Application of BP network to tests on fractal characteristics of coarse cross-sections

With the results from scanning of laser profilometer to plaster fracture surface, the dimension Ds and intercept As of coarse surface are obtained on the basis of dimension calculation theory-cubic covering method. Along x- and y-directions, the surface is cut with scanning space (0.2 mm). Based on curve dimension theory-box-dimension method, the dimensions and intercepts of all curves are calculated. Then x- and y-direction dimension average values (Dx, Dy) and intercept average values (Ax, Ay) are all gained. Logical relation among intercept average values of x-direction and y-direction-Ax and Ay, and surface intercept As are explored on the premise of different surface dimensions. By constructing BP artificial neural network with early stopping, mutual relations between the two parameters-dimension Ds and intercept As and four parameters, i.e. x- and y-direction dimension average values (Dx, Dy) and intercept mean values (Ax, Ay) are disclosed.