Skip to main content

Research Repository

Advanced Search

Design of a hybrid artificial intelligence system for real-time quantification of impurities in gas streams: application in CO2 capture and storage. [Dataset]

Contributors

Efenwengbe Nicholas Aminaho
Data Collector

Ndukaegho Sabastine Aminaho
Data Collector

Konyengwaehie Augustus Aminaho
Accompanist

Abstract

This study proposed a new sensor calibration methodology and the design of a hybrid artificial intelligence system for real-time quantification of impurities in gas streams. Furthermore, machine learning models were developed in this study to explore how impurities in gas streams can be quantified. In the case study for the machine learning models, a nitrogen gas (N2) system with impurities was used for the demonstration. The study compares the performance of sensors in quantifying the concentrations of component gases in binary gas and multi-component gas mixtures. The binary gas mixture is made up of N2 and NO2, while the multi-component gas mixtures are mixtures of N2 with two-gas, three-gas, or four-gas combinations of NO, NO2, C3H8, and NH3; each gas mixture has a constant 10% O2 by volume (as oxygen reduction is a requirement in electrochemical sensors). The hybrid artificial intelligence system of quantifying gas concentrations is scalable and can be applied in complex gas mixtures found in different industries. For instance, in a CO2 gas stream, the binary gas mixture could represent a mixture of CO2 with NO2; while the multi-component gas mixtures could represent CO2 with two-gas, three-gas, or four-gas combinations of NO, NO2, C3H8, and NH3.

Citation

AMINAHO, E.N., AMINAHO, N.S., HOSSAIN, M., FAISAL, N.H. and AMINAHO, K.A. 2025. Design of a hybrid artificial intelligence system for real-time quantification of impurities in gas streams: application in CO2 capture and storage. [Dataset]. Gas science and engineering [online], 134, article number 205546. Available from: https://doi.org/10.1016/j.jgsce.2025.205546

Acceptance Date Jan 9, 2025
Online Publication Date Jan 11, 2025
Publication Date Feb 28, 2025
Deposit Date Jan 17, 2025
Publicly Available Date Jan 17, 2025
Publisher Elsevier
DOI https://doi.org/10.1016/j.jgsce.2025.205546
Keywords Gas mixtures; Gas streams; Gas pipelines; Sensors; Sensor malfunction; Artificial intelligence; Machine learning
Public URL https://rgu-repository.worktribe.com/output/2663051
Related Public URLs https://rgu-repository.worktribe.com/output/2662175 (Journal article associated with this Supplementary Data)
Type of Data XLSX files and the accompanying TXT file.
Collection Date Oct 31, 2023
Collection Method This study proposes a hybrid artificial intelligence system for real-time quantification of impurities in gas streams. This gas monitoring technology can be applied in different gas systems as well as in carbon capture, utilization, and storage. Experimental data used in this study were from the work of Javed et al. (2022). They measured the responses of mixed-potential electrochemical sensors (MPES) to gas mixtures. In the present study, a regression model based on Artificial Neural Network (ANN) algorithm was developed, using Python programming language, to quantify the concentrations of gas components using data from the work of Javed et al. (2022) and to demonstrate the design of a hybrid artificial intelligence system for real-time quantification of impurities in gas streams. Data from experiments conducted by Javed et al. (2022) are used in developing the machine learning model. The original data separated into training and testing data were concatenated into a single dataset with 107103 data points. The ANN and XGBoost models predicted the gas concentrations with minimal error. For both models, the error is lowest (MAE, MdAE, maximum error, and RMSE) in the prediction of NO2 concentration, followed by NO concentrations, but highest in the prediction of C3H8 concentration. The R2 value for both ANN and XGBoost models is greater than 99%, and the mean absolute percentage error for the XGBoost model is less than that of the ANN model. Also, the R2 value of the XGBoost model is higher. The ANN and XGBoost models predicted NO2 concentrations, in the binary gas mixture, with minimal error. Also, for both models, the deviation between the metrics of performance evaluation for the training and testing datasets is negligible. This study proposes a hybrid artificial intelligence system for real-time monitoring of gas concentrations in a gas stream and to identify when sensors need to be recalibrated.