ANALYTICAL HEAT TRANSFER

Mihir Sen Department of Aerospace and Mechanical Engineering University of Notre Dame Notre Dame, IN 46556 May 6, 2008

Preface
These are lecture notes for AME60634: Intermediate Heat Transfer, a second course on heat transfer for undergraduate seniors and beginning graduate students. At this stage the student can begin to apply knowledge of mathematics and computational methods to the problems of heat transfer. Thus, in addition to some undergraduate knowledge of heat transfer, students taking this course are expected to be familiar with vector algebra, linear algebra, ordinary differential equations, particle and rigid-body dynamics, thermodynamics, and integral and differential analysis in fluid mechanics. The use of computers is essential both for the purpose of computation as well as for display and visualization of results. At present these notes are in the process of being written; the student is encouraged to make extensive use of the literature listed in the bibliography. The students are also expected to attempt the problems at the end of each chapter to reinforce their learning. I will be glad to receive comments on these notes, and have mistakes brought to my attention.

Mihir Sen Department of Aerospace and Mechanical Engineering University of Notre Dame

Copyright c by M. Sen, 2008

i

Contents
Preface i

I

Review
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
2 2 2 2 3 3 3 3 5 5 5 7 8 8 8 9 9 11 11 11 11 11 11 11 13 14 15 15 15 15

1 Introductory heat transfer 1.1 Fundamentals . . . . . . . . . . . . . . . . 1.1.1 Definitions . . . . . . . . . . . . . 1.1.2 Energy balance . . . . . . . . . . . 1.1.3 States of matter . . . . . . . . . . 1.2 Conduction . . . . . . . . . . . . . . . . . 1.2.1 Governing equation . . . . . . . . 1.2.2 Fins . . . . . . . . . . . . . . . . . 1.2.3 Separation of variables . . . . . . . 1.2.4 Similarity variable . . . . . . . . . 1.2.5 Lumped-parameter approximation 1.3 Convection . . . . . . . . . . . . . . . . . 1.3.1 Governing equations . . . . . . . . 1.3.2 Flat-plate boundary-layer theory . 1.3.3 Heat transfer coefficients . . . . . . 1.4 Radiation . . . . . . . . . . . . . . . . . . 1.4.1 Electromagnetic radiation . . . . . 1.4.2 View factors . . . . . . . . . . . . 1.5 Boiling and condensation . . . . . . . . . 1.5.1 Boiling curve . . . . . . . . . . . . 1.5.2 Critical heat flux . . . . . . . . . . 1.5.3 Film boiling . . . . . . . . . . . . . 1.5.4 Condensation . . . . . . . . . . . . 1.6 Heat exchangers . . . . . . . . . . . . . . 1.6.1 Parallel- and counter-flow . . . . . 1.6.2 HX relations . . . . . . . . . . . . 1.6.3 Design methodology . . . . . . . . 1.6.4 Correlations . . . . . . . . . . . . . 1.6.5 Extended surfaces . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . .

ii

CONTENTS

iii

II

No spatial dimension
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17
18 18 19 20 21 21 21 22 22 22 23 24 25 25 26 26 27 27 28 28 29 29 30 31 31 31 31 31 32 32 33 33 33 34 35

2 Dynamics 2.1 Variable heat transfer coefficient . . . 2.1.1 Radiative cooling . . . . . . . . 2.1.2 Convective with weak radiation 2.2 Radiation in an enclosure . . . . . . . 2.3 Long time behavior . . . . . . . . . . . 2.3.1 Linear analysis . . . . . . . . . 2.3.2 Nonlinear analysis . . . . . . . 2.4 Time-dependent T∞ . . . . . . . . . . 2.4.1 Linear . . . . . . . . . . . . . . 2.4.2 Oscillatory . . . . . . . . . . . 2.5 Two-fluid problem . . . . . . . . . . . 2.6 Two-body problem . . . . . . . . . . . 2.6.1 Convective . . . . . . . . . . . 2.6.2 Radiative . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . 3 Control 3.1 Introduction . . . . . . . . . . . . . . . 3.2 Systems . . . . . . . . . . . . . . . . . 3.2.1 Systems without control . . . . 3.2.2 Systems with control . . . . . 3.3 Linear systems theory . . . . . . . . . 3.3.1 Ordinary differential equations 3.3.2 Algebraic-differential equations 3.4 Nonlinear aspects . . . . . . . . . . . . 3.4.1 Models . . . . . . . . . . . . . 3.4.2 Controllability and reachability 3.4.3 Bounded variables . . . . . . . 3.4.4 Relay and hysteresis . . . . . . 3.5 System identification . . . . . . . . . . 3.6 Control strategies . . . . . . . . . . . 3.6.1 Mathematical model . . . . . . 3.6.2 On-off control . . . . . . . . . . 3.6.3 PID control . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . .

III

One spatial dimension
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . of convective fin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37
38 38 38 38 39 41

4 Conduction 4.1 Structures . . . . . . . . . 4.2 Fin theory . . . . . . . . . 4.2.1 Long time solution 4.2.2 Shape optimization 4.3 Fin structure . . . . . . .

. . .1 Hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . .9. . . . . . . . . 5. . . . . 5. . . . . . .5. . . . . . . . . 4. . . . . . . . . 4. . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . .8. . . . . . . . .9. . . . . . . . . .5 Periodic inlet and ambient temperature . . . 5. . . . . . .5 Perturbations of one-dimensional conduction 4. . . . . . . . . . . . . . . . . . . . .2 Momentum equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . .6 Transient conduction .2 Energy equation . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Annular fin . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Steady state . . . . . . . . . . . . . . . . . . . .8 Nonlinear diffusion . . . . . . . . . . . . . . . . . . 4. . . . . . . .9. . . . . . . . . .1 Known heat rate . . . . . . . . .13 Multiple scales . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . 5. 4. . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Flow between plates with viscous dissipation . . . . . . . . . . . . 5. . . . . . . 5. . . . . . . . . . . . . . . . . . . . . 5. . 5. . . . 5. . . . . . . . . 4. . . . . . . . . . . . . . . . . . . .9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Single duct . . . . . . . .3. . . . . . 5. . . . . . . . . . . . . . . . . . . .6 Effect of wall . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Linear . . . . . . . . . . . .4 Two-fluid configuration . .1 Mass conservation . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . 4. . 5. 41 43 43 43 44 46 47 48 51 51 51 52 52 53 55 55 57 57 57 57 59 60 61 62 62 63 64 64 64 65 65 66 67 68 69 69 70 73 73 73 73 74 74 76 5 Forced convection 5. . . . .9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . 4. . . .3. . . .2 Thermal networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Long time behavior . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . .12 Thermal control . . . . . . . . . . .CONTENTS iv Fin with convection and radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Control with heat transfer coefficient . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. 5. . . . . . . . . . . . . . . . . . . . . 5. . . . . .1. . . . . . . . . . . . . . . . . . .5.2 Eccentric annulus . . . . . . . . . . . . .4 . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . .1 Hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Constant ambient temperature . . . . . . . .4 Temperature in long duct . . . . . . . . . . . . . . . . 5. . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Two rooms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Temperature-dependent conductivity . . . . . . .9 Thermal control . . . . . . .4. 4. . . . . . . . .2 Multiple room temperatures . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . .7 Linear diffusion . . . .10 Self-similar structures . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . Problems . . . . . . . 5. . . . . . . . .11 Non-Cartesian coordinates . . . . . . . . . . . . . . . .8 Networks . . . . . . . . . . . . . . . . . . . . . .2 Unsteady dynamics . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . .7 Radial flow between disks . 4. . . .3 Perfectly insulated duct . . . . . . . .6 Regenerator . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . .2 Convection with known outside temperature 5. . . .9 Stability by energy method . . . . . . . . . .2. .2. . . . . . .2 Nonlinear .3. . . . . . . . . . . . . . . . .

. . . . .2. . . .5 Perturbation of one-dimensional flow . . . . . . . . . .1 Modeling . . . . . . . . . 9 Forced convection 9. . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . .4.2. . . .1 Two-dimensional flow 9. . . . . . . . . . . . . . . . . . . . . . . .4 Dynamic analysis . . . . . . . . . Problems . . . 9. . . . . . . . . . . .5 Nonlinear analysis . . . . . . . .6 Thermal control . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Toroidal geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Nonlinear analysis . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Steady state. . . . . . 9. .2 Known heat rate . . . . . . . 6. . . . . . . . . . . . . . . . . .2 Axial conduction effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. .1 Steady-state problems . .1 Neumann’s solution . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . 9. . . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . 123 7. . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Mixed condition . . . . . . .6 Falkner-Skan boundary flows Problems . . . . . . 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . .4 Multiple solutions . . . . . . . . . . . . . .2 Steady State . . 6. .1 Mass conservation . . . . . 9. . 6. . . . .1. . . . . . . 124 IV Multiple spatial dimensions . . . . . . . . . . . . . . . . . . . . 8. . . . . . . . . .3 Energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Two-dimensional fin 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . .3 Leveque’s solution . . . . . .1. . . . . . . 8. . . . . . . . . . . . . . . . . . . . 6. . . .3 Radiating fins . . . . . . . .3 Known wall temperature . . . . . . . . 6. . . . . . 6.1 Low Reynolds numbers .2 Transient problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Momentum equation . . . . . . . . . . .5 Plate heat exchangers . . . . 6. . . . . . . . . . . . . . . 6. . . . . . . . . . . . .1. . . . . . . . . . . . . . . . .1 Stefan problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . .4 Non-Cartesian coordinates . . . 8. . . . 6. . . . . . . . . . . . . . . . 125 126 126 126 126 127 127 127 128 128 128 128 128 128 128 132 132 8 Conduction 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . .CONTENTS v 6 Natural convection 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Potential flow . . . . . . .1.3 Dynamic Analysis . . . . . .2 Goodman’s integral . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 7. . . . . . . . . . . . . . .1 Modeling . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 78 78 79 80 81 81 85 88 93 101 106 107 108 109 112 116 119 119 119 7 Moving boundary 123 7. . . . . . . . . no axial conduction 6. . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . 11. . . 11. . . . . . . . .1. .1 Mathematical models . 11. .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15. . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Diffusion by random walk . . . . . . . . .2 Multi-dimensional . . . . . . . . . .2. . . .3 Natural convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . .CONTENTS vi 10 Natural convection 10. . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . .1 Darcy’s equation . . . . . . . . . . .1 Governing equations . . . . . . . . . . . . . . . . . 11. . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Stefan problems . . . . . . . . . . . . . . 15. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Homogeneous nucleation . . . . . . .1 Plane wall at constant temperature . . . . . . . . . . . . . .2 Forchheimer’s equation . .3 Marangoni convection Problems . . . . 150 14 Boiling and condensation 151 14. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 Problems . . . .1. . . . . . . . . .1 Governing equations . . . . . . . . .1 Linear stability . . . . . 11. . . . .1 One-dimensional . . . . . . .2 Stagnation-point flow . . . . . . . . . . . . . . . . .2 Cavities . . . . . . . .3. . . . . . . . . . . . . 11. . . . . . . . . . . . . . . 15. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . 133 133 133 133 133 134 134 134 134 135 135 135 135 137 138 139 139 142 147 11 Porous media 11. . . . . . . . . . . . . . . . . . . . . . . . .3 Phonons . . . . . . . . . . . . . . . . . . .3 Thermal wakes . . . . 15. . . . . . . . 11. . . . . . . . . . . . . . . . . . . . . . 157 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15. . . . .2 Forced convection . . . . . . . . . .2 Steady-state inclined layer solutions Problems . . . . . . . . . . 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . 11. . . . . . . . . . . . . . . . . . . . . . . . 148 V Complex systems 149 13 Radiation 150 13. 15. . 10. . . . . . . . . . . . . 152 152 152 153 153 153 154 154 155 156 16 Bioheat transfer 157 16. . . . . . . . . .4 Thin films . . . . . . . . . . . . . . . . . 10.1 Single atom type . . . . 12 Moving boundary 148 12. . . . . . . . . . . . . . . . . . . . .3 Brinkman’s equation . . . . . . . . . . . . . . . . . . .3. .1 Relaxation-time approximation 15. . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Boltzmann transport equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Monte Carlo methods . 151 15 Microscale heat transfer 15. . . . . . . . . . . 11. . . . .2 Two atom types . . . . . . . . . . . .

. . . . . . . . . . . . . . 17. . . . 18. . . . . . . . . . . . . . . . . . . 17. .3 Heat transfer augmentation . . . . . . . . . . . . . . .10Thermal control . . .1. . . . . . . . . . 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9 Nonlinear analysis . . . . A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Cantor set . . . . . . . . . . . . . . . . . . . . .2 Artificial neural networks 18. . . . . . . . . . A. . . . . . . . . . . . . . . .1 Fractals . . . . . . . .12Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1.6 Unfolding and structural instability A. . . . . . . . .7 Transient behavior . . . . . . . . . . . . .3 Bifurcations . . . . . . . . . . . . 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.2 Routh-Hurwitz criteria . . . . . . . . . A. . . . . . . . . . .6 Radiation effects . . . .4 One-dimensional systems . . . . . . . . . .4 Maldistribution effects . . . . . . . . . . A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Microchannel heat exchangers . . . . . . . . . .4. . . . . . . . . . . . 17. . . . . . . . . .8 Correlations . . . . . . . . . . . . . . . . . . . . . . . . . 17. . . . . . . . . . . . . . . .3 Vector spaces . . .2 Koch curve . . . . . . . . . .4. . . . . . .4 Dynamical systems . . . . . . . . A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Singularity theory . . . . . . .4 Weierstrass function . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. . . . . . . . . . . . . .7 Two-dimensional systems . . . . . . . . . . . . . . . . . . . . . . 18 Soft computing 18. . . . . . . . . . . . . . . . . . . . . . . . . . . . A. . . . . . . . . .8. . . . . 17. .4.2 Porous medium analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17. . . . . . . . . . . .11.11Control of complex thermal systems 17. .CONTENTS vii 17 Heat exchangers 17. . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. . .1 Genetic algorithms . .11. . . A. . . . . . . . . . . . . . . . A. . . . . . . . . . . . . . . . . Problems . . . . . . .1 Fin analysis . . . . . . . . . . . . . . . . . . .2 Other applications . .9 Compressible flow . . . . . . . . . . . . . . . . . . . .3 Knopp function . . . . . . . . . . . . . . . . . . . .8 Three-dimensional systems .6 Mandelbrot set . . . . . . . . . . . . . . . . . . . . . . . . 158 158 158 158 158 158 158 158 159 159 159 159 161 161 162 165 165 166 166 166 166 166 VI Appendices . . .2. . . . . . . . . . . . . . . . . . . . .2 Perturbation methods . . . . . . . . . A. . 17. . 17. . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Stability . . . . . . .4. . . . . . . . . . .1. . .4. . . . . . . . .5 Julia set . . . . . . . .1 Least squares method . . . . . . . . . . . . 17. . . . . A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 170 170 171 171 171 171 172 172 172 172 172 173 173 174 174 176 177 177 179 181 181 A Mathematical review A. . . . A. A. . . . . . . . . . . . . . . . A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . .1 Hydronic networks . . . . A. . . . . . . .4. . . . . . . . . . . . . . . .1 Heat exchangers . . . . . .5 Examples of bifurcations . 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. . . . . . . . . . . . . . . . . . .4. . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. .3. . . . . . .6 Partial differential equations . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . 182 182 183 183 184 184 184 186 187 187 187 187 187 187 188 189 209 221 B Numerical methods B. . . . . . . . . . . . . . . B. B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. . . . . B. . . . . . .5 Moments . . . . . . . . . . . . . . . . . . . . . . . . . .2 Finite element methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Chebyshev Galerkin . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . A. . . . . . . . . . . . .CONTENTS viii A. . . . . . . . . . . . . . . . . . . . . . . . .4 MATLAB . . . . . . . . . .6. . . . . . . B. . . . . . . . . .4 Legendre Galerkin . . . . . . . . . .2 Trigonometric collocation B. . . . . . . . . . . .3. . . . . . . . . . . . . . . . C Additional problems References Index . . . . . . . . . . . . . .7 Waves . . . . . . .1 Finite difference methods . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . B. . . . . . . . . . . .1 Trigonometric Galerkin . . . . . . . . . . . . . . . . . . . . .3 Spectral methods . . . . . . . . .1 Eigenfunction expansion A. . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.

Part I Review 1 .

The zeroth law states that if each of two bodies are in thermal equilibrium with a third. including vibrational and rotational motion. 112. 19. 76. so that the heat flux coming in must be equal to that going out. 1.Chapter 1 Introductory heat transfer It is assumed that the reader has had an introductory course in heat transfer of the level of [12. 90. A classic work is that of Jakob [94]. The third law says that the entropy of an isolated system cannot decrease over time. then they also are in equilibrium with each other. 1 We will not distinguish between the specific heat at constant pressure and that a constant volume. 117. A surface cannot store energy. If there is no work transfer. The laws of thermodynamics govern the transfer of heat. for example. 188.1 Fundamentals Definitions Temperature is associated with the motion of molecules within a material. According to the first law. 65. 20. 212]. Two bodies are in thermal equilibrium with each other if there is no transfer of heat between them. the increase in internal energy is equal to the net heat and work transferred in. 211]. Example 1. 210. 111. 183. 207.2 Energy balance The first law gives a quantitative relation between the heat and work input to a system.1 1. 155.1. then ∂T Mc =Q (1. 122. 26. 177. 138. [206. 34. Both heat transfer and work transfer increase the internal energy of the body. 88. More advanced books are. 80. 1.1. 24.1 Show that the above statement of the third law implies that heat is always transferred from a high temperature to a low. 2 . being directly related to the kinetic energy of the molecules. The change in internal energy can be written in terms of a coefficient of specific heat1 as M c dT . Heat is the energy transferred between two points at different temperatures.1) ∂t where Q is the heat rate over the surface of the body. 22. 14. 106.

thermodynamics dictates the rules under which these changes are possible. and the radiative loss from the side are qk qh qr 2 Also dT dx = hdAs (T − T∞ ) 4 = σdAs (T 4 − T∞ ) = −ks A (1. liquids and gases as well as the transformation of on to the other. 1. The energy flows are indicated in Fig.2.2 Fins (1. . 1. 149] The Fourier law of conduction is q = −k∇T (1.1 Governing equation ∂T = α∇ (k · ∇T ) + g ∂t 1. T (x) is the temperature field. 100.2. Again.2 Conduction [31. The conductive heat flow along the fin.8) (1. 137.3) Fin effectiveness ǫf : This is the ratio of the fin heat transfer rate to the rate that would be if the fin were not there. Longitudinal heat flux Tb − T∞ ′′ ) (1.1. Conduction 3 1. 1.1. 1.2) where q is the heat flux vector.5) The transverse heat flux can be neglected compared to the longitudinal if ′′ ′′ qx ≫ qt (1. 68.2. 77.9) (1. Fin efficiency ηf : This is the ratio of the fin heat transfer rate to the rate that would be if the entire fin were at the base temperature.2.7) Consider the fin shown shown in Fig. we will define the enthalpy of transformation2 as the change in enthalpy that occurs when matter is transformed from one state to another.6) which gives a condition on the Biot number Bi = hL ≪1 k (1. For the moment. the convective heat loss from the side.10) called the latent heat.1.3 States of matter We will be dealing with solids.4) qx = O(ks L Transverse heat flux ′′ qt = O(h(Tb − T∞ )) (1. 66. and k(T ) is the coefficient of thermal conductivity. 136.

for a small enough slope. P (x) ≈ dAs /dx is the perimeter. The different types of boundary conditions for the tip are: ρAc • Convective: ∂T /∂x = a at x = L • Adiabatic: ∂T /∂x = 0 at x = L • Known tip temperature: T = TL at x = L .11) ∂ ∂T ∂T 4 − ks (A ) + P h(T − T∞ ) + σP (T 4 − T∞ ) = 0 (1. Conduction 4 T ∞ T b x L Figure 1.12) ∂t ∂x ∂x where ks is taken to be a constant. q (x)+q (x) r h T ∞ q (x) k q (x+dx) k Figure 1. 0) = Ti (x). where.2.1: Schematic of a fin. Heat balance gives ρAc from which ∂qk ∂T + dx + qh + qr = 0 ∂t ∂x (1.1.2: Energy balance. Usually the base temperature Tb is known. The initial temperature is T (x.

15) (1. 0) = f (x).13) (1.3.22) Consider a wall with fluid on both sides as shown in Fig.1 and T∞.4 Similarity variable ∂2T ∂T =α 2 ∂t ∂x 1.17) where the subscript indicates quantities at the base.14) (1.21) ∂θ ∂ − ∂τ ∂ξ a ∂θ ∂ξ + m2 pθ + ǫp (θ + β)4 − β 4 = 0 (1. .1 and Tw. The fluid temperatures are T∞.5 Lumped-parameter approximation (1. Conduction 5 • Long fin: T = T∞ as x → ∞ Taking θ τ ξ a(ξ) p(ξ) = = = = = T − T∞ Tb − T∞ ks t Fourier modulus L2 ρc x L A Ab P Pb (1. 1.2. the fin equation becomes a where m2 = Pb hL2 ks Ab σPb L2 (Tb − T∞ )3 ks Ab T∞ Tb − T∞ (1.1.16) (1.20) (1.23) (1.2.2. ∇2 T = 0 Let T (x. The initial temperature in the wall is T (x.3 Separation of variables = Steady-state coduction in a rectangular plate. y) = X(x)Y (y).18) ǫ = β 1.2 .19) (1.2. 1.2 and the wall temperatures are Tw.

2 Tw.27) The Biot number is defined as Bi = Transient hL ks (1.1 ) + h2 (T − T∞.1 − T∞.2 T∞. dT + h1 (T − T∞. 0 0 In the short time scale conduction within the slab is important.28) ∂T ks ∂ 2 T = ∂t ρc ∂x2 (1. and changes due to convection.26) (1. The ratio of the two tk /th = Bi.1 − Tw.1 Tw.2 T∞.1 − T∞. the temperature within the slab is uniform.2 T∞.1 Tw. we have h1 (T∞.2 ks T∞.2 Tw. In the long scale.3: Wall with fluids on either side.25) (1.30) .2 − T∞.1 − Tw.2 = h2 (Tw.2 − T∞.1 − T∞.2 h1 L ≪1 ks h2 L if ≪1 ks if (1.1 − Tw.1 − T∞.1 − Tw.2 Tw.1 − T∞.2 ) = 0 dt (1. Steady state In the steady state.1.2 .2 h1 L T∞. and convection from the sides is not.1 − T∞.2 T∞.2 Figure 1.1 = Tw.2 h2 L Tw.1 − T∞.1 ≪ T∞.1 = = ks T∞.2 − T∞.1 − Tw. In the long time scale it is possible to show that 0 0 Lρs c where T = Tw.2 T∞.2.2 ≪ T∞.2 ) L (1.1 − Tw. Conduction 6 T∞.1 − Tw.24) Thus we have Tw.29) There are two time scales: the short (conductive) tk = L2 ρc/ks and the long (convective) th = Lρc/h.1 ) = ks from which Tw.

(1. The governing equation is Mc dT + hA(T − T∞ ) = 0 dt (1.4: Convective cooling.35) This is shown in Fig. Convection 7 T∞ T Figure 1. is placed in an environment of different temperature. (1.3. Tf is that of the fluid. Thus. 99. 27. Tb is its temperature. 95.34) 1. 104. Convective cooling A body at temperature T .32) (1. We nondimensionalize using θ τ = = T − T∞ Ti − T∞ hAt Mc (1.31) is dθ +θ =0 dτ the solution to which is θ = e−τ (1. 21. T∞ . 133].1.5 where the nondimensional temperature goes from θ = 1 to θ = 0.4. we can write q = hA(Tb − Tf ). such as that shown in Fig. 1. and h is the coefficient of thermal convection. 98. 67. The dimensional time constant is M c/hA. 1. . Newton’s law of cooling: The rate of convective heat transfer from a body is proportional the difference in temperature between the body and the surrounding fluid.36) where A is the surface area of the body. 102.31) with T (0) = Ti .33) The nondimensional form of the governing equation (1.3 Convection [11. and is being convectively cooled.

1.37) (1.5: Convective cooling.39) = −∇p + µ∇2 V + f = ∇ (k · ∇T ) + Φ Forced convection Natural convection 1.2 Flat-plate boundary-layer theory = 0 (1.3.3 Heat transfer coefficients Overall heat transfer coefficient Fouling Bulk temperature Nondimensional groups . Convection 8 θ 1 τ Figure 1.3.1.1 Governing equations For incompressible flow ∇·V ∂V ρ + V · ∇V ∂t ∂T + V · ∇T ρc ∂t 1.3.38) (1.3.

mirror-like with angles of incidence and reflection equal) or diffuse (i. equal in all directions). unit area. 127. The absorptivity αλ . 209] Emission can be from a surface or volumetric.42) (1. the reflectivity ρλ .44) (1. For the overall energy. and transmissivity τλ are all functions of the wavelkength λ.46) Integrating over all wavelengths α+ρ+τ =1 The emissivity is defined as ǫλ = Eλ (λ.47) where the numerator is the actual energy emitted and the denominator is that that would have been emitted by a blackbody at the same temperature. while the radiosity is the energy leaving including the emission plus the reflection. Radiation 9 Reynolds number Re = UL ν ν Prandtl number = κ hL Nusselt number N u = k Stanton number St = N u/P r Re Colburn j-factor j Friction factor f = St P r 2τw = ρU 2 2/3 (1.e. The spectral intensity of emission is the radiant energy leaving per unit time. T ) Ebλ (λ. Monochromatic radiation is at a single wavelength.41) (1. T ) (1. and unit solid angle.48) (1.45) 1.4. Kirchhoff’s law: αλ = ǫλ and α = ǫ.4.50) E(T ) Eb (T ) (1. Irradiations is the radiant energy coming in.43) (1.1.40) (1. 1. unit wavelength.e.4 Radiation [25. 58. we have a similar definition ǫ= so that the emission is E = ǫσT 4 For a gray body ǫλ is independent of λ. Also αλ + ρλ + τλ = 1 (1.1 [173] Electromagnetic radiation (1.49) . The direction distribution of radiation from a surface may be either specular (i. The emissive power is the emission of an entire hemisphere.

D. J. g is the electrical conductivity. Thermal radiation is the part of the spectrum in the 0.56) (1.60) Eb = 0 Eλ dλ (1.59) Wien’s law: Putting dEλ /dλ = 0.52) where H. The frequency f and wavelength λ of a wave are related by c = fλ (1. it can be shown that ∇2 H − ǫµ ∂2H ∂H − gµ 2 ∂t ∂t ∂E ∂2E 2 ∇ E − ǫµ 2 − gµ ∂t ∂t = = 0 0 (1.2566 × 10−6 NC−2 s2 .54) (1. and ρ are the magnetic intensity. electric displacement.998 × 108 m/s. electric field. respectively.53) (1. For ρ = 0 and constant ǫ.4. J = gE (Ohm’s law).62) . current density.57) (1. g and µ. the maximum of is seen to be at λ = λm . and charge density.670 × 10−8 W/m2 K4 . magnetic induction. d2 T = λ2 T 4 dx2 (1.58) √ The speed of an electromagnetic wave in free space is c = 1/ µǫ. Maxwell’s equations of electromagnetic theory are ∇×H = J+ ∂D ∂t ∂B ∇×E = − ∂t ∇·D = ρ ∇·B = 0 (1. and µ = 1. For free space ǫ = 8.8542 × 10−12 C2 N−1 m−2 . Stefan-Boltzmann’s law: The total radiation emitted is ∞ (1. B.51) The radiation can also be considered a particles called phonons with energy E= f where is Planck’s constant.55) (1. For linear materials D = ǫE. and µ is the permeability. and B = µH. E. Radiation 10 Electromagnetic radiation travels at the speed of light c = 2.8 µmK. Blackbody radiation Planck distribution [147] Eλ = C1 λ5 [exp (C2 /λT ) − 1] λm T = C3 and C3 = 2897. where ǫ is the permittivity.1–100 µm range.1.61) = σT 4 where σ = 5. where (1.

4.6 Heat exchangers [105. . 1. A possible classification of HXs is shown in Table 1.6: Parallel and counter flow.3 1. 154] Shell and tube heat exchangers are commonly used for large industrial applications.5. Boiling and condensation 11 cold (a) parallel flow hot cold (a) counter flow hot Figure 1.5 Boiling and condensation [29.2 View factors 1. The fin efficiency concept was introduced by Harper and Brown (1922). Compact heat exchangers are also common in industrial and engineering applications that exchanger heat between two separated fluids.4 Boiling curve Critical heat flux Film boiling Condensation Nusselt’s solution 1.5.5.1.1.2 1. 42. weight and volume of the heat exchanger. 198] 1.1 1. The term compact is understood to mean a surface to volume ratio of more than about 700 m2 /m3 . The effectiveness-N T U method was introduced by London and Seban in 1941.5.5. The advantages are savings in cost.

(e) shell and tube divided flow Plate Number of fluids Two fluid Three fluid Multifluid Heat transfer Single-phase convection mechanisms on both sides Single-phase convection on one side. 1981 Types of HXs Examples Direct contact Indirect contact (a) direct transfer. two-phase convection on other side Two-phase convection on both sides Combined convection and radiative heat transfer According to Transfer processes . Heat exchangers 12 Table 1. (b) extended surface cross parallel flow.1. (b) spiral. (b) storage. (c) lamella Extended surface (a) plate fin. (c) shell and tube parallel counterflow shell and tube mixed. (b) tube fin Regenerative (a) rotary disk (b) rotary drum (c) fixed matrix Flow arrangement Single pass (a) parallel flow (b) counterflow (c) crossflow Multipass (a) extended surface cross counter flow.1: Classification of HX (due to Shah [170]. (d) shell and tube split flow.6. (c) fluidized bed Surface Compact compactness Non-compact Construction Tubular (a) double pipe (b) shell and tube (c) spiral tube Plate (a) gasketed.

1 Parallel. Energy balances give dq dq dq = U (Th − Tc ) dA = ±mc Cc dTc ˙ (1.71) is the logarithmic mean temperature difference.1.6.o )] (Th.i )] q(x) mc Cc ˙ q(x) = Th.1 ± (1.63) (1. we get 1 1 −dq = d(Th − Tc ) (1.6.73) We an write the element of area dA in terms of the perimeter P as dA = P dx.67) which can be integrated from 1 to 2 to give −U A From equation (1. and 2 the other end.65) = −mh Ch dTh ˙ where the upper and lower signs are for parallel and counterflow. Heat exchangers 13 1.75) . so that Tc (x) Th (x) = Tc. 1 the end where the hot fluids enters.64) and (1. From equations (1.69) 1 1 ± mh Ch ˙ mc Cc ˙ = ln (1.o − Tc. For parallel flow.63).o )/(Th.o − Tc.o − Tc.66).i − Tc.i ) ln[(Th. we find that −U dA 1 1 ± mh Ch ˙ mc Cc ˙ = d(Th − Tc ) Th − Tc (Th − Tc )1 (Th − Tc )2 (1.70) (1.i − Tc.68) where qT is the total heat transfer rate.o ) ln[(Th. respectively. i and o for inlet and outlet.o − Tc.i )/(Th.i − Tc.i − Tc. The last two equations can be combined to give qT = U A∆Tlmtd where ∆Tlmtd = (Th − Tc )1 − (Th − Tc )2 ln[(Th − Tc )1 /(Th − Tc )2 ] (1.66) ± mh Ch ˙ mc Cc ˙ Using (1.o ) − (Th.and counter-flow We define the subscripts h and c to mean hot and cold fluids.74) (1.72) (1. we get −qT 1 1 + mh Ch ˙ mc Cc ˙ = (Th − Tc )2 − (Th − Tc )1 (1.64) (1.1 − mh Ch ˙ (1.65).i ) − (Th. we have ∆Tlmtd = while for counterflow it is ∆Tlmtd = (Th.

one fluid mixed. (a) Counterflow 1 − exp[−N T U (1 − CR )] ǫ= (1.85) . the other unmixed If the unmixed fluid has C = Cmin . Heat exchangers 14 Thus dq + qU P dx 1 1 ± mh Ch ˙ mc Cc ˙ (1. (b) Parallel flow ǫ= (1.82) The heat capacity rate ratio is CR = Cmin /Cmax . Cc ) Assuming U to be a constant.81) (1.i ) Cmin (Th.86) (1. Effectiveness-N T U relations In general.83) 1 − CR exp[−N T U (1 − CR )] 1 − exp[−N T U (1 − CR )] 1 + CR so that ǫ → 1 as N T U → ∞. both fluids unmixed Series solution (Mason.80) Cmin = min(Ch .1.78) (1.76) With the boundary condition q(0) = 0.i ) Cc (Tc. the number of transfer units is NTU = AU Cmin (1.1 1 1 mh Ch ± mc C c ˙ ˙ 1 − exp −U P 1 1 ± mh Ch ˙ mc Cc ˙ (1. the effectiveness is a function of the HX configuration.o − Tc.6.6.i ) ǫ = = = where (1.1 − Tc.77) The HX effectiveness is Q Qmax Ch (Th.i − Th.2 HX relations Th. the solution is q9x) = 1.o ) Cmin (Th. then ǫ = 1 − exp[−CR (1 − exp{−N T U Cr })] But if the mixed fluid has C = Cmin ǫ = CR (1 − exp{−CR (1 − e−N T U )}) (1. 1954) (d) Crossflow.84) (c) Crossflow.i − Tc. its N T U and the CR of the fluids.i − Tc.79) (1.

For a perimeter corresponding to a fin slope that is not small. 2. Heat exchangers 15 (e) Crossflow.6. and σ is the ratio of free-flow area to frontal area. Af is the HX total fin area.6.88) (1. 1. ηf is the fin temperature effectiveness. If the external temperature.1. 1. T∞ (x). This is given by G2 ρ1 Aρ1 ρ1 ∆p = (Kc + 1 − σ 2 ) + 2( − 1) + f − (1 − σ 2 − Ke ) p1 2ρ1 p1 ρ2 Ac ρm ρ2 (1. the mean temperature difference. . 1. Find the temperature distribution within the wall. this method enables one to find the outlet temperatures.89) A where η0 is the total surface temperature effectiveness. ǫ.87) where Kc and Ke are the entrance and exit loss coefficients. Problems 1. its effectiveness.6.5 Correlations Extended surfaces η0 = 1 − Af (1 − ηf ) (1.11. find the general steady-state solution of the fin temperature T (x) in terms of a Green’s function. 1. The thermal conductivity varies with temperature in the form k(T ) = k0 + α(T − T1 ). Find the steady-state temperature distribution in the fin.and first-order Bessel functions of the first kind. derive Eq. A constant-area fin between surfaces at temperatures T1 and T2 is shown in Fig.8. Show that the efficiency of the triangular fin shown in Fig. and I0 and I1 are the zeroth. and its efficiency. The base is at temperature Tb and the tip at T∞ .4 1. The two sides of a plane wall are at temperatures T1 and T2 .3 Design methodology Mean temperature-difference method Given the inlet temperatures and flow rates. 1. and then the heat rate. the ambient temperature is also T∞ . 4. and A is the HX total heat transfer area. 3.7 is ηf = 1 I1 (2mL) . Consider a cylindrical pin fin of diameter D and length L.6. Effectiveness-NTU method The order of calculation is N T U . mL I0 (2mL) where m = (2h/kt)1/2 . qmax and q. Assume that there is only convection but no radiation. 5. is a function of the coordinate x. both fluids mixed (f) Tube with wall temperature constant ǫ = 1 − exp(−N T U ) Pressure drop It is important to determine the pressure drop through a heat exchanger.

x T1 T∞ (x) T2 Figure 1. show that the temperature of the plate. The fin in Problem 3 has a non-uniform diameter of the form D = D0 + ǫx. Determine the equations to be solved for a two-term perturbation solution for T (x) if ǫ ≪ 1. natural convection.1. is governed by dT + α(T − T∞ )5/4 = 0 dt Find T (t) if T (0) = T0 . T (t). 7.7: Triangular fin. 6.6. . Show that the governing equation in Problem 3 with radiation can be written as d2 T 4 − m2 (T − T∞ ) − ǫ(T 4 − T∞ ) = 0. Using a lumped parameter approximation for a vertical flat plate undergoing laminar. dx2 8.8: Constant-area fin. Heat exchangers 16 t L w Figure 1. Find a two-term perturbation solution for T (x) if ǫ ≪ 1 and L → ∞.

Part II No spatial dimension 17 .

we get dθ1 + θ1 dτ θ1 (0) the solution to which is Taking the expansion to order ǫ2 θ1 = −e−t + e−2τ dθ2 + θ2 dτ θ2 (0) (2. however.5) (2. .2) = = 0 1 (2.11) (2.6) (2.Chapter 2 Dynamics 2. the h is slightly temperature-dependent. then we have dθ + (1 + ǫθ)θ = 0 dτ which can be solved by the method of perturbations. we have dθ0 + θ0 dτ θ0 (0) which has the solution θ0 = e−τ To the next order ǫ1 .9) 2 = −θ0 (2. To order ǫ0 .1 Variable heat transfer coefficient If.4) (2.8) = −e−2τ = 0 = −2θ0 θ1 = −2e−2t − 2e−3τ = 1 18 (2. .12) .10) (2. We assume that θ(τ ) = θ0 (τ ) + ǫθ1 (τ ) + ǫ2 θ2 (τ ) + .1) (2.3) (2.7) (2.

15) (2. Separating variables.19) 2 −τ 2 = e 2. so that ln This can be rearranged to give θ= e−τ 1 + ǫ(1 − e−τ ) −1 θ(1 + ǫ) = −τ 1 + ǫθ (2.26) where . and time to be τ= and introducing the parameter β= we get σA(Ti − T∞ )3 t Mc T∞ Ti − T∞ (2.25) (2.24) dφ + φ4 = β 4 dτ φ=θ+β (2. Combining.1.32). we can write Mc dT 4 + σA(T 4 − T∞ ) = 0 dt (2. . we can find an exact solution to equation (2.1).21) If the heat loss is due to radiation.16) The condition θ(0) = 1 gives C = − ln(1 + ǫ)..2... Variable heat transfer coefficient 19 with the solution θ2 = e−τ − 2e−2τ + e−3τ And so on. (2.13) Alternatively.17) (2. . −2τ (2. we get dθ = dτ (1 + ǫθ)θ Integrating ln θ = −τ + C ǫθ + 1 (2.20) −3τ − 2e +e ) + . (2..22) Taking the dimensionless temperature to be defined in equation (1. we get θ = e−τ − ǫ(e−τ − e−2τ ) + ǫ2 (e−τ − 2e−2τ + e−3τ ) + .18) A Taylor-series expansion of the exact solution gives θ = e−τ 1 + ǫ(1 − e−τ = e −τ (2.14) (2.1 Radiative cooling −τ − ǫ(e 1 − ǫ(1 − e −τ −τ −e −2τ ) + ǫ (1 − e ) + ǫ (e 2 −τ ) + .23) (2.1.

Variable heat transfer coefficient 20 Writing the equation as the integral is 1 ln 4β 3 φ−β φ+β dφ = −dτ φ4 − β 4 − 1 tan−1 2β 3 φ β = −τ + C (2. we get dθ + θ + ǫθ4 = 0 dτ which has an exact solution τ= 1 + ǫθ3 1 ln 3 (1 + ǫ)θ3 (2.33).32) (2.30) with T (0) = Ti . . we get dθ + θ + ǫ (θ + β 4 )4 − β 4 = 0 dτ where β is defined in equation (2. . 2 +ǫ +6β 2 θ0 + ǫθ1 + ǫ2 θ2 + . .37) . . T∞ = 0. Using the variables defined by equations (1. we get d θ0 + ǫθ1 + ǫ2 θ2 + . + 4β θ0 + ǫθ1 + ǫ2 θ2 + .2. in equation (2.34) (2. + θ0 + ǫθ1 + ǫ2 θ2 + .35) (2. of we take β = 0. Substituting the perturbation series. dτ 4 3 θ0 + ǫθ1 + ǫ2 θ2 + . equation (2. =0 (2.31) If radiative effects are small compared to the convective (for Ti − T∞ = 100 K and h = 10 W/m2 K we get ǫ = 5.24). . .27) (2. . In this case 4β 3 θ0 + ǫθ1 + ǫ2 θ2 + . i.28) Using the initial condition θ(0) = 1. . .32) and (1.1.e.29) Convective with weak radiation The governing equationis Mc dT 4 + hA(T − T∞ ) + σA(T 4 − T∞ ) = 0 dt (2.67 × 10−3 ). .36) (2. we can take ǫ ≪ 1. .2). and ǫ= σ(Ti − T∞ )3 h (2. .31).2 1 1 (β + T )(β − 1) T −1 ln + tan−1 2β 3 2 (β − T )(β + 1) β + (T /β) (2.1. we get (?) τ= 2.33) dθ 4 + (θ − θ0 ) + ǫ(θ4 − θs ) = 0 dτ θ(0) = 1 As a special case.

48) (2.1 Linear analysis If we assume that θ′ is small.44) 2. Radiation in an enclosure 21 2. . The walls have no other heat loss and have different masses and specific heats. .46) (2. The steady state is T i = TH (i = 1. . from which dθ′ + bθ′ = 0 dτ θ′ = Ce−bτ so that θ → 0 as t → ∞ if b > 0.41) (2. .3. ′ (2.42) 2. The governing equations are N dTi 4 Mi ci Ai Fij (Ti4 − Tj4 ) + σAi FiH (Ti4 − TH ) = 0 (2. .40) (2.47) (2.2. then a Taylor series gives f (θ + θ′ ) = f (θ) + θ′ f ′ (θ) + . Let f (θ) = a Then we would like to show that θ → θ as t → ∞. Writing θ = θ + θ′ we have dθ′ + f (θ + θ′ ) = a dτ (2.49) where b = f ′ (θ).39) Linear stability is determined by a small perturbation of the type Ti = TH + Ti′ from which Mi ci This can be written as M dTi′ 3 3 + 4σTH Ai Fij (Ti′ − Tj′ ) + 4σTH Ai FiH Ti′ = 0 dt j=1 dT′ 3 = −4σTH AT′ dt N (2. .38) +σ dt j=1 where the view factor Fij is the fraction of radiation leaving surface i that falls on j.2.43) with θ(0) = 1.2 Radiation in an enclosure Consider a closed enclosure with N walls radiating to each other and with a central heater H. N ) (2. The solution is .45) (2.3 Long time behavior dθ + f (θ) = a dτ The general form of the equation for heat loss from a body with internal heat generation is (2.

1 Let Linear T∞ = T∞.52) (2. 2.4. we get dθ + θ = Aτ dτ (2. Thus θ′ → 0 as τ → ∞.50) (2.53) with 2.1: Convective cooling.4 Let Time-dependent T∞ Mc dT + hA(T − T∞ (t)) = 0 dt T (0) = Ti (2.1.33). f(θ ) f(θ ) θ θ Figure 2.0 Ti − T∞. 2. are both of the same sign or zero.54) Defining the nondimensional temperature as θ= and time as in equation (1.0 (2.0 + at (2.2. we get 1 d ′ 2 (θ ) = −θ′ f (θ + θ′ ) − f (θ) 2 dτ Thus d ′ 2 (θ ) ≤ 0 dτ (2.3.46) by θ′ .4. Time-dependent T∞ 22 2.51) if θ′ and [f (θ + θ′ ) − f (θ)].2 Nonlinear analysis Multiplying equation (2.56) T − T∞. as shown in Fig.55) .

2: Response to linear ambient temperature.59) θ = (1 + A)e−τ + A(τ − 1) The time shown in Fig.4.58) The condition θ(0) = 1 gives C = 1 + A. 2. Time-dependent T∞ 23 where A= The nondimensional ambient temperature is θ∞ = The solution to equation (2.4. Defining θ= T − T∞ Ti − T ∞ (2. 2.2 Let Oscillatory T∞ = T ∞ + δT sin ωt (2.64) . so that θ = Ce−τ + Aτ − A (2.62) ∆θ θ ∞ θ τc τ Figure 2.56) is aM c hA(Ti − T∞.60) (2.63) where T (0) = Ti . 1+A A (2.61) (2.57) (2.0 ) (2.2 at crossover is τc = ln and the offset is δθ = A as τ → ∞.0 ) aM c τ hA(Ti − T∞.2.

2 T T∞.73) .1 and T∞.68) (2.5. fluid 1 for instance.2 T T∞.72) (2. the nondimensional equation is dθ + θ = δθ sin Ωτ dτ where δθ = δT Ti − T ∞ ωM c hA (2. we get C = 1 + δΩ/(1 + Ω2 ).71) where T (0) = Ti . If T∞.1 Ti − T∞.1 ) + h2 A2 (T − T∞. Two-fluid problem 24 T∞.2 ) = 0 dt (2.65) Ω = The solution is θ = Ce−τ + where δθ sin(Ωτ − φ) (1 + Ω2 ) cos φ φ = tan−1 Ω (2.2 .66) (2.70) 2. we can nondimensionalize the equation using the parameters for one of them.1 h1 A1 t Mc (2. like in the two examples shown in Fig.69) From the condition θ(0) = 1.1 and T∞. so that θ= 1 + δθ Ω 1 + Ω2 e−τ + δθ sin(Ωτ − φ) (1 + Ω2 ) cos φ (2.3.67) (2.3: Two-fluid problems. 2. The governing equation is Mc dT + h1 A1 (T − T∞.5 Two-fluid problem Suppose there is a body in contact with two fluids at different temperatures T∞.2. and using equation (1. Thus we have θ τ = = T − T∞.1 T∞.33).2 are constants.1 (a) (b) Figure 2.

where α β The equation can be written as = = (2.35).1 Two-body problem Convective Suppose now that there are two bodies at temperatures T1 and T2 in thermal contact with each other and exchanging heat with a single fluid at temperature T∞ as shown in Fig.6.82) . the solution reduces to the single-fluid case.75) (2.6 2.2 Ti − T∞.4.1 − T∞. 1 2 Figure 2. 2.74) with θ(0) = 1.2.4: Two bodies in thermal contact.79) For α = 0. equation (1.78) The condition θ(0) = 1 gives C = 1 + αβ/(1 + α).6.77) with the solution (2. Otherwise the time constant of the general system is Mc t0 = (2.80) h1 A1 + h2 A2 2. from which θ= 1+ αβ 1+α e−(1+α)τ − αβ 1+α (2. The mathematical model of the thermal process is M1 c1 ks Ac dT1 + (T1 − T2 ) + hA(T1 − T∞ ) = Q1 dt L ks Ac dT2 + (T2 − T1 ) + hA(T2 − T∞ ) = Q2 M2 c2 dt L (2.76) dθ + (1 + α)θ = −αβ dτ θ = Ce−(1+α)τ − αβ 1+α (2.81) (2. Two-body problem 25 from which dθ + θ + α(θ + β) = 0 dτ h2 A2 h1 A1 T∞.1 (2.

T∞ Tmax δt Tmin t Figure 2. ǫ) surfaces for the convection with radiation problem. numerical solutions). and (b) the amplitude of oscillation of the system temperature.2 Radiative ks Ac dT1 4 4 4 4 + (T1 − T2 ) + A1 σF1s (T1 − Ts ) + A1 σF12 (T1 − T2 ) = Q1 dt L ks Ac dT2 4 4 4 4 + (T2 − T1 ) + +A1 F2s (T2 − Ts ) + A2 σF21 (T2 − T1 ) = Q2 M2 c2 dt L M1 c1 Without radiation Q1 = Q2 = − 2kA T1 − T2 L (2. 4. for a small period δt. 2. and comment on the existence of solutions.6: Ambient temperature variation. Complete the problem of radiation in an enclosure (linear stability. 7.85) Problems 1.5: Bodies with radiation. 2. varies with time in the form shown.6.83) (2. Find the steady-state temperatures for the two-body problem and explore the stability of the system for constant ambient temperature. Lumped system with convective-radiative cooling with nonzero θ0 and θs . Consider the change in temperature of a lumped system with convective heat transfer where the ambient temperature.6. T (t). T∞ (t). 5. Find (a) the long-time solution of the system temperature. Two-body problem 26 1 111111111111 000000000000 111111111111 000000000000 2 Figure 2. 3.84) (2. 6. T (t).2. Show that the temperature distribution in a sphere subject to convective cooling tends to become uniform as Bi → 0. Plot all real θ(β. Check one of the perturbation solutions against a numerical solution. .

but even then it will be impossible to go into any depth within the space available. which is closely related to and is often an integral part of thermal control.Chapter 3 Control 3. concentrate only on the basic principles relating to the theory of control as applied to thermal problems. rapid thermal processing of computer chips [84.152. 200]. 114]. radiation. but when put together presents such a massive computational problem so as to be practically intractable. Many controllers are also computer based. multi-phase flows. however. and the use of microprocessors [87. Most thermal systems are generally complex involving diverse physical processes. The most important of these are hardware considerations. each one of which can be analyzed and computed. There are many aspects of thermal control that will not be treated in this brief review. and there are many ways in which that can be done. hydrodynamic instability. a complex system can be defined as one that is made up of sub-systems. control may simply be manual. 82. turbulence. For this reason large. In this context. One can give the example of heat exchangers [85. or material properties that may not be accurately known.184. 27 . approximations due to using lumped parameters instead of distributed temperature fields. devices and plants is commonplace. The present paper will. transportation and domestic use that need to be controlled in some manner. If precise control is not required.218]. Approximate correlations from empirical data are also heavily used in practice. and many others. complex geometries. 180] and PCs in machines. nonlinearities and bifurcations. Most often some degree of approximation has to be made to the mathematical model. property variation with temperature. It is common to have large uncertainties in the values of heat transfer coefficients. Flow control. or chemical reaction. thermal packaging of electronic components [150. satellites [101.172.185].221]. commonly used engineering systems are hard to model exactly from first principles.115. This is only an introduction. or if the process is very slow. 158.1 Introduction There are many kinds of thermal systems in common industrial. has its own extensive literature [64]. environmental control in buildings [70.187] developed for measurement and actuation are used in the control of thermal systems. all kinds of sensors and actuators [59. Discrete-time (as opposed to continuoustime) systems will not be described. 72. manufacturing [54]. The two major reasons for which control systems are needed to enable a thermal system to function as desired are the approximations used during design and the existence of unpredictable external and internal disturbances which were not taken into account. otherwise some sort of mechanical or electrical feedback system has to be put in place for it to be automatic. and even when this is possible the dynamic responses of the models are impossible to determine computationally in real time. These include natural and forced convection.

ordinary or partial differential operator. There are good texts and monographs available on the basics of control theory [116. . and λ is a parameter that defines the system. u. 144. integral. u. It is important to point out that output controllability does not imply system controllability. The stability of a system is a property that leads to a bounded output if the input is also bounded. nonlinear control [91].1 Systems without control The dynamic behavior of any thermal system (often called the open-loop system or plant to distinguish it from the system with controller described below).2.93. x may be a spatial temperature distribution. 3. 3. A system is said to be observable if its state x can be uniquely determined from the input u and output y alone.151]. and mathematics of control [9. These are all topics that include and are included within thermal control. In fact. The relation between the two may be written as Lo (y. It is output controllable if the same can be done to the output. 179] that can be consulted. may be mathematically represented as Ls (x. A system is said to be controllable if it can be taken from one specific state to another within a prescribed time with the help of a suitable input. For example Ls may be an algebraic.89].2: Schematic of a system with comparator C and controller. process control [28.1: Schematic of a system without controller. it is hard to be specific at this stage. Figure 3. w(t) is some external or internal disturbance. In general.1) where Ls is a system operator. 3. t is time.2) where Lo is the output operator. The system is single-input single-output (SISO) if both u and y are scalars. 132. x(t) is the state of the system. in practice for many thermal systems the former is all that is required. while x may be a finite-dimensional vector or a function. (3. λ) = 0. For example.75. 157]. All possible values of the output constitute the reachable set. u(t) is its input. u(t) is usually a low-dimensional vector.3.2 Systems Some basic ideas of systems will be defined here even though. Each one of these quantities belongs to a suitable set or vector space and there are a large number of possibilities. Systems w(t) λ c u(t) E plant x(t) c c yr (t) E C e(t) j E controller u(t) E w(t) λ c plant x(t) c 28 p y(t) E p y(t) E Figure 3. infinite-dimensional systems [39. it has been reported that most industrial plants are controlled quite satisfactorily though they are not system controllable [156]. while y are the readings of one or more temperature measurement devices at a finite number of locations. (3. because of the generality involved.2. λ) = 0.1. schematically shown in Fig. w. and the interested reader should look at the literature that is cited for further details. x. w. the output of the system y(t) is different from its state x(t).

etc. This is usually used in heating or cooling systems in which the heat coming in or going out is reduced to zero when a predetermined temperature is reached and set at a constant value at another temperature. the figure actually shows unit feedback. (3. The controller itself is either entirely mechanical if the system is not very complex. relay.4) form a set of equations in the unknowns x(t). y(t) and u(t). The control process can be written as Lc (u. to the input side of the system. Since this is exclusively for linear systems. Another method is Proportional-Integral-Derivative (PID) control [214] in which t u = Kp e(t) + Ki 0 e(s) ds + Kd de . which is usually taken to be e = yr − y. The actuator can be a fan or a pump if the flow rate is to be changed. it receives the error in the output e(t) and puts out an appropriate control input u(t) that leads to the desired operation of the plant. If the changes desired in the output are very slow then manual control can be carried out. the output in effect is then insensitive to changes in input or disturbances. This is done using feedback from the output. This is a passive method of control that is used in many thermal systems.9)–(3. (3. A self-controlling approach that is sometimes useful is to design the system in such a way that any disturbance will bring the output back to the desired value. The controller designer has to propose a suitable Lc . It will work if the behavior of the system is exactly predictable. and then Eqs. Open-loop control is not usually effective for many systems. if precise output control is not required.3.5) 3. and tracking if it is function of time.3. It is common to use on-off (or bang-bang. Linear systems theory 29 3. Sensors that are commonly used are temperature-measuring devices such as thermocouples.) control. e.2.3) The key role is played by the controller which puts out a signal that is used to move an actuator in the plant. and that is also frequently done.4) where Lc is a control operator. (C.3 Linear systems theory The term classical control is often used to refer to theory derived on the basis of Laplace transforms.2 Systems with control The objective of control is to have a given output y = yr (t). In any case. The problem is called regulation if yr is a constant. 3.2. Internal disturbances may be noise in the measuring or actuating devices or a change in surface heat transfer characteristics due to fouling. that is one would have to determine u(t) such that y = yr (t) using the mathematical model of the system alone. or if the output of the system is not very sensitive to the input. while external ones may be a change in the environmental temperature. or a heater if the heating rate is the appropriate variable. λ) = 0. resistance thermometers or thermistors. Choice of a control strategy defines Lo and many different methodologies are used in thermal systems. as measured by a sensor. In open-loop control the input is selected to give the desired output without using any information from the output side. dt (3. For these cases closed-loop control is an appropriate alternative. There is a comparator which determines the error signal e(t) = e(yr . where the reference or set value yr is prescribed. For thermal systems contributing factors are the uncertainties in the mathematical model of the plant and the presence of unpredictable disturbances. y). or is a digital processor with appropriate software. we will be using the so-called modern control or state- . as shown in Fig.

9) and (3. B. . From Eq. Also. . and D can be functions of time in general. CA2 B . is of rank n. Similarly. It must be emphasized that the u(t) that does this is not unique. C ∈ Rp×n .3. T (3. . . 3. a u(t) can be found to take x(t) from x(t0 ) at t = t0 to x(tf ) = 0 at t = tf if and only if the matrix . .6) is t x(t) = eA(t−t0 ) x(t0 ) + t0 eA(t−s) Bu(s) ds. .2) take the form dx dt y = Ax + Bu. CA2 . Control theory can be developed for different linear operators. . . (3. Linear systems theory 30 space analysis which is based on dynamical systems. = Cx + Du. . it can be shown that the output y(t) is controllable if and only if . For a linear system. B.9) define the state x(t) and output y(t) if the input u(t) is given. . (3.7) where x ∈ Rn . . . . It can be shown that for the system governed by Eq. . P = C . . C. x. Though A. (3. . . . .1 Ordinary differential equations Much is known about a linear differential system in which Eqs. D ∈ Rp×m .6) (3. AB . . CAn−1 B ∈ Rp×(n+1)m . (3.3. C.3.9) Eqs. mainly because it can be extended to nonlinear systems. CA . . the state x(t) is observable if and only if the matrix .8) and (3. . A ∈ Rn×n . CB . CAB . . . we get t y(t) = C eAt x(t0 ) + t0 eA(t−s) Bu(s) ds + Du. .10) is of rank n. N = D . 2! 3! with I being the identity matrix. and some of these are outlined below.12) .6). (3. M = B . is of rank p. . and D are matrices of suitable sizes. u ∈ Rm . the issue is only one of preference since the results are identical. The system is then controllable.7). A2 B .8) where the exponential matrix is defined as eAt = I + At + 1 1 2 2 A t + A3 t3 + . here they will be treated as constants. . controllability from one state to another implies that the system can be taken from any state to any other. An−1 B ∈ Rn×nm . (C. u and y are vectors of different lengths and A. .11) ∈ Rpn×n (3. . (3. Where they overlap. . y ∈ Rp . B ∈ Rn×m . The solution of Eq. . (3. CAn−1 . (3. .

(3.3. (3. so that variables such as x and u in Eqs. flow rates and the like. 3. (3. some of which are ordinary differential and the rest are algebraic. u).4 Nonlinear aspects The following are a few of the issues that arise in the treatment of nonlinear thermal control problems. heat rates. dt (3.2 Controllability and reachability General theorems for the controllability of nonlinear systems are not available at this point in time.2 Algebraic-differential equations This is a system of equations of the form E dx = A x + B u. Eq.6) such as dx = f (x. but it may have restricted or R-controllability [45].7) are now nonlinear since the sum of solutions may fall outside the range in which x . If one is interested in local behavior about an equilibrium state x = x0 .4.13) cannot be converted into (3.4. 3. Thus regions of state space that are unreachable with the linearized equations may actually be reachable. Results obtained from the linearized equations generally do not hold for the nonlinear equations.4.15) where x = x0 + x′ and u = u′ . being temperatures.6) by substitution.2).9) and (3. 3. are evaluated at the equilibrium point.15) has the same form as Eq. The index of the system is the least number of differentiations of the algebraic equations that is needed to get the form of Eq. but one that can be used is a generalization of Eq.14) dt where f : Rn × Rm → Rn . 3. (3. The reason is that in the nonlinear case one can take a path in state space that travels far from the equilibrium point and then returns close to it.6) and (3. (C. (3. The Jacobian matrices (∂f /∂x)0 and (∂f /∂u)0 .4.6). even systems locally governed by Eqs. (3. If this happens.13) where the matrix E ∈ Rn×n is singular [113].6). are bounded.3 Bounded variables In practice. In a thermal convection loop it is possible to go from one branch of a bifurcation solution to another in this fashion [1]. u = 0.3. As a result of this. Eq. This is equivalent to a set of equations. due to hardware constraints it is common to have the physical variables confined to certain ranges. Nonlinear aspects 31 3. (3.1 Models There are no general mathematical models for thermal systems. this can be linearized in that neighborhood to give dx dt = ∂f ∂x x′ + 0 ∂f u′ ∂u 0 = Ax′ + Bu′ . The system may not be completely controllable since some of the components of x are algebraically related. (3.

a form of Ls is assumed with unknown parameter values. t3 . i=1 −∞ −∞ ki (t.5 System identification To be able to design appropriate control systems. like y(t) = y0 (t) + .3. (3. . of which there are several different possible definitions [10. one of which is the closed-affine model dx = F1 (x) + F2 (x)u (3.18) • Fractional-order derivatives. are also used. As an example. (3. 148] can be used in differential models. System identification 32 exists and thus may not be a valid solution. for a controllable system in which only u is bounded in a neighborhood of zero. (3. . other models that are used include the following.17. . This may be accompanied by hysteresis where the relationship also depends on whether the input is increasing or decreasing.5. .6). . (3. 135. t2 . 129].134.4 Relay and hysteresis A relay is an element of a system that has an input-output relationship that is not smooth. the Riemann-Liouville definition of the nth derivative of f (t) is n a Dt f (t) = where a and n are real numbers and m is the largest integer smaller than n.14). • Functional [71].. Through optimization routines the values of the unknowns are chosen to obtain the best fit of the results of the model with experimental information. x can reach any point in Rn if the eigenvalues of A have zero or positive real parts. . Mathematical models of these systems can be obtained in two entirely different ways: from first principles using known physical laws. . one needs to have some idea of the dynamic behavior of the thermal system that is being controlled. There are many different ways in which this can be done. t1 .4. Apart from the linear Eq. The latter is the process of system identification. 3..19) . the most common being the fitting of parameters to proposed models [141]. dti (3.17) for a SISO system. . u(ti ) dt1 . by which a complex system is reduced to mathematical form using experimental data [75. difference [23] or delay [57] equations such as dx = A x(t − s) + B u dt also appear in the modeling of thermal systems. • There are many forms based on Eq. 1 dm+1 Γ(m − n + 1) dtm+1 t a (t − s)m−n f (s) ds. 3. On the other hand. and the origin is reachable if the eigenvalues of A have zero or negative real parts [179].16) dt The bilinear equation for which F1 (x) = Ax and F2 (x) = N x + b is a special case of this. ∞ ∞ ∞ • Volterra models. ti )u(t1 ) . 121. and empirically from the analysis of experimental information (though combinations of the two paths are not only possible but common). Valves are typical elements in flow systems that have this kind of behavior. In this method. it may be discontinuous or not possess first or higher-order derivatives.

22) that θ → θr as t → ∞. and the control objective is to maintain the temperature of the body at a given level by manipulating the heat source. L is a characteristics length dimension of the body. Tmax − Tmin (3. (3. In practice.20) dt where M is the mass of the body.2 On-off control In this simple form of control the heat rate in Eq. If Bi < 0. to do this the dimensional parameters hAs and T∞ must be exactly known.24) 0 off dt the solution for which is θ= 1 + C1 e−t C2 e−t on . and goes off when it is rises above TU . Open-loop operation to maintain a given non-dimensional temperature θr is easily calculated. (3. this equation becomes Mc dθ + θ = Q(t) dt (3. c is its specific heat.1.26) (3. The other variables are now nondimensional. (3. it can be shown from the solution of Eq. off (3. Tmax − Tmin TU − Tmin . it is is either Q = Q0 or Q = 0. It also has an internal heat source Q(t) to compensate for this heat loss. 3.1 Control strategies Mathematical model Consider a body that is cooled from its surface by convection to the environment with a constant ambient temperature T∞ . Choosing Q = θr .23) Tmax − Tmin the governing equation is dθ 1 on +θ = . With x = θ. (3.21). Taking the non-dimensional temperature to be T − Tmin θ= (3.21) Here θ = (T − T∞ )/(Ti − T∞ ) where T (0) = Ti so that θ(0) = 1.27) . With the system in its on mode. Since this is usually not the case some form of feedback control is required. (3. respectively.6). (3. T → Tmax = T∞ + Q0 /hAs as t → ∞. Control strategies 33 3.20) has only two values.6 3. Under this lumped approximation the energy balance is given by dT + hAs (T − T∞ ) = Q(t). we find from Eq. depending on whether the heater is on or off.3. u = Q. that is θ(t) = (1 − θr )e−t + θr . The Biot number for the body is Bi = hL/k.6. (3. These lower and upper bounds are non-dimensionally θL θU = = TL − Tmin . where h is the convective heat transfer coefficient.6.10) that rank(M )=1. and in its off mode. so the system is controllable. and k is its thermal conductivity. n = m = 1 in Eq.6. the body can be considered to have a uniform temperature T (t).25) We will assume that the heat source comes on when temperature falls below a value TL . Using M c/hAs and hAs (Ti −T∞ ) as the characteristic time and heat rate. and As is the surface area for convection. T → Tmin = T∞ .

1 − θU θU ln . (3. It can be shown that the on and off time periods are ton toff = = ln 1 − θL . so that the derivative of Eq. The result of applying this form of control is an oscillatory temperature that looks like that in Fig.2 θL 0. (3.6.5 0..1 0 0 1 2 3 4 5 t Figure 3. 3. (3.30) If we make a small dead-band assumption.3 0. we can write θL θU where δ ≪ 1. dt dt .21) gives dθ d2 θ (3.31) (3.33) = θr − δ.. (3. Control strategies 34 1 0. A Taylor-series expansion gives tp = 2 δ 1 1 + θr 1 − θr + .3.3.29) respectively. The frequency of the oscillation increases as its amplitude decreases. θL (3. = θr + δ.8 θ 0.28) (3. 3. θL (1 − θU ) (3.7 0.3 PID control The error e = θr − θ and control input u = Q can be used in Eq.4 0.6 0.5).34) (Kd + 1) 2 + (Kp + 1) + Ki θ = Ki θr .9 θU 0.3: Lumped approximation with on-off control.6. the period and amplitude of which can be chosen using suitable parameters.32) The period is thus proportional to the width of the dead band. The total period of the oscillation is then tp = ln θU (1 − θL ) .

36) The response of the closed-loop system can be obtained as a solution. Ki and Kd will give overdamped or underdamped oscillatory or unstable behavior of the system.1.5 0 10 20 30 40 50 60 70 t Figure 3.6. Problems 1.5 −1 −1. A 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 B (a) Check that the system is controllable. The steady-state for Ki = 0 is given by θ = θr .35) at t = 0.9.4 shows two examples of closed-loop behavior with different parameter values. (3. Fig. (a) Kp = −0. Ki = Kd = −0.5 (b) 1 θ 0. It is desired to control their temperatures at TA and TB using separate internal heat inputs QA and QB . Control strategies 35 1.1. θr = 0.5 (a) 0 −0. Two lumped bodies A and B in thermal contact (contact thermal resistance Rc ) exchange heat between themselves by conduction and with the surroundings by convection. . 3. (b) Kp = −0. It can be appreciated that different choices of the controller constants Kp .5.3.4: Lumped approximation with PID control. Kp + 1 Kp θ r = − θ0 + Kd + 1 Kd + 1 (3. with the initial conditions θ dθ dt = θ0 at t = 0.

circulant. . with each at a uniform temperature Ti (t). ˙ (b) Find the steady state temperatures.3.. . 2. Determine the condition for linear stability of the control system. (c) Write the dynamical system in the form x = Ax and determine the condition for stability1 .. A number of identical rooms are arranged in a circle as shown. . . . 0 6 6 . 0 6 a b c .. 2. . . Plot TA (t) and TB (t) for selected values of the parameters. . 4 0 . Check for phase synchronization. 3. 0 a b c 0 . Show that the case of two independent bodies is recovered as Rc → ∞. where j = 1. . each room has a heater that is controlled by independent but identical proportional controllers.. . Control strategies 36 (b) Set up a PID controller where its constants are matrices. 0 a 3 a 0 7 7 0 7 7 7 7 7 c 5 b . Apply an on-off controller to the previous problem. . and nondimensionalize.6. To maintain temperatures.. Each room exchanges heat by convection with the outside which is at T∞ . . 0 6 6 0 a b . . . (c) Calculate and plot TA (t) and TB (t) for chosen values of the controller constants.. . and with its neighbors with a conductive thermal resistance R. of an N × N .. (a) Derive the governing equations for the system. banded matrix of the form 2 b c 0 . 6 . N . 6 .. .. T∞ i−1 i T∞ i+1 1 Eigenvalues are λj = b + (a + c) cos{2π(j − 1)/N } − i(a − c) sin{2π(j − 1)/N }..

Part III One spatial dimension 37 .

1: Complex conductive structures.1 Fin theory Long time solution ∂θ ∂ − ∂τ ∂ξ j The general fin equation is a a ∂θ ∂ξ + f (θ) = 0 (4.2) from which T0 = (4.1) For each branch i ki Ai (Ti − T0 ) = 0 Li i k i Ai Li Ti k i Ai i Li (4.1 shows a complex shape consisting of conductive bars. At each node qi = 0 i (4. 38 .3) 4.4) i Figure 4.1 Structures Fig.Chapter 4 Conduction 4. 4.2 4.2.

6) a + f (θ + θ′ ) − f (θ) = 0 a − ∂τ ∂ξ ∂ξ where θ′ is the perturbation from the steady state. If we also assume that I2 ≤ 0 (4. we have ∂θ′ ∂θ′ ∂ (4. The boundary conditions for θ′ are homogeneous. 4.14) dx2 where 1/2 2h (4. Fin theory 39 where f (θ) includes heat transfer from the sides due to convection and radiation.2.5) − a + f (θ) = 0 dξ dξ with the same boundary conditions. 4.13) then equation (4.15) m= ks δ . this is a consequence of the Second Law of Thermodynamics.2.2 Shape optimization of convective fin Consider a rectangular fin of length L and thickness δ as shown in Fig.12) = − a 0 since the first term on the right side of equation (4.10) θ′ ∂ ∂ξ a ∂θ′ ∂ξ = − 0 θ′ f (θ + θ′ ) − f (θ) dξ Integrating by parts we can show that I1 ∂θ′ = θa ∂ξ ′ 1 1 1 0 − dθ′ dξ a 0 2 dθ′ dξ dξ 2 dξ (4.5). The dimensional equation is d2 T − m2 (T − T∞ ) = 0 (4. The steady state is determined from d dθ (4.13) holds if [θ′ and f (θ + θ′ ) − f (θ)] are of the same sign or both zero. Condition (4.8) that E is nonnegative.9) (4.8) dξ (4. Substituting θ = θ + θ′ in equation (4. Multiplying by θ′ and integrating from ξ = 0 to ξ = 1. The boundary conditions are either Dirichlet or Neumannn type at ξ = 0 and ξ = 1.4) and subtracting (4.7) 1 2 1 a(θ′ )2 dξ 0 (4.11) (4. Thus we know from the above that I1 is nonpositive and from equation (4. we have dE = I1 + I2 dτ where E I1 I2 = = 0 1 1 (4.11) is zero due to boundary conditions.2. Thus the steady state is globally stable.7) tells us that E must decrease with time until reaching zero.4.

we get q = ks δ 2h ks δ 1/2 (4.2: Rectangular fin.4.4192. constant fin volume.19) x=0 = ks δ(Tb − T∞ )m tanh mL Writing L = Ap /δ.22) This is equivalent to where (4.21) Keeping Ap constant.24) Numerically. i. We will take the boundary conditions T (0) dT (L) dx The solution is The heat rate through the base per unit width is q = −ks δ T = T∞ − (T − T∞ ) [tanh mL sinh mx − cosh mx] dT dx (4. Fin theory 40 T b T ∞ δ L Figure 4. we find that βopt Ap δopt = 1.26) .18) = Tb = 0 (4.2. the heat rate can be maximized if δopt sech2 1/2 Ap δ 2h ks δopt Ap 2h 1/2 3 −5/2 1 −1/2 Ap (− )δopt + δopt tanh ks 2 2 δ 3βopt sech2 βopt = tanh βopt 2h ks δopt 1/2 =0 (4.17) (4.25) 1/2 2/3 Lopt = (4.23) (4.e. Thus βopt = δopt = Ap βopt βopt 2h ks δopt 1/2 2/3 ks Ap 2h ks Ap 2h (4.20) (Tb − T∞ ) tanh Ap δ 2h ks δ 1/2 (4.16) (4.

Fin structure 41 4. so that − Convective With only convective heat transfer.34) (4.1 If T (0) T (L) then show that " r « „ q !# sinh x hP kA hP „ q « + T0 cosh kA sinh L hP kA r hP kA ! = = T0 T1 (4.4 Fin with convection and radiation Steady state solutions Equation (4. 4.1 with convection.33) T (x) = T1 − T0 cosh L x (4.4) reduces to − Uniform cross section For this case a = p = 1.32) (4.27) d2 θ + m2 θ + ǫ (θ + β)4 − β 4 = 0 dξ 2 (4.28) d2 θ + m2 θ = 0 dξ 2 (4.30) = = 1 0 (4.4.3. For example. we have − the solution to whiich is θ = C1 sinh mξ + C2 cosh mξ the constants are determined from the boundary conditions.29) (4. if θ(0) dθ (1) dξ we get θ = − tanh m sinh mξ + cosh mξ Example 4.35) d dξ a dθ dξ + m2 pθ + ǫp (θ + β)4 − β 4 = 0 (4.36) .3 Fin structure Consider now Fig.31) (4. 4.

48) (4.49) . Fin with convection and radiation 42 Radiative The fin equation is − Let φ=θ+β so that − d2 φ + ǫφ4 = −ǫβ 4 dξ 2 (4.38) (4. dφ1 (1) = 0 dξ (4. (4.44) φ0 (0) = 1 + β.. dξ 2 which gives φ0 = 1 + β To the next order dφ 1 = φ4 − β 4 . .40) (4.37) As an example. The lowest order equation is dφ 0 = 0.45) with the solution φ1 = [(1 + β)4 − β 4 ] The complete solution is ξ2 − [(1 + β)4 − β 4 ]ξ 2 ξ2 − [(1 + β)4 − β 4 ]ξ 2 (4.42) = = 1+β 0 (4.. 0 dξ 2 φ1 (0) = 0. dφ0 (1) = 0 dξ (4.43) (4. we will find a perturbation solution with the boundary conditions φ(0) dφ (1) dξ We write φ = φ0 + ǫφ1 + ǫ2 φ2 + . .4.. (4.46) φ = (1 + β) + ǫ [(1 + β)4 − β 4 ] so that θ = 1 + ǫ [(1 + β)4 − β 4 ] Convective and radiative + .47) ξ2 − [(1 + β)4 − β 4 ]ξ 2 + .39) d2 θ + ǫ (θ + β)4 − β 4 = 0 dξ 2 (4.41) (4.4..

5.54) = Tb = 0 (4.55) − m2 θ = 0 θ(0) = 1 dθ (L) = 0 dξ (4.58) where m2 = Introduce P hL2 Ak0 (Tb − T∞ ) (4. (4.56) (4. Perturbations of one-dimensional conduction 43 4. .4. 4.51) (4.5 4.53) (4. .4.2 Show that under suitable conditions.52) Consider the special case of a linear variation of conductivity k(T ) = k0 1 + ǫ so that (1 + ǫθ) d2 θ +ǫ dξ 2 dθ dξ 2 T − T∞ Tb − T∞ (4.1 [13] Perturbations of one-dimensional conduction Temperature-dependent conductivity The governing equation is d dx k(T ) dT dx − Ph (T − T∞ ) = 0 A (4. the temperature distribution in a two-dimensional rectangle tends to that given by a one-dimensional approximation.59) θ(ξ) = θ0 (ξ) + ǫθ1 (ξ) + ǫθ2 (ξ) + .1 Annular fin Example 4.57) (4.50) with the boundary conditions T (0) dT (L) dx we use the dimensionless variables θ ξ = = T − T∞ Tb − T∞ x L (4.60) .5.

65) (4. (4.3: Eccentric annulus. 4.68) = −m2 (1 − tanh2 m) cosh 2mξ − m2 tanh m sinh 2mξ = 0 = 0 Steady-state conduction in a slightly eccentric annular space.2 Eccentric annulus = −θ0 d2 θ 0 − dξ 2 dθ0 dξ 2 1 (4.69) Solving for r. where 2 r2 = a2 + r2 + 2ar cos ψ. We will use polar coordinates (r. as shown in Fig.70) . The two circles are at r = r1 and r = r. Perturbations of one-dimensional conduction 44 r 2 a Ο2 Ο 1 ^ r T=T 2 ψ r T=T 1 Figure 4. ψ) with the center of the small circle as origin.3 can be solved by regular perturbation [13].5.4.62) (4. we have r= 2 r2 − a2 (1 − cos2 ψ) − a cos ψ.5.64) (4.61) (4. (4. Collect terms of O(ǫ0 ) d2 θ 0 − m2 θ = 0 dξ 2 θ0 (0) = 1 dθ0 (1) = 0 dξ The solution is To O(ǫ ) d2 θ 1 − m2 θ 1 dξ 2 θ1 (0) dθ1 (1) dξ The solution is 4.66) (4.63) θ(ξ) = cosh mξ − tanh m sinh mξ (4. The radii of the two circles are r1 and r2 .67) (4.

(4. ψ) + ǫθ1 (R. .75) (4. ψ) + ǫθ1 (0.86) (4.77) = T1 . = 0. . . (4.76) (4. (4. ψ) = 0.78) and θ(0. = T2 . ψ) + ǫ2 θ2 (R. . ψ) + ǫθ1 (R.81) The perturbation expansion is θ(R. ψ) = 0. ψ) With the variables θ R d ǫ we get = = = = T − T2 .80) (1 + d)2 − ǫ2 (1 − cos2 ψ) − ǫ cos ψ − d.71) ∂ ∂2 ∂2 1 1 + + ∂R2 R + d ∂R (R + d)2 ∂ψ 2 (4. ψ) where R(ψ) = = r − r1 r2 − r1 = = 1. T1 − T2 r − r1 . ψ) = θ0 (R. ψ) θ(R. . 0. we get ∂2 ∂ ∂2 1 1 + + ∂R2 R + d ∂R (R + d)2 ∂ψ 2 θ0 + ǫθ1 + ǫ2 θ2 + . (4. = 1.73) T (r. = 0. r2 − r1 r1 .72) (4. the positive sign corresponding to the geometry shown in the figure has been kept. r2 − r1 a . The governing equation for the temperature is ∂2 1 ∂ 1 ∂2 + + 2 2 ∂r r ∂r r ∂ψ 2 The boundary conditions are T (r1 .74) (4.84) (4. . 2 θ0 (R. ψ) T (r. r2 − r1 θ(R.83) (4. (4. Substituting in the equations. (4.82) R(ψ) = 1 − ǫ cos ψ − ǫ2 (1 − cos2 ψ) + . . . ψ) + . ψ) + ǫ2 θ2 (0. ψ) + ǫ θ2 (R. . .4.85) (4. ψ) + . 2 θ0 (0. .79) (4. Perturbations of one-dimensional conduction 45 In the quadratic solution. .5. . ψ) + .

ψ) θ0 (1. ψ) = 1 − To order O(ǫ1 ) ∂2 ∂ ∂2 1 1 θ1 + + 2 2 ∂ψ 2 ∂R R + d ∂R (R + d) θ1 (0. Transient conduction 46 Using Eq.93) (4.99) (4. ψ) = = = = The solution is θ1 (R.101) x √ 2 κt (4. 1. (1 + h) ln(1 + 1/h) (4. .88) (4.91) R − 2h R cos ψ . we get ∂2 ∂ ∂2 1 1 θ0 + + 2 2 ∂ψ 2 ∂R R + d ∂R (R + d) θ0 (0.90) dθ0 (1. ψ).84).6 Transient conduction Let us propose a similarity solution of the transient conduction equation ∂2T 1 ∂T − =0 ∂x2 κ ∂t as T = A erf Taking derivatives we find ∂T ∂x ∂2T ∂x2 ∂T ∂t x2 1 exp − 4κt πκt x x2 = −A √ exp − 4κt 2 πκ3 t3 x x2 = −A √ exp − 4κt 2 πκt3 = A√ (4. ψ) + ǫ θ1 (1. (1 + 2h) ln(1 + 1/h) R + h (4. 0. dR cos ψ . ψ) which has the solution θ0 (R.94) (4.87) (4.100) (4. ψ) + . ψ) − cos ψ Collecting terms to order O(ǫ0 ).86) can be further expanded in a Taylor series around R = 1 to give θ0 (1. ln(1 + 1/h) = = = 0. = 0.95) ln(1 + R/h) . (4. cos ψ dθ0 (1.6.92) (4. . dR (4.97) (4.4.98) is a solution to equation (4.89) (4. 0.98) so that substitution verifies that equation (4. ψ) = ± 0. (4.96) 4. . (4.97). ψ) θ1 (1.

with the boundary and initial conditions T (0. since Eq.7. The operator is self-adjoint.108) ai (t)φi (x).7 Linear diffusion Let T = T (x. and T (x.110) t . 0) = f (x). t) = i=1 (4.107) 2 iπx sin . The steady state solution is T (x) = T1 + With T ′ (x. Its eigenvalues are λi = − and its orthonormal eigenfunctions are φi (x) = Thus we let θ(x. as t → ∞. T ′ (L.105) but with the conditions: T ′ (0. Following the methodology outlined in Section A.112) The solution shows that T ′ → 0.111) Cj exp −α jπ L 2 t 2 iπx sin . t) and ∂T ∂2T =α 2 (4. t) = i=1 ∞ jπ L aj . t) = 0.103) (4. L L ∞ (4. 0) = f (x) − T . . L (4.102) ∂t ∂x in 0 ≤ x ≤ L. jπ L 2 (4. Thus θ = 0 is a stable solution of the problem. L L (4. (4. we consider the eigenvalue problem d2 φ = λφ dx2 with φ(0) = φ(L) = 0. t) = T2 . t) = T − T we have the same equation ∂T ′ ∂2T ′ =α 2 ∂t ∂x (4. t) = T1 .109) so that daj = −α dt with the solution aj = Cj exp −α Thus T ′ (x.4.104) T2 − T1 x. (4. t) = 0. 2 (4. and T ′ (x.105) was already linear.6. It must be noted that there has been no need to linearize. Linear diffusion 47 4.106) i2 π 2 .1. T (L. L2 (4.

. The initial condition is taken to be T (x. Nonlinear diffusion 48 4. ∂t ∂t1 ∂t2 Assuming an asymptotic expansion of the type T = T0 (x.119) dr = t1 + θ(x. and ǫ ≪ 1.123) (4. ∂t ∂x with −∞ < x < ∞.114) (4. 1 + eλx (4. (4. and a slow. . ∂t1 with the solution T0 1/2 (4. t2 ).5. Applying the initial condition gives g(x) dr θ= . a fast.8 Nonlinear diffusion The following diffusion problem with heat generation is considered in [81] ∂T ∂2T = ǫ 2 + f (T ).115) (4.121) f (r) 1/2 The terms of O(ǫ) are ∂ 2 T0 ∂T0 ∂T1 = f ′ (T0 )T1 + − 2 ∂t1 ∂x ∂t2 Differentiating Eq.122) (4. t1 . long scale t2 = ǫt.8.116) (4.124) . we have a Taylor series expansion f (T ) = f (T0 ) + ǫf ′ (T0 ) + . f (r) (4.117) (4. 0) = g(x) 1 = .120) The lower limit of the integral is simply a convenient value at which g(x) = 0. t2 ) + . . 4. we get ∂T0 = f (T0 ).120 with respect to x gives ∂T0 ∂θ = f (T0 ) . t ≥ 0. .4. t1 .118) (4. short one t1 = t. Thus ∂ ∂ ∂ = +ǫ .113) Consider two time scales. Substituting and collecting terms of O(ǫ0 ). ∂x ∂x so that ∂ 2 T0 ∂x2 ∂T0 ∂θ ∂2θ + f (T0 ) 2 ∂x ∂x ∂x 2 ∂θ ∂@2θ = f ′ (T0 )f (T0 ) + f (T0 ) ∂x ∂x2 = f ′ (T0 ) (4. t2 ) + ǫT1 (x.

4. differentiating with respect to t2 gives ∂T0 ∂θ = f (T0 ) ∂t2 ∂t2 Substituting in Eq.122. t2 ) + t1 ln f (T0 ) f (T0 ) (4. t1 ) 2 ∂x ∂t2 where κ = f ′ (T0 ). 4. Let so that its derivatives are ∂w ∂x ∂2w ∂x2 ∂w ∂t2 = κeκθ ∂θ ∂x ∂θ ∂x 2 ∂θ ∂x 2 = 0.133) ln f (T0 ) f ′ (T0 ) ∂2θ ∂θ − + f ′ (T0 ) 2 ∂x ∂t2 f (T0 ) + T1 f ′ (T0 ). t2 ) = eκθ . 4.136) + κeκθ ∂2θ ∂x2 (4.129) The solution is T1 = A(x.137) (4.138) = κ2 eκθ = κeκθ ∂θ ∂t2 .135) (4. (4. (4. To suppress the secular term in Eq.119 has been used.127) (4.130) which can be checked by differentiation since ∂T1 ∂t1 = ∂θ ∂2θ − + f ′ (T0 ) 2 ∂x ∂t2 + A + t1 = ∂θ ∂2θ − ∂x2 ∂t2 ∂θ ∂x + ∂θ ∂x 2 2 f (T0 ) ∂θ ∂x 2 (4. we take ∂θ ∂2θ − + κ(x.125) .126) ∂T1 ∂θ ∂ ∂2θ = f ′ (T0 )T1 + f (T0 ) − + ln f (T0 ) . where Eq. (4.132) (4. ∂t1 ∂x2 ∂t2 ∂t1 ∂2θ ∂θ − + ∂x2 ∂t2 ∂θ ∂x 2 (4. 4. Nonlinear diffusion 49 Also. ∂T1 ∂θ ∂2θ = f ′ (T0 )T1 + f (T0 ) − + f ′ (T0 ) 2 ∂t1 ∂x ∂t2 Since ∂ ln f (T0 ) ∂t1 we have f ′ (T0 ) ∂T0 f (T0 ) ∂t1 = f ′ (T0 ) = (4.130.134) w(x.131) ∂T0 ∂t1 (4.8.128) ∂θ ∂x 2 (4.

147) is Fisher’s equation: As an example.153) . 4. 4. This can be confirmed by finding the derivatives ∂2w ∂x2 ∂w ∂t2 = = √ 2 R′′ (x + 2r t2 )e−r dr √ 2 r R′ (x + 2r t2 ) √ e−r dr t2 ∞ ∞ (4.146) exp κ 1/2 1 1 dr = t1 + ln √ f (r) κ π ∞ ∞ √ 2 R(x + 2r t2 )e−r dr (4.115. so that the integral in Eq. 0)] g(x) (4. Nonlinear diffusion 50 We find that ∂2w ∂w − 2 ∂x ∂t2 = κeκθ = The solution is 1 w= √ π 1 √ π 1 √ π ∞ ∞ ∞ ∞ ∂θ ∂2θ − +κ 2 ∂x ∂t2 ∂θ ∂x 2 (4.143) (4. R(x) = = so that the final (implicit) solution is T0 1/2 ∞ − 2 √ √ 2 R′′ (x + 2r t2 )2 t2 e−r dr (4.150) Thus w= and from Eq.141) where R(x) = w(x.144) ∞ 1 √ π ∞ ′′ ∞ R (x + 2r t2 )e−r dr √ exp [κθ(x. we get T0 T0 dr = ln r(1 − r) T0 − 1 = = 1 1 + e−(t1 +θ) w w + e−t1 T0 e−t1 1 − T0 (4.149) (4. 0) = e −λx (4.145) dr f (r) (4.152) (4. (4.120 T0 1/2 Substituting in the equation.4. ∞ ∞ √ 2 R(x + 2r t2 )e−r dr. (4.139) (4.151) R(x) = w(x.148) (4. 0).140) 0. we take f (T ) = T (1 − T ). Also.8.142) √ 1 = − √ R′ (x + 2r t2 ) 2 π = and substituting.

160) with T (0. w = exp −λx + λ2 t2 so that T0 = 1 1 + exp [x − t(1 + λ2 ǫ)/λ] (4.7. 4. 4. dt Thus E → 0 as t → ∞ whatever the initial conditions. Also dE dt L = 0 T′ L ∂T ′ dx ∂t ∂2T ′ dx ∂x2 ∂T ∂x ′ L L = α = α = −α so that 0 T′ L T′ 0 L 0 0 2 − dx 0 ∂T ∂x ′ 2 ∂T ′ ∂x dx (4. t) = 0.4. t) = T1 and T (L. The steady state.141 and integrating.159) Let us now re-do the problem for a bar with temperature-dependent conductivity.9. T (x). is governed by d dx k(T ) dT dx = 0.161) .9 4. (4.156) =α 2 ∂t ∂x with T ′ (0.155) This is a wave that travels with a phase speed of (1 + λ2 ǫ)/λ. t) = T2 . t) = 0.2 Nonlinear (4.9. (4. 4.1 Stability by energy method Linear As an example consider the same problem as in Section 4. Define 1 L ′2 E(t) = T dx (4.158)  dE ≤ 0. The deviation from the steady state is governed by ∂T ′ ∂2T ′ (4.9.157) 2 0 so that E ≥ 0. Stability by energy method 51 Substituting in Eq. T ′ (L.154) (4. Thus ∂T ∂ = ∂t ∂x k(T ) ∂T ∂x .

The beginning is at temperature T0 and the ambient is T∞ .. .4 in which each line i (indicated by i = 0.167) ∂θ = θk(θ) ∂x 0 − k(θ) 0 Due to boundary conditions the first term on the right is zero. t) = T (x. Let E(t) = so that E ≥ 0.. 0 (4. t). Thus E → 0 as t → ∞. 1.4.10. Self-similar structures 52 Let the deviation from the steady state be θ(x. The total length of the structure is LT = L0 + 2L1 + 4L2 + 8L3 + .166) dx.10 Self-similar structures Consider the large-scale structure shown in Fig. bipolar. (4. 4. . with θ(0. t) − T (x). The length of each bar is Li = L/β i and its diameter is Di = D/β i .162) (4.) is a conductive bar. = . so that dE/dt ≤ 0. (4. Thus ∂ ∂θ = ∂t ∂x k(θ) ∂θ ∂x . (4. . t) = θ(L.169) Both of these are finite if β < βc = . . Then dE dt L 1 2 L θ2 dx.165) ∂θ ∂x L = dx ∂θ ∂x 2 (4..164) = 0 L θ θ 0 ∂θ dx ∂t ∂ ∂x k(θ) L (4.. t) = 0. (4. . 4 = . 4. . 4. The steady state is θ = 0. . . .11 Non-Cartesian coordinates Toroidal.168) The total volume of the material is VT = π 2 2 2 2 D0 L0 + 2D1 L1 + 4D2 L2 + 8D3 L3 + ..163) where θ = θ(x. .

then it has real eigenvalues and a complete orthonormal set of eigenfunctions φm (ξ). (4.4: Large-scale self-similar structure..170) ∂t with homogeneous boundary and suitable initial conditions. 2 . 4. (4. Fig. valid for Bi ≪ 1.12 Thermal control Partial differential equations (PDEs) are an example of infinite-dimensional systems that are very common in thermal applications [2.5 shows a fin of length L with convection to the surroundings [4]. If A is self-adjoint. with m = 0. which forms a complete spatial basis for X. The system is then described by PDEs that represent a formidable challenge for control analysis. . It is thin and long enough such that the transverse temperature distribution may be neglected.172) ∂t ∂ξ . The state X(ξ.171) The lumped approximation in this chapter. is frequently not good enough for thermal systems. 4. t) is a function of spatial coordinates ξ and time t. Determination of approximate controllability is usually sufficient for practical purposes. φm = 0. The temperature field is governed by ∂2T ∂T = α 2 − ζ(T − T∞ ). It is known [110] that the system is approximately state controllable if and only if all the inner products B. where A is a bounded semi-group operator [2]. and is approximate if it can be taken to a neighborhood of the target [118]. 110]. Thermal control 53 $T_\infty$ 3 3 2 2 3 3 1 T 0 0 1 3 3 2 2 3 3 Figure 4. and the spatial variation of the temperature must be taken into account. and B is a another linear operator. The simplest examples occur when only one spatial dimension is present. Exact controllability exists if the function representing the state can be taken from an initial to a final target state.4. Consider a system governed by ∂X = AX + Bu. (4. .12. 1.

0) = 0. ∂t ∂ξ with the initial and boundary conditions (∂θ/∂ξ)(0. The operators in Eq. and ζ = hP/ρcAc where h is the convective heat transfer coefficient. For simplicity it will be assumed that ζ is independent of ξ.12. (4. P is the perimeter of the cross section. we can arbitrarily assume the fin to be initially at a uniform temperature. B = ζ. ρ is the density.172) becomes ∂θ ∂2θ (4. and c is the specific heat.171) is satisfied for all m. t) = 0. (a) Distributed control: The boundary temperature T (L.175) term is also used in other senses in control theory. A is a self-adjoint operator with the eigenvalues and eigenfunctions βm φm = − = (2m + 1)2 π 2 − ζ. Eq. It can be shown that the same problem can also be analyzed using a finite-difference approximation [5]. (4.170) are A = α∂ 2 /∂ξ 2 − ζ. t) = TL is fixed. the domain [0. and θ(ξ. and u = θ∞ . so the system is indeed state controllable. L] is divided into n equal parts of size ∆ξ. ∂θ ∂2θ = α 2 − ζθ + ζθ∞ (t) ∂t ∂ξ (4.174) becomes dθi = σθi−1 − (2σ + ζ)θi + σθi−1 . Thermal control 54 ξ=0                         E ξ T∞ ξ=L TL Figure 4. where T (ξ. Eq. θ(L.172) becomes. The end ξ = 0 will be assumed to be adiabatic so that (∂T /∂ξ)(0. To enable a finite-difference approximation [5]. dt 1 This (4. so that Eq.5: One-dimensional fin with convection. and θ(ξ.174) = α 2 − ζθ. θ(L.173) with the homogeneous boundary and initial conditions (∂θ/∂ξ)(0. t) = 0. (b) Boundary control: Using the constant outside temperature T∞ as reference and defining θ = T − T∞ . and ξ is the longitudinal coordinate measured from one end. 4L2 (2m + 1)πξ 2 cos . Using it as a reference temperature and defining θ = T − TL . There are two ways in which the temperature distribution on the bar can be controlled: in distributed control1 the surrounding temperature T∞ is the control input and in boundary control it is the temperature of the other end T (L. t) = TL (t) − T∞ . t) = 0. t) is the temperature distribution that represents the state of the system. t is time. Since a linear system that is controllable can be taken from any state to any other. T∞ is the temperature of the surroundings. t) = 0. Inequality (4. L 2L respectively. 0) = 0. The thermal diffusivity is α. Ac is the constant cross-sectional area of the bar.4. (4. (4. t) of the fin. .

174) can be discretized to take the form of Eq. The controllability matrix M is  0 ··· ··· 0  0 ··· 0 σ n−1   . Thus we find   −(2σ + ζ) 2σ 0 ··· 0   . T (0) = T + .180) where ǫ ≪ 1.. . . . . 2 show that the magnitude of the heat rate is independent of the sign of ∆T if we can write k(x. (4. · · · . .. Multiple scales 55 where σ = α/∆ξ 2 . . 2. . where i = 1 is at the left and i = n + 1 at the right end of the fin in Fig. . . and with a step change in temperature at one end. .179) (4.13.5. . σ 0 ··· 0 σ −(2σ + ζ) B = [0.   . .176)  ∈ Rn×n . σ −(2σ + ζ) σ     . n + 1. . T ) = 0. A =  (4. (3. .  .. Let t = t0 + ǫt1 (4. indicating that the state of the system is also boundary controllable. T ) = A(x) λ(T ). . 3  0 0 σ ···   0 σ2 −2σ 2 (2σ + ζ) ··· σ −σ(2σ + ζ) σ 3 + σ(2σ + ζ)2 · · · at the left end the fin is σn ··· . 4. . . 2 ∆T T (L) = T − . where x is the vector of unknown θi . . With this Eq. (4.178) 4. σ]T ∈ Rn . .177) The boundary conditions have been applied to make A non-singular: adiabatic. The nodes are i = 1. .181) Problems 1.   .      ···   ···  ···  (4. and at the right end θn+1 is the control input u. 0     . M = . The rank of M is n. From the governing equation for one-dimensional conduction » – d dT k(x.4. dx dx with boundary conditions ∆T . .13 Solve Multiple scales ∂T1 ∂t ∂T2 ∂t ∂ 2 (T1 − T2 ) ∂x2 2 ∂ (T2 − T1 ) = Rα ∂x2 = α (4.6). .

Assume that the base temperature is known. Neglect convection from the thin sides. 3. and (b) the heat flow at the base of the fin. Graph the results for several values of the parameters.4. radiation and Dirichlet boundary conditions. Calculate numerically the evolution of an initial temperature distribution at different instants of time. T ∞ T b x L Figure 4. where δ = [1 − (x/L)]2 δb . Find (a) the temperature distribution in the fin. δb is the thickness of the fin at the base. Find (c) the optimum base thickness δb .13.6: Longitudinal fin of concave parabolic profile. Consider a rectangular fin with convection. Optimize the fin assuming the fin volume to be constant and maximizing the heat rate at the base. Multiple scales 56 2. . Consider a longitudinal fin of concave parabolic profile as shown in the figure. and (d) the optimum fin height L.

1. 5. We will take a one-dimensional approach and neglect transverse variations in the velocity and temperature.1. 57 . and p is the pressure in the fluid. Since the mass of the element is ρA ds. In addition.5) where p1 and p2 are the pressures at the inlet and outlet respectively.1 Hydrodynamics Mass conservation For a duct of constant cross-sectional area and a fluid of constant density. shown in Fig. 5. we can write the momentum equation as ρA ds from which we get dV = fv + fp dt (5. the mean velocity of the fluid.3) dV τw P 1 ∂p + =− dt ρA ρ ∂s 4τw p1 − p2 dV + = dt ρD ρL (5.1 5. the pressure force.1) (5. we have (5.2 Momentum equation The forces on an element of length ds. for simplicity. we will assume that fluid properties are constant and that the area of the pipe is also constant. is also constant. We can write fv fp = −τw P ds ∂p = −A ds ∂s (5.4) Integrating over the length L of a pipe. V .2) where τw is the magnitude of the wall shear stress. and the hydraulic diameter is defined by D = 4A/P . in the positive s direction are: fv . 5. the viscous force and fp .1.Chapter 5 Forced convection In this chapter we will considering the heat transfer in pipe flows.

8) um 2 (5. Laminar The fully developed laminar velocity profile in a circular duct is given by the Poiseuille flow result ux (r) = um 1 − 4r2 D2 (5.9) 4 = µum D 8µV = D The wall shear stress is linear relationship τw = αV where α= so that T (|V |)V = 8µ D 32µ V ρD2 (5.15) . β = 1/ρL.6) dt where T (V ) = |4τw /ρDV | is always positive. and ∆p = p1 − p2 is the pressure difference that is driving the flow. Hydrodynamics 58 p s p+dp fp f v ds Figure 5. so that we can write dV + T (|V |)V = β ∆p (5. The wall shear stress is estimated below for laminar and turbulent flows. um is the maximum velocity at the centerline.13) (5.11) (5.7) where u is the local velocity.1.12) (5.14) (5. and acts in a direction opposite to V . r is the radial coordinate. and D is the diameter of the duct. For fully developed flow we can assume that τw is a function of V that depends on the mean velocity profile. The mean velocity is given by V = Substituting the velocity profile.1: Forces on an element of fluid. we get V = The shear stress at the wall τw is given by τw = −µ ∂ux ∂r (5.10) r=D/2 4 πD2 D/2 ux (r) 2πr dr 0 (5.5.

21) (5. or similar expressions. which is one-fourth the Darcy-Weisbach value. the Fanning friction factor.7 Re f 1/2 (5.6).1 Consider a long. 2D (5. 5.1. The flow velocity in the steady state is a solution of T (V )V = β ∆p (5.22) (5.0 log f 1/2 e/Dh 2. and may be calculated from the Blasius equation for smooth pipes f= 0. Hydrodynamics 59 Turbulent For turbulent flow the expression for shear stress at the wall of a duct that is usually used is τw = so that T (V ) = f 4 1 2 ρV 2 f |V |V. In the flow in a length of duct. Find the resulting time-dependent flow. Make the assumption that the axial velocity is only a function of radial position and time. thin pipe with pressures p1 and p2 ate either end. without acceleration.3 Long time behavior Consider the flow in a single duct of finite length with a constant driving pressure drop.17) Here f is the Darcy-Weisbach friction factor1 .51 + 3. L. The friction factor is also a fucntion of |V |. The governing equation for the flow velocity is equation (5.5. p1 − p2 is a nonzero constant. p1 − p2 = 0 and there is no flow. is used in the literature.23) 1 Sometimes. the pressure drop is given by ∆p A = τw P L where A is the cross-sectional area. or the Colebrook equation for rough pipes 1 = −2. confusingly. Thus ∆p = f = f L 1 2 ρV 4A/P 2 1 L ρ|V |V D 2 (5. For t ≤ 0.18) where the Reynolds number is Re = |V |D/ν.19) where e is the roughness at the wall.3164 Re1/4 (5. For t > 0.20) Example 5.16) (5. . and P is the inner perimeter.1.

we find that dE dt dV ′ dt = −V ′ T (V + V ′ )(V + V ′ ) − T (V )V = V′ ′ ′ ′2 (5. (5. say nonnegative. equation (5.2. c is the specific heat at constant pressure and k is the coefficient of thermal conductivity.32) Q− = ρAV cT − kA ∂s where the first term on the right is due to the advective and second the conductive transports. we get dV ′ = −T (V + V ′ )(V + V ′ ) + T (V )(V ) dt Defining E= so that E ≥ 0.31) (5.23).24) (5.2 Energy equation Consider a section of a duct shown in Fig.33) Q+ = Q− + ∂s The difference between the two is Q+ − Q− = = ∂Q− ds ∂s ∂2T ∂T − kA 2 ρAV c ∂s ∂s ds (5. Energy equation 60 where ∆p and V are both of the same sign. The heat rate going out is ∂Q− ds (5.27) (5.25) 1 ′2 V 2 (5. and V = V is globally stable to all perturbations. where an elemental control volume is shown. We can show that under certain conditions the steady state is globally stable.34) . The heat rate going in is given by ∂T (5.29) If we assume that T (V ) is a non-decreasing function of V .28) ′ = −V V T (V + V ) − T (V ) − V T (V + V ) (5.30) 5.26) (5.5.6) becomes dV ′ + T (V + V ′ )(V + V ′ ) = β ∆p dt Subtracting equation (5. 5.2. we see that V ′ V T (V + V ′ ) − T (V ) ≥ 0 regardless of the sign of either V ′ or V . E(V ) is a Lyapunov function. so that dE ≤0 dt Thus. Writing V = V + V ′ .

40) ∂θ d2 θ ∂θ + −λ 2 =1 ∂τ ∂ξ dξ λ= k LV ρc (5. θ = θ1 at ξ = 1.34) and (5. Substituting equations (5. .2. is known all along the duct.2: Forces on an element of fluid.38) (5. Energy equation 61 Q Q _ s + Q ds Figure 5.39) (5. Furthermore.42) Boundary conditions may be θ = 0 at ξ = 0. An energy balance for the elemental control volume gives Q− + dQ = Q+ + ρA ds c ∂T ∂t (5.35) where the last term is the rate of accumulation of energy within the control volume.37) The heat rate per unit length.41) where (5. Defining ξ θ τ gives = = = x L (T − Ti )ρV AC Lq tV L (5.5.2. q(s). heat is gained from the side at a rate dQ.35) in (5. which can be written as dQ = q ds where q is the rate of gain of heat per unit length of the duct.36) we get the energy equation ∂T q k ∂2T ∂T +V = + ∂t ∂s ρAc ρc ∂s2 The two different types of heating conditions to consider are: 5.36) (5.1 Known heat rate (5.

3. and an external temperature of T∞ (s). We assume that the flow is one-dimensional. 5.5.49) 5. Single duct T∞ (t) Tin (t) V T (s. and the outlet temperature is Tout (t). Using the same variables to represent non-dimensional quantities. and neglect axial conduction through the fluid and the duct. t) and the ambient temperature is T∞ (t). 5. t) Tout (t) 62 Figure 5. q = P U (T∞ − T ) (5.46) ∂θ ∂θ d2 θ + − λ 2 + Hθ = H ∂τ ∂ξ dξ k LV ρc U ρL ρV Ac (5.44) (5.45) (5. and the fluid velocity is V .2.47) where λ H = = (5.50) ∂τ ∂ξ where the nondimensional variables are s ξ = (5. Thus.43) ξ θ τ gives = = = x L T − Ti T∞ − Ti tV L (5.48) (5.2 Convection with known outside temperature Defining The heating is now convective with a heat transfer coefficient U .53) θ = ∆T .3 Single duct Consider the duct that is schematically shown in Fig.3.51) L tV τ = (5.52) L T − T∞ (5. U is the overall heat transfer coefficient and P the cross-sectional perimeter of the duct. where the local fluid temperature is T (s. the governing non-dimensional equation is ∂θ ∂θ + + Hθ = 0 (5. The duct is subject to heat loss through its surface of the form U P (T − T∞ ) per unit length. The inlet temperature is Tin (t).3: Fluid duct with heat loss.

60) To lowest order. we have (5. zero. 5. Notice that the nondimensional mean ambient temperature is. The ambient temperature is T∞ (t) = T ∞ + T∞ (t) (5.61) (5.e. Single duct 63 The characteristic time is the time taken to traverse the length of the duct.59) We have (5. .3. the residence time. .65) (5.58) λ X= This gives the equation ξ−1 λ (5. i.56) with boundary condition θ(0) = 0 is The boundary layer is near ξ = 1.54) where the time-averaged and fluctuating parts have been separated.5. where we make the transformation dθin d2 θin − − λH(θin − 1) = 0 2 dX dX λ dθin d2 θin − =0 2 dX dX θin = A + BeX (5. by definition. We can use a boundary layer analysis for this singular perturbation problem. . The characteristic temperature difference ∆T will be chosen later.55) (5. so that θin = A + (θ1 − A)eX The matching conditions is so that The composite solution is then θouter (ξ = 1) = θin (X → −∞) A = 1 − e−H θ = 1 − e−H + (θ1 − 1 + e−H )e(ξ−1)/λ + .64) (5.62) (5.1 Steady state No axial conduction The solution of the equation dθ + H(θ − 1) = 0 dξ θ(ξ) = 1 − e−Hξ With small axial conduction d2 θ dθ − − H(θ − 1) = 0 (5.3. and with the boundary conditions θ(0) = 0 and θ(1) = θ1 .66) with the solution The boundary condition θin (X = 0) = θ1 gives θ1 = A + B. The parameter γ = U P L/ρAV c represents the heat loss to the ambient. The outer solution is θout = 1 − e−Hξ (5.57) dξ 2 dξ where λ ≪ 1.63) (5.

3. The following are some special cases of equation (5.68) T (s. 0) = T0 (s) are shown in Fig. t) = ′ t T0 (s − t)e−Ht + He−Ht 0 eHt T∞ (t′ ) dt′ for t < s for t ≥ 1 for t < 1 (5. t) = f (s − t) + γ eHt T∞ (t′ ) dt′ e−Ht 0 ′ (5. t) = Tin (t) and T (s.2 Unsteady dynamics t The general solution of this equation is T (s. Tout (t). The temperature. but with an exponential drop due to heat transfer.5.70) The outlet temperature is the same as the inlet temperature. 5.67) The t < s part of the solution is applicable to the brief. at the outlet section. 5. T∞ . t) = Tin (t) d ‚               t<s   © T (s.69). but at a previous instant in time. s = 1. The solution becomes ′ t Tin (t − s)e−Hs + He−Ht t−s eHt T∞ (t′ ) dt′ for t ≥ s (5.3. Single duct 64 t     t=s t>s T (0. after an initial transient. transient period of time in which the fluid at time t = 0 has still not left the duct.3. . The later t > s part depends on the temperature of the fluid entering at s = 0.71) This is similar to the above. is given by Tout (t) = Tin (t − 1)e−H + He−Ht T0 (1 − t)e−Ht + He−Ht ′ t eHt T∞ (t′ ) dt′ t−1 t Ht′ e T∞ (t′ ) dt′ 0 The boundary conditions T (0. 5.69) It can be observed that.3. 0) = T0 (s) s Figure 5.3 Perfectly insulated duct If H = 0 the outlet temperature simplifies to Tout (t) = Tin (t − 1) T0 (1 − t) for t ≥ 1 for t < 1 (5. the inlet and outlet temperatures are related by a unit delay. 5. and equation (5.4: Solution in s-t space.4. γ.69) becomes Tout (t) = Tin (t − 1)e−H T0 (1 − t)e−Ht for t ≥ 1 for t < 1 (5. and the ambient temperature fluctuation. The outlet temperature is also affected by the heat loss parameter.4 Constant ambient temperature For this T∞ = 0.

72) (5. 5. Single duct 65 Figure 5.76) (5.3.77) (5. There are several complexities that must be considered in practical applications to heating or cooling networks. 5. some of which are analyzed below.78) .5.3.6 Effect of wall The governing equations are ∂T ∂2T ∂T + ρV Ac − kA 2 + hi Pi (T − T∞ ) ∂t ∂x ∂x ∂Tw ∂ 2 Tw ρw Aw cw − kw Aw + hi Pi (Tw − T ) + ho Po (Tw − T∞ ) ∂t ∂x2 ρAc = = 0 0 (5.5: Effect of wall.69) becomes   T in + Tin sin ω(t − 1) e−H   −H −2H Tout (t) = sin(Ωt + φ) +T∞ H −2e Hcos 1+e 2 +Ω2   √  H −Ht T0 (1 − t)e + H 2 +Ω2 T∞ H 2 + Ω2 sin(Ωt + φ′ ) where tan φ = − tan φ′ H(1 − e−H cos 1) + e−H Ω sin 1 Ω(1 − e−H cos 1) − He−H sin 1 Ω = − H for t < 1 for t ≥ 1 (5.75) The outlet temperature has frequencies which come from oscillations in the inlet as well as the ambient temperatures. A properly-designed control system that senses the outlet temperature must take the frequency dependence of its amplitude and phase into account.74) (5.5 Periodic inlet and ambient temperature We take Tin (t) T∞ (t) = T in + Tin sin ωt = T∞ sin Ωt (5.3.73) so that equation (5.

4 Two-fluid configuration Consider the heat balance in Fig.88) (5.90) dT + Hw θ = 0 dξ Hw = w w Hin Hout w + Hw Hin out (5. Neglecting axial conduction.89) (5. we get θw = The governing equation is d2 θ w + Hin (θw − θ) + Hout θw dξ 2 Hin θ Hin + Hout = 0 0 (5.4.84) = = = = x L tV L T − T∞ Ti − T∞ Tw − T∞ Ti − T∞ (5.83) (5.87) (5.5.6. 5.82) In the steady state and with no axial conduction in the fluid dθ + Hin (θ − θw ) = dξ −λw If we assume λw = 0 also.79) (5.85) (5.86) (5. using ξ τ θ θw we get ∂θ ∂θ ∂2θ + − λ 2 + Hin (θ − θw ) = 0 ∂τ ∂ξ ∂ξ 2 ∂ θw ∂θw + Hin (θw − θ) + Hout θw = 0 − λw ∂τ ∂ξ 2 where λw Hin Hout = = = kw ρw V Aw cw L hin Pin L ρw Aw cw V hout Pout L ρw Aw cw V (5. we have . Two-fluid configuration 66 Nondimensionalize.92) 5.91) where (5.80) (5.81) (5.

flat plates at y = −h and y = h.93) (5. laminar flow of an incompressible. where k has been taken to be a constant. The energy equation can be written as k d2 T + µ(T ) dy 2 du dy 2 =0 (5. Newtonian fluid between fixed.5. We non- .98) Due to symmetry du/dy = 0 at y = 0 so that C = 0. ρw Aw cw ∂Tw + h1 (Tw − T1 ) + h2 (Tw − T2 ) ∂t ∂T1 ∂T1 ρ1 A1 c1 + ρ1 V1 c1 + h1 (T1 − Tw ) ∂t ∂x ∂T2 ∂T2 + ρ2 V2 c2 + h2 (T2 − Tw ) ρ2 A2 c2 ∂t ∂x = = = 0 0 0 (5.99) with T = T0 at y = ±h. The plane walls are kept isothermal at temperature T = T0 . Integrating. The second term corresponds to viscous heating or dissipation.95) 5. (5.96). and the viscosity is assumed to be given by Eq.96) with boundary conditions u = 0 at y = ±h. The flow velocity u(y) is in the x-direction due to a constant pressure gradient P < 0.) The momentum equation is then d dy µ(T ) du dy =P (5.6: Two-fluids with wall. Flow between plates with viscous dissipation 67 1 2 Figure 5.94) (5.5. There is also other evidence for this. and the viscosity is assumed to decrease exponentially with temperature according to µ = exp(1/T ). we get µ(T ) du = Py + C dy (5.5 Flow between plates with viscous dissipation Consider the steady.97) (5.

for instance) for the boundary-value problem represented by Eqs.4 −0.9.8 where S is the slope of the temperature gradient on one wall.104) for a = 1.5. (5.2 0. 5.9: Schematic of regenerator. Regenerator 68 6 25 5 20 4 15 S 10 5 3 θ high−temperature solution 2 1 low−temperature solution 0 −1 −0. There are other solutions also but they do not satisfy the boundary conditions on the velocity. As examples.2 0 η 0.101) d2 θ + a η 2 eθ = 0 dη 2 a= βP 2 h4 kµ0 (5.6.7.102) with boundary conditions (5.6 0.104) The boundary conditions are There are two solutions that can be obtained numerically (by the shooting method.6 −0. Figure 5. The bifurcation diagram corresponding to this problem is shown in Fig. 5.7: Two solutions of Eq.105) .8: Bifurcation diagram. 5.4 0. (5.6 Regenerator A regenerator is schematically shown in Fig.104) for a < ac and above which there are none. dimensionalize using θ η The energy equation becomes = β(T − T0 ) y = h (5.102) where (5. θ = 0 for η = ±1 5.102) and (5. Figure 5.8 1 0 0 1 2 3 a 4 5 6 Figure 5.100) (5.8 −0.103) (5. Mc dT + mc(Tin − Tout ) = 0 ˙ dt (5. two numerically obtained solutions for a = 1 are shown in Fig.

5.106) (5. C r dT dr ur qr = (5.10. we get d dT d (rur T ) = k (r ) dr dr dr For the boundary conditions T (r1 ) = T1 and T (r2 ) = T2 .2 Redo the previous problem with a slightly eccentric flow.8 Networks A network consists of a number of ducts that are united at certain points.10: Flow between disks. we must have Ai Vi = 0 (5. 5. the momentum equation is dVi + T (Vi )Vi = β pin − pout + ∆p i i dt (5.110) i where Ai are the areas and Vi the fluid velocities in the ducts coming in. for each duct.111) . the temperature field is T (r) = T1 ln(r2 /r) − T2 ln(r1 /r) ln(r2 /r1 ) (5.7 Radial flow between disks This is shown in Fig. the sum being over all the ducts entering the junction.108) Example 5. Furthermore. With dqr /dr = 0.109) (5.7. At each junction.107) = ur 2πrHT − k2πrH where H is the distance between the disks. 5. Radial flow between disks 69 r 1 r 2 r Figure 5. 5.

The unknowns are the E velocities in the ducts and the V pressures at the junctions. . respectively.8. junctions and circuits. if there happens to be one on that line. and each is connected to all the rest. In a general network. We take the diagonal terms in Tij to be also infinite. so as to have Vii = 0. Also. n (5. . To simplify the analysis the network is considered fully connected. we have E =V +F −2 (5. The momentum equation in the branches produce E independent differential equations.8. respectively.115) where Aij is a symmetric matrix. n (5. The mass conservation equation at junction j for all flows arriving there is n Aij Vij = 0 i=1 for j = 1. The number of equations to be solved is thus quite large if n is large. (a) Two-dimensional networks: A planar or two-dimensional network is one that is topologically equivalent to one on a plane in which every intersection of pipes indicates fluid mixing.114) The network properties are represented by the symmetric matrix βij . Thus the number of (b) Three-dimensional networks: For a three-dimensioanl network. we know that E =V +F −1 (5. assume that there are n junctions. The symmetry of Aij and antisymmetry of Vij gives the equivalent form n Aji Vji = 0 i=1 for j = 1. vertices and faces. pi is the pressure at junction i. We must distinguish between two possible geometries. . For such a graph. . . 5.116) which is simply the mass conservation considering all the flows leaving junction j. . The number of unknowns thus are E + V − 1. except for one pressure that must be known. . so that Vii which has no physical meaning is considered zero.113) If there are n junctions. while mass conservation at the juntions give V − 1 independent algebraic relations. but Tij is infinite for those junctions that are not physically connected so that the flow velocity in the corresponding branch is zero. The number of circuits is then (n2 − 3n + 4)/2.5.112) where E. . . The resistance Tij may or may not be symmetric. they can have a maximum of n(n − 1)/2 lines connecting them. V and F are the number of edges. The flow velocity matrix Vij is anti-symmetric. In the present context. The momentum equation for Vij is dVij + Tij (Vij )Vij = βij (pi − pj ) dt (5.1 Hydrodynamics The global stability of flow in a network can be demonstrated in a manner similar to that in a finite-length duct. and Vij is the flow velocity from junction i to j defined to be positive in that direction. Networks 70 where ∆p is the pressure developed by a pump. these are better referred to as branches.

114)–(5.125) The pressure terms vanish since n n ′ Aij Vij p′ i j=1 i=1 n n ′ Aij Vij p′ i i=1 j=1 n =  (5.124) j=1 i=1 n n j=1 i=1 Aij ′ ′ ′ V Tij (V ij + Vij )(V ij + Vij ) − Tij (V ij )V ij βij ij ′ Aij Vij (p′ − p′ ) i j (5.8.122) Defining E= we get dE dt n n 1 2 n n j=1 i=1 Aij ′ 2 V .123) = = − + j=1 i=1 n n ′ Aij ′ dVij Vij βij dt (5.116). Networks 71 The steady states are solutions of Tij (V ij )V ij n = βij (pi − pj ) n (5.118) Aij V ij = 0 or i=1 i=1 Aji V ji = 0 We write Vij pi ′ = V ij + Vij = p + p′ i (5.117) (5.128) n n = j=1 p′ j i=1 ′ Aij Vij (5.117) and (5.120) Substituting in equations (5.118) we get ′ dVij dt n ′ Aij Vij = 0 or i=1 i=1 = ′ ′ − Tij (V ij + Vij )(V ij + Vij ) − Tij (V ij )V ij + βij (p′ − p′ ) i j n ′ Aji Vji = 0 (5.127) (5.121) (5.129) (5.119) (5.126)  n j=1 = i=1 = and n n ′ Aij Vij p′ j j=1 i=1 0  p′ i ′ Aij Vij  (5.5. βij ij βij > 0 (5. and subtracting equations (5.130) = 0 .

5.133) and (5.135) (5. For this reason the steady state is also unique.131) Ai0 Vi0 = 0 (5.134) ′ Substituting Vi0 = V i0 + Vi0 and p0 = p0 + p′ in equations (5.134). we find that ′ dVi0 dt 3 X i=1 ′ Ai0 Vi0 = = ˜ ˆ ′ ′ − Ti0 (V i0 + Vi0 )(V i0 + Vi0 ) − Ti0 (V i0 )V i0 − βi0 p′ 0 0 (5.138) = − 3 3 X X Ai0 ˜ ˆ ′ ′ ′ ′ Ai0 Vi0 Vi0 Ti0 (V i0 + Vi0 )(V i0 + Vi0 ) − Ti0 (V i0 )V i0 − p′ 0 βi0 i=1 i=1 (5. the steady state is globally stable.132) and subtracting equations 0 (5.132) In the steady state 3 X i=1 Ti0 (V i0 )V i0 Ai0 V i0 = = βi0 (pi − p0 ) 0 (5.11 is globally stable. Since E ≥ 0 and dE/dt ≤ 0. equation (5. For branches i = 1.110) at the junction gives 3 X i=1 (5. Networks 72 p u 20 2 u p 1 10 p 0 u 30 p 3 Figure 5. V20 and V30 are the unknowns.133) (5.30) and satisfy the same inequality.139) .3 Show that the flow in the the star network shown in Fig.11: Star network. 2.131) and (5.6) is dVi0 + Ti0 (Vi0 )Vi0 = βi0 (pi − p0 ) dt Equation (5. The pressures p1 . p2 and p3 are known while the pressure p0 and velocities V10 .137) (5. Example 5.125) are similar to those in equation (5.8. 3.136) If we define E= we find that dE dt = 3 ′ X Ai0 dVi0 V′ βi0 i0 dt i=1 3 1 X Ai0 ′ 2 V i 2 i=1 βi0 (5. The terms that are left in equation (5. 5.

142) j (ma + |ma |)Tia + qi ij ij where T e is the exterior temperature.9. Thus the steady solution is not unique unless one of the room temperatures is known.1 5. (i) Steady state with U = 0 (a) The equality (ma + |ma |)Tja − (ma + |ma |)Tia = 0 ji ji ij ij i j i j (5. the sum of equations (5.144) can be shown by interchanging i and j in the second term.8. By definition mij = −mji .9 5.9.140) Since E ≥ 0 and dE/dt ≤ 0.5.145) i which is a necessary condition for a steady state.141) and (5.136). 5. we know that ma = 0 ji j (5.141) (ma + |ma |)Tja ji ji (5.2 Thermal control Control with heat transfer coefficient Multiple room temperatures Let there be n interconnected rooms.142) are the 2n temperatures Tiw and Tia . Also.2 [61] Thermal networks 5.142) for all rooms gives an identity. the set of equations is not linearly independent.143) Analysis The unknowns in equations (5. Thermal control 73 The last term vanishes because of equation (5. mij is an anti-symmetric matrix.9. Using this result. Thus 3 3 X Ai0 ˆ ˜ X Ai0 ′ 2 dE ′ ′ ′ Vi0 V i0 Ti0 (V i0 + Vi0 ) − Ti0 (V i0 ) − V i0 Ti0 (V i0 + Vi0 ) =− dt β β i=1 i0 i=1 i0 (5. The wall temperature of room i is Tiw and the air temperature is Tia . E is a Lyapunov function and the steady state is globally stable.141) and (5. (b) Because the sum of equations (5.142) for all rooms gives qi = 0 (5. . Since mii has no meaning and can be arbitrarily taken to be zero. from mass conservation for a single room. mij is the mass flow rate of air from room i to room j.141) and (5. The heat balance equation for this room is Mia cw dTiw dt a a a dTi Mi c dt = hi Ai (Tia − Tiw ) + Ui Ae (T e − Tiw ) i 1 = hi Ai (Tiw − Tia ) + ca 2 1 − ca 2 j (5.

Tiset ) ji (5.149).3 Two rooms Consider two interconnected rooms 1 and 2 with mass flow m from 1 to 2.148) and (5. 128] and buildings [8. and leakage out of room 2 to the exterior at the same rate.4 Temperature in long duct The diffusion problem of the previous section does not have advection. Transport of fluids in ducts introduces a delay between the instant the particles of fluid go into the duct and when they come out. The fluid inlet temperature Tin is kept .150) One example of a control problem would be to change m to keep the temperatures of the two rooms equal. Tia .146) (5.9.147) Similar on-off control schemes can also be proposed. The literature includes applications to hot-water systems [43. Thermal control 74 Control The various proportional control schemes possible are: • Control of individual room heating qi = −Ki (Tia − Tiset ) • Control of mass flow rates ma = fij (Tja .149) The overall mass balance can be given by the sum of the two equations to give M1 ca dT1 dT2 + M2 ca dt dt = U1 A1 (T e − T1 ) + U2 A2 (T e − T2 ) + |m|T e 1 1 + (m − |m|)T1 − (m + |m|)T2 2 2 +q1 + q2 (5. respectively. 5. transport [197] and heater [40. which creates a difficulty for outlet temperature control. 5.9. schematically shown in Fig. Delay can be introduced by writing T2 = T2 (t − τ ) and T1 = T1 (t − τ ) in the second to last terms of equations (5. 7]. 5. Also there is leakage of air into room 1 from the exterior at rate m.148) M2 ca dT2 dt (5. A long duct of constant cross section.5. illustrates the basic issues [4. where τ is the time taken for the fluid to get from one room to the other.9.12 where the flow is driven by a variable-speed pump. 41] delay and the effect of the length of a duct on delay [46] have also been looked at. The energy balances for the two rooms give M1 ca dT1 dt 1 = U1 A1 (T e − T1 ) + (m + |m|)(T e − T1 ) 2 1 − (m − |m|)(T2 − T1 ) + q1 2 1 = U2 A2 (T e − T2 ) − (m − |m|)(T e − T2 ) 2 1 + (m + |m|)(T1 − T2 ) + q2 2 (5. 159].

h is the coefficient of heat transfer to the exterior. Thermal control 75 ξ=0 Tin E ξ flow =⇒ T∞ ξ=L Tout (t) velocity = v(t) temperature = T (ξ. ρ is the fluid density. Applying the boundary condition at ξ = 0 gives t 1 = e−t f − v(s) ds . i. so that the ξ = 0 end is always the inlet and ξ = L the outlet. and D is the hydraulic diameter of the duct. . t) is the fluid temperature.155) Eqs. ds 0 (5. ξ is the distance along the duct measured from the entrance.151) is ∂θ ∂θ +v + θ = 0. is t θout (t) = e−t f 1− v(s). t) = e−t f ξ− v(s) ds . The other variables are now non-dimensional.13 shows a typical result using PID control in which the system is unstable. v(t) is the flow velocity. With a one-dimensional approximation. t) = 1. the non-dimensional version of Eq. c is its specific heat. with θ(0. ρcD/4h for time.e. and hL/ρcD for velocity. Fig. where L is the length of the duct. flow velocity and residence time of the fluid in the duct. at ξ = 1. this can be solved to give t θ(ξ. 174].154) The temperature at the outlet of the duct. energy conservation gives ∂T ∂T 4h +v + (T − T∞ ) = 0. The flow velocity is taken to be always positive. as it does in other applications [16. (5. constant. f is an arbitrary function.151) with the boundary condition T (0. all of which ultimately achieve constant amplitude oscillations. Knowing v(t). 69. The problem is non-linear if the outlet temperature Tout (t) is used to control the flow velocity v(t).155) must be simultaneously solved to get the outlet temperature θout (t) in terms of the flow velocity v. ∂t ∂ξ ρcD (5. Shown are the outlet temperature.5. t is time. ∂t ∂ξ (5. The delay between the velocity change and its effect on the outlet temperature can often lead instability. (5.152) where θ = (T − T∞ )/(Tin − T∞ ). 5. The temperature of the fluid coming out of the duct is Tout (t). 74. t) = Tin . t) Figure 5.154) and (5. Using the characteristic quantities of L for length. 0 (5. 0 (5. and there is heat loss to the constant ambient temperature T∞ through the surface of the duct. where T (ξ.153) where the initial startup interval in which the fluid within the duct is flushed out has been ignored.12: Schematic of duct.9.

The fluid inlet and outlet temperatures given.A.5 0. where gβ(Ts − T∞ )D3 . . Churchill’s correlation for natural convection from a sphere is 0. No. 2003. u(r).5. A sphere. and that the material properties are all constant. and find a two-term perturbation solution. Hooman and A. u′′ + r −1 u′ − s2 u + s2 Da = 0. Problems 2. 7. 30. 1 + (0. Ranjbar-Kani.4 1. Vol. να Assume that the temperature within the sphere T (t) is uniform. Thermal control 76 0.4 Tout 0.2 1 0.2 1. 1015–1026.13: Outlet temperature. Derive the governing equation. Consider one-dimensional steady-state flow along a pipe with advection and conduction in the fluid and lateral convection from the side. initially at temperature Ti is being cooled by natural convection to fluid at T∞ . Forced convection in a fluid-saturated porous-medium tube with isoflux wall.589 RaD Nu = 2 + h i4/9 .5 [4]. where s and the Darcy number Da are parameters. 4.3 0. A WKB solution for small Da has been reported as2 " # e−s(1−r) u = Da 1 − . Use the nondimensional version of the governing equation to find the inner and outer matched temperature distributions if the fluid thermal conductivity is small. Determine the single duct solutions for heat loss by radiation q = P ǫσ(Tsur − T 4 ). √ r 3.8 0. velocity and residence time for Ki = −5 and Kp = 2. International Communcications in Heat and Mass Transfer. for forced convection in a cylindrical porous medium is given by Re-do to check the analysis.5 1.469/Pr )9/16 RaD = 1/4 4 1. 2 K. pp.9. The velocity field.6 0 50 100 150 τ 0 50 100 150 t Figure 5.5 0 50 100 150 v 1 0.

1. as shown in Fig. Show that no solution is possible in Problem 4 if the boundary layer is assumed to be on the wrong side. There is conduction along the fluid as well as along the wall of the tube. Consider one-dimensional unsteady flow in a tube with a non-negligible wall thickness. 7. 6. Thermal control 77 5. 111111111111111111111 000000000000000000000 111111111111111111111 000000000000000000000 111111111111111111111 000000000000000000000 111111111111111111111 000000000000000000000 111111111111111111111 000000000000000000000 Figure 5. 17. There is also convection from the outer surface of the tube to the environment as well as from its inner surface to the fluid. Find the governing equations and their boundary conditions.5.9. Nondimensionalize. .14: Flow in tube with non-negligible wall thickness. Plot the exact analytical and the approximate boundary layer solutions for Problem 4 for a small value of the conduction parameter.

1 Mass conservation Consider an elemental control volume as shown in Fig. 78 . so that the mass flow rate into and out of the control volume must be the same. 6. We will also approximate the behavior of the fluid using one spatial dimensions. there is no accumulation of mass within an elemental control volume. In general both u and T are functions of space s and time t.Chapter 6 Natural convection 6. Thus. The loop is heated in some parts and cooled in others. shown in Fig. though we will find that u = u(t).e. m− = m+ .1: A general natural convective loop. The spatial coordinate is s. of length L and constant cross-sectional area A filled with a fluid.2) For a fluid of constant density. measured from some arbitrary origin and going around the loop in the counterclockwise direction.1. 6. we will assume that the velocity u and temperature T are constant across a section of the loop. For s g Figure 6. The temperature differences within the fluid leads to a change in density and hence a buoyancy force that creates a natural circulation.2. 6. The mass fluxes in and out are m− m+ = ρ0 uA = m− + ∂m ds ∂s − (6.1) (6. i. We will make the Boussinesq approximation by which the fluid density is constant except in the buoyancy term.1 Modeling Let us consider a closed loop.1.

3. shown in Fig.2: Mass flows in an elemental control volume. 6. The integral around a closed loop should vanish. 6. g is its component in the negative s ˜ direction.e. we will assume a linear relationship between the wall shear stress and the mean fluid velocity. with z being measured upwards. however. so that L g (s) ds = 0 ˜ 0 (6.2 Momentum equation The forces on an element of length ds. this would be 8µ α= (6. fp . It is impossible to determine the viscous force fv through a one-dimensional model.8) where g is the usual acceleration in the vertical direction. We can write fv fp fg = −τw P ds ∂p = −A ds ∂s = −ρA ds g ˜ (6. this implies that u is the same into and out of the control volume. in the positive s direction are: fv .9) The density in the gravity force term will be taken to decrease linearly with temperature. τw = αu. which is strictly not the case here but gives an order of magnitude value for the coefficient. the viscous force. since it is a velocity profile in the tube that is responsible for the shear streass at the wall. i. and dz is the difference in height at the two ends of the element.7) (6. For simplicity.3) (6. For Poiseuille flow in a duct. and p is the pressure in the fluid.6) D The local component of the acceleration due to gravity has been written in terms of g (s) ˜ = g cos θ dz = g ds (6. so that ρ = ρ0 [1 − β(T − T0 )] (6. and fg . the component of the gravity force.6.5) where τw is the wall shear stress.1. the pressure force. and must be a function of t alone.1. Modeling 79 + m _ m s ds Figure 6.4) (6. Thus u is independent of s.10) . a loop of constant cross-sectional area.

16) Furthermore.14) Q− = ρ0 Aucp T − kA ∂s where the first term on the right is due to the advective and second the conductive transports.13) Fig. cp is the specific heat at constant pressure and k is the coefficient of thermal conductivity. Figure 6. we find that the pressure term disappears. The heat rate going in is given by ∂T (6.6. 6. Since the mass of the element is ρ0 A ds.17) .4: Heat rates on an elemental control volume. t).3: Forces on an element of fluid.1. which can be written as Q = q ds where q is the rate of gain of heat per unit length of the duct. (6. 6.12) Integrating around the loop.3 Energy equation L T g (s) ds ˜ 0 (6.11) du Pα 1 ∂p + u=− − [1 − β(T − T0 )] g ˜ dt ρ0 A ρ0 ∂s (6.1. heat is gained from the side at a rate Q. Modeling 80 p+dp p s θ ds fv fg fp _ Q s Q + Q ds gravity Figure 6.15) Q+ = Q− + ∂s The difference between the two is Q+ − Q− = = ∂Q− ds ∂s ρ0 Aucp ∂T ∂2T − kA 2 ∂s ∂s ds (6.4 shows the heat rates going into and out of an elemental control volume. The heat rate going out is ∂Q− ds (6. and Pα β du + u= dt ρ0 A L where u = u(t) and T = T (s. we can write the momentum equation as ρ0 A ds from which we get du = fv + fp + fg dt (6.

21) (6. and q(s) < 0 cooling.23) where T (0) = T0 . q(s). 165] The simplest heating condition is when the heat rate per unit length. For zero mean heating.2.1 Steady state.18) we get the energy equation ∂T q k ∂2T ∂T +u = + ∂t ∂s ρ0 Acp ρ0 cp ∂s2 (6.22) q(s) ρ0 Acp The solution of equation (6. Using equation (6. the steady-state governing equations are Pα u = ρ0 A u dT ds = β L L T (s)˜(s) ds g 0 (6.24) β P αLcp L 0 s L 0 0 s q(s′ ) ds′ g (s) ds ˜ (6.26) .21).18) where the last term is the rate of accumulation of energy within the control volume.19) 6.9) it can be checked that T (L) = T0 also.20) q(s) > 0 indicates heating. Known heat rate 81 An energy balance for the elemental control volume gives Q− + Q = Q+ + ρ0 A ds cp ∂T ∂t (6.25) q(s′ ) ds′ g (s) ds ≥ 0 ˜ (6.17) in (6. 6.2 Known heat rate [164. Substituting in equation (6. no axial conduction Neglecting axial conduction.16) and (6. Substituting equations (6.22) gives us the temperature field T (s) = 1 ρ0 Acp u s q(s′ ) ds′ + T0 0 (6.2.6. is known all along the loop. we have L q(s) ds = 0 0 (6. we get Pα β u= ρ0 A ρ0 Acp Lu from which u=± Two real solutions exist for 0 L 0 0 s q(s′ ) ds′ g (s) ds ˜ (6.

The integral H in the four cases is: (a) H = 0. and none otherwise.9) and (6. (b) Constant heating between points c and d.30) where p(0) = p0 .5: Bifurcation with respect to parameter H.1 Find the temperature distributions and velocities in the three heating and cooling distributions corresponding to Fig. (c) H = −ˆL/8. and the total length of the loop is L. The constant value is q .31) (6. (d) H = q L/4.2. Example 6. and constant cooling between e and g. and constant cooling between h and a. (b) H = q L/8. and the coordinate s runs counterclockwise.7. and constant cooling between g and h.6.29) from which p(s) = p0 − P αu s − ρ0 A s g (s′ ) ds′ + ˜ 0 β Acp u s 0 0 s′′ q(s′ ) ds′ g (s′′ ) ds′′ ˜ (6.27) The pressure distribution can be found from equation (6.34) u=± P αLcp .33) F (s)g(s) Z L G(s) ds 0 The functions F (s) and G(s) are shown in Fig.12) dp ds = − P αu − ρ0 1 − β(T − T0 ) g ˜ A s β P αu ˜ q(s′ ) ds′ g ˜ − ρ0 g + = − A Acp u 0 (6. 6.28) (6. it can be shown that p(L) = p0 also. and constant cooling between h and a. (c) Constant heating between points d and e. 6. The origin is at point a.32) (6. ˆ Let us write F (s) G(s) H = = = Z s q(s′ ) ds′ 0 (6.6. Known heat rate 82 u H Figure 6.24). where L s H= 0 0 q(s′ ) ds′ g (s) ds ˜ (6. (a) Constant heating between points c and d. Thus there is a bifurcation from no solution to two as the parameter H passes through zero. ˆ q ˆ The fluid velocity is s βH (6. (d) Constant heating between points a and c. Using equations (6.

6.2. Known heat rate

83

g

f

e

h

d

a

b

c

Figure 6.6: Geometry of a square loop.

F

F

s

s

G

G

s (a) (b)

s

F

F

s

s

G

G

s (c) (d)

s

Figure 6.7: Functions F (s) and G(s) for the four cases.

6.2. Known heat rate

84

No real solution exists for case (c); the velocity is zero for (a); the other two cases have two solutions each, one positive and the other negative. The temperature distribution is given by T − T0 = F (s) ρ0 Acp u (6.35)

The function F (s) is shown in Fig. 6.7. There is no real; solution for case (c); for (a), the temperature is unbounded since the fluid is not moving; for the other two cases there are two temperature fields, one the negative of the other. The pressure distribution can be found from equation (6.29).

Example 6.2
What is the physical interpretation of condition (6.26)? Let us write H = = = Z Z
L

0

0

L »Z s

»Z

s

q(s′ ) ds′
0 s

– –
0

g (s) ds ˜ d
s

(6.36) –
0

q(s′ ) ds′
0

»Z

q(s′ ) ds′
0

–L »Z
0

»Z

s

g (s′ ) ds′ ˜
0

(6.37) Z L» q(s)
s

g (s′ ) ds′ ˜

–L
0

Z

g (s′ ) ds′ ˜
0

ds

(6.38)

The first term on the right vanishes due to equations (6.20) and (6.9). Using equation (6.8), we find that Z L H = −g q(s)z(s) ds (6.39)
0

The function z(s) is another way of describing the geometry of the loop. We introduce the notation q(s) = q + (s) − q − (s) where q+ = and q− =  q(s) 0 0 −q(s) for for for for q(s) > 0 q(s) ≤ 0 q(s) ≥ 0 q(s) < 0 (6.40)

(6.41)

Equations (6.20) and (6.39) thus becomes Z L Z q + (s) ds =
0


L

(6.42)

q − (s) ds
0

(6.43) Z – L q − (s)z(s) ds (6.44)

H

=

−g

»Z

L 0

q + (s)z(s) ds −

0

From these, condition (6.26) which is H ≥ 0 can be found to be equivalent to RL − RL + q (s)z(s) ds 0 q (s)z(s) ds < 0R L RL + (s) ds − 0 q 0 q (s) ds

(6.45)

This implies that the height of the centroid of the heating rate distribution should be above that of the cooling.

6.2. Known heat rate

85

6.2.2

Axial conduction effects

To nondimensionalize and normalize equations (6.13) and (6.19), we take t∗ s∗ u∗ T∗ g∗ ˜ q∗ where V ∆T τ G Substituting, we get du∗ + u∗ dt∗ ∂T ∗ ∂T ∗ + G1/2 u∗ ∗ ∂t∗ ∂s where K=
1

= = = = = =

u V G1/2 T − T0 ∆T G1/2 g ˜ g q qm

t τ s L

(6.46) (6.47) (6.48) (6.49) (6.50) (6.51)

= = = =

P αL ρ0 A P 2 α2 L βgρ2 A2 0 ρ0 A Pα qm βgρ2 A2 0 P 3 α3 Lcp

(6.52) (6.53) (6.54) (6.55)

=
0

T ∗ g ∗ ds∗ ˜ ∂2T ∗ ∂s∗2

(6.56) (6.57)

= G1/2 q ∗ + K kA P αL2 cp

(6.58)

The two nondimensional parameters which govern the problem are G and K. Under steady-state conditions, and neglecting axial conduction, the temperature and velocity are T (s) u∗

=

1 u∗

s∗ 0 1

q ∗ (s∗ ) ds∗ 1 1
s∗ 0

(6.59) g ∗ (s∗ ) ds∗ ˜ (6.60)

= ±

0

q ∗ (s∗ ) ds∗ 1 1

All variables are of unit order indicating that the variables have been appropriately normalized.

6.2. Known heat rate

86

For α = 8µ/D, A = πD2 /4, and P = πD, we get G = K = 1 Gr 8192π P r 1 32 P r D L D L
2 4

(6.61) (6.62)

where the Prandtl and Grashof numbers are µcp k qm gβL3 Gr = ν2k respectively. Often the Rayleigh number defined by Pr = Ra = Gr P r (6.63) (6.64)

(6.65)

is used instead of the Grashof number. Since u∗ is of O(1), the dimensional velocity is of order (8νL/D2 )Gr1/2 . The ratio of axial conduction to the advective transport term is ǫ = = K G1/2 8π Ra (6.66)
1/2

(6.67)

Taking typical numerical values for a loop with water to be: ρ = 998 kg/m3 , µ = 1.003 × 10−3 kg/m s, k = 0.6 W/m K, qm = 100 W/m, g = 9.91 m/s2 , β = 0.207 × 10−3 K−1 , D = 0.01 m, L = 1 m, cp = 4.18 × 103 J/kgK, we get the velocity and temperature scales to be V G1/2 ∆T G and the nondimensional numbers as G = K = Gr = Ra = ǫ = 1.86 × 10−2 4.47 × 10−7 (6.70) (6.71) (6.72) (6.73) (6.74)
1/2

= =

(6.68) (6.69)

3.35 × 1011 2.34 × 1012 3.28 × 10−6

Axial conduction is clearly negligible in this context. For a steady state, equations (6.56) and (6.57) are
1

u∗ d2 T dT ǫ ∗2 − u∗ ∗ ds ds
∗ ∗

=
0

T g ∗ ds∗ ˜

(6.75) (6.76)

= −q ∗ (s∗ )

Integrating over the loop from s∗ = 0 to s∗ = 1, we find that continuity of T and equation (6.20) ∗ imply continuity of dT /ds∗ also.

88) . along with conditions that T 0 and dT 0 /ds have the same value at s = 0 and s = 1.6. gives T 0 = an arbitrary constant. Thus s s′′ 1 s′′ T1 = 0 0 q(s ) ds ′ ′ ds + s 0 0 ′′ q(s′ ) ds′ ds′′ + T1 (0) (6. and collecting terms of O(λ0 ).86) A = A respectively.82) = −q(s) + u0 The second equation can be integrated once to give dT 1 =− ds and again s s′′ s q(s′ ) ds′ + A 0 (6. for convenience. Known heat rate 87 Conduction-dominated flow If λ = G1/2 /ǫ ≪ 1. from which 1 s′′ A= 0 0 q(s′ ) ds′ ds′′ (6. the asterisks have been dropped.78) where.79) (6.81) (6. axial conduction dominates. . .84) Continuity of T 1 (s) and dT 1 /ds at s = 0 and s = 1 give 1 s′′ B = − q(s′ ) ds′ 0 o ds′′ + A + B (6. Substituting into the governing equations.2. . We can write u T (s) = u0 + λu1 + λ2 u2 + .87) and that B can be arbitrary. = T 0 (s) + λT 1 (s) + λ2 T 2 (s) + . we have 1 u0 d2 T 0 ds2 = 0 T 0 g ds ˜ 0 (6.77) (6.80) = The second equation. (6. .83) T1 = − q(s′ ) ds′ 0 0 ds′′ + As + B (6. The first equation gives u0 = 0. The terms of O(λ) give 1 u1 d2 T 1 ds2 = 0 T 1 g ds ˜ dT 0 ds (6.85) (6.

94) To O(ǫ0 ). Known heat rate 88 where T (0) is an arbitrary constant. . therefore. we have u T = u0 + ǫu1 + ǫ2 u2 + .81). Advection-dominated flow The governing equations are 1 u = 0 T g ds ˜ d2 T ds2 (6. gives 1 s 0 0 s′′ 1 s′′ 1 u1 = − q(s′ ) ds′ 0 ds′′ g ds + ˜ 0 0 q(s′ ) ds′ ds′′ 0 s˜ ds g (6.3 Toroidal geometry The dimensional gravity function can be expanded in a Fourier series in s.89) The temperature distribution is determined by axial conduction. = T 0 + ǫT 1 + ǫ2 T 2 + . (6. Expanding in terms of ǫ. rather than by the advective velocity.92) (6. so that the resulting solution is unique.2.2.93) (6.97) (6.100) . to give ∞ g (s) = ˜ n=1 c gn cos 2πns 2πns s + g1 sin L L (6. Substituting in equation (6. .6.98) = ± ˜ q(s′ ) ds′ g ds Axial conduction.91) u dT ds = q+ǫ where ǫ ≪ 1.90) (6. we get 1 u0 u0 from which T0 u0 = 1 u0 0 = 0 T 0 g ds ˜ (6. 6.96) dT 0 ds = q s q(s′ ) ds′ 0 1 0 s (6.99) The simplest loop geometry is one for which we have just the terms c g (s) = g1 cos ˜ 2πs 2πs s + g1 sin L L (6.95) (6. slightly modifies the two solutions obtained without it. . .

and take φ0 = 0 so that g (s) = g cos (2πs/L) ˜ (6.108) u The homogeneous solution is dT ds = − sin(2πs − φ) + ǫ T h = Beus/ǫ + A (6.104) The nondimensional gravity component is g = cos(2πs) ˜ where the * has been dropped. from three o’clock point.2.102) Without loss of generality.107) (6.111) (6.106) u = 0 T (s) cos(2πs) ds d2 T ds2 (6.100) becomes g (s) = g cos ˜ 2πs − φ0 L (6.101) (6. Using g2 φ0 equation (6. i.103) = = c s (g1 )2 + (g1 )2 c g tan−1 1 s g1 (6. Known heat rate 89 corresponds to a toroidal geometry. we have dTp u − Tp ds ǫ 1 cos(2πs − φ) 2πǫ 1 = − [cos(2πs) cos φ + sin(2πs) sin φ] 2πǫ = − T p = a cos(2πs) + b sin(2πs) from which dT p = −2πa sin(2πs) + 2πb cos(2πs) ds (6.109) The particular integral satisfies 1 d2 T p u dT p − = sin(2πs − φ) ds2 ǫ ds ǫ Integrating. we can measure the angle from the horizontal.6.114) .105) (6. Assuming also a sinusoidal distribution of heating q(s) = − sin(2πs − φ) the momentum and energy equations are 1 (6.113) (6.e.110) Take (6.112) (6.

117) (6.122) gives an additional spurious solution u = 0.119) Since T (0) = T (1).115) (6. we get u − a + 2πb ǫ u −2πa − b ǫ The constants are a = b The temperature field is given by T = Th + Tp 1 + u2 /ǫ2 u u 1 1 cos φ + sin φ cos(2πs) + − cos φ + sin φ sin(2πs) 2 2πǫ ǫ ǫ 2πǫ2 (u/2πǫ2 ) cos φ + (1/ǫ) sin φ 2(4π 2 + u2 /ǫ2 ) 1 ǫ cos φ − sin φ = 0 4π 2 (6. Taking the other arbitrary constant A to be zero. we have T = 4π 2 (6.2.122) On the other hand substituting ǫ = 0 in equation (6.107) and (6.124) (6.116) The momentum equation gives u= which can be written as u3 + u 4π 2 ǫ2 − Special cases are: • ǫ=0 Equations (6.121) (6. .118) = − cos φ 2πǫ sin φ = − 2πǫ (6. Known heat rate 90 Substituting and collecting the coefficients of cos(2πs) and sin(2πs). • ǫ→∞ • φ=0 We get that u → 0.120) = (u/2πǫ2 ) cos φ + (1/ǫ) sin φ 4π 2 + u2 /ǫ2 −(1/ǫ) cos φ + (u/2πǫ2 ) sin φ 4π 2 + u2 /ǫ2 (6.6.125) The last two solutions exist only when ǫ < (16π 3 )−1/2 .108) can be solved to give T u = 1 [cos(2πs) cos φ + sin(2πs) sin φ] 2πu cos φ = ± 4π (6.123) (6. we must have B = 0. We get u=   0      1 4π − 1 4π − 4π 2 ǫ2 − 4π 2 ǫ2 (6.

there is only one.06 0.6.5 0 -0.1 ε -1 -3 -2 -1 0 φ 1 2 3 -0.01 u 0.5 0 φ=0. Known heat rate 91 1 u ε = 0.126) Figure 6.1 φ=0 -0.11.04 0.04 0.02 0. A cubic equation x3 + px + q = 0 has a discriminant (6. Figure 6.06 0.122) is D= D= The result is shown in Fig. shown in Figure 6.2 1 u 0.01 -0.001 u 0.9 and 6.129) .04 0.5 -0. 1 27 4π 2 ǫ2 − 1 cos φ 4π 3 + 1 2 (ǫ sin φ) 4 (6.8: u-φ curves.2.08 0.1 ε φ Figure 6.1 -1 -3 -2 -1 0 1 2 3 -0.2 u 0.8 shows u-φ curves for three different values of ǫ. The discriminant for the cubic equation (6.2 0.5 ε = 0.1 φ=−0.08 0. there are three real solutions.01 -0.1 ε φ 1 u ε = 0. The velocity is a solution of u3 + u4π 2 ǫ2 − ǫ =0 2 (6.10 show u-ǫ curves for different values of φ. It is also instructive to see the curve u-Ra.02 0. 6. The bifurcation set is the line dividing the regions with only one real solution and that with three real solutions.2 0.08 0. • φ = π/2 Figure 6.06 0.127) p3 q2 + (6.1 0.1 0.02 0.9: u-ǫ curves.1 0.5 0 -0.2 0. and for D > 0.128) 27 4 For D < 0.5 -1 -3 -2 -1 0 1 2 3 0.1 -0. since the Rayleigh number is directly proportional to the strength of the heating.12.2 0.

08 Figure 6.11: u-Ra for φ = 0.02 0.02 u 0.025 0.08 0.08 u 0.6.2 Figure 6.1 20000 30000 40000 Ra -0. .02 0.2. Known heat rate 92 u 0.1 0.06 0.12: Region with three solutions.05 10000 0.04 ε 0.02 0.01 radians.04 0.045 0.08 0.01 0.1 -0.1 0. 0.2 0.1 0.2 0.04 0.2 0.1 0.2 0.04 0.1 -0.06 0. 0.06 0.15 φ=π/4 u 0.035 0.03 0.06 0.005 0 −100 −80 −60 −40 −20 0 20 40 60 80 100 φ (degrees) Figure 6.02 φ=π/4 0.04 φ=π/4 0.015 0.10: More u-ǫ curves.1 -0.

We take g = cos(2πs) and q = − sin(2πs − φ).135) and +u n=1 c s [−nTn sin(2πns) + nTn cos(2πns)] = −G [sin(2πs) cos φ − cos(2πs) sin φ] ∞ −2πn2 G1/2 K Integrating.4 Dynamic analysis We rescale the nondimensional governing equations (6.130) (6. Known heat rate 93 6.137) Multiplying by cos(2πms) and integrating (6. Expanding the temperature in a Fourier series.56) and (6.57) by u∗ T∗ to get du +u dt 1 ∂T ∂T + u ∂t 2π ∂s 1 = = 1 u 2πG1/2 1 T 2πG1/2 (6.2. we have du 1 c + u = T1 dt 2 c dT c dTn dT0 + cos(2πns) + n sin(2πns) dt dt dt n=1 ∞ ∞ (6.139) .6.138) Now multiplying by sin(2πms) and integrating s m c 1 1 dTm s − uTm = − G cos φ − πm2 G1/2 KTm 2 dt 2 2 (6.133) = Gq + where the hats and stars have been dropped. ˜ we get ∞ T (s.131) = 0 T g ds ˜ G1/2 K ∂ 2 T 2π ∂s2 (6.132) (6.134) Substituting.2.136) dT0 =0 dt c m s 1 1 dTm c + uTm = G sin φ − πm2 G1/2 KTm 2 dt 2 2 (6. we get c s [Tn cos(2πns) + Tn sin(2πns)] n=1 (6. t) = T0 (t) + n=1 c s [Tn (t) cos(2πns) + Tn (t) sin(2πns)] (6.

and from equation (6. we get the local form  ′     ′   0 x −1 1 0 x d  ′   y = −z −c −x   y ′  +  −x′ z ′  dt ′ z′ y x −c x′ y ′ z (6. except in different variables.141) (6.6.149) (6.150) (6.146)-(6.144) (6.155) (6. while a and b can have any sign. we get x3 + x(c2 − b) − ac = 0 This corresponds to equation (6.147) (6. so that y−x = 0 = 0 = 0 (6. and z is the vertical temperature difference.149).145) (6. we have y = x.151) a − xz − cy −b + xy − cz From equation (6. The parameter c is positive.151). y is the horizontal temperature difference.2.154) (6.156) .122).140) (6.152) Substituting in equation (6. Known heat rate 94 Choosing the variables x = u 1 c y = T 2 1 1 s T z = 2 1 and the parameters a = b = G sin φ 2 G cos φ 2 2πG1/2 K (6.150). The critical points are found by equating the vector field to zero. we get z = (−b + x2 )/c. y.153) (6.143) (6.148).148) The physical significance of the variables are: x is the fluid velocity. To analyze the stability of a critical point (x.142) c = we get the dynamical system dx dt dy dt dz dt = y−x = a − xz − cy = −b + xy − cz (6. z) we add perturbations of the form x = x + x′ y = y + y′ z = z + z′ (6. Substituting these in equation (6.146) (6.

157) to get  ′   ′   −1 1 0 x x d  ′   y (6.160) The bifurcation diagram is shown in Figure 6.152).6.162) . for a = 0 we get x3 + x(c2 − b) = 0 from which   0 √ b − c2 x=y=  √ − b − c2   −b/c −c z=  −c (6.158) (6. The linearized version is   −1 x′ d  ′   y = −z dt y z′   ′  x 1 0 −c −x   y ′  z′ x −c (6. c = 0) From equation (6. Stability of conductive solution The critical point is (0. we look at the linearized equation (6.161) = b/c −c 0   y ′  dt z′ 0 0 −c z′ The eigenvalues of the matrix are obtained from the equation −(1 + λ) 1 0 b/c −(c + λ) 0 0 0 −(c + λ) which simplifies to (c + λ) (1 + λ)(c + λ) − b =0 c (6.163) =0 (6.2.13: Bifurcation diagram for x.159) The z coordinate is (6. −b/c).13.157) No tilt. with axial conduction (a = 0. 0. To examine its linear stability. Known heat rate 95 x c2 b Figure 6.

169) This is the condition for stability. In fact.166) (6. one can also prove global stability of the conductive solution.174) (6.175) (6.172) b (c + 1)2 − 4(c − ) < 0 c (6. we have dx′ dt dy ′ dt dz ′ dt Let = y ′ − x′ = b ′ x − cy ′ − x′ z ′ c (6.177) (6.167) (6. The other two are solutions of b λ2 + (c + 1)λ + (c − ) = 0 c which are λ2 λ3 = = 1 −(c + 1) − 2 1 −(c + 1) + 2 b (c + 1)2 − 4(c − ) c b (c + 1)2 − 4(c − ) c (6.165) (6. y.173) Thus (6.156) to equation (6. z) = x′2 + y ′2 + z ′2 c 1 dE 2 dt b ′ dx′ dy ′ dz ′ x + y′ + z′ c dt dt dt b ′2 2b ′ ′ ′2 = − x + x y − cy − cz ′2 c c b b ′ = − (x − y ′ )2 − (c − )y ′2 − cz ′2 c c = (6.176) Since E dE dt ≥ 0 ≤ 0 (6.161).2. λ3 is negative as long as −(c + 1) + which gives b < c2 (6. Restoring the nonlinear terms in equation (6.178) .170) (6.6.168) = −cz ′ + x′ y ′ b E(x.171) (6.164) λ2 is also negative and hence stable. Known heat rate 96 One eigenvalue is λ1 = −c Since c ≥ 0 this eigenvalue indicates stability.

157). Its eigenvalues are solutions of −(1 + λ) 1 c −(c + λ) √ √ b − c2 b − c2 This can be expanded to give λ3 + λ2 (1 + 2c) + λ(b + c) + 2(b − c2 ) = 0 (6. Also the determinants D1 D2 D3 = = = 1 + 2c 1 + 2c 2(b − c2 ) 1 b+c 1 + 2c 2(b − c2 ) 0 1 b+c 0 0 1 + 2c 2(b − c2 ) c(1 + 4c) 1 − 2c c(1 + 4c) b> 1 − 2c (6. b.188) (6.157).148) simplifies to dx dt dy dt dz dt = y−x = a − xz = −b + xy (6.190) .180) √0 − b − c2 −(c + λ) =0 (6. From equation (6.181) (6. This requires that b< if if c < 1/2 c > 1/2 (6.6.189) √ √ √ √ √ √ The critical points are ±( b.187) (6.186) With tilt. Known heat rate 97 for 0 ≤ b ≤ c2 . Stability of convective solution √ √ For b > c2 . no axial conduction (a = 0. a/ b). and the critical point is stable to all perturbations in this region.183) should be positive.182) (6. a/ b) will be analyzed. −c) will be considered. b. c = 0) The dynamical system (6. We use the linearized equations (6.146)-(6. b − c2 . which they are. The bifurcation at b = c2 is thus supercritical. the other being similar.184) (6.2. E is a Liapunov function. the solutions of −(1 + λ) 1 0 √ √ −a/ b −λ − b √ √ b b −λ =0 (6.179) The Hurwitz criteria for stability require that all coefficients be positive. The linear stability of the point P + given by ( b.185) (6. only one critical point ( b − c2 .

Using x = b for P + and equation (6.194) √ should also be positive.195) for stability.199) The stability regions for both P + and P − are shown in Figure 6. for the value of G in equation (6.7◦ .192) (6. In fact. Also shown is the √ stability of the critical point P − with coordinates −( b. a/ b).198) 2 cos φ The stability condition (6. P + is stable for the tilt angle range φ > 7. Known heat rate 98 are the eigenvalues.2. ±i 2b.195) thus becomes tan φ < x (6.144). This simplifies to a λ3 + λ2 + λ(b + √ ) + 2b = 0 b (6. the equation can be factorized to give the three eigenvalues −1. we get x2 G = (6. substituting a = b3/2 in √ equation (6.191) For stability the Hurwitz criteria require all coefficients to be positive.144).195) can be written as sin φ > cos3/2 φ G 2 1/2 (6. This gives the condition (b + a/ b) − 2b > 0.195 in the zone 0 < b < c.14.196) As a numerical example. The effect of a small nonzero axial conduction parameter c is to alter the Figure 6. Using equations (6.7◦ . The determinants D1 D2 D3 = 1 = = 1 2b √ 1 b + a/ b 1 2b √ 1 b + a/ b 0 1 0 0 2b (6. the stability condition (6.6. for P + . which they are. The stable and unstable region for P + is√ √ shown in Figure 6. Thus the √ nondimensional radian frequency of the oscillations in the unstable range is approximately 2b.191). The dashed circles are of radius G/2. we have a > b3/2 (6. for G ≪ 1. and the angleof tile φ is also indicated.15. The loss of stability is through imaginary eigenvalues.193) (6. from which. the stability condition for P + can be approximated as 1/2 G (6.70). In fact. and P − is stable for φ < −7.197) φ> 2 √ The same information can be shown in slightly different coordinates.143) and (6. b. .

5 x 1 1.14: Stability of critical points P + and P − in (b. Known heat rate 99 x a 3/2 a=b P+ exists but unstable + P stable G/2 φ Both P + and P unstable b _ P+stable −π −π _ P stable π φ π _ P stable 3/2 a=-b Figure 6.6 2.5 2 1.2 0 0 5 10 15 t 20 25 30 y −0.6.15: Stability of critical points P + and P − in (φ.9.5 0 0. Figure 6.5 0. x) space.2.5 1. _ P exists but unstable Figure 6.16: x-t for a = 0.5 1 0.5 0. z .5 2 0.4 2 1.9. a) space. b = 1. 1.8 0 −0.5 2.5 1 0. b = 1.4 1.8 1.6 Figure 6.2 1 x 0.17: Phase-space trajectory for a = 0.

Known heat rate 100 2.5 2 4 3 2 1 x z 0.5 2 1 0 −1 −1 0 5 10 15 t 20 25 30 y −2 −1 0 −0.55. 2. b = 1. . b = 1.5 Figure 6.19: Phase-space trajectory for a = 0. b = 1.2.5 1 2 2.5 0 1 0 −1 −2 4 3 −0.20: x-t for a = 0.5 1.5 1. Figure 6. Figure 6.5 x 0.5 1 2 2. b = 1.5 1.5 x 0.55.5 2 4 3 2 1 z 0 0.18: x-t for a = 0.5 −1 −2 0 −3 4 3 −0.6.5 2 1 0 −1 −1 0 5 10 15 t 20 25 30 y −2 −1 0 −0.53.53.5 x 1 1.5 Figure 6.21: Phase-space trajectory for a = 0.

22: x-t for a = 0. Figures 6.9 and a = 0.5 Nonlinear analysis Numerical Let us choose b = 1.17 show the x-t and phase space representation for a = 0.23. 6. z ) = ± x ¯ ¯ √ √ a b.24 shows that vestiges of the shape of the closed curves for a = −0.16 and 6. √ b (6. The strange attractor is shown in Figures 6.55. We start with the dynamical system which models a toroidal thermosyphon loop with known heat flux dx dt dy dt dz dt = y−x = a − zx = xy − b (6.18 and 6. y .2. and Figures 6.21 for a = 0.2.22 and 6. b.20 and 6. b = 1.6. Figure 6. b = 1.9.200) For b > 0 two critical points P + and P − appear (¯.201) .23: Phase-space trajectory for a = 0. and reduce a.19 for a = 0. Figures 6. Analytical The following analysis is by W. Known heat rate 101 3 6 2 4 2 0 −2 x −1 −4 −6 5 1 0 −2 z 0 −3 −5 −4 −2 x −1 0 1 2 3 0 50 100 150 t 200 250 300 y −10 −4 −3 Figure 6.9 can be seen in the trajectories in a = 0. Comparison of the three figures in Figures 6. Franco.53.

.2.9 −4 −6 4 2 0 0 −2 y −4 −4 x −2 2 4 Figure 6.6.24: Phase-space trajectories for b = 1.0.9 −4 −6 4 2 0 0 −2 y −4 −4 x −2 2 4 6 4 2 z 0 −2 a = . Known heat rate 102 6 4 2 z 0 −2 a=0 −4 −6 4 2 0 0 −2 y −4 −4 x −2 2 4 6 4 2 z 0 −2 a = 0.

202) √ 3 3 For stability a > b 2 . thus a nonlinear analysis through 3 the center manifold projection is possible. dropping the primes and regarding the perturbation the system becomes dx dt dy dt dz dt = y−x = β2 + ǫ β x − βz (6. the solution x(t) of (6. this is usually less complex than the initial dynamics and can be described by far simple evolution equations. then S is an invariant .206) is in S for | t |< T where T > 0.206) where x ∈ Rn . Known heat rate 103 The local form respect to P + is dx′ dt dy ′ dt dz ′ dt = y ′ − x′ √ a = − √ x′ − bz ′ b √ ′ √ ′ bx + by = (6. Let’s introduce a perturbation of the form a = b 2 + ǫ and the following change of variables α β a √ b √ = b = rewriting the local form.206) if for x0 ⊂ S. In many dimensional systems.2.205) The center manifold projection is convenient to use if the large-time dynamic behavior is of interest.203) = βx − βy for stability α > β 2 .204) (6. If we can always choose T = ∞.± 2bi.6. We first state the definition of an invariant manifold for the equation x = N (x) ˙ (6. A set S ⊂ Rn is a local invariant manifold for (6. Let’s apply the following transformation: √ 2β 2 2 w2 + 2 x = w1 + 2 2β + 1 2β + 1 y = 2w2 √ 2 2 β2 + 1 2β z = −βw1 − 2 w2 + w3 2β + 1 2β 2 + 1 in the new variables −1 w= 0 ˙ 0  0 √0 2β  0 √ ˆ − 2β  w + Pw + l(w) 0 (6. the system often settles into the same large-time dynamics irrespective of the initial condition. At a = b 2 the eigenvalues are −1.

207) and h is smooth.211) Normal form: Now we carry out a smooth nonlinear coordinate transform of the type w = v + ψ(v) (6. Now we calculate.xy and y 2 . ∂w2 ∂w3 w3 ˙ We seek a center manifold 2 2 h = aw2 + bw2 w3 + cw3 + O(3) (6.h′ (0) = 0. y) ˙ (6. h) (6. w3 . w3 ) (6. The system in the new coordinates is   2 2 √ (νv1 − γv2 ) v1 + v2 − 2β  √0 + (6. then it is called a center manifold if h(0) = 0.213) v= ˙ 2β 0 2 2 (νv2 + γv1 ) v1 + v2 .207).208) The last equation contains all the necessary information needed to determine the asymptotic behavior of small solutions of (6.211) by transforming away many nonlinear terms. we obtain   w ˙ ∂h ∂h  2  w1 = ˙ = −h + l1 (w2 . y ∈ Rm and A and B are constant matrices such that all the eigenvalues of A have zero real parts while all the eigenvalues of B have negative real parts.205) and using the chain rule.207) where x ∈ Rn . h(x)) ˙ (6.212) to simplify (6. or at least approximate the center manifold h(w).6. Substituting w1 = h(w2 . Consider the system x = Ax + f (x.2.209) . Known heat rate 104 manifold.209) √  − 2βw3  = (2aw2 + bw3 . w3 ) √ = 2βw2 + s3 (w2 . y) ˙ y = By + g(x. we find that √ a = k1 − b 2β √ c = k3 + b 2β √ k2 + 2 2β (k1 − k3 ) b = 8β 2 + 1 The reduced system is therefore given by w2 ˙ w3 ˙ √ = − 2βw3 + s2 (w2 . If y = h(x) is an invariant manifold for (6. 2cw3 + bw2 )  √ 2βw2  2 2 − aw2 + bw2 w3 + cw3 + 2 2 k1 w2 + k2 w2 w3 + k3 w3 + O(3) Equating powers of x2 . The flow on the center manifold is governed by the n-dimensional system x = Ax + f (x. w3 ) in the first component of (6.210) substituting in (6.

A ˆ of the transformed equation. In polar coordinates we have r ˙ ˙ θ where = µr + νr3 √ 2β = (6. A represents a linear expansion of order µ in the parameters above that point.205).214) where the Jacobian A has been evaluated at a point in the parameter space where all its eigenvalues ˆ are on the imaginary axis.213) as v= ˙ √0 2β √ − 2β 0 v+ p22 p32 p23 p33 v+  f1 (v) f2 (v)   (6.211). to leading order.2.217) √ ǫ 2 µ=− 2 2β (2β 2 + 1) (6. Known heat rate 105 where ν and γ depend on the nonlinear part of (6.215) a1 a3 a2 a4 a1 + a4 ω where the perturbation matrix comes from (6. This is the unfolding of the Hopf bifurcation.214) is transformed to ˆ u = A + AB − BA + A z ˙ For the Hopf bifurcation if ˆ A= then µ= Therefore for ǫ small we can write (6. The eigenvalues are then close to the imaginary axis but not quite on it. The perturbation parameter represents the size of the neighborhood in ˆ the parameter space.6.216) u+ 2β µ 2 2 (νu2 + γu1 ) u1 + u2 which is the unfolding for the perturbed Hopf bifurcation. one can also present a perturbed version. We stipulate the order of µ such that the real part of the eigenvalues of A + A ˆ does not change the coefficients of the leading order nonlinear terms is such that.218) . Although the normal form theory presented in class pertains to a Jacobian whose eigenvalues all lie on the imaginary axis. Consider the system ˆ v = Av + Av + f (v) ˙ (6. a perturbed Jacobian. Applying a near identity transformation of the form v = u + Bu the system becomes   √ (νu1 − γu2 ) u2 + u2 1 2 µ − 2β  √ u= ˙ (6. The linear part A + A of perturbed Hopf can always be transformed to µ −ω ω µ The required transformation is a near identity linear transformation v = u + Bu such that the linear part of (6.

221) 1 (fxxx + fxyy + gxxy + gyyy ) 16 1 (fxy (fxx + fyy ) − gxy (gxx − gyy ) − fxx gxx − fyy gyy ) + 16ω √ in our problem f = f1 . . and an external temperature of Tw (s). we get d e−γs/u T ds Integrating. ν = (6.227) sign of γ appears to be wrong.224) (6.3 Known wall temperature The heating is now convective with a heat transfer coefficient U .219) k1 k2 k3 = − = = 8β β 2 + 1 3 (2β 2 + 1) 8β β 2 + 1 (2β 2 + 1) = − 3 (2β 2 + 1) √ √ β 2 + 1 4 2 − 8 2β 2 3 (6. y = v2 and ω = 2β.3.225) = γ T − Tw (s) where γ = U P/ρ0 Acp 1 . q = P U (T − Tw ) (6.6.223) Neglecting axial conduction Pα u ρ0 A u dT ds = β L L T (s)˜(s) ds g 0 (6.222) 6.220) p22 p23 From the literature ǫ β (2β 2 + 1) √ ǫ 2 = − 2 2β + 1 (6. we get T = eγs/u − 1 The γ = − e−γs/u Tw u e−γs /u Tw (s′ ) ds′ + T0 ′ (6. Multiplying the second equation by e−γs/u /u. x = v1 . g = f2 . Known wall temperature 106 ν=− Appendix 40β 6 + 40β 4 + 12β 3 + 10β 2 + 12β + 3 4 (8β 2 + 1) (2β 2 + 1) 4 (6.226) γ u s 0 (6. Thus.

we get γ eγL/u T0 = u The velocity is obtained from β Pα u= ρ0 A L L 0 L −γs′ /u e Tw (s′ ) 0 eγL/u − 1 ds′ (6. Thus for part of the loop the wall temperature is known so that q = P U (T − Tw (s)). to have one part of the loop heated with a known heat rate and the rest with known wall temperature.233) (6.6.4 Mixed condition The following has been written by A.225) can be written as d(T − Tw ) The solution to this is T − Tw = γ ds u (6. Example 6.3 Show that there is no motion if the wall temperature is uniform. As an example. while q(s) is known for the rest. especially in experiments.230) T = Tw + K eγs/u (6. Then equation (6. Hence T = Tw . u = 0.231) where K is a constant.232) 6. Pacheco-Vega.234) . consider q= where T0 and q0 are constants. It is common. P U (T − T0 ) for q0 for φ ≤ s ≤ π + 2π φ π + 2π < s < 2π + φ 2π φ 2π (6. Continuity of T at s = 0 and s = L gives K = 0.228) eγs/u − γ u s 0 e−γs /u Tw (s′ ) ds′ + T0 ′ g (s) ds ˜ (6. Mixed condition 107 Since T (L) = T (0).224). from equation (6.4. Assume the wall temperature to be Tw (s) = − sin(2πs − φ) The temperature field is T = b r2 − r1 cos(2πs − φ) 2 2π(1 + r2 /4π 2 ) − cos(2πs − φ) 2π(1 + r1 62/4π 2 ) (6. and. Take Tw to be a constant.229) This is a transcendental equation that may have more than one real solution.

25).238) = +  2 ∂t R ∂θ q. 2π d=2r g ~ θ o α Cooling water out R θ=π Cooling water in Uniform heat flux q Figure 6. the momentum equation in the θ-direction can be written as ρπr2 Rdθ dv dp = −πr2 dθ − ρgπr2 Rdθ cos(θ + α) − τw 2πrRdθ dt dθ (6. v = v(t). π < θ < 2π r . (6. with the shear stress at the wall being approximated by that corresponding to Poiseuille flow in a straight pipe τw = 8µv/ρw r2 .4.236) around the loop using the Boussinesq approximation ρ = ρw [1 − β(T − Tw )]. the expression of the balance in Eq. 0 ≤ θ ≤ π ∂T v ∂T ρw cp (6.236) Integrating Eq.6.1 [1] Modeling If we consider a one-dimensional incompressible flow.25: Schematic of a convection loop heated with constant heat flux in one half and cooled at constant temperature in the other half. Mixed condition 108 Constant wall temperature Tw θ=0. 6.235) Taking an infinitesimal cylindrical control volume of fluid in the loop πr2 dθ. see Figure (6. the equation of continuity indicates that the velocity v is a function of time alone. (6. the temperature of the fluid satisfies the following energy balance equation  2h  − r (T − Tw ).237) Neglecting axial heat conduction. (6.4.236) modifies to 8µ βg dv + v= dt ρw r2 2π 2π 0 (T − Tw ) cos(θ + α) dθ (6. Thus.

the nondimensional time.240) φ cos(θ + α) dθ 0 (6. (6.239) 2πR/V V q/h respectively. Eq. 0≤θ≤π π < θ < 2π (6.4. (1979). 2D. (6.245) The resulting temperature filed is D e−(Dθ/πw) w 1−e−(D/w) . These are A= D 1 w 1 − e(−D/w)      B= D 2 e(−D/w) − 1 w 1 − e(−D/w) (6.4. (6.243) The steady-state governing equations without axial conduction are w= and π 4D 2π φ cos(θ + α) dθ 0 (6. Mixed condition 109 Following the notation used by Greif et al. (6.246) φ(θ) =  D θ + B. φ= (6. Eqs. where V = Accordingly.237) and (6. such that φ(0) = φ(2π) and φ(π − ) = φ(π + ) the constants A and B can be determined.247) where w and φ are the steady-state values of velocity and temperature respectively. D w θ π φ(θ) = 0≤θ≤π .245) can be integrated to give   A e−(Dθ/πw) .6. 0 ≤ θ ≤ π (6. w= .241) −2Dφ.238) become dw πΓ + Γw = dτ 4D and ∂φ ∂φ + 2πw = ∂τ ∂θ 2π gβRrq 2πcp µ 1/2 .248) + 2e−(D/w) −1 1−e−(D/w) . π < θ < 2π πw  − D φ.244) Applying the condition of continuity in the temperature. velocity and temperature are defined as t v T − Tw τ= . 0 ≤ θ ≤ π dφ  πw =  D dθ π < θ < 2π πw .242) where the parameters D and Γ are defined by D= 6.2 Steady State 2πRh ρw cp rV Γ= 16πµR ρw r2 V (6. π < θ < 2π (6.

two and three solutions can be identified. multiplying the numerator and denominator by e(D/2w) and rearranging terms leads to the expresion for the function of the steady-state velocity G(w.5◦ < α < α0 .252) The steady-state solutions of the velocity field and temperature are shown next. the two branches are not symmetric while for α = 90◦ and α = 135◦ . It can be seen the increase in the temperature as α takes values of α = 0◦ . Figure 6.5◦ where α0 varies from 90◦ at a value of D = 0. Three velocities are obtained for −α0 < α < −32.26 shows the w − α curves for different values of the parameter D. D) is an even function of w.251) For α = 0.001 to α0 = 32.5◦ and 147.0 the temperature decays exponentially and rises linearly.250) D θ 2e−(D/w) − 1 + w π 1 − e−(D/w) D e−(Dθ/πw) sin θ dθ w 1 − e−(D/w) sin θ dθ (6. For α = 45◦ . As D increases the variation in temperature between two opposit points also increases. The temperature distribution in the loop. symmetric steady-state solutions for the fluid velocity are possible since G(w. The regions of no possible steady-state velocity are: −180◦ < α < −147. Similar but more drastic change in temperature is seen when D = 2. There is only one velocity for the ranges −147.5◦ < α < −α0 and α0 < α < 147. except for the zero-inclination case which has two possible steady-state velocities. The positive and negative values of the velocity are equal in magnitude for any value of D. From the φ − θ curves it can be seen the dependence of the temperature with D. leads to w= + π π cos α 4D 2π π 0 − + π sin α 4D 2π π D θ 2e−(D/w) − 1 + w π 1 − e−(D/w) π 0 D e−(Dθ/πw) cos θ dθ w 1 − e−(D/w) cos θ dθ and integration around the loop. α. followed by an expansion of cos(θ + α). When has a value D = 0.5◦ when D = 100. D) = w2 − cos α (D/w) cos α + π(D/w)2 sin α coth(D/2 w) = 0 − D 2 2 4 1 + πw (6.6.251) reduces to w2 = (D/w) 1 + e−(D/w) 1 . one. .5◦ < α < 180◦ . Figure 6.245) in Eq.5. for three values of the parameter D and α = 0 is presented in Figure 6. while at a value of D = 1. For α = 0 we have two branches of the velocity-curve which are symmetric.27. In this case Eq. Regions of zero.249) As a final step.29 shows the steady-state velocity as a function of D for different angles of inclination α.1 the heating and cooling curves are almost straight lines. Mixed condition 110 Substituion of Eq. (6.28 shows the φ − θ curves for three different inclination angles with D = 2. only the positive branch exist.(6.244). + 2 4 1 + D 2 1 − e−(D/w) πw (6.5◦ and 32. This behaviour is somewhat expected since the steady-state velocity is decreasing in value such that the fluid stays longer in both parts of the loop. α = 90◦ and α = 135◦ . (6.4. (6. 0. gives the steady-state velocity as w2 = cos α (D/w) cos α + π(D/w)2 sin α + D 2 2 4 1 + πw 1 + e−(D/w) 1 − e−(D/w) .5. Figure 6.

.26: Steady-State velocity field.5 D=1.4 −0.5° 32.2 −0.8 α=0 ° α=90 ° α=45 ° 5 0.1 D=1 D=2.4 α=135 ° φ(θ.2 −0.4 −0.5.5° 1. Figure 6.4 0.6 −0.5 2.4.6 0.8 0.8 −1 −150 −100 −50 0 α 50 100 w -147.5° -32.8 −1 α=0 ° 0 0 1 2 3 θ 4 5 6 7 10 −1 10 0 D 10 1 10 2 Figure 6.28: Nondimensional temperature distribution as a function of thermosyphon inclination α for D = 2.D=2.5 2 φ(θ.2 0 −0.1 1 0.27: Nondimensional temperature distribution as a function of the parameter D for α = 0.6 0.6.5.5 150 0 0 1 2 3 θ 4 5 6 Figure 6.D.α) 4 0. Figure 6.5° D=0.0 D=0.6 α=45 ° 2 1 −0.5 D=2. Mixed condition 111 3 1 D=10 0.2 3 α= 0 ° w α= 90° 0 −0. 6 α= 135 ° 1 0.α=0) 147.29: Velocity w as a function of D for different thermosyphon inclinations α.

such that ∞ φ = φ0 + n=1 [φc (t) cos(nθ) + φs (t) sin(nθ)] n n (6.256) Multiplying by cos (mθ) and integrating from θ = 0 to θ = 2π D dφc m + 2πm w φs = −Dφc + m m dτ π ∞ n=1 n=m φs (−1)m+n − 1 n 2n n2 − m2 (6. we have dw π2 Γ π2 Γ + Γw = cos αφc − sin αφs 1 1 dτ 4D 4D and dφc dφs dφc n 0 + cos (nθ) + n sin (nθ) dτ dτ dτ n=1 ∞ ∞ (6.261) (6.260) (6.4.255) from θ = 0 to θ = 2π we get dφc 1 0 = −D φc − 0 dτ π φs n [(−1)n − 1] − 1 n n=1 ∞ (6. n n = (6. Mixed condition 112 6.3 Dynamic Analysis The temperature can be expanded in Fourier series.253) Substituiting into Eqs.254) +2πw   −2D {φc + 0  2D. (6.257) Now multiplying by sin (mθ) and integrating from θ = 0 to θ = 2π D dφs m − 2πm w φc = −Dφs + m m dτ π ∞ n=0 n=m φc (−1)m+n − 1 n − 2m m2 − n2 (6.255) Integrating Eq. (6.258) 2D [1 − (−1)m ] πm for m ≥ 1.259) (6. Choosing the variables w C0 Cm Sm = w = φc 0 = φc m = φs m (6.242). n=1 ∞ n=1 [−nφc sin (nθ) + nφs cos (nθ)] n n 0≤θ≤π π < θ < 2π [φc (t) cos(nθ) + φs (t) sin (nθ)]} .4.6.241) and (6.262) .

267) (6. so that π2 π2 cos α C 1 + sin α S 1 = 0 4D 4D ∞ Sn 1 [(−1)n − 1] = 0 (C 0 − 1) − π n=1 n w− 2πm w S m + D C m − D π ∞ (6. D and Γ are positive. and S is the vertical temperature difference. The physical significance of the variables are: w is the fluid velocity. Γ and α. The Fourier series expansion is ∞ φ= n=0 C n cos(nθ) + S n sin(nθ) (6. a convenient alternative way to determine the critical points is by using a Fourier series expansion of the steady-state temperature field solution given in Eq. Mixed condition 113 we get an infinte-dimensional dynamical system dw dτ dC0 dτ dCm dτ dSm dτ = −Γw + π2 Γ π2 Γ cos α C1 − sin α S1 4D 4D ∞ D Sn = −D C0 + [(−1)n − 1] + D π n=1 n D π ∞ (6.265) = −2πm w Sm − D Cm + = 2πm w Cm − D Sm + + 2D [(−1)m − 1] πm n=1 n=m Sn (−1)m+n − 1 D π ∞ n=0 n=m Cn (−1)m+n − 1 for m ≥ 1.272) = n=0 Cn 0 cos(nθ) cos(mθ) dθ + n=0 Sn 0 sin(nθ) cos(mθ)dθ .266) (6. C is the horizontal temperature difference.269) n=1 n=m S n (−1)m+n − 1 C n (−1)m+n − 1 + 2n n2 − m2 2m m2 − n2 = 0 D 2πm w C m − D S m + π ∞ n=0 n=m 2D [(−1)m − 1] = πm 0 (6. while α can have any sign.6.263) (6.248).264) 2n n2 − m2 2m m2 − n2 (6. (6.271) Performing the inner product between Eq. The critical points are found by equating the vector filed to zero. The parameters of the system are D. (6.270) However.4.248) and cos(mθ) we have 1 D w 1 − e−(D/w) 2π π π e−(Dθ/πw) cos(mθ) dθ 0 D + w ∞ 2π θ 2e−(D/w) − 1 + π 1 − e−(D/w) ∞ 2π cos(mθ) dθ (6.268) (6.

278) (6.284) Cm S1 ′ C m + Cm ′ S 1 + S1 Sm Sm + ′ Sm . C m . = ′ C 1 + C1 D 1 − e−(D/w) cos(mπ) + 2 2 [1 − cos(mπ)] π m w 1 − e−(D/w)    D 3 1 − e−(D/w) cos(mπ)  πw = −  m m2 + D 2  1 − e−(D/w) πw D 2 πw D 2 m2 + πw (6. · · · .282) (6. Mixed condition 114 Now the inner product between Eq.4.273) = n=0 Cn 0 cos(nθ) sin(mθ) dθ + n=0 Sn 0 sin(nθ) sin(mθ)dθ from which we get C0 Cm = = 1 D 1+ 2 w 3 2e−(D/w) − 1 + 2 1 − e−(D/w) (6.281) (6.276) (6.274) (6.248) and sin(mθ) gives 1 D w 1 − e−(D/w) 2π π π e−(Dθ/πw) sin(mθ) dθ 0 D + w ∞ 2π θ 2e−(D/w) − 1 + π 1 − e−(D/w) ∞ 2π sin(mθ) dθ (6. C 0 .275) Sm To analyze the stability of a critical point (w.283) (6.277) (6. (6. · · · . . C 1 . .280) (6. .279) (6. S 1 . = = . . S m ) we add perturbations of the form w C0 C1 = w + w′ ′ = C 0 + C0 = .6.

Mixed condition 115 Substituting in Eqs.266) can be written as dx = f (x) dt as (6.288) The linearized version is dw′ dτ ′ dC0 dτ ′ dCm dτ π2 Γ π2 Γ ′ ′ cos α C1 − sin α S1 4D 4D ∞ D (−1)n − 1 ′ ′ = −D C0 + Sn π n=1 n = −Γw′ + ′ ′ = −2πm w Sm − 2πm S m w′ − D Cm (6. (6. (6. and in general form dx = Ax dt |A − λI| = 0 (6.293) The eigenvalues of the linearized system given by Eqs.4. the system given by Eqs.289) (6. (6. such that .289) to (6.295) are obtained numerically. we obtain the local form dw′ dτ ′ dC0 dτ ′ dCm dτ π2 Γ π2 Γ ′ ′ cos α C1 − sin α S1 4D 4D ∞ D (−1)n − 1 ′ ′ = −D C0 + Sn π n=1 n = −Γw′ + ′ ′ = −2πm w Sm − 2πm S m w′ − D Cm (6.285) (6.287) ′ dSm dτ = + ′ ′ 2πm w Cm + 2πm C m w′ − D Sm D π ∞ n=0 n=m m2 2m ′ ′ (−1)m+n − 1 Cn + 2πm w′ Cm − n2 m≥1 (6.290) + ′ dSm dτ D π ∞ n=1 n=m 2n ′ (−1)m+n − 1 Sn n2 − m2 m≥1 (6.263) to (6.291) = + ′ ′ 2πm w Cm + 2πm C m w′ − D Sm D π ∞ n=0 n=m m2 2m ′ (−1)m+n − 1 Cn − n2 m≥1 (6.266).294) (6.292) In general.6.292).286) + D π ∞ n=1 n=m n2 2n ′ ′ (−1)m+n − 1 Sn − 2πm w′ Sm − m2 m≥1 (6.263) to (6.

The neutral stability curve is obtained numerically from the condition that ℜ(λ) = 0. it is to be notice that the bifurcation occurs at a higher value of the tilt angle when D is smaller. The corresponding eigenvalues of the bifurcation point are shown in Figure 6. In this plot. new eigenvalues appear in such a way that they are placed symmetrically farther from the real axis and aligned to the previous set of slave complex eigenmodes. In this figure.8 0.30: Stability curve D and Γ for a thermosyphon inclination α = 0. It is clear that the natural branches which correspond to the first and third quadrants are stable.4 0. whereas the antinatural branches. However. Along the line of neutral stability. and λ are the eigenvalues.33 illustrates a view of several stability curves. For Γ > Γcr . On the other hand. I is the identity matrix. When α = 0.31 shows the plot of w − α curve for a value of the parameters D = 0.6 −0.5◦ increasing the region of instability. a Hopf-type of bifurcation occurs.2 Γ w 15 0 −0. the only presence is of a strange attractor. and Γ = 0.6. second and fourth quadrants are unstable.5. . where A is the Jacobian matrix corresponding to the vector field of the linearized system.4 Hopf Bifurcation Stable Unstable Unstable 10 Stable 5 −0. The number of eigenvalues in this figure is 42 which are obtained from a dynamical system of dimension 42.5 1 D 1. Hopf bifurcations occur for each branch of ecah particular curve. Figures 6. the stable and unstable regions can be identified.4.5 2 2. these being a critical point and a strange attractor of fractional dimension.5◦ to −32. between the second and fourth quadrants can be notice as well. Figure 6.8 −1 0 0 0.36 show the w − τ time series and phasespace curves for Γ = 0.20029. and.95Γcr . 6. each for a different value of the tilt angle α in a D − Γ plane at α = 0. Figure 6.30 for α = 0. α for D = 0.01Γcr .5 −150 −100 −50 0 α [°] 50 100 150 Figure 6. A schematic of the neutral curve is presented in Figure 6.4. The plots suggest the appearance of a subcritical Hopf bifurcation. and increase Γ.37 and 6. This system results from truncating the infinite dimensional system at a number for which the value of the leading eigenvalues does not change when increasing its dimension. where stable and unstable regions can be identified. Figure 6. This behaviour seems to be a characteristic of the dynamical system itself. When we increase the size of the system.20029.1. Figure 6. Two attractors coexist for Γ ≤ Γcr .34 illustrates the linear stability characteristics of the dynamical system in a w − α plot for a fixed Γ and three values of the parameter D.6 Unstable Hopf Bifurcation Stable 20 0.2 −0.4 Nonlinear analysis Let us select D = 1. the neutral curves appear to unfold when decreasing the tilt angle from 75.32.35 and 6.1 and Γ = 0. a Hopf bifurcation for both the positive and negative branches of the curve can be observed. The stable and unstable regions can be observed. whereas Figures 6.38 show the results for Γ = 1. The symmetry between the first and third quadrants.31: Stability curve w vs. Mixed condition 116 1 25 0.

3 −0.32: Eigenvalues at the neutral curve for D = 0.0 150 S Figure 6.5.1.D = 2.5 ° −100 5 −150 −0. Mixed condition 117 150 100 25 Unstable 50 ℑ(λ) 20 −50 Γ 0 15 10 α=3. α Γ = 4.2 w 0 −0.0 D=1.33: Neutral stability curve for different values of the tilt angle α. 1 0. Γ = 0.5 ° α=-32.5 ° α=75.5 ° α=21.4 −0.5 −0.34: Curve w vs.6.5 ° α=57.1.20029 and α = 0.5 D=0.1 0.2 −0.5 ° α=39.1 S U Γ=4.5 1 D 1.6 −0. .8 −1 −150 −100 −50 α° 0 50 100 D=2.4 −0.5 ° α=-14.6 0.8 0.1 ℜ(λ) 0 0.1 D=1. Figure 6.0 D=2.0.D = 1.4 0.5 Figure 6.5 2 2.2 0 0 Stable 0.5 U D=0.4.2 −0.0 and D = 0.

5.01Γcr . Γ = 1. Γ = 0.01Γcr .5 1 0.5.5 −2 −2.5 w 0 −0.4.37: Curve w vs. Mixed condition 118 2.5 2 1.5 2 1.5 −1 −1. τ for D = 1. Γ = 0.5 1 0.5 450 460 470 480 τ 490 500 510 φs 1 2 1 0 −1 −2 −3 4 2 0 −2 φc 1 4 2 0 −2 −4 −4 w Figure 6.95Γcr . Γ = 1. 2.5 w 0 −0.38: Phase-space trajectory for D = 1. τ for D = 1.5.5 −2 −2.6.5 1900 1920 1940 1960 1980 2000 φs 1 2 1 0 −1 −2 −3 4 2 0 −2 τ φc 1 4 2 0 −2 −4 −4 w Figure 6.35: Curve w vs.5 −1 −1.95Γcr . Figure 6. .5. Figure 6.36: Phase-space trajectory for D = 1.

1Γcr .6. In terms of control algorithms. Γ = 1. Both known heat flux and known wall temperatures may be looked at.9 0. For known wall temperature heating.5 1 0.99Γcr .39 presents a plot of the w − τ curve for Γ = 0. Now we choose D = 0. For the steady-state problem.40 and 6. Figure 6.8 0 500 1000 1500 2000 2500 2 1. Figure 6. and Figures 6. There is constant heating between points a and c.5 −1 −1. for Γ < Γcr we have stable solutions.2 w w 1.1 1 0. 5.5 1.1.6 Perturbation of one-dimensional flow Thermal control Consider the control of temperature at a given point in the loop by modification of the heating. Consider the same square loop but tilted through an angle θ where 0 ≤ θ < 2π. For Γ = 1.41 show the time series plots and phase space representation for Γ = 1. and increase Γ.3 1.42 and 6.5 0 −0.5. Plot (a) typical temperature distributions for different tilt angles. Find the temperature field and velocity for known heating if the total heating is not zero.01Γcr . 2. Figures 6. τ for D = 0.44 and 6. Problems 1.39: Curve w vs. Find the steady-state temperature field and velocity for known heating if the loop has a variable cross-sectional area A(s). Perturbation of one-dimensional flow 119 1. Find the velocity and temperature fields for known heating if the heating and cooling takes place at two different points. What is the effect on the known heat rate solution of taking a power-law relationship between the frictional force and the fluid velocity? 7.40: Curve w vs.1. τ for D = 0.43 for Γ = 20Γcr .5 6.5 −2 1980 τ 1985 1990 τ 1995 2000 Figure 6. the temperature field is uniform and the velocity is zero. one may use PID or on-off control. and constant cooling between e and g. and (b) the velocity as a function of tilt angle. In this case. What the condition for the existence of a solution? 6.45. determine the temperature distribution and the velocity as a function of θ. The strange attractor is shown in Figures 6. Γ = 0.99Γcr .1.4 1. 4. Find the pressure distributions for the different cases of the square loop problem.01Γcr the figures show a possible limit cycle undergoes a period doubling. 6. . show that if the wall temperature is constant. This implies a supercritical Hopf bifurcation. 3.

1Γcr .1.2 φc 1 0 φs 1 1985 1990 τ 1995 2000 1 0.2 −0.1Γcr .6.44: Phase-space trajectory for D = 0.1Γcr .4 −0.2 −0.43: Phase-space trajectory for D = 0.8 0. Figure 6.42: Curve w vs.8 1088 1090 1092 1094 τ 1096 1098 1100 1102 Figure 6.4 0.6 −0.41: Phase-space trajectory for D = 0.1.5 0 −0.6 0. Thermal control 120 1 0. Γ = 1. 0.6 −0.1.2 0 −0.5 φs 1 0 −0.6.4 −0.4 0.5 φc 1 1 0.2 −0.5 −1 1 2 1 0 −1 −1 −2 w −0. Γ = 1.4 0.2 0 φc 1 −0. τ for D = 0.4 0. Figure 6.8 −1 1980 1985 1990 τ 1995 2000 Figure 6.6 0.6 −0.1. Γ = 1. .8 0.6 0.8 −1 1980 −0. Γ = 20Γcr .

Neglect the small horizontal sections and state your other assumptions.2 −4 w Figure 6. Consider a long. vertical tube that is open at both ends. Compare the cooling rate achieved by an ionic liquid to water in the same loop and operating under the same temperature difference. 10.2 0. Find the steady state velocity in the loop.46: Tall natural circulation loop.4 0.1 −0. multiplicity of solutions and bifurcation diagrams. Illustrate typical cases with appropriate graphs. Consider a tall natural circulation loop shown in Fig.2 0 φc 1 4 2 0 −2 −0.2 0.1.45: Phase-space trajectory for D = 0. Thermal control 121 0. 13. Γ = 20Γcr . Make any assumptions you need to. vertical pipe compare the wall shear stress to mean flow velocity relation obtained from a twodimensional analysis to that from Poiseuille flow.1 φs 1 0 −0. xs . Study the steady states of the toroidal loop with known wall temperature including nondimensionalization of the governing equations.46 consisting of two vertical pipes of circular cross sections.6. Find the flow rate of the air due to natural convection. 12. The heat rate per unit length coming in and going out are both q. in the toroidal natural convection loop equations dx dt dy dt dz dt = = = y−x a sin φ − xz −b cos φ + xy .6. 6. 8. thin. q q L Figure 6. For a thin. The heating pipe has a diameter D.6 0. 9. Set up a controller for PID control of the velocity x to a given value. and that of the cooling side is 2D. 11. axial conduction and tilting effects. The air in the tube is heated with an electrical resistance running down the center of the tube. Find the combination of fluid parameters that determines the rate of heat transfer from a closed loop with known temperature distribution.

6. and show numerical results. Thermal control 122 where a and b are held constant. Use the tilt angle φ as control input.6. .

7) so that it satisfies equations (7. The temperature at the x = 0 end is reduced to zero for t > 0.6) x T1 = A erf √ 2 κ1 t x x T1 = A erf √ T2 = T0 − B erf √ 2 κ1 t 2 κ2 t 123 (7. Thus ∂T1 ∂T2 dX k1 − k2 = Lρ (7.5).8) .3) Furthermore the difference in heat rate into the interface provides the energy required for phase change.1) and (7. indicated by subscripts 1 and 2.1. t) to be = → 0 at T0 as x=0 x→∞ (7. t) (7.1 Neumann’s solution The material is initially liquid at T = T0 . Similarly (7. t) = T2 (X. the conduction equations is ∂ 2 T1 1 ∂T1 − ∂x2 κ1 ∂t 1 ∂T2 ∂ 2 T2 − ∂x2 κ2 ∂t = = 0 0 (7.1) (7. Thus T1 T2 Assume T1 (x.2) Stefan problems At the interface the temperature should be continuous. In each phase. are separated by an interface at x = X(t).1 [44] The two phases.5) (7.4) ∂x ∂x dt 7. so that T1 (X.Chapter 7 Moving boundary 7.

Using the remaining condition (7. we get k1 Ae−λ − k2 B This can be written as 2 √ κ1 −κ1 λ2 /κ2 = λLκ1 ρ π e κ2 (7.6).2) and (7.10) where λ is a constant.14) .13) (7.12) The temperatures are T1 T2 7.11) √ 2 2 k2 κ2 (T0 − T1 )e−κ2 λ /κ2 e−λ λL π 1 − = √ erf λ c1 T1 s k1 κ2 T1 erfc(λ κ1 /κ2 ) (7. Stefan problems 124 satisfies equation (7.4).3) requires that x x A erf √ = T0 − B erf √ = T1 2 κ1 t 2 κ2 t This shows that √ X = 2λ κ1 t (7.2 Goodman’s integral = x T1 ) erf ( √ erf λ 2 κ1 t T0 − T1 x = T0 − ) erfc( √ 2 κ2 t erfc(λ κ1 /κ2 ) (7.1. The.7.9) (7.1. condition (7.

Part IV Multiple spatial dimensions 125 .

y) = X(x)Y (y) (8. the Biot number.3) Consider a square of unit side with θ = T − T∞ being zero all around.6) (8.5) This leads to one equation (8.1) ∂2T ∂2T 1 ∂T = + − m2 (T − T∞ ) α ∂t ∂x2 ∂y 2 (8.2 8. 8. In the steady state ∂2T ∂2T + − m2 (T − T∞ ) 2 ∂x ∂y 2 (8.7) with X(0) = X(1) = 1.1 Steady-state problems See [206].4) so that 1 d2 X 1 d2 Y =− + m2 = −λ2 2 X dx X dy 2 d2 X + λ2 X = 0 dx2 X(x) = A sin λx + B cos λx 126 (8.2.Chapter 8 Conduction 8. Let θ(x.1 Transient problems Two-dimensional fin The governing equation is cρdx dyL which simplifies to ∂T = Lk ∂t ∂2T ∂2T + ∂x2 ∂y 2 − hdx dy(T − T∞ ) (8.2) where m2 = h/kL. except for one edge where it is unity. Thus .

where due to the boundary conditions.4 Radiating fins Non-Cartesian coordinates Toroidal.1: Two-dimensional fin. B = 0 and λ = nπ. 3. Shape factor. Problems 1. The two cylindrical surfaces shown as v = 1 and v = 2 are kept at temperatures T1 and T2 .edu. Use the eigenfunction expansion method to reduce the governing equation to an infinite set of ODEs and solve.wolfram. but in parabolic cylindrical coordinates. Set up and solve a conduction problem similar to Problem 2. . Radiating fins 127 L x y Figure 8. .9) The condition Y (0) = 0 gives B = 0. Consider conduction in a square plate with Dirichlet boundary conditions.3. bipolar. Use Morse and Feshbach’s notation as shown in http://www.8) 8. 2.8. Show that the separation of variables solution for ∇2 T = 0 for a rectangle can also be obtained through an eigenfunction expansion procedure. Thus θ(x. 2. 5.html with a = 1. n = 1. .math. Another equation is d2 Y − m2 + λ 2 Y = 0 dy 2 with Y (y) = A sinh m2 + n2 π 2 y + B cosh m2 + n2 π 2 y (8. Consider an unsteady one-dimensional fin of constant area with base temperature known and tip adiabatic.htm 4.cn/mathency/math/p/p059.10) (8.sdu.com/BipolarCylindricalCoordinates. Sketch the geometry of the annular material between v = 1 and v = 2 and find the temperature distribution in it by solving the Laplace’s equation ∇2 T = 0. y) = An sin nπx sinh m2 + n2 π 2 y (8.3 8.. Moving boundary problem at a corner. Consider steady-state conduction in bipolar coordinates shown in http://mathworld. respectively. . Find the appropriate eigenfunctions for the Laplacian operator for this problem.

1) 9. respectively. 9.4 Multiple solutions See [166]. Consider a rectangular plate of size Lx × Ly in the x.2.Chapter 9 Forced convection See [206]. with an overall heat transfer coefficient of U .1 Two-dimensional flow (9.5 Plate heat exchangers There is flow on the two sides of a plate. 128 .and y-directions. 9.1 9. as shown in Fig.3 Leveque’s solution Leveque (1928) 9.2 Low Reynolds numbers Potential flow [169] If the heat flux is written as q = ρcuT − k∇T the energy equation is ∇·q=0 Heat flows along heatlines given by dx dy dz = = ρcT ux − k(∂T /∂x) ρcT uy − k(∂T /∂y) ρcT uz − k(∂T /∂z) The tangent to heatlines at every point is the direction of the heat flux vector. 9.3) (9. 1 and 2.2) (9.

For the flow in the x-direction. we get ∂Ty Tx = Ty + Cy R (9. The overall heat transfer coefficient between the two fluids is U . Plate heat exchangers 129 x flow y flow Figure 9. we have 2 Ty 1 ∂Ty 1 ∂Ty + + 2R =0 Cy ∂x Cx ∂y ∂x∂y (9. From equation (9. The mass flow rate of the flow is mx per unit transverse length.8) Nusselt (Jakob.9. and Cx = cx mx . For the other fluid ∂Ty = Tx − Ty (9.5). y) and Ty (x.6) 2Cy R ∂y These equations have to be solved with suitable boundary conditions to obtain the temperature fields Tx (x.4) ∂x where cx is the specific heat of that fluid.6). we get 2Cx R ∂Tx = Ty − Tx ∂x (9. y). 1957) gives an interesting solution in the following manner. 9. Simplifying. Let the plate be . which we will take to be a constant. y) and my . the steady heat balance on an elemental rectangle of size dx×dy gives ∂Tx cx mx dy dx = U dx dy (Ty − Tx ) (9. The flow in the other side of the plate is in the y-direction with the corresponding quantities Ty (x.1: Schematic of crossflow plate HX.5) where R = 1/2U is proportional to the thermal resistance between the two fluids. y).5. The flow on one side of the plate is in the x-direction with a temperature field Tx (x.7) ∂y Substituting in equation (9.1.

i Tx. η ′ )ebη dη ′ (9. we find the Volterra integral equation ξ η 0 θx (ξ.15) (9. from equation (9. η) = e−aξ + ae−aξ 0 θy (ξ ′ . η ′ )ebη dη ′ (9.i Ty − Ty. Nondimensional variables are ξ η θx θy = = = = x L y W Tx − Ty.16) at ξ = 0 at η = 0 (9. η) = be−bη 0 η (9.17) (9.21) ′ θx (ξ.14) a = b The governing equations are then = a(θx − θy ) = − b(θx − θy ) = with boundary conditions θx θy Equation (9.i Tx.22) Using the same procedure. we get C=0 so that θy (ξ.23) Substituting for θy .5. η ′ )eaξ +bη dξ ′ dη ′ ′ ′ (9.11) (9. η)eaξ dξ ′ ′ (9.9.9) (9.16) can be written as = = 1 0 ∂θx ∂ξ ∂θy ∂η (9.24) . η) = e−aξ + abe−(aξ+bη) 0 θx (ξ ′ .15) we get ξ θx (ξ. Plate heat exchangers 130 of dimensions L and W in the x.10) (9.18).19) θx (ξ.and y-directions.18) ∂θy + bθy = bθx ∂η Solving for θy we get θy = e−bη C(ξ) + b 0 η ′ (9.13) (9.i UWL Cx UWL Cy (9.12) (9.20) From the boundary condition (9.i − Ty.i − Ty.

and taking λ = 1.40) (9. η ′ )eaξ +bη dξ ′ dη ′ ′ ′ (9.41) . .32) The solutions are φ0 φ1 φ2 φ3 . η ′ )eaξ +bη dξ ′ dη ′ ′ ′ φn (ξ.1 Find a solution of the same problem by separation of variables.9. η) + φ1 (ξ.26) This can be substituted in the integral equation. η) φ2 (ξ.27) φ0 (ξ ′ .29) (9. η) . η) + . . . ξ η 0 = e−aξ ξ η 0 ξ η 0 (9. + λn φn (ξ.31) (9.33) (9. η) = φ0 (ξ. Example 9.37) (9. η) = φ0 (ξ. Thus φ0 (ξ.36) (9.26). y) = X(x)Y (y) Substituting and dividing by XY . . Since λ is arbitrary. . η) (9. η) + . η) + λφ1 (ξ.34) (9.38) Substituting into the expansion. Plate heat exchangers 131 for the unknown θx . . η) + λ2 φ2 (ξ. equation (9.5. . we get 1 dY 1 ∂Tx + + 2R = 0 Cx ∂x Cy dy (9. . η) φ1 (ξ. η) = abe−(aξ+bη) 0 φn−1 (ξ ′ . φn = e−aξ = aξe−aξ (1 − e−bη ) 1 2 2 −aξ = a ξ e (1 − e−bη − bηe−bη ) 2 1 1 = a3 ξ 3 e−aξ (1 − e−bη − bηe−bη − b2 η 2 e−bη ) 2×3 2 = 1 1 n n −aξ a ξ e (1 − e−bη − bηe−bη − . . − bn−1 η n−1 e−bη ) n! (n − 1)! θx (ξ.25) Let us express the solution in terms of a finite power series θx (ξ. where ξ η 0 θx (ξ. η ′ )eaξ +bη dξ ′ dη ′ ′ ′ (9. We will first solve the Volterra equation for an arbitrary λ. η ′ )eaξ +bη dξ ′ dη ′ 0 ′ ′ = abe−(aξ+bη) = abe−(aξ+bη) 0 (9.35) (9. Taking Ty (x.39) (9. η) where the φs are given above. η) + φ2 (ξ. η) = e−aξ + abλe−(aξ+bη) 0 θx (ξ ′ . . we get (9. the coefficient of each order of λ must vanish. + φn (ξ.28) (9.30) φ1 (ξ ′ .

(9. we get – » c x y Tx = exp − + a−R cx mx (a + R) cy my (a − R) The rate of heat transfer over the entire plate. is given by Z L−y Q = Cx [Tx (Lx .44) (9.42) (9. Solving the two equations and taking their product. each must be a constant.6 Falkner-Skan boundary flows Problems 1.7).6. Substituting in equation (9. Thus we can write dX 1 + X dx cx mx (a + R) dY 1 + Y dy cy my (a − R) = = 0 0 (9.45) (9. Falkner-Skan boundary flows 132 Since the first term is a function only of x.43) where a is a constant. This is a problem . Q.9. y) − Tx (0. we have – » c x y Ty = exp − + a+R cx mx (a + R) cy my (a − R) where c is a constant. and the second only of y.47) 9. y)] dy 0 – » Ly lx − −2 = cCx Cy exp − Cx (a + R) C2 (a − R) The heat rate can be maximized by varying either of the variables Cx or Cy .46) (9.

133 .2 10. This is a problem. Problems 1.3 Governing equations Cavities Marangoni convection See [204].1 10.Chapter 10 Natural convection 10.

1 Darcy’s equation (11. 201] 11. 11.2) where f is the body force per unit mass. 134 . From equations (11.1. There is still slip at a boundary. It is similar to the incompressible Navier-Stokes equation with constant properties where the inertia terms are dropped and the viscous force per unit volume is represented by −(µ/K)V.1.1) and (11.3) Forchheimer’s equation which is often used instead of Darcy’s equation is ∇p = − µ V − cf K −1/2 ρ|V|V + ρf K (11.1) For the momentum equation.4) where cf is a dimensionless constant.Chapter 11 Porous media [103. Sometimes a term cρ0 ∂V/∂t is added to the left side for transient problems.2 Forchheimer’s equation (11.2). allowing for slip in the tangential direction.1 Governing equations The continuity equation for incompressible flow in a porous medium is ∇·V =0 11. The condition on the velocity is that of zero normal velocity at a boundary. Here K is called the permeability of the medium and has units of inverse area. but it is normally left out because it is very small. the simplest model is that due to Darcy ∇p = − µ V + ρf K (11. 130. for f = 0 we get ∇2 p = 0 from which the pressure distribution can be determined.

14) (11.13) u = − v is u v For = U = 0 K µ K = − µ (11.1.11) (11.15) Ux = P ex ≫ 1 αm (11. An equivalent form is σ where αm σ See [1]. = = km ρcp (ρc)m ρcp (11.3 Brinkman’s equation µ V + µ∇2 V + ρf ˜ K Another alternative is Brinkman’s equation ∇p = − (11.12) (11. and (ρc)m (ρc)m = φρcp + (1 − φ)(ρc)m The energy equation is (11.2.11. Forced convection 135 11.6) where km (11.8) 11. In this model there is no slip at a solid boundary. ˜ 11. The subscripts m refers to the solid matrix.1 Forced convection Plane wall at constant temperature ∂u ∂v + ∂x ∂y The solution to = 0 ∂p ∂x ∂p ∂y (11.2.2 11.4 Energy equation ∂T + ρcp V · ∇T = km ∇2 T ∂t is the effective thermal conductivity. and φ is the porosity of the material.9) (11.7) is the average heat capacity.5) where µ is another viscous coefficient.16) .1.10) ∂T + V · ∇T = αm ∇2 T ∂t (11.

11. Forced convection 136 the energy equation is u or ∂T ∂T ∂2T +v = αm 2 ∂x ∂y ∂y U ∂2T ∂T = αm 2 ∂x ∂y (11.26) (11.27) dθ ∂η dη ∂y U dθ dη αm x d2 θ U (T∞ − Tw ) 2 dη αm x 1 θ′′ + η θ′ = 0 2 θ(0) 2 (11.17) (11.32) (11.28) with = 0 1 (11.18) The boundary conditions are T (0) = Tw T (∞) = T∞ Writing η = y U αm x T − Tw T∞ − Tw (11.25) (11.19) (11.33) e−η ′2 /4 dη ′ + C2 .20) θ(η) = we get ∂T ∂x = = ∂T ∂y = = ∂2T ∂y 2 so that the equation becomes = (T∞ − Tw ) (T∞ − Tw ) (T∞ − Tw ) (T∞ − Tw ) dθ ∂η dη ∂x dθ dη −y U 1 −3/2 x αm 2 (11.22) (11.30) θ(∞) = We multiply by the integrating factor eη /4 to get (11.23) (11.31) 2 d eη /4 θ′ = 0 dη The first integral is θ′ = C1 e−η Integrating again we have θ = C1 0 η 2 /4 (11.29) (11.24) (11.21) (11.2.

the velocity field is u = Cx v The energy equation is Cx = −Cy (11.2.43) Example 11.2.44) (11.38) (11. 11.41) y=0 = x = = 1 Ux √ π αm 1 √ P e−1/2 π x (11.46) .39) 2 √ π erf η/2 0 e−x dx η 2 2 (11.37) (11.35) (11. we find that C1 = 1/ π and C2 = 0.2 Stagnation-point flow For flow in a porous medium normal to an infinite flat plate.42) (11.1 Find the temperature distribution for flow in a porous medium parallel to a flat plate with uniform heat flux.40) (11.11. Forced convection 137 With the change in variables x = η ′ /2. the solution becomes η/2 θ = 2C1 0 e−x dx 2 (11.34) √ Applying the boundary conditions.36) The local Nusselt number is given by N ux = hx km ∂θ ∂y (11. Thus θ = = The heat transfer coefficient is defined as h q ′′ Tw − T∞ ∂T km = − Tw − T∞ ∂y ∂θ = km ∂y = (11.45) ∂T ∂T ∂2T − Cy = αm 2 ∂x ∂y ∂y (11.

we get 1 θ′′ = − (θ + ηθ′ ) 2 ∂θ ∂η ∞ (11.48) (T − T∞ ) dy (11.49) become = = 0 at η = 0 1 (11.56) can be written as θ′′ = − which integrates to 1 d (ηθ) 2 dη (11.58) θ dy −∞ The equation (11.51) = = 0 at y = 0 ∞ ∂2T ∂T = αm 2 ∂x ∂y (11.11. Forced convection 138 11.50) Ux αm (11. the governing equation is U where the boundary conditions are ∂T ∂y q′ Writing η = y U αm x T − T∞ q ′ /km (11.57) (11.55) q′ km q′ km αm dθ U U x dη αm x αm dθ U U x dη αm x Substituting in the equation.2.59) 1 θ′ = − ηθ + C1 2 (11.52) (11.49) (ρcp )U −∞ θ(η) = we find that ∂T ∂x ∂T ∂y ∂2T ∂y 2 = θ = = q′ km αm U 1 − x−3/2 2 + dθ y dη U αm 1 q′ − x−3/2 2 km αm Ux (11.56) The conditions (11.2.3 Thermal wakes Line source For P ex ≫ 1.47) (11.53) (11.48)-(11.60) .54) (11.

1: Stability of horizontal porous layer. Natural convection 139 Since θ′ = 0 at η = 0.11.1 Natural convection Linear stability This is often called the Horton-Rogers-Lapwood problem. Integrating again.69) (11.61) 1= −∞ θ dη = C2 −∞ e−η 2 /4 √ dη = 2 πC2 (11.3. B y x A gravity Figure 11. and consists of finding the stability of a horizontal layer of fluid in a porous medium heated from below. we find that C1 = 0.63) Thus the solution is or (11. The governing equations are ∇·V µ −∇p − V + ρg K ∂T + V · ∇T σ ∂t ρ = 0 = 0 = αm ∇2 T = ρ0 [1 − β (T − T0 )] (11.68) (11.67) (11. we get θ = C2 e−η Substituting in the other boundary condition ∞ ∞ 2 /4 (11.2 Show that for a point source T − T∞ = q −U r 2 exp( ) 4πkx 4αm x (11.62) from which 1 C2 = √ 2 π 2 1 θ = √ e−η /4 2 π (11.65) 1 q′ T − T∞ = √ 2 π km ( −U y 2 αm ) exp( ) Ux 4αm x Example 11.3. The geometry is shown in Fig.3 11.70) .1. 11.64) (11.66) 11.

89) (11.88) (11. We apply a perturbation to each variable as V T p Substituting and linearizing = V + V′ = T + T′ = p + p′ (11.78) (11.82) (11.71) (11.75) (11.74) (11.76) ∇ · V′ µ ′ −∇p′ − V − βρ0 T ′ g K ∆T ′ ∂T ′ − w ∂t H Using the nondimensional variables x∗ t∗ V∗ T∗ p∗ the equations become.3.80) (11. The basic steady solution is V T p = 0 (11. Natural convection 140 where g = −gj.90) ∇2 w = Ra∇2 T H where ∇H = ∂2 ∂x2 . on dropping *s ∇·V = = = = = = 0 = 0 = αm ∇2 T ′ (11.86) (11.11.84) = = 0 0 (11.81) (11.72) y2 − 2y H (11.85) (11.83) (11.73) y = T0 + ∆T 1 − H 1 = p0 − ρ0 g y + β∆T 2 ′′ For constant heat flux ∆T = qs /km .79) x H αm t σH 2 HV′ αm T′ ∆T Kp′ µαm (11.87) −∇p − V + Ra T k ∂T −w ∂t where Ra = From these equations we get = ∇2 T ρ0 gβKH∆T µαm (11.77) (11.

For a self-adjoint operator L. t) T (x.100) V ∇ · (P i + Qj) dV n · (P i + Qj) = S If n·(P i+Qj) = 0 at the boundaries (i. as shown below. Thus.99) (11. we must have (u.11. y.101) (11. then L is self-adjoint. t) Substituting into the equations we get d2 − k 2 − s Θ = −W dy 2 d2 − k 2 W = −k 2 Ra Θ dy 2 where 2 2 k 2 = kx + ky = W (y) exp (st + ikx x) = Θ(y) exp (st + ikx x) (11. y. v) If vL(u) − uL(v) = then [vL(u) − uL(v)] dV = = V (11. Lv) = (Lu. z.95) Isothermal boundary conditions The boundary conditions are W = Θ = 0 at either wall. the wavenumbers kx and ky must be real.102) W = k 2 Ra W (11.91) (11.92) (11.98) (11. For the solutions to remain bounded as x.e.104) . marginal stability occurs when s = 0. Furthermore. impermeable). for which d2 − k 2 Θ = −W dy 2 d2 − k 2 W = −k 2 Ra Θ dy 2 from which d2 − k2 dy 2 2 (11. Natural convection 141 Using separation of variables w(x.3.96) ∂Q ∂P + ∂x ∂y ∂P ∂Q + ∂x ∂y (11.103) The eigenfunctions are W = sin nπy (11. it can be shown that s is also real. which is the case here. y → ∞. since the eigenvalue problem is self-adjoint.97) dV (11.93) (11.94) (11. z.

. 11. y). where u v Darcy’s equation becomes − µ ∂ψ ∂p − ∂x K ∂y ∂p µ ∂ψ − + ∂y K ∂x = ρ0 g [1 − β(T − T0 )] sin φ = ρ0 g [1 − β(T − T0 )] cos φ (11.3. . Constant heat flux conditions Here W = dΘ/dy = 0 at the walls.109) (11.3. we get Rac = 12. For the zeroth order system d 2 W0 =0 dy 2 d2 W1 dy 2 d2 Θ1 + W1 dy 2 (11. .108) (11. 3. The solutions is W0 = 0. Ra = Ra0 + α2 Ra1 + . Finally. Natural convection 142 where n = 1.2 Steady-state inclined layer solutions Consider an inclined porous layer of thickness H at angle φ with respect to the horizontal shown in Fig.116) (11. Θ = Θ0 + α2 Θ1 + . . We write W = W0 + α 2 W1 + .113) with W1 = dΘ1 /dy = 0 at the walls. .117) ∂ψ ∂y ∂ψ = − ∂x = (11.105) For each n there is a minimum value of the critical Rayleigh number determined by dRa =2 dk n2 π 2 +k k − n2 π 2 +1 k2 (11.11.107) (11. which gives Rac = 4π 2 for the onset of instability. To the next order = W0 − Ra0 Θ0 = Θ0 (11. as long as Ra = n2 π 2 +k k 2 (11.110) (11..114) (11. .111) with W0 = dΘ0 /dy = 0 at the walls.112) (11. . 2. Θ0 = 1. .2 Introducing the streamfunction ψ(x.115) . 11. .106) The lowest critical Ra is with k = π and n = 1.

Taking ∂/∂y of the first and ∂/∂x of the second and subtracting.126) (11.2: Inclined porous layer. at at x=± A 2 1 y=± 2 (11.124) (11.122) = −Ra = ∂T ∂T cos φ − sin φ ∂x ∂y ∂2T ∂2T + ∂x2 ∂y 2 (11.127) (11.123) (11.119) .121) = αm ∂T ∂T cos φ − sin φ ∂x ∂y ∂2T ∂2T + ∂x2 ∂y 2 (11.120) (11. ∂y ψ = 0.11.118) (11.125) With a parallel-flow approximation. we assume [167] ψ T = ψ(y) = Cx + θ(y) (11.3. we have ∂2ψ ∂2ψ ρ0 gβK + =− ∂x2 ∂y 2 µ The energy equation is ∂ψ ∂T ∂ψ ∂T − ∂y ∂x ∂x ∂y Side-wall heating The non-dimensional equations are ∂2ψ ∂2ψ + ∂x2 ∂y 2 ∂ψ ∂T ∂ψ ∂T − ∂y ∂x ∂x ∂y where the Rayleigh number is Ra =? The boundary conditions are ∂T =0 ∂x ∂T = −1 ψ = 0. Natural convection 143 y x φ Figure 11.

11.3. Natural convection

144

The governing equations become ∂2ψ dθ − Ra sin φ + RC cos φ = ∂y 2 dy dψ d2 θ −C = dy 2 dy
1/2 −1/2

0 0

(11.128) (11.129)

An additional constraint is the heat transported across a transversal section should be zero. Thus uT − ∂T ∂x dy = 0 (11.130)

Let us look at three cases. (a) Horizontal layer For φ = 0◦ the temperature and streamfunction are T ψ = Cx − y 1 + = − RaC 2 4y 2 − 3 24 (11.131) (11.132)

RaC 4y 2 − 1 8

Substituting in condition (11.130), we get C 10R − Ra2 C 2 − 120 = 0 the solutions of which are (11.134) 1 10(Ra − 12) (11.135) C = Ra 1 10(Ra − 12) (11.136) C = − Ra The only real solution that exists for Ra ≤ 12 is the conductive solution C = 0. For C > 12, there are two nonzero values of C which lead to convective solutions, for which ψc Nu = = RaC 8 12 12 − RaC 2 (11.137) (11.138) C = 0 (11.133)

For φ = 180◦ , the only real value of C is zero, so that only the conductive solution exists. (b) Natural circulation Let us take C sin φ > 0, for which we get ψc Nu where α2 B = RC sin φ 1 + C cot φ = − cosh α 2 (11.141) (11.142) B α 1 − cosh C 2 α = − 2B sinh α + αC cot φ 2 = (11.139) (11.140)

11.3. Natural convection

145

and the constant C is determined from C− B2 2C sinh α α 2 α − 1 − B cot φ cosh − sinh α 2 α 2 =0 (11.143)

(c) Antinatural circulation For C sin φ < 0, for which we get ψc Nu where β2 B and the constant C is determined from C− End-wall heating Darcy’s law is ∇2 ψ The boundary conditions are ψ = 0, ∂T = −1 ∂x ∂T =0 ψ = 0, ∂y at at A 2 1 x=± 2 x=± (11.150) (11.151) = R ∂T ∂T sin φ + cos φ ∂x ∂y (11.149) B2 2C sin β β 2 β − 1 − B cot φ cosh − sinh β 2 β 2 =0 (11.148) = −RC sin φ 1 + C cot φ = − cosh β 2 (11.146) (11.147) = B C 1 − cosh
β 2

β 2

(11.144) (11.145)

= −

β 2B sinh + βC cot φ

With a parallel-flow approximation, we assume ψ T The governing equations become dθ dψ −C 2 dy dy dθ ∂2ψ − R cos φ − RC sin φ = 0 ∂y 2 dy An additional constraint is the heat transported across a transversal section. Thus
1/2 −1/2

= ψ(y) = Cx + θ(y)

(11.152) (11.153)

=

0

(11.154) (11.155)

uT −

∂T ∂x

dy = 1

(11.156)

11.3. Natural convection

146

Let us look at three cases. (a) Vertical layer For φ = 0◦ the temperature and streamfunction are T ψ where α2 = −RC Substituting in condition (11.130), we get C 10R − R2 C 2 − 120 = 0 the solutions of which are C C C = = 0 1 R 1 R (11.161) 10(R − 12) 10(R − 12) (11.162) (11.163) (11.160) (11.159) = Cx + = B1 B2 sin(αy) − cos(αy) α α B1 B2 cos(αy) + sin(αy) + B3 C C (11.157) (11.158)

= −

The only real solution that exists for R ≤ 12 is the conductive solution C = 0. For C > 12, there are two nonzero values of C which lead to convective solutions, for which ψc Nu = = RC 8 12 12 − RC (11.164) (11.165)

For φ = 180◦ , the only real value of C is zero, so that only the conductive solution exists. (b) Natural circulation Let us take C sin φ > 0, for which we get ψc Nu where α2 B and the constant C is determined from C− B2 2C sinh α α 2 α − 1 − B cot φ cosh − sinh α 2 α 2 =0 (11.170) = RC sin φ 1 + C cot φ = − cosh α 2 (11.168) (11.169) B α 1 − cosh C 2 α = − 2B sinh α + αC cot φ 2 = (11.166) (11.167)

(c) Antinatural circulation

11.3. Natural convection

147

For C sin φ > 0, for which we get ψc Nu where β2 B and the constant C is determined from C− B2 2C sin β β 2 β − 1 − B cot φ cosh − sinh β 2 β 2 =0 (11.175) = −RC sin φ 1 + C cot φ = − cosh β 2 (11.173) (11.174) = B C 1 − cosh
β 2

β 2

(11.171) (11.172)

= −

β 2B sinh + βC cot φ

Problems
1. This is a problem.

1 Stefan problems 148 .Chapter 12 Moving boundary 12.

Part V Complex systems 149 .

1 [209] [219] is a review of the radiative properties of semiconductors. The steady-state temperature distribution in a one-dimensional radiative fin is given by dT + hT 4 = 0 dx Is the solution unique and always possible? 3.Chapter 13 Radiation 13. Monte Carlo methods Problems 1. the dynamics of the small-body temperature is governed by dT1 4 4 M1 c1 = −A1 F12 σ(T1 − T2 ) + Q1 . Show that between one small body 1 and its large surroundings 2. Consider an unsteady n-body radiative problem. The temperature of the ith body is given by Mj cj dTj dt = − n X 4 Ai Fij σ(Ti4 − Tj ) + Qj 4 Fji σ(Ti4 − Tj ) + Qj = What kind of dynamic solutions are possible? Aj i=1 n X i=1 2. dt 150 .

Chapter 14 Boiling and condensation 14.1 Homogeneous nucleation 151 .

119. and [36] of photonic devices.1 Diffusion by random walk One-dimensional Consider a walker along an infinite straight line moving with a step size ∆x taken forward or backward with equal probability in a time interval ∆t. x + dx/2] be P (x.. t − ∆t) = P± ∂P ∂P ∆x − ∆t ∂x ∂t 1 ∂2P ∂2P 1 ∂2P 2 + ∆x2 ± ∆x∆t + ∆t 2 ∂x2 ∂x∂t 2 ∂t2 1 ∂3P 1 ∂3P 1 ∂3P 1 ∂3P 3 ± ∆x3 − ∆x2 ∆t ± ∆x∆t2 − ∆t 3 2 ∂t 2 6 ∂x 2 ∂x 2 ∂x∂t 6 ∂t3 +.1) we get ∂2P 1 ∂2P 1 ∂3P ∂P −D 2 = ∆t − ∆x2 + .1) Assuming that ∆x and ∆t are small. Substituting in Eq. 178. t − ∆t] . x − ∆x + dx/2] or [x + ∆x − dx/2. x + ∆x + dx/2]. we have P (x. [86. 202.. . 219] 15.Chapter 15 Microscale heat transfer [108] is a review of microscale heat exchangers. . t). . t). and the probability that the walker is in an interval [x − dx/2.1 15.1. t − ∆t) + P (x + ∆x. (15.. t) = 1 [P (x − ∆x. Let the current position of the walker be x. Since there is equal probability of the walker in the previous time step to have been in the interval [x − ∆x − dx/2. 2 (15. 2 ∂t ∂x 2 ∂t 2 ∂x2 ∂t (15. 186. If we let ∆x → 0 and ∆t → 0 such that D is constant.2) where the terms on the right side are evaluated at (x. (15. we get the diffusion equation ∂P ∂2P (15.3) where D = ∆x2 /2∆t. Taylor series expansions give P (x ± ∆x.4) =D 2. ∂t ∂x 152 .

t − ∆t) = P + etc. v).5) where S is a sphere of radius |∆r| centered on r.9) . (a) Fourier’s law Assume ∂/∂t = 0. ε. t)f (r.12) (15.2 Boltzmann transport equation The classical distribution function f (r. t) is defined as number of particles in the volume dr dv in the six-dimensional space of coordinates r and velocity v. ∂P ∂P 1 ∂2P ∆ri − ∆t + .. scat (15. t). Thus ∂f ∂f f0 − f + v · ∇f + a · = .1. and ∆r be a step in any direction where |∆r| is a constant. t) = where D(ε) is the density of energy states ε.. 123. 124] ∂f ∂f + v · ∇f + a · = ∂t ∂v ∂f ∂t .2. Boltzmann transport equation 153 15. Following a volume element in this space. dT (15. t − ∆t) dS (15. we can write ∇f0 = df0 ∇T.8) Under this approximation = scat (15. Also P (r + ∆r.7) where a = dv/dt is the acceleration due to an external force.6) 15. t)εD dε. ∂t ∂v τ Several further approximations can be made. a = 0 and ∇f = ∇f0 in the left side of Eq. v.2 Multi-dimensional Let P = P (r. The heat flux is then q(r. Then P (r. (15. 15.10) so that f = f0 − τ v · ∇f0 . t) = S P (r + ∆r. Introducing explicitly the dependence of f on temperature. we have the balance equation [109.11) (15.10) f0 − f .15. τ v(r. The term on the right side is due to collisions and scattering. (15.2.1 Relaxation-time approximation ∂f ∂t where τ = τ (r. ∂ri ∂t 2 ∂ri ∂ri (15.

(15. Newton’s second law gives m d2 xn dt2 = c(xn+1 − xn ) − c(xn − xn−1 ) = c(xn+1 − 2xn + xn−1 ).16) ∇T vεD dε = − ∂t dT τ Using Eq.8).19) (15. 15. (15.3.1. (15. Let xi = xei(nka−ωt) . (15. Phonons 154 xi−1 xi xi+1 Figure 15.18) (15.3.14). and a is the mean distance between the atoms.13) (15. The mass of each atom is m. (15.15) v τv · df0 dT εD dε. then the dispersion relation is ω= 2c m 1/2 1/2 (15. (15.1: Lattice of atoms of a single type Using Eqs. For a typical atom n. (15. Eq. This is Cattaneo’s equation that can be compared to Fourier’s law.20) (1 − cos ka) . (15. where k= since vf0 εD dε = 0.13).21) .12) in (15. (15. the spring constants are c.17) ∂t where k is given by Eq.1 Phonons Single atom type A lattice of atoms of a single type is shown in Fig.11) and (15.8).3 15. q+τ 15.15. (15. we get q = −k∇T. Multiply Eq.14) (b) Cattaneo’s equation Assume τ = constant and ∇f = (df0 /dT )∇T . this gives ∂q = −k∇T.10) by vεD dε and integrate to get ∂ df0 1 v· vf εD dǫ + vf εD dε.

23) .25) c 2m 1/2 2c mk 2 1/2 (1 − cos ka) a sin ka (1 − cos ka) c m 1/2 1/2 .26) (15. (15. (15.2: Lattice of atoms of two different type The phase velocity is vp = and the group velocity is vg = For ka → 0.31) d2 y i m2 2 dt (15.27) (15.29) (15.2 Two atom types Newton’s second law gives m1 d2 xi dt2 = c(yi − xi ) − c(xi − yi−1 ) = c(yi − 2xi + yi−1 ).28) (15. = c xe ika (15.32) (15.36) 2 − 2y + x .33) (15.30) (15.22) 1/2 . Phonons 155 xi yi Figure 15. (15. = yei(nka−ωt) .3. We can also write k= 1 cvg l 3 (15. we have vg = a The thermal conductivity k = ke + kp where ke and kp are those due to electron and phonon transports.35) (15.34) so that −m1 xω 2 −m2 yω = c y − 2x + ye−ika . = c(xi+1 − yi ) − c(yi − xi ) = c(xi+1 − 2yi + xi ).3. (15. . and l is the mean free path.24) where c is the specific heat. Let xi yi = xei(nka−ωt) .15. 15.

4. Thin films 156 which can also be written as 2c − m1 ω 2 −c(1 + eika ) This means that (2c − m1 ω 2 )(2c − m2 ω 2 ) − c2 (1 + e−ika )(1 + eika ) = 0. which simplifies to m1 m2 ω 4 − 2c(m1 + m2 )ω 2 + 2c2 (1 − cos ka) = 0. 2 1 (15.4 Thin films .15. The solution is ω2 = 1 2c(m1 + m2 ) ± 2c 2m1 m2 m2 + m2 + 2m1 m2 cos ka .37) The positive sign corresponds to the optical and the negative to the acoustic mode.40) (15.39) (15. (15.38) −c(1 + e−ika ) 2c − m2 ω 2 x y = 0 0 . 15.

Chapter 16 Bioheat transfer 16. 53].1 Mathematical models Good references are [35. 157 .

1 Fin analysis Analysis of Kraus (1990) for variable heat transfer coefficients.7 Transient behavior Ontko and Harris (1990) For both fluids mixed dT in out in out + m1 c1 (T1 − T1 ) + mc2 (T2 − T2 ) = 0 ˙ ˙ dt For one fluid mixed and the other unmixed. 17.3 17. we have Mc ρAc2 ∂T2 ∂T2 + ρV2 Ac2 + hP (T − T1 ) = 0 ∂t ∂x 158 (17. p. 17.5 Microchannel heat exchangers Phillips (1990).2) . 87 [130].4 Heat transfer augmentation Maldistribution effects Rohsenow (1981) 17. 17.Chapter 17 Heat exchangers 17.1) (17.2 Porous medium analogy See Nield and Bejan.6 Radiation effects See Ozisik (1981). Shah (1981) 17.

17.t). Tt (L. they appear in nonlinear form in Eqs. Fig. water and tube material. Correlations 159 17. t) and Tw (ξ. (17. The boundary and initial conditions are w Tt (0. ˙ . n y(x) = k=0 a(k)xk dk where − 1 < x < 1 A power-law correlation of the form y = cxn satisfies the invariance condition given by equation (A. . 0). and in the wall of the tube are ma ˙ a a ca (Tin − Tout ) = ho 2πro (Ta − Tt ). t) = Tw (0. t) = Tin . The governing equations on the outside of the tube. It is also seen that the outlet water temperature is hard to control for large water flow rates. . and the flow rates ma and mw can all be used as control ˙ ˙ w a inputs to obtain a desired outlet temperature.10 Compressible flow Thermal control A cross-flow heat exchanger model. 17.9 17. .8 17. . and Tt (ξ. and the outlet temperature is bounded. This example includes all the conductive.6) y(x) = a0 + a1/2 x1/2 + a1 x + a3/2 x3/2 + a2 x2 + .10) ). Water is the in-tube and air the over-tube fluid in the heat exchanger. ρw and ρt are the water and tube material densities.1 Correlations Least squares method n Possible correlations are y(x) = k=0 ak xk (17.8) and (17. and kt is the thermal conductivity of the tube material. cw and ct are the specific heats of the air.8.9) ∂t ∂ξ 2 2 2 ∂ Tt 2 2 ∂Tt = kt π(ro − ri ) 2 + 2πro ho (Ta − Tt ) − 2πri hi (Tt − Tw(17. The flow rates present a special difficulty. Tout or Tout . ri and ro are the inner and outer radii of the tube.1). Tt (ξ. ρt ct π(ro − ri ) ∂t ∂ξ respectively.2 shows the steady-state range of values of Tout that can be achieved on varying mw . w 17. In addition. Ta (t) is the air temperature surrounding the tube.4) (17. ˙ ˙ a a Tin and Tout (t) are the inlet and outlet air temperatures.8. 17. a w The inlet temperatures Tin and Tin . hi and ho the heat transfer coefficients in the inner and outer surfaces of the tube. L is the length of the tube. t) = Tw (L. schematically shown in Fig. has been studied using finite differences [4–6]. (17.3 shows the results of w applying PI control on mw to obtain a given reference temperature Tout = 23◦ C.9).3) (17. ma (t) and mw (t) are the mass flow rates of air and water. ca . advective and convective effects discussed before. As an example of control dynamics.1. Fig. Suitable numerical values were assumed for the computations. a a the air temperature is assumed to be Ta = (Tin + Tout )/2. temperatures ˙ outside this range cannot be obtained. (17.7) 17. in the water. t) are the tube-wall and water temperatures. y(x) = axm + bxn + . (17.5) (17.8) L ∂Tw 2 ∂Tw ρw cw πri + mw cw ˙ = hi 2πri (Tt − Tw ). 0) = Tw (ξ.

2 0.1 0.3 .1 23 0.2 0.4 a Tout[°C] 23.6 23. ma = 23.5 23.9 0 0. 23.8 0.23 kg/s 23.6 0. Thermal control 160 Air in Water in ξ Water out L Air out Figure 17.9 1 .7 0.4 0. . ma [kg/s] a ˙ ˙ Figure 17.2: The relation between Tout and mw for different ma [4].5 0.10.3 0.1: Schematic of single-tube cross-flow heat exchanger.17.17 kg/s 22.

each control method works differently and are labeled VF. For most practical thermal systems. Mathematical models of the dynamics of a piping network lead to differential-algebraic systems [61]. The momentum equation governing the flow in each pipe is differential. it turns out that only certain flow rates may be controllable. we will look first at thermal components and then combine them in networks.000 100 22.11 Control of complex thermal systems The previous sections examined systems that were fairly simple in the sense that mathematical models could be written down and their behaviors studied. 17. Fig. the others being dependent on these. the leaving water temperatures Tw (t). a secondary loop and a bypass that has the three strategies as special cases. 17. while the conservation of mass condition at each of the junctions is algebraic.5 and 17. Figs. In the present context this means that a complete understanding of the behavior of components does not necessarily mean that large networks formed out of these components can be modeled and computed in real time for control purposes.1 Hydronic networks The science of networks of all kinds has been put forward as a new emerging science [15].5 5 a Tout[°C] 24 Kp = 100.11. while the secondary has a water-air cooling coil which serves as a thermal load. 17. and Vγ to control L the air temperature leaving the cooling coil.17. Thus. There are at present many different strategies for the thermal control of networks. and the bypass pressure difference ∆pbp to step . Controllability issues of heat exchanger networks are reported in [203]. Vβ . 17. The primary loop includes a chiller. 62]. Ta (t).11.4 shows a network in which three specific control strategies can be compared [61.5 23 10.5 22 21. this is difficult to do with any degree of precision. MCF and BT in the following figures (details are in [61]).000 1. and comparative studies based on mathematical models can be carried out. Control of complex thermal systems 161 25 24.3: Behavior of Tout as a function of time for Ki = 50 and different Kp [4].000 23.6 show the dynamic response L L of Ta (t). The network has a primary loop. In the following. Integral controllers are used to operate the valves Vα .5 0 200 400 600 800 1000 1200 1400 1600 1800 2000 t [s] a Figure 17.

the system is becomes unstable and the variables oscillate in time. stabilization and control of convection in horizontal fluid layers [18. changes in the air velocity over the coil. the configuration is shown in Fig. 205. The input voltages v7 .7. water flow to HXBT and HXCF is zero when testing VF. that is. α(t). Each curve represents one independent run.11. and so on. β(t). The system is taken to the nominal operating conditions. Agent-based controls have been proposed by complex thermal systems such as in buildings [213]. thermal radiation [143]. 215. and materials processing and manufacturing [55. 160]. 125. 83. The hot water flow is diminished by changing its controller set point.4: Network used to study control strategies [61]. 17. and then the hot water flow is decreased by a constant value every 1800 s. v4 and v1 that control flow and the hot water temperature at the heat exchanger inlet T21 are also shown. 17. Control of convection is an important and active topic. 217]. . steady values.8 shows the secondary hot water flow rates q8 .17. q7 and q4 to the heat exchangers for the three different control strategies. this includes the study of convection loops [175. 176. It is seen that for certain control parameters. microwave heating [120]. and in porous media [189].2 Other applications There are a large number of other thermal problems in which control theory has been applied. 190–195]. 145.11. Figure 17. The controls drive the system to different operational points while coping with the changes. There are some oscillations in all the variables before they settle down to stable. Control of complex thermal systems 162 B c P2 1−B P3 a P1 i b g d T wE T aL T aE V α CC ch Vγ h f e T wL Vβ Figure 17. Laboratory experiments with a network of water-to-water heat exchangers have been reported in [61–63]. and γ(t) are the respective closing fractions of the valves which change dynamically in response to the error signal.

11. − ∗ − method VF.5 0 ∆p bp(N/m 2 ) 3 2 1 0 1 x 10 4 100 125 150 175 t 200 225 250 275 300 100 125 150 175 t 200 225 250 275 300 γ TaL (oC) 0.17. L Tw (oC) 7 6 5 1 100 125 150 175 t 200 225 250 275 300 α 0. Ta = 30 C [61]. and E L − £ − method MCF.set = 20000 (N/m ). Ta.5 0 27 26 25 100 125 150 175 t 200 225 250 275 300 100 125 150 175 200 225 250 275 300 t (s) L Figure 17.set = ◦ ◦ 2 E 5. Tw. ∆pbp.6: Dynamic response of control system to drop in air velocity with method BT.5: Dynamic response of control system to drop in air velocity. .5 0 75 100 125 150 175 200 225 250 275 300 t (s) Figure 17. Control of complex thermal systems 163 27 TaL (oC) 26 25 75 100 125 150 175 200 225 250 275 300 t (s ) 1 β 0.5 C.set = 26◦ C. Ta = 30◦ C [61].

The dashed vertical lines are instants at which the thermal load is changed [61]. 11 3 3 q1 x 10 −4 q8 . − ∗ −V F .11.v7 (Volts) 2 1 0 0 1000 2000 3000 4000 5000 6000 40 T21(oC) 35 30 0 1000 2000 3000 4000 5000 6000 t (s) Figure 17. .7: Layout of hydronic network [61].q7 .v4 . − £ −CF .17.8: Secondary hot water flows and T21 : − ◦ −BT .q4 (m3/s) 3 2 1 0 3 0 1000 2000 3000 4000 5000 6000 v1 . Control of complex thermal systems 164 PID P 1 HT1 HT2 T21 ∆p V10 10 q4 T8 HX BT 7 4 q7 T14 HX CF 1 q8 T 18 HX VF T7 T6 T5 q3 T12 T12 T11 q6 T17 T16 T15 M4 M3 Control Valve 8 5 Pump P2 6 Manual Valve ∆p V P 3 M5 q2 q5 T9 9 8 2 M6 q9 T19 ∆p V 13 T1 P4 HX m 12 q1 T3 ∆p P 4 T4 M1 M2 T10 T20 T2 Figure 17.

The governing equations cover a wide range of possibilities. To cite one specific example. Phenomena such as diffusion. control theory itself is a vast subject. it is known that unexpected synchronization may result even when multiple dynamical systems are coupled weakly [181].12. however. Problems 1. and stochastic control [37] that are well developed. with specialized branches like optimal [79]. The study of thermal control will continue to grow from the point of view of fundamentals as well as engineering applications. The fundamental ideas in this subject are firmly grounded on the mathematics of systems and control theory which should be the starting point. There are many outstanding problems and issues that need to be addressed. nonlinear and poorly predictable dynamically. Furthermore. robust [38]. convection and advection are common and the systems are usually complex. a few aspects that are particularly characteristic of thermal systems. from ordinary and partial differential equations to functional and differential-algebraic systems. There are.12 Conclusions This has been a very brief introduction to the theory of thermal control. Many of the tools in these areas find applications in thermal systems. Conclusions 165 17. This is a problem .17. It is hoped that the reader will use this brief overview as a starting point for further study and apply control theory in other thermal applications. networking between a large number of coupled components will become increasingly important.

Chapter 18 Soft computing 18. both are effective.1 shows the test facility in which the experiments were conducted.2.2 shows the results of using neurocontrol compared with PID.2 18. after that the neurocontroller brings the system back to normal operation. 220] and applications [216] are available. [51. 49–52]. 139. 18. The objective is to control a the outlet air temperature Tout . An approach that is becoming popular in these cases is that of artificial neural networks (ANN) [73] for prediction of system behavior both for time-independent [48. 166] [162] Problems 1. in thermal systems this may come from effects such as the presence of fouling over time or from changes in system configuration as could happen in building heating and cooling systems. Figs. and then the water flow is shut down between t = 40s and t = 70s. The a heat exchanger is stabilized at Tout = 36◦ C. It is particularly suitable for systems for which experimental information that can be used for training is available. 153. 182]. 168] and other soft control methodologies [32. 142] and time-dependent operation. Reviews of artificial neural network applications to thermal engineering [163. 140. 18. which are generally very complex in that they cannot be realistically computed in real time for control purposes [92.1 Artificial neural networks Heat exchangers The most important of the components are heat exchangers. This is a problem 166 .1 [141] Genetic algorithms 18. 18. 131. A stabilized neurocontrol technique for heat exchangers has been described in [47. Fig.3 shows the result of an disturbance rejection experiment. Fig. A neural network-based controller is able to adapt easily to changing circumstances.

5 ANN PID PI a T [ C] out 34 33.2: Time-dependent behavior of heat exchanger using ANN.2. PID and PI control methodologies [47].5 36 35.18.2 ANN PID a T [ C] out 36 35. a (b) heat exchanger with water and air flows indicated.4 PI t [s] 100 120 140 160 180 200 220 240 260 280 300 36.5 0 50 100 150 200 250 300 350 t [s] Figure 18.1: Experimental setup: (a) heat exchanger test facility with wind tunnel and in-draft fan.5 35 34. Artificial neural networks 167 a Tout Air flow Water flow (a) (b) Figure 18. .8 35.5 32 31. 36.4 36. Tout is the air outlet temperature [47].6 35.5 33 32.

5 0 t [s] 0.4 0.2 0.18.3: Time-dependent behavior of heat exchanger with neurocontrol for disturbance rejection experiment showing flow rates and air outlet temperature [47]. .5 a T [ C] out 36 35.2. Artificial neural networks 168 mw [kg/s] 300 200 100 0 0 50 100 150 200 250 300 350 400 t [s] 36.3 0.1 0 50 100 150 200 250 300 350 400 t [s] Figure 18.5 m a [kg/s] 0.5 35 50 100 150 200 250 300 350 400 34.

Part VI Appendices 169 .

170 . Surface of size l × l: Nǫ = (l/ǫ)2 . we present here the Hausdorff-Besicovitch dimension D. Point: Nǫ = 1. in that they are fractal within a range of length scales. D = 1 3.1 [60] Fractals are objects that are not smooth. A function f (x) is invariant under change of scale if there exists constants a and b.e. Volume of size l × l × l: Nǫ = (l/ǫ)3 . they are geometrical shapes in which the parts are in some way similar to the whole. 1.e. such that [199] f (ax) = bf (x) (A. This self-similarity may be exact. Though there are many definitions in current use. Many physical objects are fractal-like. may look exactly like the whole fractal. If there are Nǫ units of a measuring stick of length ǫ. if magnified. any part of it cannot be of finite length) and homogeneous (i. D = 2 4.1) A fractal curve must be nowhere rectifiable (i.e. If Nǫ is the number of ‘boxes’ of side ǫ needed to cover an object. D = 3 A fractal has a dimension that is not an integer. where D is the dimension. Before discussing examples we need to put forward a working definition of dimension. then D = lim ǫ→0 Fractals ln Nǫ ln(1/ǫ) (A. Coastlines are among the geographical features that are of this shape. a piece of the fractal.Appendix A Mathematical review A. any par6 is similar to the whole).2) We can check that this definition corresponds to the common geometrical shapes. i. Line of length l: Nǫ = l/ǫ. the measured length of the coastline will be of the power-law form ǫNǫ = ǫ1−D . D = 0 2.

Repeat the process to get k = 2. The middle third of each side of the triangle is removed. . This is shown as k = 1.4 Weierstrass function ∞ 2t for 2(1 − t) for 0 ≤ t ≤ 1/2 1/2 < t ≤ 1 (A.its dimension is D = ln 2/ ln 3 = 0. A. In terms of this function we can also define t D(t) = that is called the devil’s staircase. The process is continued. and g(t) is the periodic triangular function g(t) = defined on [0. Since Nǫ = 2k and ǫ = 1/3k .63 .1.1 Cantor set Consider the line [0.1. what is left is called the Cantor set. Take away the middle third to leave the two portions.. and two sides of a triangle drawn on that. closed curve that is nowhere smooth. and in the limit gives a continuous.1..6) . A. 3. Fractals 171 k=0 k=1 k=2 k=3 Figure A.A.. its integral over the interval is unity. . A. If we define a function Ck (t) at the kth level so that Ck (t) = (3/2)k if t belongs to the set and zero otherwise. If k → ∞.1. A.1: The Cantor set.2 as k = 0.26 .1] corresponding to k = 0 in Figure A. .3 Knopp function ∞ This is the function K(t) = 2−nH g(2n t) n=0 (A.3) Here we start with an equilateral triangle shown in Figure A.4) where 0 < H < 1. . the dimension of the Koch curve is D = ln 4/ ln 3 = 1. this is shown as k = 1. Since Nǫ = 3 × 4k and ǫ = 1/3k .1.5) This is the function W (t) = ω −nH cos (ω n t + φn ) n=0 (A. .1. .1].2: The Koch curve. Figure A. .2 Koch curve 0 k→∞ lim Ck (t′ ) dt′ (A.

A.2. Perturbation methods

172

Figure A.3: Mandelbrot set where a is real, b is odd, and ab > 1 + 3π/2. It is everywhere continuous, but nowhere differentiable. The related Weierstrass-Mandelbrot function

Wm (t) =
n=−∞

ω −nH (1 − cos ω n t)

(A.7)

satisfies the invariance relation A.1. A.1.5 Julia set

An example of this comes from the application of Newton’s method to find the complex root of the equation z 3 = 1. In this method the following iterative scheme is set up: zk+1 = zk −
3 zk − 1 2 3zk

(A.8)

Each one of the three roots has a basin of attraction, the boundaries of which are fractal. A.1.6 Mandelbrot set

This is the set of complex numbers c for which
2 zk+1 = zk + c

(A.9)

stays bounded as k → ∞. The boundaries of this set shown in Figure A.3 are again fractal.

A.2 A.3

Perturbation methods Vector spaces

Functional analysis, norms, inner product, complete space, Banach and Hilbert spaces, operators, eigenvalues problem, adjoint and self-adjoint operators. Finite-dimensional spaces, linear algebra. Eigenfunction expansions. [13, 78, 107]

A.4

Dynamical systems

A dynamical system is a set of differential equations such as dxi = fi (x1 , x2 , . . . , t; λ1 , λ2 , . . . , λp ) for i = 1, . . . , n dt (A.10)

The x1 , . . . , xn s are state variables and the λ1 , . . . , λp are bifurcation parameters. The mapping f : X × Rp → Y is a vector field. If f1 , . . . , fn do not depend on time t, the system is autonomous.

A.4. Dynamical systems

173

A nonautonomous system can be converted into autonomous one by the change in variable xn+1 = t, from which we can get the additional equation dxn+1 =1 dt (A.11)

We will assume that the solutions of the system are always bounded. The system is conservative if the divergence of the vector field ∂fi /∂xi is zero, and dissipative if it is negative. An attractor of a dissipative dynamical system is the set of {xi } as t → ∞. The critical (or singular, equilibrium or fixed) points, xi , of equation (A.10) are those for which fi (x1 , x2 , . . . , t; λ1 , λ2 , . . . , λp ) = 0 (A.12)

There may be multiple solutions to this algebraic or transcendental equation. Defining a new coordinate x′ = xi − xi that is centered at the critical point, we get the local i form dx′ i (A.13) = fi (x1 − x1 , . . . , xn − xn ) dt Sometimes we will use the notation dxi = g(x1 , . . . , xn ) dt to indicate the local form, the origin being one critical point of this system. A.4.1 Stability (A.14)

The stability of the critical points is of major interest. A critical point is stable if, given an initial perturbation, the solutions tends to it as t → ∞. Linear stability The vector field in equation (A.14) can be expanded in a Taylor series to give dxi = dt The eigenvalues of the Jacobian matrix A= ∂gi xj + . . . ∂xj
0

(A.15)

j

∂gi ∂xj

(A.16)

determine the linear stability of the critical point. The critical point is stable if all eigenvalues have negative real parts, and unstable if one or more eigenvalues have positive real parts. A.4.2 Routh-Hurwitz criteria a0 sn + a1 sn−1 + . . . + an−1 s + an = 0 has roots with negative real parts if and only if the following conditions are satisfied: (i) a1 /a0 , a2 /a0 , . . . , an /a0 > 0 (ii) Di > 0, i = 1, . . . , n

The polynomial equation

A.4. Dynamical systems

174

The Hurwitz determinants Di are defined by D1 D2 D3 = a1 a1 = a0 = a1 a0 0 a1 a0 0 0 . . . 0 with ai = 0, if i > n. Global stability Consider the dynamical system in local form, equation (A.14). If there exists a function V (x1 , . . . , xn ) such that V ≥ 0 and dV /dt ≤ 0, then the origin is globally stable, that is, it is stable to all perturbations, large or small. V is called a Liapunov function. [56] A.4.3 Bifurcations a3 a2 a3 a2 a1 a3 a2 a1 a0 . . . 0 a5 a4 a3 a5 a4 a3 a2 . . . 0 ... ... ... ... . . . ... a2n−1 a2n−2 a2n−3 a2n−4 . . . an

Dn

=

The critical point is one possible attractor. There are other time-dependent solutions which can also be attractors in phase space, as indicated in the list below. • Point (steady, time-independent) • Closed curve (limit cycle, periodic) • Torus (periodic or quasi-periodic) • Strange (chaotic) For a given dynamical system, several attractors may co-exist. In this case each attractor has a basin of attraction, i.e. the set of initial conditions that lead to this attractor. A bifurcation is a qualitative change in the solution as the bifurcation parameters λi are changed. A.4.4 One-dimensional systems dx = f (x) dt Example A.1
Consider the linear equation dx = ax + b dt (A.18)

A one-dimensional dynamical system is of the type (A.17)

A.4. Dynamical systems

175

The critical point is x=− On defining x′ = x − x, the local form is obtained as dx′ = ax′ dt We can take one of two approaches. (a) Solving (A.20) b a (A.19)

x′ = x′ eat 0

(A.21)

The critical point is a repeller if a > 0, and an attractor if a < 0. (b) Alternatively, we can multiply equation (A.20) by 2x′ to get dV = 2aV dt (A.22)

where V = x′2 . Since V is always nonnegative, the sign of dV /dt is the sign of a. Thus V will increase with time if a > 0, and decrease if a < 0. In either case we find that the critical point is unstable if a > 0 and stable if a < 0.

Example A.2
The nonlinear equation ˆ ˜ dx = −x x2 − (λ − λ0 ) dt has critical point which are solutions of the cubic equation ˜ ˆ x x2 − (λ − λ0 ) = 0 x(1) x(2) x(3) = = = 0 p λ − λ0 p − λ − λ0 (A.23)

(A.24)

Thus

(A.25) (A.26) (A.27)

where x(i) (i = 1, 2, 3) are the three critical points. The bifurcation diagram is shown in Fig. A.4. (i) Critical point x(1) = 0 To analyze the local stability of x(1) = 0, we obtain the local form dx′ = x′ (λ − λ0 ) − x′3 dt Neglecting the cubic term x′3 , this becomes dx′ = x′ (λ − λ0 ) dt Thus x(1) = 0 is locally stable if λ < λ0 , and unstable if λ > λ0 . To analyze the global stability, equation (A.28) can be written as 1 dV = V (λ − λ0 ) − V 2 2 dt where V = x′2 . For λ < λ0 , V ≥ 0, dV /dt ≤ 0, so that x(1) = 0 is globally stable. (ii) Critical point x(2) = √ λ − λ0 (A.30) (A.29) (A.28)

4.7: Saddle-node bifurcation.32) A. A. √ (iii) Critical point x(3) = − λ − λ0 This is similar to the above. Figure A. • Subcritical: Fig. A. Linearizing. A.4 x stable stable stable unstable λ unstable S-N unstable λ Figure A. • Saddle-node: Fig.6.A. (A. • Transcritical: Fig. Dynamical systems 176 x stable x stable stable S-N stable λ 0 unstable λ linearly stable stable stable unstable unstable unstable λ unstable stable Figure A.31) dx′ = −2(λ − λ0 )x′ dt so that this critical point is linearly stable.4.5: Subcritical version of Fig.4. .5 Examples of bifurcations • Supercritical: Fig.7.4: Supercritical pitchfork bifurcations x Figure A.5. A.6: Transcritical bifurcation. A. we get The local form of the equation around this critical point is – “p ” »“p ”2 dx′ =− λ − λ0 + x′ λ − λ0 + x′ − (λ − λ0 ) dt (A.

y) = fy (x. we get f (x) = −x x2 − (λ − λ0 ) + ǫ The critical points are solutions of x3 − (λ − λ0 )x − ǫ = 0 To see the nature of the curve. (a) ǫ > 0.37) Figure A.4.35) (A.34) (A. The dynamical system without ǫ is thus structurally unstable. we get λ′ x′ = −ǫ (A.7 Two-dimensional systems = λ0 + λ ′ (A.39) .8: Imperfect bifurcations. (b) ǫ < 0. Dynamical systems 177 x stable stable unstable λ0 stable λ x stable stable λ λ0 unstable stable Figure A. x) (A.A. A.4.4.33) (A.23).8 shows the result of adding the imperfection ǫ. λ = λ0 and write x = x′ λ For small x′ and λ. we make an expansion around x = 0. A.36) Consider dx dt dy dt = fx (x.38) (A.6 Unfolding and structural instability Adding a small constant to the vector field in equation (A.

the eigenvalues are complex and the solution in phase space is a spiral.46) Example A. .41) 1 −P ± 2 P 2 − 4Q (A.45) axx ayx axy ayy (A. and the solutions are damped oscillations. if in addition P > 0. the system is dissipative. and if P < 0. and periodic if λ > λ0 . the spiral is stable. it is unstable. Example A. Dynamical systems 178 The linearized equation in local form has a Jacobian matrix A= The eigenvalues satisfy a quadratic equation λ2 + P λ + Q = 0 from which λ= The sign of the discriminant D = P 2 − 4Q (A.40) (A.4.4 dx dt dy dt = = y −ω 2 x − σy (A.42) This is a conservative system which is equivalent to d2 x + (λ − λ0 )x = 0 dt2 The solutions are exponential if λ < λ0 . If D < 0.48) For σ > 0. The occurrence of periodic solutions in two-dimensional systems. permits a Hopf bifurcation. D > 0.3 dx dt dy dt = = y −(λ − λ0 )x (A. on the other hand. If. the solutions do not oscillate in time but move exponentially towards (if all eigenvalues are negative) or away (if at least one eigenvalue is positive) from the critical point.43) determines the nature of the solution.44) (A.A. (A. which is a transition from a time-independent to a periodic behavior through a pair of complex conjugate imaginary eigenvalues.47) (A. the eigenvalues are real.

i.8 Three-dimensional systems Forced Duffing equation Since a non-autonomous equation can be converted into a three-dimensional autonomous system. we will include the case of the forced Duffing equation here1 . dx dt dy dt 1 Sometimes = y = x − x3 + f (t) (A. . at which dr/dt = 0.56) This is a Hopf bifurcation at λ = λ0 ..55) (A.4.53) (A.50) can be converted to polar coordinates. a similar analysis of equation (A.52) which simplifies to = = ` ´ r λ − λ0 − r 2 (A.49) and (AHopftwo) shows that the origin is stable for λ < λ0 .58) defined with a negative sign in front of x in the second equation.e r = 0 and r = λ − λ0 . at which point the solution goes from time-independent to periodic. The first is a critical point at the origin.51) (A. There is thus a Hopf bifurcation at λ = λ0 .5 The dynamical system dx dt dy dt = = (λ − λ0 )x − y − (x2 + y 2 )x x + (λ − λ0 )y − (x2 + y 2 )y (A. A linear analysis of equations (A.57) (A. and the second a circular periodic orbit that exists only for λ > λ0 .4. Dynamical systems 179 Example A. Example A. Substituting x = r cos θ and y = r sin θ. we get −r sin dr dθ + cos θ dt dt dθ dr r cos + sin θ dt dt dr dt dθ dt = = (λ − λ0 )r cos θ − r sin θ − r 3 cos θ r cos θ + (λ − λ0 )r sin θ − r 3 sin θ (A.49) (A.53) √ indicates that r = λ − λ0 is a stable orbit.6 dx dt dy dt = = y ` ´ −ω 2 x − σ λ − x2 − y 2 (A.54) 1 √ There are two values of r. A.A. For λ > λ0 .

the origin becomes unstable. (b) x = y = b(λ − 1).A.66) . (a) x = y = z = 0 Small perturbations around this point give  ′   −σ x d  ′   y λ = dt z′ 0 σ −1 0 The characteristic equation is  ′  0 x 0   y′  z′ −b (A. 0 z    b(λ − 1) − b(λ − 1) b(λ − 1)  .65) λ3 + (σ + b + 1)λ2 + (σ + λ)bλ + 2σb(λ − 1) = 0 (A.64) from which we get the eigenvalues −b.61) where σ and b are taken to be positive constants. 1 [−(1 + σ) ± (1 + σ)2 − 4σ(1 − λ)]. z = λ − 1 Small perturbations give  ′   x d  ′   y = dt z′ The characteristic equation is −σ 1 b(λ − 1) σ −1 b(λ − 1)  ′  0 x − b(λ − 1)   y ′  z′ −b (A. The bifurcation parameter will be λ.62) A linear stability analysis of each critical point follows.60) (A. For 0 < λ < 1. For λ > λ1 . At λ = λ1 . where λ1 = 1.  − b(λ − 1)  λ−1 λ−1 (A.63) (λ + b)[λ2 + λ(σ + 1) − σ(λ − 1)] = 0 (A. since (1 + σ)2 > 4σ(1 − λ).4. there is a pitchfork bifurcation with one zero eigenvalue. Dynamical systems 180 Lorenz equations An important example is the Lorenz equations: dx dt dx dt dx dt = σ(y − x) = λx − y − xz = −bz + xy (A. with σ > b + 1. The critical points are obtained from y−x = 0 λx − y − xz = 0 −bz + xy = 0 which give      x 0  y  =  0 .59) (A. the 2 eigenvalues are real and negative.

E s and E c .5.66) can be factorized to give the eigenvalues −(σ + b + 1). are tangents. (c) Re(λ) = 0 with the generalized eigenspace E c . i.67) At λ = λ3 the characteristic equation (A. The Hurwitz determinants are D1 D2 = σ+b+1 σ + b + 1 2σb(λ + 1) = 1 (σ + λ)b = σb(σ + b + 3) − λb(σ − b − 1) σ + b + 1 2σb(λ − 1) 0 1 (σ + λ)b 0 = 0 σ + b + 1 2σb(λ − 1) = D3 2σb(λ − 1)[σb(σ + b + 3) − λb(σ − b − 1)] Thus the real parts of the eigenvalues are negative if λ < λ3 . respectively. The periodic solution which is created at this value of λ can be shown to be unstable so that the bifurcation is subcritical. becomes unstable at λ = λ1 . corresponding to a Hopf bifurcation. There exist manifolds W u . • Just below λ3 . W s .e. The type of bifurcations that occur with a single bifurcation parameter λ has been discussed in the previous . stable and center manifolds. the two critical points are stable to small perturbations. W u .e. where λ3 = σ(σ + b + 3) σ−b−1 (A.9 Nonlinear analysis Center manifold theorem [30] Consider a vector field fi (x) with fi (0) = 0. W s .A. in the range λ2 < λ < λ3 .5 Singularity theory We have seen that the critical points of a dynamical system. • For λ > λ3 . The eigenvalues λ of ∂fi /∂xj at the origin are of three kinds: (a) Re(λ) > 0 with the generalized eigenspace E u .10). A. • Two other critical points are created for λ > λ1 . respectively. these are linearly stable in the range λ1 < λ < λ3 . by finding the singularities of the function fi . Singularity theory 181 Using the Hurwitz criteria we can determine the sign of the real parts of the solutions of this cubic equation without actually solving it. and W c to which E u . and W c are the unstable. are found by solving an equation of the type (A. Here is a summary of the series of bifurcation with respect to the parameter λ: • Origin is a stable critical point for λ < λ1 . and ±i 2σ(σ + 1)/(σ − b − 1). equation (A. all initial conditions produce chaos (except for periodic windows). but for large enough perturbations produce chaos. A.12). i. (b) Re(λ) < 0 with the generalized eigenspace E s .4.

71) .4 and µ = ǫ will give Fig. such that Lφi = λi φ. t) and ∂T = L(T ) (A. (A. Let φi (x) be the eigenfunctions of L. A.6. . Here we increase the number of parameters to λ1 . Example A.6 Partial differential equations Classification Boundary conditions A. A. There are three real solutions if the discriminant D = > 0. systems in which fI = ∂φ/∂xi . . λp and look at the consequent changes in the bifurcations. Figure A. . section. there are only seven. µ) surface in the (λ.70) ∂t in a domain Ω.69) can be classified in terms of eleven canonical surfaces. i. Partial differential equations 182 µ One solution Three solutions λ x One solution µ λ Figure A. .68) 4λ3 + 27µ2 Figure A. A. λp at which equation (A. A bifurcation (or catastrophe) set is the set of points in parametric space λ1 . A section of this surface at µ = 0 will give Fig. The quadratic surface a1 x2 + a2 y 2 + a3 z 3 + a4 xy + a5 xz + a6 yz + a7 x + a8 y + a9 z + a10 = 0 (A. where L is a linear operator with spatial partial derivatives. . µ) coordinates. . It is projection of the x = x(λ.7 Examine the one-dimensional vector field f (x) = −(x3 + px + q) (A.6.9: Surface x = x(λ.10 in (λ. The boundary and initial conditions are T = 0 at ∂Ω and T = f (x) at t = 0. µ) plane.10: Bifurcation set.9 shows the surface x = x(λ. The bifurcation set is shown in Fig.8. .12) is satisfied. µ).A. . For gradient systems. and only one otherwise. A. µ).1 Eigenfunction expansion Let T = T (x.e.

φj .80) vg = dk Problems 1.77) A wave can be of the form [208] If we follow a point at constant phase φ = kx − ωt = k(x − ωt/k). Carry out an imperfection analysis on the transcritical bifurcation above. 2. 4. i=1 (A.A. dt (A. (A. and is given by dω . Thus φi . (A. 3. we will be moving with the phase velocity vp . t) = i=1 ai (t)φi (x). we get ∞ i=1 dai φi = dt ∞ ai (t)L(φi ). so that the λi are real and the φi are orthonormal. Show that dx = −x2 + (λ − λ0 ) dt has a saddle-node bifurcation. Using eq.75) A. φj = δij . The group velocity is the velocity at which the energy of the wave moves. we get daj = λj aj .71).73) Taking the inner product with φj .79) is called the dispersion relation. Investigate the bifurcations in ` ´ dx = −x x4 − 2x2 + 2 + λ dt dx = −x [x − (λ − λ0 )] dt .76) (A.78) k The relation between the frequency ω and the wave number k ω = ω(k) (A. φj = dt ∞ i=1 ai (t) L(φi ). where ω vp = .7 Waves f (x.72) Substituting in the equation. (A. t) in the form ∞ T (x. Waves 183 Expand T (x. (A. (A. Show that has a transcritical bifurcation.74) At this point we restrict ourselves to the special case in which L is a self-adjoint operator.7. (A. t) = Aei(kx−ωt) . we find that ∞ i=1 dai φi .

196] Method of weighted residuals Let us apply the following numerical methods to the one-dimensional fin equation d2 T −T =0 dx2 with the boundary conditions T (0) = TL and T (1) = TR . . each of length ∆x = 1/N . −1 c where c = 2 + ∆x2 .1. . . .2) where Ti = T (xi ).   . B. T2 . we get −Ti+1 + (2 + ∆x2 )Ti − Ti−1 = 0 (B. . 1] into N intervals. . . 96. .             TR TN −1 0 0 . we get N − 1 algebraic equations in the unknowns T1 . . . (B.1 Finite difference methods Divide [0. . . (B. B. 2. 0  T1   TL           −1 c −1 0 .   . N so that xi = i∆x. . so that from Eq.   T2   0            0 −1 c −1 . . . 2. .1) B. TN −1 . 171.  . . . 1. Thus we have      c −1 0 . . . . N − 1. . . . 126. . . . 97. . 146.4) =   .3) at i = 1. .Appendix B Numerical methods [33.2 [161] Finite element methods 184 . . .  . and number the nodes as i = 0. . . Write the second derivative at x = xi as T ′′ (xi ) = 1 (Ti+1 − 2Ti + Ti−1 ) ∆x2 (B. Applying the boundary conditions.  T     N −2   0   . . . which can be solved.

Notice that φ1 (0) = φ2 (∆x) = 1 and φ1 (∆x) = φ2 (0) = 0. 1] into N intervals or finite elements. . 2. . . Number the nodes as i = 0. where Ti−1 = Tx=xi−1 = Tξ=0 and Ti = Tx=xi = Tξ=∆x .2: Finite element. (and multiplying by −∆x for convenience) we . Once again. we have the two equations ′ −Ti−1 − Ti−1 Ti′ + Ti−1 ∆x 1 + ∆x 3 1 ∆x − ∆x 6 + Ti − Ti 1 ∆x − ∆x 6 1 ∆x + ∆x 3 = = 0. where k = 1. Using the Galerkin method ∆x 0 d2 T −T dξ 2 φk (ξ) dξ = 0. Doing this for each of the N element. respectively. 0. N so that xi = i∆x. 2.5) (B. Figure B. 1. divide [0.6) within element i. For k = 1 and k = 2.B. Finite element methods 185 i = 0 1 2 N ξ=0 ξ ξ=∆x x Figure B.2. respectively. . Integrating by parts ∆x ∆x T φk 0 ′ − 0 dT dφk + T φk dξ dξ dξ = 0. Let us use the linear test functions φ1 (ξ) φ2 (ξ) = = 1− ξ ∆x ξ ∆x (B. each of length ∆x = 1/N . In each element use a local coordinate ξ that goes from ξ = 0 to ξ = ∆x and correspond to the global coordinate x = xi and x = xi+1 = xi + ∆x. where T ′ = dT /dξ.1: Finite differencing. so that we can write T (ξ) = Ti−1 φ1 (ξ) + Ti φ2 (ξ).

15) For the following.  T     N −2   0   .             c2 TR TN −1 0 0 . . 0  T1   c2 TL           −c2 c1 −c2 0 . −c2 c1 −c2 . .  . c2 = 1 − ∆x2 /6.8) (B.9) cancels the T1 term..16) (B. .B.14) =  . it can be noticed that adding Eqs. . . .9) (B. The dependent variable is expanded in terms of orthonormal functions φn (x) that satisfy the boundary conditions. . .12) (B. (B.13) There are 2N unknowns including T and T ′ at each of the N + 1 node minus the values of T0 and TN at the two ends that are known. However.18) . −c2 c1 where c1 = 2(1 + ∆x2 /3).   T2   0            0  .8) and (B. . . (B.   . (B.  . . . .3. . add the rest in pairs to get the reduced set      c1 −c2 0 . 0 0 0 0 (B. So we discard the first and last equations. Thus N θ(x) = n=1 an φn (x). . so that we can solve for the unknowns at this stage if ′ we wish.17) Substituting in the differential equation gives the residual N an n=1 d 2 φn − φn dx2 − g = ǫ. let so that with the homogeneous boundary conditions θ(0) = θ(1) = 0.11) ′ ∆x TN −1 + (1 + ∆x ∆x )TN −1 − (1 − )TN 3 6 2 2 ∆x ∆x ′ −∆x TN − (1 − )TN −1 + (1 + )TN 6 3 = = 0 0 (B.3 Spectral methods θ = T − TL − (TR − TL )x d2 θ −θ dx2 = TL + (TR − TL )x = g(x) (B. .10) (B.  .. . . . .7) (B. B.  . Spectral methods 186 end up with 2N algebraic equations. ∆x2 )T0 − (1 − 3 2 ∆x ′ )T0 + (1 + −∆x T1 − (1 − 6 ∆x2 ′ ∆x T1 + (1 + )T1 − (1 − 3 ∆x2 ′ −∆x T2 − (1 − )T1 + (1 + 6 ′ ∆x T0 + (1 + 2 ∆x2 )T1 6 2 ∆x )T1 3 ∆x2 )T2 6 ∆x2 )T2 3 2 = = = = . (B. .

4 MATLAB The PDE Toolbox uses a finite-element code to solve basic partial differential equations needed for conduction heat transfer.3 Chebyshev Galerkin Using ξ = 2x − 1. Solve. so that it satisfies the boundary conditions and θ(x) = aφ(x).3. Let φ(x) = x(1 − x). 1]. (B. For higher-order accuracy.20) (B. .3. B. and ψ1 (x) = 1. ψ2 (x) = x. ψm = 0 This reduces to an algebraic equation for each m. . B. .2 Trigonometric collocation Trigonometric Galerkin √ 2 sin(nπx) (B.5 Legendre Galerkin Moments 1 − ξ2. .18) to zero in N equally-spaced points.3.3. .B. B.3. φ2 (x) = x2 (1 − x). φ3 (x) = x(1 − x)2 .4 B. where the Chebychev polynomials Tn are T1 T2 T3 T4 = 1 = ξ = = . solve. take φ1 (x) = x(1 − x). transform the independent variable to the domain [−1.1 Here φn (x) = ψn (x) = with respect to the L2 inner product. MATLAB 187 Making the residual and functions ψm (x) orthogonal gives ǫ. .19) Equate the residual in Eq. This gives algebraic equations.4. . 2ξ 2 − 1 4ξ 3 − 3ξ Take ψn (ξ) = φn (ξ) and an inner product with respect to the weight function 1/ B. The orthonormal √ Chebychev functions are φn (ξ) = 2n Tn (ξ)/ 2π. . . ψ3 (x) = x2 . Take ψ(x) = 1 to determine a. B.

(e) Galerkin finite elements: Divide into N elements. and solve. assume linear functions. and solve. with T (0) = 1. (b) Trigonometric Galerkin: Expand in terms of N trigonometric functions. take inner products. (a) Finite differences: Divide into N parts. Solve using the above methods. write derivatives in terms of finite differences. T (1) = 0. In each case show convergence. substitute in equation. Consider the convective fin equation d2 T − T = 0. (f) Polynomial moments: Assume dependent variable to be a N th-order polynomial that satisfies boundary conditions. apply boundary conditions. take inner products. substitute in equation.4. (c) Chebyshev Galerkin: Expand in terms of N Chebyshev polynomials. apply collocation. Solve using the following methods. reduce to algebraic equations. Consider the convective fin equation with heat generation d2 T − T = 1. assemble all equations. reduce to algebraic equations. integrate by parts. with T (0) = 1. and solve. reduce to algebraic equations.B. dx2 where 0 ≤ x ≤ 1. MATLAB 188 Problems 1. and solve. apply boundary conditions. take inner products. and solve. T ′ (1) = 0. 2. . You may have to transform the dependent or independent variable differently for each method. substitute in equation. obtain algebraic equations by taking moments. and solve. dx2 where 0 ≤ x ≤ 1. (d) Trigonometric collocation: Expand in terms of N trigonometric functions.

2. ǫ) surfaces for the convection with radiation problem.1) (C. Two bodies at temperatures T1 (t) and T2 (t). T∞ (t). numerical solutions). 3. 5. T∞ Tmax δt Tmin t Figure C. Find (a) the long-time solution of the system temperature. 4. and (b) the amplitude of oscillation of the system temperature.2) Derive the equations above and explain the parameters. T (t). Complete the problem of radiation in an enclosure (linear stability. Consider the change in temperature of a lumped system with convective heat transfer where the ambient temperature. T (t). The temperatures are governed by M1 c1 dT1 + hAc (T1 − T2 ) + hA(T1 − T∞ ) = dt dT2 + hAc (T2 − T1 ) + hA(T2 − T∞ ) = M2 c2 dt 0 0 (C. are in thermal contact with each 189 . respectively. Plot all real θ(β. and comment on the existence of solutions. for a small period δt. Two bodies at temperatures T1 (t) and T2 (t). Find the steady-state temperatures and explore the stability of the system for constant T∞ . varies with time in the form shown.1: Ambient temperature variation.Appendix C Additional problems 1. respectively. are in thermal contact with each other and with the environment.

where x is the coordinate along the edge. 13. respectively. each one of which can be independently controlled. take the ambient temperature. Consider a rectangular fin with convection. Take the ambient temperature. Analyze an annular fin with a prescribed base temperature and adiabatic tip. unstable. Graph the results for several values of the parameters. stable. to be constant. 11. (b) 10◦ C. Assume that the base temperature is known. etc.190 other and with the environment. and (d) the optimum fin height L. Calculate numerically the evolution of an initial temperature distribution at different instants of time. The temperatures are governed by dT1 + hc Ac (T1 − T2 ) + hA(T1 − T∞ (t)) dt dT2 + hc Ac (T2 − T1 ) + hA(T2 − T∞ (t)) M2 c2 dt M1 c1 = Q1 (t) = Q2 (t) where Q1 and Q2 are internal heat generations. and (d) 10 + sin(πx) ◦ C.). then you can take it to be (a) constant. 12. Determine its fin efficiency and plot. The temperatures are governed by M1 c1 dT1 + hc Ac (T1 − T2 ) + hA(T1 − T∞ (t)) dt dT2 + hc Ac (T2 − T1 ) + hA(T2 − T∞ (t)) M2 c2 dt = Q1 (t) = Q2 (t) where Q1 and Q2 are internal heat generation sources. If you do the problem analytically. 6. The temperatures on each side are (a) 10◦ C. A plane wall initially at a uniform temperature is suddenly immersed in a fluid at a different temperature. . radiation and Dirichlet boundary conditions. but if you do it numerically. and (b) oscillatory. Write a computer program to do the previous problem numerically using finite differences and compare with the analytical results. 7. Consider a longitudinal fin of concave parabolic profile as shown in the figure. where δ = [1 − (x/L)]2 δb . Find the temperature profile as a function of time. Consider a square plate of side 1 m. δb is the thickness of the fin at the base. Find (a) the temperature distribution in the fin. Assume all parameter values to be unity. T∞ . Optimize the fin assuming the fin volume to be constant and maximizing the heat rate at the base. Choose different grid sizes and show convergence. 10. Find the steady-state temperature distribution analytically. oscillatory. Two bodies at temperatures T1 (t) and T2 (t). Find (c) the optimum base thickness δb . show numerical results for the different types of responses possible (damped. T∞ . are in thermal contact with each other and with the environment. 8. show analytical or numerical results for the temperature responses of the two bodies. Write a computer program to do the previous problem numerically using finite differences and compare with the analytical results. Using PID control. 9. Using on-off control. Neglect convection from the thin sides. and (b) the heat flow at the base of the fin. (c) 10◦ C. to be (a) constant and (b) oscillatory. each one of which can be independently controlled.

Using cubic expansions for u/u∞ and θ.7) numerically by the shooting method for different Pr and compare with results in the literature.9) (C. derive expressions for the boundary layer thicknesses.6) (C.2: Longitudinal fin of concave parabolic profile. Solve equations (C.191 δ(x) w δb x L Figure C.4) (C.6) and (C. For Problem 1. 14. derive the momentum and energy integral equations. 16. 17.8) (C. show that the governing boundary layer equations ∂u ∂v + ∂x ∂y ∂u ∂u u +v ∂x ∂y ∂T ∂T +v u ∂x ∂y = 0 ∂2u ∂y 2 (C. Consider the hydrodynamic and thermal boundary layers in a flow over a flat plate at constant temperature. Starting from the boundary layer equations ∂u ∂v + ∂x ∂y ∂u ∂u +v u ∂x ∂y ∂T ∂T u +v ∂x ∂y = 0 (C.5) = ν ∂2u ∂y 2 ∂2T = α 2 ∂y change to variables f (η) and θ(η) and derive the boundary layer equations 2f ′′′ + f f ′′ Pr ′ fθ θ′′ + 2 and the boundary conditions.7) = gβ(T − T∞ ) + ν = α ∂2T ∂y 2 .3) (C. For natural convection near a vertical plate.10) = = 0 0 (C. 15.

(c) CD cooling with heat flux qs . For parallel and counterflow heat exchangers.1 − Th. (From Incropera and DeWitt) A single-pass. The thermal conditions at the walls of the cavity are: (a) AB heating ′′ ′′ with heat flux qs . Solve equations (C. 19. having thermophysical properties similar to air. cross-flow heat exchanger uses hot exhaust gases (mixed) to heat water (unmixed) from 30 to 80◦ C at a rate of 3 kg/s.1 1 − exp{−U P 1 ± (mh ch /mc cc ) ˙ ˙ Th. estimate the required area. 21. If the overall heat transfer coefficient is 200 W/m2 K. derive the expression for the effectiveness as a function of the NTU. 18. For a counterflow heat exchanger. respectively.1 − Tc.11) and (C. ± mh ch ˙ mc cc ˙ 1 1 ± x} mh ch ˙ mc cc ˙ for the hot and cold fluids. (b) BC adiabatic. respectively.1 − Tc.12) numerically by the shooting method for different P r and compare with results in the literature. I obtained the temperature distributions Th (x) = Th. Please check. Write the governing equations for natural convection flow in an inclined rectangular cavity. C H L y x B D gravity = = 0 0 (C. 1 is the end where the hot fluid enters (from where x is measured) and 2 is where it leaves. and nondimensionalize them.1 ± (mc cc /mh ch ) ± 1 ˙ ˙ 1 1 x} .192 can be reduced to f ′′′ + 3f f ′′ − 2(f ′ )2 + θ θ′′ + 3 P r f θ′ with appropriate boundary conditions. As usual the upper sign is for parallel and the lower for counterflow.11) (C. . Find from the literature if there is any experimental confirmation of the result. The exhaust gases. Derive the Nusselt result for laminar film condensation on a vertical flat plate.1 1 − exp{−U P Tc (x) = Tc. 22. (d) DA adiabatic. and also the NTU as function of the effectiveness.12) A α 20. 23. enter and exit the exchanger at 225 and 100◦ C.

C.5.1987 C2 Also show that (Stefan-Boltzmann’s law) ∞ Eλ dλ = 0 C1 π 4 4 4 T 15C2 You can use a symbolic algebra program in this problem. T2 = 500K. respectively. 27. Consider conductive rods of thermal conductivity k joined together in the form of a fractal tree (generation n = 3 is shown in Fig. the dimension of the curve is D = 1 and the length L → ∞. four-surface enclosure shaped in the form of a tetrahedron (made of four equilateral triangles). 29. 28. The base and tip temperatures are T0 and T∞ . The heat exchanger operates with both fluids unmixed. see Fig.4. respectively. (c) What is the surface area of the heat exchanger? 25. and centered along the length of the rectangles) as an obstacle between the two rectangles. (From Incropera and DeWitt) A cross-flow heat exchanger used in cardiopulmonary bypass procedure cools blood flowing at 5 liters/min from a body temperature of 37◦ C to 25◦ C in order to induce body hypothermia. Show that the energy spectrum for blackbody radiation (Planck’s law) Eλ = λ5 C1 C2 exp λT − 1 has a maximum at λ = λm where (Wien’s law) λm T = 0. The length and cross-sectional area of . An “Aoki” curve is defined as shown in Fig. 26. C. and the overall heat transfer coefficient is 750 W/m2 K. Write a numerical code to evaluate the view factor between two rectangular surfaces (each of size L × 2L) at 90◦ with a common edge of length 2L. which reduces metabolic and oxygen requirements. Calculate the view factor again but with a sphere (diameter L/2. C. The temperatures and emissivities of three sides are T1 = 700K. Determine its temperature. center at a distance of L/2 from each rectangle. gray. ǫ1 = 0. Compare with the analytical result. 30. see Fig. Show that when n → ∞.3. (b) Calculate the water flow rate. (a) Determine the heat transfer rate for the exchanger. The coolant is ice water at 0◦ C and its flow rate is adjusted to provide an outlet temperature of 15◦ C. C. the fractal is obtained in the limit n → ∞). T3 = 300K.5 ǫ3 = 0. (From Incropera and DeWitt) Consider a diffuse.193 24.7 ǫ2 = 0.3 The fourth side is well insulated and can be treated as a reradiating surface.6. The density and specific heat of the blood are 1050 kg/m2 and 3740 J/kg K.

6: Fractal tree.4: Two rectangular surfaces with sphere.5: Aoki curve. .3: Two rectangular surfaces. n=0 n=1 n=2 Figure C.194 Figure C. T4 T3 T0 0 2 3 T2 1 T1 n=3 Figure C. Figure C.

∂x T = T1 at x = L . Using a complete basis. are in thermal contact with each other and with the environment. and T∞ is the ambient temperature. A is the convective heat transfer area. show that f (T ∗ ) has a point of inflexion at T ∗ = 1/2. Show that the conductive heat transfer through rod 0 is q= kA β (T0 − T∞ ) 1 − L 2 31. Two bodies at temperatures T1 (t) and T2 (t). The dependence of the rate of chemical reaction on the temperature T is often represented by the Arrhenius function f (T ) = e−E/T . and so on. T 2 ) and determine its stability. Plot also the L2 -error in the same range for different orders of the approximation.∞ (T2 − T∞ ) = Q2 M2 c2 dt M1 c1 where Mi is the mass. and Qi is internal heat generation. and there is heat loss by convection from the tank. show that the temperature of the tank T is determined by dT Mc = ae−E/T − hA(T − T∞ ) dt where M is the mass. 34. If e−E/T is proportional to the heat generated within a tank by chemical reaction. Nondimensionalize the equation to ∗ dT ∗ ∗ = e−1/T − H(T ∗ − T∞ ) ∗ dt ∗ For T∞ = 0. where E is the activation energy. ci is the specific heat. and those of bar 1 are 2L/β and A/β 2 . and determine the bifurcation points. a is a proportionality constant. respectively. respectively. draw the bifurcation diagram with H as the bifurcation parameter. c is the specific heat. 33. 32.195 bar 0 is L and A. h is the convective heat transfer coefficient. the Cs are thermal conductances. Plot f (T ∗ ) in the range T ∗ = 1/16 to T ∗ = 4 as well as its Taylor series approximation to various orders around T ∗ = 1/2.1. ∂T = q0 at x = 0. Find the steady state (T 1 .∞ (T1 − T∞ ) = Q1 dt dT2 + C(T2 − T1 ) + C2. where 1 ≤ β < 2. Show that the temperatures are governed by dT1 + C(T1 − T2 ) + C1. expand the solution of the one-dimensional heat equation ∂T ∂2T =α 2 ∂t ∂x with boundary conditions −k as an infinite set of ODEs. Writing T ∗ = T /E.

with three terms. For the two-dimensional.5 and (b) λ = 2. plot the five solutions (with one term.196 35. 39. = x + (λ − λ0 )y − (x2 + y 2 )y. Show that the governing equation of the unsteady. λ = 28 and b = 8/3. Assuming ǫ = 0. convective fin can be written in the form ∂T ∂T ∂ a(x) + b(x)T = 0 − ∂t ∂x ∂x Show that the steady-state temperature distribution with fixed temperatures at the two ends x = 0 and x = L is globally stable. i. etc.) in the range 0 ≤ τ ≤ 1. variable-area. with two terms. Find the dimension of the strange attractor for the Lorenz equations dx dt dy dt dz dt = σ(y − x) = λx − y − xz = −bz + xy where σ = 10. Use the method of counting the number of points N (r) within a sphere of radius r from which D = ln N/ ln r. 37. Consider two cases: (a) a = 0.1. q = 1. Nondimensionalize and solve the radiative cooling problem Mc with T (0) = Ti .1) at time t = 0.e. where u = y v = x − x3 + a cos t determine the pathline of a fluid particle which is at position (1. β = 1. For these two cases find where 11 × 11 points uniformly distributed within a square of size 0. Choose a suitably long final time. 40. dx dt dy dt = (λ − λ0 )x − y − (x2 + y 2 )x. for dθ 4 + θ + ǫ (θ + β) − β 4 = q dτ with θ(0) = 1. using symbolic algebra determine the regular perturbation solution up to and including terms of order ǫ4 . and (b) a = 1. unsteady velocity field ui + vj. dT 4 + σA T 4 − T∞ = 0 dt . 36.1) end up. Choose λ0 = 1 and (a) λ = 0.01 and centered on (1. For heat transfer from a heated body with convection and weak radiation. 38. Show numerically that there are two different types of attractors for the following dynamical system.

197

41. Consider a body in thermal contact with the environment Mc dT + hA(T − T∞ ) = 0 dt

where the ambient temperature, T∞ (t), varies with time in the form shown below. Find (a) the long-time solution of the system temperature, T (t), and (b) the amplitude of oscillation of the system temperature, T (t), for a small period δt.
T∞ Tmax δt

Tmin

t

Figure C.7: Ambient temperature variation.

42. For a heated body in thermal contact with a constant temperature environment Mc dT + hA(T − T∞ ) = Q dt

analyze the conditions for linear stability of PID control. 43. Two heated bodies at temperatures T1 (t) and T2 (t), respectively, are in thermal contact with each other and with a constant temperature environment. The temperatures are governed by dT1 + C(T1 − T2 ) + C1,∞ (T1 − T∞ ) dt dT2 + C(T2 − T1 ) + C2,∞ (T2 − T∞ ) M2 c2 dt M1 c1 = Q1 (t) = Q2 (t)

where Q1 (t) and Q2 (t) are internal heat generations, each one of which can be independently controlled. Using PID control, choose numerical values of the parameters and PID constants to show numerical results for the different types of responses possible (damped, oscillatory, etc.). 44. Show numerical results for the behavior of two heated bodies in thermal contact with each other and with a constant temperature environment for on-off control with (a) one thermostat, and (b) two thermostats. 45. Analyze the system controllability of two heated bodies in thermal contact with each other and with a constant temperature environment for (a) Q1 (t) and Q2 (t) being the two manipulated variables, and (b) with Q1 (t) as the only manipulated variable and Q2 constant.

198

46. Run the neural network FORTRAN code in http://www.nd.edu/ msen/Teaching/IntSyst/Programs/ANN/ for 2 hidden layers with 5 nodes each and 20,000 epochs. Plot the results in the form of exact z vs. predicted z. 47. Consider the heat equation ∂T ∂2T = ∂t ∂x2 with one boundary condition T (0) = 0. At the other end the temperature, T (1) = u(t) is used as the manipulated variable. Divide the domain into 5 parts and use finite differences to write the equation as a matrix ODE. Find the controllability matrix and check for system controllability.

48. Determine the semi-derivative and semi-integral of (a) C (a constant), (b) x, and (c) xµ where µ > −1. 49. Find the time-dependent temperature field for flow in a duct wih constant T∞ and Tin , but with variable flow rate V (t) = V0 + ∆V sin(ωt) such that V is always positive. 50. Write a computer program to solve the PDE for the previous problem, and compare numerical and analytical results. 51. Derive an expression for heat transfer in a fractal tree-like microchannel net1 . 52. Find the steady-state temperature distribution and velocity in a square-loop thermosyphon. The total length of the loop is L and the distribution of the heat rate per unit length is  for L/8 ≤ x ≤ L/4,  Q −Q for 5L/8 ≤ x ≤ 3L/4, q(x) =  0 otherwise.
g Q

Q x

L/4

1 Y. Chen and P. Cheng, Heat transfer and pressure drop in fractal tree-like microchannel nets, International Journal of Heat and Mass Transfer, Vol. 45, pp. 2643–2648, 2002

199

53. Show that the dynamical system governing the toroidal thermosyphon with known wall temperature can be reduced to the Lorenz equations. 54. Draw the steady-state velocity vs. inclination angle diagram for the inclined toroidal thermosyphon with mixed heating. Do two cases: (a) without axial condution2 , and (b) with axial conduction. 55. Model natural convection in a long, vertical pipe that is being heated from the side at a constant rate. What is the steady-state fluid velocity in the pipe? Assume one-dimensionality and that the viscous force is proportional to the velocity. 56. Using the center manifold projection, find the nonlinear behavior of dx = x2 y − x5 , dt dy = −y + x2 , dt and hence determine whether the origin is stable. 57. The Brinkman model for the axial flow velocity, u∗ (r∗ ), in a porous cylinder of radius R is µeff µ d2 u ∗ 1 du∗ + ∗ ∗ − u∗ + G = 0, ∗2 dr r dr K

where u∗ = 0 at r∗ = R (no-slip at the wall), and ∂u∗ /∂r∗ = 0 at r∗ = 0 (symmetry at the centerline). µeff is the effective viscosity, µ is the fluid viscosity, K is the permeability, and G is the applied pressure gradient. Show that this can be reduced to the nondimensional form 1 d2 u 1 du + − s2 u + = 0, dr2 r dr M where M = µeff /µ, Da = K/R2 , s2 = (M Da)−1 . (C.13)

58. Using a regular perturbation expansion, show that for s ≪ 1, the velocity profile from equation (C.13) is 1 − r2 s2 u= 1− 3 − r2 + . . . 4M 16 59. Using the WKB method, show that the solution of equation (1) for s ≫ 1 is u = Da 1− exp {−s(1 − r)} √ + ... r

60. Consider steady state natural convection in a tilted cavity as shown. DA and BC are adiabatic while AB and CD have a constant heat flux per unit length. It can be shown that the governing equations in terms of the vorticity ω and the streamfunction ψ are ∂2ψ ∂2ψ + +ω ∂x2 ∂x2 2 2 ∂T ∂T ∂ ω ∂ ω − Ra P r + cos α − sin α 2 2 ∂x ∂y ∂x ∂y ∂ψ ∂T ∂2T ∂2T ∂ψ ∂T − − − ∂y ∂x ∂x ∂y ∂x2 ∂y 2 = = = 0 0 0

∂ψ ∂ω ∂ψ ∂ω − − Pr ∂y ∂x ∂x ∂y

2 M. Sen, E. Ramos and C. Trevi˜ o, On the steady-state velocity of the inclined toroidal thermosyphon, ASME J. n of Heat Transfer, Vol. 107, pp. 974–977, 1985.

200

where P r and Ra are the Prandtl and Rayleigh numbers, respectively. The boundary conditions are: A ∂ψ ∂T at x = ± , ψ = = 0, = 0, 2 ∂x ∂x ∂ψ ∂T 1 = 0, = 1. at y = ± , ψ = 2 ∂y ∂y where A = L/H is the aspect ratio. Find a parallel flow solution for ψ using ψ T (x, y) = ψ(y) = Cx + θ(y)

C H L y x B D gravity

A

α

Figure C.8: Problem 5.

61. Determine the stability of a fluid layer placed between two horizontal, isothermal walls and heated from below. 62. Obtain the response to on-off control of a lumped, convectively-cooled body with sinusoidal variation in the ambient temperature. 63. Determine the steady-state temperature field in a slab of constant thermal conductivity in which the heat generated is proportional to the exponential of the temperature such that d2 T = exp(ǫT ), dx2 where 0 ≤ x ≤ 1, with the boundary conditions T (0) = T ′ (0) = 0. 64. In the previous problem, assume that ǫ is small. Find a perturbation solution and compare with the analytical. Do up to O(ǫ) by hand and write a Maple (or Mathematica) code to do up to O(ǫ10 ). 65. The temperature equation for a fin of constant area and convection to the surroundings at a constant heat transfer coefficient is d2 − m2 θ = 0, dx2

Consider the hydrodynamic and thermal boundary layers in a flow over a flat plate at constant temperature. (d) Use the summation rule to show that Fab−cd = (e) Show that Fab−cd = AA FAB /Aab . Add radiation to a convective fin with constant area and solve for small radiative effects with boundary conditions corresponding to a known base temperature and adiabatic tip. Complete the details to derive the Nusselt result for laminar film condensation on a vertical flat plate. Using the shooting method and the appropriate boundary conditions. 67. 70. 66. The steps are: (a) Assuming the strings to be imaginary surfaces. Prove Hottel’s crossed string method to find the view factor FAB between two-dimensional surfaces A and B with some obstacles between them as shown. x 68. . Use a similarity transformation on the boundary layer equations to get 2f ′′′ + f f ′′ Pr ′ θ′′ + fθ 2 = = 0.201 where θ = T − T∞ . 0. Determine the eigenfunctions of the differential operator for each combination of Dirichlet and Neumann boundary conditions at the two ends x = 0 and x = L. 2Aab 69. Find from the literature if there is any experimental confirmation of the result. show that Fab−ac = (c) For abd find Fab−bd in a similar way. (f) Show the final result FA−B = Abc + Aad − Aac − Abd 2AA Abc + Aad − Aac − Abd 2Aab Aab + Aac − Abc . apply the summation rule to each one of the sides of figure abc. solve the equations for different Pr and compare with the results in the literature. (b) Manipulating these equations and applying reciprocity. The dotted lines are tightly stretched strings. Find the temperature distribution in a slightly tapered 2-dimensional convective fin with known base temperature and adiabatic tip.

8 to calculate hc . Write a computer program to do the previous problem numerically using finite differences and compare with the analytical result. Derive the unsteady governing equation for a two-dimensional fin with convection and radiation. A plane wall of thickness 1 m is initially at a uniform temperature of 85◦ C. Write a computer program to do the previous problem numerically using finite differences and compare with the analytical result. and 250◦ C isotherms. Edges DA and BC have temperatures of 100 and 0 units. where T is in K. write a program to redraw Fig. respectively. . 72. (c) 200◦ C. 11 must be solved numerically. (a) What is the heat generation rate per unit volume. AB and CD are adiabatic and the circle is at a temperature of 200 units. Going around. (b) 100◦ C.202 a 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 B b 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 d cc A 71. 150. The steady-state temperature distribution in a plane wall of thermal conductivity k and thickness L is given by T (x) = 4x3 + 5x2 + 2x + 3. in the wall? ′′ (b) Determine the heat fluxes. q(x). Suddenly one side of the wall is lowered to a temperature of 20◦ C. 2 on a sunny day (Cl = 0) and a cloudy day (Cl = 1). 5th edition) Consider a square plate of side 1 m. and (d) 300◦ C. Save the M-file and e-mail it to: wcai@nd. qx . Choose different grid sizes and show convergence of the heat flux at any wall. Find the steady-state temperature distribution analytically. Plot the 75. Use Eq. Draw the isotherms. t). 73. 75. (From Brauner and Shacham. Find the view factor of a semi-circular arc with respect to itself. 79. while the other side is perfectly insulated. and the constants in appropriate units. 77. At a corner of a square where the temperature is discontinuous. (From Incropera and DeWitt. 74. x in m. at the two faces x = 0 and x = L. 11. the temperatures on the sides are (a) 50◦ C. show how the finite difference solution of the steady-state temperature behaves ompared to the separation-of-variables solution. Assume the thermal diffusivity to be 1 m2 /s. Assume Ta = 37◦ C. 80. 78. Solve the steady state conduction equation ∇2 T = 0 in the area in the figure between the square and the circle using the MATLAB Toolbox. Eq. 76.edu. 1995) Using Eq. Note: since the physical properties are to be taken at the mean temperature between Ts and Ta . Find the time-dependent temperature profile T (x.

(a) Show that the transient governing equation for a constant area fin with constant properties that is losing heat convectively with the surroundings can be written as 1 ∂θ ∂2θ − m2 θ. circulant. . .203 81. ... To maintain temperatures. of an N × N . . where j = 1. initially at temperature Ti is being cooled by natural convection to fluid at T∞ . . . with each at a uniform temperature Ti (t). (a) Derive the governing equations for the system. gβ(Ts − T∞ )D3 .. 0 a 6 a b c . RaD = 84. A number of identical rooms are arranged in a circle as shown. 2.469/Pr ) where 1/4 9/16 4/9 . each room has a heater that is controlled by independent but identical proportional controllers. and that the material properties are all constant. Determine the steady temperature distribution in a two-dimensional convecting fin... 0 0 7 7 6 7 6 . 7 6 . (c) Write the dynamical system in the form x = Ax and determine the condition for stability3 . . 0 a b c 5 c 0 . = α ∂t ∂x2 3 Eigenvalues are λj = b + (a + c) cos{2π(j − 1)/N } − i(a − c) sin{2π(j − 1)/N }. (b) Find the steady ˙ state temperatures.589 RaD 1 + (0. . . and with its neighbors with a conductive thermal resistance R. . banded matrix of the form [3] 3 2 b c 0 . 0 0 7 7 6 6 0 a b . . Each room exchanges heat by convection with the outside which is at T∞ . . . 82. Derive the governing equation. .. Churchill’s correlation for natural convection from a sphere is Nu = 2 + 0. 7 6 4 0 . and find a two-term perturbation solution. .. . 0 a b . A sphere. and nondimensionalize. T∞ i−1 i T∞ i+1 83. να Assume that the temperature within the sphere T (t) is uniform.... . N .

Cantor Sets: Construct a fractal that is similar to the Cantor set shown in class. (c) Box Counting (graphical). Space Filling Curves: Shown below is a Peano curve. Calculate its Hausdorff dimension using each of the following methods: (a) Dh = log P/ log S. (b) applying Cauchy’s formula once with n = 5.10: Initiator and generator for a Peano (space filling) curve. 85. Calculate its Hausdorff dimension and state if the Peano curve is indeed a fractal. (b) Box Counting (analytical). 89. 86.204 (b) With prescribed base and tip temperatures. use an eigenfunction expansion to reduce to an infinite set of ordinary differential equations. (c) Show that the steady state is attracting for all initial conditions. Menger’s Sponge: Shown below is Menger’s Sponge. but instead remove the middle 1/2 from each line. Show that the fractal dimension is 1/2. The generator is recursively applied to generate the Peano curve. Show that the perimeter of the cross-section is infinite while the flow area is finite.9: Menger’s Sponge 87. (a) Initator (b) Generator Figure C. 88. a single line that completely fills a unit square. Cauchy’s formula: Verify Cauchy’s formula for repeated integration by (a) integrating f (t) five times. Figure C. (c) applying Cauchy’s formula . A duct carrying fluid has the cross-section of Koch’s curve.

which would be of interest in this problem. . Consider the periodic heating and cooling of the surface of a smooth lake by radiation. The surface is subject to diurnal heating and nocturnal cooling such that the surface temperature can be described by Ts (t) = To + Ta sin ωt. the calculation we used would not be suitable because of the enormous number of summands in the calculation of the derivative and because of the accumulation of round off errors. Also assume the lake to be a semi-infinite planer medium (a lake of infinite depth) with T (∞. 1/2 Note: In large time intervals (t very large). Hint: It is easiest to calculate the binomial coefficients recursively. Numerical Evaluation of a Fractional Derivative: Consider the example worked in class of calculating the heat flux in a blast furnace. which is simply the derivative of order α = 1/2 of the temperature difference at the surface. t) ∂T (x. 92. 0) = To and T (0. Assume the heat diffusion to be one-dimensional and find the heat flux at the surface of the lake. k k cpλ 0 Dt g(t). 91. 90. The following steps might be useful: (a) Assume transient one-dimensional heat conduction: ∂ 2 T (x. (a) calculate D3 f (t) (take m = 4. ∂t ∂x2 with initial and boundary conditions T (x.205 twice. showing that t f (τ )dτ = J 5 f (t) = J 3 J 2 f (t) 0 for f (t) = 16t3 .5 f (t) and plot the function. Calculate the fractional derivative numerically using the first 2 hours of data and plot both the heat flux at the surface and g(t). α = 3) and verify that you get the same result as traditional differentation by comparing to d3 f /dt3 . α α−k+1 . The heat flux was calculated to be q ′′ (t) = where g(t) = Tsurf (t) − T0 . The derivative with this “short α memory” assumption is typically written as (t−L) Dt . according to the recursion formula: α 0 α k+1 = = 1. t) −α = 0. t) = 0. giving the discrete data set gi = 14 sin(πi/60). t) = Ts (t). In these situations. Caputo Derivative: Take f (t) = 2t5 and using the Caputo RHD. Assume that the function g(t) is given as g(t) = 14 sin(πt/60) where t is in minutes and the thermocouples sample once per minute. (b) Calculate D2. the principle of “short memory” is often applied in which the derivative only depends on the previous N points within the last L time units. once to f (t) with n = 2 and then to the result with n = 3.

For small ǫ. Assume constant properties. . t)/∂x. s) (with 0 initial conditions) L ∂tα and ∂α ∂ α f (x. Now take the inverse Laplace transform of ∂Θ/∂ξ and convert back to the original variables. 94. The following simplifications might be helpful: ∂ 1/2 C C = 1/2 ∂t1/2 πt ∂ α g(t) ∂ α [Cg(t)] =C ∂tα ∂tα (h) The solution should look now look like ′′ qs (t) = α1/2 k d1/2 (Ta sin ωt) dt1/2 93. 0) = f (x). s) by applying the transformed boundary conditions. s) = L α ∂x ∂xα (d) Solve the resulting second order differential equation for Θ(ξ. t) − To .12 has a fixed temperature and the other is adiabatic. Consider radiation between two long concentric cylinders of diameters D1 (inner) and D2 (outer). determine the temperature distribution in the wall T (x. respectively. C.206 (b) Non-dimensionalize the problem with a change of variables ξ = α−1/2 x and θ(x. q ′′ (x. TR − TL 96. (e) Find ∂Θ/∂ξ and then substitute Θ(ξ. C.11 are TL and TR . With an initial condition T (x. 95. t) = −k ∂T (x. t)/∂x. One side of a plane wall shown in Fig. s) into the result. t) F (x. (c) Use the following Laplace transform properties to transform the problem into the Laplace domain: ∂ α f (x. (b) Find F22 and F21 in terms of the cylinder diameters. (a) What is the view factor F12 . Temperatures at the two sides of a plane wall shown in Fig. Be careful! The derivative of order 1/2 of a constant is not zero! (see simplifications in (g) to simplify) (f) You should now have an expression for ∂T (x. Substitute this into Fourier’s Law to calculate the heat flux. t) = sα F (x. Describe how you would solve this problem using more typical methods and what other information would be required. t) at any other time. (g) Evaluate this expression at the surface (x = 0) to find the heat flux at the surface of the lake. find a perturbation steady-state temperature distribution T (x) if the dependence of thermal conductivity on the temperature has the form k(T ) = k0 1 + T − TL ǫ . t) = T (x.

1111 0000 111 000 1111 0000 111 000 √ √ 101.11: Plane wall in steady state. L T = Ts x ∂T /∂x = 0 Figure C. The heat rate in and out at the top and bottom. Show that the functions φ1 (x) = 2 sin πx and φ2 (x) = 2 sin 2πx are orthonormal in the interval [0. A room that loses heat to the outside by convection is heated by an electric heater controlled by a proportional controller. 98.12: Plane wall in unsteady state. State your other assumptions. Using an eigenfunction expansion. Using these as test functions. Determine the constraint on the controller gain for the system response to be stable. 1] with respect to the L2 norm. 1111 0000 111 000 1111 0000 111 000 . where 1111 0000 111 000 1111 0000 111 000 1111 0000 111 000 Tb Ta 1111 0000 111 000 1111 0000 111 000 β > 1. and so on. 97. L ω L 100. Consider rotational forces but not gravity. is Q while the rest of the loop is insulated. respectively. An infinite number of conductive rods are set up between two blocks 1111 0000 111 000 1111 0000 111 000 1111 0000 111 000 at temperatures Ta and Tb . With a lumped capacitance approximation for the temperature. 1111 0000 111 000 . the third A3 = A/β .207 L T = TL x T = TR Figure C. What is the temperature of the room after a long time? 99. A turbine blade internally cooled by natural convection is approximated by a rotating natural circulation loop of constant cross-section. What is the total steady-state heat transfer rate between the 1111 0000 111 000 . . use the Galerkin method to find an approximate solution of the steady-state fin equation T ′′ − T = 0. the second A2 = A/β. and 1111 0000 111 000 1111 0000 111 000 1111 0000 111 000 1111 0000 111 000 1111 0000 111 000 that there is no convection. with T (0) = 0. The first rod has a cross-sectional area 1111 0000 111 000 1111 0000 111 000 1111 0000 111 000 2 1111 0000 111 000 1111 0000 111 000 A1 = A. reduce the governing PDE in Problem 96 to an infinite set of ODEs. T (1) = 1. 1111 0000 111 000 two blocks? Assume that the thermal conductivity k is a constant. set up a mathematical model of the system. Find the steady-state velocity in the loop.

If heat loss is by convection on the external surface only. A B R 104. 105. A lamp of q W is radiating equally in all directions. Determine the temperature of the ball as a function of time T (t) if heat loss is by convection to the atmosphere. A ball with coefficient of restitution r falls from height H and undergoes repeated bouncing. . Determine the steady-state temperature distribution in the triangle shown. d C 103. find the steady-state temperature distribution.208 q 102. Assume that the energy loss at every bounce goes to heat the ball. if the hypotenuse is adiabatic. Heat at the rate of q per unit volume is generated in a spherical shell that lies between R and R + δ. one of the sides is at one temperature and the other is at another. Set up the governing equation for the temperature T (r) in the disk.

M. Fortuna. Convection Heat Transfer.T. Alotaibi. Department of Aerospace and Mechanical Engineering. 1995. Prentice-Hall. NJ. Aziz and T. Yang. and D.-H. Renewable Energy. MA. Numerical Heat Transfer A. 47:913–924. 2004. Alotaibi. Goodwine. Antonopoulos and C. 1997. IN. Flow-based control of temperature in long ducts. [5] S. 2000. Ramos. 2001. Ahmed. New York. [7] S. U. L. and E.N.References [1] R. Sen. Nonlinear Noninteger Order Circuits and Systems–An Introduction.und Stoff¨bertragung. [10] R. 2004. Hemisphere Publ. Sen. Michel.S. [6] S. Controllability of cross-flow heat exchangers. 1998. Stephan. and A. [12] V. Barab´si.Y. Alotaibi. 1999.U.T. Perseus. PhD thesis. Semigroup Theory with Applications to Systems and Control. R. DC.T. 1987. Goodwine.J..W. Kao. W. World Scientific. Notre Dame. [11] Arpaci and Larson. [3] R. Sen. and K. Special Matrices of Mathematical Physics: Stochastic. a u [2] N. J. S. and K. International Journal of Heat and Mass Transfer. A correlation for the thermal delay of building. World Scientific. Na. 41:229–244. Goodwine M. 1984. a 209 . Upper Saddle River. Harlow. McGraw-Hill.K. Tzivanidis. Yang. University of Notre Dame. Sen. [13] A. Cambridge. 2002. Numerical simulation of the thermal control of heat exchangers. Arpaci. and K. Singapore. Baehr and K. Longman. Prentice-Hall. Selamet. [9] P. Antsaklis and A. Washington. M. International Journal of Heat and Mass Transfer.J. W.D. Temperature Controllability in Cross-Flow Heat Exchangers and Long Ducts. Introduction to Heat Transfer. Single-phase natural circulation in a tilted square loop. Acosta.. Corp. Aldrovandi. Perturbation Methods in Heat Transfer. Arena. Alotaibi. Porto. 21(5):269–275. [4] S. Singapore. Linked: The New Science of Networks. [8] K. 47:4995–5009. 2002.-L. Heat and Mass Transfer. 1984. Springer.A. [14] H. 6(7):687–699. New York. Linear Systems.J. [15] A. Yang. Circulant and Bell Metrices. Caponetto. 2003. 1991. W¨rme. M.

Charny. 42(7):1327–1341. New York. John Wiley. 1992. and Mass Transfer. Hemisphere Publ. 1984. Brewster. Momentum. 1999. Jaeger.O. New York. Thermal Radiative Transfer and Properties. Conduction of Heat in Solids. editor. Springer-Verlag. Liquid-Vapor Phase-Change Phenomena. John Wiley. Model Predictive Control in the Process Industry. Cooke. 1963. New York. Physica-Verlag.. International Journal of Heat and Mass Transfer. [34] A. Camacho and C. chapter 1. editor. pages 19–155. New York.S. [29] V. [26] L. 2001. Convection Heat Transfer.C. Bejan. New York. Springer. 1988. Cois. [20] M. 39(3):374–389.N. Physical and Computational Aspects of Convective Heat Transfer. Heat transfer in micro.K. Bejan.and nanoscale photonic devices. Bau. 1992. New York. DC. Bordons. Becker. [21] A. Comp. [35] C. Macmillan. Convection Heat Transfer. [25] M. New York.J. McGraw-Hill. New York. 2001. 1987. Begell House. Elements of Heat Transfer. Carey. New York. Castillo and P. [32] O. [33] Cebeci and Bradshaw. and O.. Advances in Heat Transfer. 1996. [24] C. 2nd edition. [19] Y. Academic Press. Burmeister. [17] J. New York. volume 22. Carr. Le Lay. 1981. Batsale. UK. volume 7.C. Applications of Centre Manifold Theory. Bayazitoglu and M. Chen. Differential-Difference Equations. New York. 2000. Clarendon Press. Oustaloup. New York. Academic Press. . Plenum Press. Burmeister. [23] R.F. [36] G. J. 3rd edition. John Wiley. Heat Transfer. Oxford.I. Heat Transfer.L. Darrell. Soft Computing for Control of Non-Linear Dynamical Systems. Math. Dover. Heat. Mathematical models of bioheat transfer. In C. [30] J. [31] H.C. Heat flux estimation through inverted non-integer identification models. Chapman.Q. Bellman and K. Washington. Annual Review of Heat Transfer. Bennett. Battaglia. Appl. 117:137–158.-L. A Modern Approach. Melin. Heat Transfer. Control of Marangoni-Benard convection.L. Ozisik. A. J.H. [28] E. Tien.P. Corp. 1993. 1982. 3rd edition. 1982. Fundamentals of Heat Transfer. Carslaw and J. New York. 2000. [27] L. L. McGraw-Hill. International Journal of Thermal Sciences. In Y. Springer-Verlag.C. 1959. On stability of systems of delay differential equations. London. John Wiley. Cho. [22] A. [18] H. Heidelberg. 1992. 1986. 1984. 1993.REFERENCES 210 [16] C Baruch and S.

2000. A time-delay control algorithm for an industrial electrical heater. 5(3):195–208. D´ M. Chu.T.I. D´ M.T. Stabilization of thermal neurocontrollers. IN. D´ ıaz. McClain. 1992. To ıaz.T. [40] J. On the delayed hot water problem. [43] C. 4:225–230.T. Sen.L. M. Effect of delay in thermal systems with long ducts. McClain. 44(9):1671–1679. AIChE Journal. [53] K. Notre Dame.REFERENCES 211 [37] G. Academic Press.. 2002. Fourligkas. Boca Raton. McGraw-Hill. 5(1):3–8. 123(3):417–612. Proc.D.T. D´ ıaz.L.. ASME Journal of Manufacturing Science and Engineering. K. [46] D. [52] G.K. 2001. 3(4):220–224. Sen. Free and Moving Boundary Problems. D´ ıaz. pages 157–357. K. Cho. [48] G. . [49] G. 1974. McClain. McClain. Sen. Doumanidis and N. Nonlinear and Robust Control of PDE Systems. Singular Control Systems. PhD thesis. In Y. University of Notre Dame. Adaptive neurocontrol of heat exchangers. Chen. Dai. Hu. M.T. To be ıaz. Convective Boiling and Condensation. Crank. be published in Applied Artificial Intelligence. 2004.L. CRC Press. Linear Stochastic Control Systems. [42] Collier. Oxford. 118(4):571–578. and R. Sen. 1984. Springer-Verlag. Yang. Feed-back control of hyperbolic PDE systems. Chu. [51] G. Berlin. Diller. McClain. International Journal of HVAC&R Research. [44] J. published in International Journal of Thermal Sciences. D´ ıaz.L. Yang. [45] L. Yang. Distributed-parameter control of the heat source trajectory in thermal materials processing. [41] J. 1995.. a [39] P. Proc. Yang. Yang. [38] P. Simulation and Control of Heat Exchangers using Artificial Neural Networks. J. Boston. and R. Comstock. 1989..D. 412:3063.R. Cont. J. 1995. Advances in Heat Transfer. 1978. Christofides. [54] C. Birkh¨user. Dynamic prediction and control of heat exchangers using artificial neural networks. 2004. [47] G. Cont. 96:166–171.L. Christofides and Dautidis P. 2002. Simulation of heat exchanger performance by ıaz. artificial neural networks. Application of a discrete optimal tracking controller to an industrial electrical heater with pure delay. K. 1999. Heat and Mass Transfer. and R. Stabilization of a neural network-based temperature controller for heat exchangers. 1996. and R. ASME Journal of Heat Transfer. Su H. M. K. Proceedings of the 12th International Heat Transfer Conference. Models of bioheat transfer processes at high and low temperatures. editor. Yang M. 1984. Sen. Department of Aerospace and Mechanical Engineering. Clarendon Press. U. New York. 2001. D´ K. International Journal of Heat and Mass Transfer. Chen and G. 1993. volume 22. Sen. and x. and K. and R. [50] G.

Longman. 1985. [70] L. Automatica. Yang. 2004.K. R. PI stabilization of first-order systems with time delay. [65] V.. 2001. Sensors for Measurement and Control. Henryk. McGraw-Hill. 1984. [59] P. and R. Feder. Corp. Driver. Dynamics of thermal-hydraulic network control strategies. 1992. [58] D. 1998. Washington. Active and Reactive Flow Management. Cambridge University Press. Springer-Verlag. Proceedings of the Institution of Mechanical Engineers.G. Thermal responses in control loop in indirect control of indoor environment of non-air-conditioned space with quasi-steady-state model. [72] E. [73] S. Part I: Journal of Systems and Control Engineering. 2000. Franco. 2003.L. S. 1981. Radiation Heat Transfer Notes. 2003. 1993.M.. and R. [62] W. 3rd edition. . Hemisphere Pub. Hemisphere Publ. Hale and S.. Guillermo. Department of Aerospace and Mechanical Engineering. Introduction to Functional Differential Equations. Hansen. Analysis and Synthesis of Time Delay Systems.. Fourligkas. 217:35–47. Macmillan. Cambridge University Press. Springer-Verlag. Experimental Heat Transfer.K. New York. Franco. Essex. Drazin. University of Notre Dame. Notre Dame. Corp.L. OK. Corp. New York.K. 118(4):571–578. Heat Conduction and Mass Diffusion. McGraw-Hill. Hydrodynamics and Control in Thermal-Fluid Networks. 1999. Franco. and S. Plenum. 1982. Grigull. Sen. [71] J. PennWell Pub. U. Applied Heat Transfer. Piotr. Tulsa.P. Chichester.. G. Elgar. New York.K. 1996. Gad-el-Hak. [56] P. Distributed-parameter control of the heat source trajectory in thermal materials processing. Jaluria. IN. 1993. U. New York. [69] J. New York. Nonlinear Systems.G. Hemisphere Pub. M. DC. Mahajan. New York. Hydronic System Design and Operation. Gebhart. Bhattacharyya. Flow Control: Passive. Y. McClain. McClain. Lunel. M. Co..D. Katoh.REFERENCES 212 [55] C. JSME International Journal Series C-Mechanical Systems Machine Elements And Manufacturing. Aniruddha. Yang. [74] G. [66] B. Cambridge. Hach and Y. [63] W. 2003. 1988. Cambridge. 1977. U. Adam. John Wiley Sons. Neural Networks: A Comprehensive Foundation. Comparison of thermal-hydraulic network control strategies. PhD thesis. Washington. Ordinary and Delay Differential Equations. K. Sen. Sammakia. Doumanidis and N. 37:2020–2031. Buoyancy-Induced Flows and Transport. [68] U. D. [60] J.C. Ganapathy. Haykin. 17(3):161–179.T. 1988. New York.. Edwards. and K. [57] R. K. D.K. [67] B.. ASME Journal of Manufacturing Science and Engineering.L. and B. 1989. [64] M. 46(1):197–211. [61] W. Gebhart.T. Fractals. Heat Conduction.

London.E. [79] K. editors.P. [90] V. 1994. Howle. Zhao. [80] J. Liu. Sukomel. Microcontroller-Based Temperature Monitoring and Control. 159:68–76. New York. International Journal of Heat and Mass Transfer. Irena and R. Jaluria. Krabs. . New York. Fairmont Press. 5th edition. Oxford. Moscow. Introduction to Control Theory. Hewitt. W. Nonlinear Control Systems.K. Dewitt. Hoffman and W. Holmes. u [87] D. [88] F. and M. Hill and J. 41. New York. 1990. [86] H. 2nd IFAC Symposium on Distributed-Parameter Systems. a source book for compact heat exchanger analytical data. 1977. John Wiley. 1995. Heat Transfer. 1990. [81] M. U. Cambridge. McGraw-Hill. Control of Rayleigh-Benard convection in a small aspect ratio container. Heat Transfer. V. Selection of measurement locations for the control of rapid thermal processor. 1980. 7th edition.J. [84] C. Prentice Hall. 2000. 40(4):817–822. Vols. Proc. HVAC Control in the New Millennium. Triggiani. L¨th. Berlin. 1993. Isachenko. 1995. [76] G.H. 2001..E.P. Y..P. Pergamon Press. 1999. Osipova. CRC Press.S. Yu. Springer. Liburn. New York. Huang. Laminar flow forced convection in ducts. Holman.P. Springer. Jakob.REFERENCES 213 [75] M. GA. [94] M. John Wiley. Isidori. Newnes. I and II Encyclopedia of Mathematics. volume II.A. SpringerVerlag. Oxford University Press. 1957. [93] O. 1997. Dewynne. [89] L.K. New York. [85] G. Lecture Notes in Control Information Sciences. Oxford.-H. Cambridge University Press. L. Boca Raton. Blackwell Scientific Publications. 2002. 1991. 2002. Natural Convection Heat and Mass Transfer. Kanoh. Jacobs. Temperature control system of heat exchangers. MWR approximation and modal control of parallel and counterflow heat exchangers. Nonlinear Process Control.J. Process Heat Transfer. Shen. Q. Nie. Optimal Control of Partial Differential Equations. [78] E. Hinch. NJ. Masubuchi. Yang. C.R. Henson and D.F. Springer-Verlag. Incropera and D. Mir. [92] A.F. Wu. Fundamentals of Heat and Mass Transfer. Ibrahim. 2nd Ed. 36(5):705–715. Seborg. 1991. [83] L. 1987. and L. an application of DPS theory. 1991. and S. Perturbation Methods. Solid-State Physics. [82] M. [95] Y. and A. [91] A.H. Upper Saddle River. New York.C. U. H. 1997.N.A. Cambridge University Press. Heat Conduction. Heat Transfer. Deterministic Control Theory for Infinite Dimensional Systems.M. Automatica. Oxford.. Ito. 1978. Ibach and H. [77] J. Huang.L. Hordeski.

[115] J. FL. S. Auckland. Hartnett. and W. [104] W.-H. Leigh. Kays and A. Convective Heat and Mass Transfer. [108] M. Jaluria and K. FL. Kakac and Y. In J. 1984. 1987. Control System Fundamentals. 1998. [98] S. W. Aung. HVAC Controls and Systems. San Francisco. Kays and M. 1980. Kittel and H. 2nd edition. Boca Raton. Co. Freeman & Co. Irvine. Shah. Principles of Heat Transfer. 1980. [113] A. [109] C. New York.K. 1993.. Control of Nonlinear Differential Algebraic Equation Systems. Webb. Levenhagen and D. 1995.R.H.S. London. Crawford. editors.M. Boca Raton. [106] D. Chapman and Hall/CRC. Compact Heat Exchangers. Kreith and W. VA. Kern. R. .S. Kroemer.I. McGraw-Hill. Peregrinus.P. [112] F. S. Dordrecht. 3rd edition. Springer. Handbook of Single-Phase Convective Heat Transfer. Boca Raton. [116] W.H. Kluwer. 5th edition. Taylor and Francis. Kaviany. CRC Press. Washington. 2nd edition. Springer-Verlag. Cho. 1999. New York. Mehendale. Mc-Graw-Hill. Black.F. [102] M..L.D. Multiple Scale and Singular Perturbation Methods. Kumar and P. Basic Heat Transfer.D. 1994. 1986. 2003. Heat Conduction. [111] F. Taylor and Francis. Greene. Jaluria. [100] S.Y. Lee. 2000. Netherlands. 1950. [114] J. New York. and G. Kevorkian and J.I. Principles of Convective Heat Transfer.E. [103] M. 3rd edition. West Publ. Cole. pages 297–429. McGraw-Hill. Reston. New York.K. Harper and Row. St. Temperature Measurement and Control. editors.REFERENCES 214 [96] Y.S. Kreith and M. [110] J. Computational Heat Transfer. 2nd edition. 1983. New York. [99] S. London.E. Kakac. [97] Y. Spethmann. 1993. Levine. 1991. Microscale heat exchanger design for evaporator and condenser applications. Washington. and R.. John Wiley. T. Karam. AIAA. Torrance.L. 1996. [105] W. Daoutidis. Thermal Physics. Bohn. [107] J. Springer-Verlag.M. New York. McGraw-Hill International. 1995. 1993. Computational Heat Transfer. Controllability of Dynamical Systems. 2nd edition. New York. CRC Press. San Francisco. Amsterdam. Satellite Thermal Control for Systems Engineers. volume 37.Q.A. MN. Process Heat Transfer. DC. Advances in Heat Transfer. U. Kaviany. Convective Heat Transfer. Kakac. Kim. Principles of Heat transfer in Porous Media. Klamka. Y. [101] R. Paul. Cambridge. 1993. Yener. Academic Press. editor. DC. 3rd edition.Z. 2003.

[127] M. Berlin. [128] W. Poulsen. Introduction to Convective Heat Transfer Analysis. Schneider. Modern Control Engineering. McGrawHill. Springer-Verlag. and H. [135] A. second edition. Modest.M. Gerner. Lienhard. Springer. Active control of natural convection in a fluid layer with volume heat dissipation. Cho.. Liou and Y. Mir. Nørgaard. chapter 8. Prentice-Hall. Oustaloup. J. Taylor and Francis. Paris. Applied Mechanics. [126] W. chapter 1. 35(9-10):1095–1118. [133] P. Nield and A.K. Prentice-Hall. 1981. and L. New York. International Journal of Heat and Mass Transfer. In C. [118] J. [125] T. Ljung. 1997. Majumdar. The occurrence of limit-cycles during feedback control of microwave heating. ASME J. Nelles. Washington. 1954.P Hartnett. Amsterdam. [122] A. Academic Press. 2nd edition. N. brief notes. Acad. 1983. Rohsenow.R. Majumdar. The delayed hot-water problem. New York. Paris. 1998. Handbook of Numerical Heat Transfer.J. Mathematical and Computer Modelling. Ravn. Luikov. 21(10):?– ?. Englewood Cliffs. Lions.M.E. A. 2002. Moscow. 1998.H. N. DC. and R. A.REFERENCES 215 [117] J. 2003. [121] L. Handbook of Heat Transfer. editors. Upper Saddle River. [129] O. Microscale Energy Transport. [124] A. editors. editors. and Y. 1998. Naylor. [134] A. Pletcher. Fang. Hansen. Springer.I. El Moctar. Marchant.H. [132] K. G. Convection in Porous Media. John Wiley. Nonlinear System Identification. E.H.W. [119] W. 2000. A Heat Transfer Textbook. Natl. Bejan. Minkowycz. [131] M. e e e . Peerhossaini.O. System Identification: Theory for the User.-L. 1990. 2001. 1995. Microfluid Mechanics: Principles and Modeling. Tien. Syst`mes Asservis Lin´aires d’Ordre Fractionnaire: Th´orie et Practique. Oosthuizen and D. Microscale transport phenomena. La D´rivation Non Enti`re. Neural Networks for Modelling and Control of Dynamic Systems. Prentice-Hall. On the controllability of distributed systems. N. Liu and T. New York. USA. Sci. 1999. [123] A. 3rd edition. New York. 1980.A. 2002. Sparrow.J. McGraw Hill.. Microscale energy transport in solids.K.J. NJ. 45(3):667–678. Mase e e son. In W. London. Oustaloup.L. 2006. Ogata. and F. [120] B. Munk. 1988. McGraw-Hill. [130] D. Marimbordes. Englewood Cliffs. 1999. Herm`s. 94:4828–4835. Majumdar.M. Heat and Mass Transfer. New York. Radiative Heat Transfer. second edition. O.

Visier. University of Notre Dame. 123(2):348–354. Advanced Process Control. PhD thesis. [150] D. and Ganic. [151] W. [140] A. Department of Aerospace and Mechanical Engineering. and J. editors. 1983. 6:137–142. [147] M. [138] M. 3rd edition. 2001. New York. . Influence of sensor position in building thermal control: criteria for zone models. G. Pacheco-Vega. New York. [153] W. Dover Publications.H. Paraskevopoulos. New York. [152] P. Heat Conduction. Numerical Heat Transfer and Fluid Flow. 1994. Heat rate predictions in humid air-water heat exchangers using correlations and neural networks.N. McClain. Neural network analysis of fin-tube refrigerating heat exchanger with limited experimental data. Podlubny. Lee. [142] A. Modern Control Engineering. 2001. Riederer. Simulation of Compact Heat Exchangers using Global Regression and Soft Computing. CA. 1980. 1981.T. H. Prentice-Hall. Ozisik. New York.J.C. M. Marcel Dekker. Ray.M. Patankar. 1980. John Wiley.C. Energy and Buildings. [154] Rohsenow. Poulikakos. Yang. Notre Dame. Sen.N. Genetic-algorithm based prediction of a fin-tube heat exchanger performance. and R. Proceedings of the 11th International Heat Transfer Conference. New York. NJ. 1999. 1998. Heat and Mass Transfer. [144] P. NJ. Heat Exchanger Design Handbook. McGraw-Hill. A review of selected thermal management solutions for military electronic systems. 1959. Pacheco-Vega. 2002. WIT Press. ASME Journal of Heat Transfer. IN. 2002. K. and R.T. and R. Park and W. 2002. 1999. 2002. Pacheco-Vega. Englewood Cliffs.L. McClain. McClain. McGraw-Hill/Hemisphere. Prentice-Hall. M. Feedback control of the Rayleigh-Benard convection by means of mode reduction. Pacheco-Vega. San Diego. M. 2nd edition. Heat Transfer. K. Boston. Hartnett. Price. International Journal of Heat and Mass Transfer. Dynamic behaviour of Heat Exchangers. The Theory of Heat Radiation. 40(7):927–949. Ozisik.REFERENCES 216 [136] M.V.L. Active control and parameter updating techniques for nonlinear thermal network models. Xuan. Computational Heat Transfer. New York. Fractional Differential Equations. 26(1):26–39. Papalexandris and M. 1968. Yang. Roetzel and Y. 2001. Rolle. [141] A.L. 34(6):785–798. John Wiley. A Basic Approach. McGraw-Hill. 27(1):11–22. Boundary Value Problems of Heat Conduction. Sen. [145] H. [148] I.C. [149] D. D´ ıaz. 2000. 2003. New York. Yang.T. 44:763–770. D. [139] A. [143] M. [155] K. Marchio. Dover Publications. IEEE Transactions on Components and Packaging Technologies. Upper Saddle River.N. Planck. [137] M. [146] S. 1985. Milman. Computational Mechanics. International Journal for Numerical Methods in Fluids. Sen. Ozisik.N. K. Academic Press.

Chapter 14. Vasseur.K.W. Ohnishi. [163] M. 30(10):2097– 2113.H.M. Multiple steady states for unicellular natural convection in an inclined porous layer. 1985. 2000. DC. Shah and A. 107. Sen. ASME Journal of Heat Transfer. Modeling. Kailath. Vasseur. E. Laplace’s equation for convective scalar transport in potential flow. A review: thermal processing in fast ramp furnaces. 1989. Analysis of the delay hot/cold water problem. K. Notre Dame. 1994. Trevi˜o. University of Notre Dame. Heat Transfer in Convective Flows. Ramos. Rosenbrock. [161] L. pages 1–37. M. and L. Segerlind. 620– 661. 455(2004):3041– 3045. [169] M. Sen. 28(1):219–233. Sen and P.D. Washington. A. 123(3):Section 4. 24(3):757–769. Kaushal. Sen and K. Lecture Notes on Intelligent Systems. pages 267–272. P. 2001.T. Amsterdam. and C. [166] M. 21(5):395–400.24. Sen and J. [162] M.T. Soft computing in control. and control of rapid thermal-processing systems. Academic Press. On the steady-state velocity of the inclined toroidal n thermosyphon. New York. [167] M. John Wiley. Corp.REFERENCES 217 [156] H. Ramos. Intern national Journal of Heat and Mass Transfer. and Y. 2000. Moslehi. [157] E. Physical and Engineering Sciences. E. [168] M. International Journal of Thermophysics. Goodwine. 1996. Mahdi.. 2004. 2003.J. [170] R. Journal of the Electrochemical Society.O. 1984. Shimazaki. Robillard. Numerical Heat Transfer. 1978. 1985. and C. Hemisphere Pub. [172] K. Applications of artificial neural networks and genetic algorithms in thermal engineering. and T. The MEMS Handbook. London. . Yang. Journal of Electronic Materials. Energy. 1984. HTD-Vol. [159] N. Proceedings of the National Heat Transfer Conference. International Journal of Heat and Mass Transfer. Analysis of multiple solutions in plane poiseuille flow with viscous heating and temperature dependent viscosity. Sen. Development of spectral selective multilayer film for a variable emittance device and its radiation properties measurements.L. Schaper. Laminar flow forced convection in ducts. 2nd edition. and S. John Wiley. identification. [160] C. Merchant. Roxin.C. 141(11):3200– 3209.M. 1987. Roy. Proceedings of the Royal Society A: Mathematical. [171] T. [164] M. Control Theory and its Applications. Nagasaka. 1970. 30(12):1578–1583. [165] M. a source book for compact heat exchanger analytical data. Gordon and Breach. The CRC Handbook of Thermal Engineering. Department of Aerospace and Mechanical Engineering. S. New York. The toroidal thermosyphon with known heat flux. Saman and H. Sen and K. 2001.M. New York. Applied Finite Element Analysis. State-Space and Multivariable Theory. Trevi˜o.K. [158] P. IN 46556. Sen. 1997. 107(4):974–977. Saraswat. Shih. Yang.

H. U. Stabilization of the no-motion state in the Rayleigh-Benard problem. Feedback-control stabilization of the no-motion state of a fluid confined in a horizontal porous layer heated from below. Sontag. Tabib-Azar. Magnetic. Kluwer Academic. Sync: The Emerging Science of Spontaneous Order. Stabilization of the no-motion state of a horizontal fluid layer heated from below with Joule heating.H. Sund´n and M. 125(1):164–174. [189] J.H. 1995. Bau. Wang. Tang and H. Smith and H. London. Thermal Radiation Heat Transfer. Sissom and D. UK. Tang and H. 3rd edition. Bau. 1989. [179] E.R.H. DC. [186] P. Clarendon Press. Pitts. Microactuators: Electrical. Computational e Mechanics Publications. Bau. Springer. Sweetland and J. [188] J.Z. Singer and H. Boston. 1993. Lienhard. Springer.H. ASME Journal of Heat Transfer. 1972.V.H. N. Stabilization of the no-motion state in Rayleigh-Benard convection through the use of feedback-control. Faghri. Howell. Southhampton. Chemical & Smart Structures. Introduction to Microfluidics.. 2005. Prentice-Hall. Engineering Heat Transfer. Bau. Transport Phenomena. Corp. [181] S.H. Tabeling. [178] H. Oxford University Press. [187] M. [191] J... Tang and H. Microcomputer Control of Mechanical and Thermal Systems. Minneapolis/St. Jensen. Hemisphere Publ. 2003. 23(9):1055–1065.. Tang and H.F. McGraw-Hill. West Pub. Y. New York. Computer Simulations in Compact Heat Exchangers. ASME Journal of Heat Transfer. Journal of Fluid Mechanics. Physical Review Letters. Stoecker. 1998. [175] J.H. 3(12):2859–2865. 447(1931):587–607. Delay Effects on Stability. Stoecker and P..A. [176] J. 1998. [174] N. [177] L.H. Mechanical. 2003. [192] J. Mathematical Control Theory. Oxford. 66(9):1123–1125. Washington. .K. and H. Van Nostrand Reinhold. Silviu-lulian. [185] M. New York.. Elements of Transport Phenomena. New York. 2001. [182] B. [190] J. Taine. 1995. Applied Thermal Engineering. 1993.C. Oxford. Active thermal control of distributed parameter systems with application to testing of packaged IC devices. NASA thermal control technologies for robotic spacecraft.J. 1991. Suryanarayana. Strogatz. 1993. 1992. [183] N. 1994. New York. Swanson and G. Bau.D. 2003. Englewood Cliffs. 1991. Physical Review Letters. Active control of convection. Proceedings of the Royal Society of London Series A-Mathematical and Physical Sciences. Heat Transfer. Controlling a chaotic system. Singer. 1998. 257:485–505. Co. Bau.REFERENCES 218 [173] R. Paul.D. [180] W. Thermal. 117(2):329–333.K.R. Theia. U.E. [184] T. 1989. Physics of Fluids A-Fluid Dynamics. Siegel and J. Birur. 70(12):1795–1798. Optical. Lecture Notes in Control and Information Sciences.

Evaluation of rapid thermal processing systems for use in CMOS fabrication. .S. [209] J. [202] D. Tseng. 1996. Numerical investigation of the stabilization of the no-motion state of a fluid layer heated from below and cooled from above. K. Wolf.M. Whitham. 102(2):?–?. 237:479–498. 2nd edition.. ASHRAE Transactions. 2004. Singer. 1997. 1997. Bau.H. Springer-Verlag. U. 1966.REFERENCES 219 [193] J. Analytical investigation of b´nard-marangoni convection e heat transfer in a shallow cavity filled with two immiscible fluids.to Microscale Heat Transfer: The Lagging Behavior. DC. Varga. J.H. M. Holt. [201] J. 1998. White. Tin-Tai. [196] J. New York. DC. 48:35–53. Weigand. Vasseur.H. John Wiley & Sons. 2nd edition. [205] Y. and P. Wang. Tang and H. Bau.C. Applied Scientific Research. Berlin. New York. Tang.M. Linear and Nonlinear Waves.W. Taylor and Francis. New York. [207] F.S. Hangos. 1974. Control Engineering Practice. New York. Washington. D. Controllability of heat exchanger networks in the time-varying parameter case. Springer. Wang. 2002. 2003. 1991. Radiant Heat Transfer in Porous Media. 1995. Cambridge.K. Physics of Fluids. 1997. [210] R. Wiebelt. [206] B.H. Heat Transfer. 1998. Experiments on the stabilization of the no-motion state of a fluid layer heated from below and cooled from above. Taylor and Francis. and F.Y. New York. Numerical investigation of the stabilization of the no-motion state of a fluid layer heated from below and cooled from above. Trivedi and S. 363:153–171.H. Marcel Dekker. Macro. Computational Fluid Mechanics and Heat Transfer. 1988.V. [200] V.A. Solid-State Electronics. Engineering Radiation Heat Transfer. Curves and Fractal Dimension. Boiling Heat Transfer and Two-Phase Flow. and R. 1995. DC. 3(10):1409–1419. Heat Transfer. Harper and Row. [194] J.H. MA. Controlling chaos in a thermal-convection loop. Bau. Numerical modeling of thermal behavior of fluid conduit flow with transport delay.S.C. 1997. Journal of Fluid Mechanics. 10(7):1597–1610. and H. [199] C. Pletcher. 1990. Anderson. Intermediate Heat Transfer. Winterton. 10(7):1597–1610. 1982. Rinehart and Winston. [197] C. Physics of Fluids. Taylor and Francis. Washington. Oxford University Press. Washington. Wong. Tzou. 1998.H. Analytical Methods for Heat Transfer and Fluid Flow Problems. Pearton. [203] E. [208] G. Tang and H. 46(5):777–783. [198] L. [212] K. Heat and Mass Transfer. [204] C. Tong and Y.J. Tricot. 1992.B. Tang and H. Journal of Fluid Mechanics. Reading.I.Z. [195] J. Tannehill. Bau.-F. Springer-Verlag. Sen. [211] H.A. Addison-Wesley. Szigeti.

Yu. 53(4-10):533–540. Agent networks for intelligent dynamic control of complex hydronic HVAC building systems–Part 1: Framework for agent network development. Jamshidi. Advances in Heat Transfer. editors. editors.H. and Q. 1992. and G. Academic Press. Nelson.H. Yuen and H. Amsterdam. [216] P.theory and experiments. Greene. Yang and M. volume 37. Leofanti.T. Neural Networks. Irvine.A.J. Marchetti.K.K. 1999. Design of a small deployable satellite. [220] A. 1998. Controlling chaotic convection using neural nets . Hartnett. Y. Boca Raton. 317:91–109. 2002. 98(1):43–48. J.F. Springer. Zilouchian and M. 1996. [218] Z. [221] S. [215] P.-C. C. Bau. 1999. T. Sen. Rendering a subcritical Hopf bifurcation supercritical. and M. 2003. Physics of Fluids. Zhu. 3(1):43–50.I. Intelligent Control Systems Using Soft Computing Methodologies. Journal of Fluid Mechanics. ASHRAE Transactions. Zhang. CRC Press. FL.REFERENCES 220 [213] K. [217] P. 2001.M. S. International Journal on Architectural Science. Yuen and H. Optimal and adaptive control of chaotic convection . Yuen and H. [214] C.theory and experiments. Corradi. pages 179–296. Fu. Acta Astronautica. Bau. In J. London. 11(6):1435–1448.P. 2003. Bau. Allegri. Parametric analysis of a building space conditioned by a VAV system. Zhang and R. Cho.Z. Zumbo.H.K. 11(3):557–569. M. Autotuning of PID Controllers. . G. Optical and thermal radiative properties of semiconductors related to micro/nanotechnology. [219] Z.

127 Matlab. 128 forced convection. 121 fouling. 146 plate. 132 potential flow. 122 maldistribution. 145 Neumann solution. 10 boiling. 129 natural convection. 146 cooling. 146 annular. 174 phonons. 118 nondimensional groups. 144 numerical methods finite difference. 154 Goodman’s integral. 122 radiation. 172 spectral. 133 stagnation-point flow. 19 fin. 2 convection. 122 set Cantor. 10 cavity.Index artificial neural networks. curve. 159 Weierstrass. 10 conduction. 143. 6 correlations. 40 fins radiation. 128 equations Lorenz. 147 computational methods Monte Carlo. 145 porous media. 7 fractals. 17 enclosure. 119 heat exchanger. 154 boiling. 160 equation Brinkman’s. 158 function Knopp. 129 Darcy’s. 13 221 counterflow. 167 Forchheimer’s. 159 Mandelbrot. 147 Leveque’s solution. 143 condensation. 159 genetic algorithms. 39 Reynolds number low. 175 microscale heat transfer. 127 compressible flow. 160 . 8. 35 fin. 131 thermal wakes. 172 finite elements. 146 Maragnoni convection. 162 least squares. 7 nucleation homogeneous. 147 dynamical systems. 13 microchannel. 6 cooling. 128 Duffing. 14. 122 Hurwitz determinants. 14 Fin. 10. 168 extended surfaces.

162 temperatute bulk. 169 thin films. 22 . 169 stability. 23 two-fluid.INDEX 222 singularity theory. 145 two-body. 7 theorem center manifold. 161 global.