Higher Education: The Story Behind Skand's Drone/Machine Learning Envelope Inspection
Emily Pollock posted on October 29, 2018 |
Skand founder Brett Chilton presents his company’s envelope inspection to the Year in Infrastructure jury. (Image courtesy of Bentley.)
Skand founder Brett Chilton presents his company’s envelope inspection to the Year in Infrastructure jury. (Image courtesy of Bentley.)

An Australian startup with fewer than 10 employees developed an award-winning envelope inspection that uses machine learning to find and classify defects based on drone footage. Skand mapped an entire university campus’ building envelope without needing to put a single worker up on a building rooftop, creating a complex and easily searchable 3D mesh map.

“What does this mean?” asked Skand Founder Brett Chilton during his presentation at Bentley’s Year in Infrastructure awards. “Well, it leads to a safer and more productive building, across all metrics.”

The Royal Melbourne Institute of Technology (RMIT) wanted a building envelope inspection project for its Brunswick campus, which covers six buildings and 65,000 square meters. The university wanted a project that would incorporate drone imagery of the buildings’ roofs and walls, and which would catalog and attach a priority system to any defects that were identified. Additionally, the school wanted a system that would interface with its 40-year-old, award-winning asset life-cycle program.

Several aspects of the project made it tricky to complete. The most difficult factor was site access, which was restricted both by the fact that the buildings were still in use and the fact that some of them backed onto a major roadway. Civil aviation rules meant that Skand couldn’t fly drones over the road, or within 30 meters horizontal of pedestrians on the campus. These rules meant that Skand had to carefully plan both its flight plan and flight timing to avoid breaking the law, and that it couldn’t just fly conically around all of the buildings to gain a 360-degree view.

Additionally, RMIT wanted all the building defects to be analyzed in a homogenous fashion, with no room for differences in human analyses. These exacting specifications meant that the Skand team would have to use AI/machine learning to classify any defects. Finally, the data that Skand collected had to be put on a platform that was intuitive for everyone involved in the maintenance process, from senior engineers to student workers.

To catalog the campus, the Skand team performed three different drone passes, each in a double grid system to gather the maximum number of data points. The highest pass was 380 feet off of the facade, the highest Skand could legally fly its drones, while the lowest pass was between 10 and 15 feet, the lowest possible legal limit. The team performed both visual and thermal analyses of the buildings.

When the team had collected its point cloud data, it translated the data into a 3D reality in ContextCapture. While competitors’ products let users draw defects onto the mesh to mark them off, ContextCapture automatically found defects and classified them into 16 categories (debris, thermal bridges, sediment buildup, etc.) and three severity categories.

From that data, the team created an inspection report and a publicly available 3D campus model that was hosted on its web-based application.

“I want to make it clear that we’re not here saying we’re replacing the human touch,” Chilton cautioned his audience. Although the company’s machine learning algorithm was able to find and classify defects, it needed human input to determine what needed to be done about the defects. Rather than a replacement for workers, Chilton sees the project as just “a new tool for the people that interact with buildings.”


Recommended For You