Learning Machine Based on Optimized Dimensionality Reduction Algorithm for Fault Diagnosis of Rotor Broken Bars in Induction Machine

Abstract

Induction machine health monitoring is considered a developing technology for the online detection of faults that occur even at the initial stage. The objective of this study is to present an artificial intelligence (AI) technique for the detection and localization of adjacent and distant broken bar faults in the induction machine, through a multi-winding model for the simulation of these cases. In this work, it was found that the application of Artificial Neural Networks (ANN) based on Mean Squared Error (MSE) and Random Forest (decision tree) plays an important role in detecting and locating defaults. The stator current signal Ias of a motor in the dynamic state was acquired from a healthy and faulty motor with a broken rotor bar fault. 9 statistical features and 8 wavelet packet parameters are extracted from the stator current signal. These features were employed as an input vector to train and test the ANN and random fores29t and determine whether the motor was running under normal conditions or defective. For optimizing the rotor bar defect classification procedure, feature selection algorithms are adopted, such as BBAT and BPSO. For feature reduction, we used the principal component analysis (PCA) algorithm, to reduce the number of features. The results showed that the random forest classifier based on statistical parameters and wavelet packet parameters followed by PCA can detect the defective with high accuracy (98.3333%) compared to other methods.