Si² Lab
Our research
-
Large language models (LLMs) for structural analysis
-
Digital twin (DT) of infrastructure systems
-
Life-cycle management
-
Uncertainty quantification and optimization
Large language models (LLMs) for structural analysis
We develop LLMs and AI agents that can reliably construct finite element models to perform structural analysis. You provide a natural language description of the structures, our LLM will do the analysis for you. With domain knowledge integrated, our LLMs significantly outperform commercial LLMs including Gemini and ChatGPT.
Digital twin (DT) of infrastructure systems
We focus on developing system digital twins for large-scale infrastructure systems. We have developed computational frameworks and algorithms to address the computational issues when we scale up the digital twins. As an example, a system digital twin is established to monitor bridge network risk using condition rating, river, and traffic data. The system digital twin covers Miami-Dade County with >1000 bridges and >5000 data sources.
Life-cycle management of bridges and ships
Life-cycle management is used to cost-effectively ensure acceptable levels of safety and serviceability of civil infrastructure. Our group has investigated time-based and condition-based maintenance strategies. Recently, we are more interested in adaptive maintenance strategies that are solved using deep reinforcement learning.
Uncertainty quantification and optimization
Our group focuses on how to efficiently perform uncertainty quantification and optimization through knowledge transfer. We developed meta-learning-based surrogate modeling (MLSM) framework. By utilizing surrogate models from prior similar tasks, MLSM can quickly establish a surrogate model for a new task. Our group also relies on advanced algorithms from uncertainty quantification and optimization to support reliability analysis and life-cycle management.
