A Multi-Objective Optimization Method For The Design Of Suspension Parameters Of A High-Speed Train

The optimization of suspension system parameters in high-speed trains aimed at the comprehensive improvement of straight-track-operation stability, curve-passing safety and wear performance is a multi-objective problem in nature and involves highly non-linear functions. Traditional, dynamic-equation-based modelling methods and single-objective or normalized multi-objective optimization techniques are no longer suitable for this problem. Surrogate modeling enables the quick statistical approximation of the relation between the dynamic performance indexes of the high-speed train and its suspension parameters, thus avoiding the repeated evaluation of highly non-linear dynamics equations. Meanwhile, Pareto multi-objective optimization algorithms, a group of genetic-algorithm-based calculation techniques such as the NSGA-Ⅱ algorithm, provide the set of solution that compromises among a group of incomparible and conflicting objective functions to be further chosen as it best fits the purpose specific to the design task. This paper proposes a new approach based on the surrogate modeling technique and the Pareto multi-objective optimization algorithm for the optimal design search for high-speed train suspension parameters.The CRH2 EMU is taken as template for the form of vehicle to be optimized. A sufficient number of suspension parameter sample points are selected using the OptLHD design-of-experiment method, and their corresponding straight-track running stability, curve passing safety and wear performance indexes are found through multibody dynamic simulation. The samples, along with their calculated dynamic responses, are then used to train an RBF-NN model as the surrogate model for the relation between suspension parameters and dynamic performance indexes. Finally, using the NSGA-Ⅱ multi-objective genetic algorithm, optimization calculations with three different objective function combinations, namely critical velocity-derail coefficient, critical velocity-wear number, and critical velocity-derail coefficient-wear number, are completed based on the surrogate model, obtaining the Pareto-optimal suspension parameter design points. Suggested values for designs of trains with different purposes are given based on these design points.The three optimization runs obtain 343,563 and 1223 Pareto-optimal design points respectively, allowing for a maximum improvement of 9.97% for critical velocity, 22.71% for wear number, and 10.03% for derailment factor compared with the template vehicle. Among them,958 design points exhibit improvements in all three objective functions, making them good designs for an overall improved version of the template vehicle. The other 1171 design points provide a gain in certain dynamic performances at the sacrifice in others, and are favorable for the design of derived versions of the template vehicle operating on different lines under different speeds.