Optimización y control del flujo de materiales en procesos de producción flexibles utilizando aprendizaje profundo

SAAVEDRA SUELDO, CAROLINA; PÉREZ COLO, IVO; DE PAULA, MARIANO; VILLAR, SEBASTIÁN A.; ACOSTA, GERARDO G.

San Juan

Congreso; RPIC 2021; 2021

INAUT (UNSJ-CONICET)

Industry 4.0, currently on the rise, demands increasing flexibility and adaptation of production systems to changing products demands and external factors. The adaptation of the production systems implies frequent and often abrupt changes in the configurations of the shop floors and consequently the movement of materials must be re-planned. Material handling is significant in terms of operative costs and times and it does not add value to the end products. It is desired to optimize the performance of the system based on the degree of movements, buffer usage and waiting times, such that the combination of these minimizes the impact on the process costs. Machine learning algorithms in combination with powerful computational simulators can be mutually leveraged to give rise to solve these kinds of real-world problems, typical of smart factories. In this work, for the optimization approach, we develop a closed-loop decision-making system with a deep reinforcement learning algorithm based on a discrete-event simulation model for material handling. In addition, our proposed approach uses the communication architecture Simulai, which allows interfacing a computational discrete-event simulator and the proposed deep learning-based decision-making algorithm. The functionality of our proposal is evidenced through the obtained results and an optimal solution for the problem stated is reached, proving that an intelligent agent can collaborate in making multiple decisions for smart factories.