Research On Thermal Error Modeling And Optimization For Heavy Duty CNC Machine Tools

Thermal error is one of the main factors affecting the machining accuracy of heavy CNC machines.It is the key to achieve thermal error compensation and effectively improve machining accuracy by establishing a thermal error model based on temperature field monitoring data.However,the traditional modeling method needs to select different key temperature measurement points according to different working conditions to improve the prediction accuracy and generalization ability,and avoid overfitting phenomenon.Therefore,it is difficult to obtain a thermal error model with good generalization ability for different working conditions.In addition,due to the influence of the processing environment and internal factors of the machine tool,the historical model can not adapt to the distribution of new data,and needs to be dynamically adjusted with the addition of new data.Therefore,researching the thermal error model that can take into account the prediction accuracy and generalization ability under multiple working conditions,and improving the description ability of the existing model according to the new data is of great significance for ensuring the reliability of heavy-duty CNC machine tool.In this paper,the heavy-duty CNC machine tool ZK5540 A is taken as the object.A thermal error model of heavy-duty CNC machine tools based on real-time monitoring data of fiber bragg grating is established and a thermal error model optimization algorithm is proposed based on incremental learning.The main research contents are as follows:(1)According to the complex characteristics of heavy-duty CNC machine tools,this paper studies the structural characteristics of the machine tool and the distribution of heat sources,determines the data measurement method and the arrangement of fiber grating sensors and displacement sensors,performs data acquisition test on temperature and spindle thermal error,and analyzes the relationship between temperature of the heat source point and thermal error.Pre-processing and enhancement of temperature and thermal error data lays the data foundation for subsequent modeling.(2)In view of the problem that the traditional thermal error modeling methods are weak in generalization under multiple conditions,an improved thermal error model based on convolution neural network is proposed.In order to solve the problem that the hyper parameters of convolution neural network are based on empirical design,an adaptive convolution neural network based on mutual information is constructed.Analyze the mutual information between the intermediate feature layer and the target vector,and selecte the feature map with large mutual information,so as to optimize the number of convolution kernels and determine the structure of the convolution neural network.In order to improve the feature selection ability of the model under multiple conditions,avoid the influence of redundant features,analyze the contribution of neural network features,and combine the traditional feature evaluation methods into the training process,and propose a feature enhancement model of convolution neural network.(3)Since the convolution neural network unable to self-learn new samples after constructing models based on complete data sets,it needs to retrain the model and change network parameters,which will result in huge time consumption and loss of existing knowledge.In view of this problem,a model optimization algorithm based on incremental learning is proposed.The algorithm can adjust the local parameters according to the new sample,update the convolution neural network model based on incremental learning and branch extension method,and realize the dynamic adjustment of the model.