Surrogate Model of a Manufacturing System Using Random Forest Method

Abstract

This paper investigates the development of a surrogate model for a simulation model of a manufacturing system made in Tecnomatix Plant Simulation. Discrete-event simulations have a need of significant computational resources, which limits the scope for analysis and optimization. As an alternative, the use of machine learning methods to estimate system behavior is proposed. Based on a generated dataset (over 70,000 observations), a random forest model is trained to predict key performance indicators: throughput, failure time and congestion. It is shown that the random forest provides high accuracy $\left(\mathrm{R}^{2} > 0.86\right)$ and allows for the identification of key factors influencing the system's behavior.