Research On Metal Fracture Images Classification And Fatigue Period Measurement

Wear, corrosion, fracture are the three primary failure modes for metal component, among which the fracture failure is the most harmful. With the rapid development of materials science, the upgrading of equipment and the improvement of pattern recognition technology , automatic quantitative analysis and intelligent diagnosis of fracture component have become two of the frontier research fields of fracture failure. The texture characteristics of fracture image are not obvious due to the constraints of the fracture status and the environment where we take the pictures. Especially in fatigue fracture, the fatigue striations are often very vague, which make it difficult to recognize the fatigue striations even with the naked eye, not mentioned to measure the striation period accurately. Therefore, fracture texture image segmentation and fatigue fracture period measurement are the key problems in automatic quantitative analysis and intelligent diagnosis of fracture component. This paper is focus on the two key problems, and the discussions are performed mainly in the metal fracture image classification, fatigue fracture image identification, segmentation and the fatigue striation period measurement. Meanwhile, a system platform of fracture image quantitative analysis is developed. The main work and results of the paper are the following:1. The classification of fatigue, dimples, intergranular and cleavage are discussed. Firstly, according to the problem of uneven illumination of fracture image, the improved image filtering method is discussed in the paper. Secondly, on the issue of feature extraction, due to the shortcomings of GLCM in descripting features of fracture, the corresponding fuzzy membership function is constructed based on the different distances which between the real gray-level and the mean of quantization gray-levels area, and then, a newly co-occurrence matrix, namely fuzzy gray level co-occurrence matrix (FGLCM) which applied to metal fracture image recognition is proposed in the paper, and experiment shows that FGLCM is better than the GLCM in fracture images feature description.2. Feature selection is one of the important issues in fatigue fracture image identification. It is a combinatorial optimization problem. In this paper, an improved multi-objective genetic algorithm is proposed based on the density of individuals around. Then a rational multi-objective function is designed, and applied to feature selection of fatigue fracture image. Experiment improves that the improved multi-objective genetic algorithm is effective in feature selection of fracture.3. Fracture image segmentation and determination of striation region are the precondition in the measure of fatigue striation period. A two-level fatigue fracture image segmentation method is proposed in the paper, the rough segmentation is applied in fracture texture image according to the texture information, and then the striation region of fatigue is obtained. Then the method of cumulative squared difference is adopted to determine striation direction, and accurate segmentation is worked in tangent and normal direction of fatigue striation, and then the rectangular area of fatigue striation can be determined accurately.4. Fatigue striation is the typical microscopic characteristics of fatigue fracture. And the period of fatigue striation is an important parameter when quantitative calculating fatigue life and fatigue stress. Usually the period is obtained by artificial. An automatic fatigue striation period measurement method is firstly proposed in the paper, and a triangular mathematical model of period calculated is established. Then three striation period measure methods such as image thinning, EMD based on linear prediction and autocorrelation based on statistical method are compared with each other. Actual measurements show that the method of improved autocorrelation can work at high efficiency and accuracy in measuring fatigue striation period.The propoesd ideas and methods have been programmed into appliction software for the fracture image quantitative analysis under visual c++ 6.0 platform.Various of experments show that the measurement error is less than 5% which satisfy the requirement of engineering.