Skip navigation

Reliability meta-modelling of power components

Reliability meta-modelling of power components

Stoyanov, Stoyan ORCID: 0000-0001-6091-1226 , Tilford, Timothy ORCID: 0000-0001-8307-6403 , Shen, Yaochun and Hu, Yihua (2024) Reliability meta-modelling of power components. In: 47th IEEE International Spring Seminar On Electronics Technology. 15th - 19th May 2024, Prague, Czech Republic. IEEE Xplore . Institute of Electrical and Electronics Engineers (IEEE), Piscataway, New Jersey, pp. 1-7. (In Press)

47241_STOYANOV_Reliability_meta-modelling_of_power_components.pdf - Accepted Version

Download (752kB) | Preview


Finite element (FE) modelling integrated with lifetime prediction models is an attractive and powerful approach for predicting and improving the thermal fatigue reliability of power electronic components and modules subjected to temperature cycling loads. The challenge with the FE-based modelling approach is the model development effort, device characterisation data requirements and the computational cost of the high-fidelity simulation. This paper presents a modelling methodology for developing fast and user-friendly damage prediction models for power components that benefit from the combined deployment of meta-modelling and machine learning (ML), using physics-informed damage data. The main attribute of the meta-modelling framework is the FE-like mapping of the spatial distribution of the thermal fatigue damage parameter in the local domain of the failure site. The thermal fatigue predictions for the planar solder interconnection layer in a conventional wire-bonded, Si-based Insulated-Gate Bipolar Transistor (IGBT) power electronic module (PEM) are demonstrated using the proposed methodology under parameterised thermal cycling load. The results show that the metamodel with location-dependent model parameters can retain the accuracy of damage predictions obtained with the full-order FE simulations and can accurately inform on the damage spatial distribution in the solder layer. The metamodel is found to have superior performance compared to a unified Neural Network model with the same spatial damage prediction attribute.

Item Type: Conference Proceedings
Title of Proceedings: 47th IEEE International Spring Seminar On Electronics Technology. 15th - 19th May 2024, Prague, Czech Republic
Uncontrolled Keywords: power components; IGBT; reliability; solder interconnects; metamodels; thermal fatigue; damage; Machine Learning
Subjects: Q Science > Q Science (General)
Q Science > QA Mathematics > QA75 Electronic computers. Computer science
T Technology > T Technology (General)
Faculty / School / Research Centre / Research Group: Faculty of Engineering & Science
Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified: 22 May 2024 14:37

Actions (login required)

View Item View Item


Downloads per month over past year

View more statistics