UNCERTAINTY QUANTIFICATION FOR THE SIMULATION OF AUTOMOTIVE COMPONENT LOADS USING STOCHASTIC SURROGATE MODELS

To design reliable automotive components, it is important to understand the expected
operational loads. Early in development, virtual load generation helps derive these loads realistically. Many loads depend on vehicle speeds, making velocity simulation an essential part
of virtual load generation. The velocity profiles are determined using a stochastic simulator,
which employs a stochastic process with variable parameters to model the variation in driving behavior and generate random velocity profiles for a given route and driver. The velocity
profiles are propagated to component loads using use case specific system simulations. The
derived velocities and loads are subject to model and data uncertainty. To address this, we propose a framework that models input parameters as random variables in a Bayesian framework,
allowing for the handling of uncertainties. To enable the identification of these uncertainties
and sampling from the posterior distribution, we apply Polynomial Chaos Expansion and Generalized Lambda Models to derive stochastic surrogate models as a substitute for the velocity
simulation. We show that the parameter distributions derived with the surrogate provide valid
... mehr
descriptions of the velocity simulation uncertainty and validate the results with respect to measured velocities. As a result, our study yields a broad probabilistic framework for data-driven
reliability prediction of automotive components and enables the consideration of load estimation uncertainties in an early stage of component development.