Spindle Thermal Error Prediction Based on Back Propagation Neural Network and Hybrid Taguchi Genetic Algorithm for a Computer Numerical Control Machine Tool

In this study, a search was made for the optimal training parameters for a back propagation neural network (BPNN) using the hybrid Taguchi genetic algorithm (HTGA). This was conducted to enhance the predictive accuracy of the model and solve the difficult problem of BPNN parameter adjustment. The 10-fold cross-validation method was used for verification and to assess the pros and cons of the model as well as optimize the training parameters. Practical spindle thermal deformation experiments were also conducted to verify the prediction results using a computer numerical control milling machine at different spindle speeds using contact thermal sensors and an eddy current sensor to measure deformation. The findings of this research demonstrate that the training parameters for the BPNN, when optimized using the HTGA, exhibit superior performance compared with those obtained through the conventional genetic algorithm methodology. The results of the experiment in thermal deformation and displacement indicate that the root-mean-square error of the predicted displacement and the actual displacement for the optimized BPNN training parameter model using HTGA were within 6 mu m, and the results were better than those found by conventional methods.