State Estimation and Life Prediction of IGBT Devices Based on Particle Filtering Algorithm

Aiming at the traditional analytical life model which is difficult to consider the IGBT device parameters and service life dispersion, this paper proposes a particle filtering algorithm based IGBT device state estimation and life prediction method based on the accelerated aging experimental data of IGBT devices and considering the device dispersion. Firstly, a distributed parameter-based degradation model is established based on the evolution of collector-emitter saturation voltage $V_{\text {CE }}$ during IGBT degradation. Secondly, a state-space equation is formulated using the distributed parameter model, and a particle filter algorithm incorporating state weighting and kernel smoothing is employed for parameter estimation. A particle filter-based prediction model for IGBT state and remaining useful life is then developed. Eventually, power cycling test data are used to validate the proposed method under different prediction horizons. Results demonstrate that the proposed method achieves accurate state estimation and remaining useful life prediction under different prediction times. The remaining lifetime follows a normal distribution, with increasing regularity in degradation states and reduced variance in remaining useful life probability distribution as prediction time advances.