Study On State Monitoring Technology Of Tool Wear Based On Hilbert-Huang Transform

To study monitoring technology of tool wear states can not only monitor tool states inreal time and extend the life of cutter, but also can reduce the cost and improve the processingquality, etc. In automatic production system, the safety of workers can be guaranteed verywell. Therefore, this research has the very high practical value.Monitoring technology of tool wear states mainly includes three steps, the selection ofmonitoring signal, the technology of information processing and diagnosis and identificationof tool wear states. In the cutting processes, acoustic emission signal has high correlation withtool wear states, and acoustic emission sensor is easy to operate, so this article selects theacoustic emission signal as monitoring signal.HHT (Hilbert Huang Transform) analysis technique and the filtering pretreatment of bestwavelet packet are used in information processing technology. Because the signals whichcollect from acoustic emission sensor often have some interference, they can not be directlyused to recognize tool wear states and must be filtered firstly. This paper discusses threefiltering methods, including Best Wavelet Packet filtering, Stein Unbiased Risk Estimate filteringand HHT filtering, the comprehensive contrast result is that the Best Wavelet Packet filteringis suitable for the preprocessor of signal analysis. HHT method is a kind of adaptive signalprocessing method, it can make the various frequency component of signal separated fromtime curve in Intrinsic Mode Function form, but HHT is a kind of new method, there aremany areas in need of improvement. End effect, stop standard and artificial components ofempirical mode decomposition are improved in this paper. In three tool wear states,normalized energy of the IMF components and EMD energy entropy were extracted ascharacteristic vectors. The experiments result shows that high frequency component of IMFincreases gradually, low frequency component decreases gradually and EMD energy entropyalso decreases gradually along with the deepening of the tool wear degree.Support vector machine has good generalization ability in small sample and nonlinear case, therefore, it is used as a classifier of tool wear state in this paper. This article will usecharacteristic vectors as training samples and testing samples of support vector machine. Dueto characteristic vectors are not linear separable signals, it need to use nonlinear supportvector machine. This paper choose least squares support vector machine to recognize toolwear stats, the result shows that it can accurately and effectively identify tool wear states.