Abstract
In an industrial group such as Safran, numerical simulation of physical phenomena plays a central role in the design processes of manufactured products. At Safran Tech, the corporate research center of the Safran Group, we develop technologies to improve these processes by constructing fast and reliable surrogate models for a wide range of physical systems. This presentation introduces several technologies developed in recent years. First, we present a physics-based reduced-order modeling approach for nonlinear structural mechanics, applied to lifetime prediction of high-pressure turbine blades. Second, we address the learning of physics simulations under non-parameterized geometrical variability using classical machine learning techniques combined with nonlinear deterministic dimensionality reduction, including morphing, principal component analysis, and Gaussian process regression. We then illustrate the generation of mechanical components constrained by performance requirements. Finally, we highlight our contributions to the open-source and open-data Scientific Machine Learning community.
