Machine Learning-Based Kinetic Modeling of the CO₂ Methanation Reaction over an Industrial Catalyst.

Area of Science:

• Chemical Engineering
• Catalysis
• Machine Learning

Background:

• Industrial catalysts often have confidential compositions, hindering mechanistic understanding and kinetic model development.
• Accurate kinetic models are crucial for effective reactor design, but are difficult to obtain for commercial catalysts.
• Machine learning (ML) offers a potential alternative for kinetic modeling without requiring detailed reaction mechanisms.

Purpose of the Study:

• To investigate the efficacy of ML-based regressions in describing the kinetics of CO2 methanation using a commercial catalyst.
• To compare the performance of ML models against traditional kinetic modeling approaches (power law and LHHW).
• To assess the applicability of ML-derived kinetics in reactor modeling and design.

Main Methods:

• Kinetic experiments were conducted on a commercial Ni/ZrO2-based CO2 methanation catalyst in a fixed bed reactor.
• A comprehensive dataset was generated across various temperatures, partial pressures, and space velocities.
• Gaussian Process Regression and a 1-layer neural network were employed as ML models and compared to power law and LHHW models using RMSE.

Main Results:

• ML-based models, specifically Gaussian Process Regression and a 1-layer neural network, demonstrated superior performance compared to the power law model.
• The ML models achieved comparable accuracy to the Langmuir-Hinshelwood-Hougen-Watson (LHHW) model in describing the catalyst's kinetic regime.
• When integrated into a reactor model, ML-derived kinetics accurately predicted methane yield, even outperforming LHHW in nonisothermal conditions.

Conclusions:

• ML-based kinetic models are a viable and promising approach for CO2 methanation, especially when detailed mechanistic insights are unavailable.
• These ML models can be effectively utilized for robust reactor design, offering comparable or superior performance to traditional methods.
• The study validates the use of ML for developing predictive kinetic models from experimental data, accelerating process development.