Research On Technologies Of Tool Wear Prediction And Process Parameter Optimization Based On Data-driven

Intelligent machine tools improve CNC(Computerized Numerical Control,CNC)machine tools by applying the new generation of information technology and realize intelligent functions such as self-perception,self-learning and self-decision in the cutting process,which are of great significance for improving the modernization level of our country’s equipment manufacturing system.Tool health operation and maintenance and machining performance optimization are the key technologies to realize the intelligent cutting process,the cores of which are to carry out the research of tool condition monitoring and process parameter optimization technologies.Based on the data-driven method,this thesis carried out in-depth and systematic research on the above intelligent key technologies from the aspects of multi-sensor fusion technology,monitoring sensor signal type,deep learning theory and multi-objective optimization method and actively explored to improve the intelligent level of CNC machine tools by developing intelligent application algorithms.The main research work done in this thesis was summarized as follows:In view of the problem of accurate prediction of tool wear,this thesis proposed a novel tool wear prediction model based on multi-sensor feature fusion and the Bayesian optimized stacked denoising autoencoder.The feature quantities of time domain,frequency domain and wavelet domain were extracted from the three commonly used monitoring signals of cutting force,vibration and acoustic emission.Correlation analysis was performed to select the combination of features sensitive to tool wear changes,which were input into stacked denoising autoencoder with Bayesian optimized model hyperparameters for deep fusion feature learning to establish the tool wear prediction model.The comparative validation analysis of different prediction models showed that the proposed model had better the accuracy and stability for tool wear prediction.In view of the problem that the tool tip temperature signal is rarely used in tool condition monitoring research,this thesis proposed a new tool wear prediction model based on tool cutting area temperature monitoring signal and stacked sparse autoencoder.Through the wear experiments on tools with the thin-film thermocouple embedded in the tool tip area,the temperature signals of the cutting area and the corresponding tool wear values in the tool life cycle were obtained.The stacked sparse autoencoder took the temperature raw signals as input for deep feature learning to establish the tool wear prediction model.The comparative validation analysis of different prediction models showed that the proposed model can achieve accurate prediction of tool wear and outperformed other comparative models.In order to further improve the accuracy of tool wear prediction based on tool cutting area temperature monitoring signal,a tool wear prediction model based on intergrated stacked sparse autoencoder was proposed.By selecting stacked sparse autoencoders with different activation function configurations,the tool wear prediction models with different prediction performances were established.The ensemble strategy based on the stacking learning method was utilized to fuse and integrate different stacked sparse autoencoder models and the tool wear prediction model with higher prediction accuracy was established to reduce prediction error by an average of 14.01%.In view of the point that tool wear is an important optimization variable in cutting processing,in order to further improve processing performance,a multi-objective particle swarm optimized neural network system was proposed for multi-objective optimization research of process parameters.Considering the material removal stage in the manufacturing process of precision parts to maximize material removal rate and minimize tool wear and the surface forming stage to minimize tool vibration intensity and obtain optimal surface quality as optimization variables,orthogonal experiments were designed to obtain the experimental results of the corresponding cutting-related variables under different cutting conditions.The proposed optimization system established the mapping model of cutting parameters and cutting-related variables through artificial neural network and the improved multi-objective particle swarm optimization algorithm was applied to solve multi-objective optimization problems to obtain the Pareto solution set of process parameters.In addition,the optimal combination of process parameters for each processing stage was determined by maximum deviation theory.In conclusion,this thesis systematically summarized the main research content of the full text and looked forward to the next research trend and future work focus of key technologies in intelligent machine tools.