Deep Neural Network Based Modelling of Chemisorption Process on Surface of Oxide Based Gas Sensors: DNN BASED MODELLING OF CHEMISORPTION PROCESS FOR GAS SENSORS

Rahul Gupta; Pradeep Kumar; Dinesh Kumar

doi:10.56042/jsir.v82i11.1978

Authors

Rahul Gupta J. C. Bose University of Science and Technology, YMCA, Faridabad 121 006, Haryana, India & University Institute of Engineering and Technology, Kurukshetra University, Kurukshetra 136 119, Haryana, India
Pradeep Kumar J. C. Bose University of Science and Technology, YMCA, Faridabad 121 006, Haryana, India
Dinesh Kumar Department of Electronic Science, Kurukshetra University, Kurukshetra 136 119, Haryana, India & Gurugram University, Gurugram 122 003, Haryana, India

DOI:

https://doi.org/10.56042/jsir.v82i11.1978

Keywords:

Chemisorption, Deep neural network, Gas sensor, Numerical modelling

Abstract

The sensor response of the metal oxide based gas sensor has been simulated using Deep Neural Network (DNN) model. The neural network designed for the modelling of the sensor has single input layer, three hidden layers and single output layer. The linear regression algorithm has been used to compute the electrical conductance of the sensor at given temperature and pressure. The data generated through modified Wolkenstein method has been used for training, validation and testing of the developed network. The data for materials Tin (IV) oxide (SnO2), Tin (II) oxide (SnO) and Copper (I) oxide (Cu2O) with different Eg values has been utilized. The other input parameters like Temperature, ND, NC, NV, EF−ESSand ECS−EF are varied for the specific range to collect a variety of data for calculation of electrical conductance of the sensor. The total data used for training, validation and testing was 1,90,512 data points. The plots for training, validation and testing phase have been plotted. The sensor response computed through the proposed model is validated with the results of already published mathematical model. The sensor response shows steep change when the gas concentration of the target gas reaches above 10−8 atm. The proposed model can be retrained or transfer learning can be applied for using the same model for other types of materials for gas sensing applications.