Acoustic Fault Diagnosis of Rolling Bearings in Induction Motors Using Time-Frequency Image Analysis

Detecting bearing failures in induction motors is essential for preventing unexpected breakdowns and minimizing industrial downtime. The evaluation of acoustic signals and their spectral characteristics provides a non-invasive method for identifying bearing faults. Over the years, various monitoring techniques have been introduced, incorporating both sound and vibration sensors. Among these, rolling element bearing analysis has become a well-established approach for monitoring rotating machinery conditions. However, the implementation of sophisticated accelerometer-based bearing monitoring systems, which are widely adopted in the industry, may not be cost-effective for non-critical machinery due to extended return on investment periods. Consequently, human auditory perception often serves as an initial diagnostic tool for interpreting bearing noise. This study examines the effectiveness of machine learning and deep learning techniques in diagnosing rolling element bearings. It further explores a classification method that converts 1-D bearing sound signals into 2-D scalogram images and utilizes transfer learning with a pre-trained model. This approach substantially reduces the time required for feature extraction and selection in traditional bearing diagnostics while preserving high accuracy for the dataset employed in this research.