Virtual Sensor for Door Opening Detection and Anomaly Detection Using Machine Learning

Currently, several studies are being developed using machine Learning with the aim of replacing real sensors with virtual sensors. One of the biggest challenges is obtaining quality data to enable the system modeling. This article proposes the replacement of a light dependent resistor sensor, applied to a refrigerator, which detects whether the door is open or closed and an anomaly test when the door is forgotten open for too long. The use of tiny machine learning is applied to ESP32 to enable the use of regression and classification models to detect the door opening and anomaly. The tested metrics for the regression were the root mean squared error. The comparative results between the virtual and real sensors are satisfactory with an accuracy of 99.59% for classification and a root mean squared error value of 0.01 for regression. Furthermore, a prototype was developed, and the model was embedded in the ESP32 microcontroller.