An EMD And ANN-based Grinding Chatter Detection Method For Large Grinding Machine

CNC rail grinder，which is a high precision,large tonnage machine, is the key equipment ofmachine tools and other machinery and equipment manufacturing industry. State monitoring andfault diagnosis for this kind of equipment to ensure the safe operation has important practicalsignificance and economic value.Grinding chatter is one of the major forms of the host fault inthe process of grinder working,and machine vibration signal is feature signal carrier of theequipment fault,when the grinder occurs chatter signs,it would be reflexed in machine vibrationsignal.So predicting the grinding chatter can be achieved by monitoring grinder vibration signals.Because the grinder vibration signals are mostly nonlinear and non stationary signals,thetraditional Fourier analysis cannot effectively reflex the feature of vibration signals.But theempirical mode decomposition method has excellent characteristics of time frequency analysisand self adaption to different signal,which can analysis and extract non stationary grinder chattersingal feature more effectively.An experimental method is presented to acquire different vibration state of differentgrinding parameters for the CNC guideway grinder KD4020X16from Hangzhou HangjiMachine Tool Co., Ltd. The IEPE piezoelectric accelerometer sensor with the supportingdynamic signal test and analysis system names TST5912was used to collect grinding vibrationsignal.Finally,80groups of grinding vibration signal of different vibration state is obtained,ofwhich46groups of stable grinding signal and34groups of chatter signal.Combines the grinding mechanism and vibration characteristics of grinder, firstly, themethod of empirical mode decomposition is used to decompose grinder vibration signalcollected from experiment into a series of intrinsic mode functions(IMFs),and then eliminatefalse vibration mode with the method of mechanical vibration model validation based oncorrelation coefficient. Second, extract signal features which is sensitive to grinderchatter—real time variance and instantaneous energy to form chatter feature vector.Finally,takethe BP neural network as classifier of grinding vibration signal. Randomly select60groups ofdata with different vibration states from chatter feature vector to training BP network, and withthe other20groups of data to test the trained network and analyze test results.The test results of BP network show that the design of chatter prediction system in thispaper based on EMD and ANN method has good recognition rate, and also proved the feasibility of BP neural network technology for the identification of grinder chatter.