GridSense

Inspiration

I've had too many appliances break on me suddenly, leading to hundreds of dollars in repair costs that I wish I could have avoided. So, I built GridSense to predict those problems in advance and provide the user with clear next steps to resolve the issue. With all analysis done locally and double check across a library of 240 recordings, GridSense runs smoothly on all computers, phones, and devices, regardless of their components.

What it does

GridSense is a tool that predicts issues in appliances long before they ever break. GridSense does this by analyzing a short sound clip from the user of an appliance, cross-checking it over a database of 240 different test motors, and then providing the user with a health score from 1-100 and clear next steps the user should take to prolong the appliance lifespan.

How we built it

Python 3 · Librosa & SciPy (signal processing) · scikit-learn (RandomForest) · Streamlit (dashboard) · Matplotlib (waveform + spectrogram). A physics-based synthetic-audio generator lets the entire system be trained and demoed with zero physical hardware: motor hum, high-frequency grinding, and low-frequency wobble are synthesised with controllable severity and realistic background noise.

Challenges we ran into

Many of the challenges I ran into regarded sourcing information, like the example sound clips, and analyzing user inputted sound. At first, the analysis was wildly inaccurate, even when inputting clean motor sounds. However, through many trials and errors, I was able to resolve these issues and create a clean, working tool that actually predicted motor issues accurately.

Accomplishments that we're proud of

The biggest accomplishment that I am proud of is how accurate GridSense is. So far, GridSense has had an accuracy rate of 98.3%: absolutely astounding. This result genuinely makes me happy after hours upon hours of trial and error to smooth out each bug and make the user interface clean and accessible.

What we learned

The biggest thing I learned from this project is that impact has no correlation with project size. GridSense, a relatively small project, has immeasurable effects on lives across the world, ultimately with the potential to save millions of dollars yearly for the average citizen. However, GridSense didn't need any large backend, months of research and development, or a fully fleshed out idea. Instead, GridSense works on the premise of an idea that has potential and then building it out along the way. GridSense has taught me that ideas do not need to be grand at the beginning, they just need to have some potential, and a lot time for iterations.

What's next for GridSense

Field recordings from real machinery to validate accuracy beyond synthetic data; more fault classes (electrical arcing, cavitation, gear-mesh wear); a trend view that tracks a machine's Health Index over time; and a Raspberry Pi edge build.