基于最佳代理模型的液力变矩器多目标优化设计

Multi-Objective Optimization Design of a Torque Converter Based on Optimal Surrogate Model

摘要: 为了缩短优化设计周期，应用最佳代理模型结合基于非支配排序遗传算法II (NSGA-II)的多目标优化方法进行液力变矩器优化设计。建立了液力变矩器参数化流道模型，应用灵敏度分析方法筛选出对液力性能影响较大的关键设计参数用于优化设计变量。分别对比分析了多项式响应面法、径向基神经网络法、广义回归神经网络法、克里金法和支持向量回归法对液力性能的预测精度，结果表明，对于本文设计优化问题，支持向量回归法具有最佳的预测精度。基于最佳预测代理模型对液力变矩器动力性、经济性和与发动机匹配性进行多目标优化，选择一组合适的优化设计参数进行样件制造并进行试验验证。结果表明，本文方法可以为液力变矩器的优化设计提供有效的解决方案。

Abstract: To reduce the total design and optimization time, the optimal surrogate model coupled with a multi-objective optimization method based on non-dominated sorting genetic algorithm II (NSGA-II) is used to design optimum impellers for an automotive torque converter. A parameterized flow channel model for the torque converter is established, and sensitivity analysis is applied to select key design parameters which have a significant impact on hydraulic performance for optimizing design variables. We compared and analyzed the prediction accuracy of hydraulic performance using polynomial response surface method, radial basis function neural network method, generalized regression neural network method, kriging method, and support vector regression method. The results showed that for the optimization problem designed in this paper, support vector regression method has the best prediction accuracy. Based on the optimal surrogate model, multi-objective optimization is carried out on the power performance, economy, and matching performance with engine. A suitable group of optimization design parameters is selected for sample manufacturing and experimental verification. Results demonstrate that the presented method can provide an effective solution for the optimization design of torque converters.