Intelligent scanning for optimal rock discontinuity sets considering multiple parameters based on manifold learning combined with UAV photogrammetry

Abstract

The Sichuan-Tibet railway, which spans many alpine canyon regions, is being built in southwestern China. Investigating the characteristics of rock discontinuity sets is the basis for identifying dangerous rock masses above the tunnel portals. The traditional methods of identifying discontinuity sets usually consider orientations and ignore other parameters, which results in incorrect guidance for rock engineering. To this end, the affinity propagation (AP) algorithm based on modified isometric mapping (Isomap) is proposed for partitioning discontinuity sets based on orientation, trace length, and aperture. The new unsupervised algorithm (ISOAP) uses manifold learning to complete the transformation process for orientations from spherical vectors to scalars and avoids selecting the initial clustering center to achieve global optimization. The Silhouette Index is used to intelligently scan the optimal clustering results. The proposed algorithm is tested on a complex artificial data set and on Shanley and Mahtab's data set. Since accurately obtaining discontinuity information is impossible by traditional means (i.e., using geological compasses and measurement tapes) due to the existence of a mass of high and steep slopes, the ISOAP algorithm is combined with semiautomatic technology based on unmanned aerial vehicle (UAV) photogrammetry and applied to a rock slope located along the railway. The introduction of manifold learning is beneficial for quickly applying abundant unmodified clustering algorithms to rock engineering and searching the optimal algorithm suitable for analyzing the structural characteristics of a specific fractured rock mass. The proposed method can simplify rock engineering analyses and provide more reasonable results.