(46c) Topological and Parametric Optimization of Membrane Systems for H2 Purification from Syngas Using Carbon Molecular Sieve Membranes

While hydrogen can be efficiently produced from gasifying a variety of feedstocks, separating it from the large number of contaminants present in the syngas from such gasification processes to a high purity level requires multiple, costly unit operations which increase the capital and operating costs as well as the footprint of the overall separation process. Furthermore, traditional separation processes such as solvent absorption, stripping, cryogenic distillation, multi-bed pressure swing adsorption, etc. are not desirable for distributed H₂ generation processes.

Membranes provide an opportunity for less energy- and cost-intensive separation compared to traditional gas separation mechanisms.¹ Carbon molecular sieve membranes (CMSMs) are among the most promising varieties for H₂ purification, but their industrialization is not yet realized as fabrication, design, and implementation procedures are being improved and perfected.^2,3 Computational/machine-learning models can help assess specific design choices,⁴ but full-scale operational models are needed for maximizing the economics of these membranes for high-purity H₂ production on industrial scales. While models for CMSMs exist with ranging considerations on non-ideality⁵, there is a lack of models for such membranes which consider rigorous transport phenomena in multi-component systems applied to H₂ purification. There is also a lack of literature on optimal superstructure synthesis and optimization of design and operating conditions of CMS membranes using detailed multi-component models for H₂ purification.^6–8

In this work, a one-dimensional, steady-state model of a hollow-fiber CMSM is developed using literature and data for select-gas permeation in CMSMs derived from Matrimid®-5218 polyimide precursor. For rigorous properties modeling of the multi-component system, the multi-stage membrane model is developed in Aspen Custom Modeler (ACM). The phenomenological-based dual-mode sorption (DMS) model is used for transport⁹, providing component-dependent parameterization to each species in select effluents. The permeation model is spatially-variant and incorporates composition and pressure effects as expected non-idealities. The model is validated against literature and our in-house data and used for optimization. A cost model is developed that can estimate the capital and operating costs for multi-stage membrane systems with inter-stage compression and recycle streams. The optimization objective considered is minimization of levelized cost of hydrogen separation (LCOHS) and includes design and operational decision variables (membrane areas, pressure ratios, recycle ratios) and configurational decision variables (number of stages number, flow configuration, number of compressor stages for inter-stage compressors, recycle location and splitting) thus leading to a mixed-integer, nonlinear programming (MINLP) problem. Currently, ACM can solve only NLPs; to handle this issue, a two-stage approach is developed. First, surrogate models are developed for functional representation of the cost vs the number of compressor stages for a given inter-stage compression ratio. These models are then used to optimize the number of stages for the inter-stage compressor. Since exhaustive enumeration of the number of membrane stages is feasible, this approach converts the overall optimization problem to a series of NLP problems to be solved at the second stage. Results show that there is a strong trade-off between recovery vs purity, and that topological and operational variation can alleviate some of these trade-offs. It is also observed that as the desired purity of H₂ increases, the optimal number of stages and their design greatly vary, with steepest penalties seen in separations leveraged to produce ultra-pure H₂ that exceeds 99.9%.