(652f) Artificial Neural Network-Based Surrogate Models for Rapid Simulation, Optimization of Pressure Swing Adsorption

Adsorption based processes such as Pressure Swing Adsorption (PSAs), due to its cyclic nature and the presence of multiple design variables are notoriously challenging to simulate and optimize. Using detailed models that are based on first principles require hundreds of core hours of computation to accurately simulate and optimize a given process for a given material. In this talk we introduce MAPLE (Machine-assisted Adsorption Process Learner and Emulator) a framework that is capable of learning from a training-set obtained from detailed modelling of thoroughly sampled operational space and with the potential of evaluating PSA performance at any other arbitrary operating condition. As an example CO₂ capture from post-combustion flue gas is considered and two different process cycles are evaluated. For each of these cases, the trade-off between training effort and accuracy and the model’s ability to predict PSA performance are evaluated. The MAPLE framework is then employed to optimize PSA performance for a variety of objective functions. The talk will also discuss how the framework can be suitably adopted for reliably screening large databases of adsorbents.