Artificial Intelligence for Structural Simulation and Health Monitoring

Principal Investigator: Professor Jiaji WANG (Assistant Professor from Department of Civil Engineering)

This project is showcased in the third exhibition – Technology for Future.

About the scholar

Professor Jiaji WANG

Research interests:
• Physics-informed machine learning
• Data-driven metamodeling and Digital twin of engineering structure
• Intelligent structural design based on composite structures


Project information
To address challenges and future developments in structural engineering, Dr. Wang has developed a deep-learning-based platform, which contains (1) physics-informed neural networks (EPINN) for structural simulation and (2) state-of-the-art neural operator (VINO) for structural health monitoring.
AI for Structural Engineering: from Model-driven to Data-driven. Blue means model-driven and red manes data-driven approach. EPINN is more focused on model-driven approach. VINO (Vehicle-bridge Interaction Neural Operator) is more focused on data-driven approach.
Project video
Project images
Exact Dirichlet boundary Physics-informed Neural Networks (EPINN) to solve solid mechanics problems: (a) Using Convolutional Neural Networks, EPINN can exactly reformulate 1D global shape function of truss elements in finite element method; (b) Learning curves of 3D bracket under uniform load using EPINN (green) and conventional PINN (blue); (c) Finite element simulation result of displacement of bracket after 1140s calculation on CPU; (d) EPINN solution of vertical displacement of bracket after 685s calculation time on GPU.
Vehicle-bridge Interaction Neural Operator (VINO) for bridge health monitoring (BHM): (a) Field test of steel bridge before demolishing; (b) Detection of bridge damage using accelerators in drive-by vehicle; (c) Vehicle-bridge Interaction test with two damages: Damage 1 is inclined cut and Damage 2 are three slits cut; (d) Proposed Vehicle-bridge Interaction Neural Operator (VINO) to serve as digital twin of bridge structures; (e) Damage detection results based on VINO framework. The severe damage above 10% are well captured by VINO.
Other projects