A Machine Learning Approach for Adaptive Classification of Power MOSFET Failures

In electric drives, semiconductor devices are susceptible to faults due to high switching stresses. An unsupervised principal component analysis (PCA) approach that utilizes the on-state median values of voltage and current signals as features is proposed; it is computationally efficient for realtime operation. A key innovation is handling variations of the operating point in the feature space using exponentially weighted samples to recursively compute the time evolution of means of on-state features used in a PCA-based anomaly detector. Data from accelerated tests on commercial SiC devices are used to validate the effectiveness of the proposed approach in predicting incipient semiconductor faults. An experimental setup is also used to demonstrate real-time implementation of the proposed method.