Setting production card levels for the CONWIP material control strategy in multiproduct manufacturing systems

We develop an analytical method to set production card counts for the Pull-CONWIP material control strategy. We use queueing network models and Mean Value Analysis with the Schweitzer-Bard approximation (SB-MVA) for this analysis. Based on certain assumptions, we first analyze the Pull-CONWIP strategy as a closed queueing network and set the number of cards to achieve target throughputs. Then, we analyze Pull-CONWIP as a mixed queueing network and determine the number of cards to achieve target service levels.; Queueing models of Pull-CONWIP do not have a product form solution. Hence we first verify the accuracy of SB-MVA for these models by comparing SB-MVA throughput predictions with observations from simulation. During this verification we show that not all target throughputs can be obtained in these networks. The main reason for this inability to achieve target throughputs is the integer values of card counts Ni for product type i. SB-MVA involves solving a system of nonlinear equations and there is no restriction for Ni to be integer. This leads to several steps: (i) we get solutions to these equations which achieve target throughputs exactly, however the resulting Ni are real numbers; (ii) we interpret these values to represent the average number of cards of each product type in the network; (iii) we implement a control algorithm to ensure that the actual (integer) number of cards in the network is manipulated to achieve this target value; (iv) we extend SB-MVA to handle the case where Ni < 1. Then we propose a mixed integer nonlinear optimization model to set Pull-CONWIP card counts to achieve target throughputs. Simulation results verify that this method achieves target throughputs quite accurately.; Next we analyze the mixed queueing network model of the Pull-CONWIP strategy in order to set card counts to achieve a target service level. For our research we propose an iterative approach based on the aggregation method along with markov chain analysis. Simulation results verify that the card counts estimated by this method achieve the desired service levels quite accurately.