Applied Mathematical Modelling xxx (2013) xxx–xxx

Contents lists available at ScienceDirect

Applied Mathematical Modelling
journal homepage: www.elsevier.com/locate/apm

A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths
Salah Alden Ghasimi ⇑, Rizauddin Ramli, Nizaroyani Saibani
Department of Mechanical and Materials, Engineering Faculty of Engineering and Built Environment, Universiti Kebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia

a r t i c l e

i n f o

a b s t r a c t
The competitive environment of global markets has forced many manufacturers to select the most appropriate supply chain network (SCN) for reduction of total costs and wasted time. Cost reduction and selection of the appropriate length of each period are two important factors in the competitive market that are often not addressed comprehensively by researchers. In our study, we proposed genetic algorithms (GAs) for optimising a novel mathematical model of the defective goods supply chain network (DGSCN). In the proposed model, we assumed that all imperfect-quality products are not repairable, whereas those considered as scrap are directly sold to customers at a low price. The objective of the proposed model is to minimise the costs of production, distribution, holding and backorder. In addition to minimising the costs, the model can determine the economic production quantity (EPQ), the appropriate length of each cycle (ALOEC) and the quantities of defective products, scrap products and retailer shortages using Just-In-Time logistics (JIT-L). We used the GAs and a Cplex solver with probability parameters and various dimensions for validation of the studied model in real-life situations, and we compared the outputs to demonstrate the performance of the model. Additionally, to identify the appropriate length of each cycle (ALOEC), we needed to solve the model using exact parameters and same dimensions and prefer to use Lingo for this application. Ó 2013 Elsevier Inc. All rights reserved.

Article history: Received 9 October 2012 Received in revised form 23 May 2013 Accepted 8 August 2013 Available online xxxx Keywords: Cplex Genetic algorithm Length of each-cycle Supply chain Lingo

1. Introduction To survive in the current competitive environment of global markets, manufacturers have been forced to choose an appropriate supply chain network (SCN) for cost reduction in their inventory systems [1–5].One aspect of the supply chain (SC) is logistics, which is the process of planning, implementing and controlling the flow and storage of goods or services. The SC also involves the related information flow from a point of origin to a point of consumption for the purpose of conforming to customer requirements [6,7]. Furthermore, in recent years, one of the most important developments in manufacturing systems has been the Just-inTime (JIT) system, which has been demonstrated to increase productivity. The JIT approach was initially used at Toyota in the 1960s and subsequently spread throughout the world. JIT has been used by many American companies (i.e., Xerox, GE, and Hewlett Packard) to decrease the level of inventory [8–10]. Kawtummachai and Van Hop [11] considered price and JIT delivery as objective programming parameters, and their results showed that the JIT delivery date lies in a different logical domain. Shyur and Shih [12] applied JIT to delivery, production quality, price/cost, facilities, technology, response to

⇑ Corresponding author. Tel.: +60 126267952.
E-mail address: salah340@eng.ukm.my (S.A. Ghasimi). 0307-904X/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.apm.2013.08.023

Please cite this article in press as: S.A. Ghasimi et al., A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths, Appl. Math. Modell. (2013), http://dx.doi.org/10.1016/j.apm.2013.08.023

imperfect goods production is quite normal in manufacturers. Another regularly adopted model is the economic production quantity model for defective goods [24]. They evaluated their proposed model by comparing the results of GAs with that of a GAMS optimiser. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. and the third objective was to minimise the total amount of raw material. decision vari- Scrap goods Defective goods store Perfect goods Manufacturers Distributors Retailers Fig. they presented a solution approach based on steady-state genetic algorithms (ssGA). Ramazani et al. For evaluation of the ssGA. however.08. In this paper. these researchers did not consider the reworking function of defective goods. Additionally. Moreover. Lee and Kim [21] proposed a hybrid method for solving the optimisation problems of production–distribution planning in Supply Chain Management (SCM). retailer shortage. In their model. We used the GAs and a Cplex solver with probability parameters and various dimensions for validation of the studied model in real-life situations.org/10. [27] proposed a two-objective stochastic model for the supply chain to evaluate the relationship between profit and quality. to identify the appropriate length of each cycle (ALOEC). reworking of defective goods can be quite profitable if the cost of production is high or if raw materials are limited and costly [16–18]. increasing the quality and selecting the appropriate length of each-cycle (ALOEC) are important aspects in a competitive market. and we compared the outputs to demonstrate the performance of the model. Amasaka [14] presented a new JIT method that is not limited only to TPS or TQM but also includes software and hardware systems. limitations. less attention has been comprehensively focused on this issue. and exit quality criteria. Farahani and Elahipanah [15] solved the JIT distribution problem for a SCM by presenting a model with two target functions designed for a three-level distribution network in a supply chain. (2013).apm. and the machine types. In their model. Most of the classical production models assume that all goods generated by production machines are characterised by perfect quality. Wahab and Jaber [25] presented an economic order quantity model for defective goods. [30] investigated new GAs for a serial-parallel production line (S-PPL) to maximise production efficiency by optimising three decision variables: the buffer size between each pair of work stations. and the second was to minimise the defective raw material.doi. In most cases. However. [29] proposed a mathematical model for a three-stage SCN with a single source and multiple products.. and indirect costs and determining the economic production quantity. the machine numbers in each of the work stations. assumptions. The first objective was to maximise the profit. transportation. we propose a mathematical model for the DGSCN. To solve the proposed model.A. scrap products. studies of inventory control models that oversee the basic concepts of warehouse needs and imperfect products after manufacturing have attracted researchers’ attentions [19. Flapper and Teunter [22] considered the defective products of single-item manufacturers that used the same facilities and production machines for reworking. improvement. they compared their results with those obtained by Cplex. Guo and Tang [13] proposed a unified model for the optimisation of supply chains based on JIT and confirmed the accuracy of the cost optimisation. it showed that defective goods and shortages might reduce the overall profit of the firm. Their numerical results showed that a decrement in the defective raw material led to an increment in the profit and a decrease in the financial risk. reducing costs. the production and distribution capacities of a machine are limited.2 S. Section 2 studies the model structure. and production in the Toyota factory. have been routinely used to generate useful solutions for optimisation and search problems. The proposed mathematical model can determine the ALOEC according to the importance of customer lead time using JIT logistics in addition to minimising the costs of production. In reality. quality and time are three key elements that should be noted in the design of a SCN.20]. Genetic Algorithms (GAs).2013. Appl. The remainder of the paper is organised as follows. maintenance. cost. we needed to solve the model using exact parameters and same dimensions and prefer to use Lingo for this application. The first objective was to maximise total profit. [26] studied a three-objective stochastic model for a forward/reverse logistic network. The three level supply chain network (SCN).023 .1016/j. Therefore. Modell. Additionally. Therefore. defective goods. Ghasimi et al. Abu Qudeiri et al. According to the literature. [28] proposed a mathematical model for a closed-loop battery-recycling SCN by solving with GAs and a GAMS optimiser. Altiparmak et al. Please cite this article in press as: S. http://dx. These factors are not fully addressed in other models. Math. Kannan et al. the costs of defective goods are considered essential in supply chain optimisation. His findings showed that JIT had a positive impact on sales. 1. Eroglu and Ozdemir [23] presented a model of defective goods and the ordered amount that is returned as shortage. / Applied Mathematical Modelling xxx (2013) xxx–xxx customer needs. design. the second objective was to maximise the customer service level in both the forward and reverse networks.A. Ghasimi et al. Franca et al. which are search heuristics that mimic the process of natural evolution. they used numerical examples that showed that defective and useless goods could reduce the overall profits.

apm. Shipping cost of each defective goods l at manufacturer i during period t (from production to defective goods storage and vice versa). and parameters of the mathematical model. Rework cost per defective good l by manufacturer i during period t Holding cost of products l at distributor j during period t. 10. Shipping cost of each product l from distributor j to retailer k during period t. and the inspection time is zero. 8. / Applied Mathematical Modelling xxx (2013) xxx–xxx 3 ables. III. The storage capacities for each perfect product are limited. 5. Parameters pilt : cilt : hjlt : hilt : hilt : T ht : lilt : T lt : Pcilt : TC ijlt : T 0 C jklt : T 00 C ilt : 0 Percentage of defective goods l produced by manufacturer i during period t. Modelling We propose a mathematical model for a DGSCN that is able to determine the EPQ and ALOEC using JIT logistics in addition to minimising the costs of production. Math. 7. Rework time required for goods l by manufacturer i during t. Percentage of scrap productsl produced by factory i during period t. Finally. The capacities of the manufacturers. All demands must be satisfied during the planning horizon. Appl. distributors.A. the following assumptions were established: 1.doi. Store capacities and allocated storage capacities for defective goods are limited. Modell. Time required to produce goods l by manufacturer i during period t. (continued on next page) ailt : filt : bklt : Please cite this article in press as: S.1. Shortage cost for each product l at retailer k during period t. Production cost per item by manufacturer i during period t.2. 2.1016/j. 2. Section 5 presents conclusions. (2013). Shipping cost of each product l from manufacturer i to distributor j during period t. Furthermore. Discount cost per scrap goods l for sale by factory i during period t. retailers and suppliers are fixed.A. The capacities of the manufacturer. Assumptions and limitations of the model In developing the proposed model. http://dx. distributor and retailers are limited. 6. whereas those considered as scraps. 2. 3. Ghasimi et al.2013.2. products and multi-periods. rework of defective goods. 1.S.1. Products are temporarily stored at the distributors before delivery to the retailers. 2. IV.. 9. Total production time during period t. The production and reworking times are limited. 11. We assumed that all imperfect quality products are not repairable. Rework of imperfect products is allowed. The amount of demand was assigned to the manufacturers at the beginning of the period. and Section 4 investigates the results and discussion.08. Decision variables and parameters The parameters and decision variables used in the cost optimisation are defined as follows: 2. The duration of each period was equal to the sum of the production and rework times. II. maintenance.org/10. The proposed model is divided into three levels: level 1 indicates the manufacturers. Holding cost of products l for defective goods stored at manufacturer i during period t. V. distributors. Ghasimi et al. level 2 represents the distributors and level 3 denotes the retailers. The locations of plants. Retailer shortage is allowed. The inspection process for the products is perfect. several limitations were considered in the proposed model: I. scrap products and retailer shortages due to the production of defective goods and indirect costs. the distributors and the retailers are known. The supply chain network used in this study is shown in Fig. The model is designed for multiple manufacturers. Section 3 presents the proposed GAs. retailers. transportation.023 . 4. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. There is no inventory at the distributors at the beginning or end of the planning horizon. Total rework time during period t.

l.A. ð6Þ Defilt ¼ pilt :Q ilt Scilt ¼ ailt :Defilt ð7Þ ð8Þ Please cite this article in press as: S. 2.023 . A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. t . l. t . ð4Þ 8 k. t . Amount of scrap goods l produced by factory i during period t. Amount of perfect products l produced by factory i during t after rework. Storage capacity of retailer k for product l during period t. 8 j. l. Amount of defective goods l produced by factory i during period t. l . t. Total storage capacity of distributor j during period t. Storage capacity of distributor j for product l during period t. Production capacity of factory i for product lduring period t. http://dx.2. Ghasimi et al. Ghasimi et al. Variables X ijlt : Q ilt : Defilt : Scilt : Coilt : TCoilt : Y jklt : Bklt : Tt : Amount of products l transported from factory i to distributor j during period t. Amount of products l transported from distributor j to retailer k during period t. EPQ of products l by factory i during period t. Inventory of product l at distributor j during period t. Total storage capacity of retailer k during period t. 2.doi.. 8 i.4 S. Amount of shortage of products l in retailer k during period t. ALOEC according to the importance of lead time by customers in period t. Math.08.1016/j. ð5Þ 8 k . 8 i. t . Demand of retailer k for product l during period t.2013. ð2Þ ð3Þ I X L X X ijlt 6 W jt i¼1 l¼1 I X X ijlt 6 W jlt i¼ 1 J X L X Y jklt 6 W 0kt j¼1 l¼1 J X Y jklt 6 W 0klt j¼1 8 j. Modell. Mathematical model The objective function is presented as follows: Z min ¼ J X I X L X T L X T t X I t X K X X X X Pcilt :Q ilt þ hjlt : X ijls À Y jkls i¼1 l¼1 t ¼1 j¼1 l¼1 t ¼1 ! þ I X L X T I X L X T X X 0 hilt :Defilt þ cilt :Defilt i¼1 l¼1 t ¼1 i¼1 l¼1 t ¼1 s ¼ 1 i¼ 1 s¼ 1 k ¼ 1 J X J X I X L X K K X L X T I X L X T I X L X T X X X X þ TC ijlt :X ijlt þ T 0 C jklt :Y jklt þ T 00 C ilt :Defilt þ filt :Scilt i¼1 j¼1 l¼1 t¼1 j¼1 k¼1 l¼1 t ¼1 i¼1 l¼1 t ¼1 i¼1 l¼1 t ¼1 K X L X T T X X þ Bklt :bklt þ Indt :T t : k¼1 l¼1 t ¼1 t ¼1 ð1Þ The constraints and their explanations are presented in the following: Q ilt 6 Silt 8 i. Amount of perfect products l produced by factory i during t before rework.A. / Applied Mathematical Modelling xxx (2013) xxx–xxx Indt : Inljt : dklt : Silt : W jt : W jlt : W 0kt : W 0klt : Indirect costs (such as stagnant capital and maintenance) during periodt. l. t. Appl.3. t. (2013).org/10.2.apm.

Eq. l. transportation from the manufacturers to defective good storage and vice versa.A. ð13Þ K K I X X X Bklt ¼ dklt À Coilt k¼1 k¼ 1 i¼1 8 l. (11) states that total production is equal to the sum of demand and scrap goods. Eqs. Modell. Eq.2013. (9) calculates the amount of perfect goods before reworking. Eqs. including the costs of production. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. l. k .doi. t . 8 i. Eq. transportation from the manufacturers to the distributors. (15)explains how the total demands during the planning horizon are supplied. ð17Þ s¼ 1 T X I T X K X X X ijlt ¼ Y jklt t ¼ 1 i¼ 1 t ¼1 k¼1 8 j. respectively. Eq. Q ilt . t : ð23Þ Eq. (5) and (6) consider the delivery capacity limitations of the retailers for all product types and each product type. Ghasimi et al. 8 l. Ghasimi et al. holding at the distributor. TCoilt P 0 8 i . (2013). t . t – T . ð12Þ I K X X Coilt 6 dklt i¼ 1 k¼1 8 l. http://dx. defective good storage.apm.08. l. Bklt . (7) calculates the amount of defective goods. t . Defilt . there is no inventory at the distributors in the beginning and at the end of the planning horizon. Eq. Eq. (1) represents an objective function that minimises the total costs of the supply chain. ð20Þ lilt :Defilt 6 T lt 8 t. ð15Þ K X t I X t X X Y jkls 6 X ijls k¼1 s¼1 i¼1 s¼1 8 j. t . respectively.org/10. j.1016/j. and indirect costs such as stagnant capital and maintenance. retailer shortages due to defective goods production. ð21Þ ð22Þ T ht þ T lt 6 T t X ijlt . Eqs. Scilt .. (12) shows that the total demands are covered by the amount of perfect goods. 8 t. (2) states the restriction of production capacity. Eq. Appl. t . t. Math. ð16Þ I X t K X t t X X X X ijls À Y jkls ¼ Inljs i¼1 s¼1 k ¼ 1 s¼ 1 8 j. l. transportation from the distributors to the retailers. t. (16) and (17) represent the inventory at the distributors. l. Eq. l . t – T . l . / Applied Mathematical Modelling xxx (2013) xxx–xxx 5 Coilt ¼ ð1 À pilt Þ:Q ilt 8 i. (13) and (14) investigate the amounts of shortage at the retailers due to defective goods. rework of defective goods. (18) shows the balance between the total inputs Please cite this article in press as: S. (8) represents the amount of scrap goods.023 . ð18Þ J X X ijlt 6 TCoilt j¼1 I X L X hilt :Q ilt 6 T ht i¼1 l¼1 I X L X i¼1 l¼1 8 i.A. (10) expresses the total amount of perfect goods after reworking. Y jklt . (3) and (4) denote the delivery capacity limitations of the distributors for all types of products and each type of product. Coilt .S. t ð 9Þ ð10Þ ð11Þ TCoilt ¼ Coilt þ Defilt À Scilt K I I X X X dklt þ Scilt 6 Q ilt k¼1 i¼ 1 i¼ 1 K I X X dklt 6 TCoilt k¼1 i¼ 1 8 l. Eqs. the discount cost of scrap goods for sale. ð14Þ J X Y jklt ¼ dklt j¼1 8 k . ð19Þ 8 t. and Eq. l.

And Eq. the amount Q ilt should be expressed as follows [Eq. (ii) The GA has the ability to work with many decision variables. the number of infeasible answers is low and the probability of reaching the global optimum is high. and after crossover and mutation operations. (1) Q ilt 6 Silt 8 i.6 S. the algorithm must be started again from the beginning. Matrix Q ðIÃLÞÃT Q 111 B . any change in variable Q ilt leads to certain changes in other variables. We attempted to select the best chromosomes in the proposed GAs that would give us good results and require a low run-time.1. C . Finally. A ÁÁÁ Q ILT ðIÃLÞÃT ð24Þ 3. However. Therefore. Therefore. Math.023 . most other algorithms do not run in parallel and can only search the problem space from one direction at a time. (iii) The large number of dimensions in the proposed model requires along run-time because the GA has several starting points. Eq. Initial population A certain number of chromosomes were randomly created according to: Eq. mutation. . The chromosomes are arranged based on the value of the objective function and the best chromosomes are selected for the next generation. . (19) assures that the total goods shipped from the manufacturers to the distributors are of perfect quality.3. as shown in Eq.1. we chose the GA due to its many advantages: (i) Because the answer space is extensive.org/10. the variable Q ilt has both direct and indirect relationships with the variables Scilt . and if one fails to find the result.doi. scrap goods. Ghasimi et al. (25)]: K I I X X X dklt < Q ilt 6 Silt k¼1 i¼1 i¼ 1 8 l. and thus. (10) K I I X X X dklt þ Scilt 6 Q ilt k¼1 i¼ 1 i¼ 1 8 l Next. In the studied mathematical model. Coilt . Q IL1 0 1 Á Á Á Q 11T C .1016/j. t ð25Þ The initial population is randomly produced. Appl. First. (23) indicates that the production amount. 2. Each gene in the matrix was randomly created by Matlab software. Ghasimi et al. (21) shows the available time limitations of the production facilities for rework processes. (24). defective goods. Chromosomes One of the important components of the GA is the selection of chromosomes. The chromosomes are displayed in a matrix with the columns denoting the number of periods (T) and rows (I⁄L). Eq. which are crossover. Bklt . the parents and produced children are placed in the pool.3.08.. and the fitness of each individual is subsequently evaluated. The columns will be combined for each chromosome selected for the crossover. the scrap goods are removed from the system. http://dx.2. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. (2013). Genetic operations The following describes the main operations of the GA.A.. t. (22) shows the length of each period. Crossover To perform the crossover.apm. the chromosomes to be combined should be identified and allowed to mix. retailer shortages. in other words. The flowchart of the proposed GA is shown in Fig.A. B ¼@ . deliveries to warehouses and retailers. . the variable Q ilt was chosen as the chromosome. Modell.2013. l. the next generation of chromosomes is selected from among these chromosomes according to the population size. . 3. (20) explains the available time limitations of the production facilities for all production processes. Eq. Eq. 3. / Applied Mathematical Modelling xxx (2013) xxx–xxx and outputs of goods moving to and from the distributors during the planning horizon. The individuals are sorted based on their fitness. X ijlt and Y jklt . two chromosomes must be merged. Proposed GA In this study. 3. the other directions can continue. 3. it can simultaneously search the problem space from multiple directions. and selection. The value of the objective function of the children is calculated. and the intersection Please cite this article in press as: S. if the answer represents a local optimal answer or the result is a subset of the main answer. therefore. Eq. and perfect goods before and after reworking should all have positive values.

A. the values of both sides of the matrix are exchanged. if the mutaPlease cite this article in press as: S. the intersection point is chosen.08. Flowchart of the proposed GA. First. 3). / Applied Mathematical Modelling xxx (2013) xxx–xxx 7 Generation =1 Create initial random population Pop. Ghasimi et al. In this study.2013.023 .8 No The number of children produced is sufficient? Yes All parents and children produced are put into a pool. 3. Math. The intersection point shown in this example denotes the first period (Fig. (2013). if we work with two chromosomes with four periods and three rows. For example.apm.3 Selection ofparents for crossover and mutation operations Crossover operation Crossover probability=0. For example. Thereafter.S. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths.size = 100 Evaluate fitness for each individual Store the averagepopulation in each generation Sorting the population of each generation Store the best value fitness in each generation Mutation operation Mutation probability=0. http://dx. 2.org/10. Ghasimi et al.A. a number of genes must be selected to undergo mutation. size)are selected for the population of the next Yes Generations =150 No Generation = Generation+1 Fig.1016/j.2.3. each chromosome receives a certain number of genes that are assumed to change. Appl. Mutation The mutation probability refers to the probability of change in any gene.doi.. they are sorted and 100(pop. and this number is derived by multiplying the total number of genes by the mutation probability. point will be used to combine the chromosomes. the combination of the chromosomes will proceed as follows. Modell. Accordingly.

Display of the mutation operation. and T = 4.023 . Q111 Q211 Q311 Q112 Q212 Q312 Q113 Q213 Q313 Q114 Q214 Q314 Q '111 Q211 Q311 Q112 Q212 Q113 Q114 Q214 Q314 Q '213 Q313 Q '312 Parent and selected genes Produced child Fig. scrap goods and retailer shortages due to defective goods production.. / Applied Mathematical Modelling xxx (2013) xxx–xxx Q111 Q211 Q311 Q112 Q212 Q312 Q113 Q213 Q313 Q114 Q214 Q314 Q '111 Q '211 Q '311 Q '112 Q '212 Q '312 Q '113 Q '213 Q '313 Q '114 Q '214 Q '314 Selected Parent 1 Selected Parent 2 Q111 Q211 Q311 Q '112 Q '212 Q '312 Q '113 Q '213 Q '313 Q '114 Q '214 Q '314 Q '111 Q '211 Q '311 Q112 Q212 Q312 Q113 Q213 Q313 Q114 Q214 Q314 Produced Child 1 Produced Child 2 Fig.A. Please cite this article in press as: S. Additionally. L = 1.2013. Results and discussion This study proposes a novel mathematical model for optimising the supply chain costs of imperfect products. The aim of the model is to optimise total costs. three genes should undergo mutation. 3. and the value of the objective function of the children was calculated. The resulting value is rounded off to the nearest whole number. Ghasimi et al. 4. Thus.3. These three genes are randomly selected. reworking of defective goods.8 S. 3. (2013).org/10. First. Selection Different strategies can be applied to perform the selection function.3 and I = 3.1016/j. Math.apm. in this example.3. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. the total number of genes required to undergo mutation will calculated as follows: Round (0. and their values are changed (Fig. Modell. 4. the next generation of chromosomes was selected according to the population size. this model is able to determine the economic production quantity (EPQ). http://dx.6) = 3. Display of the crossover operation. including production. holding.08. and the elite strategy was chosen in this study. tion probability is 0. the appropriate length of each cycle (ALOEC) and the number of active manufacturers and distributors. The chromosomes were arranged based on the value of the objective function and the best chromosomes were selected for the next generation.3 ⁄ 3 ⁄ 1 ⁄ 4 = 3. shipping. Finally.A.doi. 4). from among these chromosomes. the parents and produced children were placed in the pool. Ghasimi et al. Appl.

A.1016/j.6) x 10 7 5 Best Average 4 Zmin 3 2 1 0 50 100 150 Number of generations Fig. 2) Normal (10.apm. 2) (12. 2) cilt hjlt hilt hilt 0 lilt TC ijlt T 0 C jklt T 00 C ilt filt bklt ailt dklt Indt Normal (5.doi. 3) (8. Appl. 1000) Normal (0. http://dx. Please cite this article in press as: S. 3) Normal (6. Modell.023 . 0. 0. 0. Math.S.09) Normal (15. 3) (5. Ghasimi et al. (2013). Ghasimi et al. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths.9) Normal (2000.org/10. 2) Normal Normal Normal Normal (6..2013.A. 3) Normal (6. 5. 2) Normal (11. 1) Uniform (0. / Applied Mathematical Modelling xxx (2013) xxx–xxx Table 1 Twelve sample problems with different dimensions. 3) Normal (6.08. GA output for Z min (best and average) of sample problem 11. Sample problem 1 2 3 4 5 6 7 8 9 10 11 12 I 2 3 4 3 4 3 4 6 2 6 10 8 J 2 2 5 6 3 4 3 2 3 4 9 9 K 2 3 3 2 4 4 6 4 2 5 8 10 L 2 2 1 2 1 3 1 1 2 1 1 1 T 2 3 3 2 4 3 4 3 4 7 6 10 9 Table 2 Parameters and values Parameters pilt Pcilt Value Uniform (0.

015 0.doi. and we compared the results of both methods.008 0.016 0.97 0.31 0.A.2013.021 0. http://dx.org/10. This figure shows the fine function of the proposed GAs. increasing the generation size increased the run-time of the GA but resulted in minimal changes in the answers. In the proposed GA.9 1..024 1. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths.A. (2013).08.9 0.014 0. to demonstrate the performance accuracy of the model and the GA.74 1.27 2. The curves for Z min (best) and Z min (average) for Sample Problem 11 are shown in Fig. For solution of the model and accuracy of the results. The appropriate generation size was 150.48 0. and the Z min (average) is the average of 100 answers in each generation. Ghasimi et al. Additionally. the population size is 100. Sample problem GA Z min (least) 1 2 3 4 5 6 7 8 9 10 11 12 340958 853500 319010 408040 720100 1521860 1101520 583300 681030 1651390 2010230 3275006 Run time (sec) 161 254 281 301 296 396 284 234 274 561 1372 2586 CPLEX Z min (least) 335900 848300 315009 396843 716660 1510600 1080900 572009 674900 1646211 2000650 3267006 Run time (sec) 0. Ghasimi et al. Modell. 100 answers (Z min ) were produced.016 0.10 S.015 0. each sample problem was solved nine times by Please cite this article in press as: S. Sample problem 1 2 3 4 5 6 7 8 9 10 11 12 Z min (Run 1) 358400 875600 338280 428500 743100 1588000 1133600 583300 685800 1667000 2024000 3388000 Z min (Run 2) 365020 866700 319010 408040 740980 1587410 1125980 620560 690600 1659540 2110000 3278060 Z min (Run 3) 351474 857654 347800 459270 752780 1591460 1110320 603500 681030 1671580 2090830 3350800 Z min (Run 4) 342600 873840 358650 416080 739930 1521860 1137060 632930 688500 1688010 2180340 3409850 Z min (Run 5) 355289 853500 329300 425120 724310 1579210 1101520 618070 693090 1669210 2040700 3275970 Z min (Run 6) 370550 860200 347030 412034 720100 1532080 1112600 595900 701230 1707000 2125870 3360320 Z min (Run 7) 340958 880090 350790 445006 742700 1529660 1133030 584883 696710 1657800 2140650 3343860 Z min (Run 8) 360840 854990 320500 436090 722770 1523820 1127070 615610 691928 1651390 2010230 3389480 Z min (Run 9) 341649 860500 352340 420830 725840 1590040 1135540 604380 684200 1672160 2060890 3275006 Table 4 Results of nine runs for ten sample problems solved with Cplex. To mimic a real-life situation.6 1.48 0.5 0.82 0.015 0. The Z min (best) is the smallest amount of answers.023 . and the average of 100 answers was reported together with the best answer in each generation.24 Gap% = [(GA – CPLEX)/ CPLEX] ⁄ 100 Twelve sample problems with different dimensions were used to validate the correctness and fine functions of the model. The studied GA was tested with different generation sizes. Math. / Applied Mathematical Modelling xxx (2013) xxx–xxx Table 3 Results of nine runs for twelve sample problems solved with the GA. in each generation. statistical distributions were used to select the values of the parameters.018 0. Therefore.016 0. 5. Sample problem 1 2 3 4 5 6 7 8 9 10 11 12 Z min (Run 1) 350418 862530 333880 429609 733679 1522900 1134000 575000 674900 1648908 2027005 3288840 Z min (Run 2) 367402 868921 329040 403490 740006 1590707 1123000 601590 690000 1658980 2050500 3267006 Z min (Run 3) 341049 854004 340076 439007 743006 1510600 1127098 613530 680003 1647890 2091800 3288000 Z min (Run 4) 350712 863490 315009 417540 741800 1532000 1136000 642956 677450 1692006 2006750 3268900 Z min (Run 5) 345510 848900 318781 410500 716660 1566008 1090660 572009 690600 1648000 2102000 3303000 Z min (Run 6) 349100 855609 351500 414007 719993 1540088 1080900 580400 675909 1661009 2077098 3351089 Z min (Run 7) 351034 848300 318007 410005 732005 1511866 1141000 583787 701006 1650775 2000650 3310090 Z min (Run 8) 335900 859100 322110 396843 718002 1515700 1082707 573009 692300 1657000 2008590 3267980 Z min (Run 9) 339240 849231 316903 420087 725008 1570600 1088970 581098 683600 1646211 2121008 3270098 Table 5 GA and CPLEX results for 12 sample problems with different dimension. Instead. as shown in Table 1. we did not rely only on an exact optimiser such as Cplex.1016/j.017 0. we proposed a GA.apm. as shown in Table 2. Appl.

A. / Applied Mathematical Modelling xxx (2013) xxx–xxx 6 x 10 3.08.08 T 00 C 311 = 2 d311 = 8000 0 h311 = 7 TC 3111 = 8 TC 3211 = 9 T 0 C 1211 = 6 T 0 C 2311 = 5 Please cite this article in press as: S.5 3 2.35 b211 = 14 c211 = 6 h211 = 6 l211 = 2 f111 = 9 Pc211 = 25 p211 = 0.org/10.3 f111 = 7 Pc111 = 30 p111 = 0. Modell. 6. Sample problems Fig. Appl.doi.2013.5 1 0.05 T 00 C 211 = 6 d211 = 6000 h211 = 7 h211 = 4 TC 2111 = 4 TC 2211 = 3 T 0 C 2111 = 3 T 0 C 1311 = 9 0 a311 = 0.27 b311 = 12 c311 = 2 h311 = 5. Ghasimi et al. Comparison of the GA and Cplex results.1016/j. http://dx. (2013). A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. Scilt Sc111 = 76 Sc112 = 16 Sc121 = 0 Sc122 = 262 Sc211 = 0 Sc212 = 0 Sc221 = 256 Sc222 = 0 Defilt Def111 = 291 Def112 = 191 Def121 = 0 Def122 = 335 Def211 = 0 Def212 = 0 Def221 = 373 Def222 = 0 Coilt Co111 = 4190 Co112 = 2585 Co121 = 0 Co122 = 4307 Co211 = 0 Co212 = 0 Co221 = 5195 Co222 = 0 TCoilt TCo111 = 4406 TCo112 = 2760 TCo121 = 0 TCo122 = 4379 TCo211 = 0 TCo212 = 0 TCo221 = 5303 TCo222 = 0 Bklt B111 = 0 B112 = 174 B121 = 108 B122 = 0 B211 = 215 B212 = 0 B221 = 0 B222 = 72 Qilt Q111 = 4482 Q112 = 2777 Q121 = 0 Q122 = 4642 Q211 = 0 Q212 = 0 Q221 = 5568 Q222 = 0 Xijlt X1111 = 0 X1112 = 397 X1121 = 0 X1122 = 0 X1211 = 4406 X1212 = 2362 X1221 = 0 X1222 = 4379 X2111 = 0 X2112 = 0 X2121 = 0 X2122 = 0 X2211 = 0 X2212 = 0 X2221 = 5303 X2222 = 0 Yjklt Y1111 = 0 Y1112 = 397 Y1121 = 0 Y1122 = 0 Y1211 = 0 Y1212 = 0 Y1221 = 0 Y1222 = 0 Y2111 = 2050 Y2112 = 0 Y2121 = 2703 Y2122 = 1958 Y2211 = 2355 Y2212 = 2362 Y2221 = 2600 Y2222 = 2421 Table 7 Values of the parameters for the considered sample problem in Fig.5 l311 = 2.A.3 b111 = 10 c111 = 4 h111 = 4 l111 = 1. Ghasimi et al.5 h111 = 8 h111 = 2 TC 1111 = 6 TC 1211 = 5 T 0 C 1111 = 6 T 0 C 2211 = 4 0 clit hilt lilt filt Pcilt pilt T 00 C ilt dklt Indt hjlt hilt TC ijlt T 0 C jklt 0 a211 = 0. Table 6 Outputs of the model for the best amount of sample problem 4.5 0 0 2 4 6 8 10 12 No.S.apm.023 .5 11 Z min (GA) Z min (Cplex) Z min 2 1.. 7.5 f111 = 10 Pc311 = 27 p111 = 0. Math.07 T 00 C 111 = 3 d111 = 3000 Ind1 = 0. Parameters Value of parameters ailt bklt a111 = 0.

5 and 6 and Table 5 show the correctness and fine function of the model and the GA. the best result of each sample problem (out of nine runs.1 1.2 7. The findings for the proposed model (i. We used the GA and Cplex together with probability parameters for validation of the studied model in a real-life situation and compared the outputs to demonstrate the good performance of the model. Appl. In addition. defective goods. The results from Fig.3 7. from Tables 3 and 4) is selected and presented in Table 5 together with its run time.08. Relationship between the length of the period and Z min .9 1 1. l = 1 and t = 1.apm. Table 5 and Fig.org/10. Due to the statistical distributions chosen for the values of the parameters. http://dx. we prefer to use Lingo.7 6.A. the GA and Cplex using the parameters of Table 2. 6 show the fine function of the presented model. Sample problem 1 2 3 4 5 6 7 8 9 T h1 69500 77000 85000 93000 100418 108000 115000 123000 130000 T l1 1600 1800 2000 2200 2448 2650 2900 3100 3300 T1 71100 78800 87000 95200 102866 110650 117900 126100 133300 P sc P Def P Co P TCo P B P Q Z min 731339 715358 699035 682712 666132 670023 673648 677748 681348 364 356 347 339 332 332 332 332 332 1215 1183 1143 1103 1080 1080 1080 1080 1080 16149 16173 16204 16235 16251 16251 16251 16251 16251 17000 17000 17000 17000 17000 17000 17000 17000 17000 851 827 796 764 748 748 748 748 748 17364 17356 17347 17339 17332 17332 17332 17332 17332 5 x 10 7.4 5 x 10 Length of period ( T ) 1 Fig. 7. 8.7 0.1 b Z min 7 6. / Applied Mathematical Modelling xxx (2013) xxx–xxx R1 M1 W1 M2 R2 W2 M3 R3 Fig. Ghasimi et al. the total costs. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths.8 0.8 f 6. Ghasimi et al. Modell. we needed to solve the model using the exact parameters.To reach this answer.2 e g k h 1.12 S. The best result of each run from the GA and Cplex are illustrated in Tables 3 and 4. Another objective of the proposed model was to find the appropriate length of each cycle (ALOEC). Please cite this article in press as: S.6 0.1016/j. perfect goods and amounts of products) that should be transferred between the nodes of the supply chain network for sample problem 4 are shown in Table 6.9 c d 6.doi.e..3 1.. j = 2.4 a 7.2013. Math. retailer shortage. Scheme of a sample problem with dimensions i = 3. Table 8 Results of nine sample problems with the same dimensions. scrap goods. the outputs are different.A. amounts of EPQ. k = 3.023 . In this case. (2013).

J. M. 47 (6) (2005) 399–410. Amasaka. According to the dimensions.Z. Van Hop. J. Cordeau. 24 (6) (2008) 735–743. A. J. Berman.A. The relationship between the total costs and the length of the period is shown in Fig. K. [3] M. [4] D. Robot. We considered a supply chain network with dimensions i ¼ 3. Indust. Optimal batch sizing in a multi-stage production system with rework consideration. J. Product. These factors are not addressed comprehensively in other models. Econ. Trienekens. An optimized supply chain planning model for manufacture company based on JIT. Math. Inf. Shih. W. M. Her. [20] J. Inventory models for defective items incorporating marketing decisions with variable production cost. C. J. http://dx.023 . The total costs decreased with increments in the production and repair times (see sample problems 1.apm. Guo.-C. Manuf. Exploring defect causes in products developed by virtual teams. [11] R. Wang.S. Mondal.-D. Solomon. Appl. Maiti. Additionally. 33 (6) (2009) 2845–2852. Math. t ¼ 1 and the exact parameters shown in Table 7. P. [10] S. Indust. Eng. Inbound logistic planning: minimizing transportation and inventory cost. transport. Int. in the online version. The logistics of Just-in-Time between parts suppliers and car assemblers in Japan. we were required to solve the model using the exact parameters and prefer to use Lingo in this case. Kaneko.1016/ j. 7).A.023. 80 (2) (2002) 135–144. there are 18 routes from the manufacturers to the retailers in the network (see Fig. any change in the T h and T l has a direct effect on the variables Q ilt and Defilt and consequently.org/10. 16 (3) (2008) 155–173. [19] K. F. Res. Jacobs. k ¼ 3. A hybrid MCDM model for strategic vendor selection. Grazia Scutellà. b. If no time limitation exists. Lin. Oper. retailer shortages and indirect costs (i. Minner. Oper. Farahani. References [1] L. 120 (2) (2009) 437–451. 172 (1) (2006) 179–200. Point e represents the minimum. B. Modell. ‘New JIT’: a new management technology principle at Toyota. Please cite this article in press as: S. 3 (11) (2008) 129–133.. 36 (4) (2012) 1703–1717. A genetic algorithm to optimize the total cost and service level for just-in-time distribution in a supply chain. Shyur. To model the real states accurately. Distribution network design: new problems and related models. [18] B. 111 (2) (2008) 229–243. A production-inventory model with remanufacturing for defective and usable items in fuzzy-environment.G. An integrated model for logistics network design. A. Ghasimi et al. 44 (2006) 749–761. (20)–(22). Bhunia. Math. Appl. Conclusions In this paper. Multiple-supplier inventory models in supply chain management: a review.D. Comput. Supply chain models for an assembly system with preprocessing of raw materials: a simple and better algorithm. M. Model. Roy. The GA and Cplex techniques were used to solve the proposed mathematical model. Sci. J. [7] O. 40 (3) (2006) 287–299. 144 (1) (2006) 59–82.To obtain this answer.3. defective goods. 8). Eng. J. [12] H. on Scilt . H.doi. Supplementary data Supplementary data associated with this article can be found. Kusters.E. Kar.2013. at http://dx. Additionally.K. Cárdenas-Barrón. Econ. Sawik. M. Int. Nojiri. Additionally.R. 8). Another objective of the investigated model was to identify the appropriate length of each cycle (ALOEC). Optimal models for a multi-stage supply chain system controlled by kanban under just-in-time philosophy. Maiti.08. [13] R. Model. [5] S.M. Q. Product. scrap products. Sarker. Math. The calculated gap between the best results of the GA and Cplex demonstrates the accuracy and fine function of the proposed model. Order allocation in a multiple-supplier environment. the values of the parameters were treated as probabilities with normal and uniform distributions..-F. because T ht þ T lt ¼ T t 8 t.M. 94 (2005) 231–238.R. Int.doi.G. Product. Business Manag. [8] Y.6 and 8). Annal. A new approach to tactical and strategic planning in production-distribution networks. Appendix A. 165 (3) (2005) 610–624. Comput. Talebi.apm. J. C. Oduoza. Jamal. S.4 and 5 in Table 8 and points a. Econ. Math. S. J. Oper. T e Š. Proverbs. Transp. the model searches for the best path using the parameters in Table 7 to obtain the minimum cost (see sample problem 5 in Table 8 and point e in Fig. Krause.org/10. Coilt . Tang. J. We illustrate this scenario using an example. J. we have proposed a new mathematical model for a three-echelon defective goods supply chain network (DGSCN) using JIT logistics that not only minimises the costs of production. Softw. Mondal. [6] J. 56 (1) (2009) 193–197. [14] K. Comput. stagnant capital and maintenance) but also determines the economic production quantity (EPQ) and the appropriate length of each period (ALOEP) according to the importance of customer lead time (see Tables 5. J. A systematic modelling and simulation approach for JIT performance optimisation.1016/j. Ghasimi et al. Res. Product. [9] J. This period is dependent on the significance of the customer order time. Elahipanah. T t has a direct effect on the mentioned variables.2. Table 5 shows that the results and run time from Cplex are better than those of the GA solver. the smallest time for production is needed to supply the retailer demands (see sample problem 1 in Table 8 and point a in Fig. Ambrosino. R. Maity.e. Appl. (2013). A. and any further increment in the times leads to increase in the total costs due to indirect costs such as stagnant capital and maintenance.F. Coordinated supply chain scheduling. [16] B. 81–82 (2003) 265–279. Porter. 8. A two-warehouse inventory model with imperfect quality production processes. 56 (1) (2009) 87–96. Bashiri. Appl. / Applied Mathematical Modelling xxx (2013) xxx–xxx 13 According to the restrictions of Eqs. (2013). Model.-J. Model. [17] A. Bklt and Z min . c. d and e in Fig. S. Res. holding. [15] R. Eur. Int. l ¼ 1. 5. Pasin. Int. Chung. Econ. Eur. Badri. Geo. Sarker. D. Int. Econ. Brombacher. van Moll.2013. H. Oper. Integrat. Kawtummachai. Wang. J. Q. N. j ¼ 2. Transp. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths. Comput. This model can also determine the number of active manufacturers and distributors and is applicable for all producers faced with problems of defective goods. 184 (3) (2008) 915–929. Sandanayake. [2] T.08. Technol. 8). J. Product. Eur.M. and the best length for the period lies in the interval½T a . M. Res.

Appl. Comput. Gen. Product. Lin. A. 88 (1) (2004) 51–59.H. Kannan. Altiparmak. Mondal. Richards. pp. Eng. Jamali. [30] J. 1252–1259. Ramezani. B. S. different holding costs.B. A genetic algorithm for optimizing defective goods supply chain costs using JIT logistics and each-cycle lengths.D. Life Robot. M. Econ. Karaoglan. 106 (2) (2007) 544–549. Devika. An economic order quantity model with defective items and shortages. Tavakkoli-Moghaddam. Ramli. Jones. Eng.R. Econ. H. Flapper. [27] R. Jamal. J. [28] G. M. R.org/10. [25] M. Model. (2013).N.A. G. [29] F. [26] M. Sasikumar. Product. Bashiri. Ozdemir. http://dx.08. J. 37 (1–2) (2013) 328–344. Franca. 47 (1) (2004) 77–89. [22] S. A genetic algorithm approach for solving a closed loop supply chain model: a case of battery recycling. C. Eng. Model.1016/j.023 . [23] A.. L.M. Comput. Int. Sarker. M. Math. Wahab. 127 (2) (2010) 292–299. R. 58 (1) (2010) 186–190. 2000. 12 (1–2) (2008) 102–106. H.P. Eroglu. Math. Artif. Econ.M.Y. A steady-state genetic algorithm for multi-product supply chain network design.P. Indus. Modell. I. 56 (2) (2009) 521–537.doi. in: Procedings of the 2000 Winter Simulation Conference. Int. Economic order quantity model for items with imperfect quality.A. 34 (3) (2010) 655–670. J. Lee. P. Please cite this article in press as: S. Appl. Carlson. J. Optimal production-distribution planning supply chain management using a hybrid simulation-analytic approach. / Applied Mathematical Modelling xxx (2013) xxx–xxx [21] Y.14 S. Ghasimi et al. and learning effects: a note.C. Abu Qudeiri. Optimal manufacturing batch size with rework process at a single-stage production system.2013. Multi-objective stochastic supply chain modeling to evaluate tradeoffs between profit and quality. Sook Han Kim. Teunter. K. Product. Genetic algorithm for buffer size and work station capacity in serial-parallel production lines.apm. A new multi-objective stochastic model for a forward/reverse logistic network design with responsiveness and quality level. [24] A. Yamamoto. E. Jaber.I. Ghasimi et al. Int. Comput. R. Math. Appl. Indus. Indus. Logistic planning of rework with deteriorating work-in-process.