Structure-aware Identification and Gaussian Process Regression-based 3D Reconstruction of Faults

This repository contains two complete Python implementations supporting the research titled Structure-aware Identification and Gaussian Process Regression-based 3D Reconstruction of Faults and Their Influence on Geostress Redistribution. The codes realize structural clustering of scattered fault point clouds and high-precision three-dimensional fault surface reconstruction based on Gaussian Process Regression (GPR). The first script adopts a geometry-aware DBSCAN clustering algorithm integrating spatial coordinates, surface normal vectors and plane offset features to automatically segment independent fault bodies from raw 3D scattered point data, filter noise and small invalid clusters, and output segmented fault point subsets with interactive 3D visualization via Plotly. The second workflow constructs a local UVW parametric coordinate frame using fault boundary constraints, fits a Matérn kernel exact GPR model to predict fault undulation offsets within the fault boundary, reconstructs continuous triangulated 3D fault surfaces, calculates core geometric attributes (fault surface area, central line length, relief distribution) and geological orientation parameters (strike, dip, dip direction), and evaluates reconstruction accuracy via hold-out RMSE metrics. All geometric results, prediction point clouds, boundary polylines, fault centerlines and quantitative evaluation indicators are exported as structured CSV/JSON files, with an interactive 3D visualization HTML report generated for post-processing geological analysis and geostress redistribution numerical simulation preprocessing. Dependencies include NumPy, Pandas, Scikit-learn, SciPy, Openpyxl and Plotly, tailored for underground engineering, tectonic geology and in-situ stress research.