Operating Condition Prognosis of Multi-Phase Electric Drives with Machine Learning Models

Multi-phase electric drives are increasingly adopted in industrial applications due to their fault tolerance, efficiency, and high performance under different operating conditions. Reliable operation requires accurate classification into healthy, degraded, and faulty (H/D/F) states to support condition monitoring, reduce downtime, and enable predictive maintenance. Traditional model-based prognosis methods often struggle with nonlinear system behavior, limiting their effectiveness. To address this, a machine learning (ML)-based framework is proposed for operating condition classification in multi-phase drive systems. The framework involves simulating normal and degraded sensorlevel scenarios, generating training data, training a regression neural network, and validating the trained model. Although sensor degradation is used as a representative case, the framework is generalizable to other fault types. The trained model achieves a maximum average error of 0.29% and a maximum absolute error of 1.5% during training, and less than 3% error in offline validation, confirming its accuracy and real-time applicability.