Heat Exchanger Network (HEN) costs and performances estimation for multi-period operation Raaele Bolliger, Francesca Palazzi,

Franois Marechal
Industrial Energy Systems Laboratory(LENI), Ecole Polytechnique Fdrale de Lausanne, CH-1015Lausanne

Abstract
This paper addresses the problem of modeling and estimating the cost of the heat exchanger network (HEN) in multi-period operation using pinch technique concepts. The developed method improves the vertical heat exchanges technique originally proposed by Ahmad et al., by estimating the contribution of each stream to the overall HEN area instead of equally distributing the area between the computed minimum number of exchangers. This method allows to consider the available area for each stream and to manage the fact that some streams may not be active during some particular periods of operation. The method proposed computes the near-optimal T min /2 contribution, associated to the streams in each of the periods, from one single reference value. This allows to reduce the number of decision variables to one, when computing the energy saving/investment trade-o in thermo-economic optimisation models.
Key words:

Pinch analysis, minimum approach temperature, multi-period

1. Introduction

Composite curves are conventionally employed to compute the maximum energy recovery between hot and cold streams and to model the heat exchanger network (HEN) [1] in process systems. In thermo-economic process design of energy conversion systems [2,3], the HEN model based on composite curves computes the heat recovery and the integration of the energy conversion devices in order to close the energy balance of the system. This is done without having to consider and optimise the HEN layout. This approach is particularly attractive when solving process design problems where the operating conditions will result in dierent pinch points and in the potential selection of
Corresponding author: francois.marechal@ep.ch
Preprint submitted to Elsevier February 1, 2008

dierent integrated energy conversion technologies. In such superstructures, the Heat Recovery Approach Temperature (HRAT or T min ) is the decision variable that is used to represent the trade-o between energy recovery eciency and the required investment in the heat transfer system. The HEN cost is conventionally estimated considering the vertical heat exchanges in the balanced composite curves and assuming the minimum number of connections calculated by applying the Graph theory. If the validity of the HEN cost estimation using vertical exchange model has been analysed by Ahmad et al.[4], there was a need to further study the model and extending it to account for varying heat transfer conditions. We propose in this paper a method which extends the conventional approach to the study of energy systems operating in multi-period conditions. We also propose a mathematical relation allowing to obtain near-optimal T min from one single reference value and thus to reduce the number of degrees of freedom of the problem formulation.
2. Background

Considering together the process and the utility system streams, the optimal owrate of the utility streams u will be computed by solving the transhipment MILP problem originally proposed by Papoulias and Grossmann [5,6]. Equations (1)-(6) are an extended + version [7] taking into account the electricity produced by the utilities (E ), sold (Egrid ) and purchased from the grid (Egrid ). Two variables are therefore associated with any utility technology u : the integer variables yu represent the presence of the technology u in the optimal conguration and fu represents its level of utilisation. cu,1 , cu,2 , c+ and c el el are constants used for the computation of the operating costs respectively of the utilities and the electricity. The objective function is the operating cost (CU ). This formulation can be easily extended to solve multi-period problems [7].
Nu

Minimize CU =
u=1 Nu

+ (cu,1 yu + cu,2 fu ) + c+ Egrid c Egrid el el

(1)

s.t.
u=1 Nu

fu qu,r +
i=1

Qi,r + Rr+1 Rr = 0

r = 1 . . . Nr+1

(2)

Electricity consumption:
u=1 Nu

+ fu eu + + Egrid E 0 Egrid

(3)

Electricity exportation :
u=1

+ fu eu + + Egrid

E =0 u = 1 . . . Nu , yu {0, 1} r = 1 . . . Nr+1

(4) (5) (6)

Use of technology u :
+ Egrid , Egrid 0

fu,min yu fu fu,max yu R1 = 0, RNr+1 = 0, Rr 0

The heat recovery and the optimal heat exchange between the hot and the cold streams of a system is computed by solving the heat cascade (eq. 2). In order to estimate the HEN investment cost, the overall heat exchange area is calculated considering the vertical heat exchange between the hot and cold composite curves. Considering Nj vertical sections dened by the slope change of the streams, the exchange area Aj associated to the 2

vertical sections j is computed by summing the contribution of all the streams i and their respective heat transfer lm coecient i,j , as shown in equation (7).
Nj Nj

Atot =
j=1

Aj =
j=1

Ni,j Qj 1 ( )) Tlm,j i=1 i,j

(7)

In the conventional approach, the overall HEN cost is computed knowing the minimum number of exchangers target NHX,min and assuming that the overall heat exchange area Atot will be equally distributed among the heat exchangers (eq. 8). We propose a new method accounting the contribution of each stream to the overall area, as shown later in equation (19).
CI = NHX,min ahx + bhx Atot NHX,min
chx

(8)

To optimize the HEN layout, the annual cost of the system CT ot is minimized, considering the annualized ( ) investment cost and the annual operating cost (eq. 9) and by searching for the optimal T min value.
CT ot = 1 CI + CU Tyear

(9)

The value of T min is obtained by summing the contribution of the hot and the cold stream at the pinch points. The value of T min /2i being related to the heat transfer lm coecient (i ), the optimization of the exchange area Atot should therefore be performed by considering each T min /2i as decision variable. This optimisation is however dicult to solve since only the T min /2i of the streams creating the pinch points really inuence the objective function. Townsend et al.. [8] suggested relation (10), that links T min /2i with the corresponding lm transfer coecient i where K constant can be derived from a reference state, which is dened by T min /2ref and ref . The optimization of the HEN cost can then be simplied by using K as the only decision variable.
T min /2i
3. Optimal

i = K

(10)

T min

estimation for steady-state conditions

The formulation of Townsend et al.. was established considering a linear cost of the heat exchangers. For non-linear cost functions (eq. 8), we propose to replace the K constant with a more complex relation. Let us analyse the denition of the T min calculation for one single heat exchanger as described on gure 1(a). The heat exchanged between the hot and cold stream is dened in equation (11).
Q if both streams change phase Q(T min ) = mcp,h (Th,i (Tc,i + T min )) if mcp,h mcp,c mcp,h (Tc,o + T min Th,o ) otherwise

(11)

According to the relative value of the specic heat of the two streams, the logarithmic mean temperature dierence is computed as equation (12). 3

Tlm

T min /2h + T min /2c = T min = (Th,i Tc,o ) (Th,o Tc,i ) T T ln h,i c,o
Th,o Tc,i

if mcp,h = mcp,c otherwise

(12)

Considering ccu and chu the unitary price of the hot and cold utilities, the cost (eq. 13) to be minimised can be expressed as a function of the heat exchanged. Parameters ahx , bhx and chx are the cost estimation constants, Tyear is the yearly working time and is the annualisation factor.
CT ot 1 = ahx + bhx
1 Q ( +
c

1 ) h

chx

Tlm

Tyear (ccu Qcu + chu Qhu )

(13)

The optimal value of T min is obtained by solving equation (14).

(14)

(a) Single heat exchanger denition.

(b) Results of the problem optimization.

Figure 1. Heat exchanger denition and optimization results.

Assuming that T min /2c = T min /2h = T2min and that the optimal value of T min /2ref has been found for a given reference stream with given heat transfer coecient (ref ), the optimality condition (eq. (14)) will be used to compute the value of T min /2i for any stream i knowing the value of its heat transfer coecient i . When considering a heat exchanger that operates with a temperature dierence equal to T min during all the heat exchange (i.e. the two streams involved in the heat exchange have the same specic heat and Tlm = T min , equation (14) simplies into equation (15).
T min /2i = T min /2ref (Th,i Tc,i 2T min /2i ) (Th,i Tc,i 2T min /2ref ) T min /2i and Th,i Tc,i
c chx 1 chx +1

ref c chx ) hx +1 i

(15)

In this case, when Th,i Tc,i is simplied to equation (16).

T min /2ref , equation (15)

ref c hx T min /2i T min /2ref ( ) hx +1 = i

(16)

When the investment cost function (eq. 8) is linear (i.e. chx = 1), the relation given by equation (16) is identical to the one presented by Townsend et. al (eq. 10). In the situation where the cold stream evaporates and the hot stream condenses, the energy saving does not continuously depend on the value of T min . The exchange is either acceptable or not if the annual investment cost is higher than the energy saving benet. However, if a reference T min /2ref is known, the dependence of the heat transfer coecient remains given by equation (16).
4. Extension to multi-period operation

Applying the heat cascade model in multi-period situations reveals another diculty since the value of the T min /2i should be updated to represent the fact that the available area is constant for all the Np periods even if the operating conditions of the streams are changing. To do so, a reference area Ai,pref will be attributed to each stream. Assuming the more constrained situation where the stream is exchanging with a stream of similar temperature variation and heat transfer coecient, equation (17) will be used to establish the relation between the T min /2i,p in any period p by comparison with the T min /2i,pref computed for the attributed area Ai,pref in the reference period.
Ai,p = Ai,pref = Qi,pref i,pref Qi,p i,p = 4 T min /2i,p 4 T min /2i,pref

(17)

As given in equation (16), T min /2i,pref can be computed from a global reference T min /2ref,ppref . Equation (17) can then be rearranged into equation (18).
T min /2i,p Qi,p i,pref = T min /2ref,pref Qi,p i,p
ref

ref,pref i,pref

chx chx +1

(18)

In order to estimate the HEN cost, all the streams appearing in the problem formulation must be taken into account. Eq. (7) must be adapted to account for streams that would be active only in some periods. We consider therefore a contribution of each stream to the total area and, at the same time, a contribution to the HEN overall investment cost. This can be deduced again from the analysis of the heat exchanger with identical temperature variations. In this case, the investment cost (CI ) is the sum of the contribution of the costs of the hot and the cold streams. Assuming that both streams have the same heat transfer coecient i half of the exchanger price will be attributed to each stream. We obtain then equation (19) which denes the cost attributed to a given stream i. In this equation, Qi,j,p is the heat load of stream i in the vertical section j of period p.
Ni

CI =
i

ahx + bhx 2chx 1 max 2 p{1Np } j=1

Nj,p

chx i,j,p Q i,p Tlm,j,p

(19)

5. Application

In order to demonstrate the method, we applied it to the problem 1 of Floudas et al.. [9]. This simple problem presents two hot and two cold streams and three periods of 5

operation. Figure 1(b) reports the results of two multi-objective optimizations realised using an evolutionary algorithm. The gure represent the energy savings/investment trade-o by two objective functions: the HEN investment cost and the operating cost of the utilities. auto refers to the calculation where the only decision variable is T min /2ref,pref and where the other T min /2 have been computed using equation (18). single refers to the optimization where all the T min /2i,p are considered as decision variables. This generates a variable per stream and per period. Only evolutionary algorithms are ecient to solve such problems, since only T min /2i,p involved in pinch points are really inuencing the objective functions. The two Pareto frontiers do coincide, which demonstrates the validity of the approach.
6. Conclusions

A method to estimate the HEN area and cost has been proposed to solve multi-period problems using pinch techniques. The method includes a technique allowing to estimate a near-optimal T min /2 contribution of all the streams of the problem as a function of the operating conditions in the dierent periods of operation. This technique requires a reference state where the optimal relationship between T min /2ref and ref is known. This modelling method will be used to model HEN in thermo-economic design of energy conversion systems.
References

[1] Z. Kravanja, P. Glavic, Cost targeting for hen through simultaneous optimisation approach: a unied pinch technology and mathematical programming design of large hen., Computers and Chemical Engineering 21 (1997) 833853. [2] M. Gassner, F. Marechal, Thermo-economic optimisation of the integration of electrolysis in a wood to methane process, ECOS 2006 proceedings, Greece, 12-14 July 2006 2 (2006) 553561. [3] F. Marechal, F. Palazzi, J. Godat, D. Favrat, Thermo-economic modelling and optimisation of fuel cell systems, Fuel cells - From Fundamentals to Systems 5 (1) (2005) 524. [4] S. Ahmad, B. Linnho, R. Smith, Cost optimum heat exchanger networks2. targets and design for detailed capital cost models, Computers & Chemical Engineering 14 (7) (1990) 751767. [5] S. A. Papoulias, I. E. Grossmann, Structural optimization approach in process synthesis - i., Computers And Chemical Engineering 7 (6) (1983) 695706. [6] S. A. Papoulias, I. E. Grossmann, Structural optimization approach in process synthesis - ii., Computers And Chemical Engineering 7 (6) (1983) 707721. [7] F. Marechal, B. Kalitventze, Targeting the integration of multi-period utility systems for site scale process integration, Applied Thermal Engineering 23 (2003) 17631784. [8] B. Linnho, D. Townsend, et al., A user guide on process integration for the ecient use of energy., The Institution of Chemical Engineers. [9] C. Floudas, I. Grossmann, Automatic generation of multiperiod heat-exchanger network congurations, Computers And Chemical Engineering 11 (2) (1987) 123  142.