The Nanfen mine, located roughly 25 kilometers south of Benxi City, is a vast excavation: the pit stretches 3.41 kilometers north to south and 1.75 kilometers east to west, with a closed contour at 286 meters elevation and a pit bottom specified at 70 meters as of April 2025. The surrounding terrain is dominated by low mountains and hills, with natural relief of 300 to 400 meters and a maximum elevation of 963.7 meters. The east slope, the focus of the new analysis, sits above historical underground workings whose voids could interact with the open pit in ways that conventional two-dimensional stability charts cannot capture.

The stakes are not abstract. On 19 June 2021, extensive sliding and subsidence developed on the footwall slope, extending from an elevation of 710 meters near the crest down to 298 meters near the toe, a vertical span of roughly 410 meters. The deformation reached northward from a 2016 landslide area to exploration line 6, with an affected width of approximately 480 meters in the upper part and 550 meters in the lower part. It was the largest deformation event recorded during the mining period, and it underscored how quickly a slope under active excavation can transition from monitored stability to large-scale movement.

To model such a system realistically, the research team first rebuilt the mine in three dimensions. Surface topography was captured using unmanned aerial vehicle oblique photogrammetry and laser scanning, producing point clouds that were converted into a surface solid in Rhino, a three-dimensional modeling platform. Lithological interfaces were reconstructed from surveyed geological sections: contour lines for each rock type were matched between sections and lofted into continuous three-dimensional surfaces. The resulting model spans 1,500 meters by 900 meters in plan, with a nominal vertical extent of 800 meters that varies with the topographic profile, and it was built at a 1:1 scale with tolerances of 0.01 in absolute terms and 1 degree in angular terms.

Meshing such a heterogeneous model posed its own challenges. Boolean partitioning of the geology can leave duplicate surfaces along lithological contacts, and meshing without coplanar processing would create overlapping elements and misaligned nodes that compromise mesh continuity. The team made the duplicate surfaces coplanar so each contact was represented by a single interface, then generated a tetrahedral mesh with Rhino’s Griddle plug-in and imported it into COMSOL Multiphysics in BDF format. The final mesh contains 5,564,628 elements and 1,016,142 nodes, with a nominal element edge length of about 10 meters, fine enough to resolve both the slope geometry and the underground voids.

The mechanical heart of the study is an elasto-damage constitutive model, implemented in COMSOL with MATLAB-assisted customization. Rather than treating rock as a purely elastic-plastic material, the model introduces a scalar damage variable that degrades stiffness as microscopic defects initiate and coalesce under loading. Separate tensile and compressive strain thresholds govern damage initiation, and historical maximum strain variables record the largest equivalent strains each element has ever experienced. Because damage is irreversible, these historical maxima are inherited through unloading and subsequent excavation stages, so the calculation remembers every episode of deterioration even after the damaged rock itself is mined away. The maximum damage value is capped at 0.99, retaining one percent of the undamaged elastic stiffness to support numerical convergence.

Seven lithologies are represented in the model: mica schist, chlorite schist, green amphibolite, iron ore, mica quartz schist, quartz granulite, and granite, each with its own bulk modulus, shear modulus, cohesion, friction angle, tensile strength, and density. The east slope cutting area receives a separate, weaker parameter set, with a cohesion of 0.1 megapascals and a tensile strength of just 0.02 megapascals. These rock-mass parameters were supplied directly by the mine’s technical department, anchoring the simulation in site-specific engineering values rather than generic literature estimates.

The excavation sequence is what gives the study its unusual continuity. The team first established self-weight equilibrium with the underground void regions represented as solid rock, then excavated the voids and re-established equilibrium, resetting displacement to zero while retaining stress and damage histories. From that reference state, the model formed the actual 2024 open-pit geometry, then progressed continuously through the actual end-of-2025 configuration, which combined production excavation with slope cutting, and on to the planned end-of-2026 and end-of-2027 design geometries. At every stage, excavated elements were removed or deactivated without remeshing, and the retained material inherited stress, strain, displacement, damage, and historical strain states from the preceding step.

The results trace a striking evolution in slope behavior. In the 2024 baseline geometry, the maximum settlement magnitude on the slope surface reached 27.5 millimeters, concentrated in the middle and upper slope. By the actual end-of-2025 geometry, the mean settlement over the remaining slope surface was approximately 20 millimeters. The end-of-2026 design geometry showed a maximum below 7 millimeters and a mean near 5 millimeters, and by the end-of-2027 configuration the remaining slope surface exhibited sub-millimeter settlement magnitudes. The researchers caution that these statistics describe the slope surface retained at each stage, so the comparison reflects both changing mechanical response and the evolving sampled domain as material is removed.

Perhaps the most intriguing finding lies in the spatial organization of damage. Across all four configurations, the damage concentrations associated with the underground voids appear spatially separate from those associated with the slope itself, with an intervening region of lower displayed damage between them. In the end-of-2025 configuration, localized stress concentration appeared near the pit bottom as excavation approached the void roofs, a zone the authors flag for targeted monitoring and site-specific assessment of excavation increments and support. The team emphasizes that the separation remains a qualitative pattern observed in full-field plots with stage-specific color scales; quantifying the geometry and persistence of the intervening rock zone will require common-threshold connectivity analysis. They also note that independent field validation remains a priority because monitoring data were unavailable for this analysis, and that future work should incorporate groundwater, blasting vibration, long-term weathering, and mesh sensitivity studies. For now, the model offers mine planners a screening tool: a way to see, before the excavators arrive, where the rock beneath Nanfen’s slopes is quietly accumulating damage.

Subject of Research: Three-dimensional progressive damage and geohazard assessment of an open-pit mine slope affected by subsurface voids and staged excavation

Article Title: Three-dimensional progressive damage and geohazard assessment of an open-pit slope affected by subsurface voids and staged excavation: a case study of the Nanfen mine

Article References: Li, H., Gong, R., Jiao, S., Zhao, Y., Tian, Y., Yang, T., Deng, W., & Li, Y. (2026). Three-dimensional progressive damage and geohazard assessment of an open-pit slope affected by subsurface voids and staged excavation: a case study of the Nanfen mine. Frontiers in Earth Science, 14, Article 1964731. https://doi.org/10.3389/feart.2026.1964731

Image Credits: AI Generated

DOI: 10.3389/feart.2026.1964731

Keywords: open-pit mining, slope stability, underground voids, damage mechanics, 3D numerical modeling, UAV photogrammetry, Nanfen mine, geohazard assessment, excavation sequence, rock mechanics, landslide, COMSOL

The study, published in Earth Science Informatics, was led by Ke Li of the Chongqing AI Geological and Mineral Research Institute together with colleagues from the Chongqing Bureau of Geology and Minerals Exploration and partner institutions. Their starting point was a problem familiar to anyone who has worked in applied geoscience: the knowledge needed to assess a dangerous rock slope is scattered across heterogeneous documents, including field investigation reports, engineering records and monitoring archives, each organized differently and none designed for machine reading. This fragmentation complicates three things at once, the authors note: modeling the chains of events that turn a trigger such as a storm into a disaster, extracting reliable facts from messy text, and retrieving both specific entity-level details and broader mechanism-level context when a question is asked.

To solve it, the team built what they call an ontology-guided LightRAG framework, a system with three interlocking components. The first is a domain ontology, a formal specification of the concepts and relationships that matter in precarious rock hazard work. Ontologies have a long pedigree in geoinformatics, tracing back to foundational work on semantic web terminology for earth and environmental science, and they serve as a shared vocabulary that both humans and machines can use. The researchers’ ontology contains 31 entity categories organized across five core evolutionary stages: environmental background, triggering factors, disaster evolutionary processes, mitigation decisions and effectiveness evaluation. In effect, it encodes how geologists think about a rockfall hazard from first causes to final remediation.

The second component is a four-stage conversational extraction process driven by GeoGPT, an open-source geological language model fine-tuned from Qwen2.5-72B-Instruct by its original developers and deployed locally by the team without further modification. In the first stage, the system performs an ontology-aware semantic scan of a document, using a sliding window to pull in surrounding paragraphs so that cross-sentence context is preserved. In the second stage, it anchors entities and resolves co-references, mapping phrases such as ‘this rock mass’ or a shorthand label like ‘W01’ to a single unique identifier. The third stage involves recursive attribute backfilling, in which the model hunts through the text for missing physical parameters such as volumes, fracture orientations and stability coefficients, and the fourth applies logical verification, checking that numbers and qualitative descriptions agree before stripping away redundant modifiers and emitting structured data.

The authors illustrate the pipeline with a real example from Chaqi Mountain, where a precarious rock unit designated W01 sits on a second-tier cliff above a school housing roughly 4,000 students and staff. From a single passage, the system extracts the unit’s irregular geometry and volume of about 66.4 cubic meters, its dominant sliding-falling failure mode, the controlling fracture set with its attitude expressed as strike and dip angles, the weak Lower Jurassic mudstone pedestal that softens and forms cavities under weathering, and three stability coefficients, 8.62 under natural conditions, 2.00 under storm conditions and 5.48 under seismic conditions, all confirming a currently stable state. That kind of multi-fact extraction, performed consistently across an entire archive, is precisely what manual digitization struggles to deliver.

The third component is the retrieval architecture. The team adopted LightRAG, a graph-based retrieval-augmented generation approach, and paired it with a hybrid document retrieval strategy. Questions and document chunks are encoded with the BGE-M3 embedding model into 1,024-dimensional dense vectors, while a classic BM25 keyword retriever runs in parallel; the two ranked lists are merged using Reciprocal Rank Fusion with a constant of 60, and the top ten chunks are retained. On the graph side, entities in a question are linked to knowledge-graph nodes by exact, alias or semantic matching, then the system traverses incoming and outgoing relations up to two hops, keeping only simple paths without repeated nodes. The ten best paths, ranked by cosine similarity between question and path embeddings, are serialized as head-relation-tail triplets and handed to the language model alongside the document context. This dual retrieval lets the system answer both pinpoint factual queries and broader questions about failure mechanisms.

The performance numbers are striking. With the complete extraction configuration, the system achieved a precision of 89.00 percent, a recall of 87.74 percent and an F1-score of 88.36 percent against a ground-truth dataset annotated by five professor-level senior and senior engineers, each with more than ten years of experience in engineering geology. Ablation analysis showed that the four-stage conversational extraction process was the primary driver of that performance, with GeoGPT itself contributing a smaller additional benefit under the same extraction setting. The annotation protocol itself was rigorous: two experts independently labeled 1,200 text chunks, from which 212 entities were drawn, reaching an initial agreement rate of 91.8 percent and a Cohen’s Kappa of 0.894, with a third senior expert adjudicating all disputes. Most disagreements, 73.5 percent of them, clustered at the boundary between environmental background and triggering factors, a reminder that even expert judgment finds some geological distinctions genuinely ambiguous.

For question answering, the team ran a blinded pairwise evaluation across four precarious rock task categories, comparing GeoGPT against two powerful general-purpose models, DeepSeek-V3 accessed through its official API and Qwen2.5-72B-Instruct deployed locally on eight A100 GPUs with BF16 precision and greedy decoding at temperature zero. The GeoGPT-based system achieved higher overall pairwise win rates than both competitors across all four categories, with the largest margins appearing in the most demanding tasks: stability assessment and mitigation planning. A qualitative example shows why. Asked to analyze collapse mechanisms, DeepSeek-V3 assembled a competent descriptive summary of physical parameters, while GeoGPT structured its answer along explicit geological causal chains, incorporating formal failure mode classifications such as sliding, toppling and composite modes, and even geomechanical expressions like the toppling coefficient equation, aligning its output with domain standards for hazard assessment.

The implications reach well beyond one mountain range. Landslides and rockfalls are among the deadliest geological hazards worldwide, and climate change, slope urbanization and aging infrastructure are expected to increase exposure in many regions. Knowledge graphs have already proven their value in geoscience for link prediction and disaster cause analysis, and large language models are being applied to everything from landslide image analysis to earthquake fatality estimation. What this study adds is a template for combining them responsibly: an ontology to constrain what the machine can extract, a staged conversational process to keep extraction faithful to the source text, and graph-aware retrieval to ground answers in verified facts rather than the model’s own tendencies toward hallucination, a known weakness of generative systems.

There are honest limits. The datasets analyzed in the study cannot be released publicly because of multi-party data-sharing agreements and confidentiality restrictions involving the regional geological survey institutions of the Three Gorges Project, though the authors say data are available upon reasonable request with formal permission from the competent geological authorities. The evaluation also rests on a specific corpus and a specific panel of expert annotators, so the reported win rates describe performance within that setting rather than a universal guarantee. Even so, the framework demonstrates something quietly transformative: that the tacit expertise embedded in thousands of pages of engineering reports can be formalized, verified and made conversational. For the geologists watching rock slopes above schools and highways, an assistant that answers mechanism-level questions in seconds, grounded in their own field’s ontology, could change how hazard assessments get done.

Subject of Research: An ontology-guided LightRAG framework for knowledge-graph construction and question answering on precarious rock hazards

Article Title: An ontology-guided LightRAG framework for knowledge-graph construction and question answering on precarious rock hazards

Article References: Li, K., Pan, Y., Yi, Z., Guo, S., Liu, P., Dong, L., Wang, L., Wang, Y., Qin, W., Cheng, X., She, N., & Yuan, J. (2026). An ontology-guided LightRAG framework for knowledge-graph construction and question answering on precarious rock hazards. Earth Science Informatics, 19(11), Article 202. https://doi.org/10.1007/s12145-026-02258-9

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02258-9

Keywords: precarious rock masses, knowledge graph, ontology, LightRAG, retrieval-augmented generation, GeoGPT, large language models, geohazard assessment, Three Gorges, information extraction, question answering, landslide hazards