一种基于电磁声纹的IGBT器件多工况老化状态诊断方法

Health status monitoring of power electronic devices is essential for ensuring their safe and reliable operation. In recent years, monitoring switching transient electromagnetic voiceprint (EMVP) signals has emerged as a promising new indicator for assessing device health. However, the identification of aging states based on EMVP still relies on manual interpretation, leading to low accuracy. This paper presents a health status monitoring method for Insulated Gate Bipolar Transistor (IGBT) devices using EMVP signals, applicable across various operating conditions. Specifically, we propose a spatiotemporal feature fusion cross attention neural network for aging state identification. Experimental results demonstrate that the network achieves an accuracy of over 95% in detecting the aging state of IGBT devices. Furthermore, a transfer learning approach is introduced to improve the model’s effectiveness and generalization ability when working with small sample datasets. The proposed monitoring method facilitates accurate and rapid evaluation of IGBT device aging states under diverse operational conditions.