Data-driven surrogate model establishment and performance optimization for cab and suspension systems

A surrogate model establishment and vibration performance optimization method is proposed for multi-body dynamics models of cab and suspension systems. First, the 6-degree-of-freedom excitation inputs of the system were obtained through virtual load iteration, and key design variables were screened using an optimal Latin hypercube design. Next, the influence of hyperparameters on fitting accuracy was systematically investigated for response surface model (RSM), Gaussian process model (GPM), and neural network model (NNM). The results validated the significant advantage of neural networks in capturing nonlinear relationships, and a high-fidelity surrogate model was established with a fidelity of 96.4%. Through multi-objective optimization, the comprehensive performance index was improved by 19.4%, while the optimization computation time was reduced to 1/365.6 of the original. This study yielded several key engineering insights: Reducing front suspension spring stiffness, increasing front damper damping, and enhancing rear spring stiffness can significantly improve vertical/roll/pitch posture performance, whereas bushing stiffness parameters exhibit limited optimization potential. Monte Carlo robustness analysis identified critical variables contributing to failure probability, guiding a fine adjustment of the shock absorber stroke to meet all constraints. Furthermore, a notable theoretical finding emerged: increasing the number of hidden layers or neurons does not necessarily enhance model fidelity and must be carefully optimized to avoid overfitting. Notably, RSM with linear terms, GPM with linear kernel functions, and NNM with linear activation functions are mathematically equivalent. This research provides a high-precision and low-cost solution for multidisciplinary optimization of complex suspension systems, demonstrating significant engineering application value.