Using Machine Learning and Deep Learning Algorithms for Downtime Minimization in Manufacturing Systems: An Early Failure Detection Diagnostic Service

Abstract

Abstract Accurate detection of possible machine failure allows manufacturers to identify potential fault situations in processes to avoid downtimes caused by unexpected tool wear or unacceptable workpiece quality. This paper aims to report the study of more than 20 fault detection models using Machine Learning (ML), Deep Learning (DL), and Deep Hybrid Learning (DHL). Predicting how the system could fail based on certain features or system settings (input variables) can help avoid future breakdowns and minimize downtime. The effectiveness of the proposed algorithms was experimented with a synthetic predictive maintenance dataset published by the School of Engineering of the University of Applied Sciences in Berlin, Germany. The fidelity of these algorithms was evaluated using performance measurement values such as accuracy, precision, recall, and the F-Score. Final results demonstrated that Deep Forest and Gradient Boosting algorithms had shown very high levels of average accuracy (exceeded 90%). Additionally, the Multinominal Logistic Regression and Long Short Term Memory based algorithms have shown satisfactory average accuracy (above 80%). Further analysis of models suggests that some models outperformed others. The research concluded that, through various ML, DL, and DHL algorithms, operational data analytics, and health monitoring system, engineers could optimize maintenance and reduce reliability risks.