Comprehensive comparison of different earlyBearing fault detection techniques

Bearing fault detection is a part of predictive maintenance for rotating machinery to provide early warnings of pending breakdowns, preventing sudden stops in production. This study presents two advanced methods for bearing fault detection utilizing the Case Western Reserve University (CWRU) and the HUST Bearing Datasets: Support Vector Machines (SVMs) optimized by Grey Wolf Optimization (GWO), Particle Swarm Optimization (PSO), and the novel Kepler Optimization Algorithm (KOA), and a deep learning approach using Bidirectional Long Short-Term Memory (BiLSTM) networks. The SVM parameters, box constraint and kernel scale were tuned with GWO, PSO, and KOA to improve fault detection efficiency. These results were compared with those of a BiLSTM-based deep learning model. Our comparison showed that the BiLSTM model significantly outperformed the optimized SVM m odels. Although the optimized SVMs achieved considerable improvements over non-optimized SVM models in fault detection accuracy, they were still inferior to the BiLSTM model. Evaluated based on accuracy, the BiLSTM model consistently performed outstandingly across different fault types and sizes, reaching 100% accuracy on small fault sizes, and accuracies as high as 99.92% on bigger ones on the CWRU Dataset and accuracies as high as 99.58% on the HUST Dataset. The proposed model outperformed several modern models regularly utilized for bearing fault detection. This research highlights the potential of deep learning techniques, specifically BiLSTM, in bearing fault detection, demonstrating their advantage over traditional machine learning models even when optimized with advanced algorithms. This study adds value to the field by showcasing the capabilities of deep learning to enhance predictive maintenance systems.