Multi gas sensors using one nanomaterial, temperature gradient, and machine learning algorithms for discrimination of gases and their concentration

Nguyen Xuan Thai; Matteo Tonezzer; Luca Masera; Hugo Nguyen; Nguyen Van Duy; Nguyen Duc Hoa

doi:10.1016/j.aca.2020.05.015

Multi gas sensors using one nanomaterial, temperature gradient, and machine learning algorithms for discrimination of gases and their concentration

Anal Chim Acta. 2020 Aug 8:1124:85-93. doi: 10.1016/j.aca.2020.05.015. Epub 2020 May 13.

Authors

Nguyen Xuan Thai¹, Matteo Tonezzer², Luca Masera³, Hugo Nguyen⁴, Nguyen Van Duy⁵, Nguyen Duc Hoa⁶

Affiliations

¹ ITIMS, Hanoi University of Science and Technology, Hanoi, Viet Nam; Vietnam Metrology Institute, 8 Hoang Quoc Viet Road, Hanoi, Viet Nam.
² IMEM-CNR, Sede di Trento - FBK, Via Alla Cascata 56/C, Povo, TN, Italy; University of Trento, Via Calepina 14, Trento, Italy. Electronic address: matteo.tonezzer@cnr.it.
³ DISI, University of Trento, Via Sommarive 9, Povo, Trento, Italy.
⁴ Uppsala University, Department of Material Science, Uppsala, Sweden.
⁵ ITIMS, Hanoi University of Science and Technology, Hanoi, Viet Nam. Electronic address: nguyenvanduy@itims.edu.vn.
⁶ ITIMS, Hanoi University of Science and Technology, Hanoi, Viet Nam.

PMID: 32534679
DOI: 10.1016/j.aca.2020.05.015

Abstract

In this work, four identical micro sensors on the same chip with noble metal decorated tin oxide nanowires as gas sensing material were located at different distances from an integrated heater to work at different temperatures. Their responses are combined in highly informative 4D points that can qualitatively (gas recognition) and quantitatively (concentration estimate) discriminate all the tested gases. Two identical chips were fabricated with tin oxide (SnO₂) nanowires decorated with different metal nanoparticles: one decorated with Ag nanoparticles and one with Pt nanoparticles. Support Vector Machine was used as the "brain" of the sensing system. The results show that the systems using these multisensor chips were capable of achieving perfect classification (100%) and good estimation of the concentration of tested gases (errors in the range 8-28%). The Ag decorated sensors did not have a preferential gas, while Pt decorated sensors showed a lower error towards acetone, hydrogen and ammonia. Combination of the two sensor chips improved the overall estimation of gas concentrations, but the individual sensor chips were better for some specific target gases.

Keywords: Gas sensor; Machine learning; Nanowires; Selectivity; Tin oxide.