Many-objective optimization design method for rear seat backrest frame of a passenger car based on hierarchical optimization and adaptive radial basis function neural network

Addressing the “dimensional catastrophe” in optimal design while ensuring a balance between optimization efficiency and time cost has become a central focus for researchers. This paper proposes an efficient design strategy for automotive seatback optimization based on the concepts of adaptive approximation-optimization and many-objective hierarchical optimization. To begin with, the finite element analysis models of the rear seat of a passenger car are developed and validated using experimental results. Considering factors such as total cost, seat material quality, and safety performance index, the seat structure parameters are categorized into primary and secondary sensitivity parameters through comprehensive parameter sensitivity analysis. Then, an advanced optimization strategy was developed, integrating optimal Latin hypercube experimental design, adaptive radial basis function neural network approximation model, non-dominated sorting genetic algorithm-III, modified entropy weight method, and Euclidean distance method to enhance grey relational analysis techniques. The strategy was subsequently applied to optimize an automotive seat skeleton. When compared with a non-hierarchical many-objective optimization strategy, the proposed approach demonstrated significant advantages in optimization search performance, time cost, and iterative convergence speed. Finally, the optimized automotive rear seat achieves a 6.78% reduction in weight and a 19.3% reduction in cost of the backrest skeleton, while also enhancing comfort and occupant protection to varying degrees.