| Thermal errors have become the major contributor to the inaccuracy of machinetools. The motorized spindle is the main heat source and thermal deformation ofmotorized spindle impact the performation of the whole machine tool directively. Inthe thermal error compensation technology, the accuracy and robustness of the thermalerror model is one of hot and difficult problems in modern precision engineering. Thisdissertation studies on the following key techniques about this subject: quantitativeanalysis of heat source and heat transfer mechanism, the optimization of the thermalkey points and motorized spindle thermal error modeling method in different operatingconditions. Conclusions from this dissertation are all verified by simulation andexperiment application. The main research contents are shown as follows:(1) Temperature field of motorized spindle are distributed in unbalance, whichleads to thermal error. The heat generations of two internal heat sources are closerelated with speed of spindle, and the relationship between heat source and speed isestablished. Using the basic theory of heat transfer, the heat transfer model is foundand the heat transfer equations for complex temperature boundarys of spindle unit arederived. The model of motorized spindle is built and analyzed by FEM. Thetemperature field of spindle is obtained, by applying thermal and structural loads onthe FEA model. Finally, the correctness of theoretical analysis is validated byexperiments.(2) According to the choice of temperature sensors placement in thermal errormodeling, a new method for optimizing the locations of thermal sensors is proposed.Temperature measuring points are divided into groups by using fuzzy clusteringmethod, in terms of the measured temperatures and thermal errors. Grey relationalmodel is adopted to analyze emphasis of each measured point to thermal error intemperature field distribution of motorized spindle and the most sensitive points arepicked out. The best combinations is choosed by using the modified coefficient ofdetermination, which will reduce the number of temperature sensors and the modelingtime. Finally, the optimized strategy of chosing the variable is presented depending ondifferent test conditions and request.(3) The neural network model based genetic algorithm and autoregression modebased on temperature are put forward respectively and the performances of twomodeling method are studied. According to different representations of generation mechanism of motorized spindle thermal error, operation efficiency and curve fitprecision of these two models are compared. The results indicate that the estimationranges of two models are different, that the MVAR model has higher forecast precisionin short-term prediction, while the GARBF neural network model has higher forecastprecision in mid-long term forecasting. Finally, by comparing several commonmodeling methods, an autoregressive analysis model and a gray dynamic model areused to predict thermal error respectively and on the basis of those models, a hybridprediction model based on radial basis function neural network is proposed. The testresults show that the prediction of the combined prediction model for motorizedspindle thermal errors is effectively improved.(4) According to the generation mechanism of motorized spindle thermal error, thecombined forecasting model based on the fuzzy logic is proposed under two typicalworking conditions. The model can be expected to improve the forcasting accuracy byusing autoregressive analysis method and grey system theory combined with differentweight. The combined model synthetically uses the above methods and makes the bestuse of them. Through the prediction study on thermal error model under the constantrunning condition and the progressive running condition in experiment platform ofmotorized spindle, experiment results demonstrates intelligence combinationforcasting model has higher precision and stronger robustness, which provides newideas for forecasting and controling thermal error. |
PDF Full Text Request