A vehicle crash simulation can be divided into three parts: compression of low-stiffness pedestrian protection structures, compression of high-stiffness longitudinal members, and the rebound phase after reversal of motion. To represent these, the spring-mass-damper model uses an exponential spring for the pedestrian protection and a linear high-stiffness spring. The weighted mass displacement represents the vehicle’s center of gravity motion. The model stiffnesses are identified from many FE crash simulations so that, for each combination of input parameters (impact angle, impact velocity, offset), a set of stiffness parameters is computed. A neural network is then trained to predict stiffness given inputs.