A Novel Pulse Heating Approach for Gas Sensors with Concentration Estimation through Back Propagation Neural Network

Abstract

This work aims to heat gas sensors with a novel approach by applying pulse heating instead of conventional DC voltage heating. Since the change in the resistance of the sensor is inapparent, a specific program is adopted to process the origin data, which extracts both rising and dropping edges of the data plot and transforms them into a bitmap. The origin data are plotted into resistance-time figure, and an algorithm is applied to extract the rising edges of each pulse. Then the edges are transformed into a pixel line and further processed. The algorithm for analysis of data and estimation of concentration is realized by utilizing multilayer back propagation neural network (BPNN) and deep learning. Practical tests are conducted on two commercial ethanol sensors MP-3B, with one of them heated by pulse and the other by stable voltage, as calibrated sensor. Currently, a relative error of 10% is achieved when the concentration varies from 0 ppm to 200 ppm. This research shows the potential of applying pulse heating approach for gas sensor testing and quantitively analyzing the concentrations of single VOC gas by BPNN.