| As fiber reinforced composites extensively in the application of aerospace, automobile,construction filed and other areas in daily life, its security problem such as damage is fundamental andcrucial for production and in use. Acoustic emission technology is an in-situ non-destructivetechnology; it can forecast damages in composites and structures much earlier than other non-destructive methods. During the loading process, there are a large amount acoustic emission signalsemitted, and the acoustic signal consists of useful information of the damage sources. The challenge ofAE analysis is determining connections between AE signals according to the nature of the damageevents they originate from. One of the generally accepted ways of discriminating AE signals is clusteranalysis, which is a synonym for unsupervised pattern recognition technique．In this study, cluster analysis algorithms were studied and discussed in detail. A useful featureselection method with Laplacian score combined with correlation analysis was proposed, the clustervalidity was checked by silhouette coefficient and Davies-Bouldin index. According to these clusteranalysis methods and process, cluster analysis toolbox was designed based on Matlab R2013a program.There have been many reports on acoustic emission used for fiber reinforced composites, but many ofthem concentrated on unidirectional and cross-ply thermoset composites, studies on2D and3D fabriccomposites with complex preforms are limited, in addition, studies on thermoplastic composite are stillnot enough. In this thesis, based on the cluster analysis of acoustic emission signals, comprehensiveresearch was carried on the damage mechanics of complex preform reinforced thermoset andthermoplastic composites. For thermoset compsites,2D and3D glass and carbon woven compositeswith similar structure were studied in this paper. For thermoplastic composites, Polyethylene self-reinforced graded thermoplastic composites were considered. By cluster analysis methods, the acousticemission signals of these different materials were grouped, and then combined with Digitalphotography technology, and Weibull predicted fiber bundle model, scanning electronmicroscopy(SEM) to investigate the connections between clusters and different damage modes. By useof the damage threshold method and results of clusters and corresponding damages, the damageinitiation and propagation process were discussed. By wavelet analysis, the difference of AEregistration of thermoset and thermoplastic composites were made comparision and discussed. Themain research contents of this thesis can be summarized as follows:1. Considering the importance of feature selection, this study investigates feature selectionmethods, which is the first and import step in cluster analysis process. A useful feature selectionmethod with Laplacian score combined with correlation analysis was proposed, this can eliminateredundant features and selected those features with high classification ability. Before cluster analysis,nine acoustic emission features were selected by feature selection method, and four features were selected, and by the eigenvalues of principal components, two important features were found, whichcan characterize acoustic signals, and make the analysis and discussion of cluster results much easierand reasonable. Based on Matlab R2013a software program, cluster analysis toolbox was designed,which can be easily used for comprehensive cluster analysis of unknown acoustic emission signalsgenerated from different materials.2. Making comparison of six cluster algorithms, which are k-means, fuzzy c-means (FCM), k-means++, SOM+k-means, SOM+FCM and Adaptive affinity propagation (AAP) algorithms foracoustic emission signals of fiber composites. k-means++and AAP algorithm are the better among thesix algorithms，but AAP needs more computation time and cost, and good data seperation is needed,otherwise it will fall into the infinite iterative process. k-means++, which is an optimised initial clustercenter selection algorithm, needs shorter computation time, faster computation speed, it is moreapplicable for cluster analysis for AE data generated by composites under loading.3. Damage mechanisms of the glass fiber fabric reinforced thermoset composites wereinvestigated. The acoustic emission signals generated from the tensile damage process of2D and3Dglass/epoxy composites were studied by cluster analysis technology. AE events can be discriminated infour clusters based on peak amplitude, peak frequency, frequency of centroid and RA value: lowfrequency&low amplitude cluster, moderate frequency&low amplitude cluster, low to moderatefrequency&high amplitude cluster, high frequency cluster. Peak amplitude (PA) and peak frequency(PF) are the most important parameters in this discrimination. Moreover, for all the studied test variantsthe boundaries of the clusters in PA-PF coordinates are the same.4. Damage mechanisms of carbon fiber fabric reinforced thermoset composites with similarstructure were studied. Cluster analysis is performed on the acoustic emission signals generated fromthe tensile damage process of2D and3D carbon/epoxy composites, and the cluster results werecompared with that of glass counterparts. Results showed that cluster shape and cluster boundaries aresimilar for similar structures and the cluster bounds of amplitude and frequency for the carbon/epoxycomposites are higher than for the glass/epoxy. The acoustic emission signals for2D and3D glass andcarbon weave composites tensile in different directions can be summarized to three clusters: lowamplitude&low frequency cluster, high amplitude&low frequency cluster, broad amplitude&highfrequency cluster.5. This research efforts are done to establish correlation between mode of the damage andparameters of the AE signals originated from it. The statistics of high frequency acoustic emissionevents are compared with the estimates obtained from a fiber bundle model based on Weibull fiberstrength statistics. The number of AE events agrees well with the number of groups of carbon fibersthat fail simultaneously. This finding gives additional grounds for the hypothesis that high frequencyAE events represent fiber breakage. Then AE events registered during tensile loading of a plain weaveglass/epoxy laminate are correlated to actual damage, according to its characteristics with observedoptically under transmitting light. And reasonable correlations between the clusters and damage modeswere established: cumulative number of transversal cracks generated in the90°of the yarnscorresponds well to the number of AE events in the low frequency&high amplitude cluster. The studyvalidates use of cluster analysis of AE for identification of damage models in woven glass fibrereinforced laminates.6. This study investigates the mechanical properties and damage mechanisms of thermoplasticcomposites by cluster analysis and damage thresholds methods, the studied thermoplastic materialswere unidirectional and cross-ply self-reinforced graded polyethylene composites. Firstly,unidirectional and cross-ply PE/PE graded composites were prepared and their tensile properties underfiber and45°loading direction were investigated and compared with non-graded composites. Secondly, in order to further investigate the damage sources of acoustic emission, much concern was on cross-plyPE/PE graded composites under fiber direction, and the acoustic emissions were studied by clusteranalysis method. Results revealed that cluster number5is optimal for different specimens, and peakamplitude and frequency centroid are the important features and were used for discussion of clusterresults. Through the results of amplitude and frequency characteristics from literatures on simple modelPE/PE specimens and typical damage modes, the connections of five clusters of acoustic emissionsignals of cross-ply PE/PE graded composites and five damage mechanisms were established, and bywavelet analysis, the difference of AE registration of thermoset and thermoplastic composites werecompared, the duration of corresponding frequency components AE signals for thermoset compositesis longer than thermoplastic composites, thus attenuation is slower, and the frequency distribution oflow frequency AE signals of thermoplastic composites is concentrated. |
PDF Full Text Request