Enhancing Predictive Maintenance in Industrial Machinery Using Hybrid Deep Learning Models with Sensor Fusion and Anomaly Detection

Abstract

This study examines the use of hybrid deep learning architectures for improving predictive maintenance practices in industrial equipment. The method incorporates sensor fusion methodologies to take advantage of information from various sensor modalities (vibration, temperature, pressure) and utilizes anomaly detection methods to detect abnormal performance from normal working conditions. A hybrid architecture, which uses Convolutional Neural Networks (CNNs) for features extraction and Long Short-Term Memory (LSTM) networks for modeling temporal dependencies, is suggested. The performance of the model is verified on an actual dataset of industrial pump operation, exhibiting excellent gains in prediction accuracy and minimized false alarm rates over conventional methods. The findings indicate the proposed approach's potential for proactive maintenance scheduling, downtime reduction, and operational efficiency optimization.