Parametrized design software is not a recent invention. This software is based on predetermined, fixed algorithms, leaving most of the work to the designer. Sweco, a leading engineering consultancy, is now exploring how artificial intelligence (AI) could take design automation in the Architecture, Engineering, and Construction industry to the next level.
“We want to build a system that can help the designers in their daily repetitive tasks by suggesting possible viable outcomes for a design. For example, if a designer starts working on a concrete sandwich panel, the machine should be able to suggest a valid detailed solution, adjust it to the present geometry and learn from the process. This will speed up the process and improve the quality of the design over time,” says Ricardo Farinha, BIM Application Development Manager at Sweco. “We call this the Design Recommendation System.”
The development of the solution started in a KIRA-Digi experimentation project, and it now continues in a Business Finland funded project (DiCtion) that aims to digitize construction workflows. Development and piloting is taking place in Finland, but Sweco will eventually roll out the system throughout the Group.
Learning from the Past and Present
Sweco’s solution is a virtual companion that not only suggests solutions but also learns as the designer chooses between the options the system recommends. The initial training data was extracted from thousands of fully detailed precast concrete panels created by Sweco engineers in the past.
“We’re now working with Tekla Structures models. First, we ran data mining routines on the existing models. Then we transformed the data so that it works for AI,” says Farinha. “The process is totally automated.”
“We are moving into an era in which static software tools are being replaced by dynamic software tools that are continuously learning and adapting,” Farinha says.
The system will monitor the actions and decisions of the designer in detailing the panels but “All the suggestions made by the AI are verified by a qualified engineer and changed when needed,” says Atte Leppänen, Business Development Director.
For example, the system might not have previous data on how to create a rebar mesh for a concrete wall panel of a certain size. When the designer adds the mesh to this particular panel, the system stores those actions and creates rules about how to create the panel in similar situations in the future.
“For a human, it’s obvious that you don’t add a five-meter-long rebar mesh to a four-meter-long concrete wall. Surprisingly, that’s not the case with AI, and we have to teach this simple rule to the machine,” Mauri Laasonen, the project manager, chuckles.
In-House Data and Development
Currently, Sweco develops the software in-house. It can be used to design any type of structure. Sweco chose to start with sandwich panels because of the fact that many of its designers are involved in detailing precast concrete structures in the Nordic region, which means that the benefits can quickly be scaled up.
Several startups and other companies are promoting AI or even doing AI development similar to the work of Sweco. Laasonen and Farinha agree that Sweco’s advantage over the competition is its access to relevant historical data.
Sweco has done hundreds of projects in the past, but previoulsy the knowledge and lessons learned from those projects were stored only in the minds of the designers, not as machine-readable data. With AI, any designer can access this huge collection of design solutions and improve the quality of the data available on every project.
“Efficiency is certainly a big benefit. Another big benefit is that junior designers will be able to quickly build up expertise similar to their more senior colleagues,” says Laasonen. “This improves the quality of our designs. In addition, we will be able to proactively provide multiple alternative solutions, something that is valuable to the customer.”
Improving Design Quality
Today, designers may employ slightly different methods in solving similar engineering challenges. An AI-supported process will eventually lead to less variations in outcome. This makes the industrial manufacturing of building parts more efficient.
AI can help in implementing a circular economy because building parts will become more standardized and will therefore fit together better when they are reused in another project. Furthermore, AI allows structure optimization based on a variety of criteria, e.g., reusability or price.
A group of engineers are currently testing the system and providing valuable feedback to help make the application as useful as possible.
When talking about AI, the question about machines replacing humans is inevitable. So far, designers at Sweco have been excited – they see AI as a tool that will make their lives easier. A virtual sidekick will free up time for more demanding engineering tasks, such as determining how to make a building more usable and how to make it recyclable and more energy-efficient, among other things.
The project will last until the summer of 2020, but Farinha expects piloting on actual construction projects to take place in 2018, Laasonen says. “The best way to achieve practical solutions is to perform focused pilots with a few clients.”
You can learn more about this and other AI projects at the World Summit on Digital Built Environment WDBE 2018 taking place in Helsinki on September 11–12, 2018. With over 100 presentations and 10 keynotes, it is a must-attend event for construction innovators and game changers.