This research introduces a two-stage binary classification methodology for fault detection and identification (FDI) in the stator and rotor of cage induction motors. Utilizing a comprehensive simulation model, stator current signatures were obtained from both healthy motors and those exhibiting broken bar and inter-turn faults. The system was fully implemented and assessed within a MATLAB simulation framework, encompassing data acquisition, fault detection, and fault identification. In the initial stage of the FDI system, inputs are classified as either healthy (no fault) or faulty (broken bar or inter-turn fault). If deemed healthy, no further action is taken. Conversely, detected faults proceed to the second stage, where the system discriminates between broken rotor bars and inter-turn faults. Key signal features, specifically Peak Amplitude and Bandwidth Power were identified as the most distinctive for this classification task. The FDI system employed a 7-fold cross-validation approach to ensure robustness. Among the models tested, the Support Vector Machine (SVM) exhibited the highest classification speed. The classification accuracies achieved were 97.3% for Linear SVM, 99.3% for Quadratic SVM, and 95.6% for Artificial Neural Network (ANN). Additionally, the two-stage classification approach demonstrated significant speed enhancements over the conventional single-stage method, with processing times reduced from 8.2 to 0.64 seconds for Linear SVM, from 12 to 6.7 seconds for Quadratic SVM, and from 14.8 to 8.22 seconds for ANN. These results indicate the developed system's efficacy in providing accurate and efficient fault detection and identification in cage induction motors, offering substantial improvements in both classification speed and accuracy.
Key words: Fault Detection and Identification (FDI); Cage Induction Motor; Two-Stage Classification; Support Vector Machine (SVM); Condition Monitoring
|
