Inspiration
retrofitting a building or factory with new energy efficient equipment - something thats becoming mandatory due to regulation in California (2025/2026) and across the country - is an incredibly difficult process that requires months of planning, dozens of contractors, and potentially millions of dollars just to get an idea of what to do.
Our goal is to reduce that burden and make the process easier for small to mid-sized companies that have a few buildings or factories in their portfolio by creating rough estimates that are 'good enough' and don't need the same extent of research as mega-corporations.
This will solve two problems: 1) Help reduce the burden that smaller corporations feel when there's a talent crunch, especially for something that's becoming required by regulations and needed for compliance. 2) Save money, time, and energy by creating an accurate assessment of how much a building can be improved by a retrofit, and choosing which buildings to retrofit first for the largest impact.
What it does
We built a web app that allows a user to input a video of the interior of a building, room, or factory. This is then passed into a computer vision model that counts the number of windows, light bulbs, and light bars inside said area. Afterwards our tool uses calculations, conversions, and estimates from the Environmental Protection Agency, Department of Energy, and various non-profits to estimate the electricity cost of older equipment versus newer equipment. Then we calculate how much of a difference this makes across an entire year, and we plug that data into datasets from the EPA and National Renewable Energy Laboratory and calculate how much of a reduction there is in carbon emissions. We graph out these results with forecasting using the NREL algorithm up to 2050 for carbon emissions, and share results of electricity savings for the year
How we built it
We used a web app running flask/node.js, a cv model from aws sagemaker/rekognition, and a python script internally. We used a lot of resources/datasets/etc from online government organizations.
Challenges we ran into
Setting up a cv model to accurately classify was tough because we had to balance the size of the video, and the accuracy of the results. We also had to tie this all into a web app which was difficult when trying to deal with errors and bugs. The largest time sink though was trying to do the calculations and figure out the math behind finding how much electricity an average window loses, especially when compounded with doing that for various areas across the nation.
Accomplishments that we're proud of
Getting aws set up to accurately (enough) run and find the number of specific appliances was something we were very proud of, and getting the math figured out for all this was also very good.
What we learned
How to actually deploy models in the real world and how to do background research on important problems and how to whiteboard and solve them. That was what we learned the most about, and coding was also a close second after that.
What's next for Factory retrofit webapp
We're going to expand this to more appliances, rather than just windows, lights bulbs, and light bars so it can be used in a factory or building to truly save time for a building owner or manager.
Log in or sign up for Devpost to join the conversation.