| 作成者 |
|
|
|
|
|
|
|
|
|
|
|
| 本文言語 |
|
| 出版者 |
|
| 発行日 |
|
| 会議情報 |
|
| 出版タイプ |
|
| アクセス権 |
|
| 権利関係 |
|
|
|
| 関連DOI |
|
| 関連DOI |
|
|
|
|
|
| 関連ISBN |
|
|
|
|
|
| 関連HDL |
|
|
|
|
|
| 関連HDL |
|
| 関連情報 |
|
| 概要 |
Bond wire lift-off is a major failure mode in IGBT modules, and monitoring the gate-emitter voltage (Vge) waveform provides an effective indicator for detection. This study applies convolutional neura...l networks (CNNs) to evaluate waveform sensitivity and its impact on classification accuracy under various gate drive conditions. Here, sensitivity refers to the ability to detect fault-related waveform features, while accuracy represents the overall correctness of the classification. Waveform segments analysis reveals that specific regions contribute strongly to detection, while a performance map clarifies the learning progress between training and testing accuracy. The results show that two-step vector control (2-sVC) greatly improves sensitivity, and CNNs can still recognize subtle waveform changes under conventional control. These findings demonstrate CNN’s capability for robust and generalizable health monitoring of power modules.続きを見る
|
Mendeley出力
