| With the continuous growth in computer technique, the research and design work ofcomplex mechanical products are often done by using CAD or CAE tools. As finite elementtechnique is being completed, using FEA to analyze the properties of products can reduce thetime cost of the real physical experiment to a certain extent. However, due to the complexproduct structural design, constructing the finite element models can be time costly, and themeshing can be complicated, which are still obstacles for designers. Using the proper DOEmethod and approximation model, can guarantee the accuracy of prediction, reduce the timesof experiments and cut down the compute costs.Aiming at solving the problem of structure optimization design for complex mechanicalproducts can be handled in two aspects: design of experiments method and approximationmodeling, in order to construct the solving model, and briefly analyze the experiment designmethods and approximation modeling techniques. According to the demand of space-fillingexperiment design method, considering both the space-filing property and the correlationbetween different dimensions, we construct optimal LHDs by using the uniform designmethod, taking the orthogonal design and maximin deisgn as a reference, and by means ofILS to improve the quality of sampling points. Analyze the effectiveness of the new designcomparing with the Hammersley sequence sampling method. GSA, GA, PSO and usingimproved LHD to initiate PSO methods to search for the kernel function parameter2andthe penalty factor C, in order to improve the accuracy of approximation models with the samesampling points in SVR. Combine with the engineering project of28meters vertical millinglathe supporting plates as an example, using the methods mentioned in this thesis to optimizethe design of the structure to evalutate its efficiency so that the designer can take it as areference. |
