During the £1.5bn upgrade of the A14 in Cambridgeshire, an archaeologist found what was believed to be the earliest evidence of beer brewing in Britain, dating back around 2,000 years.
Generating as much excitement, for different reasons, was the introduction of a very modern concept on the same scheme. The project team pioneered artificial intelligence (AI) and machine-learning technology to successfully predict times when an accident was more likely to happen – and to take action to stop it. By collecting swathes of information and using the AI, data scientists were able to spot problems before they occurred.
AI has also been popping up in infrastructure planning and even remote assessments during the pandemic.
So is it the solution for improving safety and efficiency the industry has craved? And if so, why is it not already more widely used?
Highways England’s A14 scheme involved improvements to a 21-mile section of road between Cambridge and Huntingdon. To ensure that both hazards and good practice could be recorded, data was collected from staff via a specially built app for iPhone and Android. At its peak, 4,500 observations were recorded in a month. As a result, a daily profile was generated that identified the days that were thought to raise the risk of an incident occurring. The use of an app meant that reporting was more efficient than the old paper-based system, and was instantly logged.
“Previously, you would have a paper card, which would get left on someone’s desk for a week or so, then input by hand,” says Mark Tootell, Highways England’s head of the A14 project management office.
The new system meant the AI was able to model the days that presented the greatest risk based on factors that were found to include work that was carried out after 6pm; for more than nine hours; in high winds; following heavy rainfall; and immediately after a bank holiday or national sporting event. One of the most significant factors was staff fatigue.
A joint venture of contractors Balfour Beatty, Costain, Skanska and Atkins delivered the A14 project. Balfour deputy project director Julian Lamb says the AI allowed supervisors to focus on areas of concern. “It doesn’t mean you run around saying ‘today is going to be a really risky day’, but what you can do is try to target the things where it [the AI] is suggesting there might be issues,” he explains. “So you are concentrating on and inspecting certain operations.”
Tootell says the model was helped by the amount of raw data received from workers on the ground: “It was that volume of information that enabled the AI to really be as accurate as it was. They [staff] are your eyes and ears on the site, so [they’re] your best leading indicator.”
To encourage site workers to input data, they were offered a variety of incentives, ranging from a canteen meal to a fully paid day off. The key was getting as many people involved as possible, Tootell says.
The results led to an accident-frequency rate of 0.028 after 14.3 million hours worked – less than half the 0.07 average rate for Highways England contractors in 2019/20. Around 2,500 workers were on site at the project’s peak. Tootell says he is proud of the health and safety record on the project, which saw only “four RIDDOR reportable injuries during the whole project – three fractures and a twisted knee”.
He encourages others to also look to use this type of system. “You need somebody who can operate the model,”Tootell says. “I’m not saying every project should have a data scientist in the team, but absolutely they should have access to that type of resource going forward. The power that a good data-analytics team can bring to a project is enormous.”
Consultants are also increasingly incorporating AI into their projects. In Shanghai, Arup developed a machine-learning tool on a project to tackle the city’s growing flood and river-pollution issues.
