Research On Heterogeneous Information Fusion Method Of CNC End Mill Wear Condition Monitoring

The wear of CNC tool not only affects the dimensional precision and the quality of the workpiece's surface,but also affects the machining efficiency and the cost of production.Therefore,the monitoring of tool's wear conditions is particularly important and has become one of the popular points researched by domestic and foreign scholars.During the process of researching the monitoring method based on detecting the cutting texture on the surface of workpiece,it is found that this method is easily be interfered by the scratch caused by chips and difficult to guarantee the accuracy of the monitoring results,because this method only considers the single-type information of texture.Therefore,a method of fusing the heterogeneous information was researched to monitor the wear conditions of CNC end mill on-line.This method combined the advantage of strong anti-interference ability of the multisource information fusion.The main research contents are as follows:1.A method for identifying tool's wear conditions by combining the information of the image and the spindle current was proposed.Aiming at the characteristics that the data is heterogeneous and the information has uncertainty,a method of decision-level information fusion combining the PSO-SVM model and the improved weighted evidence theory was adopted.This method was verified on the sample set obtained by the cutting experiment.The results showed that after fusing the information,the average recognition accuracy rates of different conditions were between 97.21% and 98.42%,and higher than the accuracy rates of recognizing based on only one of the information source.In addition,the superiority of the information fusion algorithm was verified by comparing a variety of recognition models and comparing a variety of evidence fusion methods.2.Due to the angle deviation of the visual detection device,the texture directions of images are inconsistent.The inconsistency does affect the accuracy of recognition.To deal with this problem,a texture feature extracting method was proposed,which could keep the feature value unchanged when the image was rotated.The relationship between the results of the multi-angle Radon transform and the main direction of texture was obtained by analyzing experimental data.According to this relationship,the direction of the image could be corrected.Then the Gabor-GLCM was used to enhance and extract the information of texture feature.The verification results of this method showed that the average recognition accuracy of image samples was improved by 4.17% after adopting this method.For current data,its time/frequency and wavelet packet energy features were extracted.In order to remove the features which were weakly related to tool's wear conditions and redundant in the original feature set,the feature selection method combining RF and SVM was verified.3.Based on the CNC on-line exchangeable visual detection device and the secondary development package of FANUC system,a set of on-line monitoring system was developed on Lab VIEW platform combining with MATLAB programming.This system could be effortlessly integrated into CNC system.Its monitoring process was controlled by CNC code,and the machining stage for detection can be set freely according to the needs.This system was tested through the actual machining.Through this testing,the feasibility of the heterogeneous information fusion monitor method was verified in the actual machining site.