DEEP LEARNING FOR PREDICTIVE MAINTENANCE IN MANUFACTURING: A DATA-DRIVEN APPROACH TO FAULT DETECTION AND SYSTEM OPTIMIZATION

Abstract

This research evaluates deep learning approaches particularly hybrid CNN and LSTM networks to enhance industrial predictive maintenance systems for determining equipment RUL and fault detection.  The evaluation of four deep learning models comprised CNN and RNN and LSTM as well as their hybrid CNN-LSTM model through assessing the performance measures that included RUL prediction accuracy, precision, recall, F1-score, and Mean Absolute Error (MAE).  The hybrid CNN-LSTM model proved superior in both fault detection precocity and failure prediction accuracy since it delivered 92.5% accuracy and 13.1 hours MAE in forecasting RUL.  The resilient fault detection capability of the model produced a high F1-score that reached 93.4%.  The hybrid technique decreased false negative outcomes in the confusion matrix to ensure prompt defect discovery for the reduction of unplanned system downtime.  The model received evaluation by applying live industrial data ensuring both practical application and compatibility with real manufacturing settings.  The research demonstrates that hybrid deep learning models represent practical approaches to improve industrial predictive maintenance thus resulting in reduced costs and improved operational efficiency.