QUANTIFYING THE EPISTEMIC UNCERTAINTY LINKED TO THE DEVELOPMENT PROCESS OF AN EARLY-STAGE COMPONENT FOR CRASHWORTHINESS

We propose in this work a new method to quantify the epistemic uncertainty linked to the early-stages of the development process of non-linear dynamic systems. It is here applied to the development of the crashworthiness of the front of a new car. More specifically, we investigate the effects on a component of the parts around it. The uncertainty derives from the many unknowns that drive the design of a component to change. Nevertheless, it is quantifiable thanks to the solution space methodology. It is then propagated through a simplified Finite Elements environment that represents the front of a newly developed car. A model quantifies the uncertainty by exploiting the force ranges computed in the solution space methodology. However, the model inherits a high number of variables, making the propagation of the uncertainty problematic. To solve the dimensionality issue we propose a new uncertainty model that couples the variables together. Experiments on early-stage designs of components are used to express the dependencies. They are designed to translate the corridors of solution space into new force ranges that capture the crash behavior of a component. Therefore, from an initial number of possibly 29 independent variables, we end up with only 6; one per component making up the crash sub-assembly of the front of the car. The proposed method yields results similar to those yielded by the model with 29 variables. This shows that we effectively solved the dimensionality problem without losing information. Our method estimates the effects on a component of its neighboring systems. This allows a developer to quickly identify in the early-stage phases of development the designs that would not work well in the final product. Thus, the method reduces the chances of costly re-design in a late-stage phase of the development.