A rapid kinematic analysis method considering the spatial proximity and rock joint connectivity using non-contact measurement

Traditional kinematic analysis methods are limited in representing fragmented discontinuities, variable slope geometries, and spatial relationships between joint surfaces. This study presents a point-cloud-based kinematic analysis framework integrating a shallow artificial neural network, DBSCAN clustering, PCA-based orientation calculation, co-planar reconstruction, and spatial proximity analysis. These components enable direct, integrated processing from point-cloud acquisition to kinematic assessment. Rock joints are identified from normal vectors and curvatures, and fragmented patches are connected using distance-based clustering. Equivalent trace lengths visualize the reconstructed structures. The framework evaluates planar, wedge, and toppling failure criteria while considering actual spatial intersections, and iteratively determines the maximum kinematically admissible excavation angle. Applications to three field datasets show that calculated joint orientations agree with manual and field measurements within 5°. The proposed framework reduces processing time by 63.5% compared with the conventional workflow. Comparison with Dips produces the same primary kinematic conclusion, while spatial filtering reduces theoretical intersection lines from 12,403 to 36 in Case C. The framework identifies joints and locations potentially associated with structurally controlled failure, but direct localization of complete unstable rock bodies requires further integration of linear rock joints, hidden internal joint networks, site-specific physical parameters, and field-based evaluation.