Real Time Optimization of Gas Separation Cynara Membrane Through AI & ML for Adaptive CO2 Removal and Maximum CH4 Recovery Under Variable Feed Conditions

Abstract

Abstract Gas processing facility that processes sour feeds, CO2 removal is required to meet pipeline specifications for the natural gas transmission pipeline. The separation is particularly challenging when high CO2 streams must be to processed. The present study does not consider alternative separation technologies, but rather systematically optimizes an existing membrane type process to optimize the removal of CO2 so as to avoid the production of off specification gas, which is costly to be generated. It is based on the commercially available hollow fiber membrane commercial cellulosic triacetate (CTA) modules, nominal capacity of 12T 0.72 MMCFD/membrane module as per Simulation. Deployments and optimizations to these modules will be explained in real time thus pressure of about 600 psi is applied to the membrane feed for optimum separation, furthermore it is mandatory to maintain a minimum CO2 concentration of the Nonpermeate within regulatory requirements (3 mol % set by Oil and Gas Regulatory Authority, OGRA Pakistan) for feeds containing approximately 22 mol % CO2. In line pressure, flow rate and temperature data are measured, while composition data are measured simultaneously from an online gas chromatograph sampling the feed, Nonpermeate and permeate streams. This data is fed to a supervisory machine learning layer which predict the behavior of the Nonpermeate CO2 concentration and provide appropriate set point values for required feed pressure, flow rate and temperature. If an excess of CO2 is detected, the controller provides predictive value for the CO2 rich stream thus decreasing the CO2 concentration in the stream prior to exceeding the Nonpermeate specification by controlling the flowrate. On the other hand, a decrease CO2 in the feed provide an operational slack that will allow a safe increase in throughput. This has resulted in a closed-loop, human in the loop configuration with significant improvements in process stability, hydrocarbon recovery, and significant reduction in off specification events using Artificial Intelligence. The study thereby presents a feasibility of real time sweetening of natural gas by machine learning & Artificial Intelligence in an industrially viable way.