HVAC Performance Optimization

For our Challenge Climate Change contest (done in collaboration with QuickLogic and hackster.io) we had two primary categories:  battery powered applications and line powered applications.  Previously we discussed the battery powered winners, now it’s time to take a look at the clever line powered application winners.

Our first place finisher in the line powered category was Ralph Yamamoto’s HVAC Performance Optimization Project.  His project was focused on tackling climate change by improving home energy efficiency.  As we all know, heating and cooling a home consumes more energy than just about any other residential use.  Any application that can save even a bit of energy in a single home could have a huge impact if deployed on a large scale.

Ralph’s approach tackled two different HVAC-related issues.  The first was comparing temperature and humidity measurements for both floors of his home and then adjusting the airflow to get each floor more evenly matched.  That way neither floor would be too hot nor too cold and the amount of heating or cooling coming from the HVAC system would be “just right” for the whole house. The second used AI-based analysis of vibration and sound data to ensure that his HVAC equipment was running at optimal efficiency and didn’t have any potential operational problems that might require maintenance work.

Since we’re particularly interested in the application of AI for helping to solve the problem of climate change, let’s zoom in a bit on that second issue.  Ralph focused initially on vibration sensing using the accelerometer available as part of the QuickFeather open source development kit from QuickLogic.  That same kit features the QuickLogic EOS S3 SoC with an embedded Arm Cortex MF4 processor, which he used to implement AI-based analysis of the real-time vibration activity exhibited by his HVAC system.  He then added a Wi-Fi module that enabled the QuickFeather board to communicate wirelessly with his computer.

Next, he used the SensiML Data Capture Lab in our free Community Edition of the Analytics Toolkit to train his system.  He mounted his box with the QuickFeather board (including the accelerometer) to the side of his furnace and began streaming data as he turned his furnace from off to on.  That gave him a dataset showing both “no operation” and “normal operation” for his furnace.  That dataset was then used to train the software to recognize the state of the furnace and ultimately to create an ultra-low-power tinyML model capable of running on processors as small as 8-bits.  He tested his system and found it to have a very high degree of accuracy, though additional work was required to add sound-based problem detection and the identification of nuanced conditions such as dirty furnace filters.

Although this system demonstrated basic functionality, we were impressed by the creativity of Ralph’s ideas and the fact that his design created a nice foundation for adding new and more sophisticated features.  We thought this was a good use of AI technology with the potential for significant impact due to the broad use of HVAC systems around the world.  

See Winning Hackster Line Powered Project Details:

HVAC Performance Optimization by Ralph Yamamoto