State of Health Estimation of SiC MOSFETs Using Convolutional Neural Networks and Long Short-Term Memory Recurrent Neural Networks

Abstract:
The State of Health of emerging wide-bandgap semiconductor devices is an important precursor in the overall powertrain safety. One way to estimate this health index is by implementing data-driven approaches, which require a significant amount of accelerated aging or condition monitoring data. Due to lack of data available, the data for the Machine Learning model was generated using MATLAB Simulink. The powertrain has been parametrised accordingly to an existing inverter powertrain set up. To analyse the aging effect, the on-resistance of the switches was altered every driving cycle as a percentage increase of its initial value up to a familiar point where the first wire-bond failure typically occurs. The powertrain simulation contains sensors for the on-state voltage and the inverter current, and based on the power losses, the junction temperature of the Silicon Carbide Metal Oxide Semiconductor Field Effect Transistors is estimated. These signals are used to develop a Convolutional Neural Network followed by a Long Short-Term Memory Neural Network. Feature extraction is also performed to estimate the most relevant failure mechanisms of the Silicon Carbide Metal Oxide Semiconductor Field Effect Transistors. The output of the model indicates the State of Health of the device. The model is then integrated into Simulink for condition monitoring of the Silicon Carbide Metal Oxide Semiconductor Field Effect Transistors in the powertrain.