Abstract:
This paper proposes a method to measure the crack propagation length during structural deformation. A convolutional neural network is established to eliminate the interference of surface marks and noises and identify crack features. The initial area of the crack zone is obtained through the prediction of the convolutional neural network. Based on the initial area, an improved crack tip identification algorithm is proposed to calculate the precise position coordinates of the crack tip. According to the position coordinates, the crack length information is obtained. By increasing the number of cameras, cracks in different directions and positions can be detected at the same time. Using this method, the relationship between the crack growth length and the load information (such as fatigue cycles) can be obtained. The effectiveness and accuracy of the method are verified by fatigue tests of the center hole specimen and full-scale bending tests of X80 pipeline steel.