Research On Cutter Wear State Recognition Base On Milling Force Signal

With the constant improvement of the production automation degree, cutting state monitor of the cutter has become a key technology for achieving machining process automation, especially for the material which is hard-to-cut. After having investigated and researched on the spot for milling High Manganese Steel (ZGMn13) Rail, it was discovered the frequency of "cutter Breakage" highly occurred when cutting High Manganese Steel before-mentioned in hard alloy cutter (YT767), and cutter must be changed every five or six minutes. In this case, monitoring in real time for wear state of the cutter, and discovering the severe wear in time, so that "tool breakage" can be avoided, it is vital important for prolonging the cutter service life, enhancing the productivity and Reduction of cost.The dissertation, takes the course of milling hard-to-cut material-High ManganeseSteel as the research object, establishes experimental system of milling cutter wear state monitor which regards the milling force as the monitor signal, and carries out theory analysis systemically and experimental research in order to de-noise and feature extraction of the milling force signal and cutter state recognition.Based on the average force method, the dissertation establishes the milling force model and the coefficient quadratic regression equation model, and proposes a new method ofcoefficient recognition-main component method. After experimental validation andemulator, it is indicated that the coefficient model is suitable on the different work condition and the recognition method is faster and more available than ever.Through analyzing the characteristic of milling process systemically and the cutter wear type in the course of milling, this dissertation proposes a view of using the average wear value of the back surface in all cutter tooth as monitoring index. By the wear experiment of the cutter, wear curve is demarcated, and corresponding wear value range of each wear phase is marked off, a milling cutter wear state monitoring scheme and experimental system with X and Y direction force as the monitoring signal is built up, and on the basis of all above, the data of milling force signal is collected.Milling force signal is a sort of non-stationary signal with serious noise. Wavelet package Transform not only is the potent method for analyzing the non-stationary signal, but also can be detailed in analyzing the high frequency part of the signal. The dissertation adopts wavelet package theory to analyze and de-noise the milling force signal, and extracts energy feature of the signal as import vectors of Neural Network.Based on strong non-linear mapping and classified capability of Neural Network, this dissertation adopts the combination of wavelet package analysis and BP(Back Propagation) Neural Network to recognize cutter wear state, and establishes the Pattern-recognition configuration of BP network, constructs net training swatch and testing swatch. The results of training, simulating and validating to network indicate the net can exactly recognize cutter wear state and has significant realistic meaning to cutter online monitor.