توسعه نمودار تجربي به منظور ارزيابي پايداري شيب تحت شكست نوع دايره اي

Slope stability is one of the fundamental issues in geotechnical an‎d mining engineering. Classical approaches in this field, such as limit equilibrium methods an‎d numerical methods, often require simplifying assumptions an‎d computationally expensive analyses. The aim of this study is to transform the knowledge embedded in data-driven models into a simple, explicit, an‎d reliable stability chart for rapid slope stability assessment. To this end, a database comprising 176 cleaned samples was compiled. Each sample was characterized by six input parameters affecting slope stability: slope height, slope angle, unit weight, cohesion, internal friction angle, an‎d pore water pressure ratio. These six parameters were reduced to two principal components using Principal Component Analysis (PCA). Together, these two principal components explain approximately 60% of the total variance in the dataset, enabling the entire database to be represented in a two-dimensional plane. Subsequently, seven different approaches were employed on this plane to extract the boundary separating stable an‎d unstable regions an‎d were compared with one another. The first six approaches consisted of Support Vector Machine (SVM) models with three kernels—linear, polynomial, an‎d sigmoid—each eva‎luated both in its default configuration an‎d after hyperparameter tuning using the Particle Swarm Optimization (PSO) algorithm. The seventh approach was a graphical boundary derived from the continuous stability index surface obtained through Kriging interpolation. The results showed that the performance of each kernel was highly dependent on hyperparameter tuning. In particular, the sigmoid kernel, which exhibited the weakest performance in its default configuration (test accuracy of 44.44%), became one of the two best-performing models after optimization using PSO. The optimized sigmoid Support Vector Machine an‎d the graphical boundary derived from the Kriging surface achieved the highest performance, with a test accuracy of 83.33%. In addition to its predictive accuracy, the continuous Kriging surface offers the advantage of representing the relative degree of stability at each point rather than providing merely a binary response. Overall, the findings demonstrate that integrating data-driven methods with graphical representation can lead to a practical tool that benefits from the predictive capability of machine learning models while maintaining the simplicity an‎d transparency of conventional design charts.

ابراهيم قاسمي ورنوسفادراني

محمود بهنيا , محمدحسين كدخدايي

حسن طباطبائي , امين ازهري