In the last few years, the demand for general-purpose Finite Element (FE) vehicle models with fine mesh and small elements has increased the size of these models dramatically. The FE simulation of these models requires extensive CPU time, which makes the simulation cost an important issue to consider. The main objective of this research is to develop an accurate and computationally inexpensive method to predict a vehicle’s crash performance in the event of a collision. This becomes very important as the demand for performing several impact scenarios for each vehicle becomes excessive. This demand is driven by the desire to investigate different impact scenarios and to study the effect of the impact velocity, the offset-barrier ratio, and the impact angle on the dynamic behavior of the vehicle structure in crash events.
In the last decade, Artificial Neural Networks (ANN) emerged as a reliable tool for solving nonlinear problems in variety of applications. The most important feature in ANNs is its ability to infer the nonlinear characteristics of any complex system, even if the mathematical model of the system does not exist. This is an extremely important feature when dealing with highly nonlinear dynamic problems such as vehicles collision. In a previous research conducted by the authors, advanced ANNs were developed and trained to model vehicles frontal impacts. This paper extends the concept and technique in order to use ANNs in modeling vehicles offset-barrier impacts.
Special ANNs were developed, trained and tested through numerical examples for two different offset impact cases. The first case was 50% offset-barrier impact at five different impact velocities, while the second case was 35 mph frontal impact at five different offset-barrier ratios. Validated FE vehicle model was used to perform FE simulations for many different offset-barrier impacts. The crash profiles obtained from the FE simulations were used to train and test the developed ANNs. The results of these numerical examples indicated the ability of the ANNs to accurately capture the nonlinear dynamic characteristics of the vehicle structure for offset impacts. The trained networks could then be used to predict the crash profiles of any offset impact scenario within the training range.