After many months of research and development, we recently launched the rapid noise analysis. This predictive analysis and new AI-powered design tool complements our existing noise analysis that you’re already familiar with. What’s the difference between the two, you might be asking? To find out more, we caught up with Machine learning engineer and rapid noise project lead Renaud Danhaive from our Boston office. He gives us a quick glimpse behind the scenes into creating this exciting new feature.
So Renaud, what is the rapid noise analysis and how does it work?
Rapid noise is a new analysis that predicts noise levels on the ground instantaneously. With these approximations, our users can have an immediate indication of how their buildings impact noise conditions on the site. They can then make iterations much faster and more fluidly before running the regular noise analysis, which takes several minutes, to get the exact results for verification. An easy way to think of it is that the rapid noise analysis makes an educated guess about the noise conditions and the regular noise analysis provides the accurate calculations.
Through machine learning, we use a dataset of tens of thousands of noise simulations to train a neural network to predict ground noise results. Visually this analysis looks similar to our regular noise analysis – green means quiet, red means noisy and yellow means somewhere in between. Rapid noise also takes rail and road noise conditions into account. Only now you’ll notice that the noise levels on the ground respond instantly as you make changes to your building. Like other AI-powered functions within Spacemaker, this feature is designed to think along with our users and help them reach better outcomes faster.
What makes this so game-changing?
For designers, this means they get immediate feedback on their design while they’re designing which is a massive change to how they’re used to working. As they’re now able to make iterations much more quickly and fluidly, this helps them stay in a good design flow.
From a technical perspective, this is one of the first times that a machine learning model like this, a so-called surrogate, is commercially available within an Autodesk product. In the past, surrogates were mostly experimental or used only in very niche products. Here it’s accessible and easy for everyone to use, no matter how tech savvy they are.
How does it work with Spacemaker’s regular noise analysis?
They’re designed to complement each other but serve a different purpose: rapid noise predicts noise levels using data from previous simulations to offer instant but approximate results, and noise analysis uses site-specific environmental data to calculate accurate results for when you need to verify your designs. In that sense rapid noise is designed more as a design tool to speed up the iteration process – it also lives within Design while our regular analysis lives in Analyze.
Did you work with users to test the tool?
Yes, we invited selected noise ‘super users’ to many user sessions to test the analysis. Their feedback was overwhelmingly positive, and they also encouraged us to expand noise predictions to facades, which we’re exploring at the moment. At present, to calculate noise levels on facades, users can continue using our regular noise analysis.
Can you give a sneak of other instant analyses you’re developing?
In terms of the machine learning analyses, we’re expanding our rapid analyses to wind. But that’s just the start, and we’ve got some exciting things in the pipeline.
There are already other types of rapid analyses available as you design in Spacemaker. You can enable key figures and area metrics that update as you design. You can enable the rapid sun analysis, a sun analysis that utilizes your GPU to estimate impact of sun on your project as you design. Plus you also have access to the Constraints rapid analysis, that warns you if you exceed volumetric constraints as defined in your Layers.
For more information about rapid noise, please visit our Help center (requires Spacemaker account).