Rodrigo Sarlo receives National Science Foundation CAREER award to create structural digital twins on campus

Rodrigo Sarlo, assistant professor in the Charles E. Via, Jr. Department of Civil and Environmental Engineering has been awarded a $674,582 National Science Foundation Faculty Early Career Development (CAREER) award to drive the digitization of the civil engineering industry. His research will facilitate the conversion of sensor data to structural models, which has applications in structural monitoring for safety and maintenance.

The CAREER award is the most prestigious award given by the National Science Foundation for early career faculty, encouraging them to serve as academic role models in research and education and to lead advances in the mission of their organizations. Sarlo will use the award to partner with Virginia Tech's Division of Campus Planning, Infrastructure, and Facilities to create a living laboratory on the Blacksburg campus for students and researchers to explore new and emerging technologies. 

The big data paradox

Civil engineers now have an unprecedented ability to measure large volumes of data on the built environment. One example is 3D geometric models from drone images, which are used in architecture and construction to take measurements, create maps, and visualizations. These technologies have not revolutionized the industry as many predicted they would. In practice, “big data” takes up a lot of space, and engineers are often unsure about how to make the best use of it.

Why? Sarlo believes it is because data does not necessarily equal information. Distilling data to information requires a significant degree of engineering domain knowledge and concerted effort, which is not always practical.

“Right now, researchers are focusing on different types of data separately,” he said. “I want to put all of this data together to make a detailed model that covers everything from the external geometry to what’s inside the structure and how it behaves in the real world.”

In structural engineering, sensor data can be interpreted to obtain information useful for modeling an as-built structure. For example, measuring how the structure responds to winds loads can give clues about how stiff beams and connections are. Or alternatively, imperfections in the construction process can be captured by 3D scans. The challenge is that current approaches to convert this data to working models are largely manual and often not worth the effort. Most of the time, engineers assume ideal conditions to make their job easier.

“These tools still need a lot of human interpretation,” said Sarlo. “Sensors can give us lots of details about a bridge or building, like how it is shaped and how it responds to forces. But turning that sensor data into a useful structural model is still hard and needs a human expert’s judgment.” 

Data to models

Over the course of the five-year grant, Sarlo aims to become a leader in translating sensor data to complete structural models through a system he refers to as "data to models."

“The model will represent a structure’s ‘as-built’ behavior, considering as much real-world information as is available,” he said. In other words, if a column is out of alignment or the concrete cured stiffer than expected, those will be accounted for.

Sarlo plans to incorporate data from various sources, such as drone cameras, laser scanners, ground penetrating radar, vibrations, and strains. By combining this data, digital copies of a structure could be created at any point in its lifecycle. This would allow experts to make fewer idealizing assumptions when making important decisions about maintenance or repair. 

As-built models are useful in instances such as retrofit. If there is an existing structure that has settled over time, it is important to determine whether it needs reinforcement. Sarlo’s approach will help generate a model that is representative of the new geometry and other effects of degradation so engineers can re-evaluate the design loads.

In this project, Sarlo’s team will develop algorithms to bridge the data to models gap. These algorithms will use artificial intelligence to encode structural engineering domain knowledge. Some of this knowledge will be encoded directly as structural rules while some will be trained by using thousands of example structural models.

“We will be working with Virginia Tech, particularly with new construction projects, where we can generate data at various stages of construction, starting simple and getting more complex,” he said. “One of the long-term benefits for Virginia Tech is that we will be creating a digital catalog of several campus buildings.”

Integrating technology into existing classes

According to Sarlo, another key to accelerating the adoption of new technology in civil engineering is to make sure that the next generation is familiar with advanced technologies, such as sensors, programming, and artificial intelligence. His solution is to find creative ways to include these topics within existing classes, like his Introduction to Structural Engineering course. This will show students new technologies as part of the traditional curriculum. He hopes that this will better prepare students for future jobs.

Sarlo's research will allow future structural engineers to solve complex societal challenges. “Giving a programming angle to assignments in existing engineering courses hopefully will make students more comfortable with this approach to solving problems,” he said. “I also want to bring examples of artificial intelligence research into the classroom so students can appreciate both its potential and limitations. Sooner or later, they will encounter AI in the workplace, and I want them to understand realistically what it can and can’t do.”