Random Forest-Bayesian Optimization for Product Quality Prediction With Large-Scale Dimensions in Process Industrial Cyber–Physical Systems

Cyber-physical systems and data-driven techniques have potentials to facilitate the prediction and control of product quality, which is one of the two most important issues in modern industries. In this article, we integrate random forest (RF) with Bayesian optimization for quality prediction with large-scale dimensions data, selecting crucial production elements by information gain, and then utilizing sensitivity analysis to maintain product quality. Horizontal empirical experiments are performed to verify the superiorities of RF embedded within Bayesian optimization over classical RF, support vector machine, logistic regression, decision tree, and even background propagation neural network. Besides, we find fewer but critical features handled by RF-Bayesian optimization can realize satisfactory forecast accuracy as well as cost-effective computing time, where we interpret it with Herbert A. Simon's management decision theory and Pareto principle. Consequently, the results could provide managerial insights and operational guidance for product quality prediction and control at the real-life process industry.

I. Introduction

Cyber–physical system can connect the cyber world with the physical production process and manual operations, which is recognized as a promising accelerator for understanding the production process, improving product quality, enhancing industrial security, realizing sustainably production, and saving labor costs [1], [2]. As a vitally important part of fabrication, the process industry is facing reformulation with the emerging of cyber–physical systems [3]. Process monitoring systems together with data-driven techniques, especially machine learning, are being actively developed for implementations in modern smart factories to improve operation efficiency [4].