Throughout 2024-2025, Research Manitoba celebrates our 10th anniversary year. As we mark this achievement, we will be looking back at some of our past funded researchers to highlight their success.

Researcher: Dr. Youngjin Cha
Institution: University of Manitoba
Grant received: CFI (John R Evans Leader Fund) / Research Manitoba Matching Funds
Year: 2017
Amount Received: $399,736
Project Title: Autonomous structural health monitoring using UAV and deep learning

Researcher Young-jin Cha is a leader in developing a new and better way to keep tabs on the health of various infrastructure, including bridges, buildings and roads

Profile written by: Brian Cole

Bridges don’t collapse all that often, but when they do, they tend to attract a lot of attention.

The Fern Hollow Bridge in Pittsburgh is a case in point. Six vehicles were on or near the bridge when it suddenly collapsed in 2022, injuring several people.

Needless to say, the incident raised concerns across the continent as commuters and government officials alike asked a very simple question: What can we do to keep our bridges safe and sound?

It’s a question 48-year-old Young-jin Cha, a professor in the Department of Engineering at the University of Manitoba, has spent much of his career trying to answer.

In fact, it was the collapse in 1994 of the Seongsu Bridge in Cha’s home town of Seoul, South Korea, that first got him thinking about the structural safety of bridges and other infrastructure. Thirty-two people died and 17 more were injured when the bridge collapsed into the Han River. A subsequent report attributed the bridge’s failure to faulty welding and insufficient maintenance.

“That was my initial motivation to pursue studies in structural engineering,” he says of the incident.

To that end, Cha attended Yonsei University in South Korea, where he received a master’s degree in civil and environmental engineering in 2004, and then went to Texas A&M University, where he received his PhD in 2008. He also served as a Postdoctoral Fellow at the Massachusetts Institute of Technology in 2012.

In 2014, Cha’s career took a major turn when he accepted a position as a professor at the Department of Civil Engineering at the University of Manitoba. Among other things, the appointment paved the way for Cha to begin exploring more deeply the subject that had long held his interest: how to better keep tabs on the structural health of infrastructure, such as bridges, buildings and roadways.

And, it didn’t take long for Cha to make his mark.

Soon after his arrival in Manitoba, Cha turned his attention to a relatively new area of research that involved using computer vision (including artificial intelligence) and camera-equipped unmanned aerial vehicles (UAVs) to monitor the structural health of various infrastructure, including bridges, buildings and roadways. By 2018, he had established himself as a pioneer in the field, having developed a new approach to structural health monitoring, one that utilizes autonomous UAVs and a form of artificial intelligence known as deep learning.

As Cha explains, safety checks on infrastructure, such as bridges, have traditionally been done by having engineers or specially trained investigators carry out visual inspections, looking for cracks, corrosion and other signs of structural weakness. In the case of a bridge, this is usually done by having inspectors climb up and around the steel girders and trusses that support the structure.

More recently, bridge inspectors have also incorporated more technology into their tool kit, adding vibration sensors and cameras, among other things, to the list of equipment that is used to help monitor the health of infrastructure.

But the traditional approach has several weaknesses.

For one, Cha says visual inspections aren’t always as accurate as they could be for a variety of reasons, including variations in the skill level and training of personnel and the difficulty involved in accessing hard-to-get-to areas of a bridge.

“These (inspectors) are only human,” says Cha. “Maybe sometimes they miss a crack or other damage,” he says.

Even the addition of vibration sensors, while helpful, cannot catch all the potential problems with bridges and other infrastructure.

“Vibration based approaches were initially developed for damage detection, but (long-term research shows it is) extremely difficult to detect small, but critical damage like cracks or corrosion in (a) bridge system.” says Cha.

Another issue is that the labour-intensive nature of bridge inspections means they are generally only carried out bi-annually to control costs.

“Every two years, we do these kinds of inspections. Based on that, they (inspectors) say the bridge is safe,” says Cha.

The problem, he says, is that bridges have been known to collapse six months or a year after an inspection, presumably because of damage that went undetected at the time of inspection or because something happened to the bridge between inspections.

In 2016, for example, the newly constructed Nipigon River Bridge near Thunder Bay Ontario heaved apart without warning before it could even open.

“Those things happen world-wide,” he says.

This is where deep learning and autonomous UAVs enter the picture.

“(Deep learning) can do what humans cannot do,” he says.

Essentially, deep learning involves programming (or training) a high-powered computer system  to recognize and analyze patterns in images and other data, which makes it a perfect tool for detecting potential flaws in infrastructure, including bridges, buildings and roads.

“When I focused on computer vision-based damage detection, I encountered the challenge of identifying specific structural elements in the images,” says Cha. “I needed to first locate the element of interest before determining whether it was intact.”

“Through my literature review, I discovered that deep learning is the most effective method for finding specific objects within images.” he says. “This was the moment I realized the potential of deep learning in structural health monitoring.”

Cha likens the approach to a water purification system.

“It (the deep learning process) employs multiple filters to eliminate noise and irrelevant data, extracting only what is necessary,” he says. “These various filters can be adjusted to address different engineering challenges. So, in damage detection… we process all the images using deep learning that was (programmed) to detect only (cracks).” 

In addition to reducing inspection time and increasing the frequency of inspections, the new approach significantly enhances the accuracy in detecting potential flaws, says Cha.

“The accuracy of damage detection is greatly improved, as human inspectors typically only provide a general assessment of severity rather than specific details about the size of the damage,” he says. “Therefore, this represents a completely different level of inspection that was previously unattainable through traditional visual inspections by engineers.”

Of course, the use of deep learning to analyze images can only take place once the photographs  are taken. For that job, Cha determined that an autonomous UAV with a specially designed navigation system could be used to fly in and around the infrastructure being inspected. In fact, Cha was the first researcher anywhere to develop an autonomous UAV flight navigation system for the purposes of structural health monitoring.

And doing so was no small thing. Most UAVs are controlled manually or via GPS. But manually controlled UAVs would require two trained “pilots” to maneuver the machine around a bridge or other structure to get the required photos, which would add to the cost of the exercise. Meanwhile, UAVs guided by GPS aren’t always reliable because the signal is not always available, says Cha.

To get around these issues, Cha built his own prototype UAV, one that uses fiducial markers to navigate around a particular structure, and tested it on a on a parkade on the U of M campus. The autonomous UAV can also avoid unexpected obstacles within the planned trajectory, significantly enhancing the reliability of its real world application.

The results were positive.

“I learned that our integrated system (deep learning and an autonomous UAV) performs exceptionally well, even in windy conditions, despite the use of small UAVs. This indicates that outdoor applications for automated inspections of bridges, buildings, dams, railway systems, and large facilities are very promising,” he says.

Cha’s work in this area has not gone unnoticed.

His research has appeared in more than 100 peer-reviewed publications, including Automation in Construction, Computer-Aided Civil Infrastructure Engineering, IEEE Transactions on Industrial Electronics, and Structural Control and Health Monitoring. As a result, his work has drawn attention from scientists and academics around the world and has been cited in numerous follow-up papers.

Cha has also received the 2021 Merit Award in the Research, Scholarly Work and Creative Activities category from the joint committee of the University of Manitoba and University of Manitoba Faculty Association, and the International Association of Advanced Materials Scientist Award from the 2022 European Advanced Materials Congress.

In addition to monitoring bridges, Cha says UAVs and deep learning can be used to keep tabs on all sorts of infrastructure, including buildings and roadways.

For example, he recently completed a study into the viability of using deep learning to detect potholes on city streets.

This initiative involved equipping a bus with a camera that would take pictures of the roadway as the vehicle travels around the city.

“If we collect images or videos from cameras installed on buses or other vehicles, deep learning can process all this data to detect potholes and their corresponding depths in real time,” he says.

So far, Cha has received $1.5 million in grants from various sources, including Research Manitoba, the Canada Foundation for Innovation, and the National Sciences and Engineering Research Council of Canada, to carry out research into the use of deep learning and autonomous UAVs to assist in structural health monitoring. And although the new approach holds much promise, Cha says the technology is still in the early stages of development. More work needs to be done before an integrated autonomous UAV – deep learning system can effectively replace current infrastructure inspection methods, he says.

“At this point…our concept has already been proven,” says Cha. “Now we have to make a commercial product. To do that, we need more intensive study,” he says.

Dr. Youngjin Cha

Website: www.youngjincha.com

Dr. Youngjin Cha is a Professor in the Department of Civil Engineering at the University of Manitoba (UofM). Prof. Cha is the Director of the Laboratory for Infrastructure Science and Technology (LIST). He received PhD at Texas A&M University and had PostDoc position at MIT. He is the pioneer of deep learning based structural health monitoring (SHM) including autonomous UAV method for SHM. His focus areas include computer vision-based damage identification, thermography based external and internal damage identification, development of autonomous UAVs including robots for civil infrastructure monitoring, various high resolution 3D digital twin technologies, and remote autonomous scheduled monitoring through metaverse platform. He was listed as top 0.29%, 0.30%, 0.34%, and 0.35% most cited scientist globally in the fields of civil engineering, acoustics, all engineering, and all science and engineering for 2022, 2021, 2020, and 2019, respectively. He received Merit Award from UofM, IAAM Scientist Award research award, and UMGSA teaching award. He is currently serving as Academic Editor for Structural Control & Health Monitoring, Wiley, Associate Editor for Structural Health Monitoring, SAGE, Associate Editor for Engineering Reports, Wiley, and Associate Editor for the International Conference on Pattern Recognition (IAPR) in 2022.