Generalised Performance Estimation in Novel Hybrid MPC Architectures: Modeling the CONWIP Flow-Shop System

Abstract

The ability to supply increasingly individualized market demand in a short period of time while maintaining costs to a bare minimum might be considered a vital factor for industrialized countries’ competitive revival. Despite significant advances in the field of Industry 4.0, there is still an open gap in the literature regarding advanced methodologies for production planning and control. Among different production and control approaches, hybrid architectures are gaining huge interest in the literature. For such architectures to operate at their best, reliable models for performance prediction of the supervised production system are required. In an effort to advance the development of hybrid architecture, this paper develops a model able to predict the performance of the controlled system when it is structured as a controlled work-in-progress (CONWIP) flow-shop with generalized stochastic processing times. To achieve this, we employed a simulation tool using both discrete-event and agent-based simulation techniques, which was then utilized to generate data for training a deep learning neural network. This network was proposed for estimating the throughput of a balanced system, together with a normalization method to generalize the approach. The results showed that the developed estimation tool outperforms the best-known approximated mathematical models while allowing one-shot training of the network. Finally, the paper develops preliminary insights about generalized performance estimation for unbalanced lines.