Multi-Mems Differential Pressure Sensor Elements-Based Airflow Sensor with Neural Network Model

Haneda, Kotaro; Matsudaira, Kenei; Takahashi, Hidetoshi

doi:10.1109/mems49605.2023.10052249

Cited by 1 publication

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, we attached the calibrated sensor to the toy drone and demonstrated its measurement during flight. At the most recent conference, we briefly reported on the airflow sensor without parts of the simulation [36]. Additionally, the amount of input data points of each wind speed and direction was not equal in that case; therefore, they were equalized in this study.…”

Section: Introductionmentioning

confidence: 99%

Neural Network-Based Airflow Vector Sensor Using Multiple MEMS Differential Pressure Sensors

2023

View full text Add to dashboard Cite

As airflow information is significant in various fields, several airflow measurement devices have been developed. Differential pressure (DP)-type sensors, including pitot tubes, are extensively used because they can directly measure dynamic pressure due to airflow. However, this method has a disadvantage in that it is difficult to measure some wind directions because of the non-monotonic pressure distribution around the device. This study proposes a compact sphere airflow vector sensor using a neural network model that compensates for nonmonotonicity. It consists of three built-in microelectromechanical system (MEMS)based DP sensor elements that simultaneously measure the DPs around the spherical sensor housing. The measured DPs are converted into 2D wind speed and direction using a neural network. The network model was trained using the simulated output of each sensor, and it was confirmed that the proposed design guideline contributed to achieving high accuracy. An airflow sensor was fabricated on the basis of the designed model. A network model was trained using the sensor responses in a wind tunnel experiment. The wind speed and direction accuracies obtained by the neural network model for 2-10 m/s and 0°-359° were 0.24 m/s and 3.62°, respectively. Finally, we demonstrated a drone flight test by attaching the calibrated sensor to a toy drone. This study is expected to contribute to realizing highly accurate low airflow measurement devices.INDEX TERMS Airflow sensor, neural network, differential pressure sensor, machine learning, drone. FIGURE 1. Conceptual diagram of the proposed airflow vector sensor.

show abstract

Section: Introductionmentioning

confidence: 99%