Kawady Tamer A, Khater Wagdy M, Abu-Faty Hagar G, Izzularab Mohamed A, Ibrahim Mohamed E
Electrical Engineering Department, Faculty of Engineering, Menoufia University, Shebin El Kom, 32511, Egypt.
Dredging Department, Suez Canal Authority, Ismailia, Egypt.
Sci Rep. 2025 Jun 20;15(1):20125. doi: 10.1038/s41598-025-04462-x.
Induction motors (IMs) are vital in industrial applications. Although all motor faults can disrupt its operation significantly, stator turn to turn faults (ITFs) are the most challenging one due to their detection difficulties. This paper introduces an AI-based approach to detect ITFs and assess their severity. A simulation based on an accurate mathematical model of the IM under ITFs is employed to generate the training data. Recognizing that ITFs directly affect the motor's current balance, complex current unbalance coefficient is identified and used as the key feature for detecting ITFs. Since unbalanced supply voltage (USV) can also disrupt current balance, the AI models are trained to account for USV by incorporating complex voltage unbalance coefficient that helps to distinguish between ITF-induced and voltage-induced imbalances. After feature extraction, the AI models are trained and validated with simulation data. The approach's effectiveness is further tested using an experimental setup, where measurements from motors under various fault conditions, including USV scenarios, are considered. The results indicate that the gradient boosting model outperforms other ML models in detecting ITFs in IMs and assessing their severity. In the pursuit of achieving highest possible performance, DNN is tested and compared with ML models. The study reveals that DNN demonstrates superior performance in all tested scenarios including USV making DNN the top performer that to be used in the proposed approach. The proposed AI-based approach based on DNN offers high accuracy in fault detection and can effectively distinguish between ITFs and USV-induced anomalies, maintaining low estimation errors and robust performance across different operational conditions.
感应电动机在工业应用中至关重要。尽管所有电机故障都会严重干扰其运行，但定子匝间故障因其检测困难，是最具挑战性的一种。本文介绍了一种基于人工智能的方法来检测定子匝间故障并评估其严重程度。利用基于感应电动机在定子匝间故障情况下精确数学模型的仿真来生成训练数据。认识到定子匝间故障直接影响电机的电流平衡，识别出复电流不平衡系数并将其用作检测定子匝间故障的关键特征。由于不平衡电源电压也会扰乱电流平衡，通过纳入复电压不平衡系数来训练人工智能模型以考虑不平衡电源电压，这有助于区分由定子匝间故障引起的不平衡和由电压引起的不平衡。在特征提取之后，利用仿真数据对人工智能模型进行训练和验证。使用实验装置进一步测试该方法的有效性，实验中考虑了来自各种故障条件下电机的测量数据，包括不平衡电源电压情况。结果表明，梯度提升模型在检测感应电动机的定子匝间故障及其严重程度评估方面优于其他机器学习模型。为追求尽可能高的性能，对深度神经网络进行了测试并与机器学习模型进行比较。研究表明，深度神经网络在包括不平衡电源电压在内的所有测试场景中均表现出卓越性能，使其成为所提出方法中表现最佳的模型。所提出的基于深度神经网络的人工智能方法在故障检测方面具有高精度，能够有效区分定子匝间故障和由不平衡电源电压引起的异常，在不同运行条件下保持低估计误差和稳健性能。
