This article presents a stochastic modeling approach for simulating the mechanical behavior of pervious concrete, based on novel extensions of the lattice discrete particle model. Selected digital images of the internal mesostructure, obtained from physical specimens, are used to survey material features and produce statistically representative descriptions of the pore networks. A procedure for estimating the statistical features of the mesostructure is proposed, and samples of a spatially correlated random field are utilized to numerically reproduce the distribution of the large, connected pores in the material. The numerical samples are linked to the topology of the lattice network by means of two novel techniques proposed herein: (1) a random placement procedure for the poly-sized spheres representing coarse aggregate and (2) a ray tracing technique, which is used to evaluate the effective distributions of mass, stiffness, and strength of each lattice element according to the local distribution of porosity. Numerical results demonstrate how the proposed model is capable of simulating both the large scatter in strength and the variety of failure modes that are observed when testing physical specimens of pervious concrete. The proposed procedure is generally applicable to different types of materials with varying porosity, opening up new possibilities for the simulation of porous media by means of lattice discrete particle modeling techniques.

This article presents a novel multiscale modeling approach to simulate the evolution of the backward erosion piping (BEP) process in flood protection systems (FPSs). A multiphase description of the BEP phenomenon is proposed for the numerical solution at the local scale and validated by means of full-scale experimental results available in the literature. Results of the local scale simulations are used as the training set for a multilayer machine learning (ML) model to bridge the information between the local and system scales. Accuracy of the trained ML algorithms is demonstrated by comparing results obtained from detailed physics-based numerical models. The novelty of the proposed methodology lies in its capability of real-time predictions of the overall response at the system scale. A case study is presented where a portion of the Nashville Metro Levee System is analyzed over the span of a year, to assess the likelihood of BEP in the infrastructure. The capability of the model to accept water height data obtained from field measurements is exploited in the numerical simulations.

Lattice discrete particle models (LDPM) effectively relate concrete heterogeneity in terms of coarse aggregate–paste structure to damage patterns and load-displacement response. Interest in this type of model is enhanced by its ability to overcome issues that plague traditional methods, such as mesh dependence and need for energy regularization. Herein, an experimental procedure is developed to identify the mechanical parameters defined at the concrete mesoscale. The bounding surface of the LDPM is modified to accept the mesoscale parameters as inputs. Based on the parameter values identified through the multiscale experimental campaign, the macroscopic response to compressive load is simulated. Accuracy of the approach is demonstrated through comparisons with experimental results.