CNC Machine Tool’s Motion Error Abduction Based On Graphic Recognition

Motion error, as the final reflection of CNC machine tools error, includes the information about geometric error and control error of CNC machine tools’. It has a significant impact on precision of CNC machine tools. How to monitor the CNC machine tools motion error quickly and accurately plays an important role in inhibiting massive quality issues and improving production efficiency. However, researches on CNC machines tools precision were mostly emphasized on test, control and compensation and there were few on error abduction. Moreover, error modeling was often applied in error abduction in the literature, which is complicated and not very adaptable due to the dependency on CNC machine tools structure and type.In this paper, the circular trajectory graphs of CNC machine tools and pattern recognition technologies are used. A new corner point is defined and a corner detector operator is developed to detect the distribution of corner points in the circle trajectory graphs. Then the circle trajectory graphs are divided into 16 pieces. Characteristic matrix including average radius and number of corners in each piece are established. Supporter Vector Machine(SVM) is employed to verify the robustness of the mapping from characteristic matrix to circle trajectory graphs. Then motion error abduction network based on RBF neural network is built and results show that the proposed abduction network is fast, convenient and of good accuracy and high efficiency. Software with friendly interface is also developed. Method proposed in this paper is simple, economical and useful in practice and suitable for CNC machines tools users in error abduction and control.This paper mainly includes the following parts: corner detection and characteristic extraction for circle trajectory graphs, verifying the robustness of mapping from characteristic matrix to circle trajectory graphs and building general error abduction network.Firstly, a new defined corner is introduced to recognize circle trajectory graphs, then they are divided into 16 pieces after pre-processing, and corners detected by the developed corner detection operator are extracted in each piece.Secondly, corner distribution in each piece of circle trajectory graphs is analyzed. Mean radius and number of corners are calculated to obtain the characteristic matrix. Then mapping from characteristic matrix to circle trajectory graphs is established. SVM is applied to verify the robustness of this mapping. Experimental result shows that classification of testing samples is significant, which means the mapping established in this paper is robust.Finally, general error abduction network based on RBF neural network was proposed. Setting the characteristic matrix as input and every motion error as output, the trained network has good performances and high accuracy for abducting motion errors.