Head injury criteria is evaluated considering different equations based on the acceleration of the impact head. The main objective it to decelerate the head from an initial impact velocity with a low deceleration. This implies the design of deformation elements in both exterior of the car for pedestrian impact and interior of the car for regulations such as ECE-R 21 or FMVSS201u. Such deformation elements should produce an optimal deceleration of the impact head. To use the minimum space for deformation, this deceleration should be as close as possible to a constant value. If we design too soft elements the injury values increase as the headform hits the stiff car structure. If we design too hard elements deceleration increases and this also means injury number increases. In real automotive parts there is a statistical distribution of thickness and material properties that are known by quality inspection. In this paper we present a design approach to decide the number of simulations required to be carried out to obtain an estimation of percentage of injury numbers which do not meet the requirements. Iterations in design are evaluated and discussed with particular attention to the point where engineers ask for more space for deformation. Discussion on stability of injury values is given as for example from 400 simulations only 20 injury values do not meet the requirements. This means 95% meet the requirement, but we cannot accept a 5% possibility of failure. The major discussion is to convince the engineering team of this possibility of failure when we have just a prototype test with a good injury value. This discussion is focused on the search of designs with a lower standard deviation to ensure all values meet the requirements. The use of machine learning algorithms, particularly evolutionary algorithms is explored in this approach to obtain the best design with the minimum deformation space. These algorithms are well-suited for optimization problems. Simulations of head impact are performed using Python scripts to feed the ESI Crash input deck model from real statistical values. These simulations provide as outcomes injury values with real statistical distributions which are far from normal bell shape distribution. The best design with the minimum required space to obtain a 0% failure is obtained with this approach.
Prof. Dr. Andres-Amador Garcia-Granada, Head of Department Industrial Engineering, IQS-URL