Integrated maintenance policy optimization in buffered serial production lines using neural networks

This paper employs neural network training to minimize cost and reduce loss in production lines under conventional time-dependent maintenance policies. Using a probabilistic approach based on subsystem reliability, the dynamics ruling the flow of goods through production lines, machines, and buffers are mathematically modeled. The mathematical model presented herein is straightforward. Then, production cost is derived from preventive and corrective maintenance moments, with maintenance intervals treated as control input variables. Numerous policy scenarios are evaluated in this regard, and a neural network model predicts total loss based on multiple inputs, enabling informed decision-making for optimizing preventive maintenance in complex manufacturing systems. The neural network model is trained using 100 numerically generated cases. Eventually, the estimated loss is minimized, using the Nelder–Mead method.