Case Study: Converting AutoCAD Layers to Revit with Machine Learning

07.16.20

Many industries have embraced machine learning to solve a variety of problems. As a mechanical, electrical and plumbing engineering consulting firm that does 100 percent of its designs in Revit, Kohrs Lonnemann Heil Engineers, Inc. (KLH Engineers) was manually converting its clients’ 2D AutoCAD models to 3D Revit models. Utilizing this historical data, KLH Engineers was able to implement a machine learning model to automate the process.

The Manual Process

Architects use different labeling systems when organizing elements within their 2D design files. When KLH Engineers receives an AutoCAD model, it needs to be “cleaned” to ensure the model meets internal standards so the firm’s customized tools can convert the model automatically. One of these standards encompasses the layer names in the model so engineers and designers can differentiate between different layers such as walls and furniture and give them the proper properties in Revit.

Each layer in the AutoCAD model must be assigned a phase and a category. This was previously done one at a time. The model manager responsible for the conversion would go through each layer, look at it, and determine the phase and category based on how the layer looked. This would require hours of a skilled designer’s time devoted to a relatively simple task.

Automating the Process with Machine Learning

KLH accumulated 38,000 layer name translations made by its designers over the course of a few years. Instead of writing rules by hand, KLH used this historical data to have a training algorithm write the rules to train a machine learning model.

The algorithm is a deep neural network written using TensorFlow, a library to implement machine learning models. The algorithm analyzes each set of letters in pairs and correlates the results to historical data to make predictions. For example, a ‘New Wall’ would get analyzed as [ NE , EW , WW , WA , AL , LL ]. Compared to ‘A-Wall-New’, which breaks down to [ A- , -W , WA , AL , LL , L- , -N , NE , EW ]. Based on pairs of letters that exist, both are designated as a new wall by the algorithm because they both contain [ WA, AL, LL ] and a letter pair with an ‘N’. Not only are the predictions instantaneous, they are also accurate. The model also alerts users of its confidence level so layers may be reviewed and corrected, if needed.

The Results

As the historical dataset grows, the results are becoming so precise that it now predicts results at a higher level of accuracy than a person doing it manually. The machine learning model is able to predict the results in seconds, thus saving the designers’ and engineers’ time to do more important tasks. In 2019, nearly 2,200 hours were saved, equating to $275,000 in waste elimination.

KLH Engineers is continually leveraging its experience in machine learning, along with its ever-growing internal database, to further automate the design of building systems and reduce waste so the benefit can be felt by all of its partners and clients.

Watch the video below to see the tool in action.

About KLH Engineers:

KLH Engineers is nationally ranked among the top mechanical, electrical, plumbing and technology engineering firms. As a leader in the AEC industry, KLH leverages building information modeling (BIM) and advanced in-house technology to deliver innovative and value-driven solutions. KLH has experience in all 50 states and serves the civic, education, healthcare, hospitality, lifestyle, industrial, retail and workplace markets.

Want to help build the future?

Connect With Us
Match: Very efficient, saves a significant amount of time. This was huge for light fixture tagging.
- Kevin B., Software Customer