关键词: 尾矿坝, HVSR, 共振频率, 结构反演

Abstract:

China has the largest number of tailings ponds in the world. Failure of tailings dams can cause severe ecological damage, economic losses, and casualties in downstream areas. Conventional monitoring approaches, including surface deformation measurements such as GPS and InSAR, as well as borehole-based instruments such as pore-pressure sensors and deep displacement meters, are limited by high costs, sensor vulnerability, and their limited ability to directly characterize the internal mechanical properties of dam structures. Therefore, there is an increasing need for damage-sensitive and cost-effective monitoring techniques that can directly evaluate the mechanical properties of tailings dams during operation. In recent years, seismic ambient-noise-based methods have attracted growing attention in subsurface structure investigation. Among these methods, the characteristic resonant frequency derived from the horizontal-to-vertical spectral ratio(HVSR) of seismic ambient noise is sensitive to variations in subsurface velocity and impedance, making it particularly suitable for structural health monitoring of tailings dams.

In this study, we deployed three seismometers at different elevations of the Xiaodonggou tailings dam in Chengde, China, to analyze HVSR characteristics and infer the internal layering of the dam. Full-day seismic records were divided into 72 non-overlapping 20 min windows, and the Konno-Ohmachi algorithm was applied to smooth the HVSR curves. Steady-state data were then selected to obtain representative HVSR curves for each station. Furthermore, an evolutionary inversion algorithm based on covariance matrix adaptation evolution strategy (CMA-ES) was employed to invert the HVSR curves and characterize the layered structure of the tailings dam. The reliability of CMA-ES was evaluated by comparing its inversion results with those obtained using particle swarm optimization(PSO), grid search, and the open-source software HVSRInv.

The results show that the peak frequencies of the HVSR curves at the three stations increase from 1.4 to 3.7Hz from the crest to the toe of the tailings dam, indicating the presence of a well-defined stratified interface with substantial lateral variation in depth. The HVSR curve at the crest exhibits the largest standard deviation, likely due to interference from discharge activities, highlighting the influence of external disturbances on monitoring results. The CMA-ES inversion results indicate a strong S-wave velocity impedance contrast between unsaturated and saturated tailings sand. Moreover, the depth of the upper unsaturated tailings layer inferred from the inversion agrees best with the phreatic surface measured in the borehole profile at the dam crest. Compared with PSO and grid search, CMA-ES produces the smallest interface-depth error, with the interface depth corresponding to the HVSR peak frequency being closest to the phreatic-line depth observed in borehole data. In addition, compared with the open-source HVSRInv software, the CMA-ES inversion yields a better fit to the observed HVSR curves, further supporting its accuracy and reliability for structural inversion of tailings dams.

These findings demonstrate the strong potential of HVSR-based methods for structural health monitoring of tailings dams and confirm the reliability of CMA-ES for HVSR inversion. This study provides a scientific basis for assessing tailings-dam stability and mitigating potential geohazards. Future work should focus on improving the anti-interference capability of the HVSR method and conducting long-term observations to evaluate its applicability and stability under varying environmental conditions.

Key words: tailings dam, HVSR, resonance frequency, structure inversion