Improve Production Balance for Apparel Supply Chain Adopting VMI Replenishment Strategy

A. H. Dong, W. K. Wong, S. F. Chan, P. K. W. Yeung

Institute of Textiles and Clothing, The Hong Kong Polytechnic University, HungHom, Kowloon, Hong Kong, SAR, China
Abstract-This paper analyzed the replenishment problem faced by manufacturer who adopts the Vendor Management Inventory (VMI) replenishment strategy with different retailers in apparel supply chain. An optimization replenishment model with the consideration of production capacity constraints was proposed. A simulation program was presented to generate the VMI replenishment strategy to satisfy the customer service level (CSL) required by the retailers. With the consideration of the production capacity constraints, a rolling optimization of the VMI replenishment strategy using simulation and genetic algorithm was then formulated to optimize production balance of manufacturers. The experimental results demonstrated that the proposed replenishment strategy could benefit the manufacturers with balanced production, meanwhile maintain the CSL required by different retailers at a certain degree.

II.

PROBLEM SETTING AND LITERATURE REVIEW Problem Setting

A.

I. INTRODUCTION In recent years, Vendor Management Inventory (VMI) strategy has become more popular in supply chain management, since the necessary information and communication technology are economically available[1, 2]. The supply chain management with production capacity has been investigated for quite a long time[3-5]. However, studies on linking the manufacturers VMI replenishment strategy in the supply chain with its production balance were limited. In supply chain adopting VMI replenishment strategy, the manufacturer makes replenishment decisions how many quantities of product to be shipped to the retailers in every replenishment cycle. Usually, the manufacturer produces products for different retailers and the customer service level (CSL) required by these retailers are different. For instance, 95% of CSL is required by one retailer and 90% of CSL is required by another retailer. Owing to the fluctuating demand from the customers, the replenishment quantity from the manufacturer varies on each replenishment cycle which will directly influence the balance of product. To ensure the products to be received by the retailers on time, manufacturer needs to deliver the goods by air after production, set up extra production line or require the operatives to work overtime. All these strategies may deteriorate the production balance and cause extra cost to the manufacturer. In this paper, an optimization VMI replenishment model is devised to generate appropriate replenishment quantity in each replenishment cycle so as to balance the production of the manufacturer while maintain the different CSLs required by different retailers at acceptable levels.

The proposed replenishment model is established based on the real operation of one two-echelon apparel supply chain. The apparel manufacturer is located in China while the customers are overseas retailers in the States. Besides adopting the VMI replenishment strategy, the manufacturer also produces garments for traditional retailers under the Order-Based (OB) business agreement. The manufacturer generates initial replenishment plans with two retailers who adopt the VMI strategy requiring different CSLs before the sales season. Then the manufacturer arranges the orders from traditional retailers according to the production capacity. Once the quantity and delivery time of these orders are determined, they cannot be changed and the products must be delivered to the retailers on time. Under the VMI strategy, the actual customer demands fluctuate which are always different from the initial plans although the forecasting on customer demands has been done. In order to satisfy the actual demands during the sales season, the production of the manufacturer is not well balanced. In order to balance the production balance of the manufacturer, a comprehensive VMI replenishment model is established. The main purpose of the proposed optimization replenishment model is to minimize the imbalance of production of the manufacturer in which the production balance is defined as the absolute difference between production capacity and actual production quantity. At the same time, the lost sales percentages under VMI strategy for both retailers at certain level can be maintained. The lost sales percentage for each retailer is defined as the ratio of total lost sales to their customer demand for all Stock Keep Units (SKUs) in the whole sales season. The key idea of the proposed model is moving the exceeding replenishment quantity under VMI strategy to the successive periods provided that there are sufficient production capacities in these periods. These adjustments include decisions on replenishing the exceeding quantities in advance or postponing the delivery of extra garments to the retailers in next replenishment cycles. The proposed model also determines the priorities of the replenishment quantities belonging to different retailers when the replenishments to them conflicted with each other, i.e. in condition that the

848

1-4244-0148-8/06/$20.00 c 2006 IEEE

limited production capacity could only satisfy part of the total demands from the retailers. B. Apparel Supply Chain Simulation Model And Optimization Using Genetic Algorithm

The simulation approach has been used extensively in the literature for the selection and evaluation of policies for controlling and managing the supply chain dynamic behaviors[6-8]. In apparel industry, the problems associated with the supply chain simulation model have been attracting increasing attention as a consequence of the development of Quick Response (QR) in apparel industry in the mid 1980s[8-10]. In complex problems such as the replenishment strategy in the supply chain, it is usually difficult to define one function form for measure of goodness for the solution. Unlike other optimization techniques, Genetic algorithm (GA), first introduced by Holland[11], does not make strong assumptions about the form of the objective function[12]. The next sampling point in GA is determined based on stochastic sampling/decision rules, rather than a set of deterministic decision rules. Therefore, evaluation functions of many forms can be used, for instance, the simulation results on performance of the supply chain such as production balance in replenishment strategy could be defined as the evaluation functions of the GA.
III.

replenishment cycle PB (i, i): Percentage of PVBi being replenished in the ith replenishment cycle PA (i, i-1): Percentage of EPVAi being adjusted to the (i-1)th replenishment cycle PB (i, i-1): Percentage of EPVBi being adjusted to the (i-1)th replenishment cycle PA (i, i+1): Percentage of EPVAi being adjusted to the (i+1)th replenishment cycle PB (i, i+1): Percentage of EPVBi being adjusted to the (i-1)th replenishment cycle LOSAj: Lost sales for order A in week j, j=1 to SS LOSBj: Lost sales for order B in week j, j=1 to SS DEMj: Customer demand for both order A and B in week j, j=1 to SS The optimization of VMI strategy is based on the following assumptions: A1. The VMI replenishment quantity in the ith replenishment before optimization PVAi and PVBi should be replenished to order A and B so far as sum of PVAi and PVBi is less than space for VMI replenishment which equaled to the difference between the production limitation PLi and order based quantity OBi. A2. The VMI replenishment quantities above the upper bound of the production capacity in the ith replenishment could be shifted to the contiguous replenishments, i.e., the (i-1)th replenishment or the (i+1)th replenishment. A3. In the ith replenishment cycle, the replenishment priority of PVAi and PVBi was higher than that of the EPVA (i+1), EPVB (i+1), EPVA (i-1), and EPVB (i-1). Take the ith replenishment cycle as an example, the optimization procedures to determine replenishment quantity under VMI strategy includes: Step 1: Calculate the difference between PLi and order based quantity OBi. Step 2: Determine the percentages of PVAi and PVBi to be replenished in the ith replenishment cycle PA (i, i) and PB (i, i), particularly in condition that the sum of PVAi and PVBi is greater than the difference between the production limitation PLi and order based quantity OBi. Step 3: Determine the percentages of EPVA(i+1) , EPVB(i+1) , EPVA(i-1) and EPVB(i-1) to be shifted into the ith replenishment PA (i+1, i), PB (i+1, i), PA (i-1, i) and PB (i-1, i). Step 4: Obtain the optimal replenishment quantity under VMI strategy in the ith replenishment cycle for order A and order B OVAi and OVBi as follows:
OVAi PA(i,i)*PVAi  PA(i-1,i)*PVA(i-1)  PA(i  1,i)*PVA(i  1) (1) OVBi PB(i,i)*PVBi  PB(i-1,i)*PVB(i-1)  PB(i  1,i)*PVB(i  1) (2) This optimization procedure is repeated in each replenishment cycle.

A SIMULATION-BASED OPTIMIZATION MODEL Optimization Replenishment Strategy

A.

The proposed model links the apparel supply chain VMI replenishment with production balance in manufacturers. The following notations are used in the optimization replenishment model: SS: No. of weeks in the sales season Ord: Order of different retailers adopting VMI replenishment strategy, Ord=A or B; i: Replenishment cycle, i=1 to n, where n is the total replenishment cycles in the sales season PLi: Maximum production capacity in the ith replenishment cycle OBi: Order-based quantity in the ith replenishment cycle Vi: Maximum VMI replenishment quantity in the ith replenishment cycle, Vi= PLi -OBi PVAi: VMI replenishment quantity in the ith replenishment cycle before optimization for order A PVBi: VMI replenishment quantity in the ith replenishment cycle before optimization for order B OVAi: Optimized VMI replenishment quantity in the ith replenishment cycle after optimization for order A OVBi: Optimized VMI replenishment quantity in the ith replenishment cycle after optimization for order B EPVAi: Surplus quantity of PVAi beyond the production capacity in the ith replenishment cycle EPVBi: Surplus quantity of PVBi beyond the production capacity in the ith replenishment cycle PA (i, i): Percentage of PVAi being replenished in the ith

B.

Simulation Replenishment Model

2006 IEEE International Conference on Management of Innovation and Technology

849

In order to experimentally investigate the performance of the proposed optimization replenishment model, a simulation model that imitates the operation of the supply chain is proposed. Firstly, a replenishment simulation program is presented to generate the VMI replenishment strategy which satisfies the retailers defined CSLs. The replenishment simulation imitates customers buying at the retail store and records the inventory held by the retailer. The manufacturer regularly replenishes goods for the retailer based on this information. The simulation procedure is validated with the data from industry[13]. The replenishment quantity before optimization for order A and order B, namely, PVAi and PVBi are thus calculated. Secondly, a simulation combining with genetic algorithm (GA)[11, 14] is devised to search the optimized replenishment strategy so as to optimize the manufacturers production capacity balance as well as minimize the lost sales of the retailers. Five steps are conducted in this stage: Step 1: Predict replenishment quantity of 3 continuous periods using replenishment simulation program. The replenishment quantity for order A and order B are denoted as RVAi, RVBi, RVA(i+1) RVB(i+1), RVA(i+2) and RVB(i+2). In first replenishment cycle, these 6 variables are equal to the replenishment quantity generated before optimization, i.e. PVAi, PVBi, PVA(i+1) , PVB(i+1), PVA(i+2) and PVB(i+2). Step 2: Adjust RVAi, RVBi, RVA(i+1) RVB(i+1), RVA(i+2) and RVB(i+2) to intermediate variable GRAi, GRBi, GRA(i+1) GRB(i+1), GRA(i+2) and GRB(i+2) which are the optimal solution to satisfy the model objectives using GA (Details of GA see section C). Step 3: Set optimal replenishment quantity OVAi GRAi , OVBi GRBi . In this study, rolling horizon method[15] is adopted in which only the upcoming VMI replenishment quantity GRAi and GRBi rather than the total six intermediate variables GRAi, GRBi, GRA(i+1) GRB(i+1), GRA(i+2) and GRB(i+2) are set as the optimized VMI replenishment quantity OVAi and OVBi. This rolling procedure is adopted because it simulates the practical replenishment procedure in the apparel supply chain. Once certain quantity OVAi and OVBi are shipped to the retailer, the POS data and the replenishment quantity for the next cycles are changed compared with those of PVAi and PVBi. Therefore, the replenishment quantity for the next 2 cycles are no longer the previous PVA(i+1), PVB(i+1) , PVA(i+2) and PVB(i+2). The optimization should be conducted based on the most updated situation. Step 4: Input the optimized replenishment quantity OVAi and OVBi to the replenishment simulation program to generate the prediction of replenishment quantity RVA(i+1), RVB(i+1), RVA(i+2), RVA(i+2), RVA(i+3) and RVA(i+3)for the next 3 replenishment cycles, after imitating customers buying and POS data recording at the retailers. Step 5: Repeat steps 1-4 until the end of the sales season. The optimized replenishment strategy is thus obtained.

Fig. 1 illustrates the structure of the VMI replenishment strategy simulation model. C. Genetic Algorithms (GAs) for Optimization

In order to search the optimal solution to the replenishment quantity for order A and B in each replenishment cycle, Genetic Algorithm (GA) is employed. The first step in constructing GAs is to define an appropriate genetic representation (coding). In this study, chromosomes of the individuals are real coded in floating point numbers. Fig. 2 illustrates the chromosome which segmented in two parts used in the ith optimization procedure in our proposed model. The first part of the chromosome is composed of three coefficients CO1 to CO3 to determine the proportion of replenishment quantity for order A and order B in each replenishment cycle. Take CO1 as an example, it reflects the weight of PVAj and PVBj occupying in the jth replenishment cycles. These three coefficients should be subjected to:
COj PVAj ( PVAj  PVBj ) COj*PVAj  (1  COj)*PVBj d Vj PVAj ( PVAj  PVBj ) 0 d COj d COj*PVAj  (1  COj)*PVBj d Vj

if PVAj  PVBj d Vj (3)

otherwise

(4)

Fig.1. Structure of the VMI replenishment strategy simulation model

CO1 CO2 CO3 P1 Part 1

P2

P3

P4

P5

P6

P7

P8

Part 2
Fig.2. Representation of GA chromosomes

850

2006 IEEE International Conference on Management of Innovation and Technology

In the second part, eight real numbers P1 to P8 are employed to represent the percentage of exceeding replenishment quantity of order A and B in the ith, (i+1)th and (i+2)th replenishment cycle, to be adjusted to their contiguous replenishment cycles. Specifically, P1: percentage of EPVAi to be adjusted to (i+1)th replenishment cycle P2: percentage of EPVBi to be adjusted to (i+1)th replenishment cycle P3: percentage of EPVA(i+1) to be adjusted to ith replenishment cycle P4: percentage of EPVB(i+1) to be adjusted to ith replenishment cycle P5: percentage of EPVA(i+1) to be adjusted to (i+2)th replenishment cycle P6: percentage of EPVB(i+1) to be adjusted to (i+2)th replenishment cycle P7: percentage of EPVA(i+2) to be adjusted to (i+1)th replenishment cycle P8: percentage of EPVB(i+2) to be adjusted to (i+1)th replenishment cycle These eight percentages should be subjected to:
EPVAj EPVBj PVAj  PVAj*COj PVBj  PVBj* (1  COj ) j j j i to i  2 i to i  2 i (5) ( 6) (7)

f1

PL  (OV
k k j j2 k k j

j2

Ak

 OVBk  OBk ) j 1 to n  2  OVBk  OBk ) (10)

PL  (OV
LOS
k j j2 k j j2 Ak

Ak

 LOSBk j
k

f2

1 to n  2

(11)

DEM

PVAj*COj  PVBj* (1  COj )  P3 * EPVA( j  1)  P 4 * EPVB ( j  1) d Vj

PVAj*COj  PVBj* (1  COj )  P1 * EPVA( j  1)  P 2 * EPVB ( j  1)  P5 * EPVA( j  1)  P 6 * EPVB ( j  1) d Vj j i 1 (8) PVAj*COj  PVBj* (1  COj )  P 7 * EPVA( j  1)  P8 * EPVB ( j  1) d Vj j i2 (9)

The selection strategy of GA is based on the natural law of survival of the fittest in which the chromosomes are selected for the next generation in terms of their fitness. In this study, the Roulette-wheel-selection scheme[14] is utilized. Crossover options specify how the GA combines two individuals, or parents, to form a crossover child for the next generation. The procedure for computing offspring from two parents in this paper is based on Simulated Binary Crossover (SBX)[16]. In order to maintain diversity in a population from one generation to the next, a mutation operator is normally applied. In this paper, this mutation is implemented based on method in [17] which is similar to SBX for crossover. Both the genetic operators of crossover and mutation are randomly used. In the proposed replenishment strategy optimization approach, a crossover rate of 0.7 and mutation rate of 0.04 are chosen after testing. The process of selection, crossover and mutation repeats in each generation with the objective of reaching the global optimal solution. IV. EXPERIMENTS AND RESULTS

The GA operates on a population of chromosomes. In this study, each chromosome is randomly initialized by assigning each gene a real number. Since not all combinations of the decision variables constitute feasible solution to ensure that the range of adjustment of the surplus quantity to the contiguous replenishment cycles is from 0% to 100%, infeasible solutions are repaired as follows: 1) Set Pi 0 in condition that Pi  0 (i=1 to 8); 2) Set Pi 100% in condition that Pi ! 1 (i=1 to 8); 3) If P 3  P5 ! 100% , or P 4  P6 ! 100% discard the chromosome and randomly generate another one until the population size is satisfied. Fitness function is defined as the fitness of each chromosome so as to determine which will reproduce and survive into the next generation. Given a particular chromosome, the value of fitness function, fitness, represents its probability to survive. Since each optimization procedure includes 3 replenishment cycles, the periods of performance evaluation are set as 3 replenishment cycles. These fitness functions are calculated using simulation program. Fitness of each chromosome is defined as follows:

Experiments were conducted on the basis of the practical experience from the industry using the proposed simulation model. Table I shows the input variables of the model for two experimental tests. In each experiment, orders A and B were two orders adopting VMI replenishment strategy with different CSL requirement. Namely, for order A, the retailer defined CSL was 95%; for order B, this figure was 90%. Tables II and III illustrate the performance comparison of the VMI replenishment strategy and the optimized one in the two experiments. Since the forecasting error, including volume error, the SKU mix error and seasonality distribution error were randomly distributed to each SKU in each week during the sales season; four different distribution cases of the same forecasting error were randomly generated in each experiment. The results shown in Tables II and III are the mean values and standard deviations of the four cases. In both experiments, the improvements on the production balance were significant. In experiment 1, over 21% of the total production quantity that exceeding the production capacity limitation could be shifted to the continuous replenishment cycles while in experiment 2, this figure was about 17%. Meanwhile CSL could be kept at a high level. In

2006 IEEE International Conference on Management of Innovation and Technology

851

experiment 1, the declines of the CSL for order A and B were 1.7% and 3.7% respectively. In experiment 2, the declines of CSLs in order A and B were 1.3% and 3.6% respectively. These results showed that the optimized replenishment strategy could work well to balance the different profits for participants in the supply chain. Specifically, for the manufacturer, the cost for imbalanced production was saved; and for the retailers, the CSL declined a little.

V.

CONCLUSIONS

In this paper, an optimization VMI replenishment model is developed for manufacturers who adopting VMI strategy with its retailers in the supply chain. With the development of simulation program and genetic optimization algorithm, the proposed model is implemented. Experimental results indicate that the proposed the replenishment strategy could improve the production capacity balance of the manufacturer significantly while sacrificed limited CSL of the retailers. ACKNOWLEDGMENT

TABLE I INPUT IN EXPERIMENTS 1 & 2

Input Sales season Replenishment cycle Lead time

Experiment 1 Order A Order B 28 weeks 2 weeks 7 weeks

Experiment 2 Order A Order B 36 weeks 2 weeks 7 weeks

[1] [2] [3]

The authors would like to thank the Institute of Textile and Clothing, The Hong Kong Polytechnic University for the funding supporting. We also thank TAL Apparel Limited for data providing. REFERENCES
S. C. Kulp, H. L. Lee, and E. Ofek, "Manufacturer benefits from information integration with retail customers," Magn. Sci., vol. 50, pp. 431-444, 2004. S. M. Disney and D. R. Towill, "The effect of vendor managed inventory (VMI) dynamics on the Bullwhip Effect in supply chains," Int. J. Prod. Econ., vol. 85, pp. 199-215, 2003. R. L. Behling, "Supply chain management with capacity constraints," American Production & Inventory Control Society 25th Annual Conference Proceedings, 1982. C. Chandra and J. Grabis, "A coordinated supply chain dynamic production planning model," Proceedings of SPIE, 2001. D. Vlachos and G. Tagaras, "An inventory system with two supply modes and capacity constraints," Int. J. Prod. Econ., vol. 72, pp. 41-58, 2001. D. R. Towill, M. M. Naim, and J. Wikner, "Industrial dynamics simulation models in the design of supply chain," Int. J. Phys. Distr. logist. Magn., vol. 22, pp. 3-13, 1992. R. Mason-Jones, M. M. Naim, and D. R. Towill, "The impact of pipeline control on supply chain dynamics," Int. J. Logist. Magn., vol. 8, pp. 47-62, 1997. H. Nuttle, R. E. King, and N. A. Hunter, "A stochastic model of the apparel retailing process for seasonal apparel," J. Text. Inst., vol. 82, pp. 247-259, 1991. R. E. King and N. A. Hunter, "Demand re-estimation and inventory replenishment of the basic apparel in a specialty retail chain," J. Text. Inst., vol. 87, pp. 31-41, 1996. B. Lowson, R. E. King, and N. A. Hunter, Quick response: managing the supply chain to meet consumer demand, 1st ed. Chichester; New York, N.Y.: John Wildy & Sons, 1999. J. H. Holland, Adaptation in Natural and Artificial Systems, 1st ed. Michigan: University of Michigan Press, 1975. Z. Michalewicz, Genetic algorithms+Data structures=Evolution Programs, 3rd ed. New York: Springer-Verlag, 1996. A. H. Dong, W. K. Wong, S. F. Chan, and K. W. Yeung, "Developing an apparel supply chain simulation model with application of fuzzy logics," 17th IMACS world congress: scientific computation, applied mathematics and simulation 2005, Paris, France, 2005. D. E. Goldberg, Genetic algorithms in search, optimization, and machine learning, 1st ed. Reading, Mass.: Addison-Wesley, 1989. K. R. Baker, "An experimental study of the effectiveness of rolling schedules in production planning," Decision Sci., vol. 8, pp. 19-27, 1977. K. Deb and R. B. Agrawal, "Simulated binary crossover for continuous search space," Complex Sys., vol. 9, pp. 115-148, 1995. K. Deb, Multi-Objective Optimization Using Evolutionary Algorithms, 1st ed. New York: Wiley, 2001.

Style/Color/Size 3/4/5 5/6/7 3/4/5 5/6/7 Forecasting error 0/20/20/10/5 20/25/25/15/25/10/10/10/315/20/16/9/8 Required CSL 95% 90% 95% 90%

TABLE II COMPARISON OF PERFORMANCE IN EXPERIMENT 1

[4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

Performance Evaluation CSL Performance Evaluation CSL VMI over PL (qty) VMI over PL (%)

Order A (Statistical result of 4 cases) Before optimization Mean value SD After optimization Mean value SD

0.957 0.002 0.94 0.002 Order B (Statistical result of 4 cases) Before optimization Mean value 0.908 13874 21.91% SD 0.008 1678 0.03 After optimization Mean value 0.871 0 0 SD 0.007 0 0

TABLE III COMPARISON OF PERFORMANCE IN EXPERIMENT 2

Performance Evaluation CSL Performance Evaluation CSL VMI over PL (qty) VMI over PL (%)

Order A (Statistical result of 4 cases) Before optimization Mean value SD After optimization Mean value SD

0.961 0.002 0.948 0.005 Order B (Statistical result of 4 cases) Before optimization Mean value 0.908 11584 17.76% SD 0.008 1204 0.02 After optimization Mean value 0.872 0 0 SD 0.001 0 0

[14] [15] [16] [17]

852

2006 IEEE International Conference on Management of Innovation and Technology