Virtually every major groundbreaking application in AI you’ve heard about – self-driving cars, go-computers, etc. – rely on machine learning algorithms. This change in the final decade has been made possible by the advent of Big Data. Computing power and storage on individual PCs have grown exponentially. Nowadays, millions of servers connect everything throughout the internet. Finally, after fifty years of mathematical research, we have the tools to build some really interesting machine learning applications.
What can machine learning do for Computer Aided Design (CAD)?
A big problem in CAD is the increase in complexity. Machine learning is a possible solution. You could train the computer to understand what’s similar in your drawing and treat it the same way. The computer is an intelligent interpreting assistant that can provide help with repetitive tasks. It can even help to reduce the amount of errors in your model or suggest suitable alternatives for extensive libraries.
AI in BricsCAD
In BricsCAD, the designer always has the final word. The designer can reverse, overrule or simply neglect any design suggestion as they wish. At this point, BricsCAD already has automatic classification of basic building elements based on a solid’s geometry in BIM. For example, automatic recognition of whether any (imported) model is sheet metal or intelligent quad suggestions – now available in beta. In BricsCAD, you can find the AI in our constraint solvers and solid recognition features.
But what we actually envision goes a lot further
Imagine you start with the geometry and turn it into a full BIM model with the aid of artificial intelligence and machine learning in just five steps.
1. Automatically classify solids as building elements
At the moment, BricsCAD already detects walls, slabs and columns. Yet, this can be extended to incorporate more specialized types like composites (curtain walls, stairs, roofs, etc.) and subtypes (inner and outer walls, roof slabs, front doors, etc.). Or even automatically separate stories and buildings.
2. Repair your model
The AI can repair modeling errors, like small gaps between solids, or other small errors, when trained to do so. Of course, if your design is made to have small gaps between walls, they should be left there.
3. Smart connections
BIM allows us to add more details to a model: detailed connections, window types, etc. It’s easy to overlook some locations, if you have to add all those details manually. When you specify the connection between the roof and the outer wall at a single location, the computer will point you to all locations where it needs to be applied.
4. Suggest alternative
The AI can suggest applicable alternatives. Imagine that a manufacturer has discontinued the window series you had in mind and all the windows needs changing. The AI can intelligently apply a new alternative to all the appropriate locations.
5. Intelligent “copy style” command
You can instruct the AI to look at the projects you point to while it intelligently copies and applies certain style elements, such as wall materials, story heights and window styles to your current project. Even without having to manually assign hundreds of classifications and assignments (and potentially missing out on a few). This action can work in two mouse clicks. The first click gets you the suggested model and with the second you approve it.
In short, this is how we think that a BIM workflow should and could look like in a few years. These ideas also transfer to other CAD markets: you start with modeling (or importing) some plain geometry and you get assistance to make it a full Computer Aided Design and Data model.