The machine learning-based analytics tool Construction IQ has demonstrably improved quality and reduced safety issues on building sites. Pat Keaney, director of products for project delivery and collaboration at Autodesk, tells Stephen Cousins how it works and why AI will have revolutionised many aspects of project delivery within the next five years.
What is Construction IQ?
A machine learning and AI-based tool that analyses the quality and safety data of projects that utilise our construction management platform BIM 360. It identifies any activities that it calculates pose the greatest risk.
Why was it created?
BIM 360 users enter information on checklists that describe numerous things that can go wrong on projects, from a quality or safety perspective. We asked ourselves: are they really able to leverage that rich data to understand patterns and risks and figure out how to improve projects and are we doing anything to help them do that?
The answer to both of those questions was no, BIM 360 was being used to track problems and resolve them, but customers weren’t getting the full benefit of the rich data.
Construction IQ searches for any issues that could in any way be related to a water penetration risk, if it finds them it makes them visible to ensure that issues are addressed before the wall gets closed up.– Pat Keaney, Autodesk
How does it work?
Construction IQ sifts through data points from issues, observations, checklists, subcontractor assignments, related meta data, and historical data to help analyse, identify and prioritise risk factors related to either quality or safety each day.
We have carried out several years of R&D to develop algorithms and machine learning models that categorise the data to identify problem areas. Models have learned from over 150 million construction issues and checklist observations across nearly 30,000 real projects to develop the algorithms.
If a project has 1,000 open issues on any given day, the tool tries to identify the 50 high-risk issues and makes them visible to the relevant managers to act upon.
Any specific examples?
From the quality perspective, a large window being installed needs a sealed waterproofing system around it, but if the material is not installed correctly it can create leaks that lead to all kinds of problems downstream. It might create mold or damage flooring and result in warranty problems for the owner of the building.
Construction IQ searches for any issues that could in any way be related to a water penetration risk, if it finds them it makes them visible to ensure that issues are addressed before the wall gets closed up.
On the safety side, we sift through all safety-related information to identify behaviours that correspond to the four key ways that people tend to die on construction sites, the top one being falls from height.
We built models that detect behaviours related to unsafe practices on ladders and scaffolding, or where there is a risk related to an open elevator shaft or the edges of a high rise building.
Construction IQ makes risky behaviour visible to the safety manager or the superintendent of the project so they know how to target their toolbox talks and ensure that people are paying attention to fall protection and following good safety practice.
Why does machine learning make this more effective than normal processes?
Numerous insights are all contained in the data, but no human can read and process that much data every day, it would be physically impossible and hugely time consuming. BIM 360 is 100% cloud based and all the data is pushed into a data lake or a separate repository for analysis.
The goal is to find a signal in all that noisy data, then surface the actionable results in a very simple, easy-to-understand user experience for people on site.
BAM Ireland piloted an earlier version of the tool and managed to achieve a 20% reduction in quality and safety issues on site as a result of better decision making supported by its use.
Have you been able to prove that the issues identified as most risky are indeed the most risky?
We believe that most of our models are now about 85% accurate, based on seeing the results of the model, putting it in front of customers and running tests with them. That’s a fairly high level of accuracy, and the system gives customers the opportunity to agree or disagree with our analysis. If IQ says an issue is high risk, they can disagree and re-label it as a medium or low risk and the system ignores that issue the next time data is processed.
AI is not 100% accurate and we make it clear that Construction IQ is meant to be an assistant, not a substitute for a person with 40 years of experience.
To what extent are AI innovations going to transform building projects?
We started developing Construction IQ about four years ago, and back then we had to explain to people the basics of machine learning and AI, but today you can’t pick up a magazine or read a news website without seeing a mention of AI or machine learning, it’s everywhere.
There are a number of construction start-ups beginning to apply AI to all kinds of problems. In five years, AI will be used throughout the construction lifecycle. Autodesk wants to be a platform company and help other firms that are trying to solve problems with AI. One interesting partner is Smartvid.io, which exploits image recognition technology to identify safety risky behaviour on site.
Another company we host at our innovation centre manufactures an Internet of Things device that hangs on a crane and is able to measures things like weather conditions, logistics and productivity. By tracking weight on the crane it can work out utilisation and if the crane moves in a way that’s not normal it can search for potential safety issues.
Another interesting company is Germany’s HoloBuilder, which uses image recognition and 360 degree camera images to recognise the phase of a project based on what it sees. For example, if it sees the structure of interior office partitions without a surface material, it knows what phase the project has hit. This isn’t science fiction and it isn’t 10 years out, these start-ups have prototypes, previews and working versions today.
Given the data being captured by cameras, IoT, drones and point clouds it is almost endless the problems that AI can be applied to and we have only just started to scratch the surface.