A CUSUM-based condition monitoring algorithm for power electronics based on SOM-MQE feature extraction method

MOSFET is an important power electronic component and is widely used in high-power, high-availability applications, such as electric vehicles. The anomalous state of a MOSFET has a major impact on the operation of the entire electrical system in a vehicle. Anomaly detection can provide early warning to vehicle operators, thereby avoiding unscheduled maintenance and improving the vehicle availability. This paper presents an unsupervised data-driven method that involves a distance-based feature extraction and the Page's Cumulative sum (CUSUM) detection scheme for real-time detection of incipient faults in MOSFETs. The method utilizes the ON-state signals under a healthy condition to train a self-organizing Map (SOM) model of the device and uses the Minimum quantization error (MQE) of a test device from the SOM centroids as a Health Indicator (HI). Once the statistics of HI of a nominal device are determined, a CUSUM model monitor for changes in the device's operation. The proposed algorithm is able to process multiple measurements at once, leading to a shorter detection time and a better detection accuracy. The effectiveness of the proposed method is validated on 60 accelerated aging experimental data sets with both low-frequency and high-frequency switching. The results demonstrate that the proposed method can detect anomalies of MOSFETs under diverse operating conditions.