Decision synthesis framework approach for dynamic dispatching and preventive maintenance in large flexible manufacturing systems
Yang Yang,Bin Zhang,Cheng-Hung Wu
DOI: https://doi.org/10.1016/j.aei.2024.102417
IF: 8.8
2024-03-07
Advanced Engineering Informatics
Abstract:This research investigates the challenges of dynamic dispatching and preventive maintenance (DDPM) in large flexible manufacturing systems (FMS) that handle various products and machines. The creation of real-time DDPM policies is crucial for improving production cycle times, reducing production cost, and maximizing throughput. Markov decision processes (MDP) are used to consider stochastic characteristics, such as the random deterioration of machine health and the uncertain demands, in formulating DDPM problems. However, given that solving large MDP models is computationally inefficient, we divide the overall MDP models into two-product and single-machine (2P1M) subproblems. Then, we propose a decision synthesis framework (DSF) to coordinate those 2P1M solutions to generate near-optimal global solutions. The computational complexity of the DSF approach increases linearly with the number of machines and quadratically with the number of products. This polynomial growth of computational complexity overcomes the curse of dimensionality for the DDPM problems and makes the DSF approach scalable in large systems. In small systems where DDPM can still be optimized, the DSF method attains an average accuracy of 92.90 % compared with the optimal solutions. It reduces the computational time by more than 90 % compared with the optimal solutions and the solutions generated by deep Q-network. For large systems of 15 products and 6 machines, where the optimum solutions are unknown, the DSF approach upholds throughput optimality in numerical assessments and significantly reduces production cycle times compared with existing DDPM methods from the literature.
engineering, multidisciplinary,computer science, artificial intelligence