Induction Motor Fault Detection and Classification System Using Proposed Fusion Features and Machine Learning Techniques

A machine fault diagnostic system is vital for industrial workshops, emphasizing the essential role of diagnosing diverse faults in induction motors for efficient repair and maintenance. The research diagnoses 2.9kW of induction motor faults through the analysis of acoustic emission signals using time-frequency analysis, employing machine learning classifiers for fault classification. The diagnostic system employs a method for fault detection and classification by utilizing proposed fusion feature vectors and a machine learning algorithm. The first feature vector consists of 16 Linear Predictive Coefficients (LPC). Using the product fusion method, the LPC coefficients are multiplied by the average Short Time Energy and construct the first fusion feature vector (F1). The second fusion feature vector (F2) for fault classification is formed by multiplying Mel-frequency Cepstral Coefficients (MFCC) with the average Spectral Centroid (SC). The process of generating proposed feature vectors begins with the creation of a reference feature vector (F3) from normal acoustic signals of the induction motor. Then, the residual features are formed by subtracting the F1 and F2 feature vectors from the reference feature vector (F3). To detect the motor's condition, the RMSE threshold value is assigned based on the characteristics of the collected signals. The F2 feature vectors serve as input for a machine-learning algorithm to classify the fault types of the induction motor. Faults are classified using a Support Vector Machine if the input test signal is below the assigned threshold value. The experimental outcomes indicate that the proposed system achieved a 99% accuracy in fault detection and a 98.8% accuracy in fault classification.