A messy genetic algorithm and its application to an approximate optimization of an occupant safety system

Abstract The paper explores an adaptation of a messy genetic algorithm (mGA) and back-propagation neural network (BPN) in the context of a global approximate optimization of an occupant safety system in automotive design. The design objective is to determine the airbag vent hole size, seatbelt load limiter, steering column collapse load, and airbag tether length by minimizing both the head injury criterion (HIC) and thorax injury criterion (centre-of-gravity criterion (CGC)) for passenger protection. The analysis of the occupant safety system is highly non-linear, thus an mGA is adopted as a global optimizer in the present study. For global function approximations, BPN models are established to represent output responses of the HIC and CGC. A design of experiments is conducted to examine the interaction effect between the design variables and to generate training data for use in the BPN. The paper develops new schemes of adaptive random seed selection and adaptive operation probability to enhance the mGA performance. A number of benchmarking problems are first explored to support proposed mGA strategies, and approximate optimization is performed for the design of occupant safety system.