A New Method for System Reliability Analysis of Tailings Dam Stability

For the purpose of stability evaluation, a tailings dam can be considered as an artificial slope made of special soil materials which mainly come from mine tailings. As a particular engineering project, a tailings dam generally has experienced multi-loop hydraulic sediments as well as a long-term consolidation in the process of construction. The characteristics of sediment and consolidation result in a unique distribution of the soil layers with significant uncertainties, which come from both nature development and various human activities, and thus cause the discrete and the variability of the physical-mechanical properties dramatically greater than the natural geo-materials. Therefore, the location of critical slip surface (CSS) of the dam usually presents a notable drift. So, it means that the reliability evaluation task for a tailings dam is a system reliability problem indeed. Unfortunately, the previous research of reliability of tailings dam was mainly confined to the limit equilibrium method (LEM), which has three obvious drawbacks. First, it just focused on the variability along the slip surface rather than the whole space of the dam. Second, a fixed CSS, instead of variable one, was concerned in most cases. Third, the shape of the CSS was usually simplified to a circular. The present paper tried to construct a new reliability analysis model combined with several advanced techniques involving finite difference method (FDM), Monte Carlo simulation (MCS), support vector machine (SVM) and particle swarm optimization (PSO). The new framework was consisted of four modules. The first one is the limit equilibrium finite difference mode, which employed the FLAC3D code to generate stress fields and then used PSO algorithm to search the location of CSS and corresponding minimum factor of safety (FOS). The most value of this module was that each realization of stress field would lead to a particular CSS and its FOS. In other words, the consideration of the drift of CSS was significant to system reliability analysis. Besides, the CSS was described with non-circular shape, which was much more superior to the circular one. Moreover, compare with the LEM, the numerical method has the ability to reflect the variability in a whole space of the dam. The second one is the orthogonal design modules aimed to generate high-quality training samples according to the variability of soil layers. The third one is the response surface model based on SVM, which was designed to obtain the explicit performance function through the well-prepared training samples. The fourth one is the MCS model, which can perform steady and effective reliability analysis through the explicit performance function. As a result, a new approach for reliability analysis of tailings dam stability was presented. As a case study, the Yong-ping Copper Mine Tailings Dam in Jiangxi Province of China was analyzed in detail by this new method. It was shown that the effectiveness of the new method is considerable. In sum, this research is geared towards providing new ideas and available examples for future reliability assessment of tailings dam stability.