Machine Learning–Based Fault Detection in SEPIC DC–DC Converters

In this paper, a SEPIC converter was analyzed from the aspect of detection and classification of defects using machine learning. The combination of electrical parameters and temperature, processed with WEKA, allows for efficient identification of normal and defective states. The results confirm that machine learning, in combination with electrical and thermal measurements, is an effective approach for automatic diagnosis of defects in SEPIC converters. Although the detection of defects in passive components remains a challenge, the achieved results show significant potential for practical application and further improvement of the system. Of the four experimental setups analyzed, the Multilayer Perceptron (MLP) model achieves the best performance shows the best and most balanced results with 96.5% accuracy. The significant accuracy, excellent Kappa coefficient and reliable detection of all types of defects confirm the superiority of this model and its suitability for real-world application in intelligent systems for diagnostics of SEPIC converters.