Search for: All records

Sinkhole formation poses a significant geohazard in karst regions, where unpredictable subsurface erosion often necessitates costly grouting for stabilization. Accurate estimation of grout volume remains a persistent challenge due to spatial variability, site-specific conditions, and the limitations of traditional empirical methods. This study introduces a novel machine learning-based regression model for grout volume prediction that integrates cone penetration test (CPT)-derived Sinkhole Resistance Ratio (SRR) values, spatial correlations between CPT and grouting points (GPs), and field-recorded grout volumes from six sinkhole sites in Florida. Three data transformation methods, the Proximal Allocation Method (PAM), the Equitable Distribution Method (EDM), and the Threshold-based Equitable Distribution Method (TEDM), were applied to distribute grout influence across CPTs, with TEDM demonstrating superior predictive performance. Synthetic data augmentation using spline methodology further improved model robustness. A high-degree polynomial regression model, optimized with ridge regularization, achieved high accuracy (R2 = 0.95; PEV = 0.94) and significantly outperformed existing linear and logarithmic models. Results confirm that lower SRR values correlate with higher grout demand, and the proposed model reliably captures these nonlinear relationships. This research advances sinkhole remediation practice by providing a data-driven, accurate, and generalizable framework for grout volume estimation, enabling more efficient resource allocation and improved project outcomes. 

This work studies the feasibility of imaging a coupled fluid-solid system by using the elastodynamic and acoustic waves initiated from the top surface of a computational domain. A one-dimensional system, where a fluid layer is surrounded by two solid layers, is considered. The bottom solid layer is truncated by using a wave-absorbing boundary condition (WABC). The wave responses are measured on a sensor located on the top surface, and the measured signal contains information about the underlying physical system. By using the measured wave responses, the elastic moduli of the solid layers and the depths of the interfaces between the solid and fluid layers are identified. To this end, a multi-level Genetic Algorithm (GA) combined with a frequency- continuation scheme to invert for the values of sought-for parameters is employed. The numerical results show the following findings. First, the depths of solid-fluid interfaces and elastic moduli can be reconstructed by the presented method. Second, the frequency-continuation scheme improves the convergence of the estimated values of parameters toward their targeted values. Lastly, a preliminary inversion, using an all- solid model, can be employed to identify if a fluid layer is presented in the model by showing one layer with a very large value of Young's modulus (with a similar value to that of the bulk modulus of water) and the value of mass density being similar to that of water. Then, the primary GA inversion method, based on a fluid-solid model, can be utilized to adjust the soil characteristics and fine-tune the locations of the fluid layer. If this work is extended to a 3D setting, it can be instrumental to finding unknown locations of fluid-filled voids in geological formations that can lead to ground instability and/or collapse (e.g., natural/anthropogenic sinkhole, urban cave-in subsidence, etc.).