Shop floor simulation optimization using machine learning to improve parallel metaheuristics

Abstract

Simulation optimization is a tool commonly used as a decision-making support system on industrial problems in order to find the best resource allocation, which has a direct influence on costs and revenues. The present study proposed an open-source framework developed on Python, integrating different strategies for a novel optimization algorithm. The framework includes multicore parallelism (tested on two different types of computer sets), (two) population-based metaheuristics, and 33 machine learning methods. Moreover, the study tested the framework to optimize resource allocation on a theoretical shop floor case study, evaluating 12 optimization scenarios. The use of metaheuristic with parallelism reduced 88.3% the processing time compared with the serial metaheuristic, while the integration of metaheuristic with the selected machine learning generated an additional reduction of 59.0% on the necessary processing time. The combination of the optimization methods created a solution of 95.3% near the global optimum and time reduction of 95.2%.