Research On Thermal Error Modeling Technology Of CNC Machine Tools Based On Deep Learning

High-grade CNC machine tools are important processing equipment in the fields of intelligent equipment manufacturing and national defense and military equipment,and processing accuracy and consistency are extremely important to ensure product quality.The error factors that affect the machining accuracy of CNC machine tools include geometric error,control error,motion error,thermal error,force error,position(positioning)error,and machining error.The thermal error is the error term that accounts for the largest proportion of the error sources.The higher of the precision of machine tools or precision machining,the higher the proportion of thermal errors.After nearly 40 years of research,domestic and foreign scholars have proposed a variety of thermal error modeling theories and compensation methods.However,due to the dynamic variability of the thermal error of CNC machine tools,the use range of the model is narrow and the robustness is not high enough.This dissertation is supported by the National Natural Science Foundation of China(51775074),Chongqing Basic Research and Frontier Exploration Project(cstc2018jcyj AX0352)and Chongqing Postgraduate Research and Innovation Project(CYS19316),combined with existing modeling and compensation theoretical foundations,the temperature field transformation and high generalization thermal error models of CNC machine tools are studied.The main contributions of this dissertation are shown as follows:1)Taking the thermal error of CNC machine tools as the research object,the structural characteristics and heat source distribution of CNC machine tools were studied and analyzed.The thermal characteristics detection system of CNC machine tools was established with USB3120 data acquisition board and Lab VIEW software development platform.The measurement errors that may occur in the measurement system were analyzed,and the data processing of the measurement data was researched.The moving average filtering method was used to reduce the noise of the sensor data,and the data was normalized and enhanced to make a data set.2)According to the gray relation and K-means comprehensive analysis method,the temperature sensitive point changes of the CNC machine tools were analyzed,and the method of transforming the temperature field of CNC machine tools was introduced.Based on the deep learning theory,the thermal error model of the convolutional neural network was established,and the training method and optimization method of the network were deduced.Based on the machine learning theory,an optimized fuzzy neural network thermal error model was established.The goodness of fit,prediction accuracy and generalization index of the thermal error model were established to evaluate the pros and cons of the model.3)Aimed at the problem that the stability of the existing thermal error model shows large differences under different working conditions,a spatiotemporal convolution thermal error model based on the attention mechanism was proposed.Using the spatial feature extraction capabilities of the convolutional neural network and the temporal feature extraction capabilities of the long-short-term memory neural network,a thermal error model with two branches was established.After extracting the high-dimensional features of the original data,the attention mechanism was used to reconstruction the features by importance.Prediction experiments on different data sets show that the proposed model has high generalization.4)Embedding thermal error compensation control system based on deep learning was developed by using the integrated development environment of STM32 Cube IDE.The thermal error compensation PC software were developed by using MATLAB and C#.Compensation experiments on CNC grinders and CNC lathes verify the effectiveness of the model and the system,and the thermal error values after compensation in the idling state were all within 5?m.