Research

I lead research initiatives for the multi-scale characterization of in-placed infrastructure materials (i.e., asphalt, concrete, and soil) and subsequently evaluate the built system under extreme dynamic loading, climatic and hazardous conditions. With my experience in artificial intelligence (AI), geospatial technologies, and large-scale non-destructive technologies, I intend to develop an entirely new approach to designing, constructing, and managing infrastructure assets.

1. Data-driven AI and machine learning models for the multi-scale characterization of infrastructure materials

Unbound base/subgrade materials or soil beneath the ground surface are characterized by the resilient modulus (M_R), which directly affects the design and analysis of built structures. It is established that the degree of saturation or matric suction has a significant impact on the M_R value of the unbound material, indicating that the resilient behavior of the unbound material is stress-dependent and moisture-dependent. The relationship between soil suction and moisture content (or degree of saturation) is represented by the soil-water characteristics curve (SWCC). Most of the existing approaches, i.e., experimental methods and correlation models to determine the SWCC and M_R of unbound materials, are very time-consuming or have poor prediction accuracy. Hence, we developed an ANN model to predict the coefficients of MR and SWCC models from soil properties and climatic indicators. The main advantage of a neural network model over nonlinear regression models is that it can capture the complex nonlinear scattered relationship between input and output parameters and train the model based on evaluating the error function. Two three-layered ANN models, one for plastic and another for non-plastic soils, were constructed as shown in the Figures below.

2. Integration of multi-physical modeling approach with geospatial technologies for infrastructure asset management

With evolving challenges in the infrastructure sector and limited budget funds, there is an inherent need to change the infrastructure asset management system from an empirical and experimental approach to a digitized system. We generated a geographic information system (GIS)-based Thornthwaite Moisture Index (TMI) contour map based on the precipitation and temperature metadata from the National Climatic database. Based on this work, we later proposed a micro-scale suction energy-based mechanistic model to predict the long-term equilibrium suction and potential vertical movement (PVM) in fine-grained underlying soil considering the effects of physical properties and climatic factors. A digital contour map of the PVM for the entire continental United States (U.S.) is generated using the GIS platform to accurately determine the baseline equilibrium suction and corresponding unsaturated soil properties at any given latitude and longitude. The generated TMI and PVM contour map of the continental U.S. is shown in the Figures below. The developed GIS map will serve to evaluate the differential ground movement under roads, buildings, and other light infrastructures and thereby play a significant role in minimizing the massive threat to infrastructures against natural disasters such as extreme precipitation, hurricane, floods, and droughts in the U.S.

3. Use of rapid, nondestructive test technologies to evaluate the construction quality

Attainment of specification and material properties during the construction and identifying localized defects afterward serves to maximize infrastructure life and minimize life-cycle costs. We developed a ground penetrating radar (GPR)-based efficient nondestructive testing (NDT) approach to evaluate the quality of the flexible base layer in terms of the estimated real-time resilient modulus and stabilizer content profiles. The figure below illustrates the procedures for implementing the mechanistic-based NDT approach for field sites. The GPR outputs are analyzed by the software PaveSCM to obtain the dielectric constant profile of the flexible base. The PaveSCM software uses a Self-Consistent Micro-mechanics model to estimate the various layer properties. Finally, laboratory test results are incorporated with the dielectric constant, and the electrical conductivity profile data are input into the software LayerMAPP to estimate the resilient modulus and stabilizer content of the in-situ flexible base.