Dynamic production scheduling in virtual cellular manufacturing systems

Abstract

Manufacturing companies must constantly improve productivity to respond to dynamic changes in customer demand in order to maintain their competitiveness and marketshares. This requires manufacturers to adopt more efficient methodologies to design and control their manufacturing systems. In recent decades, virtual cellular manufacturing (VCM), as an advanced manufacturing concept, has attracted increasing attention in the research community, because traditional cellular manufacturing is inadequate when operating in a highly dynamic manufacturing environment. Virtual cellular manufacturing temporarily and dynamically groups production resources to form virtual cells according to production requirements, thus enjoying high production efficiency and flexibility simultaneously. The objective of this research is to develop cost-effective methodologies for manufacturing cell formation and production scheduling in virtual cellular manufacturing systems (VCMSs), operating in single-period/multi-period, and dynamic manufacturing environments. In this research, two mathematical models are developed to describe the characteristics of VCMSs operating under a single-period and a multi-period manufacturing environment respectively. These models aim to develop production schedules to minimize the total manufacturing cost incurred in manufacturing products for the entire planning horizon, taking into consideration many practical constraints such as workforce requirements, effective capacities of production resources, and delivery due dates of orders. In the multi-period case, worker training is also considered and factors affecting worker training are analyzed in detail. This research also develops a novel hybrid algorithm to solve complex production scheduling problems optimally for VCMSs. The hybrid algorithm is based on the techniques of discrete particle swarm optimization, ant colony system and constraint programming. Its framework is discrete particle swarm optimization which can locate good production schedules quickly. To prevent the optimization process being trapped into a local optimum, concepts of ant colony system and constraint programming are incorporated into the framework to greatly enhance the exploration and exploitation of the solution space, thus ensuring better quality production schedules. Sensitivity analyses of the key parameters of the hybrid algorithm are also conducted in detail to provide a theoretical foundation which shows that the developed hybrid algorithm is indeed an excellent optimization tool for production scheduling in VCMSs. In practice, the occurrence of unpredictable events such as breakdown of machines, change in the status of orders and absenteeism of workers will make the current production schedule infeasible. A new feasible production schedule may therefore need to be generated rapidly to ensure smooth manufacturing operations. This research develops several cost-effective production rescheduling strategies for VCMSs operating under different dynamic manufacturing environments. These strategies facilitates the determination of when-to and how-to take rescheduling actions. To further enhance the performance of such strategies in generating new production schedules, especially for large-scale manufacturing systems, a parallel approach is established to implement the developed hybrid algorithm on GPU with compute unified device architecture. The convergence characteristics of the proposed hybrid algorithm are also studied theoretically by using probability theory and Markov chain model. The analysis results show that the optimization process will eventually converge to the global optimal solution.