Abstract
This study utilised machine learning to analyse the switching waveforms of power semiconductor devices, with a particular focus on wire lift-off and the objective of detecting power cycling failures. Power semiconductor device losses generate heat during the operation of power electronics systems, resulting in power cycling failures. The gate voltage waveform ($V_{g s}$), the drain-source voltage waveform ($V_{d s}$), the source current waveform ($I_{s}$), and the combination of $V_{g s}$ and $I_{s}$ were employed as detection signals, and the accuracy of the analysis was compared using each waveform. Additionally, different types of M-shunts, with and without a coupling area, were utilised for current measurements to examine the effect of deliberate field coupling on the accuracy. The results demonstrate that the analysis employing the $V_{g s}$ waveform and the combined $V_{g s}$ and $I_{s}$ waveforms yield high accuracy for both the turn-off and turn-on conditions, with and without coupling. The use of shunts with coupling structures enables the chosen approach to distinguish between no cut and a single cut. By incorporating coupling structures, the theoretical approach gains practical applicability, as it allows for the detection of slight degradation effects in addition to discrete wire cuts.
Funding Information
  • Japan Society for the Promotion of Science (JSPS) (JPJSBP120243506)

This publication has 10 references indexed in Scilit: