Multiobjective optimization algorithm for accurate MADYMO reconstruction of vehicle-pedestrian accidents

Abstract

In vehicle–pedestrian accidents, the preimpact conditions of pedestrians and vehicles are frequently uncertain. The incident data for a crash, such as vehicle deformation, injury of the victim, distance of initial position and rest position of accident participants, are useful for verification in MAthematical DYnamic MOdels (MADYMO) simulations. The purpose of this study is to explore the use of an improved optimization algorithm combined with MADYMO multibody simulations and crash data to conduct accurate reconstructions of vehicle–pedestrian accidents. The objective function of the optimization problem was defined as the Euclidean distance between the known vehicle, human and ground contact points, and multiobjective optimization algorithms were employed to obtain the local minima of the objective function. Three common multiobjective optimization algorithms—nondominated sorting genetic algorithm-II (NSGA-II), neighbourhood cultivation genetic algorithm (NCGA), and multiobjective particle swarm optimization (MOPSO)—were compared. The effect of the number of objective functions, the choice of different objective functions and the optimal number of iterations were also considered. The final reconstructed results were compared with the process of a real accident. Based on the results of the reconstruction of a real-world accident, the present study indicated that NSGA-II had better convergence and generated more noninferior solutions and better final solutions than NCGA and MOPSO. In addition, when all vehicle-pedestrian-ground contacts were considered, the results showed a better match in terms of kinematic response. NSGA-II converged within 100 generations. This study indicated that multibody simulations coupled with optimization algorithms can be used to accurately reconstruct vehicle-pedestrian collisions.