Large-Scale \(\mathscr {H}_2\) Optimization for Thermo-Mechanical Reliability of Electronics

Optimization of transient models is required in several domains related to thermo-mechanical reliability of electronics, such as Prognostic Health Monitoring (PHM) and design optimization. A novel framework for efficient (local) parameter optimization of transient models in the \(\mathscr {H}_2\) norm is proposed. The optimization is feasible for large-scale transient models because it approximates the gradient using physics-based model order reduction (MOR), in contrast to existing approaches that typically use data-driven surrogate models such as neural networks. To demonstrate the framework an optimal fixed-order virtual sensor for PHM of a Ball Grid Array (BGA) is numerically determined.