Link to Report: Coming Soon
Background :
The proposed study aims to create an artificial intelligence (AI)-based framework that harness data from advanced sensing technologies to advance the next-generation of transportation asset management systems. The development of AI-driven diagnostic tools for infrastructure is fundamental to optimizing maintenance plans, thereby enhancing public safety and minimizing the inefficient use of economic resources.
To address the limitations associated with the availability of training data for deep learning algorithms, the proposed framework integrates heterogeneous data collected from multiple sources. Beyond the deployment of conventional fixed sensors, cutting-edge mobile sensing technologies will be incorporated to achieve unprecedented temporal and spatial resolution, thereby ensuring the scalability and adaptability of the AI-based strategy. The dynamic characteristics extracted from acceleration data acquired through smartphones will be utilized to: (i) calibrate the digital twin of the structure, (ii) identify and characterize potential structural damage, and (iii) provide essential physics-based knowledge to support the development of a physics-informed neural network for life-cycle assessment.
Objectives :
The objectives of this research are the following:
- Develop an AI-based agent capable of converting data from multiple sources into valuable structural information for damage identification and life-cycle assessment.
- Demonstrate a proof of concept for a physics-informed AI system applied to bridge monitoring.
Scope :
Task 1 – Data Collection
Data will be collected from multiple sources, as illustrated in Figure 1A. This project will focus on the use of:
- Mobile sensors – specifically, smartphones will be used to collect acceleration data.
- Fixed sensors – these will record the dynamic structural response, including acceleration and strain measurements.
Task 2 – Dynamic Identification and Digital Twin Model Updating
Recent studies have demonstrated that acceleration data acquired from smartphone represent a novel and reliable resource for extracting the dynamic characteristics of infrastructure. Building on its expertise in this emerging field, the research team will leverage smartphone-based data for bridge dynamic identification. The resulting dynamic characterization will then be used to calibrate virtual structural models through model updating techniques
Task 3 – Development of Deep-learning based Pattern Recognition System for Global Damage Detection
Bridge global damage-sensitive features (e.g., natural frequencies and mode shapes) extracted from system identification analysis of smartphone-based acceleration data will be used to train a deep learning algorithm for anomaly detection within a pattern recognition framework.
As demonstrated in the extensive datasets made available by mobile sensing technologies—characterized by their exceptional temporal and spatial resolution— enable the training of sophisticated deep learning models for the first level of the structural health monitoring hierarchy, namely damage identification. This deep learning–based global damage assessment will not only evaluate the current structural condition but also establish a reference baseline for future assessments within a statistical pattern recognition framework.
Task 4 – Development of Physics-Informed AI for Life-Cycle Assessment
A comprehensive understanding of the bridge’s dynamic behavior will pave the way for full structural prognosis. To this end, a physics-informed neural network will be developed for life-cycle assessment through the generation of virtual sensors. For life-cycle assessment, strain data are typically used to build rainflow diagrams and estimate the remaining structural life. However, the direct measurement of strain poses significant practical challenges. To overcome these limitations, this project will leverage artificial intelligence to create virtual sensors that infer strain data from acceleration measurements—an approach that is both more accessible and cost-effective.
Figure: Graph Representation: the generic nodes Vi and Vj are connected with the edge eij
Research Team :
Principal Investigator: Giulia Marasco, Ph.D.
Co-Principal Investigator: Atorod Azizinamini, Ph.D., P.E. and Arslan Khan, Ph.D.