A Two-Dimensional Flow Sensor with Integrated Micro Thermal Sensing Elements and a Back Propagation Neural Network

Ruiyi Que,Rong Zhu

doi:10.3390/s140100564

Abstract

This paper demonstrates a novel flow sensor with two-dimensional 360° direction sensitivity achieved with a simple structure and a novel data fusion algorithm. Four sensing elements with roundabout wires distributed in four quadrants of a circle compose the sensor probe, and work in constant temperature difference (CTD) mode as both Joule heaters and temperature detectors. The magnitude and direction of a fluid flow are measured by detecting flow-induced temperature differences among the four elements. The probe is made of Ti/Au thin-film with a diameter of 2 mm, and is fabricated using micromachining techniques. When a flow goes through the sensor, the flow-induced temperature differences are detected by the sensing elements that also serve as the heaters of the sensor. By measuring the temperature differences among the four sensing elements symmetrically distributed in the sensing area, a full 360° direction sensitivity can be obtained. By using a BP neural network to model the relationship between the readouts of the four sensor elements and flow parameters and execute data fusion, the magnitude and direction of the flow can be deduced. Validity of the sensor design was proven through both simulations and experiments. Wind tunnel experimental results show that the measurement accuracy of the airflow speed reaches 0.72 m/s in the range of 3 m/s–30 m/s and the measurement accuracy of flow direction angle reaches 1.9° in the range of 360°.

Highlights

Two-dimensional flow measurement is becoming more and more important in many applications such as meteorology, drag reduction research for aircraft and vessels [1,2,3,4], biomedical flow detection [5] and control enhancement for Unmanned Air Vehicles (UAVs)/Micro Air Vehicles (MAVs) [6,7,8,9]
The measuring probe of the 2-Degree Of Freedom (DOF) sensor was made up of a long wire attached to the center of a cross-shaped beam with strain gauges on the four roots
When establishing the Back Propagation (BP) neural network, we found that the direction angle of the flow vector was not a good output variable for the network because when the vector angle varies from 0°to 360°, the output of the sensing element follows approximately sine or cosine law that is not consistent with the monotonicity of angle variation

Summary

Introduction

Two-dimensional flow measurement is becoming more and more important in many applications such as meteorology, drag reduction research for aircraft and vessels [1,2,3,4], biomedical flow detection [5] and control enhancement for Unmanned Air Vehicles (UAVs)/Micro Air Vehicles (MAVs) [6,7,8,9]. Conventional techniques are mostly based on Pitot tubes (including hemispherical nose probes) [10,11,12,13] or electromechanical self-orienting vanes [12,13,14], which usually protrude outside the testing body and disturb the flow they measure, need hard mechanical ties and/or intrusive pneumatic links inside. Their fabrication and packaging processes are generally elaborate and do not meet practical requirements. The sensor was hard to fabricate and the structure was fragile and broken

Methods

Results

Conclusion