Optimization Of Acoustic Emission Signal Pocessing And Recognition Technology Of Fiber Tensile Fracture

During the production,the strength of the yarn is influenced by the fiber performance and the structure of the yarn.These factors will directly affect the quality of the production after the procedure.Some scholars have constructed a mechanical model to change the yarn structure to improve the strength of the fiber.The performance of fiber itself is one of the important indexes of fiber strength utilization in yarn.Generally,the research on fiber performance focuses on the mechanical properties of fiber,such as tensile breaking strength,breaking elongation,etc.For a variety of fiber blended yarn,due to the large differences between the surface and the intrinsic properties of different fibers.It is difficult to science comprehensively reveal the process of blended yarn faults,relying only on the mechanical properties of the fiber,or artificial assumptions or simplification of the yarn or fabric structure prediction model.The result of prediction is poor and the prediction accuracy is limited.It is not obvious that the overall strength of the yarn can be improved by improving the strength of various fibers in the blended yarn,and it is also not scientific to describe the stretching and fracture behavior of blended yarn.This paper focuses on the study of the science and the optimization of the recognition technology of the fiber tensile fracture AE signal processing.Firstly,The pre-set fiber tensile fracture acoustic emission signal acquisition platform by the research group is optimized.The tensile fracture AE signal of single fiber was collected by using the platform,and the signal was pretreated by wavelet denoising and EEMD.Then,the distribution range of time frequency and energy characteristics of the Hilbert spectrum of the fiber tensile breaking signal was explored by kernel density estimation method and the characteristics were extracted by principal component analysis.According to the signal characteristics,construct the least squares support vector machine(SVM)identification model,and complete the selection of kernel function and the parameters in the model.Finally,determine a suitable fiber tensile fracture AE signal recognition model.The research contents and main conclusions are as follows:(1)It has been modified to build the platform about the tensile fracture of the fibers and the method of the experimental sample.Using XS(08)X series electron single fiber strength tester,select a single fiber sample with similar fineness.and the tensile fracture signals of fiber were collected under the conditions of constant temperature and humidity.(2)To determine the characteristics acquisition method that the signal is intercepted after wavelet denoising and then the Hilbert transform is performed.The pretreatment of the fiber tensile fracture AE was performed by wavelet denoising and EEMD decomposition.The signal characteristics were obtained on the Hilbert spectrum,and the three-dimensional feature set are including time domain,frequency domain and energy spectrum.The characteristics of the tensile fracture of the fiber were identified from these three aspects.(3)Based on the method of kernel density estimation,analyze and study the distribution characteristics of fiber feature sets in different kernel functions and window width.Through kernel density estimation,determine fiber fracture in different periods from a statistical point,namely fiber breakage when the release of energy in different frequency.This method reduces the range of feature extraction of fiber tensile fracture AE signals,and provides support for checking the accuracy of feature extraction.(4)The principal component analysis(PCA)is used as feature extraction method to achieve the dimensionality reduction of AE signal of fiber tensile fracture.This method reduces the feature set of the fiber tensile fracture AE signal from 400 dimensions to less than 50 dimensions,and establishes a new feature set of four single fiber tensile fracture AE signals.(5)Based on the least-squares support vector machine method,construct the LS-SVM recognition module for the characteristic frequency of specific fiber tensile fracture.From the perspective of nuclear parameters and punishment factors,the paper discusses the influence of the number of support vector,recognition accuracy and recognition time of LS-SVM recognition respectively.The grid search-K-fold cross validation method is chosen as the method of parameter optimization of the fiber tensile fracture AE signal recognition module.Based on constructing and researching of multi-classification model,adopt the MOC multi-classification method.This paper compares RBF,poly and linear kernel functions with respect to the recognition accuracy and training time,and determines that the RBF kernel function is the kernel function in the four fiber LS-SVM recognition module in this experiment.The identification results show that the optimization of various parameters in LS-SVM can obtain the recognition module with good performance.(6)To study different nuclear function and different classification methods,achieve the determination of various fiber fracture conditions(types and sequence)in the stretch process of blended yarn.First,the fibers of the four components in the blended yarn are used as training samples to train the model;Then the suspected fiber break signal is intercepted before the breakage of the blended yarn;and finally,each signal in the blended yarn sample is put into the recognition module to classify,and combined with the mechanical properties of each component fiber in the hybrid yarn.The results show that the proper model can be used to determine the sequence of various fibers in the process for blended yarn.The research results of this study make it possible to put forward specific measures to enhance the contribution of various fiber strength in the blended yarn.It is of positive significance for guiding textile enterprises to improve the mechanical properties of the yarn and fabric.