(429c) Machine Learning-Based Gas Product Estimation and Feedback Control of an Experimental Proton Membrane Reactor

Steam methane reforming (SMR) is the most widely used hydrogen generation method in the chemical manufacturing industry. However, SMR is very energy intensive and requires pressure swing adsorption to separate hydrogen from the side products. An alternative method using proton membrane reformers (PMR), which is an electrochemical process powered by renewable resources to produce hydrogen from methane and water vapor, is more environmentally friendly process than the traditional SMR in terms reducing emission. Additionally, the ceramic membrane allows electrochemical separation and compression of hydrogen from the reactants, thus enable controlling the selectivity for CO and CO₂ by shifting the reaction equilibrium [1]. However, scaling up this complex technology for industrial applications poses challenges, including a lack of comprehensive process control strategies. This study focuses on developing efficient process control strategies for proton membrane reformers (PMRs) to optimize their operation and enable widespread adoption for sustainable hydrogen production. To accomplish this, machine learning models and model predictive control (MPC) will be implemented to optimize PMR operation.

An experimental PMRs system is developed, automated and tested at UCLA [2]. Data generated from this system is then used to develop ML model, such as Physics Informed Neural Networks (PINN), recurrent (RNN), and feedforward neural networks (FNN) that is expected to comprehensively capture the process behavior. Controlling the PMR system will require a multi-output-multi-input (MIMO) control system. ML-based estimators will be incorporated into an MPC to drive the process to most profitable or energy optimal set points. Thus, a comprehensive optimization scheme will be established to adequately control the reactor performance. Model construction will be supported by steady state and dynamic experimental applications to ensure comprehensive system performance prediction and control, and results will be compared with traditional SMR.

References:

[1] Malerød-Fjeld, Harald, et al. "Thermo-electrochemical production of compressed hydrogen from methane with near-zero energy loss." Nature Energy 2.12 (2017): 923-931.

[2] Richard, Derek Michael. Development and Testing of Two Lab-Scale Reactors for Electrified Steam Methane Reforming. University of California, Los Angeles, 2021.