Acoustic Feature Extraction and Classification Techniques for Anomaly Sound Detection in the Electronic Motor of Automotive EPS

Anomaly sound detection (ASD) is a field of research dedicated to the early identification of mechanical failures and the facilitation of preventative maintenance to avert potential hazards and optimize performance. The capability of ASD to detect defects in mechanical equipment within this domain largely depends on the ability to uncover faults that belong to unknown categories. However, the limited labeled anomaly data available in real-world settings remains a challenge in ASD. To overcome this problem, in this study, we introduce a novel method for ASD, namely LSTM-AE with MFCC and dynamic feature maximization transformation (DFMT). The proposed method can improve normal and anomaly sound distinction in the limited anomaly data. Additionally, the unsupervised learning LSTM-AE technique enables the model to effectively learn complex sound patterns and accurately identify anomalies and deviations from normal features. In the proposed method, MFCC and DFMT are combined to generate and integrate feature maps, removing noise while retaining robustness without losing crucial features. Experimental results reveal that the proposed method achieved an accuracy of 99.2%.