Prediction For Polishing Effect Of Virtual Axis Machine Tool Based On Artificial Neural Network

In recent years, with the development of automobile industry, electronics industry, aeronautics and astronautics industry and some new high-tech industries, higher and higher requirements are made for quality and efficiency in machining die and mould with free-form surfaces. Traditional machine tools have some problems to be solved urgently in precision machining for free-form surfaces. As a result, the research of Parallel machine meets the requirement to develop new precision machining equipments for free-form surfaces. This paper focuses on improve the polishing process quality of JDYP51virtual axis machine tool designed and produced by Jilin University, to make it better move forward to practicality.In the NC process of machine tools, the best way to control roughness of the workpiece is to control processing parameter of machining tool. How to provide the reasonable processing parameters to acquire the better surface quality of workpiece, this paper puts forward a new way that building forecast model for roughness of workpiece based on ANN theory optimizes process parameters of machine tool. With the research of experiment and prediction in JDYP51virtual axis machine tool designed and produced by Jilin University,the forecast model of process parameters and roughness of workpiece based on ANN theory is established to describe the machining process via limited experiment data.It also puts forward a new way for roughness prediction and process optimization. 1.Application of ANN theory in machining process of virtual axis machine toolANN theory,the nonlinearity tool is good at solving nonlinearity problem with the characters of nonlinearity map, self-organizing configuration and high-parallel process.It not only automatically studies previous experience from data sample and approach to the function law to describe best the data sample without reference to function form ,but also show more obviously such characters as for more complicated system functions. BP network framework is applied most broadly as a kind of ANN model, because it solves the problem that multilayer network withhidden layer studied difficultly.This paper applies BP network arithmetic as basic theory, and combines with the experiment result of polishing process in virtual axis machine tool to forecast the roughness of workpiece polished by machine tool. In the prediction model framework, polishing process parameter is as input of network model, and roughness is as output. After the establishment of the prediction model, the originally established model will be simulated to train and tested with the data sample from polishing process experiment to obtain the eligible ANN model. Concrete practice application of ANN theory is shown in fig 1,2.2.Research of polishing plane experimentIn the polishing process of virtual axis machine tool, there are some factors that affect process effect, such as spindle speed, feed speed, process force, polishing angle, polishing time that are foremost and direct for roughness of workpiece. In such factors, spindle speed, feed speed and process force affect the roughness more directly than others, polishing angle that has been adjusted before machining is not changed in the process of polishing, polishing time is rest with practicalrequirement of polishing effect. As a result, three experiments are designed to research the polishing effect in different process conditions in this paper, spindle speed, feed speed and process force will be considered as influence factors in the first experiment. Influence factors in the second experiment contain spindle speed, feed speed and process force.The above five factors will be synthetically researched in the third experiment.The result of polishing plane experiment will be applied to train the roughness prediction model based on ANN as the study and test sample.3.Improvement of ANN training arithmetic and model design This paper put forward improvements as follows to improve the forecast and self-study ability of network model: one method is that individual test error is applied into the process of training the network model to monitor the individual error of study sample, avoiding exceptional individual error(too big or too small). Another better method is that the unattached test sample data is applied to check if the network model has the strong prediction ability, the test sample data doesn't join in the study process of training network model, is only applied into the test process to monitor the change of training .Improved network training flow is shown in fig3.On the basis of the above improvement, visual program of prediction system will be developed in MATLAB software .In a word, BP network model design contains the whole framework design and determination of network training parameters. Concrete design standard of network model is listed in table 1 and table 2.4.Training and test of the roughness prediction model for polishing surface According to the result of polishing plane, the roughness prediction model is trained and tested with the new-developed program of ANN prediction system, and to check out the prediction ability and precision and practical application effect of new-developed program.Fig. 4 the final structure of prediction roughness modelThe final prediction roughness model based on ANN is shown as fig4.Simulation experiment verified the prediction roughness models based on ANN that have been trained and tested have the ability to forecast the roughness of polished surface, and the individual prediction precision is all below 5% without the abnormal excess-study. It also indicated that the prediction ability and precisionof network models meet the practical requirement.5. ConclusionThis paper comprehensively research the process factor for polishing effect such as spindle speed, feed speed, process force, polishing angle and polishing time, on the basis of surface polishing experiment and combined with ANN theory, to build the forecast model of roughness for the polishing process. The built network models that have been simulated to train have the all-right prediction ability and precision to forecast the polishing roughness and meet the practical process requirement.