An Online BEMD And LSSVM-Based Grinding Chatter Detection Method For Large Grinding Machine

Grinding is an essential process in the machinery manufacturing industry,and is widely used to create machined surfaces with high precision and smoothness.Monitoring condition and diagnosing faults of grinders ensures their safe operation,which has great practical and economic value.Most notably,grinders can enter a“chatter”state during machining operations,which leads to a series of negative effects.Thus reliable chatter detection and identification methods are essential to detect the grinding state in real-time.The vibration signals collected from grinders are mainly non-stationary,nonlinear and multidimensional.Conventional methods are mostly based on the theory of Fourier transform that cannot effectively extract signal features.BEMD is an extension of EMD,aims at decomposing a complex-valued signal into a collection of zero-mean rotating components.BEMD can not only describe the nonlinear dynamic behavior of grinders,but also save computation time and remove distortion which is caused by assumptions and man-made factors.BEMD has more powerful capability in terms of detecting early faults and extracting grinding chatter features from non-stationary,nonlinear vibration signal effectively.The experiment of this paper is mainly focused on CNC guideway grinder KD4020X16.A chatter detection platform is built according to the static and dynamic characteristics of grinder and then a multi-level grinding experiment is carried out,that testing the grinding states in various parameters of grinding wheel speed,feed speed and grinding depth.Using IEPE piezoelectric acceleration sensor and TST5912 dynamic signal analyzer to collect and save the grinding machine vibration signal,that 80 groups of grinding signals are obtained,where 45 groups are in a stable grinding state,while 35 groups are in chatter state.In this paper,the BEMD is adopted to process the complex-valued signal which is constructed by selecting parts of the x-direction and z-direction chatter data,that a series of BIMF are obtained.Then the extraction criterion based on correlation coefficient could be applied to extract the true BIMFs.The sensitive feature vectors — peak to peak,real-time variation,kurtosis value and instantaneous energy could also be extracted from the experimental true BIMFs,and then superimpose and normalize respectively.Lastly,the LSSVM is adopting as an intelligent pattern classifier of grinding chatter signal,that the identification model could be established using training samples,which utilize feature vectors of 55 signals as the training set,with the remaining 25 signals used for testing.The results reveal good recognition rate of this method.According to above method,a grinding chatter detection software is established to validate the feasibility and effectiveness of monitoring the grinding states in real-time.