Deep learning model for optimizing control and planning in stochastic manufacturing environments

Abstract

Within the context of Industry 4.0, manufacturing plants implement smart technologies, which adopt machine learning and deep learning, to identify manufacturing problems and provide sensible solutions to the experts. In literature, such applications have demonstrated high-performance in terms of feature extraction and big data analysis compared to typical machine learning methods. However, in the context of control and planning, the research efforts are still focused on integrating ad-hoc solutions, e.g., heuristic algorithms and deep reinforcement learning, thus making it difficult to generate general scheduling rules for manufacturing processes that could be easily applied to a variety of stochastic manufacturing systems. To this end, this paper considers a deep learning application for addressing the scheduling optimization problem in multi-stage stochastic manufacturing environments. The aim is the formulation of a predictive model that could derive suitable authorizations for activities under uncertainty. Findings demonstrate that the proposed predictive model generates policies that improve the performance of the integrated operations within manufacturing systems, constituting to efficient and sustainable manufacturing with effective material management and near-to-zero waste.