Solder joint reliability predictions using physics-informed machine learning
S. D.M. de Jong (Student TU Delft)
Amir Ghorbani Ghezeljehmeidan (TU Delft - Electronic Components, Technology and Materials)
Willem D. van Driel (Signify, TU Delft - Electronic Components, Technology and Materials)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
The reliability of solder joints plays an increasingly important role in power electronics. The thermal fatigue experienced due to the temperature fluctuations cause catastrophic failures. However, the ability to predict the fatigue for different thermal cycles is lacking. Experimental or simulation based approaches are typically too expensive to be conducted for a wide range of thermal loading conditions. A physics informed Long Short-Term Memory (PI-LSTM) is proposed here for predicting the plastic strain and related fatigue lifetime in solder joints. The LSTM model is trained on data generated by FEM simulations, enhanced by incorporating the flow rule into the loss function. The PI-LSTM accurately predicts the plastic strain and the stress components, enabling efficient reliability predictions. Using different reliability models, the estimated cycles to failure are found to be in close agreement with those from conventional FEM simulations, demonstrating the PI-LSTM's capability for reliability assessments.
help_outline