A Fast and Intelligent Monitoring of Solder Layer Degradation in Multichip IGBT Modules Based on Copper Substrate Temperature Field Reconstruction
Fatigue-induced delamination of the direct-bonded copper (DBC) solder layer is a critical and latent failure mode in multichip Insulated Gate Bipolar Transistor (IGBT) power modules.. To address the challenge of online diagnostics, this letter proposes a novel method for monitoring solder degradatio...
| Autores: | , , , , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universidad de Alcalá (UAH) |
| Repositorio: | e_Buah Biblioteca Digital Universidad de Alcalá |
| Idioma: | inglés |
| OAI Identifier: | oai:ebuah.uah.es:10017/67279 |
| Acceso en línea: | http://hdl.handle.net/10017/67279 https://dx.doi.org/10.1109/TPEL.2025.3629744 |
| Access Level: | acceso abierto |
| Palabra clave: | IGBT module Solder degradation Gaussian process regression Vision transformer Physics-informed learning Electrónica Electronics |
| Sumario: | Fatigue-induced delamination of the direct-bonded copper (DBC) solder layer is a critical and latent failure mode in multichip Insulated Gate Bipolar Transistor (IGBT) power modules.. To address the challenge of online diagnostics, this letter proposes a novel method for monitoring solder degradation based on the temperature field of the copper baseplate. First, sparse temperature data are collected from a thermocouple array placed beneath the module. Then, a physics-guided Gaussian process regression (PG-GPR) model embedded with a 2D steady-state heat conduction equation via a Laplace kernel is constructed to reconstruct a high-resolution temperature field. Finally, the reconstructed field is converted into pseudo-color images and fed into an enhanced Tiny Vision Transformer (Tiny-ViT) network for degradation classification. Experiments conducted on eight customized IGBT modules with varying solder void ratios validate the approach. Compared with the average of bilinear and GPR interpolation baselines, the proposed PG-GPR reduces mean absolute error (MAE) by 34.2% and improves structural similarity index measure (SSIM) by 7.6%. The classification accuracy of solder degradation states reaches 99.75% with an inference delay of only 4.37ms. These results show that using the baseplate temperature field as the diagnostic variable, together with physical priors and a lightweight vision model, enables accurate online monitoring of solder degradation. This method helps improve the safety and lifetime of high power devices. |
|---|