Computer-aided engineering (CAE) is a routinely used technology for the design and testing of road vehicles, including the simulation of their response to an impact. To increase automotive industry competitiveness by reducing physical test-based type approval and to improve road safety, recent initiatives have been taken by both industry and public authorities to promote the use of virtual testing through numerical simulation as an alternative way to check regulatory compliance. [1]
To ensure acceptance of this alternative method, the accuracy of the simulation models and procedures needs to be assured and rated independently of the modelling process, software tools, and computing platform. Similarly, it is also imperative to understand the uncertainties emerging out of different component design parameters and analyze their sensitivity towards producing deviations in the reported results as per the requirements of the regulatory standard. Simulations are however deterministic in nature and do not in general consider the uncertainty in areas such as design, manufacturing, and use of the product. Using machine learning and uncertainty quantification, engineers can predict the range of possible outcomes for a given design by accounting for such uncertainties. Training emulators (aka machine learning models, predictive model, etc.) of simulations can allow for the efficient performance of the advanced analytics required.
Design parameters of a front under-run protection device (FUPD) were evaluated using computer simulation as per Automotive Indian Standard (AIS) 069 loading conditions [2]. The process involved the use of Design of Experiments (DOEs) to collect emulator training data and uses the trained emulator to perform uncertainty propagation of manufacturer part tolerances. A sensitivity analysis study was carried out to understand the dependence of output results on crucial parameters.
The results obtained from this study lend benefit towards understanding the uncertainties involved in CAE simulation-based product assessments. It will also be useful as a benchmark to quantify uncertainties involved in other load cases wherein the regulatory standard allows the use of virtual testing as an alternative method to physical type approval.