In civil engineering, the stability of many structures depends on the physical properties of the contact between the concrete seated on the rocky foundation surface. Usually, they are critical structures of great importance to society, such as: bridges, tunnels and dams. Thus, the shear behavior of rock-concrete joints is a key factor of the structural stability. The method of cohesive zones (CZM) allows to simulate the beginning of the formation of a crack and its propagation, without knowing the crack location or when it will start. Thus, in this work, a numerical model was developed capable of simulating the failure due to interfacial delamination of the contact between a structure formed by mortar seated on a rocky granite leaning surface. The calibration of the numerical model was performed using the experimental test results of simple compression on a block of rock in contact with the mortar by an inclined interface. The test was monitored using a high-resolution digital camera. The data were processed using the digital image correlation method (DIC) to obtain displacement data of the laboratory test. The DIC results were used to calibrate the nonlinear numerical model of the test using the bilinear method of cohesive zone in the rock-mortar interfacial contact elements. By a parametric analysis varying the maximum shear stress parameter of the contact cohesive elements, the bilinear law was adjusted (Maximum tensile stress and critical displacement and damage rate). Indirectly, the rate of release of energy of critical deformation from the rupture of the contact was also estimated.

The Belo Monte Hydroelectric Complex, located on the Xingu River in Altamira, Pará, Brazil, includes two powerplants: Pimental, built directly on the riverbed, and Belo Monte, which uses water diverted from an intermediate reservoir. This reservoir is contained by 28 dikes, mostly underlain by migmatite residual soils. During construction and early operation, tubular cavities known as canaliculi and associated resurgence processes were identified downstream of several dikes. However, the spatial controls governing these phenomena remain poorly understood in tropical lateritic environments. This study proposes an integrated geostatistical–machine learning framework for the spatial prediction of resurgence and canaliculi. Ordinary Kriging was applied to model the elevation of the Young Residual Soil (YRS) contact surface using approximately 450 borehole records. The model was subsequently extrapolated to downstream areas lacking direct subsurface data through Support Vector Machine (SVM) regression. A digital terrain model was combined with the modeled YRS contact to identify potential outcrop zones associated with resurgence. Spatial statistical indicators based on minimum-distance metrics were developed to quantify the correspondence between predicted outcrop areas and fieldmapped occurrences. The results show strong spatial agreement between predicted YRS outcrop zones and mapped resurgence and canaliculi points across five representative dikes, with average shortest distance ranging from 4.4 to 50.6 m. Cross-validation indicated satisfactory predictive performance of the YRS digital model, with RMSE values between 4.16 and 9.56 m for kriging and between 3.36 and 10.65 m for SVM. The proposed framework provides a replicable and cost-effective predictive tool for dam safety management, supporting the definition of priority inspection corridors and early detection of anomalous resurgence and canaliculi behavior in tropical dam foundations.