Influence of matrix cementation and soft rock blocks on the mechanical behavior of soil-rock mixture: insights from triaxial experiments and interpretable machine learning

Soil-rock mixture (SRM) is a complex geological material widely distributed in nature, and its mechanical properties are typically controlled by the block skeleton. However, matrix cementation and soft rock blocks can significantly alter the properties of this skeleton, and the controlling factors of SRM mechanical behavior under such conditions remain unclear. To address this, this study combines laboratory triaxial tests, theoretical analysis, and integrated machine learning methods to systematically reveal the mechanisms by which matrix cementation and soft rock blocks influence the mechanical response of SRM, while further exploring nonlinear strength characterization and intelligent prediction of full stress-strain curves. The results indicate that: (1) The cementation effect exhibits significant sensitivity to confining pressure. (2) Rock block strength exhibits a dual influence mechanism. Its effect on the internal friction angle is positively correlated in cemented soil-rock mixture (CSRM), but negatively correlated in uncemented SRM due to stiffness mismatch. (3) Based on the quantitative relationship between particle breakage and macroscopic nonlinear strength parameter, this study proposed an optimized nonlinear strength criterion. (4) This study proposed an integrated machine learning framework combining a multilayer perceptron (MLP) and an autoencoder (AE). This framework not only achieves accurate predictions from experimental conditions to the full stress-strain curve, but also successfully quantifies the mechanical sensitivity of experimental conditions by introducing the Shapley Additive Explanations (SHAP). These findings provide theoretical support and a universal method to understand the strength mechanisms of complex geological materials, accurately predict mechanical parameters, and uncover physical laws.