Hybrid algorithm for fault detection in three-phase motors using Principal Component Analysis

Abstract

This research proposes developing an innovative algorithm for fault detection in three-phase motors using Principal Component Analysis (PCA) and PLECS-Matlab co-simulation. The method is based on the enforced training of the algorithm using normalized historical data of all observed or estimated motor variables. Additionally, the algorithm combines strategies for handling the principal component space and defines detection thresholds using the ratio of variances and Hotelling’s T${}^2$ distribution. This provides an ellipsoid threshold that separates different classes, such as regular operation or fault conditions. The results show that this approach demonstrates excellent efficacy in the timely detection of faults, forming the basis for predictive maintenance management. To record data for different fault classes, a fault injection system was developed in the electrical scheme of the simulation model and in a real squirrel cage induction motor. The simulation model proposed three typical faults for the study: eccentricity, stator harmonics, and rotor bar breakage. The results demonstrate that the fault detection method proposed in this work, validated with fault indices at 10% and 90%, achieves 100% detection of the proposed faults with a detection rate greater than 98%. In addition, similar results were obtained with measured data in a real application. Using PLECS-Matlab co-simulation allowed the development and validation of the proposed fault detection method.