Available online at www.sciencedirect.com

SciVerse ScienceDirect

Energy Procedia 37 (2013) 1500 - 1508

GHGT-11

Optimal Operation of Solvent-based Post-combustion Carbon Capture Processes with Reduced Models

Zhengxiong Li, Manish Sharma, Rajab Khalilpour, Ali Abbas*

_School of Chemical and Biomolecular Engineering, The University of Sydney, Sydney, Australia_

Abstract

This paper addresses the development of a methodology for optimal operation of solvent-based post-combustion carbon capture (PCC) with respect to techno-economical objectives. One of the main limitations in techno-economical analyses of PCC process is the unavailability of simple models for ease of use in PCC process optimization. Such mathematical models, even in reduced form, could facilitate performing efficient techno-economical studies without dealing with the complex physico-chemical models. In this study, a flowsheet PCC process model is developed and a sensitivity analysis is carried out around 1700 case studies. The resulting data were then modeled with Response Surface Methodology (RSM) to develop an explicit nonlinear reduced model. Optimal operating conditions were then found through the reduced model. Such optimal values are proposed as control set points in the PCC plant.

© 2013 The Authors. Published by Elsevier Ltd. Selection and/or peer-review under responsibility of GHGT

Keywords: post-combustion carbon capture; monoethanolamine; model; optimization; economics; response surface methodology, process control.

1. Introduction

Solvent based post-combustion carbon capture (PCC) technology is considered to be the most reliable technology for CO2 removal from the low CO2 concentrations flue gas [1, 2]. However, the associated energy penalty is relatively high compared to pre-combustion and oxyfuel combustion processes due to hefty energy consumption in solvent regeneration [3].As such, the drawback of retrofitting power plants with solvent-based PCC processes is the significant energy penalty introduced which can be in the range of 10-40% of total electricity produced [4]. Therefore, if this approach is to succeed, much more research is needed to find innovative methods to make significant reductions in carbon capture costs.

* Corresponding author. Tel.: +61 2 9351 3002; fax: +61 2 9351 2854. E-mail address: ali.abbas@sydney.edu.au (A. Abbas).

1876-6102 © 2013 The Authors. Published by Elsevier Ltd. Selection and/or peer-review under responsibility of GHGT doi:10.1016/j.egypro.2013.06.025

Fig. 1 shows a schematic of a solvent-based PCC process. The flue gas passes, with a temperature range of 40-60 oC through the absorber column (packed or tray) where the lean solvent enters from the top of the absorber in a countercurrent process. In the absorber, the solvent removes CO2 from the flue gas through physico-chemical interaction; the rich solvent then exits from the absorber bottom while the cleaned flue gas leaves absorber overheads towards stack. In the stripper column, the rich solvent is stripped of CO2 via thermal treatment at 100-120 oC. The lean solvent is recycled to absorber while CO2 is sent from overhead to compression unit.

Fig. 1: Schematic of solvent-based PCC Process.

The key parameters influencing the efficiency of solvent-based PCC process have been discussed as solvent type and solvent concentration, configuration of absorption and stripping columns, operating conditions of columns, percentage of CO2 avoided, captured CO2 purity and amount of regeneration. Recent studies have shown that that optimization of PCC process configurations and operational conditions has significant impact on lowering the cost and energy consumption of the entire PCC system. Therefore, there is still good potential for process system optimization to benefit the feasibility of CCS projects.

One of the main limitations in techno-economical analyses of PCC process is the unavailability of user-friendly models. Because of different reaction mechanisms of any individual solvent, there is no generic model for predicting the carbon capture rate with various solvents. Instead, over the time, different models have been introduced for various solvents. In terms of accuracy, the models are still in progress and there are notable discrepancies between their predictions [5]. Optimization study, however, requires a robust mathematical model including all the critical parameters. Such mathematical equations, even in reduced form, could facilitate performing efficient techno-economical studies without dealing with the complex physico-chemical models. It is the objective of this study to develop such explicit equations in order to identify the key techno-economical variables, their interactivity and their weight of impact on process performance and plant economics. Such simple models can also reduce the computation time from hours to seconds.

Rao and Rubin [6] performed a detailed techno-economical study for retrofitting of a coal-fired power plant with MEA-based PCC process. They used ProTreat for process simulation and an in-house model (IECM-cs) for the overall techno-economical modeling. Based on the results of numerous case-studies, and using Response Surface Methodology (RSM), they proposed an equation for prediction of reboiler

duty which is function of four variables, i.e. solvent concentration, CO2 mole fraction, CO2 capture percentage and lean loading. Zhou et al. [7] with access to MEA-based pilot plant data of 116 days and using RSM developed four different models; the first model predicts CO2 production rate using three variables of solvent flowrate, reboiler duty and solvent concentration; the second model predicts reboiler duty using steam flowrate at inlet and outlet of the reboiler; the third model predicts lean loading using reboiler duty and solvent concentration; and finally the fourth model predicts the carbon capture rate using lean loading. The R2 of the models are in the range of 0.77 to 0.88. Another research, with interest in the interdependency of operation variables studied the multicolinearity among the variables [8]. From the candidate variables, those with ^-values of higher than 0.05 were considered insignificant parameter and were removed from the model. Interestingly, for reboiler duty model and CO2 production model, reboiler pressure was shown to be insignificant variable. For capture rate model and lean loading model, amine concentration was insignificant. Nuchitprasittichai and Cremaschi [9] developed an economic model, employing RSM, to study a few amines (MEA, DGA, DEA, MDEA and TEA) in order to find the best amine (or mixture) with minimum operating cost. Their candidate variables were amine concentration, the absorber and stripper column heights, and the operating conditions. Although the study did not show any equation from RSM study, it concluded that absorber height, solvent flowrate and reboiler duty were the three critical components of cost whilst stripper height and the stripper inlet temperature were insignificant parameters. The best solvent, according to this study is 48 wt% DGA.

In a different study, Sipocz et al. [10] using the simulation data of CO2SIM, employed Artificial Neural Networks (ANN) to generate a model to find the optimum operation conditions of PCC process. They trained the model with six inputs (inlet flue gas temperature, inlet flue gas flowrate, CO2 fraction of inlet flue gas, solvent flowrate, lean loading, and capture rate) for finding the reboiler duty and rich loading. This study also did not show the ANN parameters, but reported prediction error of less than 2%.

In this study, we aim to develop both technical and techno-economical equations with inclusion of a wide range of variables. The ultimate goal is to use the equations in real plant scheduling and daily operational control.

2. Modeling and Optimization Framework

The optimization methodology of this paper is given in Fig. 2. The PCC process is simulated and then N case studies are carried out using multivariable sensitivity analysis. The results of the case studies are saved from which M key parameters that may have significant impact on the objective function are nominated as decision variables. For this study we have selected nine technical variables which are listed in Table 1. These are reboiler duty (xj), liquid-to-gas ratio, L/G (x2), solvent concentration (x3), rich loading (x4), lean loading (x5), stripper inlet temperature (x6), condenser duty (x7), recycle cooling duty (x8), and capture rate (x9). Having obtained the simulation data (Nx V matrix) we use RSM to develop a nonlinear equation with response function of reboiler duty as function of other variables which will be discussed later. For RSM, we use a second order model given by,

y=p0+ Zi=i ft* + Zi=i ft;*;2 + Z Z;<j Putt*} + e (1)

where y is the response (objective function) to be predicted, is the scaled value of variable i, p is the coefficient to be regressed and e is the error. The weighting factor for each of these terms is analysed to illustrate the impacts they impose on the left-hand side of the equation.

Perform N case studies in the simulator (Aspen Hysys V7.3 package)

Economic parameters (Table 3)

Operating variables (x-)

Compute OPEX for the N case studies (OX£)

Operating variables (x™) OPEX value (OZA)

Response Surface Modeling (MODDE package)

A technical nonlinear prediction of the PCC process RQ = f{Xt)

A techno-economical nonlinear prediction of the PCC process OX = f(Xi,E)

Fig. 2: Framework of the optimization

Further to the technical equation, we also develop a techno-economical equation. For this an economical model is developed for OPEX which is discussed in Section 2.2. Electricity price is critical and essential in estimating the value of OPEX when it changes and it is expected to be stimulated by upcoming carbon policies. As such, further to the nine technical variables, we introduce one more variable for OPEX, being electricity price (x10=E).

Table 1: Selected variables and their ranges

Factors Symbol Range Unit

Reboiler duty x1 486~1082 GJ/h

L/G x2 21.08~43.72 tonne-amine/tonne-CO2

Solvent concentration x3 19.13~34.62 wt%

Rich loading x4 0.408~0.537 mole-CO2/mole-MEA

Lean loading x5 0.212~0.396 mole-CO2/mole-MEA

Stripper inlet temperature x6 95~110 °C

Condenser duty x7 96.86~472.8 GJ/h

Recycle cooling duty x8 100.9~719.6 GJ/h

Capture rate x9 63.2~91.0 %

Electricity price x10 10~30 $/GJ

2.1. Process Simulation

For this study, Hysys 7.3 (Aspentech, USA) was selected as the simulator. The Peng-Robinson fluid package was selected for flue gas stream modeling whilst ASPEN Properties (Amine) package was used for solvent processes. The base case used is a 300 MWe coal-fired power plant burning pulverized black coal and emitting 256 tonne/h of CO2 with concentration of 13 mol% [11]. This flue gas was used to simulate the PCC process using MEA. Specifications of the simulated PCC process are given in Table 2.

Table 2: Specifications and design parameter of MEA-based PCC process.

Absorber

Item Value Reference

Tray No. 10

Tray Efficiency (%) 25 [12]

Temperature (°C) 50 [13]

Pressure (kPa) 101.3

Stripper

Item Value Reference

Tray No. 6

Tray Efficiency (%) 50 [12]

Reboiler pressure (kPa) 200 [14]

Feed position (tray) 2

Feed condition Saturate liquid

2.2. Techno-economic Model

We aim in this study to develop an economic multivariable model to predict the carbon capture operational expenditure (OPEX) with respect to various technical and economical parameters. The OPEX framework is developed with adoption and modification of the cost estimation models given elsewhere [15, 16]. The economic parameters are given in Table 3.

Table 3: Economical parameters (annual basis unless otherwise the unit is mentioned)

Range Value in this work

Manufacturing cost

Fixed charge

Property Insurance 1% of FCI 1

Variable production cost

Utilities (according to simulations)

MEA makeup 1.5kg/tonne-CO2-captured

Maintenance and repairs 1-10% of FCI 5

Operating labour 3% FCI 3

Supervision 15% of operating labour 15

Laboratory charges 10-20% of operating labour 15

Operating supplies 15% of maintenance and repairs 15

Plant overhead cost 50-70% of (maintenance+operating+labor+supervision) 60

General expenses

Administrative cost 15-25% of (maintenance+operating labor+ supervision) 20

R&D cost 0.5% FCI 0.5

Utilities

Electricity (H0-440V) 10.4 $/GJ (2012 $) [17]

Cooling water (30-45C) 0.4 $/GJ (2012 $) [17]

With application of the parameters of Table 3 and rearrangements, the OPEX formulation is simplified to,

OX (%/yr) = 0.12FCI + 1.07% Ut (%/yr) (2)

Where FCI is fixed capital cost and Ui refers to utility costs including MEA make up (U1), reboiler energy (U2), condenser (stripper) energy (U3), pumping power (U4), recycle cooling energy (U5), product CO2 compression power (U6), and inter-stage cooling energy of product CO2 (U7).

FCI is taken as 0.8 of Total Capital Cost (TCC) given by Harkin et al. [18]. The Chemical Engineering Plant Cost Index (CEPCI) is used to convert the economic values to 2012 dollar. It is noteworthy that we use electricity price to calculate reboiler energy (U2). This is due to the fact that the steam which is used to heat the reboiler is extracted from power plant turbines aimed for electricity generation. In this study we have assumed that one Joule of reboiler energy equals to 0.19 Joule of electricity [11].

2.3. Response Surface Modeling

The collected data from simulation case studies are treated with MODDE 7 through Response Surface Methodology (RSM). RSM provides statistical and mathematical techniques for process optimizations. The idea of RSM is to use a set of decision variables to obtain an equation for the objective function (response). It was initially developed to model and regress experimental data and later moved to the field of numerical simulations [19]. In a real experiment the inaccuracy might be caused by measurement errors while in computer-aided simulations it is a result of less complete convergence of iterations. For RSM, it is assumed that all the errors are random [20].

The MODDE software, scales all the given data into the interval of [-1, 1] in order to provide a universal tolerance of error to all the factors in regression process. Therefore the original decision variables (xi) and the scaled variables (x[ ) have the following relationship [21]:

i=1, 2, ..., V+1

where Mi is the midpoint of natural interval ((maxi + mini)/2), Ri is equivalent to half of the interval ((maxi — mini)/2), and Vis the total number of decision variables.

3. Results and discussion

3.1 Technical Model

Reboiler duty is widely used as a benchmark for analysis of solvent-based PCC process. The reboiler duty is directly attributed to the load reduction of power plants as reboiler consumes the same steam which power plant turbines use for power generation. In technical modelling, we aim to develop an equation for reboiler duty as function of other operating variables (x2-x9). Such an equation will allow the plant operator to identify the optimal operating conditions when plant supervisory schedules certain quantity of steam (GJ/h) for the PCC process. Under the given reboiler duty, the optimal values of the variables are obtained from the equation and are used as control set points in the operation of the PCC plant. Table 4 lists the coefficients of the developed equation and the related p-value for each variable. It is reminded that statistically, when a variable has p-value above 0.05, the variable is considered to have insignificant relation with the response and it is removed from the model. It is evident from the Table 4

that the maximum p-value is 0.03 (for L/G) being less than the threshold of 0.05. The R2 of the model is 0.991 which makes the equation adequate.

3.2 Techno-economical Model

Economics is almost always the main decision making objective. From PCC plant supervisory point of view, the objective is to operate the plant within a condition that results in the minimum costs. We aim to develop a techno-economic equation with objective of OPEX ($/tonne-CO2-captured) including all the key techno-economical variables. Therefore, the model will have ten variables (x1-x10). Table 4 lists the coefficients of the developed equation and the related p-value for each variable. Again, the maximum p-value is 0.01(for lean loading) being less than the threshold of 0.05 which implies that all of the variables are correctly selected. The adequacy of the model is also proved with its high R2 of 0.990.

Table 4: The details of a) technical and b) techno-economical equations developed by RSM

a) Technical model (reboiler duty)

b) Techno-economical model (OPEX)

Const Value P Const Value P

- fto 766.746 0 - A 57.5925 0

X2 (X2 - 32.35)/11.27 £2 4.72276 0.035797 Xi (X1 - 748)/298 A 0.714155 3.3E-06

(X3 - 26.8753)/7.745 £3 -16.9943 1.53E-05 X2 (X2 - 32.35)/11.27 A -0.41228 9.07E-05

X4 (X4 - 0.4725)/0.0641 A -24.8476 2.15E-05 X3 (X3 - 26.8753)/7.745 A -0.55952 3.25E-07

X5 (X5 - 0.3077)/0.0879 A -47.2927 2.42E-36 X4 (X4 - 0.4725)/0.0641 A -0.6419 1.69E-06

X6 (X6 - 102.5)/7.5 A -15.9627 0.0001034 X5 (X5 - 0.3077)/0.0879 A -0.62551 6.68E-05

X7 (X7 - 284.83)/187.97 A 8.52434 0.0171968 X6 (X6 - 102.5)/7.5 A -0.43027 0.013382

Xg (Xg - 410.45)/309.15 A 71.8721 0 X7 (X7 - 284.83)/187.97 A -0.7145 1.56E-08

(XS - 77.11)/13.89 A 39.6439 3.83E-30 X8 (X8 - 410.45)/309.15 A -0.11356 0.001601

X4 A4 -5.11077 2.29E-07 X9 (X9 - 77.11)/13.89 A -1.44453 6.69E-34

X2 X4 A4 9.44389 2.45E-31 X1 (X10 - 20)/10 Ac 10.2512 0

X2 X8 fe 7.6339 4.73E-19 • X5 A5 -0.20653 4.32E-06

X3 X4 A4 -6.05332 4.93E-05 ft99 1.87716 0

X3 X6 &6 3.62802 3.10E-15 X1 X3 ft13 0.234694 8.48E-05

X4 Xg 048 6.25004 1.87E-11 X1X10 A,10 0.681022 0

X4 X9 049 -7.09074 8.14E-17 X3 X9 £39 0.242658 1.24E-06

X5 X9 fe -10.1794 8.80E-30 x1 X7 £57 0.201234 8.8E-05

Xg 078 -7.29278 1.16E-21

ANOVA F=8465.5 P=0 F= 13117.5 P=0

3.2 Optimization

The advantage of having an explicit equation, unlike a sophisticated simulation model, is its ease of application for optimal process control under the fluctuating techno-economical inputs. For instance, in current liberated electricity market, the price of electricity is changing dynamically (usually in half-hourly intervals) and affects the operation policy of the plant, correspondingly. Therefore, the plant operator needs to frequently find the optimal operating parameters which results in minimum OPEX. Here, we try one example (Ex. 1) using the techno-economical equation developed using RSM. Let's assume that at certain time that the optimization is carried out the electricity price (x10) is 25 $/GJ. With this, if we

perform the minimization of the equation given in Table 5, the minimum OPEX will be found as 58.6 $/tonne-CO2-captured for which the operating variables are given in Table 5.

As another example (Ex. 2) we consider the case of a PCC plant built to work in base capacity of 90% CO2 capture. However, when the pool electricity price is high, the power plant supervisory reduces the steam flow to PCC plant in order to generate more electricity and sell to the grid. Consider a certain time of a day when the supervisory has scheduled only steam flow of about 450-470 GJ/h to the PCC plant reboiler. Then the problem is to find the optimal operating conditions under this given constraint. Here, the optimization task is carried out on the technical equation and the optimal conditions are found indicating to operate the plant with only 63.66% capture with other operating conditions given in Table 5.

Table 5: optimal operating parameters as control set point of PCC process for the two given examples

Symbol Description Optimal values of Ex. 1 Optimal values of Ex. 2

xi Reboiler duty 673.6 (2.95 GJ/tonne-CO2_captured) -

x2 L/G 41.48 33.89

x3 Solvent concentration 31.21 29.72

x„ Rich loading 0.440 0.474

x5 Lean loading 0.343 0.396

x6 Stripper inlet temperature 109.0 102.7

x, Condenser duty 374.29 267.36

xs Recycle cooling duty 214.4 100.9

xç Capture rate 90.83 63.66

x10 Electricity price 25 -

Objective value OPEX ($/tonne): 58.6 Reboiler duty (GJ/h): 455.7

4. Conclusion

It was the objective of this study to develop explicit equations in order to identify key techno-economical variables, their interactivity and their weight of impact on PCC process performance and plant economics. Using around 1700 simulation case-studies, we developed two explicit nonlinear equations, one for technical objectives (reboiler duty) and another for techno-economical objectives (OPEX). Both equations showed high R of about 0.99 and could be used to find optimal operating conditions to be used as control set points in the PCC plant.

Nomenclature

OX operating expenditure ($/tonne-CO2-captured) RQ reboiler duty (GJ/h) N number of simulation case studies xi operational or economical variable

scaled form of the operational or economical variable y response (reboiler duty for technical scenario and OPEX for techno-economical study)

fi coefficient

E electricity price

References

[1] P. Singh, D.W.F. Brilman, M.J. Groeneveld, Solubility of CO2 in aqueous solution of newly developed absorbents, Energy

Procedia, 1 (2009) 1257-1264.

[2] M. Mofarahi, Y. Khojasteh, H. Khaledi, A. Farahnak, Design of CO2 absorption plant for recovery of CO2 from flue gases of gas turbine, Energy, 33 (2008) 1311-1319.

[3] B. Metz, Intergovernmental Panel on Climate Change. Working Group III., IPCC special report on carbon dioxide capture and storage, Cambridge University Press for the Intergovernmental Panel on Climate Change, 2005.

[4] B. Metz, O. Davidson, H.d. Coninck, M. Loos, L. Meyer, IPCC special report on carbon dioxide capture and storage, Cambridge University Press, for the Intergovernmental Panel on Climate Change, Cambridge, 2005.

[5] X. Luo, J.N. Knudsen, D. de Montigny, T. Sanpasertparnich, R. Idem, D. Gelowitz, R. Notz, S. Hoch, H. Hasse, E. Lemaire, P. Alix, F.A. Tobiesen, O. Juliussen, M. Kopckeh, H.F. Svendsen, Comparison and validation of simulation codes against sixteen sets of data from four different pilot plants, Enrgy Proced, 1 (2009) 1249-1256.

[6] A.B. Rao, E.S. Rubin, Identifying cost-effective CO2 control levels for amine-based CO2 capture systems, Ind Eng Chem Res, 45 (2006) 2421-2429.

[7] Q. Zhou, C.W. Chan, P. Tontiwachiwuthikul, Regression analysis study on the carbon dioxide capture process, Ind Eng Chem

Res, 47 (2008) 4937-4943.

[8] Q. Zhou, C.W. Chan, P. Tontiwachiwuthikul, R. Idem, D. Gelowitz, A statistical analysis of the carbon dioxide capture process, Int J Greenh Gas Con, 3 (2009) 535-544.

[9] A. Nuchitprasittichai, S. Cremaschi, Optimization of CO2 capture process with aqueous amines using response surface methodology, Comput Chem Eng, 35 (2011) 1521-1531.

[10] N. Sipocz, F.A. Tobiesen, M. Assadi, The use of Artificial Neural Network models for CO(2) capture plants, Appl Energ, 88 (2011) 2368-2376.

[11] R. Khalilpour, A. Abbas, HEN optimization for efficient retrofitting of coal-fired power plants with post-combustion carbon capture, International Journal of Greenhouse Gas Control, 5 (2011) 189-199.

[12] L.E. 0i, Aspen HYSYS Simulation of CO2 Removal by Amine Absorption from a Gas Based Power Plant, in: SIMS2007 Conference, Gothenburg, Sweden, 2007.

[13] A. Cousins, L.T. Wardhaugh, P.H.M. Feron, A survey of process flow sheet modifications for energy efficient CO2 capture from flue gases using chemical absorption, International Journal of Greenhouse Gas Control, In Press, Corrected Proof (2011).

[14] M.R.M. Abu-Zahra, L.H.J. Schneiders, J.P.M. Niederer, P.H.M. Feron, G.F. Versteeg, CO2 capture from power plants. Part I. A parametric study of the technical-performance based on monoethanolamine, Int J Greenh Gas Con, 1 (2007) 37-46.

[15] M.S. Peters, K.D. Timmerhaus, R.E. West, Plant Design and Economics for Chemical Engineers, McGraw-Hill, 2003.

[16] M.R.M. Abu-Zahra, J.P.M. Niederer, P.H.M. Feron, G.F. Versteeg, CO2 capture from power plants: Part II. A parametric study of the economical performance based on mono-ethanolamine, International Journal of Greenhouse Gas Control, 1 (2007) 135-142.

[17] R. Turton, R.C. Bailie, W.B. Whiting, Analysis, synthesis, and design of chemical processes, Prentice Hall, 2009.

[18] T. Harkin, A. Hoadley, B. Hooper, Optimisation of power stations with carbon capture plants - the trade-off between costs and net power, Journal of Cleaner Production, 34 (2012) 98-109.

[19] G.E.P. Box, N.R. Draper, Empirical model-building and response surfaces, Wiley, 1987.

[20] L.F. ALVAREZ, Design Optimization Based on Genetic Programming, in: Department of Civil and Environmental Engineering, University of Bradford, UK, 2000.

[21] N.G. Sahoo, A. Abbas, Z. Judeh, C.M. Li, K.-H. Yuen, Solubility enhancement of a poorly water-soluble anti-malarial drug: Experimental design and use of a modified multifluid nozzle pilot spray drier, Journal of Pharmaceutical Sciences, 98 (2009) 281296.