| Due to its features, permanent magnet synchronous motor (PMSM) has gained popularity and is used in various industrial applications, including those with high downtime costs like offshore equipment. Inter-turn short-circuit (ITSC) fault is one of the most typical PMSM faults and therefore is its early diagnostics in real-time highly valuable. Solving the problem using conventional signal, model-based, or data-driven approaches faces challenges such as computational complexity, time demand, or need for detailed domain expertise. This paper presents a computationally simple, robust, and accurate method based on the 2D convolutional neural network (CNN). The proposed data-driven model has first been validated with the help of experimental data obtained from an inverter fed PMSM subject to ITSC faults in different time intervals, and secondly its performances have been compared to a model-based structural analysis approach using Dulmage-Mendelsohn decomposition tool. The comparison is based on the same data. Results show that the accuracy of the CNN model for diagnosing early faults is more than 98% without doing additional comprehensive fine-tuning. In addition, the paper presents a robust method that can be successfully used as a metric for fast fault detection benchmark. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.