Intelligent Fault Diagnosis of SEPIC and Ćuk Converters Using Machine Learning Techniques

This paper presents a comparative study on intelligent fault detection in DC–DC power converters, focusing on SEPIC and Ćuk topologies. A machine learning-based framework was developed to automatically classify converter operating conditions using electrical measurements such as voltages, currents, and ripple characteristics. A dataset was generated through simulation under normal and faulty conditions, including MOSFET, diode, capacitor, and inductor failures. Several classification algorithms were implemented and evaluated in Weka, including Random Forest, Support Vector Machines, Decision Trees, k-Nearest Neighbors and Naïve Bayes. The results show that k-Nearest Neighbors achieved the highest classification accuracy of 90%, with excellent precision and recall values. The study demonstrates that machine learning methods provide a reliable and efficient solution for real-time fault diagnosis in power electronic converters.