Field-Constrained Screening of High-Displacement Scenarios in Deep Goaf Groups Using Latin Hypercube Sampling (LHS)-FLAC3D and Static Bayesian Inference

Deep metal mines commonly contain vertically stacked goafs whose geometry and rock mass properties are incompletely documented. This study evaluates a field-constrained screening framework for high-displacement material scenarios at the Lehong Pb-Zn mine. The framework combines a Latin hypercube sampling (LHS)-FLAC3D response library with static Bayesian inference. Evidence comprised 93 goaf records, 186 Mathews exposed-surface assessments, laboratory constraints, and 40 archived numerical scenarios, whose maximum downward displacement ranged from 4.48 to 31.75 cm. Friction angle φ showed the strongest marginal Pearson correlation with displacement (r = −0.81), followed by cohesion (r = −0.55) and elastic modulus (r = −0.22). At the response library Q75 threshold of 15.04 cm, the Laplace-smoothed probability increased from 0.262 (95% credible interval, 0.142–0.403) across all scenarios to 0.600 (0.352–0.824) under joint cohesion–friction angle degradation. However, the archived design was not an ideal 40-point LHS, and bootstrap resampling retained the scenario ordering in only 45.8–54.8% of replicates. All 93 inventory identifiers matched the Mathews stability table, enabling reproducible site-level triage when Bayesian network results are combined with treatment and stability evidence. The framework is an exploratory screening tool rather than an absolute failure probability model, collapse propagation model, or dynamic early warning system.