Novel Hybrid Framework for Detection, Discrimination, and Classification of Stator Winding Faults in Induction Motors

This paper presents a hybrid fault diagnosis framework that integrates the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d–q–0</i> transformation with a Temporal Convolutional Attention Network (TCAN) for detecting, discriminating, and classifying stator winding faults in induction motors (IMs). Initially, the Modified-Discrete Fourier Transform algorithm is applied for phasor estimation of acquired current signals, while the phase angle is estimated from voltage signals using a Synchronous Reference Frame Phase-Locked Loop. Thereafter, the estimated phase angle and current phasors from both the supply and remote ends are transformed into their corresponding <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d-q-0</i> components. After extracting domain-specific features of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d-q-0</i> components, the TCAN model has been trained using 10-fold cross-validation. The experimental data were gathered from a laboratory setup capable of simulating various fault severities under different load and fault resistance conditions, including external faults with current transformer saturation. The effectiveness of the presented technique is assessed using standard evaluation metrics and compared with other deep learning models such as CNN, LSTM, GRU and TCN. Results display superior accuracy and fault discrimination capability of the presented approach, highlighting its robustness and reliability for accurate classification of healthy, internal, and external fault situations.