Learn the past and build the future: How Firmus’ AI Exceeds Royal BAM Group’s Expectations in Finding Design Errors in 3D Models
Construction is a highly competitive industry with small margins. There’s not much room for error, so catching design mistakes early can mean the difference between a profitable project and a money loser.
The Get It Right Initiative found that the construction industry “wastes 21% on error and that four of the top ten root causes of error relate to design.” That’s a significant cost can translate into millions of dollars of losses. For example, the $1.2 billion Fort Bliss Hospital replacement project incurred $165.6 million of design errors, with a single design error accounting for $9 million, alone.
To account for the risk factor of design, quality, constructability and coordination errors, construction companies include a buffer when they propose bids, and given the small margins on these projects, even a reduction in risk of just half a point can make an enormous difference to profitability and competitiveness.
Unfortunately, the process for finding mistakes in building designs still involves a great deal of manual work, making it inefficient and error-prone. As a result, a significant number of issues typically aren’t discovered until late in the process, when they are exponentially more expensive to fix.
For example, if a relatively simple error such as a light switch placed on the wrong wall is discovered during the design phase, it’s trivial to fix, costing maybe $25. But if the error isn’t found until construction begins, it could cost $250, because now there’s the additional cost of labor for the construction crew.
Even worse, if it’s not recognized until after the owner has moved in, the cost could be as high as $250,000 because now the firm has to shut down power to a section of the building, stop people from working, tear out the wall, move the cables, reset the switch and then clean up the mess. And that’s for a relatively small mistake. If the issue involves a complex system like HVAC, costs can go far higher.
Applying AI to 3D Designs
BAM Ireland is a wholly-owned subsidiary of Royal BAM Group of the Netherlands, an international construction giant with more than €9 billion in annual revenues. BAM Ireland, itself, sees €524 million in annual revenues and employs about 2,700 people.
The organization already has a high level of digitalization on its projects, which range from the 49,000-square-meter Hi-tech HQ campus at Leopardstown to the 30,000-square meter Acute Mental Health Inpatient Centre at Belfast City Hospital. These are large projects that generate a lot of data. A courthouse, for example, can generate thousands of documents, which could include emails, requests for information (RFIs), snags, non-conformance reports (NCRs), change orders and, of course, the 3D designs themselves. If the firm is using laser scanning to capture high-definition images of the site, the amount of available data grows even more.
Simon Tritschler, Technical Deployment Specialist at BAM Contractors, wanted to use this mountain of data productively. After meeting with Firmus, Tritschler agreed to a proof of concept test with Firmus’ artificial intelligence (AI) engine to see how well it could find design errors in 3D models.
The prospect of using AI to find design mistakes was particularly interesting to Tritschler, not only because catching a higher percentage of errors early could save a great deal of money and time, but also because AI is particularly well suited to the task. A large hospital could easily have more than 6,000 rooms, and manually checking that every room has a door is a tedious, repetitive task that would take many working-hours to complete. An AI doesn’t get bored and can complete the job in mere minutes. Even better, the more models an AI analyzes, the better it will become at finding mistakes.
But first, the AI needs to train on past projects to learn how to find issues, so BAM Ireland fed Firmus’ AI engine data from previous projects before giving it a similar model — in this case, a hospital — to see how many errors it could find.
Reducing Project Risk with AI
Firmus found 211 issues in the project it analyzed which is more issues than Tritschler expected it to find and the impact on project profitability could be substantial. Using data from the American Institute of Architects and Association of General Contractors, construction management firm Hourigan estimates that the average cost of a clash — another term for a design error — is about $1,500. When you add in the additional expense of sending an RFI to the design team, which is just over $1,000, according to the 2017 Open Integration Summit Report from the Construction Progress Coalition, the average cost for each item balloons to $2,500.
Using this estimate, the potential savings for Royal BAM would be $527,500 or just over €485.700. And that’s after training Firmus’ AI on just 10 historical projects. After training on more data, the AI would find even more issues, which would, in turn, increase the savings.
Firmus’ AI exceeded our expectations and, over time, it will become even better at finding design mistakes that human beings would likely overlook. Ultimately, Firmus shows a lot of promise to significantly reduce our overall risk, which would, in turn, enable us to be more efficient and propose more competitive bids.
Technical Deployment Specialist at BAM Contractors
Royal BAM Group
Annual revenues: €7.2 billion
No. of employees: 20,000
Head office: Bunnik, Netherlands
Firmus uses the power of AI to help construction companies detect constructability, quality, design and coordination errors at the earliest stages of projects to reduce risk, increase efficiency and improve competitiveness. For more information, please visit http://firmus.ai