Machine learning-based prediction of remaining useful lifetime for solder joints based on real mission profile data

Abstract In an increasingly digitalized world, electronic systems are central to the operation of essential services. Ensuring the reliability of these systems is crucial, particularly in safety-critical applications where failures can have serious consequences. On circuit boards, solder joints are one of the key interconnection components whose degradation over time poses a significant challenge for system longevity. To address this, the Prognostic and Health Monitoring (PHM) domain has emerged as a promising solution framework to enable early detection of potential failures and support condition-based maintenance strategies. In this article, we present a methodology to estimate remaining useful lifetime (RUL) of solder joints under field operation. A light rail vehicle (tram) was equipped with sensors to measure mission profile loads (mainly temperature data) near the critical electronic components. In parallel, a finite element-based digital twin model of a solder joint was developed to generate a training data set. Then, a multilayer perceptron (MLP) was used to correlate the temperature characteristics with fatigue evolution in the solder joint. The MLP was able to predict fatigue increments with good accuracy which enables the estimation of further lifetime consumption. The major challenge with insufficient loading data, especially operational mission profile data, was addressed by supplementary synthetic data. Finally, the lifetime modeling method is presented in this article. Overall, this work provides a solution to estimate RUL of solder joints under mission profile loads, suitable for a portable inline PHM framework.