Acoustic Based Induction Motor Fault Detection System Using Adaptive Filtering Algorithm and Fusion Based Feature Extraction Method

The proposed machine fault diagnostic system utilizes acoustic signal processing and machine learning for early fault detection and localization in induction motors. The growth of the fault in an induction motor tends to be quick and can result in a significant failure that can lead to economic loss and huge maintenance expenses. Therefore, developing accurate and sensitive induction motor fault diagnostic procedures for the maintenance system is crucial. The main purpose of this paper was to propose an optimized noise reduction technique for an induction motor fault diagnosis system and two novel acoustic feature vectors that can be used in machine learning algorithms. The contribution of this paper is to implement the effectiveness of the fusion features of acoustic signals by concatenating them from different domains. The acoustic dataset for an induction motor is collected in a motor workshop, and the NLMS algorithm is used for background noise cancellation due to its quick adaptation, stability, and efficient error minimization. Data are segmented and normalized during pre-processing, and the induction motor fault diagnosis system is implemented using MATLAB. Zero Crossing Rate (ZCR), Spectral Entropy (SE), and Energy Entropy (EE) feature vectors are combined, and the F1 feature vector is built. Correlation calculations are employed to assess the motor's condition status, and if a fault is detected, the system proceeds with feature extraction for fault localization. In the feature extraction stage for induction motor (IM) fault localization, Gammatone Cepstral Coefficients (GTCC) and Mel Frequency Cepstral Coefficient (MFCC) features are combined to construct the second feature vector (F2). This feature vector is used as training feature data in machine learning algorithms. If the input test signal is strongly correlated with the faulty signals, the type of faults is classified using a Support Vector Machine (SVM) classifier. According to the experimental results, the proposed system achieved an average accuracy of 99% in fault detection, 97.5% in fault localization, and an error rate of 2.5%.