Early Bearings Fault Detection Using Acoustic Signals and Machine Learning Algorithms

Bearings are critical components in rotating machinery, and early fault detection is essential to prevent costly failures and ensure operational efficiency in industrial machines. Undetected bearing faults can lead to severe equipment damage, unscheduled stops, increased downtime, and safety risks. While traditional methods of monitoring, particularly vibration analysis, have been effective in identifying early-stage bearing issues, challenges such as fixation difficulties and sensitivity to frequency response persist. Acoustic sensors offer an interesting advantage by capturing sound emissions from the entire machine rather than just specific areas, thus providing broader analysis and allowing also non-contact measurements that alleviate fixation concerns. This research aims to harness recent advancements in artificial intelligence (AI) to develop innovative fault detection methodologies utilizing sound signals. The study investigates the effectiveness of sound-based monitoring in conjunction with Random Forest algorithms for the early detection of bearing faults, employing a SHAP Tree Explainer to enhance explainability. By utilizing machine learning techniques, this approach successfully identifies patterns and anomalies in acoustic data that may signal impending failures, indicating its potential as a dependable tool for predictive maintenance within the context of Industry 4.0.