Data-Driven Control System Using Machine Learning in Production Process

Many manufacturing companies control their production machines to produce good products within quality standards by using the results of research on physical or chemical models. Those models are developed from knowledge of the physical or chemical changes that occur when the products are processed and operational knowledge. However, it is difficult for some companies to research physical or chemical models. We study data-driven control systems to enable the stable production of good products when it is difficult to study and develop physical and chemical models or to use operational knowledge. In this paper, we propose an algorithm that builds an alternative model from actual operation data using machine learning and finds the optimal operating conditions under which the product is within the quality standards range using 0–1 integer programming. The effectiveness of the proposed algorithm was verified using operation data generated using a simulator for a food manufacturing process.