A Data-Based Modeling Approach for the Prediction of Front Impact (NCAP) Safety Performance of a Passenger Vehicle

Designing a vehicle for superior crash safety performance in consumer rating tests such as US-NCAP is a compelling target in the design of passenger vehicles. In today’s context, there is also a high emphasis on making a vehicle as lightweight as possible. In modern vehicle design, these objectives can only be achieved through simulation techniques often referred to as Computer-Aided Engineering (CAE). A pre-requisite for a detailed virtual crash safety assessment of a vehicle using a CAE tool commonly based on nonlinear explicit-dynamic finite element analysis is a detailed CAD (Computer-Aided Design) model of the vehicle. However, in the absence of detailed information on geometry in the early phase of design of a vehicle, especially based on a new platform, traditional CAE evaluation becomes infeasible. On the other hand, for design to progress after finalization of a styling concept, informed guidance on mechanical packaging and structural design of key body members influencing crash safety performance of the proposed vehicle is necessary. An intuitive design process primarily based on competitive vehicle benchmarking may not be a sufficiently reliable design process due to the complexity of the engineering phenomena involved. A data-driven design approach based on artificial intelligence and provision for machine learning can be of great assistance in addressing the requirement of achieving a robust initial design which can be subsequently subjected to detailed time-consuming modeling and analysis for demonstrating the desired crash safety performance with minimum modifications to the original design. Such techniques are still evolving and are not still part of the standard portfolio of design analysis tools. In order to better accomplish the objective of guiding the body design of a vehicle at its nascent stage, a performance predictor based on a Bayesian Regularization Backpropagation ANN (Artificial Neural Network) is adopted. The training database is generated by taking into account the performances of an assorted heterogeneous mix of real-world vehicles obtained with the aid of detailed CAE models of those vehicles in the US-NCAP crash testing mode. As an enhancement of past efforts in this area, design variables consisting of front packaging and geometric parameters of dominant front structural members such as front bumper, lower rails and shotguns (i.e. upper rails) are considered with variations in cross-sectional size of the lower rails being also accounted for. Performance metrics in terms of crash safety and efficiency of design are tied to the attributes of the NCAP deceleration pulse (i.e. “crash pulse”). The approach is shown to predict well the performance of a design variation that is not part of the training database.