Andreas Linnemann

In the past years, the finite element analysis (FEA) has proven to be a suitable way for fatigue prediction of electronic equipment based on the physics-of-failure-approach. For this, inelastic strain parameters like creep strain or creep energy density are evaluated in crack susceptible regions of solder joints. Due to the nonlinearity of the creep behavior, which is the basis for these simulations, the computational effort can be significant. This mostly leads to a component-focused approach. Global influences on components like local stiffness variations due to adjacent components, copper traces, or fixations of the printed circuit board (PCB) are often ignored. To make creep-based fatigue predictions suitable for complex printed circuit board assemblies (PCBA), a method for reducing computational effort needs to be established. For this matter, a machine learning-based approach for solder joints has been developed. First, the process for data generation and model training has been established. Thereafter, several methods for input parameter reduction are discussed. Finally, a model is being trained based on the generated simulation data.