Multi-Indicator Fusion-Based Unsupervised Statistical Method for Slope Deformation Anomaly Warning in Open-Pit Coal Mines

Abstract

Open-pit coal mine slope spalling events are rarely fully documented with sufficient samples, and the precision of publicly available monitoring data is inconsistent. To address these problems, this study develops a causality-oriented anomaly identification method that fuses four indicators: velocity, acceleration, persistent deviation, and phase change. Using the 25 May 2021 spalling event at the South Open-Pit Coal Mine (Xinjiang) as the main case, the analysis was based on 20 displacement measurements tabulated in the literature. Hourly replay was conducted using backward differencing and dual historical windows that contain only data already arrived at each time step, with baseline statistics frozen. Under the main setting, the fusion threshold was 1.351, and the first threshold exceedance occurred at 00:00 on the event day—9 h ahead of the field record; however, this time point involved a reversal of displacement direction and therefore cannot be directly interpreted as pre-failure acceleration. Among the 25 parameter sets tested, 6 triggered 15 h early, 17 triggered 9 h early, and 2 triggered 3 h after the event. For three externally digitized curves, the numbers of fused threshold exceedances were 1, 0, and 2, respectively; pixel-level perturbations could alter the triggering outcome, and exploratory phase-trajectory gating did not reduce exceedances on the periodic curve. The results support using the fused score for ranking historical anomalies and for local state re-examination, but are not yet sufficient to demonstrate warning reliability or a generalizable lead time on independent events. Engineering warning reliability still requires validation with independent raw monitoring data and complete event background records.