Classifying induction motor faults using spectrogram images with deep transfer learning

For industrial systems to be reliable and efficient, motor faults detection is important. Conventional methods of fault detection are typically expensive and time-consuming. The ability to detection motor faults has advanced significantly in recent years thanks to the application of deep learning techniques. Deep learning algorithms have the ability to automatically extract complicated features from big data sets, which can speed up and improve the accuracy of motor faults detection. Since collecting motor acoustic data is cost-effective and easy, it is advantageous to use it in fault detection. In this study, motor acoustic signals were converted into spectrograms and faults in induction motors were detected using a transfer learning approach with pre-trained models. Totally 8-class fault detection was performed with an accuracy rate of 91.52% using VGG16 and 92.11% using VGG19.