Study Of Chatter Identification Of Robot Milling Based On ResNet And SVM

Robot milling can not only reduce the cost of manufacturing,but also improve the level of manufacturing automation.However,compared with the traditional NC machine,the robot is less rigid and prone to vibration problems.And the dynamic characteristic of robot is position-dependent,the traditional analytical method taking the parameters of the machining system as the model input to identify the robot milling chatter,which is difficult and has limited precision.By constructing vibration data sets of different robot milling chatter status and based on artificial intelligence algorithm,a chatter identification model with the vibration signal as input can be established to accurately identify robot milling chatter.Based on the analytic method,dynamic equation of robot tool was established and solved by the time-domian method.Stability lobe diagram was plotted and a planar milling experiment was designed to collect vibration signals and surface pictures of workpieces,and the sample sets of vibration data under different chatter status were constructed.By observing the surface,frequency domain and time-frequency domain methods,the timevarying chatter components of the robot were analyzed,and the chatter status of the robot milling were divided into four classes: stability,transition,regular chatter and irregular chatter.The vibration signal was transferred to the wavelet spectrum image,while using the deep residual convolution neural network to classify the spectrum image.By using the normalization of inputs and increasing wavelet decomposition scale,the classification accuracy and convergency speed of the model was increased.And using VMD to preprocess vibration signals can further improve the average recognition accuracy,up to 95.28%.At the same time,Python and Qt were used to build a visual software platform based on the deep chatter identification algorithm,which realized the offline identification of the chatter status of the selected time-frequency spectrum image.For realizing the on-line identification of the chatter,vibration signals of robot milling were decomposed into sevreal chatter frequency bands based on VMD.The entropys of the decomposed frequency bands were calculated and used as the feature of robot milling chatter,and a vmd-svm algorithm was established to classify the chatter status.At the same time,the kMap algorithm was proposed to optimize the three super parameters of the vmdsvm through three stages without the calculation of all solutions.Compared with traditional method,kMap algorithm could quickly calculate the global sub-optimal solution which was not far from the global optimal solution,and the average identification accuracy is 92.43%.An online acquisition hardware module was built to monitor and collect the vibration signals of the robot spindle during robot cutting.Using C#,a software module for on-line chatter monitoring and identification based on vmd-svm algorithm was written.Meanwhile,the robot milling experiments were performed to use the written software.By analyzing the frequency spectrum of milling vibration signals and surface of the processed workpiece,the effect of the proposed software module was validated.