Novel Approach for Intelligent Fault Diagnosis in Electromechanical Drive Systems Using an Attention-Based Residual 1D Convolutional Neural Network

Numerous existing deep learning-based fault diagnosis models, have been implemented for intelligent fault diagnosis in electromechanical drive systems (EMDS). However, these models often are designed to detect and diagnose single fault types at specific operating conditions. These models are frequently susceptible to overfitting, suffer from the vanishing gradient problem, and struggle with long-range feature dependencies. Such limitations significantly hinder their accuracy and reliability in diagnosing complex fault scenarios for EMDS under different operating conditions using noisy data. To address these limitations, we propose a novel deep learning-based fault diagnosis model named Attention-based Residual 1D Convolutional Neural Network (AR1D-CNN). This model is designed to effectively diagnose complex faults in EMDS across various load conditions using noisy and imbalanced vibration data. Experimental results demonstrate exceptional classification accuracies of 99.98% and 100.00% on the KAIST dataset for 0 and 2 Nm load conditions, respectively, and 99.31% on the AETWPSU broken rotor bar dataset. Notably, the AR1D-CNN outperforms existing state-of-the-art methods by approximately 4% on the AETWPSU dataset, highlighting its robustness and reliability. This research significantly enhances predictive maintenance strategies by enabling early and precise fault detection, ultimately improving the operational efficiency of EMDS.