Intelligent System for Real-time Health Monitoring of Power MOSFETs

Abstract As Power MOSFETS plays a significant role in the industrial manufacturing process, it is of immense value to study the failure modes of these MOSFETS. A typical MOSFET failure condition is caused by a short circuit between the source and drain. In addition, excessive current and excessive power dissipation are causes of MOSFET failure. The focus of this research was to create an intelligent data-driven system that uses Prognosis, Neural Networks, and Machine Learning Models to forecast the Remaining Useful Life (RUL) of power MOSFETs. After executing the train and test models on LSTM using a learning rate of 0.001, 50 epochs, and the Rul activation function, it was discovered that the MAE and model loss decreased substantially. A modest hardware setup was developed later for real-time testing of trained and tested LSTM models. After implementing data driven techniques, real-time testing was done under accelerated aging at 70^∘ C. After 3 minutes of intensive accelerated aging, the MOSFET had 34.706 minutes of RUL. Future work will focus on developing a benchmarking circuit capable of operating under a wide range of temperature-controlled conditions. This will allow us to compare the activation energy derived from current models with that of the proposed model.