Rapid model generation and analysis of mechanical behaviour of electronic packaging structures by machine learning

For industrial applications of finite element (FE) simulations, model generations for complicated electronic packaging structures are challenging because of limited time engagement and personnel resources. To achieve accurate predictions of mechanical behaviour of electronic packaging structures under various thermal and mechanical loadings, complicated constitutive models and fine meshes in FE simulations are also anticipated, which requires much knowledge and experience for the personnel performing numerical analysis. Therefore, this paper proposes a simple prototype of the developed plug-in for industrial applications to realise the rapid model generation and analysis of mechanical behaviour of electronic packaging structures. Firstly, the model generation of board-level packaging structures is efficiently completed in a parameterized manner using Python to export a ready-to-run numerical model with boundary and loading conditions in the commercial finite element software ABAQUS. At the same time, high-quality coarse meshes are discretized for the FE models with thermo-elastic materials. This makes it possible to save great labour cost on the FE pre- and post-processes for the generation and analysis of large and complex models of packaging structures. Furthermore, the submodelling technique is utilized to allow to have rough estimations for the mechanical behaviour of board-level packaging structures such as displacement, strain and stress and subsequently transfer the nodal stress and strain as the boundary conditions around those solder joints of interest. Based on the obtained data from FE analysis, machine learning is promising to provide rapid solutions to make predictions of complex problems provided a sufficient database is available for the solutions.