Diagnosis of Bearing Faults Using Temporal Vibration Signals: A Comparative Study of Machine Learning Models with Feature Selection Techniques

Accurately identifying bearing defects is crucial for guaranteeing the dependability and effectiveness of industrial systems. Although the use of vibration signals for diagnosing bearing faults is highly important, there are still persistent obstacles, especially when it comes to detecting minor damage in the early stages. Relying solely on time-domain analysis for statistical feature extraction in complex multi-fault scenarios may lack robustness. The computational difficulties of frequency-domain and time–frequency approaches impede the real-time identification of emergent errors, despite their effectiveness. Although machine learning holds potential, its reliance on attributes that are not generated from the time domain poses a difficulty. Therefore, the main aim of this study is to fill these deficiencies by examining the utilization of temporal vibration signals for the purpose of diagnosing bearing defects. The vibration signals originated from the Case Western Reserve University. A total of fourteen time-domain features were derived from the vibration signal, encompassing root mean square, kurtosis, and skewness. The study employed two feature selection strategies, specifically Information Gain and Fast Correlation-Based Filter (FCBF), to identify the most important seven features for training machine learning models, including k-Nearest Neighbor (kNN), Support Vector Machines, and Naïve Bayes. Based on the acquired data, the kNN-based FCBF model (kNN-FCBF) approach exhibited superior classification outcomes in comparison to alternative methods. The evaluation metrics, including Area Under the Curve (AUC), Accuracy (AC), F1-score, Precision, and Recall, demonstrated a robust performance. The AUC attained a value of 99.1%, AC was assessed at 97%, F1-score reached 96%, Precision was 96%, and Recall earned a score of 95.7%. The benefits of the kNN-FCBF model were emphasized by a comparison analysis with prior studies. The kNN-FCBF algorithm provides a straightforward and precise approach that is less intricate and computationally affordable, while still achieving high levels of accuracy.