|
Krishna Yaduvanshi
Keywords:
Electronic Systems; Fault Detection; Fault Diagnosis; Machine Learning; Deep Learning; Support Vector Machine; Artificial Neural Network; Feature Extraction; Explainable AI; Physics-Informed Machine Learning; Real-Time Diagnosis.
Abstract:
Modern electronic systems are becoming increasingly complex due to the integration of advanced components across industrial automation, electric vehicles, renewable energy, telecommunications, medical equipment, and other applications. This complexity increases the risk of component degradation, abnormal operation, and system failure, creating a need for reliable and early fault detection and diagnosis. Conventional threshold-based and model-based diagnostic methods may have limitations under varying operating conditions, component ageing, noise, and incipient faults. Machine learning (ML) provides an intelligent data-driven approach by learning relationships between electrical and physical measurements such as voltage, current, temperature, impedance, vibration, and frequency-domain characteristics and different system health conditions. This review discusses the application of supervised, unsupervised, and deep-learning approaches, including Support Vector Machine, k-Nearest Neighbour, Decision Tree, Random Forest, Artificial Neural Network, XGBoost, CNN, LSTM, and hybrid models for electronic-system fault diagnosis. The importance of data cleaning, feature extraction, feature selection, cross-validation, and appropriate train-test splitting is emphasized to improve model reliability and prevent data leakage. Model performance should be evaluated using multiple parameters, including accuracy, precision, recall, F1-score, Matthews correlation coefficient, AUROC, computational cost, robustness, and generalization. The review further highlights Explainable AI, transfer learning, digital twins, edge AI, sensor fusion, and physics-informed ML as emerging approaches for developing transparent, adaptive, and real-time diagnostic systems. Overall, effective fault-diagnosis models should be selected according to the electronic system, fault characteristics, data availability, computational resources, and required interpretability rather than relying solely on classification accuracy.
|
|
International Journal of Recent Research and Review
ISSN: 2277-8322
Vol. XIX, Issue 3
August 2026
|
PDF View
PUBLISHED
August 2026
ISSUE
Vol. XIX, Issue 3
SECTION
Articles
|
