Optimization of Five-axis Tool Grinder Structure Based on BP Neural Network and Genetic Algorithm

Abstract

Abstract An optimization design was carried out based on a back propagation (BP) neural network and a genetic algorithm (GA) to improve the stiffness and accuracy of the self-developed MGK6030 five-axis tool grinding machine. First, finite element analysis was carried out on the whole grinding machine based on ANSYS Workbench, and the key parts were found to be the grinding wheel headstock, B axle box body, and column. Sensitivity analysis was carried out after the model parameterization, and 10 parameters, which affect the quality, maximum deformation, and first-order mode, were obtained. These parameters were used as input variables. A total of 235 sets of sample data were obtained by using the optimal overall performance of the grinder for the target (large first-order natural frequency, small deformation, and mass). The BP neural network was then used to fit the nonlinear coupling relationship between the input and the output. Thereafter, the optimization function of the GA was used to perform multi-objective optimization in the specified range. Finally, the parameters are verified by software simulation and prototype test. Results showed that the maximum deformation of the optimized machine tool is reduced by 21%, and the first four order natural frequencies are increased by 6.36%, 9%, 6.4%, and 2.84%. The maximum positioning accuracies of the linear axis and rotary axis are increased by 22% and 21%, respectively, which demonstrates the effectiveness of the optimization scheme and provides theoretical and technical support for similar optimization problems.