Data Fusion Analysis And Pattern Recognition Based On Multi-Channel Signal

In the process of cutting,the tool will gradually wear with the increase of service time and the number of workpiece,which not only directly affects the surface morphology of the workpiece,but also affects the processing quality,production efficiency and the entire processing System,so the study on the state of tool wear is of great significance.Starting with indirect method of tool condition monitoring,this paper establishes an effective tool wear monitoring system through the multi-channel signal synchronization analysis and pattern recognition,which can effectively determine the tool wear and tear.The main research results of the dissertation are as follows:Firstly,an adaptive noise-assisted multiple empirical mode decomposition method is proposed to anlyze multi-channel vibration signal synchronously.Two noise-assisted channels are added to the multi-channel signal.The combination of K?the number of projection vectors?and^?1,?2?the noise intensity of the two auxiliary white noise channels?is optimized by the adaptive weight-particle swarm optimization algorithm and the objective is to minimize the weighted orthogonal index?WIO?.The method improves the decomposition precision and suppresses the mode mixing effectively.The effectiveness of the proposed method is verified by simulation data and real case data.The adaptive noise-assisted multivariate empirical mode decomposition method can extract fault frequencies more accurately.Secondly,because the original signal inevitably contains a lot of noise,the improved Noise Assisted Multivariate Empirical Mode Decomposition?NA-MEMD?method can not directly distinguish useful signals and noise signals,which is not conducive to further extract the fault signal characteristics.In this paper,the decomposition of the multi-channel Intrinsic Mode Function?IMF?group were individually screened,then the effective IMF layer is selected by multi-channel integration.The screening of each channel is based on the similarity between original signal and the IMF's probability density function.The order of the IMF is determined by?1?the correlation between the IMF and the original signal channel signal,?2?the correlation between the IMF and the signal of other signaling channels,and?3?the correlation between the IMF and the same level of IMF in the secondary noise channel.The selected IMFs are separated by thresholds to remove noise,and the noise components in the IMFs are removed to improve the signal-to-noise ratio.Then,the hierarchical Bayesian model is used to construct the MRVM model,which solves the problem of RVM multi-classification.Then it's combined with KNN algorithm by its probabilistic output to establish a new multi-classifier.MRVM-KNN improves the recognition accuracy of MRVM,and reduce its dependence on kernel function parameters.The simulation data and some commonly used classification data sets prove the effectiveness of this recognition model and method.At last,Vibration,sound and temperature sensors are installed on the three-axis machine tool in the 149 space plant.Signals are collected by the acquisition card and PC.Then experiments are designed to get signal corresponding to three different degree of wear knife.The proposed tool condition monitoring method is used for analysis,with a high recognition rate.Based on this algorithm,a set of software is developed for the actual production.