. Wikner, M.M. Naim, M. Rudberg, Exploiting the order book for mass customizedmanufacturing control systems with capacity limitation, IEEE Transactions onEngineering Management 54 (1) (2007) 145–155.

[42] Y. Yao, P.T. Ever, M.E. Dresner, Supply chain integration in vendor managedinventory, Decision Support Systems 43 (2007) 663–674.

[43] A. Yassine, D. Falkenburg, K. Chelst, Engineering design management: an informa-tion structure approach, International Journal of Production Research 37 (1999)2957–2975.

Yohanes Kristianto is now a Project researcher in Industrial Management at University of
VU aasa, Finland. He earned a PhD in Industrial Management fromDepartment of Production,niversity of Vaasa, Finland. His research interests are in the area of supply-chain strategy/
management and production/operations management. He has 11 years of working experi-ence in the areas of quality management, logistics and process engineering.

Angappa Gunasekaran, PhD, is a Professor of Operations Management and theChairperson of the Department of Decision and Information Science at the CharltonCollege of Business, the University of Massachusetts-Dartmouth, USA. He has over200 articles published in 40 different peer-reviewed journals. He is on the EditorialBoard of over 20 journals and the Editor of several journals in the fields of operationsmanagement and information systems. He is currently interested in researching thedecision support systems in logistics and supply chain management. He is also theDirector of the Business Innovation Research Center at the University of Massachusetts-Dartmouth.

Petri Helo is a Research Professor and the head of Logistics Systems Research Group,Department of Production, University of Vaasa. He earned a PhD in IndustrialManagement from Department of Production, University of Vaasa, Finland. His researchaddresses themanagement of logistics processes in supply demand networks, which takeplace in electronics, machine building, and food industries. Dr. Helo is also partner atWapice Ltd, a solution provider of sales configurator systems and tailored masscustomization solutions; and visiting professor at Kasetsart University in Thailand, andFH Kiel in Germany.

Dr. Maqsood Sandhu is working at the Department of Business Administration, Facultyof Business and Economics at United Arab Emirates University, Al Ain since August2008. He earned a PhD from Swedish School of Economics and Business Administrationin Management. Sandhu has been working over five years in project-based industry. DrMaqsood has over 15 years' experience in academia and industry. He has been very activein initiating collaborative partnerships amongst academia, industry and policy makingbodies. He is Author or Co-author of over 25 international journals articles and haspresented over 50 papers and published about 40 articles in international conferenceproceedings. Currently, he is interested in doing research in the areas of project manage-ment, supply chain management and international entrepreneurship.

decision support systems - university of...

Documents