Structural sustainability focuses on ensuring the long-term safety and functionality of infrastructure under evolving environmental and loading conditions. A key aspect is the seismic assessment of scoured bridges, which evaluates the vulnerability of bridge foundations weakened by soil erosion during earthquakes—a critical concern for infrastructure near rivers or coastal areas. Incorporating probabilistic seismic assessment enhances this process by quantifying the likelihood of various damage states under uncertain seismic events, offering a more robust framework for risk-informed decision-making. Furthermore, the presence of cavities with cracks within structural elements poses hidden threats to integrity, as they can propagate under cyclic loading and compromise overall stability. Addressing these interconnected challenges is essential for designing and maintaining resilient infrastructure capable of withstanding both natural hazards and long-term degradation.

We develop a mesoscale modeling approach that integrates microscopic theories with macroscopic phenomena to investigate the mechanical behavior of advanced materials, such as nanoglasses (with its composites) and high-entropy alloys. This multiscale strategy allows us to bridge atomic-scale mechanisms with continuum-level responses, providing deeper insights into material performance under various loading conditions. Nanoglasses, which are a class of amorphous materials with internal glass-glass interfaces, are modeled using the kinetic Monte Carlo method to capture the thermally activated atomic rearrangements that govern their mechanical behavior. On the other hand, high-entropy alloys, characterized by their multi-principal element composition and crystalline structure, are studied using the crystal plasticity finite element method. This technique enables us to simulate the evolution of slip systems and stress distributions during deformation, offering a comprehensive understanding of strength, ductility, and failure mechanisms in complex materials.

Analysis procedure for the strengthening mechanisms of the Cr-rich CoCrFeMnNi alloy using dislocation density-based CPFEM. 

Multiphysics simulations integrate various physical fields into unified computational models, enabling comprehensive analysis of complex real-world scenarios. In practical applications—such as manufacturing and medical treatments—thermal-mechanical coupling is often essential, including cases like fire exposure, welding, and thermal therapies. Our team has successfully employed multiphysics simulations to investigate the mechanical behavior of truss structures under fire, the performance of multi-layer ceramic capacitors, and the creep and fracture mechanics of high-temperature ceramics. We also explore biomedical applications, including high-intensity focused ultrasound (HIFU) for cancer treatment and thermal therapy for rehabilitation. Recently, we extended our research to include liquid cooling systems for computer servers.

Simulated temperature distributions within samples of (a) metal implant and (b) PE implant groups by continuous wave ultrasound heating for 300 s at 3 MHz, 12 W (or 2 W/cm2). 

Snapshots of simulation results of the map of the normalized stress along the loading direction, sII/E, at (a) t* = 0.004095, (b) t* = 0.004695, and (c) t* = 0.006895 for polycrystal with a heterogeneity located at the center of the model subjected to creep fracture. 

Deep learning has garnered significant attention in recent years for its powerful capabilities in solving complex, nonlinear problems. We have leveraged deep learning techniques to address various engineering challenges. For example, we developed models for structural health monitoring of shear frames, enabling early detection of potential damages. In bridge engineering, we applied deep learning to monitor scour depth, enhancing safety during floods. Our work also includes predicting stress-strain curves of short fiber-reinforced composites, which are difficult to model using traditional methods. Additionally, we used deep learning to estimate temperature distributions in ultrasound thermal therapy, improving treatment precision and patient outcomes.

Classification scheme. Data collected from a single sensor from the structure is converted into a matrix based on Takens’ embedding theorem. The matrix is then fed into a convolutional neural network for feature extraction and classification of the underlying structure’s level of structural damage and for determining the location of damage. 

The overall framework for FEM-ML, TGML, and TPML. The DNN models were trained by the input data of composite composition and the output data of normalized principal components. For FEM-ML, the composite composition is directly taken as input. For TGML, Ec was derived by Halpin-Tsai equation and served as one of the input parameters. For TPML, Ec was predicted by polynomial regression model and served as one of the input parameters. The performances of FEM-ML, TGML. TPML were compared to show the outperformance of the TGML and TPML approach.