Many engineering products are subject to uncertainties in material properties and geometric dimensions, e.g., due to manufacturing. Even small deviations can significantly influence the performance, efficiency and lifetime of electric machines. These uncertainties can be incorporated into numerical simulations and quantified. This process increases the computational effort by a factor of thousands. Many different configurations have to be analyzed.
Incorporating uncertainties comes with additional computational costs. Therefore it is often only feasible in a high-performance-computing environment. In such an environment, the evaluation of configurations can be parallelized over different computing nodes. A straight-forward approach is Monte-Carlo sampling, which is trivially parallel and thus well-suited for such environments.
TU Darmstadt and KU Leuven develop a new hybrid method within a Inno4Scale project. The method combines multi-level Monte-Carlo sampling with parallel-in-time integration. It promises to reduce both energy usage and time-to-solution of Uncertainty Quantification in a high-performance-computing environment.
Image source: https://parallel-in-time.org