ANN vs. traditional machine learning models: A comparative study on open switch fault diagnosis in VSIs for solar pumping systems

This study presents a comparative analysis of fault detection and classification techniques for identifying single and multiple open-switch faults in insulated-gate bipolar transistors (IGBTs) within a two-level, three-phase voltage source inverter (VSI) used in solar water pumping systems. Five machine learning algorithms are evaluated: Artificial Neural Network (ANN), Decision Tree (DT), K-Nearest Neighbors (KNN), Naive Bayes (NB), and Support Vector Machine (SVM). The evaluation employs current-based features (module, surface, and angle) extracted via the Clarke transformation. Standard performance metrics, including accuracy, sensitivity, specificity, precision, and F1 score, are employed for benchmarking. The ANN achieved perfect classification with 100 % accuracy, while the DT provided a competitive result of 95.56 %. In comparison, SVM and NB demonstrated lower performance under the same conditions. These results highlight the effectiveness of ANN for robust and reliable inverter fault diagnosis, while also showing DT as a strong, lightweight alternative. The contributions of this work include establishing a unified framework for comparing machine learning classifiers under identical operating conditions, validating Clarke transformation features for fault diagnosis in noisy signals, and providing quantified benchmarks to support engineers in selecting suitable diagnostic models. By linking algorithmic performance to practical benefits such as improved reliability and efficiency of inverter-based renewable energy systems, this study offers actionable insights for both researchers and practitioners.