Detailed numerical simulation models support engineers during product development but they are often high-dimensional and computationally intensive. Detailed sensitivity studies and iterative optimizations thus rapidly exceed available computing capacities. Highest algorithmic efficiency is thus a key factor in virtual process and component optimization.

This is where mathematical-algorithmic competencies and engineering expertise come together. The KIT-Centre MathSEE offers an internal communication and cooperation platform for interdisciplinary topics between mathematics and applied science. In particular, the FAST-LBT works closely with the Department of Continuous Optimization of the Institute of Operations Research (IOR-KOP) from the field of mathematics.

The cooperation project between FAST-LBT and IOR-KOP investigates machine learning models for support engineering optimisation, so called „surrogates“. These surrogates can replace complex simulations during optimization and thus reduce the overall computation time. The goal is to create a formal understanding of surrogate-supported methods for common engieering optimization problems. Therein, process steps in composite manufacturing offer challenging use cases with highly variable process dynamics.

Fig. 2: Comparison of direct and ML-assisted optimization at the example of evolutionary algorithms; a significant speed-up of convergence is observed.Image source: KIT-FAST