CU Boulder researchers are developing an app that could reliably and quickly predict whether batches of concrete made at construction sites are safe. If successful, the work could usher in a new era of faster, more cost-effective and safer construction for all.
The work is still in its early stages and is funded by a seed grant from the Interdisciplinary Research Theme in Engineering Education and AI-Augmented Learning within the College of Engineering and Applied Science.
Assistant Professor Mija Hubler in Civil, Environmental and Architectural Engineering said the aim of the project was to develop an application capable of collecting and analyzing concrete image samples to detect possible defects using machine learning techniques based on composition, fault lines and visual cues.
“To do that today, we have to send samples to the lab where they’re then destroyed for analysis – so it’s not a very efficient process in many ways,” she said. “We’re hoping to develop something where you could open a sample on the spot, take a photo, and figure out how that batch will work mechanically.”
Hubler said the approach is similar to medical techniques, in which doctors examine images of bones or organs to perform assessments with increasingly sophisticated tools and techniques.
However, concrete is a much less homogeneous material, which makes the evaluation tricky. And users of such an application would need at least some basic training in machine learning to understand the inherent uncertainty of predictions and how to proceed with them.
“We’re talking – essentially – about a smart tool here. These kinds of tools and skills are going to become more common in construction over the next 20 years, especially with the introduction of autonomous vehicles,” he said. she said, “We quickly realized then that it was more of a pedagogy for everyone on the site. How are these skills taught? What do you need to know about machine learning?” to use these tools? That’s why it fits well with the interdisciplinary research theme of engineering education and AI-augmented learning.”
To answer these kinds of questions, Hubler is working with assistant professor of computer science education Geena Kim. Kim said that while artificial intelligence and machine learning are becoming more mainstream in his field, many other branches of engineering are just starting to use these concepts. She added that the scale and size of civil engineering data sets present interesting challenges when creating the necessary algorithms, testing with students, and understanding the results for broad applications.
“We need to get more data and insights to really understand how people are going to interact with this app and what their personal experience with AI and machine learning needs to be to use it properly,” she said. declared. “This work will also help us understand these concepts in curriculum and workforce development over time.”
Hubler said the team will continue to refine its approach, while seeking collaborators at CU Boulder and beyond.
“The primary way we assess and track our infrastructure in America is through visual inspections, so that kind of tool would be pretty powerful,” she said.