The refrigerator is one of the centerpieces in a modern kitchen, and experiencing a loss in cooling can lead to hundreds or even thousands of dollars of spoiled goods. Perhaps even more importantly, a sudden loss of medications or vaccines that heavily rely on refrigeration can have a big impact on the people that need them. Swapnil Verma wanted to solve this problem, so he came up with an idea to incorporate a simple machine learning model into a device that could monitor for failures.

When gathering datapoints for training the model, Verma began by identifying different failure modes, such as a decrease in temperature, change in humidity, or simply an abnormality. He opted to use an Arduino Nano 33 BLE Sense along with its built-in temperature/humidity and ambient light sensors. From here, data is streamed over Bluetooth® LE to a Portenta H7 and logged to a microSD card. Verma then uploaded the resulting CSV files to Edge Impulse Studio and trained an anomaly detection model that could recognize when conditions inside the refrigerator are incorrect.

Although the deployment doesn’t currently involve sending alerts, Verma did suggest that the feature could be added in the future, especially for the medical field. Want to dive into the details of project? Check out his tutorial on Edge Impulse as well as here in the Edge Impulse Studio.