Rock block shape classification and numerical simulation of soil-rock mixture based on machine learning algorithms
Existing numerical models of rock block shape characteristics either oversimplified the block shapes or did not carry out the statistics of the block shapes.A new modeling method was proposed based on the principal component analysis algorithm(PCA)and K-means clustering algorithm.Matlab programs were used to digitally process the cross-section photos of the soil-rock mixture to obtain the contour samples of rock blocks,and the standardization processings of rock block contour such as moving the centroid to the origin,rotating the long-axis to the horizontal-axis,and normalizing the polar radius were performed to obtain standardized silhouette vectors of rock blocks.The PCA was used to reduce the dimension of the contour vector of the rock blocks,and the K-means clustering algorithm was used to cluster the contour vector after the dimension reduction.The shapes of the rock blocks were classified and the frequencies of various types of rock blocks were counted.A random model of soil-rock mixture considering the shape classification and frequency,grain composition,and inclination was established.The biaxial compression numerical simulation was carried out,and the characteristics of the plastic strain and the stress-strain curves were analyzed.The models of the deformation and compression strength of the soil-rock mixture considering the rock block shape are significantly different from those of the traditional soil-rock mixture models with polygonal rock blocks,under the conditions of higher rock content and larger rock block size.
万方数据
维普
