An Efficient Fault-Prediction Mechanism for Improving Yield in Industry 5.0

Abstract

Industrial sectors are constantly under pressure to produce higher-quality goods while maximizing yield. Machine maintenance is a critical component of manufacturing, accounting for a significant portion of total production costs. Corrective, preventive, and conditional maintenance strategies only make a negligible contribution to cost and downtime reduction. With the fifth industrial revolution, industrialists can now use sensors and cyber-physical systems to perform predictive maintenance on manufacturing operations. This strategy eliminates unnecessary maintenance and minimizes downtime by continuously collecting and analyzing data to predict time to failure. Numerous approaches to fault prediction have been proposed for predictive maintenance, but most of them are prohibitively expensive due to the massive number of features in manufacturing machines. The purpose of this work is to develop a technique for reliably predicting machine problems with the fewest possible features. To select features, we used SVR-based Recursive Feature Elimination (SVR-RFE) and Random Forest Regressor (RFR), while to predict, we used Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). Our experiments on the 2018 PHM Challenge Dataset demonstrate that the proposed strategy outperforms prior approaches and reduces the mean absolute percentage error (SMAPE).