AI-enabled Early Faults and Anomalies Detection in Electric Inverters

Early fault detection plays an important role in reducing maintenance costs and preventing unexpected and costly downtimes of industrial machines. To this end, Artificial Intelligence (AI)-based mechanisms offer efficient approaches to enhance fault detection accuracy while enabling real-time responses. In this paper, we evaluate different supervised and unsupervised AI-based fault detection models (namely: k-Nearest Neighbors, k-NN; Adaptive Boosting, AdaBoost; XGBoost; Random Forest, RF; Multi-Layer Perceptron, MLP; Long Short-Term Memory, LSTM) for electric inverters, comparing them in terms of prediction accuracy and computational complexity. The experimental results show that, among the considered supervised models, XGBoost and MLP achieve the highest accuracy—approximately 99%— while maintaining the lowest computational complexity, thus positioning them as highly effective in terms of fault detection. In contrast, the considered unsupervised models exhibit lower accuracy and reliability for fault detection.