A Machine Learning Enhanced MEMS Thermal Anemometer for Detection of Flow, Angle of Attack, and Relative Humidity

By optimizing machine learning (ML), the accuracy of a thermal anemometer has been improved (511%) when compared to conventional linear regression. In addition, ML has extended the functionality allowing for additional angle of attack and humidity information to be determined. The miniature sensor (0.16 cm
²
) has been fabricated with a straightforward silicon on insulator (SOI) fabrication procedure. The sensor paired with ML could offer a cost-effective, small, and reliable solution for monitoring air in industrial and agricultural sensor grid applications, such as data centers and greenhouses. This proof of principle shows that thermal anemometers can have their accuracy and functionality enhanced through ML, enabling the estimation of multiple physical parameters with a single sensor.