Linear and Non-Linear Modelling Methods for a Gas Sensor Array Developed for Process Control Applications.

Area of Science:

• Chemical Engineering
• Sensor Technology
• Machine Learning

Background:

• Power to X technologies require robust process monitoring tools.
• Key gases in these processes include hydrogen (H₂), carbon monoxide (CO), methane (CH₄), and carbon dioxide (CO₂).
• Non-selective sensors necessitate advanced data analysis for accurate gas mixture composition determination.

Purpose of the Study:

• To develop and evaluate data processing models for a multi-sensor array used in Power to X process monitoring.
• To compare the performance of linear (MLR-PLS) and non-linear (ANN) models in predicting gas concentrations.
• To assess the long-term stability and prediction accuracy of sensor platforms over time.

Main Methods:

• Construction of a multi-sensor matrix utilizing commercial sensors with diverse transduction principles.
• Development and comparison of Multi Linear Regression-Partial Least Square (MLR-PLS) and Artificial Neural Network (ANN) models for gas mixture analysis.
• Validation of model performance using experimental data and evaluation on aged sensor platforms.

Main Results:

• Both MLR-PLS and ANN models achieved good concentration predictions for H₂, CO, and CO₂.
• ANN demonstrated superior prediction performance for methane (CH₄) compared to MLR-PLS.
• Aged sensor platforms showed that PLS predictions were affected by concentration offsets, while ANN predictions experienced a drop in sensitivity.

Conclusions:

• A multi-sensor array combined with machine learning offers a viable solution for gas mixture monitoring in Power to X applications.
• ANN models provide more accurate methane detection due to their ability to handle non-linear sensor responses.
• Sensor aging impacts model performance, necessitating strategies to address sensitivity drops and concentration offsets for long-term monitoring reliability.