Diffusive Representation: A Powerful Method to Analyze Temporal Signals from Metal-Oxide Gas Sensors Used in Pulsed Mode

Cyril Tropis,Germain Garcia,Nicolas Dufour,Chaabane Talhi,Gerard Montseny,Philippe Menini,Frédéric Blanc,Bernard Franc

doi:10.3390/electronics10212578

Abstract

The main objective of this work was to find the most efficient method to interpolate metal oxide gas sensor used in a pulsed-temperature operating mode. This pulsed thermal profile is obtained by applying 6 power steps of 2 s each on the heater resistor. The experimental values of the sensing layer resistance, with a sampling time of 4ms, were interpolated by using two different static methods: a polynomial modelling and a neural network modelling, and one dynamic method: the diffusive representation. Then, the results have been compared in terms of precision and number of useful output data, as minimum as possible for high performance and rapid data treatment which is great of interest in embedded systems. The best results are obtained with the diffusive representation; it allows converting 500 measurements into 11 output coefficients.

Highlights

Metal-oxide thin film sensors have been widely used for gas sensing applications thanks to their sensitivity toward a large variety of gases [1]
In order to simplify the output data of the sensing resistor, several models have been studied: two well known mathematical modeling and an approach based on a linear modeling
That’s why we found interesting to use the diffusive model to represent the sensor response as a linear dynamic model of input-output

Summary

Introduction

Metal-oxide thin film sensors have been widely used for gas sensing applications thanks to their sensitivity toward a large variety of gases [1]. The transient behavior of the sensing resistance, located just after a temperature change, is used to discriminate gases by comparing signal shapes [10,11] In these cases, it is necessary to use powerful interpolation systems associated with mathematical analysis, as discriminant factorial analysis or neural network for example [15–23]. Many studies have tried to model the sensors responses by physical or physico-chemical models, but always for thermodynamically stable behaviors, that is to say at a constant temperature or variation until steady state [24–26]. It is almost impossible in pulsed mode, the interest of using mathematical modeling or interpolation

Objectives

Results

Conclusion