A refrigerator is a central part of modern kitchens. Losing cooling can result in hundreds, or even thousands, of goods spoiling. Even more important, sudden losses of medicines or vaccines that heavily depend on refrigeration can have a significant impact on those who need them. Swapnil Verma was determined to find a solution to this problem. He came up with the idea of incorporating a simple machine-learning model into a device that could detect failures.
Verma started by looking for failure modes when collecting datapoints to train the model. For example, a decrease or change in temperature, or an abnormality. Verma chose to use the Arduino Nano 33 BLE Sensor with its temperature/humidity, ambient light and other sensors. Data is then streamed via Bluetooth(r) LE from the Arduino Nano 33 BLE Sense to a Portenta HS7 and logged to a microSD Card. Verma uploaded the CSV files to Edge Impulse Studio, where she trained an anomaly detection model which could detect when the refrigerator is not operating correctly.Verma suggested that alerts could be sent during deployment, even though it doesn’t involve sending them at the moment. This is especially true for medical fields.
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.
