An ultrasonically catalyzed conductometric metal oxide gas sensor system with machine learning-based ambient temperature compensation

Abstract

The ultrasonic catalysis method provides a new means of enhancing sensing performance of some types of gas sensors. But generating the ultrasound consumes electric energy and increases the system mass. Thus, it is significant to lower the energy consumption and mass of the ultrasonic excitation structure as much as possible, while enhancing the sensing performance. In this work, a thin conical metal shell excited by a piezoelectric disk has been proposed, serving as the focusing ultrasonic transducer and protection case for a SnO2 gas sensor. The ultrasonic transducer’s total mass is only 11 g. The experiments show that the ultrasonic transducer can significantly reduce the low detection limit (LDL), and measured LDL of the sonicated sensor for H2 can be as low as 107 ppb (1/3 of the same sensing element not sonicated) with a response of 4.6%. In this case, the ultrasonic transducer’s power consumption and temperature rise is 146 mW and 3 ℃, respectively. The low temperature rise indicates that the LDL may be further decreased by increasing the ultrasonic vibration. The effect of ambient temperature change on predicted H2 concentration is compensated by two convolutional neural networks for the low (1–100 ppm) and ultralow (≤ 1 ppm) concentration ranges, and the relative error of concentration prediction in these two concentration ranges is 5.7% and 12.3%, respectively.