Review on simulation models of desorption of CO2 in MEA solutions

https://doi.org/10.1016/j.jece.2017.08.024

“Commercially available process simulation software as ASPEN HYSYS, ASPEN PLUS, gPROMS, ASPEN, Custom Modeler, liean with Tsweet, Protreat have been extensively used for steady and dynamic simulations AspenAspen Plus[7]. Over the years, also in-house simulation tools have been used [6], [14]. For example, Pinto et al. [8] used Procede Process Simulator (PPS) and discussed the performance of their simulation model compared to experimental liquid loading, temperatures and CO₂ captured. Furthermore, they compared the experimental mass transfer coefficients to the simulated one separating this study from the other studies listed in Table 1. However, Pinto et al. [8] did not include the desorber.

Because chemical absorption has reached a TRL9 (Technology Readiness Level), most recent simulation studies are related to process dynamics to study the flexibility on the CO₂ absorption/desorption operation. A review on dynamic models previous to 2013 can be found in Bui et al. [7].

Overall, in the simulation models, kinetic constants, effective absorption area or heat loss (desorber) are often adjusted to make the simulations fit the experimental data [4], [5]. Often the data is fitted to very few experimental runs and the rest of the runs from the same campaign are thereafter simulated using the adjusted model. Finally simulations of industrial size plant are made [10]. However, when the fitted model is used to simulate other pilot data, considerable deviation might be seen [11]. The literature clearly shows that the choices related to the equilibrium model, mass transfer coefficients, and kinetic parameters influence the model predictions heavily making important to validate the simulation model using data from several pilot plants, with various operating conditions before using the model for process design or optimization [6], [9], [12].

As seen in Table 1, ASPEN PLUS is a commonly used software. Zhang & Chen [16] validated their ASPEN PLUS model with one pilot plant under different operation conditions (reproduced 19 runs). The thermodynamic, kinetic and transport property parameters were adjusted. The validation parameters used for the desorber were the reboiler duty, temperature and concentration profiles. Their model predicted the reboiler duty in a good agreement with the experimental data, with some deviations at high reboiler duty. Some slight under-prediction was observed in the CO₂ concentration, while the temperature reproduced well the experimental results. Lim et al. [13] also used ASPEN PLUS to represent one pilot plant. They obtained a representation of loading and temperature, in good agreement with experimental data, while the regeneration energy was over-predicted. Li et al. [24] used ASPEN PLUS to predict one pilot plant run. They adjusted the water wash section and obtained lower errors in the prediction of temperature and loading than the basis model. Von Harbou et al. [6] used ASPEN PLUS v7.3 to simulate two pilot plants with different scale and packing. The mass transfer correlations and CO₂ solubility were adjusted. The reboiler duty and loading results were mostly under 5% of error. Also Oi & Pedersen [20]used ASPEN HYSYS and compared it to ASPEN PLUS. Both modelled used the ENRTL framework. Their work aimed to validate a reduction of heat consumption by changes on the absorption process configuration. The temperatures were in good agreement with experimental results. Gaspar et al. [28], [29], [30] used ASPEN PLUS in combination with other software (dCAPCO2) to measure the transient response to step changes on compositions and flows using separately MEA and PZ.

In addition to ASPEN PLUS, other simulation packages from ASPEN have been used in the literature to simulate the dynamic response in carbon capture pilot plants. Huser et al. [32] and Posch and Haider [33] used the ASPEN Custom Modeller and the ENRTL framework for the validation of the absorber with one pilot plant. However, the validation with data from Notz et al. [25] in Huser et al. [32] was only done with data from the absorber. Akram et al. [26] used ASPEN HYSYS. The reboiler duty was under-predicted and the desorber temperature was in a good agreement with the experimental data. Chinen et al. [31] also used ASPEN software but to study dynamically the response of the system at steps in the key variables. They obtained some deviation of the temperature at medium height and small variations on the loading.

In the literature other software has also been used to simulate the dynamic performance of absorption and desorption [21], [22]. For example, Enaasen et al. [21] used dynamic K-Spice modelling, with multiflash provided by InfoChem and thermodynamics from CO2SIM. An over-prediction on loading was observed, potentially due to K-Spice configuration or due to inaccuracies in the analytical procedures during the pilot campaign. Additionally, although dynamic results were in good agreement with experiments, steady state results had a strong deviation.

In-house codes are also extensively used, as in Nagy&Mizsey [23]. In their work two experimental runs were represented to simulate coal and gas natural combustion (with high and low CO₂ content in the fluegas). In their work, the reboiler duty was represented, obtaining under-predicted values at high reboiler duty conditions. The main reason highlighted in their work is that the heat of absorption taken from ASPEN has some deviations. Additionally, the temperature and loading along the desorber were both over-predicted. Saimpert et al. [5] represented the desorber with an in-house MATLAB model, validated with two pilot plants, what is different to most of the studies. However, the model did not represent the two pilot plant campaigns with similar accuracy. Considering one pilot plant, the model showed under-prediction of the flow rate (with absolute deviation of 32%) and over-prediction of temperature (1–5 °C), mainly due to considering adiabatic conditions in the desorber. For the second pilot plant, the model agreed on temperature and CO₂ desorbed, with only slight deviations. This means that the developed simulation model was either not flexible enough to simulate these two different pilot plants or the data from the pilots were contradicting or too inaccurate. Also GarÐarsdóttir et al. [22] simulated the flexible operation at part-load and peak-loaded scenarios using Dymola, Modelica and Optimica software. Although this study did not include experimental validation, the results showed the process control strategy as tool to decrease the time response of the absorption system in both scenarios.

Many authors have based their work on the use of gProms to represent the absorption/desorption process over the years. Brand [34] used gProms without adjusting any model parameter. The model represented fairly well the data with some over-prediction of temperature and loading at the top and on the bottom of the desorber and some under-prediction at low temperature and on the bottom. In the works of Mac Dowell et al. and Mac Dowell & Shah [17], [18], [19], a more complex combination of software was used for the parameters prediction using one pilot plant. They combined gProms with superTRAPP method to predict viscosity and thermal conductivity, while statistical associating fluid theory was used to predict the potentials of variable range and SAFT-VK framework was used for the remaining parameters. Additionally, only two runs were used in Mac Dowell et al. [17] for the validation of the model and this provided the basis for the prediction of the theoretic optimum operation and CO₂ capture system integration in the works of Mac Dowell & Shah [18], [19]. As done by many authors, they also regarded the desorber as an adiabatic column [17], [18], what will be commented further below. Ahn et al. [14] used gProms to find the optimum configuration for the highest reboiler duty reduction.

The desorption step has been found indeed more complex than the absorption process. More than 30 years ago, the simulation of the desorption process was discussed by Weiland et al. [35], where the modelling of the desorption process was recognised of being more complicated due to the reversibility of the chemical reactions. The thermodynamic calculations consequently had strong influence in the process simulation. Weiland et al. [35], with a model based on physico-chemical data, expressed the overall mass transfer coefficients along the stripper. The concentrations of MEA varied between 0.5 and 5 Kmol/m³, with loadings between 0.313 and 0.325, exhibiting errors of 25%, expressed as root mean square deviations. Later, Escobillana [36] modified the interphase area and the bubble diameter to fit the simulation models to experimental data and proposed to fit the enhancement factor, obtaining a good representation of the CO₂ concentration and temperature along the sieve tray stripper.

Unlike the other works presented above, Luo et al. [11] simulated four pilot plants using ASPEN PLUS v2006.5 and compared to other three commercial software and two in-house codes. In their work, there were not significant differences between the results from the different software. The reboiler duty was used as output while the rich and lean loadings were inputs in the simulation model. The predicted reboiler duty showed some over-prediction compared to experimental results. These results mean that a higher energy investment was predicted to reach the concentrations of the lean flux. However, there was also an over-prediction of the temperature along the stripper, what indicated that the kinetics of the reversible reaction for the solvent regeneration could had some deviations compared to the reality.

“