Neural Networks Detect Inter-Turn Short Circuit Faults Using Inverter Switching Statistics

• DOI: 10.36227/techrxiv.19145444.v2
• Publication Date: 2022-07-27T14:26:42Z
• Authors: Mustafa Umit Oner; İlker Şahin; Ozan Keysan
• Keywords: Downtime; Multilayer perceptron; Speedup

<p>Early detection of an inter-turn short circuit fault (ISCF) can reduce repair costs and downtime electrical machine. In induction machine (IM) driven by inverter with a model predictive control (MPC) algorithm, the controller outputs are influenced due to fault-controller interaction. Based on this observation, study developed neural network models using switching statistics detect ISCF IM. The method was non-invasive, it did not require any additional sensors. task, area under receiver operating characteristics curve value 0.9950 (95% Confidence IntervaI: 0.9949 - 0.9951) obtained. At rated conditions, detected located 2-turns (out 104 turns per phase) 0.1 seconds, speedup more than two times compared thresholding-based method. Moreover, we published vector data collected at various load torque shaft speed values for healthy faulty states IM, becoming first publicly available dataset. Together dataset, provided performance baselines three main architectures, namely, multi-layer perceptron, convolutional network, recurrent network.</p>