DW-RBF Method Based Structrue Parameters Optimization Of EMU Axle-Box

The high-speed train axlebox is one of the key bearing devices under the primary suspension system. Reducing its quality helps improve the dynamic performance of the train. At the same time, its structure strength closely influences the safety and reliability of the train. Thus, in this paper, taking the axlebox of a certain type of EMU(Electric Multiple Units) as the research object, setting its structural performances under two different working conditions as optimization constraints, the size optimization method was used to reduce the quality of the axlebox.The conventional optimization design is based on computer simulation, resulting in low computation efficiency. To balance the efficiency and accuracy in the design optimization process, this paper adopts the approximation model based optimization strategy. The efficiency and accuracy of any approximation model depend on the following two aspects: DOE (Design of Experiments) method, and the approximation model method.First, to improve the sampling process efficiency, based on Isight multidisciplinary platform constructed box design-analysis integration process, using Isight automated sampling function obtained a set of training samples and a set of test samples.Secondly, radial basis function neural network (RBF) shows very good fitting ability when dealing with nonlinear problems. According to the anisotropy phenomena of the sample space, a dynamic-weighted Euclidean distance function based RBF (DW-RBF) was proposed in this paper. Then, this DW-RBF method was used to establish an approximation model to mimic the axlebox finite element simulation system, its approximation accuracy was compared with the one established using conventional RBF method, the comparing result shows that DW-RBF method has better performance in terms of accuracy.At last, based on the DW-RBF model, defining the optimization target function and optimization constraints functions, a complete optimization model was built. Then the Matlab optimization toolbox was used to solve the optimization model. After that, the optimization result was verified. The verification results showed that the optimization result was quite good:the optimized axlebox had been lightened for about 10%; at the same time, the optimized design could still meet the static strength design principle.The research methods in this paper for a certain type of axlebox have some general purposes, and thus could be used for reference for other similar problems.