In the past year, AI technology has made its way to the forefront of the public eye.  Many industries, including commissioning, have been left wondering how the technology will change the way they do business.  In this blog post, I will explore ChatGPT’s recommendations for possible applications to our industry, and see how useful the technology can be in practice. ChatGPT recommends a few ways it can assist a HVAC commissioning firm:  Technical Assistance, Procedure Guidance, Documentation Support, Regulatory Compliance, Client Communication, and Energy Efficiency.

Technical Assistance:

Here’s a real scenario from one of our projects that I posed to ChatGPT, where a damper actuator had begun slipping: “We have a classroom where the owner has complained about it being too cold because the space is 60 deg F.  The space is served by a Variable Air Volume (VAV) box, where the building control system shows the leaving air temperature is 110 deg F and airflow is 10% of design.  The motorized air damper is commanded 100% open.  What could be wrong with the equipment, and how can I troubleshoot this?”  ChatGPT immediately zeroed in on the correct answer, identifying the low airflow as a key issue and suggesting we check for obstructed ductwork or a faulty damper actuator.  Overall I am impressed with the options it provided in response, but from a user standpoint, it’s difficult to know when enough information has been provided for it to provide a good diagnosis.  Would it provide a better response knowing the Testing, Adjusting, and Balancing report was complete?  What if it knew the equipment was not having problems a few weeks ago?  What if it could analyze images and look at the ductwork layout on the HVAC drawings?

Procedure Guidance:

When asked, ChatGPT was not able to provide commissioning test procedures for three different models of Split DX Units.  I chose these because they would be an easy test case.  Split DX Units are typically stand-alone and do not rely on other HVAC or controls equipment in the building.  My first equipment model request attempt failed because the equipment was manufactured after January 2022, the chatbot’s most recent knowledge update.  For the other two equipment model requests, ChatGPT could not access the manufacturer’s commissioning procedures and told me I would have to find this on my own.  Unfortunately, ChatGPT was also not able to differentiate between the manufacturer’s commissioning forms (“startup”) that are completed by the installer, and the performance-based commissioning functional testing that is done by another entity, like Elevate.  For this exercise Chat GPT didn’t have, and could not collect, enough information to create a good functional testing plan.  This may be possible someday if it has access to all of a project’s product data, shop drawings, and control sequences but it doesn’t have a way to digest this information in its current implementation.

Documentation Support:

When creating commissioning reports, it can be difficult to find the right words for an executive summary.  When ChatGPT was given the bulk of an old project’s commissioning report as a prompt, it had no problem providing a well-written executive summary.  If this were for a real project, the executive summary provided could have served its purpose with only a few edits needed.  One thing it did especially well was summarize the information that was found throughout the document in a way that would require a full read-through of the content.  This could also be helpful to finish a commissioning report without knowing as much about the project’s history (if, for example, the original project PM is out on paid parental leave or similar). I’ll be circling back to this again in the future to find more applications in this category.

So far, ChatGPT has impressed me with its capabilities but comes with several major caveats.  In the next blog post, I’ll be exploring ChatGPT’s capabilities to assist with Regulatory Compliance, Client Communication, and Energy Efficiency and summarizing my overall findings.