University of Texas at Austin


Using Digital Twins of Civil Engineering Structures, Researchers Predict and Respond to Operational Failures

By Aira Balasubramanian

Published Feb. 22, 2024

Predictive digital twin framework for civil engineering structures.

Life would look drastically different without the bridges, buildings, and beams that surround us. The products of civil engineering form the framework of our lives, but failures in these systems, though rare, can be catastrophic. 

Now, imagine a world in which we could predict these accidents before they happen, and evaluate the health and safety of the structures around us in real time. Enter the digital twin - a virtual model that turns real time data from unique physical structures into targeted safety and maintenance recommendations for engineers. Though it may seem reminiscent of science fiction, this world of optimized maintenance, personalization, and risk prevention is far from an impossible utopia. 

New research from the Oden Institue of Computational Sciences and Engineering at The University of Texas at Austin, is pushing this dream one step closer to reality. Published in Computer Methods in Applied Mechanics and Engineering, “A Digital Twin Framework for Civil Engineering Structures” explores how digital twins (also used in personalized cancer care and diagnosis) can be applied in the context of civil engineering. 

Led by Willcox Group postdoctoral fellow Marco Tezzele and visiting Ph.D student Matteo Torzoni, with support from Andrea Manzoni, Stefano Mariani, and Oden Institute Director Karen Willcox, the recent paper offers a holistic, advanced approach to structural health monitoring, replacing periodic inspections with real time data that can generate up-to-date, specific predictions and recommendations that enhance the safety of civil structures. 

Unlike reliability based approaches, our framework facilitates the comparison of corrective courses of action and their potential consequences.

— Marco Tezzele

To make diagnostic predictions, the digital twin gathers sensor data in real time from its physical replica, which are assimilated using deep learning (DL) models that estimate the location and severity of structural damages. These evaluations are made based on training data for the deep learning model, which are optimized to represent the potential damages and operational conditions that a structure will be exposed to in its lifetime. This allows the digital twin to make predictions based on specific environmental and usage conditions unique to the physical structure, rather than relying on generalized risk-management techniques. 

Predictions based on the DL model are made through a probabilistic graphical model, formally known as a dynamic Bayesian network. This predictive model sets this research apart, by integrating uncertainties from both the sensor data and the predicted response of the structure to interventions. The model thereby facilitates a bidirectional flow of information between the physical structure and the digital twin, enhancing the accuracy with which we are able to maintain the structural health of civil engineering products. 

The digital twin model explicitly accounts for the costs of failure events that cause a physical structure to partially or totally lose function, as well as the impact of maintenance tasks and asset management that impact its physical state, overall performance, and safety. “Unlike reliability based approaches, our framework facilitates the comparison of corrective courses of action and their potential consequences,” said Tezzele. This allows stakeholders to “implement control policies that maximize utility by minimizing life cycle cost.” The digital twin model not only enhances the safety of physical structures, but also carries critical economic incentives in terms of maintenance and operation. 

The researchers assessed their model using two distinct case studies of physical systems in conjunction with a digital twin: a cantilever beam - often used to support balconies, roofs, and overhangs; and a railway bridge. Their results show that the digital twins of these structures are capable of tracking digital state evolution in varying environmental and operational conditions, and can preemptively prescribe reparative control actions in order to maintain the viability and safety of the physical structures. 

Tezzele hopes that this paper ‘contributes to pushing forward the research on digital twins and advancing the state-of-the-art for the predictive maintenance of civil structures.’ As interest in digital twins has grown, the interest in applying the technology to specific fields has enhanced. 

As the researchers work to further enhance their methodology, the impact on civil engineering maintenance and safety could be revolutionary. Billions of people interact with civil structures on a daily basis - predictive, real time assessments of safety and maintenance dually enhance efficiency and prevent catastrophe. 


Marco Tezzele