Quantitative Characterization And Automatic Identification Of New Different Index Of Typical Damage Of Composite Materials

Rolling bearings,as one of the traditional mechanical components,play an important role in supporting the transmission in rotating machinery.And on the road of China's mechanical industrialization development,rolling bearings have also played a pivotal role.For example,during the running of a high-speed train,the stability and safety of the train travel depends largely on the reliability of the key components(rolling bearings).At the same time,the rolling bearing is the most rotating transmission component,and it also has the problem of poor impact resistance.Irrespective damage often occurs under sudden abnormal impact.At the same time,the working environment of the bearing in the actual working conditions is also relatively harsh,the ambient temperature is too high,the pressure is too large,and the long-time overload operation will further aggravate the fatigue wear of the bearing and the deterioration of the crack.How to diagnose the fault location and degree information of the bearing with a small sample size at the beginning of the failure of the rolling bearing has great research significance for ensuring the smooth operation of the equipment and the safety of the workers.In view of the above problems,based on the Variable predictive model based on class discriminate(VPMCD)classifier,this paper analyzes the classification accuracy based on the least squares regression algorithm behind it.The problem of distribution.On the basis of the existing,three alternatives are proposed to compare the performance of VPMCD with three robust and Gaussian-based regression algorithms.Then,considering the importance of feature preprocessing methods for fault identification of rolling bearings,this paper further analyzes two common methods: normalization and standardization.Through experimental comparison and analysis,the advantages and disadvantages of these two methods in the VPMCD fault identification process,and the corresponding improvement scheme are robust to these two preprocessing methods.Finally,considering that the rolling bearing will trigger the fault frequency signal in different fault frequency bands when the fault occurs,this paper introduces the duffing chaotic vibrator into the field of bearing fault identification.By setting up a duffing system that detects different frequencies,a preliminary analysis of the bearing failure is made by the change of the phase trajectory.A new phase trajectory quantization feature is proposed,and the correlation between the distribution of the feature and the fault location of the rolling bearing is analyzed,and its contribution to the VPMCD pattern recognition process as a training set.The main innovations of this article are:(1)The basic theory of VPMCD and its least squares regression algorithm are studied.In view of the dependence of the least squares regression algorithm on the normal distribution of training samples,this paper proposes to replace the original regression algorithm with three regression algorithms: kernel ridge regression,support vector regression and Bayesian ridge regression.The results show that the enhanced VPMCD can achieve rapid convergence in the training phase,and significantly improve the classification accuracy without losing its timeliness.(2)This thesis analyzes the specific effects of two classical feature preprocessing methods on VPMCD pattern recognition.After normalizing and normalizing the same feature set,it is found that the normalized preprocessing method is susceptible to abnormal samples in the feature space.However,VPMCD based on Bayesian regression is less affected by the distribution of feature samples.In both cases,the classification accuracy is very different.For VPMCD based on support vector regression or kernel ridge regression,the preference for feature normalization processing is shown.(3)The weak signal detection mechanism based on duffing chaotic oscillator is studied and a new quantitative feature is proposed.After the blind source separation is performed by empirical wavelet decomposition of different fault frequency signals,the duffing system that inputs the purified signal into the critical state performs weak signal quantization detection.After that,the phase trajectory is subjected to k-mean discretization processing,and the discrete entropy value is counted.(4)The discrete entropy feature of the duffing phase trajectory is introduced in this thesis corresponding to the three fault frequencies into the original RQA feature set,and analyze its contribution to bearing fault identification under actual working conditions.And the correlation between the features is analyzed by Kendall correlation coefficient,and then the independence of the new features is verified.Through experimental verification and analysis,it is found that the improved VPMCD algorithm has improved the efficiency of the sample,whether it is for the processing of small sample problems or the classification accuracy and stability of the whole,which is obviously improved compared with the original algorithm.And,with the addition of new features,the performance of the classifier has been further enhanced.