Reliability-oriented Segmentation of Sublayers in Geologically Uncertain Substrate: A Case Study of the Żelazny Most TSF

In this study, a new approach is proposed for the stability analysis of tailings storage facility (TSF) dams. Given the considerable scale of such objects, the correct identification of geotechnical layers for computational models and the determination of reliable parameters can present significant challenges. Therefore, the investigation focuses on distinguishing additional sublayers within a geological substratum to enhance the credibility of stability assessment. The procedure of sublayer separation is directly based on soil parameters derived from geotechnical tests and it aims at reducing the variability in the parameters of individual sublayers. Specifically, the segmentation technique known as the Gaussian mixture model (GMM) is employed for this purpose. Subsequently, the spatial arrangement of sublayers within the finite element model domain is identified using the kriging technique. Utilizing Monte Carlo simulations statistical distribution of factor of safety (FOS) is obtained for the structure. Finally, the stability is assessed using the Cornell reliability index. The application of this developed methodology is demonstrated through a case study of the Żelazny Most TSF. Additional sublayers are identified in Neogene clays constituting the dominant layer beneath the facility. In total, six different segmentation variants are proposed and their performance is evaluated. The results show that the segmentation of additional sublayers produces a clear reduction in the scatter of the FOS. Consequently, the estimated simplified reliability index surpasses that of the case, which does not involve the separation of additional layers.