Fault Diagnostics of 3-phase Induction Motor using CNN with Park’s Vector Approach Input Data Configuration

This study proposed a method for reducing the estimation error and loss rate and improving the diagnostic accuracy in 3-phase induction motor fault diagnoses using CNN. To improve the diagnostic accuracy of CNNs, previous studies focused on the diversity of the input data by combining various sensors or solving algorithm-related problems in the CNN. However, in terms of image processing, numerous studies have focused on the preprocessing of input data and noted clear differences in the error and recognition rates according to the preprocessing method. Therefore, preprocessing input images is crucial when using a CNN to diagnose various devices, including 3-phase induction motors. Accordingly, this study proposed a preprocessing method that changes the image using Park’s vector approach (PVA) and converts the circular pattern obtained through PVA into an input image. The conventional method showed an error reduction rate of $ 0.35\%$, whereas the proposed method yielded an error reduction rate of $ 0.14\%$, thus minimizing the estimation error during CNN processing. Furthermore, according to an experiment on diagnosis and fault classification using CNN, the performance improved by $ 18.18\%$ compared to the conventional method $ 99.41\%$.