Machine Learning for MEP Firms

Let's continue our discussion about how the work of BuildingSP intersects with the major themes of Jeff Kowalski's 2016 Autodesk University Keynote. We previously wrote about artificial intelligence and how firms can prepare for a new generation of automation tools. Artificial intelligence is a broader field than many realize, and one related sphere is machine learning. Machine learning is the use of computers to change workflow behaviors without having to explicitly program the basis for these behaviors. The purpose of this post is to describe how this will affect firms that design, detail, or install mechanical, electrical, and plumbing (MEP) systems.

Two things are pivotal to the use of machine learning – data and automation. Data is used as an input to machine learning algorithms; machine learning is an outgrowth of statistics and, just like statistics, data underpins machine learning. Automation is what differentiates machine learning from data sciences, such as data mining. Machine learning algorithms change the automation behaviors and continue to adjust their parameters for ongoing optimization. You can't have machine learning without either data or automation. 

The MEP industry needs to make two changes to fully leverage machine learning:

1. Systems of Data

An example of data in MEP is thermostat data. We can all imagine the trend lines of thermostat data over time. Some of us may even have access to this data. But thermostat data alone is not a meaningful metric and would be worthless in a machine learning context. We don't need to gather more thermostat data, we need to gather and make accessible all the other data involved in the system. A given system goes through a series of steps during design and construction, which all produce data. That data is all related to the data streaming from the thermostat. What was the design load assumption? Is the system performing as designed? Is the occupancy load the same as expected? 

Machine learning tools will be able to crunch the numbers, but we, as an industry, need to make our numbers accessible.

2. Automate, Automate, Automate

Without automation, machine learning can't make the changes necessary to achieve the optimization task. The use of automation is the smarter way to work. We recently spoke with a vice president of a large mechanical subcontracting firm who was in charge of technology processes. He said, "I don't want to hire modelers anymore. I want to hire automation experts and algorithm wranglers." In our opinion, this is the wave of the future. Gone are the days of vellum and light tables and long, tedious hours looking for clashes. With recent advancements in scanning and 3D technology, we can give the computer precise plans and routing specifications and let the computer do the work for us.

The way we work is changing rapidly and machine learning is poised to completely reinvent MEP modeling. By embracing these changes now, we can stay a step ahead of the curve and be at the forefront of a new era in AEC.


As you may have noticed, we've been doing a lot of writing about the impact of computational BIM, generative design, and the future of how we work. One thing we haven't done is turn some of this thinking into more rigorous analysis. But now we're ready to do that! 

If you work for a general contractor, subcontractor, or design firm and want to collaborate on whitepapers that quantify how computational methods of working will change our industry, reach out to us and let’s talk about what we can do. We're open to collaborations worldwide and have lots of ways of measuring performance indicators to gain insight into change in our industry. Contact Brett Young at

Tags: MEP Coordination Machine Learning

Latest ClashMEP Demonstration Videos

Clash Filtering

File-vs-File Clash Testing

Webinar with A2K Technologies

Drop us your email address and stay connected with us!