Early fault detection of current sensors operating in a closed-loop control structure using transfer learning

Deep neural networks learnt in line with transfer learning are now increasingly used in electrical drive diagnostic applications. This fact is related to the possibility of completely automating the process using direct processing of the measured quantities. Transfer learning is used to develop universal algorithms to detect various faults. This paper presents the possibility of using the idea of deep convolutional network transfer learning in fault diagnosis of current sensors of induction motors operating in a closed control structure. The versatility of the approach includes the possibility, as demonstrated in the article, of expanding the range of current sensor faults to be analyzed. In the study, a drive system simulation model was used as a source of diagnostic information. Particularly notable is the very high level of precision of the developed system (more than 98.3%) at the same time as the possibility of extending the range of considered faults. Furthermore, the response time to an emerging fault was less than 0.0125 seconds.