Get it right the first time & learn smarter today

# Multistage Heat Exchanger Design Optimization

Disclaimer: This work has been submitted by a student. This is not an example of the work written by our professional academic writers. You can view samples of our professional work here.

Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of UK Essays.

Published: Thu, 31 Aug 2017

Multistage Heat Exchanger Design Optimization using Sequential Quadratic Programming, Particle Swarm Optimization and Simulated Annealing methods

Islam Gusseinov[1], Cenker Aktemur[2], Dinara Kumasheva[3]

Department of Mechanical Engineering, Eastern Mediterranean University, via Mersin 10, Famagusta, N.Cyprus

ABSTRACT

A multistage heat exchanger system is composed when it is requested to heat a single cold fluid stream with the aid of various existing hot streams. The present study concentrate on finding optimum area of three stages heat exchanger system which is connected in series. Optimum solution will be the smallest area permissible within given constraints, namely upper bounds and lower bounds. In general, this system is used to reach the required heat transfer between two fluids and get impressive heating or cooling of the fluid in one process. However, three heat exchangers connected in series will construct a huge system, which will require plenty of space to install. Another problem is the heat losses through the shell. Area of the heat exchangers increases, the heat losses will increase dramatically. That is why, we want to minimize the total of the heat transfer areas of the three exchangers by reaching the optimum solution for the efficiency.

Nomenclature

PSO=Particle Swarm Optimization

SA=Simulated Annealing

HX = Heat Exchanger

INP = Inequality Constrained Nonlinear Program

1. Introduction

A. Problem Description

Many problems can be solved and optimized with the help of different kinds of software. One of the benchmark problems in applying various methods of optimization appears to be a Golinski’s speed reducer problem. There are many studies done on this topic, and many different types of methods like Simulated Annealing (SA), Sequential Quadratic programming (SQP), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO) were carried out on Golinski’s problem. However, in this study, the same types of optimization methods will be applied on Three Stage Heat Exchanger problem, in order to find the optimal area, or in other words, the smallest area of three heat exchangers connected in series. In order to be able to perform optimization, Matlab software will be used which is a strong tool in optimization field, since it has wide range of optimization methods, prewritten by the developers. This allows us to directly use optimtools without re-writing codes again.

## B. System Representation

A fluid having a given flow rate W and specific heat Cp is heated from temperature To to T3 by passing three heat exchangers in series. In each heat exchanger (stage) the cold stream is heated with the aid of the hot fluid having the same flow rate W and specific heat Cp as the cold stream. The temperatures of the hot fluid entering the heat exchangers, t11, t21 and t31 and the overall heat transfer coefficients U1, U2, U3 of the exchangers are known constants. Drawing of the 3- Stage Heat Exchanger model is illustrated in Figure 1.

T0T1T2T3

t12t11 t22 t21 t32t31

Figure 1. Three Stage Heat Exchanger System [1]

C. Design Variables

## Design parameters are the parameters which can be modified by designer in order to attain optimum solution and to satisfy all the constraints. Three stage heat exchanger involves eight design variables which are specified properly.

A1 – Area of the first HX (x1)

A2 – Area of the second HX (x2)

A3 – Area of the third HX (x3)

T1 – Outlet temperature of first stage (x4)

T2 – Outlet temperature of second stage (x5)

t12 – The temperature of the hot fluid entering first stage (x6)

t22 – The temperatures of the hot fluid entering the second stage (x7)

t32 – The temperatures of the hot fluid entering the third stage (x8)

Table 1: Design Variables [2]

 Variable Symbol Units Lower Bound (LB) Upper Bound (UB) Area of the first stage A1 Sq.ft 100 10000 Area of the second stage A2 Sq.ft 1000 10000 Area of the third stage A3 Sq.ft 1000 10000 Outlet temperature of first stage T1 OF 10 1000 Outlet temperature of second stage T2 OF 10 1000 The temperature of the hot fluid entering first stage t12 OF 10 1000 The temperatures of the hot fluid entering the second stage t22 OF 10 1000 The temperatures of the hot fluid entering the third stage t32 OF 10 1000

The variable bounds for the problem are as follows [3]:

100 â‰¤ x1 â‰¤ 10000

1000 â‰¤ x2 â‰¤ 10000

1000 â‰¤ x3â‰¤ 10000

10 â‰¤ x4â‰¤ 1000

10 â‰¤ x5â‰¤ 1000

10 â‰¤ x6â‰¤ 1000

10 â‰¤ x7â‰¤ 1000

10 â‰¤ x8â‰¤ 1000

D. Design Constraints

Design constraints can be of different types such as constraints related to physical properties, general constraints and constraints related to design variables, which are also called bounds. Constraints can be more flexible or inequality constraints and fixed which are called equality constraints. The main difference between equality and inequality constraints is the level of difficulty. Inequality constraint problems are easier to solve, since there is a chance to go from one limit to another, however equality constraints problem restricts the solution to one exact limit. In this particular case, there are six main constraints, eight bounds (constraints on design variables) and nine constraints within the bounds. That is why in Matlab just fifteen main constraints are used. The following table provides mathematical representation of six main constraints.

Table 2: Design Constraints

 Constraint Symbol Units Limit G1 oF â‰¤ 0 G2 oF â‰¤ 0 G3 oF â‰¤ 0 G4 Sq.ft â‰¤ 0 G5 Sq.ft â‰¤ 0 G6 Sq.ft â‰¤ 0

E. Problem Formulation

Previously the definition of optimization was given as the finding optimal solution to the problem by considering all design variables and satisfying all the constraints. Where the solution can be obtained by changing the design variables and initial, starting points for the optimizer. After choosing all the variables and establishing the constraints, it is time to formulate the objective function, which will give mathematical solution, by relating all needed design variables and constraints to each other. Objective function for this particular problem was formulated as it was described in work[5]. Objective function and constraints are provided below.

Objective function: Minimize

Subject to (constraints in right-hand column)

G1

G2

G3

G4

G5

G6

II. Results and Discussion

Sequential Quadratic Programing (SQP) is one of the most popular methods in solving inequality constrained nonlinear program (INP). SQP appears to give reliable and effective results in short period of time. Within each iteration of SQP method Matlab solves quadratic programming (QP). This solution plays a guiding role for the next iteration [6]. Data from Matlab received for three stages HX is provided in table 3, down below.

Table 3: Optimization Results using SQP

 Solution Units Run 1 Run 2 Run 3 x1 Sq.ft 555.785990282356 368.842812105996 489.941608347603 x2 Sq.ft 1496.77829557695 1409.78842135742 1475.12102289728 x3 Sq.ft 5002.82363187765 5304.38006121906 5091.86834178931 x4 oF 180.016508545819 161.355189743578 174.019885417799 x5 oF 299.888637677655 287.834540297210 296.333509650773 x6 oF 219.979103072183 238.629016924913 225.956135531385 x7 oF 280.126421259455 273.511437056177 277.678961062453 x8 oF 399.888245902272 387.832237792480 396.331510973503 G1 oF -1.09709549950265e-05 -3.95154181341839e-05 -5.99530482414679e-05 G2 oF -3.62402177345178e-06 -2.30289692231267e-05 -1.85404301938918e-05 G3 oF -3.85816583214904e-06 -2.30229861267750e-05 -1.99886687483053e-05 G4 Sq.ft -2.42690607324766 -3.12624794289877 -27.7786141939578 G5 Sq.ft -2.18310707871569 -17.3791291491361 -17.5780774553423 G6 Sq.ft -2.02720958227292 -12.1425335318781 -10.4304198068567 Objective Sq.ft 7055.39 7083.01 7056.93 Optimizer parameters x0 = [550 1500 5000 200 300 200 300 400] x0 = [600 2000 6000 600 600 700 500 800] x0 = [300 1200 3000 450 550 650 450 600]

* The violated Constraints are highlted in Bold

NO

YES

Figure 2. General Flow Chart and Optimization Parameters of SQP

G. Simulated annealing

Simulated annealing (SA) is a probabilistic method of optimization, used to find the global optimum point for given function. Moreover, it is a metaheuristic to locate global optimization in a large search space. It is mostly used for finding precise local minimum, while having a fixed time for all iterations. Simulated annealing could be considered as better option, rather than brute-force search or gradient descent.

Simulated annealing takes its name from manufacturing process of slowly cooling the molten metal, in order to control the properties of the final product. In the same way, SQP slowly decreases the probability of getting bad result, by increasing the solution space [7].

Using SA method provided different optimization result for the same inputs, constraints and design variables, than SQP method did. These results are tabulated in Table 4.

Table 4: Optimization Results using SA

 Solution Units Run 1 Run 2 Run 3 x1 Sq.ft 139.724828271076 100.000104807751 258,177990492450 x2 Sq.ft 1000.64599816644 1017.44295810495 1000,00713464290 x3 Sq.ft 1000.07536409596 4439.97459544385 3390,50971323642 x4 oF 245.460108886459 91.8856667053922 365,912999644345 x5 oF 388.784149405058 269.643483880431 374,204976990304 x6 oF 971.698897292658 214.089384992577 958,928952825707 x7 oF 490.653755999378 320.353186032579 483,076316148436 x8 oF 752.433465643201 413.892877672396 472,961764726240 G1 oF 2.02949239121487 -0.235063049022586 2,31205648350210 G2 oF 0.243395435454855 0.245276619365850 0,228477155449455 G3 oF 3.66459533266688 0.442499055677654 -0,0124603267437754 G4 Sq.ft -1.12564615524209 -1.81711694639683 -1.50580160468628 G5 Sq.ft -3.8020,6645185796 -1.02549103736873 -1.06820780542949 G6 Sq.ft -9.25189489858765 -6.45744633847843 -2.03521857259860 Objective Sq.ft 2150.97 5564.30 4775.72 Optimizer parameters x0 = [300 1200 3000 450 550 650 450 600] Penalty = 2 x0 = [550 1500 5000 200 300 200 300 400] Penalty = 10 Boltzmann annealing Reannealing interval=250 x0= [600 2000 6000 600 600 700 500 800] Penalty = 50 Botzmann annealing Initial temp. = 300

* The violated Constraints are highlighted in Bold

NO

NO

YES

YES

YES

Figure 3. General Flow Chart and Optimization Parameters of SA

H. Particle swarm optimization

Particle Swarm Optimization (PSO) is a method of optimization that finds the optimum by repeatedly searching for the better solution, in comparison with the given value of the constraints and design variables. It solves a problem by sending the candidate solutions through the search space and letting them to find out the local optimum with respect to their mathematical algorithm. After finding local minimum these solutions are programed to compare their best point with the neighbouring one, that is how it is expected that the global optimum will be reached [8]. Based on the PSO method, the following results were retrieved.

Table 5: Optimization Results using PSO

Solution

Units

Run 1

Run 2

Run 3

x1

Sq.ft

100.000049003836

100

100.000035217871

x2

Sq.ft

1000.00011217875

1000

1000.00007680824

x3

Sq.ft

1000

1000.00002296095

1000.00000062339

x4

oF

121.418673285213

121.426230206343

121.427717593523

x5

oF

460.000989443898

460.005701802001

460.000785308856

x6

oF

278.507129110867

278.574674462544

278.563901342208

x7

oF

544.648869935443

544.652837594674

544.644099603815

x8

oF

560.000134288575

559.994330413195

559.999171302912

G1

oF

-0.000263305653911416

9.94556734945640e-05

2.220804404479283

G2

oF

1.20814848467031

1.20804826383677

1.20804602546329

G3

oF

3.71887125054649

-0.000218022179330424

-1.93810245309178

G4

Sq.ft

-6.80811877579254

7.03418059945398

-0.102022209626739

G5

Sq.ft

-5.25140812853351

-13.9946636939421

-0.787525329855271

G6

Sq.ft

-22.4404559549876