Intelligent Process Control for Carbon Capture Membrane Systems

Thesis Abstract

The study addresses the performance and operability challenges of carbon capture membranes (CCMs) in industrial gas streams, where conventional control strategies fail to cope with dynamic feed conditions, membrane aging, and process disturbances, leading to suboptimal CO2 separation, energy penalties, and reduced system reliability. The aim is to develop an intelligent, ICT-driven process control framework that integrates real-time sensing, data analytics, and model-based optimization to enhance separation efficiency, energy efficiency, and operational robustness of CCM systems. Specific objectives include (i) characterizing the dynamic behavior and aging effects of polymeric CCMs under variable feed compositions; (ii) designing a hybrid control architecture that combines model predictive control (MPC) with machine learning (ML) surrogates for rapid state estimation and fault detection; (iii) developing adaptive calibration procedures for membranes and modules using Bayesian updating and online parameter identification; (iv) evaluating energy savings and CO2 purity improvements under simulated and pilot-scale conditions; and (v) formulating guidelines for implementation in industrial units with consideration of cyber-physical security and data governance. The methodology adopts a mixed-methods research design combining experimental characterization, numerical modeling, and simulation-based validation. The population comprises industrial CCM modules deployed in natural gas sweetening and syngas purification contexts. A laboratory-scale pilot unit, consisting of a 10-module CCM stack with a nominal permeance of 1.2 × 10?7 mol m?2 s?1 Pa?1 and a target CO2/N2 selectivity of 40, will be used for controlled experiments. A sample of 30 operational cycles, each lasting 4 hours, will be conducted under systematically varied feed pressures (2–8 bar), CO2 partial pressures (0.1–0.8 bar), temperature (25–60°C), and flow rates, to capture transient dynamics and aging effects over a 6-month extended test campaign. Data collection instruments include high-resolution inline CO2 and CH4 sensors (ppm level), pressure transducers (0.1% full-scale accuracy), membrane permeance trackers, and automated gas chromatography–mass spectrometry (GC-MS) for off-line validation. The analytical framework will integrate (i) system identification using nonlinear autoregressive with exogenous inputs (NARX) models and sparse Bayesian learning to derive ML-based surrogates of membrane performance; (ii) model predictive control (MPC) with economic objectives and constraints on CO2 purity, recovery, and energy consumption, supplemented by a bespoke anomaly detection module using isolation forest and Hawkes-process-informed fault prognostics; (iii) adaptive parameter estimation via recursive least squares with forgetting factors and Bayesian?? to compensate for aging effects; and (iv) a probabilistic reliability assessment using Monte Carlo simulations to quantify risk under feed disturbances and membrane degradation. The theoretical underpinning draws on control theory (MPC, observer design), process systems engineering, and data-driven modeling, with explicit reference to the theory of robust optimization and the extended Kalman filter for state estimation. Validation will compare predicted and measured performance metrics, including CO2 purity, recovery, energy intensity (kWh per tonne CO2 captured), and membrane module life-cycle cost, using regression analysis and ANOVA to assess significance across scenarios. Expected findings include (i) improved dynamic response and faster stabilization of CO2 removal following feed transients, (ii) quantified energy reductions up to 15–25% relative to baseline PID control through optimized vacuum and sweep gas strategies, (iii) enhanced fault detection sensitivity for membrane aging indicators such as selectivity drift and permeance decay, (iv) robust control performance under model-plant mismatch with adaptive MPC maintaining process constraints, and (v) a validated digital twin enabling scenario analysis for plant-wide deployment. The study contributes to knowledge by demonstrating a scalable, intelligent control paradigm that integrates real-time sensing, ML-based surrogates, and adaptive parameter estimation for CCM systems, bridging gaps between membrane science, process control, and digitalization in carbon capture. Practical implications include a blueprint for industrial implementation, cybersecurity considerations for data streams, and decision-support tools for operators. The concluding recommendations emphasize standardization of sensing architectures, adoption of online learning protocols to accommodate aging, and further exploration of multi-objective optimization balancing purity, recovery, and energy, with an emphasis on lifecycle assessment and techno-economic analysis.