Experimental Analysis On Thermal Error Of High-Speed Electric Spindle And Establishment Of Prediction Model

Under the background that CNC machine tools are important equipment for modern high-precision processing,the cutter end of the high-speed electric spindle,as a vital part of CNC machine tools,is in direct contact with the workpiece by its cutting tool.So,the thermal error of electric spindle will influence the machining accuracy of workpiece largely.Because of its compact structure and high-speed rotation,there will be a lot of heat generated by motorized spindle in the cutting process,and the accumulation of heat will cause serious axial thermal deformation at the tool end,which will ultimately affect the machining quality of the workpiece.Therefore,how to solve the problem of machining accuracy brought about by thermal error of the electric spindle has become the current research hotspot.The thermal error compensation method is a common method to improve the machining performance of electric spindles.The method needs to establish a mathematical model of electric spindle temperature and thermal error,and then calculate the thermal error value through the temperature information obtained during the actual machining process to output the thermal error value(compensation value),and finally send the compensation value to the electric spindle control device to adjust its coordinate position so as to compensate and offset the electric spindle thermal error.This paper designs and conducts thermal error tests for the actual machining conditions of a company's A02 electric spindle,and proposes a high-speed electric spindle thermal error modeling method based on recurrent neural network,which can provide a theoretical basis and technical reference for the compensation of thermal errors in electric spindles.Firstly,the whole thermal error experiment scheme is designed according to the real machining conditions;select the temperature measurement points according to the electric spindle structural characteristics,design and build the electric spindle temperature and thermal error detection system;complete the synchronous collection of electric spindle temperature and thermal error data according to the test scheme.This could provide data support for the selection of electric spindle thermal sensitive points and thermal error prediction modelling.Then,based on the experimental data,cluster each temperature measurement point based on fuzzy relationship,reduce the degree of similarity between each temperature measurement point,and preferably select the temperature measurement points(thermal sensitive points)with high degree of correlation with thermal errors in each category by the gray absolute correlation analysis method;Analyze the forward and backward propagation algorithm of recurrent neural network,elaborate its timing characteristics,design and build the recurrent neural network model based on the thermal sensitive points,and establish the electric spindle thermal error prediction model.Finally,to solve the problem that recurrent neural network doesn't have the feature of robustness,the long and short-term memory network improved from recurrent neural network is proposed and compared.The results show that recurrent neural network model performing well in prediction accuracy but bad in robustness,while the long and short-term network model has both high prediction accuracy and strong robustness.