Fatigue Striation Segmentation Of Metal Fracture Image Based On Multi-feature

The fracture is defined as the mutual matching surfaces which are formed by thesample or metal components, which cracked in the process of test or use. Due to theimportance of fracture research in the material field, a new discipline "fractography"is derived. The long service of the components in the engineering and mechanicalequipment under alternating stress commonly results in fatigue fracture phenomenon,which causes serious economic losses, so a large number of researchers attached greatimportance to the study of fatigue fracture. With the development of computertechnology, the application of digital image processing and pattern recognitiontechnology has become a frontier topic in the field of the fracture.Fatigue striation is the typical microscope characteristic of fatigue fracture, andits segmentation is important to the quantitative analysis of metal fracture image,which can backwards concludes the fatigue crack propagation life and fatigue stress.Actual fracture is diverse and has mixed morphologies. Furthermore, the period offatigue striation varies greatly in different regions because of the complexity offracture process. Thus the accurate segmentation of texture region and texture borderis very difficult. Image texture analysis is based on texture feature. The main textureanalysis methods are as follows, statistic method, structural method (commonly usedfor regular artificial texture), model method and signal processing method. Theaccuracy of traditional single texture feature method is low for the segmentation ofcomplicated natural texture. Gray level co-occurrence matrix (GLCM) in spacedomain and wavelet transform method in signal field, as two classical texture analysismethods, have been widely used in a variety of texture analysis. According to theanalysis on the natural texture characteristics on the fracture surface, this paperdevelops a method based on multi-feature to segment the fatigue striation in metalfracture image. GLCM features and wavelet packet transform(WPT) features arecombined so that the dual advantages of the multi-features in time domain andfrequency domain are produced, which can be viewed as an innovation of this paper.In this paper, the main contents are as follows:(1) Firstly the research background about this paper as well as the textureanalysis methods are introduced in detail, which establish the theory foundation forthe subsequent chapters. (2) Fourteen dimensional texture features in metal fatigue fracture image wereextracted by GLCM. According to the correlation analysis, the14dimensionalfeatures space was compressed into4dimensional features space. Then FCMalgorithm was used to segment the image. Finally the results are analyzed.(3) According to the analysis of metal fracture image texture characteristics andsufficient experiments, the appropriate frequency bands were selected by WPT, whichcan express the texture features of fatigue fracture image. Then the features wereextracted and clustered by FCM algorithm to segment the image. Finally theexperimental results were analyzed and discussed.(4) The experiment results show that there are some limitations by using theabove two kinds of single texture feature method for the segmentation of fatiguestriation. They were all lack of segmentation. After the analysis of the texturecharacteristics in metal fatigue fracture, this paper proposes fatigue striationsegmentation of metal fracture image based on multi-feature. Combined with the priorknowledge, the experimental results show that the accuracy rate is improved with themulti-feature method proposed in this paper.Above all, in order to obtain the feature sets which can better describe thefracture image texture feature, this paper present fatigue striation segmentation ofmetal fracture image based on multi-feature, which can realize the morecomprehensive location to the fatigue striation areas. The experimental results showthat: the method based on multi-feature is superior to the traditional methods in theaccuracy of auto-segmentation.