Presentations

Multivariate calibration method for selective gas analysis with a single temperature-modulated sensor

Shaposhnik A.V., Moskalev P.V.¹, Vasiliev A.A.²

Voronezh State Agrarian University named after Emperor Peter the Great, 1 Michurin St., Voronezh, 394087, Russia, a.v.shaposhnik@gmail.com;

¹Moscow State University of Technology “STANKIN”, 1 Vadkovsky Lane, Moscow, 127055, Russia, moskaleff@mail.ru;

²Dubna State University, 19 Universitetskaya St., Dubna, Moscow Region, 141982, Russia, a-a-vasiliev@yandex.ru

The problem of selective gas analysis using a single semiconductor sensor based on tin dioxide modified with palladium oxide (SnO₂–PdO) is considered. In contrast to approaches requiring sensor arrays and resource-intensive data transformations, the authors propose a multivariate calibration for selective analysis (MCSA) method as a modification of multivariate calibration for quantitative and qualitative gas analysis. The idea behind MCSA is that during cyclic heating of the sensor, the time series of its resistances is treated as a multivariate vector. Regression models of the form $\hat\varphi_i = a \cdot R_i^b$ are constructed for selected elements of this vector, relating the predicted concentration of the target gas $\hat\varphi_i$ to the sensor resistance $R_i$.

The relative standard deviation $S_r$ of the set of concentration estimates obtained for different gas concentrations and compositions is used as a metric for qualitative gas analysis. If the sensor is exposed to the gas for which the model was built, the predicted concentration values ​​$\hat\varphi_i$ are consistent with the observed values, and $S_r < S_0$. For other gases, the model produces inconsistent estimates, and $S_r \geq S_0$. The critical value $S_0 = 0.45$ enabled accurate classification of hydrogen sulfide H₂S, carbon monoxide CO, hydrogen H₂, and ethanol C₂H₅OH in air at concentrations ranging from 0.5 to 200 ppm.

From a machine learning perspective, the MCSA method implements an interpretable and efficient binary classification algorithm based on comparing the empirical variance of the predicted and threshold values. Its complexity is several orders of magnitude lower than that of classification using neural networks, making it suitable for autonomous IoT devices. Thus, MCSA represents a solution to the problem of gas identification using a single sensor, combining minimal hardware requirements with high recognition accuracy. It can be adapted for other applications requiring signal identification within a given family.

This work was supported by the Ministry of Education and Science of the Russian Federation (state research contract FSFS-2024-0007).

References.

1. Shaposhnik A., Moskalev P., Vasiliev A. [et al.] Multivariate Calibration for Selective Analysis of Hydrogen Sulfide and Carbon Monoxide with Thermal Modulation of the SnO2–PdO Sensor // Chemosensors. – Vol. 13, No. 9. – 2025. – Pp. 323.

• abstract in Russian (PDF)