(474d) New Isotherm Model for S-Shaped Isotherm Data to be Used in Process Modeling and Its Model Reduction with Machine Learning Techniques

Title: New Isotherm Model for S-shaped Isotherm Data
to be Used in Process Modeling and its Model Reduction
with Machine Learning Techniques

Seongbin Ga, Sangwon Lee, Jihan Kim*, Jay H.
Lee*

Email: jayhlee@kaist.ac.kr

This work suggests a new isotherm model
for S-shaped isotherm data with an offset. This model enable the process
simulation to express many adsorption isotherm phenomena with offset such as H₂O
adsorption. Although molecular simulations can predict this type of pure
component isotherm data with the consideration of the molecular interactions,
an ordinary isotherm model covering the S-shaped isotherm has been absent. In
particular, the process level simulation requires the ordinary isotherm model
representing the sorbent property. The absence of the model hinders the process
level evaluation of many adsorbents. Furthermore, the isotherm information
provided by its experiments or its molecular simulations cannot
be directly used in the process simulation because the information is,
in many cases, about the pure component rather than about the mixture gas. For
this, the spreading pressure for the suggested isotherm model is also derived by integral of the uptake-pressure ratio in order
to use ideal adsorption solution theory (IAST), which is a method to describe
the mixture isotherm behaviors from the pure isotherm models. The computation
cost reduction from this derived analytical equation is shown
in the comparison with that of a numerical integral method. During this
computation, the implicit structure of IAST equations leads to the increased
computation cost as well as divergence issues. To solve this problem, a simple
machine learning method is used, and the neural network model with an explicit
set of equations is found. The performance of this
model is also tested.