This week, I presented our Proposal. I spent last weekend sorting out the elements we wanted to include with my team, specifically the use case requirements. I created some of the figures and worked on the technical challenges, solution approaches, and testing, verification, and metrics slides. I also practiced and revised how I should present and how I should answer questions. We were able to get a clear picture of the project we want to accomplish this semester.
I also looked into locating HVAC ducts for physical testing. I called a couple of HVAC duct repair and replace companies and inquired about where they store damaged ducts, but they were focused on business, i.e. they wanted me to get my ducts repaired even though I explained that I have no ducts. We, as a team, are continuing our search for used ducts but as a backup, I have found HVAC ducts of different sizes to buy on Amazon. I have also found fiberglass board to buy on Amazon to apply to our sheet metal ducts so that we are able to test our system on fiberglass duct board. Fiberglass duct board consists of standard sheet metal ducts that have a fiberglass lining that serves as an extra layer of insulation
I looked extensively into acoustic features that indicate corrosion and leaks. Although I had a general understanding of the type of acoustic signature that indicates these defects, I was able to find more specific information as well as some signal processing techniques to employ from research papers. These papers were done on a variety of sheet metal materials, but I was not able to find research papers on the acoustic signatures of defects on fiberglass duct board. Fiberglass duct board helps to dampen noise. My worry is that it will be difficult to generalize the ML model to fiberglass duct board because the acoustic signature that we capture with the robot will be dampened, and it may be hard to extract the acoustic features from the data. I am also worried about the conversation my team and our supervising professor have been having regarding the data to train the ML model. We had previously assumed that we could find a dataset with labeled acoustic data from the duct materials of our choice, but this is proving difficult. Our professor has recommended that we gather our own data based on our robot’s specific ability to create and collect acoustic data. Currently, we are working on getting our robot together while looking at potential datasets. Whichever dataset we use must, at the very least, include acoustic data labeled with the defects we are focusing on (corrosion and air leaks) from the duct material we are focusing on (sheet metal and fiberglass duct board).
Next Week Goals:
Looking through datasets to choose one that meets the minimum requirements stated above
Finding a viable HVAC duct source and purchasing/obtaining the ducts
Developing the design of the acoustic data processing pipeline
My work for this week can be found here: https://docs.google.com/document/d/1RLk7pQB5754Au6vyzFaVA-jm3ABx_Y89l1xmLTPi_bc/edit?tab=t.0