Comparative Analysis of Machine Learning Techniques for Fault Diagnosis of Rolling Element Bearing with Wear Defects

Rolling Element Bearings perform a vital function by ensuring the reliable and efficient operation of machinery in modern industries. Timely and accurate diagnosis of bearing faults is essential for preventing unexpected failures and minimizing downtime. This research addresses these challenges by employing advanced signal processing techniques and machine learning algorithms. The study investigates and optimizes fault diagnosis of rolling element bearings using various machine learning techniques, including Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), K-Nearest Neighbors (KNN), and Multi-Layer Perceptron (MLP). The study utilizes naturally occurring defect vibrational data obtained from continuous running in the experimental test rig. Initially, a baseline for fault classification accuracy was established using raw vibration data. Then, Signal-to-Noise Ratio (SNR) was introduced to enhance data quality and alleviate the impact of noise. The model was further refined by extracting 14 types of features from the SNR-enhanced vibration data, presenting a comprehensive depiction of fault patterns and finally, machine learning techniques were applied to categorize faults using the aforementioned datasets, facilitating a comparative analysis of results. This optimization of the signal enhancement methodology significantly improved the fault diagnosis accuracy. As per the result obtained, Random Forest method consistently outperforms when applied to the feature-enhanced SNR dataset. The findings contribute to a more accurate and reliable identification of faults, offering significant advancements in the field of machinery health monitoring and predictive maintenance.