Using AI to avoid workplace accidents waiting to happen

“Have you watched Minority Report?” DTU Professor Jochen Teizer is prone to asking when explaining his research. In the Hollywood blockbuster, Tom Cruise plays a police chief who wards off criminal acts with the help of psychics who can predict crimes before they happen. Similarly, the professor uses the Internet of Things and AI to foresee adverse events, albeit it on construction sites rather than in a metropolis. 

Because the hard-hitting reality is that even though the rate of serious workplace accidents in the construction industry has dropped significantly over the past few decades, more than 60,000 construction workers lose their life annually all around the world, according to the International Labour Organization.  

“In Denmark, we are among the leading countries for workplace safety. And yet, we are not down to accident-free construction sites even though Denmark has a strong focus on work injuries, health, and mental wellbeing,” says Jochen Teizen and continues:

“The focus so far has been on collecting and analyzing lagging safety data—statistics after workers have been killed or injured—to try and learn how to prevent these accidents in future. But that’s too late.”

Instead, his group focuses on close calls associated with hazards to do with, for example, falls from heights or heavy machinery that gets close to workers on foot. The researchers evaluate the root causes that went into these, so they can figure out how to prevent injuries or fatalities through prediction and planning.

A close call is defined as an unplanned event that had the potential to cause injury, damage, or loss but did not, often due to chance or a timely reaction. Close calls can stem from unsafe behaviour, wrong choices or equipment not operating properly.

Three-way problem-solving

The professor’s research group approaches occupational construction safety and health from three angles. Firstly, in the design phase of a project they evaluate digital 3D information models of a building or infrastructure. By identifying potential hazards along the work schedule early on, they can then design them out or suggest the application of alternative measures.

The group has, for example, developed AI-based software algorithms that assist the safety responsible in figuring out and planning where and how many temporary guardrails should be installed to avoid workers falling from heights or into holes. 

“The second core element in our safe digital twin process is how do we warn the workers fast when they get into a hazardous situation in an environment that is so dynamic and complex with lots of workers and big machines nearby,” Jochen Teizer explains.

To address this challenge his group has developed technologies such as smart hard hats and safety vests that use radio waves to sense dangerous situations in real time, and both warn the wearer to get out of harm’s way or alert the machine operator of a worker being (too) close by.

Lastly, the group builds augmented virtual learning environments that through VR headsets provide multiple workers with realistic safety training simultaneously in a setting that mirrors the actual site where they are or will be working.

In these safe worlds workers can perform their normal tasks alongside their colleagues while being exposed to the types of hazards that are present on the site to learn to anticipate and avoid them. The workers use tools and machines akin to the ones they use on the construction site, so when operating an angle grinder or a power drill in virtual reality, parts will rotate and vibrate, and they weigh and sound like their real-life counterparts, explains Postdoc Killian Speiser who works on virtual reality systems at DTU.

“This is about developing a science-based approach to training human behaviour and not just awareness by checking for example whether people can find hazards and make the right choice. This way we don’t just uncover intentions but show actual behaviour and how it must change,” he says.