Development and Application of a Numerical Framework

for Improving Building Foundation Heat Transfer

Calculations

by

Nathanael J. F. Kruis

B.S., Calvin College, 2007

A thesis submitted to the

Faculty of the Graduate School of the

University of Colorado in partial fulfillment

of the requirements for the degree of

Doctor of Philosophy

Department of Civil, Environmental and Architectural Engineering

2015
 UMI Number: 3704744

All rights reserved

INFORMATION TO ALL USERS

The quality of this reproduction is dependent upon the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.

UMI 3704744
Published by ProQuest LLC (2015). Copyright in the Dissertation held by the Author.
Microform Edition © ProQuest LLC.
All rights reserved. This work is protected against
unauthorized copying under Title 17, United States Code

ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
 This thesis entitled:
Development and Application of a Numerical Framework for Improving Building Foundation Heat
Transfer Calculations
written by Nathanael J. F. Kruis
has been approved for the Department of Civil, Environmental and Architectural Engineering

Prof. Moncef Krarti

Michael Brandemuehl

Harihar Rajaram

Marcus Bianchi

Michael Deru

Date

The final copy of this thesis has been examined by the signatories, and we find that both the
content and the form meet acceptable presentation standards of scholarly work in the above
mentioned discipline.
 iii

Kruis, Nathanael J. F. (Ph.D.)

Development and Application of a Numerical Framework for Improving Building Foundation Heat

Transfer Calculations

Thesis directed by Prof. Moncef Krarti

Heat transfer from building foundations varies significantly in all three spatial dimensions

and has important dynamic effects at all timescales, from one hour to several years. With the ad-

ditional consideration of moisture transport, ground freezing, evapotranspiration, and other physi-

cal phenomena, the estimation of foundation heat transfer becomes increasingly sophisticated and

computationally intensive to the point where accuracy must be compromised for reasonable compu-

tation time. The tools currently available to calculate foundation heat transfer are often either too

limited in their capabilities to draw meaningful conclusions or too sophisticated to use in common

practices.

This work presents Kiva, a new foundation heat transfer computational framework. Kiva pro-

vides a flexible environment for testing different numerical schemes, initialization methods, spatial

and temporal discretizations, and geometric approximations. Comparisons within this framework

provide insight into the balance of computation speed and accuracy relative to highly detailed

reference solutions.

The accuracy and computational performance of six finite difference numerical schemes are

verified against established IEA BESTEST test cases for slab-on-grade heat conduction. Of the

schemes tested, the Alternating Direction Implicit (ADI) scheme demonstrates the best balance

between accuracy, performance, and numerical stability.

Kiva features four approaches of initializing soil temperatures for an annual simulation. A

new accelerated initialization approach is shown to significantly reduce the required years of pres-

imulation.
 iv

Methods of approximating three-dimensional heat transfer within a representative two-dimensional

context further improve computational performance. A new approximation called the boundary

layer adjustment method is shown to improve accuracy over other established methods with a

negligible increase in computation time. This method accounts for the reduced heat transfer from

concave foundation shapes, which has not been adequately addressed to date. Within the Kiva

framework, three-dimensional heat transfer that can require several days to simulate is approxi-

mated in two-dimensions in a matter of seconds while maintaining a mean absolute deviation within

3%.
 v

Acknowledgements

This work was funded in part by an American Society of Heating Refrigeration and Air

conditioning Engineers (ASHRAE) Graduate Student Grant-in-Aid award.

Simulations were performed on the Janus supercomputer, which is supported by the National

Science Foundation (award number CNS-0821794) and the University of Colorado.

This work would not be possible without the help and guidance of my advisor Dr. Moncef

Krarti. You always encouraged me to explore deeper and develop a broader understanding.

I would like to thank Peter Ellis, President of Big Ladder Software, for allocating time and

resources to enable this work. Also, thanks to Joe Huang President of White Box Technologies for

giving me the idea for a faster ground temperature initialization method.

A special thanks to Ben Rohrs, for his friendship, support, and accountability, and Becky

McKay, who generously volunteered time to edit parts of this thesis.

To Dr. Mary Poel and Dr. Rick Kruis, without whom I would not have made it where I have

(let alone existed). Thank you, mom and dad, for your love, support, and the opportunities you

have enabled for me through your own hard work.

Most importantly I want to thank my wife, Cameron Frothingham Kruis, for the support

and encouragement she provided throughout working on this thesis. You demonstrated confidence

in me when I could not find it in myself.

 Contents

Chapter

1 Introduction 1

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Literature Review 5

2.1 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.2 Description of Modeling and Simulation Capabilities . . . . . . . . . . . . . . . . . . 6

2.2.1 Foundation Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2.2 Ground Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2.4 The Whole-Building Context . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.3 Computational Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.3.1 Heat Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.3.2 Moisture Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.3.3 Common Simplifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.3.4 Solution Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.3.5 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.4 Validation, Verification, and Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

 vii

2.4.1 Empirical Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.4.2 Analytical Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.4.3 Comparative Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.5 Comparison of Prominent Tools and Methods . . . . . . . . . . . . . . . . . . . . . . 44

2.5.1 ASHRAE Design Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.5.2 Interzone Temperature Profile Estimation (ITPE) Technique . . . . . . . . . 46

2.5.3 EnergyPlus Preprocessors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.5.4 eQUEST . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.5.5 Ground Heat and Moisture Transfer (GHMT) . . . . . . . . . . . . . . . . . . 48

2.5.6 TRNSYS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3 Tool Development 50

3.1 Kiva . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3.2 Development Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.3 Modeling and Simulation Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.3.1 Foundation Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.3.2 Ground Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.3.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.3.4 The Whole-Building Context . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.4 Computational Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.4.1 Heat Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.4.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.4.3 Finite Difference Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.5 Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

4 Solution Verification 76

4.1 IEA BESTEST for Ground Coupled Heat Transfer . . . . . . . . . . . . . . . . . . . 77

4.1.1 Steady-State Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

 viii

4.1.2 Steady-State Test Case Results . . . . . . . . . . . . . . . . . . . . . . . . . . 86

4.1.3 Harmonic (Unsteady) Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . 90

4.1.4 Harmonic (Unsteady) Test Case Results . . . . . . . . . . . . . . . . . . . . . 96

4.2 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

5 Two-Dimensional Approximation Methods 103

5.1 Current Approximation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5.1.1 Perimeter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5.1.2 Area-to-Perimeter Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

5.1.3 Rounded Rectangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.2 Heat Transfer from Concave Foundation Shapes . . . . . . . . . . . . . . . . . . . . . 110

5.2.1 Opposing Parallel Edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

5.2.2 Concave Corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

5.2.3 Compounding Cavity Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.3 Boundary Layer Adjustment Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

5.3.1 Opposing Parallel Edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.3.2 Concave Corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

5.4 Description of Approximation Method Tests . . . . . . . . . . . . . . . . . . . . . . . 127

5.4.1 Climates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

5.4.2 Foundation Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

5.4.3 Foundation Shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.5 Initial Shape Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

5.6 Approximation Method Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

5.7 Approximation Method Simulation Time . . . . . . . . . . . . . . . . . . . . . . . . . 140

5.8 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

6 Conclusions and Future Work 144

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

 ix

6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

6.2.1 Whole-Building Energy Simulation Tools . . . . . . . . . . . . . . . . . . . . 145

6.2.2 Foundation Design Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

6.2.3 Additional Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

6.2.4 Sensitivity to Model Parameters . . . . . . . . . . . . . . . . . . . . . . . . . 147

6.2.5 Performance Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

Bibliography 149

Appendix

A Finite Difference Discretizations 158

A.1 Continuous Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

A.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

A.3 Discretized Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

A.3.1 Explicit Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

A.3.2 Fully Implicit Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

A.3.3 Crank-Nicolson Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

A.3.4 Alternating Direction Explicit (ADE) Scheme . . . . . . . . . . . . . . . . . . 166

A.3.5 Alternating Direction Implicit (ADI) Scheme . . . . . . . . . . . . . . . . . . 168

A.3.6 Steady-State Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

A.4 Solving the Linear System of Equations . . . . . . . . . . . . . . . . . . . . . . . . . 171

B Mixed Climate Two-Dimensional Approximation Method Test Results 172

C Cold Climate Two-Dimensional Approximation Method Test Results 177

D Hot Climate Two-Dimensional Approximation Method Test Results 184

 x

E Comparisons of Minimum and Maximum Foundation Heat Transfer 191

 Tables

Table

2.1 Example roughness factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2 Terrain parameters for wind speed corrections . . . . . . . . . . . . . . . . . . . . . . 19

2.3 Comparison of prominent foundation heat transfer tools and methods . . . . . . . . 45

3.1 Example foundation types available in Kiva . . . . . . . . . . . . . . . . . . . . . . . 75

4.1 BESTEST steady-state test case specifications . . . . . . . . . . . . . . . . . . . . . 79

4.2 Mesh detail parameters selected by each tool for IEA BESTEST analyses . . . . . . 83

4.3 Selected numerical parameters for Kiva analyses . . . . . . . . . . . . . . . . . . . . 86

4.4 BESTEST steady-state test case results: Slab heat loss [W] . . . . . . . . . . . . . . 88

4.5 BESTEST steady-state test case simulation wall times [s] . . . . . . . . . . . . . . . 88

4.6 BESTEST harmonic test cases specifications . . . . . . . . . . . . . . . . . . . . . . 90

4.7 BESTEST GC40a initialization method percent difference from “ideal” for seven

years of simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4.8 BESTEST harmonic (unsteady) test case results: Annual slab heat loss [MWh] . . . 96

4.9 BESTEST GC40 harmonic parameters for each solution (shading indicates deviation

from the median value for each case) . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.10 ADE time-step size for harmonic BESTEST cases . . . . . . . . . . . . . . . . . . . 99

4.11 BESTEST harmonic (unsteady) test case results: Simulation wall time [s] . . . . . . 101

5.1 Approximation method testing conditions . . . . . . . . . . . . . . . . . . . . . . . . 128

 xii

5.2 Compactness of various shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

5.3 List of two-dimensional approximation methods tested . . . . . . . . . . . . . . . . . 134

5.4 Average simulation wall time for two-dimensional approximation methods . . . . . . 140

6.1 Recommended methods and numerical parameters for foundation heat transfer . . . 145
 Figures

Figure

2.1 Types of building foundations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Examples of insulation designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.3 Examples of thermal bridges for a slab-on-grade foundation . . . . . . . . . . . . . . 10

2.4 Illustration of ITPE zones (1, 2, and 3) for a basement foundation (adapted from

Krarti et al. [70]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.5 Concentric heat flow paths (adapted from Latta and Boileau [79]) . . . . . . . . . . 49

3.1 Two-dimensional context for defining foundation design in Kiva . . . . . . . . . . . . 54

3.2 Insulation and structural design elements in Kiva . . . . . . . . . . . . . . . . . . . . 56

3.3 Plan view illustration of far-field boundary definition in three-dimensions and foun-

dation shape vertex definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.4 Discretization regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.5 Example Kiva discretizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.6 Boundary cells and resulting zero-thickness cells. Black nodes = normal solid cells,

white nodes = boundary cells, gray nodes = zero-thickness cells, hollow nodes = air

cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.7 Example finite difference scheme (black nodes are known values, white nodes are

unknown values) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.8 Illustration of simple explicit finite difference scheme . . . . . . . . . . . . . . . . . . 65

 xiv

3.9 Illustration of fully implicit finite difference scheme . . . . . . . . . . . . . . . . . . . 66

3.10 Illustration of Crank-Nicolson finite difference scheme . . . . . . . . . . . . . . . . . 67

3.11 Illustration of ADE finite difference scheme: upward sweep . . . . . . . . . . . . . . 68

3.12 Illustration of ADE finite difference scheme: downward sweep . . . . . . . . . . . . . 69

3.13 Typical implicit matrix sparsity pattern for three dimensions (up to seven unknowns

per equation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.14 ADI matrix sparsity pattern for three dimensions (up to three unknowns per equation) 71

3.15 Example Kiva graphical output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.1 BESTEST GC10a mesh growth near the wall boundary . . . . . . . . . . . . . . . . 81

4.2 BESTEST GC10a slab heat loss sensitivity to mesh detail parameters . . . . . . . . 82

4.3 BESTEST GC10a simulation wall time sensitivity to mesh detail parameters . . . . 82

4.4 BESTEST GC10a slab heat loss sensitivity to boundary distances . . . . . . . . . . 84

4.5 BESTEST GC10a simulation wall time sensitivity to boundary distances . . . . . . 84

4.6 BESTEST GC10a slab heat loss and simulation wall time sensitivity to linear solver

tolerance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.7 BESTEST steady-state test case slab heat loss . . . . . . . . . . . . . . . . . . . . . 87

4.8 BESTEST steady-state test case simulation wall times (note logarithmic y-axis) . . 88

4.9 BESTEST GC40a comparison of initialization methods: First year slab heat loss . 92

4.10 BESTEST GC40a comparison of initialization methods: Annual slab heat loss for

seven years after initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

4.11 BESTEST GC40a comparison of first year slab heat loss for traditional and accel-

erated initialization methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.12 BESTEST harmonic (unsteady) test case annual slab heat loss (results for GC50b

are divided by 10 to maintain a consistent scale) . . . . . . . . . . . . . . . . . . . . 97

4.13 BESTEST GC40a harmonic (unsteady) test case hourly results . . . . . . . . . . . . 98

4.14 BESTEST harmonic (unsteady) test case simulation wall times . . . . . . . . . . . . 100
 xv

5.1 Typical context for two-dimensional approximations . . . . . . . . . . . . . . . . . . 104

5.2 Characteristic length of a circle in a cylindrical coordinate system . . . . . . . . . . 107

5.3 Characteristic length of a infinite rectangle in a Cartesian coordinate system . . . . 108

5.4 Dimensions of a rounded rectangle with equivalent area and perimeter . . . . . . . . 110

5.5 Determination of cavities for various shapes . . . . . . . . . . . . . . . . . . . . . . . 111

5.6 Plan view cross-section at grade height of heat flux near a concave slab corner (ex-

terior grade in the upper left) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

5.7 Plan view of parallel concave edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

5.8 Reduction in heat loss between two parallel concave edges over a range of area-to-

perimeter ratios (ksoil = 1.0 W/m·K, R-10 insulation) . . . . . . . . . . . . . . . . . 113

5.9 Reduction in heat loss between two parallel concave edges over a range of soil con-

ductivity (A/P = 4.6 m, R-10 insulation) . . . . . . . . . . . . . . . . . . . . . . . . 114

5.10 Reduction in heat loss between two parallel concave edges over a range of insulation

R-values [hr·ft2 ·R/Btu] (ksoil = 1.0 W/m·K, A/P = 4.6 m) . . . . . . . . . . . . . . 115

5.11 Illustration of a heat flux boundary layer profile for a slab with exterior horizontal

insulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

5.12 Example boundary layer profile for a slab with exterior horizontal insulation (ksoil =

1.0 W/m·K, A/P = 4.6 m, R-10 insulation) . . . . . . . . . . . . . . . . . . . . . . . 117

5.13 Shapes with equal area and perimeter but different number of concave corners . . . . 118

5.14 Reduction in heat loss due to corners over a range of area-to-perimeter ratios (ksoil =

1.0 W/m·K, R-10 insulation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5.15 Reduction in heat loss due to corners over a range of soil conductivity (A/P = 4.6

m, R-10 insulation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5.16 Reduction in heat loss due to corners over a range of insulation R-values (ksoil = 1.0

W/m·K, A/P = 4.6 m) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

5.17 Illustration of corner bisector and unaffected edge boundary profile cross-sections . . 121

5.18 Comparison of straight boundary profile and corner bisector . . . . . . . . . . . . . . 122

 xvi

5.19 Heat loss reduction from H-shaped slabs . . . . . . . . . . . . . . . . . . . . . . . . 123

5.20 Variables used to adjust perimeter of opposing parallel edges . . . . . . . . . . . . . 125

5.21 Variables used to adjust perimeter of concave corners . . . . . . . . . . . . . . . . . . 126

5.22 Annual outdoor dry-bulb temperatures from weather data files for selected climates.

Hot = Phoenix, Arizona; Mixed = Albuquerque, New Mexico; Cold = Anchorage,

Alaska . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.23 Slab insulation configurations for approximation method tests . . . . . . . . . . . . . 130

5.24 Crawlspace insulation configurations for approximation method tests . . . . . . . . . 131

5.25 Basement insulation configurations for approximation method tests . . . . . . . . . . 132

5.26 Shapes simulated in three dimensions for approximation method testing. Each plot

is 200 m × 200 m. A = Area, C = Compactness. . . . . . . . . . . . . . . . . . . . 134

5.27 Comparison of heat flux from a range of uninsulated foundation shapes . . . . . . . 135

5.28 Heat flux reduction relative to uninsulated foundations in mixed climate for a range

of foundation shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

5.29 Comparison of annual total heat transfer between three-dimensions and two-dimensional

approximations for mixed climate simulations . . . . . . . . . . . . . . . . . . . . . . 137

5.30 Comparison of annual total heat transfer between three-dimensions and two-dimensional

approximations for cold climate simulations . . . . . . . . . . . . . . . . . . . . . . . 138

5.31 Comparison of annual total heat transfer between three-dimensions and two-dimensional

approximations for hot climate simulations . . . . . . . . . . . . . . . . . . . . . . . 139

5.32 Example comparison of hourly three-dimensional heat transfer and approximated

two-dimensional heat transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

5.33 Two-dimensional approximation simulation wall times . . . . . . . . . . . . . . . . . 141

5.34 Deviation from three-dimensional annual heat transfer vs. simulation wall time for

different combinations of discretization parameters . . . . . . . . . . . . . . . . . . . 142

 xvii

B.1 Percent difference in annual heat transfer between three-dimensional square founda-

tion shapes and two-dimensional approximations . . . . . . . . . . . . . . . . . . . . 173

B.2 Percent difference in annual heat transfer between three-dimensional cross-shaped

foundations and two-dimensional approximations . . . . . . . . . . . . . . . . . . . . 174

B.3 Percent difference in annual heat transfer between three-dimensional narrow-H foun-

dation shapes and two-dimensional approximations . . . . . . . . . . . . . . . . . . . 175

B.4 Percent difference in annual heat transfer between three-dimensional wide-H foun-

dation shapes and two-dimensional approximations . . . . . . . . . . . . . . . . . . . 176

C.1 Comparison of heat flux from a range of uninsulated foundation shapes in the cold

climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

C.2 Heat flux reduction relative to uninsulated foundations in mixed climate for a range

of foundation shapes in the cold climate . . . . . . . . . . . . . . . . . . . . . . . . . 179

C.3 Percent difference in annual heat transfer between three-dimensional square founda-

tion shapes and two-dimensional approximations in the cold climate . . . . . . . . . 180

C.4 Percent difference in annual heat transfer between three-dimensional cross-shaped

foundations and two-dimensional approximations in the cold climate . . . . . . . . . 181

C.5 Percent difference in annual heat transfer between three-dimensional narrow-H foun-

dation shapes and two-dimensional approximations in the cold climate . . . . . . . . 182

C.6 Percent difference in annual heat transfer between three-dimensional wide-H foun-

dation shapes and two-dimensional approximations in the cold climate . . . . . . . . 183

D.1 Comparison of heat flux from a range of uninsulated foundation shapes in the hot

climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

D.2 Heat flux reduction relative to uninsulated foundations in mixed climate for a range

of foundation shapes in the hot climate . . . . . . . . . . . . . . . . . . . . . . . . . 186

D.3 Percent difference in annual heat transfer between three-dimensional square founda-

tion shapes and two-dimensional approximations in the hot climate . . . . . . . . . . 187

 xviii

D.4 Percent difference in annual heat transfer between three-dimensional cross-shaped

foundations and two-dimensional approximations in the hot climate . . . . . . . . . 188

D.5 Percent difference in annual heat transfer between three-dimensional narrow-H foun-

dation shapes and two-dimensional approximations in the hot climate . . . . . . . . 189

D.6 Percent difference in annual heat transfer between three-dimensional wide-H foun-

dation shapes and two-dimensional approximations in the hot climate . . . . . . . . 190

E.1 Comparison of annual minimum heat transfer between three-dimensions and two-

dimensional approximations for mixed climate simulations . . . . . . . . . . . . . . . 192

E.2 Comparison of annual maximum heat transfer between three-dimensions and two-

dimensional approximations for mixed climate simulations . . . . . . . . . . . . . . . 193

E.3 Comparison of annual minimum heat transfer between three-dimensions and two-

dimensional approximations for cold climate simulations . . . . . . . . . . . . . . . . 194

E.4 Comparison of annual maximum heat transfer between three-dimensions and two-

dimensional approximations for cold climate simulations . . . . . . . . . . . . . . . . 195

E.5 Comparison of annual minimum heat transfer between three-dimensions and two-

dimensional approximations for hot climate simulations . . . . . . . . . . . . . . . . 196

E.6 Comparison of annual maximum heat transfer between three-dimensions and two-

dimensional approximations for hot climate simulations . . . . . . . . . . . . . . . . 197

 Chapter 1

Introduction

1.1 Motivation

Building energy modeling has become an integral part of the design process for both new

building projects and existing building retrofit projects. The tools used to simulate building per-

formance allow designers to make informed, economic decisions on how to best invest in energy

efficient improvements to buildings, even in the earliest planning stages.

Simulation tools can aid in the design of building foundations, just as they serve to inform

the design of other components in buildings (walls, windows, lighting, HVAC equipment, etc.).

To accomplish this, simulation tools must be capable of estimating the heat loss through building

foundations under a range of conditions and insulation designs.

Heat loss from building foundations varies significantly in all three spatial dimensions and

has important dynamic effects at all timescales, from one hour to several years. With the additional

consideration of moisture transport, ground freezing, evapotranspiration, and other physical phe-

nomena, the estimation of foundation heat transfer becomes increasingly sophisticated and compu-

tationally intensive to the point where accuracy must be compromised for reasonable computation

time.

Traditional approaches to foundation heat transfer tend to be oversimplified or neglect the

problem entirely because the overall magnitude of foundation heat fluxes is considered to be small

relative to the other thermal loads in a building. However, as buildings become progressively more

efficient, foundation heat loss begins to represent a larger portion of the total building load, thus
 2

increasing the importance of accuracy in the calculations. It is estimated that heat transfer through

foundations can account for 30%–50% of the total heating load in certain buildings [39].

Unfortunately, the tools currently available to calculate foundation heat transfer are often

either too simplistic and limited in their capabilities to draw meaningful conclusions from the results

or too sophisticated to use in common practices.

A common practice to address the computational complexity of foundation heat transfer is

to ignore the problem altogether. Many practitioners use an adiabatic boundary condition for

ground-coupled building surfaces in their models. As a result, the foundation of the building

remains underdesigned relative to the rest of the energy-related features captured in the building

simulation.

Another practice is to avoid simulating the heat transfer from the building foundation by

simply overdesigning the foundation insulation. The Passivhaus strategy can specify up to 25.4

cm (10 in) of rigid foam insulation beneath the entire slab foundation [83]. Although some level

of design does go into the Passivhaus specification, the level of detail and accuracy involved in the

analysis is suspect, especially considering the economic implications.

In one extreme, there is the possibility of unrealized energy savings, and in the other, the

builder is likely over-paying for the diminishing returns of additional insulation.

Heat transfer from building foundations remains a large uncertainty in the estimation of

building energy use. An informal survey of energy modeling professionals indicates that the current

practice with regard to modeling foundation heat transfer is either oversimplified or ignored entirely

leading to low confidence in both simulation results and foundation design decisions. Many modelers

are under the impression that foundation heat transfer is negligible and, therefore, justly ignored.

However, when seeking a reference for such a claim, the author has found no such evidence.

The research presented in this thesis holds the following premise: Heat transfer through

building foundations is not always negligible and, therefore, warrants the careful consideration of

insulation design alternatives.

 3

1.2 Objectives

The primary hypothesis of this research is as follows: A foundation heat transfer simulation

tool can be developed to balance capability, complexity, computational performance, and algorith-

mic confidence to provide building designers with a tool that will allow them to make informed

decisions on how to insulate building foundations.

The list below states the research questions that are answered within the content of this

thesis:

(1) What algorithmic assumptions and simplifications can reduce computation time while re-

taining accuracy and the more important modelling capabilities?

(2) What are the accuracy and speed implications of these assumptions and simplifications?

To answer the research questions, the following objectives must be met:

(1) Develop a detailed numerical truth standard as a benchmark for all subsequent assumptions

and simplifications.

(2) Recommend a set of assumptions and simplifications to perform accurate, computationally-

efficient calculations of building foundation heat transfer within the context of whole-

building energy analysis.

1.3 Organization of the Thesis

This thesis is organized into five chapters:

Chapter 2: Literature Review: A review of the foundation heat transfer methods and tools

found in literature, including common capabilities, computational algorithms, and attempts

to evaluate tool accuracy.

Chapter 3: Tool Development: A description of the tool developed for the work in this thesis.
 4

Chapter 4: Solution Verification: Verification of the six solution methods implemented in said

tool. This chapter also investigates solution sensitivity to selected numeric parameters and

the efficiency of several initialization techniques within the context of a well-defined test

specification.

Chapter 5: Two-Dimensional Approximation Methods: Approaches reducing computation

time by approximating three-dimensional heat transfer using two-dimensional calculations.

Chapter 6: Conclusions and Future Work: Concluding statements and suggestions for future

work.
 Chapter 2

Literature Review

This chapter contains four sections, each describing a different aspect of foundation heat

transfer simulations:

(1) Terminology

(2) Description of Modeling and Simulation Capabilities

(3) Computational Algorithms

(4) Validation, Verification, and Testing

A final section provides a comparison of prominent methods and tools found in both the

literature and practice.

2.1 Terminology

For the sake of consistency, standard terminology used throughout this thesis is defined below.

These definitions are adapted from Fritzson [45] and Judkoff and Neymark [60]:

System: A system is an object or collection of objects whose properties are studied. The system

under consideration in this thesis is the physical construction of the building foundation

and the semi-infinite domain of ground surrounding the foundation.

 6

Model: A model is an approximate representation of a system. The model is the set of input

and output data values relevant to the study of the system. In this case, the model is the

quantitative and qualitative description of the foundation and ground domain.

Simulation: Simulation is the process of applying controlled inputs to the model and observing

the resulting outputs.

Solution Method: The solution method describes the algorithmic translation between model in-

put and model output. For a foundation and ground system, the solution method generally

refers to how the conduction-diffusion partial differential equations are solved including the

algorithms that define the boundary conditions of the system.

Tool: A software implementation of a solution method (or multiple methods) capable of performing

simulations on a defined model.

2.2 Description of Modeling and Simulation Capabilities

The following sections provide a review of modeling capabilities, both available in and missing

from current foundation heat transfer tools. This informs which capabilities are important for the

development of new tools and the challenges associated with implementing them.

2.2.1 Foundation Design

Foundation design capabilities are the tool’s ability to define the physical construction details

and engineering design of the foundation system.

2.2.1.1 Foundation Types

Figure 2.1 shows a range of foundation types found throughout the architectural domain.

Most tools can model the three standard foundation types (Figures 2.1(a)–2.1(c)). Even among

these common foundation types, tools are often limited to specific foundation depths. Few tools

have the capability to model less-common foundation types (Figures 2.1(d)–2.1(g)).

 7

(a) Basement (b) Crawlspace (c) Slab-on-grade

(d) Walk-out basement (e) Split-level (f) Sub-basement (g) Underground structure

Figure 2.1: Types of building foundations

A few case studies in literature investigate heat loss from underground structures [23, 84] and

slabs with a stepped-exterior grade [22], but these capabilities are rarely integrated into publicly

available tools.

Accommodating many of these additional foundation types requires a more detailed descrip-

tion of a three-dimensional domain.

2.2.1.2 Insulation Design

For a given foundation type, a continuous range of potential foundation designs exists. Figure

2.2 shows examples of typical insulation designs for foundations. While Figure 2.2 represents only

a small subset of the possible insulation designs, the full parameter space includes combinations of

values of the following variables:

• The insulation material

• The level of insulation (insulation thickness)

 8

• The location of the insulation (whole slab, interior horizontal, interior vertical, slab gap,

interior vertical, in-wall, exterior vertical, exterior horizontal, and all combinations thereof)

• Insulation dimensions (depth of vertical insulation, width and depth of horizontal insula-

tion)

(a) Uninsulated (b) Whole-slab insulation (c) Interior and slab gap in-
sulation

(d) Interior half-wall insula- (e) Interior wall insulation (f) Interior wall and floor in-
tion sulation

(g) Insulated concrete forms (h) Exterior wall insulation (i) Exterior vertical and hor-
izontal insulation

Figure 2.2: Examples of insulation designs

 9

Many simulation tools only model a discrete set of insulation designs either because they do

not generalize a meshing strategy for a continuous range of dimensions or because they are based

on regression models for a specific subset of designs.

2.2.1.3 Foundation Shape

Relatively few researchers have investigated the impact of foundation shape on heat loss.

Many of the tools used today apply one-dimensional or two-dimensional approximations based on

the area and perimeter of the foundations. Few tools have the capability to simulate heat transfer

from non-rectangular three-dimensional foundation shapes. The capability to simulate a wider

range of foundation shapes in three dimensions is necessary to justify approximations with fewer

dimensions. Descriptions of analyses of dimensional reduction strategies found in literature are

discussed in more detail in Chapter 5.

2.2.1.4 Multiple Interior Thermal Zones

Common simplifications for foundation heat transfer calculations only account for heat trans-

fer from the surfaces of a single thermal zone, neglecting heat transfer between thermal zones

through the ground.

Capturing these effects requires an explicit three-dimensional model with distinct boundary

conditions representing each thermal zone adjacent to a building’s foundation.

Few examples of foundation heat transfer between multiple interior thermal zones exist in the

literature. The temperature difference between interior thermal zones is often not great enough to

cause significant heat transfer. Xie et al. [119] and McDowell et al. [86] describe tools with multiple

interior zone boundaries; however, they do not quantify the magnitude of heat transfer between

the zones.
 10

2.2.1.5 Thermal Bridging

As with any construction element in a building, thermal bridging in foundations is a sig-

nificant concern. The placement of insulation to eliminate potential heat flow paths through the

most conductive building materials is an important design consideration. In foundations, thermal

bridging is common along the perimeter—at the interface of the above-grade foundation wall with

the exterior environment. Many studies investigate the optimal configuration of insulation on both

exterior and interior portions of the foundation wall to minimize heat loss through thermal bridges

such as the edge of the slab and sill plate (Figure 2.3).

The construction elements near the perimeter, which lead to thermal bridges, require a high-

resolution description of the foundation details. Such high resolutions are time consuming to model

and simulate. Many of the computational methods employed in foundation heat transfer models

do not describe the geometry at a resolution that adequately represents thermal bridging.

(a) Through slab edge (b) Through sill plate

Figure 2.3: Examples of thermal bridges for a slab-on-grade foundation

Wang [115] found that neglecting the heat transfer through the sill plate interface between

the above-grade and below-grade walls can result in a 46% underprediction in the basement wall
 11

heat transfer. He also found that exterior foundation wall insulation can have a 25%–32% difference

in heat transfer from interior insulation.

A number of other potential thermal bridges exist along foundation perimeters. Rock [99]

explored the effect of expansion joints on thermal bridging through the slab edge.

2.2.1.6 Impacts on Above-Grade Walls

Most simulation algorithms assume one-dimensional heat transfer for above-grade walls.

However, select studies show that the two-dimensional effects near the interface of above-grade

walls and the foundation can have a significant impact on the overall building heat balance [67].

2.2.1.7 System/Ground Interaction

There are a number of systems that are designed to exchange heat with the ground near

the building. These systems are rarely considered when calculating the heat transfer through the

building foundation. Examples of such systems are listed below:

• Snow-melt systems [92, 82]

• In-slab radiant heating

• Foundation heat exchangers [32]

• Ground-source heat pumps

Inclusion of such systems in the foundation heat transfer model requires an iterative solution

to ensure that the corresponding heat fluxes across the boundary between the system and the

ground domain are balanced.

 12

2.2.2 Ground Domain

2.2.2.1 Soil Spatial Variation

One of the largest challenges in creating a ground domain model is establishing the thermal

properties of the soil. The thermal properties at any location within the ground domain are largely

determined by the soil composition. Soil is typically comprised of up to six components:

• Quartz

• Clay minerals

• Organic matter

• Water

• Ice

• Air

The relative fraction of each of these components can vary significantly throughout the ground

domain. Because these variations are both difficult to measure and burdensome to describe within

a model, a majority of the work found in literature assumes that soil properties are spatially

homogeneous.

A few authors have explored the impact of spatial variation on foundation heat transfer calcu-

lations. Krarti [65] explored the effect of lateral variation in soil types and Rees et al. [96] explored

the effect of vertical variation of moisture content based on a static one-dimensional moisture pro-

file. Deru [39] demonstrated that spatial variations of moisture content can be approximated using

different properties deeper in the domain than near the surface.

The moisture distribution in soils is not static but instead is driven by determinable forces.

Simulating the actual distribution of moisture throughout the ground domain requires calculation

of moisture throughout the domain.

 13

2.2.2.2 Soil Moisture Content

Few authors have explored the impact of both heat and moisture balances on foundation

heat transfer. However, two notable contributions, Deru [39] and Janssen [56], have developed

solutions incorporating a moisture balance that drives spatial and temporal changes to the soil

thermal properties.

Authors who have compared heat-and-moisture-transfer calculations to heat-transfer-only

calculations have found that the heat-transfer-only calculation underpredicts foundation heat trans-

fer by 9%–40% [102, 39, 57].

When simulating the effect of rainfall, authors have shown disparate results. Though most

published results range between 0.1% and 10% difference from heat-transfer-only calculations [39,

52, 56], dos Santos and Mendes [42] show up to a 100% difference.

2.2.2.3 Soil Freezing and Thawing

An important aspect of structural design for building foundations is the construction of the

foundation footer below the frost line. Though it is a common consideration in structural design,

the impact of ground freezing is rarely considered in building energy analysis. Freezing provides the

building with additional thermal storage in the form of latent energy and influences the apparent

thermal conductivity of the soil. Ice within soil voids also impedes moisture transfer.

A number of authors have discussed heat and mass transfer in frozen soils, though the theory

has primarily been applied to frost heaving and structural design [61, 81]. Few foundation heat

transfer tools calculate freezing, and the overall impact of freezing has not yet been thoroughly

quantified.
 14

2.2.3 Boundary Conditions

2.2.3.1 Interior Convection

Interior convection is usually assumed to be buoyancy driven and is thus a function of the

temperature difference between the surface and the nearby air. Though several correlations exist

for defining natural convection coefficients, the correlations developed by Walton [113] are widely

accepted and account for different surface orientations:

⎧
⎪
⎪ c1 |Ts − Tair |1/3
⎨ (for unstable heat flow)
hn = c2 − |cos β|
1/3 (2.1)
⎪
⎩ c3 |Ts − Tair |
⎪ (for stable heat flow),
c4 + |cos β|
where

• hn is the natural convection coefficient [W/m2 -K],

• Ts is the surface temperature [K],

• Tair is the ambient air temperature [K]

• β is the tilt of the surface (from 0 at horizontal facing up, to π at horizontal facing down)

[radians],

• c1 is a correlation constant of 9.482 [W/m2 -K4/3 ],

• c2 is a correlation constant of 7.283,

• c3 is a correlation constant of 1.810 [W/m2 -K4/3 ], and

• c4 is a correlation constant of 1.328.

Unstable heat flow refers to situations where the heat flow is upward and enhances buoyancy—the

surface is tilted up and is hotter than the ambient air, or the surface is tilted down and is colder

than the ambient air. Stable heat flow refers to situations where the heat flow is downward.

Because the convective coefficient is a non-linear function of the surface temperature, an

exact solution requires an iterative approach. Some tools calculate the convection coefficient using
 15

the surface temperature from a previous timestep or set the convection coefficient to a constant

value.

The ambient air temperature is also determined by a heat balance performed on the adjacent

thermal zone. For the case of conditioned spaces, it is acceptable to assume a constant space

temperature; however, in cases where the space is unconditioned or the HVAC system is not able

to meet the set-point, the resulting ambient and surface temperatures must be calculated using an

iterative approach.

There are relatively few existing models that incorporate a direct coupling between the ground

domain and the thermal zones of the building.

2.2.3.2 Interior Long-Wave Radiation

Long-wave radiation is exchanged among the interior surfaces of a room based on the temper-

atures, emissivities of the surfaces, and the view factors between the surfaces. Calculating radiation

exchange becomes increasingly expensive with more surfaces, as it requires calculating all the sur-

face temperatures simultaneously in conjunction with the convective and conductive heat transfer

calculations. The EnergyPlus simulation engine [111] performs these calculations using an algebraic

method called the script-F method described by Hottel and Sarofim [53]. Cogil [30] developed a

tool for calculating heat transfer from basements, which used a similar algebraic method using

surface radiosities. Alternative algebraic methods are described by Clark and Korybalski [28].

Most tools do not solve the set of algebraic equations required for a complete radiation heat

balance. A common simplification is to treat all radiation from a surface as though it were only

radiating to a black body at the air temperature. This is is written as

 
q̇rad,lw = εs σ Ts4 − Tair
4
(2.2)

or as a heat transfer coefficient,

 
hrad = εs σ Ts2 + Tair
2
[Ts + Tair ] , (2.3)

where
 16

• q̇rad is the radiative heat flux from the surface [W/m2 ],

• εs is the surface emissivity,

• σ is the Stefan-Boltzmann constant of 5.67×10−8 [W/m2 -K4 ],

• Ts is the surface temperature [K],

• Tair is the air temperature [K], and

• hrad is the radiative heat transfer coefficient [W/m2 -K].

Like the natural convection coefficient, the radiative coefficient is a non-linear function of the

surface temperature. To calculate radiation within the context of a linear system of equations,

the coefficient is calculated using the previous time-step’s surface temperature at the loss of some

accuracy.

Many tools, for example DOE-2 [80], are not capable of simulating dynamic coefficients and

use a single constant coefficient to represent both convection and radiation from a surface.

2.2.3.3 Interior Short-Wave Radiation

Interior foundation surfaces are sometimes exposed to short-wave solar radiation transmitted

through windows. Passive solar buildings are designed and oriented specifically to expose slab

foundation surfaces to incoming solar radiation to store heat. Most whole-building simulation tools

lack the coupling between the foundation heat transfer and interior short-wave radiation to capture

the benefits of passive solar building designs.

The effect of short-wave radiation incident on interior surfaces is calculated using a constant

heat flux. However, calculating the effect of transmitted solar radiation on foundation surfaces

requires integration with a more comprehensive building simulation tool to capture the effects of

shading and transmission losses through windows.

 17

2.2.3.4 Exterior Convection

Exterior convection is typically dominated by forced, wind-driven convection, though there

is also a natural buoyancy-driven component. Numerous convection correlations appear in the lit-

erature with varying complexity. EnergyPlus combines several correlations to account for a wide

range of conditions into the somewhat misleadingly termed the “DOE-2 Model.” The algorithm

combines the MoWiTT correlations developed by Yazdanian and Klems [120] to account for wind

speed and wind direction, the natural convection algorithm described by Walton [113], some ad-

justments made by Booten et al. [18] to correct for the height of the surface, and an addition from

the DOE-2 to account for the roughness of the surface:

hconv = hn + Rf [hconv,smooth − hn ] , (2.4)

where

• hconv is the total convection coefficient [W/m2 -K],

• hn is the natural convection coefficient defined by Equation 2.1 [W/m2 -K],

• Rf is a dimensionless roughness factor, and

• hconv,smooth is the convection coefficient for a perfectly smooth surface [W/m2 -K] defined

by Equation 2.5:


hconv,smooth = h2n + h2f , (2.5)

where hf is defined by

hf = av b , (2.6)

where

• hf is the forced convection coefficient [W/m2 -K],

• v is the wind speed at the surface [m/s],

 18

• a is a correlation constant of

◦ 3.26 [W/m2 -K-(m/s)b ] for windward surfaces,

◦ 3.55 [W/m2 -K-(m/s)b ] for leeward surfaces, and

• b is a correlation constant of

◦ 0.89 for windward surfaces,

◦ 0.617 for leeward surfaces.

The roughness factor described in Equation 2.4 is a qualitatively defined factor based on example

surfaces as shown in Table 2.1.

Table 2.1: Example roughness factors

Example Surface Rf Estimated Rt [mm]

Glass 1.00 0.00
Smooth plaster 1.11 0.44
Clear pine 1.13 0.52
Concrete 1.52 2.08
Brick 1.67 2.68
Stucco 2.17 4.68

No references were found to provide surface roughness factors for roughnesses beyond the

range shown here. As none of the example surfaces describe typical ground coverings, the roughness

factor will have to be extrapolated to account for soils, asphalts, grasses, and other vegetation. As

a crude approximation, the roughness factors of other surfaces beyond the examples given here are

estimated by

Rf = 1 + Rt /Rn , (2.7)

where

• Rt is the maximum peak-to-valley surface relief [mm], and

• Rn is a normalizing constant of 4 [mm].

 19

The resulting estimated value of Rt for each of the examples is also shown in Table 2.1.

Wind speed values are typically measured at a weather station in an open area several meters

above any nearby obstructions and are not appropriate for calculating forced convection near the

ground adjacent to a building. The wind speed, v, in Equation 2.6 is often adjusted from the

weather station measurements provided in standard weather data sets using a procedure from the

ASHRAE Handbook of Fundamentals [7]:

αws αlocal
δws zveg
vground = vws , (2.8)
zws δlocal

where

• vground is the wind speed near the ground [m/s],

• vws is the wind speed recorded at the weather station [m/s],

• δws and δlocal are the boundary layer thicknesses [m] (Table 2.2) at the weather station and

the local building, respectively,

• zws and zveg are the heights [m] of the weather station and the local vegetation, respectively,

and

• αws and αlocal are the terrain exponents (Table 2.2) at the weather station and the local

building, respectively.

Table 2.2: Terrain parameters for wind speed corrections

Terrain Description δ [m] α

Ocean 210 0.10
Flat, open country 270 0.14
Towns, cities and forests 370 0.22
Urban centers 460 0.33
 20

2.2.3.5 Exterior Long-Wave Radiation

Long-wave radiation is exchanged among exterior building surfaces, the ground surface, the

outdoor air, and the sky. The exterior long-wave radiation exchange calculation requires that each

component has a defined temperature.

The air temperature is often provided by weather data for the building location, and the

ground surface temperatures are calculated through a heat balance. The ground and building

surface temperatures are rarely, if ever, solved simultaneously using algebraic methods similar to

those described for interior long-wave radiation. Instead, long-wave radiation from exterior building

surfaces is calculated assuming that the ground is at the same temperature as the air, and long-

wave radiation from the ground is calculated assuming that the ground does not have a view-factor

to the building’s surfaces.

Sky temperature is often correlated to information available in standard weather data sets.

The correlation used in both EnergyPlus and DOE-2 is shown in Equation 2.9:

1/4
Tsky = εsky Tair , (2.9)

where

• Tsky is the sky temperature [K],

• εsky is the sky emissivity [0–1] defined by Equation 2.10, and

• Tair is the exterior air temperature [K].

Tdewpoint  
εsky = 0.787 + 0.764 ln 1 + 0.0224fc − 0.0035fc2 + 0.00028fc3 , (2.10)
T air

where

• Tdewpoint is the dew point temperature of the air [K], and

• fc is the cloud covered fraction of the sky [tenths (i.e., 0–10)]

 21

The exterior long-wave radiative flux is then given by the following:

 
q̇rad,lw = εs σ Ts4 − Fsky fs Tsky
4 4
+ Fsky [1 − fs ] Tair 4
+ Fground Tground , (2.11)

where

• εs is the emissivity of the surface [0–1],

• σ is the Stefan-Boltzmann constant of 5.67×10−8 [W/m2 -K4 ],

• Ts is the surface temperature [K],

• Fsky is the view factor from the surface to the sky [0–1],

• fs is the fraction defined by Equation 2.12, which distinguishes radiation to the sky from

radiation to the air [0–1],

• Fground is the view factor from the surface to the ground [0–1], and

• Tground is the ground surface temperature [K].

fs = cos (β/2) (2.12)

The view factors can be calculated as follows:

Fsky = 0.5 [1 + cos β] , (2.13)

and

Fground = 0.5 [1 − cos β] (2.14)

where β is the tilt of the surface in radians (from 0 at horizontal facing up, to π at horizontal facing

down).

The radiative flux can be simplified when substituting the definition of sky temperature and

assuming that the ground temperature is the same as the air temperature:
 
q̇rad,lw =εs σ Ts4 − Fsky fs εsky Tair
4 4
+ Fsky [1 − fs ] Tair 4
+ [1 − Fsky ] Tair
 
=εs σ Ts4 − Tair
4
F∗ , (2.15)
 22

where F ∗ is a placeholder for the more verbose [Fsky fs [εsky − 1] + 1].

Defining the radiative heat transfer coefficient as

 
hrad = εs σ Ts2 + Tair
2
F ∗ 1/2 Ts + Tair F ∗ 1/4 (2.16)

while using the surface temperature from the previous time-step, the calculation of exterior radia-

tion can be solved in a linear system of equations:


q̇rad,lw = hrad Ts − Tair F ∗ 1/4 . (2.17)

2.2.3.6 Exterior Short-Wave Radiation

Short-wave solar radiation is calculated as a heat flux on the ground surface. Many tools

will simulate solar incidence on the ground, but often the solar absorptivity of the ground surface

is assumed to be homogeneous and does not account for shading from building surfaces or ground

cover.

A couple authors have explicitly simulated the effect of shading on the ground in their tools.

Bahnfleth and Pedersen [11] showed that shade from the building can increase the mean heat loss

from a slab by 27.4% in Phoenix, Arizona and 14% in Medford, Oregon. dos Santos and Mendes [41]

estimated that shading can cause up to a 30◦ C range of variation in the ground surface temperatures

surrounding the building.

2.2.3.7 Evapotranspiration

Evapotranspiration describes the evaporation of water from ground and plant surfaces. Many

models do not incorporate evapotranspiration despite its cooling effect on the exterior ground

surface boundary.

The rate of evaporative heat loss from the soil depends on the vapor pressure at the soil’s

surface relative to the vapor pressure of the air. In the absence of moisture transfer calculations, the

evaporative effect can only be approximated based on an assumption of the wet-bulb temperature

at the soil’s surface. The actual evapotranspiration rate is between zero and a maximum where the
 23

soil surface wet-bulb temperature is the same as the wet-bulb temperature of the air. When the

soil is saturated with moisture, it is safe to assume maximum evapotranspiration. However, for dry

soil the low surface wet-bulb temperatures can lead to unrealistic results.

Using the maximum potential evapotranspiration, Bahnfleth and Pedersen [11] demonstrated

that neglecting the effect can reduce the annual average heat loss by up to 170% (i.e., a heat gain)

in Phoenix, Arizona; 28.4% in Philadelphia, Pennsylvania; 18.7% in Minneapolis, Minnesota; and

32.3% in Medford, Oregon. Deru [39] also showed an 11.2% difference in Fort Collins, Colorado.

2.2.3.8 Snow Cover

Snow cover has several effects on the exterior ground surface boundary condition:

• Fresh snow has a solar reflectivity of 0.98.

• Snow acts as an insulating layer on the ground. Gilpin and Wong [47] refer to this as the

heat-valve effect.

Snow-related boundary conditions are discussed in more detail in references pertaining to

snow-melting system analysis [92, 82].

One significant challenge in simulating snow cover is that snow amounts are not typically

provided in standard weather data. While there are data fields available for snow depth, they are

not actually populated in weather data files.

2.2.3.9 Far-Field Boundary

The far-field boundary represents the distance from the building where there is no longer heat

transfer in the horizontal directions. This may be at a distance sufficiently far from the building

or, alternatively, along a plane of symmetry (e.g., half-way between two similar buildings).

Most tools simulate this boundary as either a zero-heat-flux condition or an independently-

calculated, one-dimensional temperature profile. If the boundary is sufficiently far away from the
 24

building, then both approaches should produce the same result. However, the zero-flux boundary

requires fewer calculations.

2.2.3.10 Deep-Ground Boundary

The deep-ground boundary is usually defined as either the distance to the water table or the

distance to a point in the ground deep enough that it is not affected by the annual temperature

fluctuations at the surface.

If the water table is shallower than the unaffected ground depth, then a constant temperature

condition is applied (usually the water table temperature is approximated as the annual average

outdoor dry-bulb temperature for the location). In a case where the water table is below the

unaffected ground depth, a zero-heat-flux condition is applied, although this is also sometimes

approximated by using a constant temperature condition equal to the annual average outdoor

dry-bulb temperature.

Bahnfleth and Pedersen [11] found that the type of boundary condition (zero-heat-flux versus

constant temperature) only accounts for a 1.3–3.5% change in average annual heat loss. Contrarily,

the depth of the boundary was found to have a fairly significant impact on results. Between depths

of 10 m and 15 m, the mean heat loss can change by up to 12%.

Several authors have explored the impact of water table depth on foundation heat transfer

[37, 11, 96, 39]; however, no tool accounts for the annual variation in water table depth observed

by Thomas and Rees [107].

2.2.4 The Whole-Building Context

Some foundation heat transfer tools are used in the context of a whole-building energy sim-

ulation engine. Current whole-building simulation tools require tens of seconds to several hours

to simulate a single building’s energy use, depending on the complexity of the building model and

the solution methodologies employed. As foundation heat transfer represents only a fraction of the

energy load on the building, the calculations should represent a relatively similar fraction of the
 25

overall computation time and not several orders of magnitude longer, which is the case for many

of the more detailed numerical solutions.

To minimize the computation time devoted to foundation heat transfer, many whole-building

simulation tools employ simplistic methods with limited capability based on preprocessed or re-

gression results generated by a more detailed foundation heat transfer tool.

Most whole-building simulation tools are not designed for multidimensional heat transfer cal-

culations. Often there is an artificial boundary created between the foundation heat transfer calcu-

lations and the one-dimensional foundation surface constructions represented in a whole-building

model.

The TRNSYS (TRaNsient SYtems Simulation) software program [104] has a very detailed

ground heat transfer library, which can be coupled to a whole-building energy simulation. The

heat balance between the thermal zone and the foundation requires a computationally expensive

iterative solution.

Decoupling is not without its advantages. Al-Anzi and Krarti [6] developed a method that

used decoupling as an opportunity to employ separate solution methods that have advantages at

different time scales. In their work titled “Local-Global Analysis,” the global domain of ground

heat transfer is calculated using a faster analytical model, and the local domain (near the interface

of the building) is calculated using a numerical method with high resolution to capture thermal

bridging effects.

Using whole-building energy simulation tools for foundation insulation design requires coupled

foundation heat transfer calculations that are both faster than the detailed three-dimensional tools

like TRNSYS and more capable than the simplistic regression-based tools.

2.3 Computational Algorithms

The real limitation in estimating heat loss from building foundations is not in our under-

standing of the physics but in the shear number of calculations required to solve the heat diffusion

equation for the spatial and temporal domains at a reasonably high resolution of detail.
 26

The following sections describe the calculations required to simulate the physics of foundation

heat transfer as it is understood in literature.

2.3.1 Heat Diffusion

An energy balance of heat flow through soil is commonly treated as a completely conductive

problem. In reality, there are several mechanisms that contribute to heat transfer [39]:

• Conduction through soil grains, water, ice, and entrained gasses (e.g., air)

• Latent heat transfer as water evaporates within the soil matrix

• Latent heat transfer as water freezes and thaws within the soil matrix

• Advection from bulk liquid flow (e.g., rainfall)

• Advection from bulk gas flow

• Diffusion of liquid and vapor through the soil matrix

• Radiation exchange in the soil voids

Of the mechanisms listed above, conduction is considered the dominant form of heat transfer

for most cases. Depending on the composition of the soil, other mechanisms may play greater roles

in the overall heat transfer.

A heat balance on an infinitesimally small volume yields a relationship between two indepen-

dent variables, T , temperature (the potential for heat flow), and Ψ, matric potential (the potential

for moisture flow):

∂T ∂Ψ
CH,T + CH,Ψ = ∇ · [k∇T ] + ∇ · [DH,Ψ ∇Ψ] + Cp,l ṁl · ∇T + q̇, (2.18)
∂t ∂t

where

• CH,T is the thermal heat capacitance [J/m3 -K],

 27

• T is the temperature at a given spatial position and point in time [K],

• t is time [s],

• CH,Ψ is the matric heat capacitance [J/m4 ]

• Ψ is the matric potential at a given spatial position and point in time [m],

• ∇ is the differential spatial gradient operator,

• k is the thermal conductivity [W/m-K],

• DH,Ψ is the matric heat diffusivity [W/m2 ],

• Cp,l is the specific heat capacity of liquid water [J/kg-K],

• ṁl is the mass flow rate vector [kg/m3 -s], and

• q̇ is the heat source or sink [W/m3 ].

The first two terms in Equation 2.18, CH,T ∂T ∂Ψ

∂t and CH,Ψ ∂t , represent the rate of thermal

energy stored as a result of changes in temperature and matric potential, respectively.

Here, CH,T is the thermal heat capacitance, which accounts for the volume-average heat

capacity of the soil and its constituents as well as the internal energy of water vapor associated

with changes in temperature:

CH,T = (1 − ηs )ρs Cp,s + θl ρl Cp,l + θv ρv Cp,v + ρl hf g Cv,T , (2.19)

where the subscripts s, l, and v refer to the soil, liquid water, and vapor water components,

respectively, and

• ηs is the porosity of the soil [0–1],

• ρ is the density [kg/m3 ],

• Cp is the specific heat capacity [J/kg-K],

 28

• θ is the volumetric fraction of a given component [m3 /m3 ],

• hf g is the latent heat of vaporization of water [J/kg], and

• Cv,T is the thermal vapor capacitance [K−1 ].

CH,Ψ is the matric heat capacitance, which accounts for the internal energy of water vapor

associated with changes in matric potential defined as follows:

CH,Ψ = ρl hf g Cv,Ψ , (2.20)

where Cv,Ψ is the matric vapor capacitance which, as Janssen [56] explains, has very little influence

on the overall solution and can be assumed to be negligible.

The third term in Equation 2.18, ∇ · (k∇T ), represents the Fourier heat conduction.

The effective thermal conductivity, k, represents the combined conductivity of all the soil

constituents (soil grains, water, air, ice, etc.) as well as the effect of vapor distillation [35].

The effective conductivity can be calculated based on the conductivities and volumetric frac-

tions of the constituents as well as the geometry of the soil grains (details of this calculation can

be found in De Vries [35] and Deru [39]):


θl kl + ξp θp kp + ni=1 ξi θi ki
k=  , (2.21)
θl + ξp θp + ni=1 ξi θi

where

• Subscript p indicates the gas-filled (i.e., air and water vapor) portion of the soil matrix

whose conductivity is calculated to include the effect of vapor distillation in Equation 2.22,

and

• ξ is the ratio of the average temperature gradient in the constituent and the average tem-

perature gradient of the medium defined in Equation 2.23.

∂ρvs
kp = ka + φhf g Da , (2.22)
∂T
where
 29

• φ is the relative humidity [0–1],

• Da is the vapor diffusion coefficient in air [m2 /s], and

• ρvs is the density of saturated water vapor [kg/m3 ].

3
ξi = . (2.23)
ki /kl + 2

The fourth term in Equation 2.18, ∇·(DH,Ψ ∇Ψ), represents heat transfer caused by moisture

diffusion. Here, DH,Ψ is the matric heat diffusivity, which is proportional to the matric vapor

diffusivity:

DH,Ψ = ρl hf g Dv,Ψ , (2.24)

where Dv,Ψ is the matric vapor diffusivity defined in Equation 2.29 in Section 2.3.2 on moisture

diffusion.

The fifth term in Equation 2.18, Cp,l ṁl · ∇T , represents the sensible heat transfer by bulk

fluid flow. The assumption here is that the bulk fluid and the surrounding media are in thermal

equilibrium.

2.3.2 Moisture Diffusion

A moisture balance on an infinitesimally small volume yields:

∂Ψ ∂T ∂K
CM,Ψ + CM,T = ∇ · [DM,Ψ ∇Ψ] + ∇ · [DM,T ∇T ] + , (2.25)
∂t ∂t ∂z

where

• CM,Ψ is the matric moisture capacitance [m−1 ],

• Ψ is the matric potential at a given spatial position and point in time [m],

• t is time [s],

• CM,T is the thermal moisture capacitance [K−1 ],

 30

• T is the temperature at a given spatial position and point in time [K],

• ∇ is the differential spatial gradient operator,

• DM,Ψ is the matric moisture diffusivity [m/s],

• DM,T is the thermal moisture diffusivity [(K-s)−1 ],

• K is the hydraulic conductivity [m/s], and

• z is the gravitational potential [m].

The first two terms in Equation 2.25, CM,Ψ ∂Ψ ∂T

∂t and CM,T ∂t , represent the rate of moisture

stored as a result of changes in temperature and matric potential, respectively.

CM,Ψ is the matric moisture capacitance, which is comprised vapor and liquid matric capac-

itances:

CM,Ψ = Cv,Ψ + Cl,Ψ , (2.26)

where
∂θl
Cl,Ψ = , (2.27)
∂Ψ T

and, as previously noted, Cv,Ψ can be assumed to be negligible.

CM,T is the matric moisture capacitance, which also has a neglibible effect on the solution

[56].

Calculating the derivatives in Equation 2.27 requires moisture retention curves for specific

soil types, which can be difficult to obtain or generalize for a range of soil compositions and grain

geometries.

The third and fourth terms in Equation 2.25, ∇ · (DM,Ψ ∇Ψ) and ∇ · (Dv,T ∇T ), represent

the diffusion of moisture driven by the matric and temperature gradients, respectively.

DM,Ψ = Ks + Dv,Ψ , (2.28)

 31

where Ks is the hydraulic conductivity of the soil [m/s] and

⎧
⎪
⎪
⎪
⎪ ρvs φg
⎪
⎪
⎪
⎪ η s D a for θl ≤ θl,c
⎨ ⎡ ρl Rv T ⎤
Dv,Ψ = (2.29)
⎪
⎪ ⎢ θa θl ⎥ ρvs φg
⎪
⎪ ⎢ θa + ⎥ Da
⎪ ⎣ for θl < θl,c ,
⎪
⎪
⎪ ηs − θl,c ⎦ ρl Rv T
⎩

where

• ηs is the porosity of the soil [0–1],

• Da is the vapor diffusion coefficient in air [m2 /s],

• ρvs is the density of saturated water vapor [kg/m3 ],

• φ is the relative humidity [0–1],

• g is the gravitational constant of 9.81 [m/s2 ],

• ρl is the density of liquid water [kg/m3 ],

• Rv is the gas constant of water vapor [J/kg-K],

• T is the temperature at a given spatial position and point in time [K],

• θ is the volumetric fraction of a given component (air a, or liquid water l) [m3 /m3 ], and

• θl,c represents the liquid moisture content, below which the liquid water is no longer con-

sidered continuous throughout the porous structure and the relative humidity drops below

100%. This usually occurs between values of 0.1 and 0.2 m3 /m3 .

The fifth term in Equation 2.25, DM,T , is the thermal moisture capacitance, which Janssen

[56] explains is negligible.

∂K
The final term in Equation 2.25, ∂z , represents the additional moisture flow potential due

to gravity.
 32

2.3.3 Common Simplifications

The equations established by Deru [39] with the assumptions explained by Janssen [56], lead

to the following system of two non-linear partial differential equations:

∂T ∂Ψ
CH,T + CH,Ψ = ∇ · [k∇T ] + ∇ · [DH,Ψ ∇Ψ] + Cp,l ṁl · ∇T + q̇ (2.30)
∂t ∂t

and
∂Ψ ∂K
CM,Ψ = ∇ · [DM,Ψ ∇Ψ] + . (2.31)
∂t ∂z

Equations 2.30 and 2.31 must be solved simultaneously using an iterative, non-linear solution

method such as the Newton-Raphson method in order to satisfy the conservation of both mass and

energy. Several of the terms in these equations are highly non-linear and can increase the number

of calculations significantly.

Solving the combined heat and moisture diffusion equations requires significantly more in-

formation about the soil properties than is typically available to building energy simulators. A

number of methods have been developed to calculate the thermal and matric properties based on

a soil’s composition [36, 88, 112], but it is not common practice to collect and characterize soil

samples before conducting an energy simulation.

Most foundation heat transfer tools will simplify the calculations by assuming that the ther-

mal properties are linear (i.e., they do not depend on the temperature of the soil) and without

calculating coupled moisture diffusion. These simplifications reduce the calculations to a single

equation:
∂T
CH,T = ∇ · (k∇T ) + q̇. (2.32)
∂t

The apparent thermal conductivity, k, in Equation 2.32 approximates all mechanisms of heat

diffusion in the porous soil medium and not solely the conduction. The value of k may vary spatially

and temporally while still solving the discretized equation using linear algebra methods; however,

the value of k cannot depend on the calculated temperatures in the domain without invoking

computationally-expensive, non-linear solution techniques.

 33

With Equation 2.32, the estimated foundation heat transfer is most sensitive to the value of

soil thermal conductivity [1]. Sterling et al. [105] conducted a survey of the apparent conductivity

of soils found in the United States, which revealed that the value can range between 0.2 W/m-K and

4.0 W/m-K depending on moisture content and soil composition. Rock [99] showed that between

soil conductivities of 0.9 W/m-K and 2.0 W/m-K there can be a 100% difference in the predicted

heat loss rate from a slab.

2.3.4 Solution Methods

When it comes to methods for solving the diffusion equations in the context of foundation

heat transfer, there is no “silver bullet.” Many such methods have been presented throughout

the literature, and each method brings unique advantages and disadvantages. For a more in-

depth review of numerical techniques used in solving foundation heat transfer problems, consult

MacArthur et al. [85]. The following sections will provide a brief description and explain the

advantages and disadvantages of several solution methods found in the literature.

2.3.4.1 Analytical Solutions

Lachenbruch [78] demonstrated one of the first analytical solutions for slab-on-grade heat

transfer using Green’s function. Delsante et al. [38] later developed a similar model using Fourier

transforms. These two models, compared by Kusuda and Bean [76], were shown to produce nearly

identical results. The major limitation of both Lachenbruch’s and Delsante’s solutions is that they

are only appropriate for rectangular, uninsulated slabs.

Many analytical solutions, including those using Fourier transforms, take the form of an

infinite series. The calculated solution is then an approximation where the series is calculated up

to a large number, N , after which the series is truncated. Larger values of N will improve the

numerical accuracy of the solution but will require more computation time.

Analytical methods are typically faster than the numerical alternatives where finer discretiza-

tions require proportionally more computation time. The major drawback of analytical methods
 34

is solutions only exist for a small subset of problems with highly constrained—and not always

realistic—boundaries and geometries.

Hagentoft and Claesson [48, 26, 51, 49, 50, 52] have developed a number of solutions using

the superposition of analytical solutions and conformal mapping.

Krarti [62] is another prominent contributing author of analytical solutions. His work is

described in more detail in Section 2.5.2.

2.3.4.2 Finite Difference Method

The finite difference method is one of the more prominent solution methods encountered in

foundation heat transfer tools. Variants of the method can be found in work by Bazjanac et al.

[13], Bahnfleth [9], Cogil [30], and Rock [99].

The finite difference method often uses spatial discretization aligned with the axes of the

coordinate system. This imposes a limitation on the range of geometries that can be simulated.

However, this same constraint enables computationally efficient storage and execution of the linear

algebra routines.

Furthermore, building foundations are often constructed using orthogonal surfaces and rarely

require geometries that are not aligned with a Cartesian coordinate system. The restriction that

the finite difference method imposes on the shape of the domain only poses a problem for a small

minority of design cases.

Finite difference methods are typically categorized as either explicit or implicit schemes de-

pending on how the solution is calculated relative to the solution of the previous time-step. Explicit

schemes do not require the use of linear algebra algorithms and, therefore, require fewer calcula-

tions. Conversely, implicit schemes yield more reliably stable and realistic results. For the standard

explicit scheme formulated in Cartesian coordinates with a uniform spatial discretization and ho-

mogeneous thermal properties, the stability criterion is defined as follows:

αΔt αΔt αΔt 1

2
+ 2
+ 2
≤ , (2.33)
Δx Δy Δz 2
 35

where

• α is the thermal diffusivity [m2 /s],

• Δt is the time-step [s], and

• Δx, Δy, and Δz are the spatial cell size [m] in the x, y, and z directions, respectively.

Several other variants of the finite difference method are discussed in greater detail in Section

3.4.3.

2.3.4.3 Finite Element Method

Finite element methods are implicit in nature and require the use of linear algebra algorithms

to solve the set of simultaneous equations resulting from the domain discretization. Computation

times are similar to those of the implicit finite difference methods.

Unlike finite difference methods, the spatial discretization does not have to align with the

coordinate system axes. The unconstrained discretization enables the simulation of a wider range

of domain geometries. However, off-axes geometry makes the discretized mesh difficult to generate.

The finite element method is used in the foundation heat transfer tools developed by Mitalas

[87], Baylon and Kennedy [12], Beausoleil-Morrison et al. [17], Deru [39], and Janssen [56].

2.3.4.4 Response Factors

Response factors are a set of values that relate the heat transfer between two surface bound-

aries to a series of temperature histories. DOE-2 uses response factors to calculate heat conduction

between interior and exterior surfaces of above-grade walls, roofs, ceilings, and floors. EnergyPlus

uses a modified response factor approach called conduction transfer functions or CTFs [111]. A

number of other response-factor-type calculations are employed in foundation heat transfer tools

developed by Davies et al. [34], Krarti et al. [73], Zhang and Ding [121], and Wentzel [116]. Bahn-

fleth and Amber [10] developed Ground Conduction Functions (GTFs) and demonstrated that the

GTFs produce results with 1% of the results from a detailed finite difference model.
 36

Response factors must be generated before the simulation, either through analytical calcu-

lations or a numerical preprocessor. Response factors are determined by recording the system’s

response to temperature pulses at the boundaries. Once established, the calculation of heat transfer

based on thermal history can be very computationally effective. The trade off between computation

time and numerical accuracy is determined by the additional computation time required by the

preprocessor relative to the computation time saved in the simulation.

One concern of accuracy when using response factors is the amount of time required to estab-

lish a reasonably consistent thermal history before the results can be valid for use in a simulation.

The ground domain has a very slow response to temperature variations, with some effects verging

on the magnitude of several years. Even if the response factors were to capture an adequately long

thermal history, the conduction calculated in the simulation would still have to be initialized with

an adequate history before the simulation commenced.

Response factors are also generated for a specific set of thermal properties that are assumed

to be static for the duration of the simulation. This approach works for most instances of the

diffusion equation but does not work for any non-linear or time-varying thermal properties.

Another approach related to response factors is establishing one-dimensional calculations that

have an equivalent thermal response. Xie et al. [119] describe a method for creating two (or more)

layered equivalent slabs that represent the foundation heat transfer for the more important diurnal

and annual harmonics.

2.3.4.5 Regression Methods

Regression methods have been established by fitting correlations to calculated or experimental

data. Mitalas [87] developed a regression method that provides steady-state and periodic heat loss

factors to describe below-grade heat loss. Although originally intended for use in deep basements,

the method was later adapted to describe shallow basements and slabs-on-grade. The below-grade

surfaces are separated into subsurfaces, usually an upper and lower wall and a perimeter and core

floor. The heat flux through each section is calculated by steady-state and dynamic heat loss factors
 37

derived from a two-dimensional finite element program and tabulated for nearly 100 combinations

of insulation configurations, foundation types, and soil properties.

Mitalas’s finite element tool was later used to develop the more general BASESIMP regression

by Beausoleil-Morrison and Mitalas [15]. Krarti [64] created a separate regression based on neural

networks, and later compared the methodology to the BASESIMP regressions with consistent

results [14]. Additional regression methods are described by Krarti and Choi [68], Bazjanac et al.

[13], and Winkelmann [118].

Most regression methods require little computation time, but the modeling capability of

most regression methods is limited by the robustness of the data set that originally informed

the regression. For example, Mitalas’s regression factors were developed for use in Canada, with

consistent snow cover on the ground, limiting the applicability of the regressions to snowy locations.

2.3.5 Discretization

Of the solution methods described above, the analytical methods are too constrained to be

broadly applicable to foundation insulation design. The other four methods are either numerical

in nature or are dependent on a numerical method as a preprocessor. A numerical method will

inevitably play a role in developing more capable and computationally efficient foundation heat

transfer simulation tools.

All numerical approximations of the diffusion equation require that both spatial and tem-

poral domains are divided into discrete cells or time-steps, respectively. The total computational

requirement for a simulation is directly proportional to the number of cells in the spatial domain

multiplied by the number of time-steps in the temporal domain. This section describes how the

numerical discretization is defined and some methods developed to reduce the number of associated

calculations.
 38

2.3.5.1 Spatial Domain

The spatial domain can be idealized as a semi-infinite solid extending laterally outward and

extending downward from the exterior grade. In a numerical model, the domain is bounded by

the lateral far-field boundaries and the deep-ground boundary. These boundaries must extend

sufficiently far from the building to capture the influence on the soil.

As part of the IEA BESTEST: In-Depth Diagnostic Cases for Ground Coupled Heat Transfer

Related to Slab-On-Grade Construction, Thornton [109] demonstrated that these boundaries had

less than a 0.01% impact when the deep-ground boundary was more than 30 meters below the

building and far-field boundaries were more than 20 meters from the building. A smaller domain

requires fewer discretized cells but may not reflect a semi-infinite nature.

The cells created by the discretization are numerical approximations of the infinitesimally

small elements for which the partial differential equations were derived. As cell size decreases

the approximation approaches the continuous solution. Many authors demonstrate that below a

certain cell size, the results achieve “discretization independence” and are no longer sensitive to

the numerical approximation. This is the lower bound of cell size.

The upper bound on cell size is often determined by the level of detail required in the model.

For the case of foundations, the upper bound is determined by the size of the thinnest component

of the foundation construction represented in the thermal model. For example, a board of rigid

insulation may be as thin as an inch, though cells elsewhere in the domain can be larger.

Often it is necessary to have fine cell spacing near the foundation perimeter, but the spacing

can gradually increase in the direction away from the perimeter. Usually, the cell spacing increases

according to a geometric series:

Δxi = fg Δxi−1 = fg2 Δxi−2 = . . . = fgN Δxi−N , (2.34)

where fg is a constant growth coefficient, typically somewhere between 1.05 and 1.50. Geometric

cell growth drastically reduces the number of cells in the domain.

 39

The total number of cells in the domain is the product of the number of cells in each of the

coordinate axis directions. Utilizing planes of symmetry eliminates half of the cells in the direction

perpendicular to the plane; however, this only works if the entire problem is symmetric, including

the geometry, thermal properties, and boundary conditions. This means convection algorithms that

are dependent on wind direction and solar incidence on non-horizontal surfaces must be generalized

to account for both halves of symmetry.

Another effective way to reduce the number of cells is to eliminate an entire computational

dimension. Foundation heat transfer is dominantly two-dimensional with some three-dimensional

effects near the corners of the foundation. Many tools either neglect the third dimension (along one

of the horizontal axes) or approximate the three-dimensional effects in a two-dimensional coordinate

system.

Several attempts to incorporate the three-dimensional effects into a two-dimensional model

are discussed in Chapter 5.

2.3.5.2 Temporal Domain

Most whole-building energy simulation tools simulate a single year of operation, largely be-

cause they are driven by a single year of weather data. Additionally, most of the energy-related

features of a building do not have effects at longer time scales. This is not the case for features

that interact with the ground. After a building is first heated, the ground can take several years

to come into a steady periodic equilibrium with the building. IEA BESTEST [90] showed that

the initialization period for slab-on-grade simulations can take anywhere from three to ten years to

reach a steady-periodic state, within 0.1% annual heat loss of the previously simulated year. Baz-

janac et al. [13] used an initialization period of four years for the simulations that informed their

regression method. Other mechanisms which transfer heat to the ground, including ground-source

heat pumps [44] and seasonal heat storage [103], also demonstrate similarly long time responses.

For single year analyses, it is best to use simulation results for steady periodic ground conditions

since this represents the state of the building for most of its existence.
 40

The number of years required to arrive at steady periodic ground conditions in a simulation

is dependent on how well the initial temperatures reflect the state of the ground at the beginning of

the simulation period. Poor estimates of initial ground temperatures can lead to additional years

of “warm-up” simulation prior to the actual simulation period.

The literature describes several approaches to estimating initial ground temperatures:

Constant temperature: Setting all the cells to a single spatially constant temperature is the

simplest approach to initializing ground temperatures. Many authors have used the annual

mean dry-bulb temperature as specified by Neymark and Judkoff [90] for the IEA BESTEST

work.

Kusuda: Kusuda and Achenbach [75] developed a one-dimensional analytical solution for undis-

turbed soil. This solution is based on heat diffusion into a semi-infinite domain driven by

a harmonic temperature variation at the surface:

    
ΔTm π 2π z ty
T (t, z) = Ty,ave − exp −z cos t − Δtshif t − , (2.35)
2 ty α ty 2 πα

where

• T is the temperature at depth z and time t [K],

• Ty,ave is the annual average outdoor dry-bulb temperature at the location [K],

• ΔTm is the range of monthly average dry-bulb temperatures throughout a year [K],

• ty is the time of an annual cycle, 3.1536×107 [s],

• α is the thermal diffusivity of the soil [m2 /s], and

• Δtshif t is the time difference between the calendar cycle and the temperature cycle

[s]. This is the time from the beginning of the calendar year to the time of the annual

minimum outdoor dry-bulb temperature.

This approach approximates the entire ground domain as undisturbed by the presence of

a heated building. Though it does not account for lateral variation in soil temperatures,
 41

the Kusuda approach is considered a better estimate than a spatially constant temperature

field.

Steady-state solution: A steady-state solution provides a rough approximation of spatial varia-

tion in all dimensions, but it does not account for the long-term thermal response provided

by the Kusuda estimation.

Each of these approaches, along with a new initialization method, are compared in Chapter

4 to illustrate the effect each has on reducing the duration of the initialization period.

The discretized time-steps in the temporal domain must be short enough to both capture the

variation in boundary conditions (typically, hourly weather data) and ensure numerical stability in

the solution as exemplified in Equation 2.33.

2.4 Validation, Verification, and Testing

Judkoff and Neymark [60] characterize three primary methods of evaluating the accuracy of

building energy simulation tools. These methods are described as follows:

Empirical Validation: Calculated results from a tool or method are compared to monitored data

from a real building, test cell, or laboratory experiment.

Analytical Verification: Outputs from a tool or method are compared to results from a known

analytical solution or generally accepted numerical method for isolated heat transfer under

very simple, highly-constrained boundary conditions.

Comparative Testing: A program is compared to itself or to other programs.

Each of these methods is discussed in further detail, along with relevant efforts in evaluating

the accuracy of foundation heat transfer calculations.

 42

2.4.1 Empirical Validation

Empirical data provides quantifiable results that are accurate within the uncertainty of the

measurements, but such measurements can be expensive and time consuming. This is particularly

true in the case of foundation heat transfer where measurements may require multiple seasons or

years to produce an adequate data set. A well-designed experiment requires that all unknown

quantities be carefully measured, which in the case of foundation heat transfer would require the

measurement of the thermal properties of the soil at every point in the spatial and temporal domain

of the experiment.

Another disadvantage of empirical validation is that the results are only applicable to the

specific experimental setup at the time of measurement. The range of foundation designs, ground

domain properties, and boundary conditions that affect heat transfer are difficult to capture through

experiment, meaning that the results of the experiments would only serve to validate a small subset

of the parameter space.

There have been relatively few efforts in the past to validate the accuracy of foundation heat

transfer tools. Experimental validation has traditionally been cost and time prohibitive and has

produced only a limited set of measurements. Two of the more significant efforts of experimental

validation are authored by Rees et al. [94, 93, 107, 97, 98, 108] and Adjali et al. [1, 3, 4]. They

found that in many cases the models produce reasonably accurate results and that most of the

errors were related to uncertainty in the model input (namely the thermal properties of the soil)

and the inability to model the impacts of rain and snow. Additional empirical validation efforts

can be found in Trethowen and Delsante [110] and Rantala and Leivo [91]. A more comprehensive

review of empirical case studies is available in Adjali et al. [5].

In many cases, attempts to validate foundation heat transfer tools becomes an exercise in

calibrating a model rather than a true validation of the tool due to the high degree of uncertainty

in the model inputs.

 43

2.4.2 Analytical Verification

Analytical solutions can be used to verify the implementation of the algorithms in computer

code. Since analytical solutions are well defined and highly constrained, analytical verification does

not introduce the same problem of input uncertainty inherent to experimental validation. However,

these constraints also limit the application of the verified results to a similarly small subset of the

parameter space.

Numerical solutions demonstrating discretization independence are used as a surrogate in

analytical verification for cases where an analytical solution cannot be derived. These solutions

must first be verified against the analytical solutions, where they exist, to anchor the veracity of

their implementation.

For analytical verification, once the solution has been established, it is relatively quick and

inexpensive to verify other tools against it. However, verification is limited to the ground domain

subsystem of the building where a problem is reasonably well-defined and cannot encompass the

added complexity at the whole-building level without straying from the well defined analytical

anchor.

The most comprehensive example of analytical verification is the International Energy Agency

Building Energy Simulation Test and Diagnostic Method (IEA BESTEST ) for ground coupled

heat transfer related to slab-on-grade construction [90]. The IEA BESTEST procedures include

verification against the analytical solution developed by Delsante et al. [38] and several verified

numerical solutions.

The IEA BESTEST for ground coupled heat transfer related to slab-on-grade construction

establishes test cases for rectangular, uninsulated slabs with constant thermal properties of soil and

constant convective boundary conditions. IEA BESTEST has yet to produce additional test cases

to explore other foundation shapes, types, insulation designs, and boundary conditions.
 44

2.4.3 Comparative Testing

Comparisons of simulation results among different tools or methods reveal inaccuracies where

there is poor agreement. However, agreement under comparative testing does not guarantee ac-

curate results since there is no absolute truth standard like there is in empirical validation and

analytical verification.

Like analytical verification, comparative tests can be executed inexpensively to quickly iden-

tify discrepancies. The tests can also be designed to focus on any portion of the system as well as

establish high-level comparisons at the whole-building level.

Previous whole-building comparative tests described in the Home Energy Rating System

(HERS) BESTEST include four cases highlighting heat loss through foundations [59]. These cases

include insulated and uninsulated slabs, and insulated and uninsulated basements, illustrated by

Deru et al. [40]. Several comparative test cases in the IEA BESTEST specification [90] revealed

disagreements in predicted heat loss between 1%–24% across a number of simulation tools.

The BESTEST procedures have laid the foundation for improving the accuracy of foundation

heat transfer tools but are limited in the range of capability they can test. Additional validation

is still needed for a wide range of foundation designs, ground domain properties, and boundary

conditions.

2.5 Comparison of Prominent Tools and Methods

This section takes a closer look at some of the more prominent foundation heat transfer tools

and methods either frequently used by energy modeling practitioners or cited often in the literature.

The methods and tools discussed in this section are qualitatively compared side-by-side along with

the tool developed as part of this thesis work in Table 2.3. Kiva is described in detail in Chapter

3. Claridge [27], Adjali et al. [2], Krarti [66] and Deru [39] also provide thorough reviews of the

various foundation heat transfer methods and tools found throughout literature.
 Table 2.3: Comparison of prominent foundation heat transfer tools and methods

Method/Tool ASHRAE ITPE EnergyPlus Preprocessors eQUEST GHMT TRNSYS Kiva

Huang &
Author Latta Wang Krarti Bahnfleth Cogil Deru Thornton Kruis
Winkelmann
Foundation Type:
Slab-on-Grade -    -    
Basement  -  -     
Crawlspace  -  -     
Insulation Design:
Slab, Crawl
Interior Horizontal  -      
Only
Interior Vertical     -    
Exterior Horizontal - -  -  Slab Only   
Exterior Vertical - - - - -    
Whole Slab  -  - - -   
Boundary Conditions:
Interior Air Temperature Constant Constant Constant Sinusoidal Simulated Constant Constant Simulated Constant
Interior Surface Convection - - - Constant Constant Constant Constant Constant Correlation
Interior Surface Radiation - - - -  - - - -
Weather Weather Weather Weather Weather
Exterior Air Temperature Constant Constant Sinusoidal Sinusoidal
File File File File File
Exterior Surface Convection - - - Constant Correlation - Correlation Correlation Correlation
Exterior Surface
- - -   -   
Long-Wave Radiation
Exterior Surface
- - -   -   
Short-Wave Radiation
Other Capabilities:
Multiple Interior Thermal Zones - - - - - - -  -
Thermal Bridging -  - - -    
Moisture Balance - - - - - -  - -
Preprocess Preprocess Preprocess Full Preprocess
Simulation Integration - - - -
Calculation Calculation Library Integration Calculation
Number of Dimensions 1 1 2-3 3 3 2 2 3 2-3
Scaling Parameter P P A, P A, P A, P P P - A, P
Characteristic length(s) - - A/P A/P A/P A/P width/2 - Multiple
Finite Finite Finite Finite

45
Finite
Solution Method Analytical Regression Analytical Difference Difference Regression Difference Difference
Element
(Explicit) (ADI) (Implicit) (Multiple)
Mesh Generation - - - Automatic Automatic - Manual Manual Automatic
 46

2.5.1 ASHRAE Design Calculations

Some of the oldest foundation heat transfer calculation methods can be found in the ASHRAE

Handbook of Fundamentals [7]. The handbook describes separate methods corresponding to the

type of foundation. Heat flow from insulated basement walls is calculated based on a method

developed by Latta and Boileau [79] in 1969 where steady-state conduction is approximated through

a series of one-dimensional concentric radial paths (Figure 2.5).

The basement wall and floor are separated into discrete sections, each with its own radial

path with only angular variation in thermal properties. This allows for relatively simple calculation

using polar coordinates where each section of the wall or floor can have different levels of insulation.

This method can also be applied to crawlspace walls and floors.

Heat transfer through slabs is defined by a perimeter conductance factor found in look-up

tables. The factors are provided for only four slab configurations based on regressions from finite

element simulations [115].

2.5.2 Interzone Temperature Profile Estimation (ITPE) Technique

The ITPE technique developed by Krarti et al. [71, 70, 72] is the most complete analytical

solution for foundation heat transfer. The basic premise of this method is to break up the domain

into zones so that each zone is a rectangular region with unique boundary conditions. The heat

diffusion equation is solved for each zone using an estimated temperature profile as the boundary

condition between zones, which satisfies a continuity of heat flux across the boundary. This method

has been adapted to work in either two- or three-dimensions. Applications of the ITPE technique

can be found in several references [63, 65, 22, 23, 25].

2.5.3 EnergyPlus Preprocessors

Bahnfleth [9] and Bahnfleth and Pedersen [11] established a three-dimensional finite difference

solution for slab-on-grade floors using a standard explicit solution scheme. This solution was
 47

originally used to examine sensitivity to shading caused by the building, evapotranspiration, and

different deep-ground boundary condition assumptions.

Later, Cogil [30] developed a similar solution for basements using an alternating direction

implicit (ADI) finite difference scheme. The solution included a detailed radiation exchange model

and heat balance for the basement space (including heat losses through the sill box) that was used

to calculate the interior surface boundary condition.

Bahnfleth’s and Cogil’s tools were adapted by Clements [29] to provide preprocessed results

for use in the EnergyPlus simulation engine [111]. In the process of adapting the code for use in

EnergyPlus, the shading algorithm in the slab program and the radiative exchange model in the

basement program were removed to reduce computation time. The preprocessors output monthly-

average surface temperatures for ground-coupled surfaces. These temperatures serve as boundary

conditions for the Conduction Transfer Functions (CTFs) used to calculate one-dimensional un-

steady heat flow through buildings surfaces.

2.5.4 eQUEST

eQUEST (the QUick Energy Simulation Tool) is a popular graphical user interface to the

DOE-2 simulation engine [55]. Huang et al. [54] integrated a fully implicit, two-dimensional finite

difference solution into DOE-2 to establish foundation insulation guidelines for the Building Foun-

dation Design Handbook [77]. A large set of simulation data, 88 foundation configurations in 13

cities, was generated and used to establish the regression method employed by Winkelmann [118],

where perimeter conduction factors characterize the effective steady-state resistance of an under-

ground surface. The dynamic effects of heat transfer through surfaces in DOE-2 are calculated

using a form of response factors. As implemented in DOE-2, the response factors are limited to a

thermal history of 100 hours and are unable to capture the longer time-frame responses typically

seen in foundation heat transfer [13]. To best represent the thermal response under this constraint,

the actual wall is modeled as a layered construction with one foot of soil.
 48

Krarti and Choi [69] found that this approach tends to underestimate the mass effect of the

ground and cannot properly represent indoor temperature changes within the course of a day.

2.5.5 Ground Heat and Moisture Transfer (GHMT)

Deru [39] developed one of the most complete solutions with coupled heat and moisture trans-

fer. His solution employs a two-dimensional finite element method with moisture and temperature

dependent thermal properties. The tool can also simulate the effects of evapotranspiration and

ground freezing.

2.5.6 TRNSYS

TESS (Thermal Energy Systems Specialists) created a ground coupling library for use in the

TRNSYS simulation engine [104]. The library employs a three-dimensional finite difference model

that can interface with any number of zones in a building. The TRNSYS ground-coupling library is

currently one of the only tools where the foundation heat transfer calculations are directly coupled

to the zone heat balance model through iterative simulation. TRNSYS is the only whole-building

simulation tool with a ground heat transfer solution that was used as a secondary truth standard

in IEA BESTEST for ground-coupled heat transfer [90]. TESS also compared TRNSYS to several

of the other simplified methods described above [86].

 49







Figure 2.4: Illustration of ITPE zones (1, 2, and 3) for a basement foundation (adapted from Krarti
et al. [70])

Figure 2.5: Concentric heat flow paths (adapted from Latta and Boileau [79])
 Chapter 3

Tool Development

3.1 Kiva

Kiva is a free and open-source C++ program developed by the author for the purposes of

this thesis and the source code is currently hosted at GitHub.com [74]. The name comes from the

subterranean ceremonial structures used by the ancient Puebloan peoples of the present-day states

of Colorado, Utah, Arizona, and New Mexico. Kivas are among the earliest constructed structures

in North America and are still actively preserved today.

The need for Kiva was established by observing the difficulty of the current tools to achieve

an acceptable balance of capability, complexity, computational performance, and algorithmic con-

fidence. Such a balance is a subjective assessment and will vary depending on the application of

the tool. However, in the context of whole-building energy simulation, there are some precedents

that must be set before such a tool is widely adopted:

Capability: The tool should provide methods of describing a large range of potential foundation

insulation designs. Such a range cannot be objectively defined, but some of the more

common design considerations should be accommodated by the tools capability. Examples

include the following:

• Basement, crawlspace, and slab foundation types.

• A range of foundation shapes beyond a rectangle.

• Continuously variable foundation depth (for different basement and crawlspace heights).
 51

• Continuously variable exterior/interior and horizontal/vertical insulation placement.

• User-specified thermal properties (for insulation, structural elements, and surrounding

soil).

Complexity: The level of complexity required to describe a foundation system should be within

the expected domain knowledge of the user. In this case, users are typically mechanical

engineers or architects with domain knowledge pertaining to buildings and heat transfer.

Users should be able to confidently define a foundation system without being confronted

by the numerical settings that are only tangentially related to estimating building energy

use (e.g., spatial discretization, or meshing, and convergence tolerance for a linear solver).

While it is important for these settings to be available to a user, the settings should not be

required inputs that distract from modeling information relevant to the foundation design.

Computational Performance: Current whole-building simulation tools require from tens of sec-

onds to several hours to simulate a single building’s energy use, depending on the complexity

of the building model and the solution methodologies employed. As foundation heat trans-

fer represents only a fraction of the energy load on the building, the calculations should

represent a relatively similar fraction of the overall computation time and not several orders

of magnitude longer, which is the case for many of the more detailed numerical solutions.

Computation performance in this paper is measured in wall time, or “real world” time,

as it is perceived by the user. This is used to differentiate from CPU time, which is the

collective calculation time of each processing core.

Algorithmic Confidence: The calculations performed should reflect knowledge of the physical

phenomena of the heat transfer process. Several methods and tools attempt to simplify the

calculations without justification. There are currently few highly-detailed tools available

to provide an accurate benchmark for such simplifications. Common algorithmic simplifi-

cations for ground-coupled heat transfer include the following:

 52

• One- or two-dimensional approximations of three-dimensional heat transfer.

• Convective and radiative boundary conditions approximated with constant coefficients

(combining convection and radiation) or constant temperatures.

• Coarse representations of the foundation components that cannot characterize influ-

ential thermal bridges.

• Regression models that do not expose important variables.

Kiva’s framework is a collection of interchangeable calculation routines specific to foundation

heat transfer. The calculation of foundation heat transfer is comprised of several components

including:

• the spatial and temporal discretization of the governing partial differential equation,

• the numerical scheme used to progress the simulation from one time-step to the next,

• the initialization of temperatures in the ground surrounding the building, and

• the coordinate system and number of dimensions describing the problem.

Each component listed can be addressed using a variety of different approaches, with each approach

having different impacts on the computational performance and accuracy of the solution. The Kiva

framework allows for testing of different combinations of approaches while maintaining consistency

and reusing common computational methods and data structures.

The purpose of Kiva is to provide a framework with a highly capable and detailed reference

solution against which the approximations and simplifications of subsequently developed solutions

can be verified. At the time of this work, Kiva is a standalone computer program which simulates

heat transfer through building foundations into the surrounding ground. Though it is not cur-

rently integrated into any whole-building simulation engines, the framework has been generalized

to minimize such an effort.

 53

3.2 Development Philosophy

For the most part, tools used for building energy simulation have been developed in a closed

environment. This approach has led to several overlapping efforts to provide many of the same

capabilities (e.g., DOE-2, EnergyPlus, and TRNSYS). There are occasions where this redundancy

is beneficial, such as comparative testing. However, it also means that researchers must often

reinvent the wheel. This is especially true for foundation heat transfer tools.

Several tools have been developed for graduate thesis research, including but not limited to

Bahnfleth [9], Davies [33], Deru [39], Janssen [56], Krarti [62], Shen [101]. The code from several

of these theses has since been incorporated in some fashion into currently used simulation engines:

EnergyPlus (Bahnfleth [9]), DOE-2 (Shen [101]), SUNREL (Deru [39]), and APACHE (Davies [33]).

Without available source code, each of these authors spent a significant amount of time

developing a software tool to perform their analyses. If the code were to be made available in an

open-source environment, future researchers could use more of their time improving and enhancing

the code that already exists in collaboration with other researchers rather than investing time

creating yet another foundation heat transfer tool.

Many of the existing foundation heat transfer tools are written in Fortran, which is well

established for its computational performance as a compiled language. However, the popularity,

usage, and support of the Fortran language has declined considerably over the years as more

modern languages, such as C++, have introduced improvements such as object-oriented design

with a relatively small difference in computational performance. Using a more supported and

modern language makes tool development accessible to a broader audience.

3.3 Modeling and Simulation Capabilities

The following sections describe the modeling and simulation capabilities of Kiva. As an open-

source framework, anyone is free to adapt Kiva to include additional features and capabilities.
 54

3.3.1 Foundation Design

The description of the foundation design is provided within the two-dimensional context

illustrated in Figure 3.1. This profile is applied along the entire perimeter of the foundation.




  



  

 

 


Figure 3.1: Two-dimensional context for defining foundation design in Kiva

Vertical dimensions are defined relative to the exterior grade at a zgrade = 0:

zwall-top = zgrade + Δzwall-above-grade (3.1)

zf loor = zwall-top − Δzf oundation (3.2)

zdeep-ground = zgrade − Δzdeep-ground (3.3)

where Δzwall-above-grade and Δzf oundation are foundation design user-inputs and Δzdeep-ground is a

simulation input with a default value of 40 meters.

The foundation type is defined by the combination of Δzwall-above-grade and Δzf oundation

illustrated in Table 3.1.

The foundation insulation and structural design are described by six elements illustrated

in Figure 3.2. The components of the insulation and structural design elements each reference a
 55

set of thermal properties: thermal conductivity, specific heat, and density. The locations of the

rectangular insulation components are specified relative to the foundation wall.

Foundation Wall: The composition of the foundation wall is defined by a series of layers extending

away from the interior wall surface. Although the footer is not explicitly described, the

wall height user-input, Δzwall , can be extended to represent the footer depth.

Slab: The slab is comprised of a series of layers extending below zf loor . Whole-slab insulation is

modeled by assigning one of these layers to an insulating material. If no slab is specified

by the user, the foundation floor will be directly exposed soil.

Interior Vertical Insulation: Interior vertical insulation begins at zwall-top and extends down-

ward and inward to a user-specified depth and thickness. The depth can be specified to

model partial interior wall insulation.

Interior Horizontal Insulation: Interior horizontal insulation begins at the wall’s interior sur-

face and extends inward and downward to a user-specified width and thickness at a user-

specified depth at or below zf loor .

Exterior Vertical Insulation: Exterior vertical insulation begins at zwall-top and extends down-

ward and outward to a depth (below grade) and thickness specified by the user.

Exterior Horizontal Insulation: Exterior horizontal insulation begins at the wall’s exterior sur-

face and extends outward and downward to a user-specified width and thickness at a user-

specified depth at or below zgrade .

The slab and wall elements also describe the long-wave emissivity and short-wave absorptivity of

the surface.

The level of specification available for each of the insulation and structural elements allows

the user to describe the foundation system in great enough detail to simulate some of the important

effects of thermal bridging.

 56

 
 

  
  

 
 

 



 

 

 

 

 
 
 

Figure 3.2: Insulation and structural design elements in Kiva

The foundation shape for three-dimensional simulations is defined by a rectilinear polygon,

meaning that it must be comprised of right angles. However, circles can also be simulated in two-

dimensions using axisymmetry in a cylindrical coordinate system. The polygon is specified by a

list of x-y Cartesian vertices tracing the foundation perimeter in a clockwise fashion (see Figure

3.3).

The perimeter of this polygon defines the location of the interior surface of the foundation

wall in the profile shown in Figure 3.1. The positioning of the foundation insulation and structural

components are translated into three-dimensional space internally.

The two-dimensional context for the foundation system description reduces the complexity

for the user to create a fully three-dimensional spatial domain. However, this convenience imposes

some limitations on the capability of the tool. For instance,

• Foundation types defined by variations in elevations cannot be modeled. Examples include

walk-out basements, split-levels, and underground structures.

 57

• It is not possible to simulate heat transfer between multiple thermal zones through the

ground since only one interior environment is defined.

The far-field boundaries define the extent of the domain in all four of the lateral directions.

These boundaries are created at a distance extending outward from a rectangle bounding the

polygon specified by the user (Figure 3.3). The far-field distance is a simulation input that defaults

to 40 meters.


  















  

 

Figure 3.3: Plan view illustration of far-field boundary definition in three-dimensions and founda-
tion shape vertex definition

3.3.2 Ground Domain

A reference to material definition sets the thermal properties (i.e., thermal conductivity, spe-

cific heat, and density) of the soil. Currently, the domain soil has spatially homogeneous properties

and does not account for effects such as moisture content or freezing.
 58

3.3.3 Boundary Conditions

The following sections describe the algorithms used to determine boundary conditions in

Kiva.

3.3.3.1 Interior Convection

The user assigns a temporally constant temperature for the interior thermal zone. This

temperature is used to calculate interior convection according to Equation 2.1. The user can also

override the interior convection coefficient to a constant value to replicate the idealistic constraints

required for analytical verification.

3.3.3.2 Interior Long-Wave Radiation

Interior long-wave radiation calculation does not utilize an algebraic method to predict radia-

tive exchange among interior surfaces. Instead, surface radiation is approximated as an exchange

with a black-body enclosure at room temperature using Equation 2.3.

3.3.3.3 Interior Short-Wave Radiation

Kiva is programmed to account for incident short-wave solar radiation on the interior foun-

dation surfaces. As a stand-alone tool, Kiva does not calculate transmitted solar through windows

and assumes that interior solar incidence is zero.

3.3.3.4 Foundation Wall

The boundary at the top of the foundation wall is adiabatic. In the context of a whole build-

ing simulation, conduction through above-grade walls can still be simulated as a one-dimensional

problem, implying that multi-dimensional impacts at the wall interface are neglected. The wall

boundary can also be set as a constant linear temperature profile to replicate the idealistic con-

straints required for analytical verification.

 59

3.3.3.5 Exterior Convection

Weather data in the EnergyPlus Weather (EPW) file format drives the exterior boundary

conditions in Kiva. Exterior convection is simulated using the relationships described in Section

2.2.3.4. Kiva assumes that the weather station is located in flat, open country at a height of 10

meters and adjusts the wind speed using a user-specified boundary layer thickness and terrain

exponent from Table 2.2. A user-specified vegetation height defines the height for the wind speed

adjustment and the roughness factors used for forced convection.

3.3.3.6 Exterior Long-Wave Radiation

Exterior long-wave radiation is calculated according to Equation 2.16 to account for radiation

exchange between the surface and the ground, air, and sky.

3.3.3.7 Exterior Short-Wave Radiation

The EPW file provides the direct normal and diffuse horizontal components of solar radiation.

The location latitude and longitude in the EPW file are used to calculate the solar incidence angle

according to the solar position equations provided by Duffie and Beckman [43]. Kiva contains an

experimental capability for simulating the effect of shading on exterior surfaces. This technique

uses the graphics processing unit of the computer to count the pixels of a surface rendered from the

perspective of the sun. The application of pixel counting for use in solar shading calculations was

originally presented by Jones and Greenberg [58]. While Jones and Greenberg [58] demonstrated

that this technique greatly improves the speed of shading calculations, their tests were applied to

cases where the number of shading surfaces was high relative to the number of receiving surfaces.

Preliminary tests of this technique applied to Kiva, where there are potentially thousands of re-

ceiving surfaces (one for each exterior surface cell), yielded prohibitively long computation times.

The effect of shading is not explored further in this thesis, but remains an area for improvement

within the Kiva framework. For the results presented here, Kiva applies unimpeded solar incidence

to each surface based on orientation relative to the sun’s position.

 60

3.3.3.8 Other Boundaries

Kiva does not calculate the effects of evapotranspiration or snow cover.

The far-field boundary is applied as a zero-flux boundary. By default, the deep-ground bound-

ary is a zero-flux boundary, but the user may choose to apply a constant temperature boundary

using the annual average dry-bulb temperature or a user-specified temperature.

3.3.4 The Whole-Building Context

As an open-source framework, relatively little overhead effort is required to reuse the code in

the context of an established whole-building energy simulation tool. Portions of the code have been

written explicitly to accommodate a more dynamic interaction between a building and the ground

domain, including variable interior temperatures, incident short-wave heat fluxes, and internal heat

sources (such as radiant slab heating or ground-loop heat exchangers).

3.4 Computational Algorithms

This section describes the solution methods used in Kiva to calculate heat transfer from

building foundation and through the surrounding soil.

3.4.1 Heat Diffusion

Heat transfer in the ground domain is solved using the diffusion equation

∂T
CH,T = ∇ · [k∇T ] (3.4)
∂t

which assumes there is no moisture flow. The thermal properties, k and CH,T = cp ρ, are assigned

by the materials referenced for the soil and the foundation design elements.
 61

3.4.2 Discretization

3.4.2.1 Spatial Domain

Equation 3.4 is solved for the solid portion of the domain bounded vertically by zwall-top

and zdeep-ground , and laterally by the far-field boundaries illustrated in Figure 3.3. For three-

dimensional simulations the user can request that the domain use planes of symmetry if they exist

for the foundation shape defined by the user. Kiva automatically detects symmetry in both the x

and y directions and applies zero-flux boundaries to reduce the domain size. Orientation-specific

boundary conditions, such as solar incidence and wind driven convection, are applied to the reduced

domain using the average values of mirrored surfaces.

Kiva is capable of approximating the three-dimensional domain using a two-dimensional

coordinate system. The effectiveness of several two-dimensional approximation methods is the

focus of Chapter 5.

The spatial discretization is defined for four distinct regions of the domain. The region

bounding the foundation wall and insulation elements defines the near-field region. All other

regions are defined either laterally (interior and exterior regions) or vertically (deep region) relative

to the near-field region. The near-field, interior, exterior, and deep regions are illustrated in Figure

3.4.

Cells grow geometrically towards the far-field, deep-ground, and symmetry boundaries. Cells

grow towards the center of each element within regions that do not border one of the domain

boundaries. Figure 3.5 illustrates cell growth and distribution for two examples, one in profile view

and the other in plan view.

Cell size and growth are determined by five parameters: four which determine the cell growth

coefficient in each of the regions and another parameter defining the minimum cell size for the entire

domain. The minimum cell size defines the allowable number of cells between element or region

boundaries. The cells’ sizes are allowed to increase to fit within the element or region boundaries

according to the growth and distribution of the cells. For example, the length of N consecutively
 62



  

   

  
   

 

 

(a) Profile view in two-dimensional context (b) Plan view

Figure 3.4: Discretization regions

(a) Profile: basement with exterior vertical and (b) Plan: uninsulated, J-shaped slab. Growth coef-
horizontal insulation. Growth coefficient = 1.20 ficient = 1.20 (for all regions), minimum cell size =
(for all regions), minimum cell size = 0.01 meters 0.1 meters

Figure 3.5: Example Kiva discretizations

growing cells is given by


N
l = Δxmin + Δxmin fg + Δxmin fg2 + . . . + Δxmin fgN = Δxmin fgi , (3.5)
i=0
 63

where Δxmin is the minimum cell size [m] and fg is the growth coefficient. If l is greater than the

distance between two element or region boundaries at N but not at N − 1, then the actual cell size

is set using
l
Δxactual = N
−1
. (3.6)
fgi
i=0

Kiva creates extra cells along the boundaries to represent surface temperatures and heat

fluxes. These cells have no thickness and are completely planar with the surface boundaries. The

other cells within the same plane as these boundary cells also have no thickness and require modified

material properties if the normal cells on either side of them are different materials. Figure 3.6

exemplifies zero-thickness cells resulting from the wall and floor boundaries. In this figure, a node

represents the center of each cell. The interior air cells are set to the interior air temperature and

have no impact on the calculation of the other cell temperatures.

Figure 3.6: Boundary cells and resulting zero-thickness cells. Black nodes = normal solid cells,
white nodes = boundary cells, gray nodes = zero-thickness cells, hollow nodes = air cells.
 64

3.4.2.2 Temporal Domain

Kiva can simulate any number of years of building operation. The user can select any

time-step duration. There are four initialization methods available: three traditional initialization

methods (constant temperature, Kusuda, and steady-state) and an accelerated initialization method

with several variants. The effectiveness of each of these initialization methods at predicting the

steady-periodic conditions of the ground prior to a simulation is discussed in Chapter 4.

3.4.3 Finite Difference Methods

Of the various mathematical solutions employed by other tools and methods described in

Table 2.3, the finite difference methods offer several benefits over the analytical and finite element

methods. Finite difference methods allow for more flexibility in describing boundary conditions

than analytical methods and are less mathematically abstract than finite element methods.

In general, the variants of finite difference methods are characterized by how the new time-

step is solved relative to the previous time-step.

Consider a very simple one-dimensional problem, discretized into three cells (one internal

cell, i, and two boundary cells, i − 1 and i + 1) as shown in Figure 3.7.

i-1 i i+1
t

t+1

Figure 3.7: Example finite difference scheme (black nodes are known values, white nodes are
unknown values)
 65

The problem is always defined at the previous time-step, t (as defined by the initial conditions

or subsequent time-step solutions). For some finite difference schemes, the solution for the new time-

step, t+1, can depend solely on previous time-step’s known values and can be solved independently

(explicitly) to find the solution for the new time-step. Other solutions may depend on other

unknown values and must be solved simultaneously (implicitly) using linear algebra techniques.

The manner in which each scheme solves the discretized solution will affect the computation

time and the stability of the solution as well as how physically realistic the solution is. Though

descriptions of each scheme can be found in the literature, they are rarely illustrated together using

consistent notation. Understanding the differences between the schemes’ mathematical implemen-

tations is essential to understanding the differences observed in the results.

3.4.3.1 Explicit Scheme

The simplest explicit scheme is one which only uses values from the previous time-step to solve

for the new time-step. Each of the three unknowns shown in Figure 3.7 can be solved independently

and sequentially based on the previous values of the node and its neighbors (indicated by the arrows)

as shown in Figure 3.8.

i-1 i i+1 i-1 i i+1 i-1 i i+1

t t t

t+1 t+1 t+1

(a) Step 1 (b) Step 2 (c) Step 3

Figure 3.8: Illustration of simple explicit finite difference scheme

 66

Mathematically, the dependencies are as follows:

t+1
 
Ti−1 = f T ti−1 , T ti
 
Tit+1 = f T ti−1 , T ti , T ti+1 (3.7)

t+1
 
Ti+1 = f T ti , T ti+1 ,

where boldface terms values are known values.

The simple explicit scheme is conditionally stable and does not always provide physically

realistic solutions. The solution becomes more stable with smaller time-steps and larger cell sizes.

3.4.3.2 Implicit Scheme

In an implicit scheme, all unknowns must be solved simultaneously since they cannot be

explicitly calculated from known values. In the basic implicit scheme (often referred to as “fully”

implicit), each unknown in the new time-step is calculated from its previous time-step value and

the values of its neighbors in the new time-step. In Figure 3.9, there is only one step since all

solutions must be simultaneous (as indicated by the bidirectional arrows).

i-1 i i+1
t

t+1

Figure 3.9: Illustration of fully implicit finite difference scheme

 67

Mathematically, the dependencies are as follows:

t+1
 
Ti−1 = f T ti−1 , Tit+1
 
Tit+1 = f T ti , Ti−1
t+1 t+1
, Ti+1 (3.8)

t+1
 
Ti+1 = f T ti+1 , Tit+1

Since each equation has multiple unknowns, implicit schemes require linear algebra techniques

to solve the simultaneous set of equations. The linear system can be solved either directly (using

Gaussian elimination) to arrive at an exact solution (within the floating point precision of the

machine), or they can be solved iteratively within a specified convergence tolerance. In general,

direct solvers are more robust but can require substantially more memory. While Kiva has the

capability to use both types of solvers, it uses an iterative solver by default.

The fully implicit scheme is both unconditionally stable and always provides physically real-

istic solutions.

3.4.3.3 Crank-Nicolson Scheme

The Crank-Nicolson scheme is another implicit scheme, but it relies on the average neighbor-

ing values from both the previous time-step and the new time-step as a better approximation of

the time derivative (Figure 3.10).

i-1 i i+1
t

t+1

Figure 3.10: Illustration of Crank-Nicolson finite difference scheme

 68

Mathematically, the dependencies are as follows:

t+1
 
Ti−1 = f T ti−1 , T ti , Tit+1
 
Tit+1 = f T ti−1 , T ti , T ti+1 , Ti−1
t+1 t+1
, Ti+1 (3.9)

t+1
 
Ti+1 = f T ti , T ti+1 , Tit+1

3.4.3.4 Alternating Direction Explicit (ADE) Scheme

A promising explicit scheme is the alternating direction explicit (ADE) scheme developed by

Saul’yev [100]. This scheme is similar to the simple explicit scheme except that it uses the known

values of the subsequently calculated neighbor nodes to calculate the value for the new time-

step. The unknown values are solved independently using sweeps starting from each boundary (see

Figures 3.11 and 3.12).

i-1 i i+1 i-1 i i+1 i-1 i i+1

t t t

t+1 t+1 t+1

(a) Step 1 (b) Step 2 (c) Step 3

Figure 3.11: Illustration of ADE finite difference scheme: upward sweep

Mathematically, the dependencies for the upward sweep temperatures, U , and the downward

sweep temperatures, V , are as follows:

t+1
 
Ui−1 = f U ti−1 , U ti
 
Uit+1 = f U t+1
i−1 , U i , U i+1
t t
(3.10)

t+1
 
Ui+1 = f U t+1
i , U i+1
t
 69

i-1 i i+1 i-1 i i+1 i-1 i i+1

t t t

t+1 t+1 t+1

(a) Step 1 (b) Step 2 (c) Step 3

Figure 3.12: Illustration of ADE finite difference scheme: downward sweep

t+1
 
Vi+1 = f V ti , V ti+1
 
Vit+1 = f V ti−1 , V ti , V t+1
i+1 (3.11)

t+1
 
Vi−1 = f V ti−1 , V t+1
i

The final temperature, T , is then the average of the two independent solutions:

U +V
T = (3.12)
2

Because each solution is independent, they can be solved in parallel to further reduce com-

putation time.

The ADE scheme has been shown to be unconditionally stable and give physically realistic

results when used with a uniform grid. However, Fukuyo [46] has shown that the ADE method is in

fact conditionally stable for nonuniform grids with some fairly loose stability requirements related

to cell size and the relative growth of cells. For applications of foundation heat transfer, the ADE

solution is significantly more stable than the standard explicit solution.

3.4.3.5 Alternating Direction Implicit (ADI) Scheme

Each of the schemes described so far can be explained with a simple one-dimensional exam-

ple. The alternating direction implicit (ADI) scheme applies only to problems with two or more
 70

dimensions. In the typical implicit schemes described (i.e., fully implicit and Crank-Nicolson), each

nonboundary cell requires an equation with three unknowns (i.e., the cell itself, the cell before, and

the cell after). In a three-dimensional solution, each cell requires an equation with seven unknowns

(i.e., itself, top, bottom, east, west, north, and south).

A common way to represent these problems is to show a matrix sparsity pattern. This plot

shows a row for every equation and a column for every unknown in the discretized system. A block

is filled in for each unknown that appears in the given equation. The sparsity pattern for a typical

implicit scheme is shown in Figure 3.13.

Figure 3.13: Typical implicit matrix sparsity pattern for three dimensions (up to seven unknowns
per equation)

Unlike the ADE method, where “alternating direction” refers to the order in which the equa-

tions are solved (using alternating sweeps), the alternating direction in ADI refers to something

entirely different. Here, each time-step is divided into smaller sub-time-steps: one for each dimen-

sion in the problem. Each sub-time-step is solved with only the unknowns along the corresponding

dimension (e.g., x, y, or z) being solved implicitly (at the new sub-time-step), with the values of the

other directions using the values from the previous time-step similar to the simple explicit scheme.

The equations for each sub-time-step can be arranged so that all the unknowns fall along the

three inner-most diagonals in the sparsity pattern (as shown in Figure 3.14).
 71

Figure 3.14: ADI matrix sparsity pattern for three dimensions (up to three unknowns per equation)

This linear algebra problem can be solved quickly and directly using a standard tri-diagonal

matrix algorithm. This algorithm is significantly faster than an iterative solver and other more

generalized direct solvers. It is, however, less efficient than the explicit schemes because it performs

two serial passes over the data elements (i.e., it cannot be parallelized like the ADE approach), and

it must calculate additional sub-time-steps within each time-step corresponding to each dimension

in the problem.

The specific ADI scheme used in this work is described by Chang et al. [21]. This scheme

introduces some nuances that can improve solution stability and accuracy, even at larger time-steps.

Although one cannot guarantee stability for applications of foundation heat transfer, the stability

of the ADI solution has not been found to be an issue.

3.4.3.6 Discretized Equations

The formulations of the discretized approximations of Equation 3.4 for each of the finite

difference methods described above are provided in Appendix A for a two-dimensional axisymmetric

cylindrical coordinate system. Formulations for Cartesian two- and three-dimensional coordinate

systems can be easily extended by omission or addition of certain terms as described in the appendix.
 72

3.4.3.7 Solving Linear Systems of Equations

All implicit finite difference schemes require linear algebra techniques to solve a set of si-

multaneous equations resulting from discretized equations. The implicit schemes available in Kiva

include fully implicit, Crank-Nicolson, and Alternating Direction Implicit (ADI) schemes. As pre-

viously mentioned, the tri-diagonal nature of the ADI scheme lends itself to a straightforward,

computationally efficient solution for linear system of equations. The other schemes utilize iter-

ative linear solvers available in the Library of Iterative Solvers (Lis) [106]. Lis is an open-source

library with several types of iterative solvers including

• Conjugate gradient (CG),

• Biconjugate gradient (BiCG),

• Conjugate gradient squared (CGS),

• Biconjugate gradient stabilized (BiCGSTAB),

• Generalized product biconjugate gradient (GPBiCG),

• Transpose-free quazi-minimal residual (TFQMR),

• Orthomin,

• Generalized minimum residual method (GMRES),

• Jacobi,

• Gauss-Seidel, and

• Successive over-relaxation (SOR).

Lis also provides several preconditioners for accelerating the solution process including

• Jacobi,

• Incomplete LU factorization (ILU),

 73

• Symmetric successive over-relaxation (SSOR), and

• Smoothed aggregation - algebraic multigrid (SA-AMG).

Each combination of a solver and, if selected, a preconditioner impacts the efficiency of the

iterative process differently. A limited test of several combinations revealed that the biconjugate

gradient stabilized (BiCGSTAB) solver, coupled with an incomplete LU factorization (ILU) precon-

ditioner, provided the fastest convergence and most consistent results relative to other combinations

tested. This is the default combination used in Kiva.

3.5 Outputs

Kiva has the capability of providing both time-series and graphical output. The time-series

outputs are as follows:

• surface average heat flux [W/m2 ],

• surface average temperature [K],

• surface average effective temperature [C],

• surface total heat transfer rate [W],

• foundation average heat flux [W/m2 ],

• foundation average temperature [K], and

• foundation total heat transfer rate [W];

where effective temperature is the temperature that can be applied as a preprocessed boundary con-

dition for foundation surfaces in a whole-building simulation tool. The optional surfaces available

for output are the

• slab core,

• slab perimeter, and

 74

• foundation wall;

where the core and perimeter surfaces of the slab are defined by a user-specified distance inward

from the foundation perimeter.

Kiva can output images representing any two-dimensional slice of the spatial domain and any

point within the temporal domain. A series of images can be created at a user-specified frequency

for compilation into an animation. Figure 3.15 demonstrates this capability with a profile slice and

a plan slice for the same domains illustrated in Figure 3.5. Kiva can create plots of temperature,

heat flux magnitude, and heat flux in the direction of any of the coordinate system axes.

(a) Profile: basement with exterior vertical and hori- (b) Plan: uninsulated, J-shaped slab at 3 meters
zontal insulation (isotherms equally spaced at 2.5◦ C) depth (isotherms equally spaced at 5.0◦ C)

Figure 3.15: Example Kiva graphical output

 75
Table 3.1: Example foundation types available in Kiva

Foundation Type Δzf oundation [m] Δzwall-above-grade [m] Illustration

Deep basement 2.4 0.3

Shallow basement 2.4 1.2

Below-grade
1.0 0.3
crawlspace

Shallow below-grade
0.6 0.3
crawlspace

Above-grade
1.0 1.0
crawlspace

Slab-on-grade 0.0 0.2

Slab-in-grade 0.0 0.0

Elevated slab 0.0 0.6

 Chapter 4

Solution Verification

It is important to distinguish between simulation validation and solution method verification.

Simulation validation describes the effort of determining the accuracy of simulation outputs when

compared to carefully measured results from a controlled experiment. Solution method verification

describes the process of testing the correctness of the algorithms as programmed in a tool.

While, ultimately, an empirical validation may provide the best truth standard for deter-

mining a solution’s accuracy, it is difficult to design an experiment where the properties of the

foundation and ground domain are known within a reasonable degree of uncertainty. When creat-

ing a model of the foundation and ground system, this uncertainty propagates through the solution

resulting in large error bounds on the output results of the simulation. Most attempts to validate

a simulation of a foundation heat transfer result in a calibration of uncertain model inputs rather

than a validation of the simulation outputs.

For this reason, validation is not an objective of this thesis. This is not intended to undermine

the importance of validation, but given the challenges it presents and the evident need to improve on

the computational performance and algorithmic confidence of current solution methods, validation

is considered beyond the scope of this work.

Kiva provides a framework with several variants of the finite difference method allowing direct

comparisons of the accuracy, computation time, and stability of each method as they are applied to

solving the foundation heat transfer problem. Such comparisons have not been documented before

in the literature.
 77

Before the Kiva framework can be used to develop faster simplified solutions, each of the

finite difference solution methods implemented in Kiva must be independently verified against an

existing mathematical truth standard. This will establish Kiva as an accuracy benchmark for

justifying future simplifications.

This chapter focuses on the verification of the finite difference solution methods implemented

in Kiva and discussed in Section 3.4.3. This chapter also investigates solution sensitivity to selected

numeric parameters as well as the efficiency of several initialization techniques within the context

of an established, well-defined test specification.

4.1 IEA BESTEST for Ground Coupled Heat Transfer

As all numerical discretizations are approximations of the continuous nature of heat diffusion,

the goal is to compare the accuracy of these approximations against a known analytical solution. For

foundation heat transfer, the known analytical solutions tend to be highly constrained, mathemat-

ically idealized, physically unrealistic cases that can only provide a starting point for verification.

Beyond these solutions, the only available comparisons are to other independently developed nu-

merical solutions. While these numerical solutions cannot be considered truth standards like the

analytical solutions, a general consensus among detailed numerical solutions can establish a general

range of acceptance.

The International Energy Agency Building Energy Simulation Test and Diagnostic Method

(IEA BESTEST ) [90] describes a methodology to verify solutions for foundation heat transfer

against analytical solutions and verified numerical solutions. A single analytical solution developed

by Delsante et al. [38] was used as a steady-state mathematical truth standard for a single test

case. For the same test case, three detailed numerical solutions were developed and shown to give

very close agreement with the analytical solution. These three solutions, TRNSYS [109], FLUENT

[89], and MATLAB [31], were then used as secondary mathematical truth standards for the other

steady-state and harmonic (unsteady) test cases. These are all idealized test cases with several

inherent limitations:
 78

• The test cases are for slab-in-grade foundations only. They do not test slabs-on-grade,

basement, or crawlspace foundations.

• The test cases do not test the sensitivity to any level of insulation. The slabs are all

uninsulated.

• The entire domain has homogeneous soil properties, meaning that the concrete of the slab

and foundation wall has the same thermal properties as the soil.

• The boundary conditions are approximated with constant film coefficients. There is no

distinction among natural convection, forced convection, and long-wave radiation.

• There is no incident solar radiation on the exterior soil surface.

Results from the analytical truth standard and the three verified numerical solutions are

compared to the solution methods used in Kiva for each of the BESTEST test cases. Each case

uses symmetry to reduce the size of the domain and, consequently, the number of computations

(to 1/4). The BESTEST procedures do not specify any sort of pass/fail criteria for the solutions

outside of subjective comparison. For the purposes of this paper, verification implies there are no

statistical outliers among the different solutions.

Before this work, the accuracy and computation time of different finite difference schemes have

not been compared side-by-side in a common framework for foundation heat transfer calculations.

Traditionally, researchers have selected schemes based on (1) ease of implementation (typically

explicit schemes), (2) stability (fully implicit methods), (3) or availability of commercial software

(usually finite element). Each of the three secondary mathematical truth standards uses some form

of implicit scheme requiring use of a linear solver:

• TRNSYS: Fully implicit finite difference with iterative solver

• FLUENT: Finite element method with iterative solver

• MATLAB: Fully implicit finite difference with direct solver

 79

The original source of secondary mathematical truth standards is not available to measure

their computation performance under a consistent computational environment, nor were the simu-

lation times provided in the BESTEST literature. However, due to their implicit nature, each of

the solutions is expected to require computation times on the same order of magnitude as Kiva’s

fully implicit solution.

4.1.1 Steady-State Test Cases

Table 4.1 shows the specification of each of the steady-state test cases. The only distinction

between cases not listed in Table 4.1 is that the boundary at the wall/ground interface is a constant

temperature profile for the analytical test case (GC10a) and adiabatic for all other cases.

Table 4.1: BESTEST steady-state test case specifications

Slab dimensions hint hext Ground depth Far-field width Soil conductivity
Case
[m x m] [W/m2 ·K] [W/m2 ·K] [m] [m] [W/m·K]
GC10a 12 x 12 ∞ ∞ ∞ ∞ 1.9
GC30a 12 x 12 ∞ ∞ 30 20 1.9
GC30b 12 x 12 100 100 15 15 1.9
GC30c 12 x 12 7.95 ∞ 15 8 1.9
GC60b 12 x 12 7.95 100 15 15 1.9
GC65b 12 x 12 7.95 11.95 15 15 1.9

Other characteristics similar to all steady-state test cases include the following:

• Interior air temperature = 30 ◦ C,

• Foundation wall thickness = 0.24 m,

• Outside air temperature = 10 ◦ C, and

• Deep ground temperature = 10 ◦ C.

The IEA BESTEST User’s Manual specification (Part I of Neymark and Judkoff [90]) for

the analytical steady-state comparison case, GC10a, states the following:

Demonstrate that the current level of modeling detail yields negligible (≤ 0.1%)
change in results versus a lesser level of detail; examine the effects of shallower deep
 80

ground boundary, shorter far-field boundary distance, and less detailed mesh (if
applicable). For ground depth and far-field length variations, increase the number
of mesh nodes proportionally to the increase in soil volume being modeled.

The following sections describe the sensitivity of the GC10a results to changes in mesh detail,

boundary distances, and solver convergence tolerance. These analyses were performed using the

steady-state finite difference solution. Other solutions are expected to have similar results, when

scaled accordingly. The results from the sensitivity analysis are used to inform the numerical

parameters used elsewhere in this study.

4.1.1.1 Sensitivity to Mesh Detail

Each of the three secondary mathematical truth standards uses a nonuniform mesh where the

mesh size increases geometrically with the distance from the foundation wall boundary. This is also

the case with Kiva where the mesh is uniform along the wall boundary (in the x and y directions)

and increases geometrically downward (towards the deep-ground boundary), inward (towards the

center-of-slab planes of symmetry), and outward (towards the far-field boundaries). Figure 4.1

shows a profile of the mesh generated by Kiva near the wall boundary for the GC10a test case.

Sensitivity to mesh detail was tested by varying the minimum allowable cell dimension,

Δxmin , and the geometric growth factor, fg . The size of each cell, i cells away from the boundary

cell is defined by Equation 4.1:

Δxi = Δxmin · fgi−1 (4.1)

The minimum allowable cell size is used to set the size and number of cells in the uniform

mesh region. The actual cell size is allowed to increase in order to fit uniformly into the dimensions

allotted (in this case, the width of the wall boundary). Similarly, the growth factor is allowed to

decrease in order to fit an exact number of cells within the domain’s boundaries.

To demonstrate that the tested mesh detail yields negligible change in results versus a less

detailed mesh, Kiva was tested under a combination of different minimum cell dimensions and

growth factors over a range of values. The results are shown in Figure 4.2. The approximated slab
 81

Figure 4.1: BESTEST GC10a mesh growth near the wall boundary

heat loss increases as the geometric growth factor decreases before settling near the value of the

analytical solution at geometric growth factors less than 1.3. The slight peak at growth factors

of 1.3 may be related to the actual sizing of the cells relative to the input minimum cell size and

maximum growth factor. Simulations at higher levels of detail (i.e., fg < 1.10) were beyond the

memory limitations of the computing hardware, so it is difficult to say definitively whether the

trends towards convergence continues.

No report from the three secondary mathematical truth standards indicated a complete

parametric sweep across ranges of these two parameters. Instead, mesh independence was tested

either with a single variable sweep (i.e., either Δxmin or fg , but not both) or with a small subset

of combinations. Therefore, it is difficult to know whether the other tools demonstrated similar

trends towards convergence.

Figure 4.3 shows how the simulation wall time increases significantly as the geometric growth

factor, fg , and the minimum cell dimension, Δxmin , decrease.

 82

2500
Min. cell dim., ¢xmin
2480 3 mm 9 mm Analytical 2%

Slab heat loss [W]

6 mm 12 mm +/- 0.1%
2460 1%

2440
0%
2420
-1%
2400

-2%
2380
2.0 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1
Geometric growth factor, fg

Figure 4.2: BESTEST GC10a slab heat loss sensitivity to mesh detail parameters

35
Min. cell dim., ¢xmin
30
Simulation wall time [s]

3 mm
6 mm
25
9 mm
12 mm
20

15

10

0
2.0 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1
Geometric growth factor, fg

Figure 4.3: BESTEST GC10a simulation wall time sensitivity to mesh detail parameters

The combination of fg and Δxmin selected for the Kiva BESTEST solutions is based on

simulation wall time as well as the values selected by the other secondary mathematical truth
 83

standards. These parameters are shown in Table 4.2 along with the selected values by the other

tools.

Table 4.2: Mesh detail parameters selected by each tool for IEA BESTEST analyses

TRNSYS FLUENT MATLAB Kiva

Δxmin [mm] 25.4 6.0 9.0-20.0 6.0
fg 1.20-1.30 1.10 1.15-1.50 1.20

4.1.1.2 Sensitivity to Boundary Distances

The analytical solution for GC10a represents the heat transfer through the slab into ground

that extends infinitely outward and downward, so the numerical representation must demonstrate

that variation in the boundary distances (i.e., the deep-ground depth and far-field width) yields

negligible differences in the results.

The sensitivity to the boundary distances (far-field width and deep-ground depth) are shown

in Figure 4.4. Generally, there is little sensitivity when the boundary distances are both 20 m or

greater.

Figure 4.5 shows that simulation wall time has relatively little sensitivity to the boundary

distances because the geometrically increasing mesh size requires very few additional cells near

the boundary edges for increased boundary distances. Since there is little computation cost for

improved accuracy, both boundary distances were set to 40 m for the GC10a test case.

4.1.1.3 Sensitivity to Convergence Tolerance

IEA BESTEST specifies a similar test for the convergence of iterative solvers [90]:

If the software being tested allows user specification of convergence tolerance,

demonstrate that the current level of heat-flow or temperature convergence toler-
ance yields negligible (≤ 0.1%) change in results versus the next finer convergence
tolerance.
 84

2500

2480 2%

Slab heat loss [W] 2460 1%

2440
0%
2420

Far-field width -1%

2400 10 m 30 m Analytical
20 m 40 m +/- 0.1%
-2%
2380
10 15 20 25 30 35 40
Deep-ground depth [m]

Figure 4.4: BESTEST GC10a slab heat loss sensitivity to boundary distances

35
Far-field width
30 10 m
Simulation wall time [s]

20 m
25 30 m
40 m
20

15

10

0
10 15 20 25 30 35 40
Deep-ground depth [m]

Figure 4.5: BESTEST GC10a simulation wall time sensitivity to boundary distances

Kiva uses the Library of Iterative Solvers (Lis) [106] to solve the system of equations at

each time-step for the fully implicit, Crank-Nicolson, and steady-state solutions. As with any
 85

iterative solver, it is necessary to specify a convergence tolerance. For the linear system, Ax = b,

convergence in Lis is defined as such:

b − Ax2 ≤ tol · b2 , (4.2)

where b − Ax2 is the 2 -norm of the residual.

Figure 4.6 shows the sensitivity of the slab heat loss and simulation wall time to tighter

tolerances. By a tolerance of 10−6 there is less than a 0.1% change in results. Tighter convergence

tolerance has relatively small impact on simulation wall time.

2500 35
Heat flow
Analytical 30

Simulation wall time [s]

2480
+/- 0.1%
Slab heat loss [W]

Time 25
2460
20
2440
15
2420
10

2400 5

2380 0
10-4 10-5 10-6
Tolerance

Figure 4.6: BESTEST GC10a slab heat loss and simulation wall time sensitivity to linear solver
tolerance

4.1.1.4 Sensitivity Analysis Summary

The previous sections demonstrate that the level of modeling detail for the GC10a test case

yields a negligible change in results when compared to a lesser level of detail.

The numerical sensitivity results were derived from steady-state solutions of case GC10a.

Extrapolating the results beyond this case implies the following assumptions:
 86

• The accuracy for other cases, and other solution schemes, will show similar sensitivity.

• The computation wall time for other numerical schemes will demonstrate similar sensitivity

and scale proportionally to that of the steady-state solution.

The selected parameters affecting the numerical approximation are shown in Table 4.3 and

are used for all the remaining test cases and analyses unless otherwise specified.

Table 4.3: Selected numerical parameters for Kiva analyses

Parameter Value
Time-step, Δt [min] 60
Minimum cell dimension, Δxmin [mm] 6.0
Maximum cell growth factor, fg 1.20
Far-field width [m] 40
Deep-ground depth [m] 40
Tolerance 10−6

4.1.2 Steady-State Test Case Results

The slab heat loss for the steady-state BESTEST cases are shown in Figure 4.7 and in Table

4.4. It is evident that each of Kiva’s solution methods produce results within the statistical variation

of the secondary truth standards.

All of Kiva’s solution methods provided accurate results, but, in some cases, the discretiza-

tion parameters had to be relaxed in order to ensure stable solutions. If instabilities arose for a

particular solution, the time-step was reduced first to ensure that there were no changes to the

spatial discretization between solution methods. If a minute-long time-step was not enough to

satiate the stability requirement, subsequent changes were made to the minimum cell dimension

(Δxmin ).

For all the steady-state cases, the explicit solution method required a minute-long time-step

and a minimum cell dimension of 20 mm. The ADE solution method required 30 minute time-

steps for the cases with infinite convective coefficients (i.e., the a and c cases). These additional
 87

3000

2500

Slab heat loss [W] 2000

1500

1000

500

0
GC10a GC30a GC30b GC30c GC60b GC65b
Analytical FLUENT Kiva: Steady-State Kiva: Explicit Kiva: Implicit
TRNSYS MATLAB Kiva: ADE Kiva: ADI Kiva: Crank-Nicolson

Figure 4.7: BESTEST steady-state test case slab heat loss

time-steps resulted in a proportional increase in simulation wall time. Figure 4.8 (note logarithmic

y-axis) and Table 4.5 show the total simulation time required for each solution method and for each

case. Each Kiva simulation was performed on a 24 core machine with 96 GB of available memory.

The CPUs are Intel Xeon X5660 2.8 Ghz with 12 MB cache.

The IEA BESTEST User’s Manual (Part I of Neymark and Judkoff [90]) specification for

steady-state cases states the following:

Initially (time < 0) the ground and slab are at 10 ◦ C (50 ◦ F) throughout; at the
beginning of the simulation (time = 0) the zone air temperature steps to 30 ◦ C (86
◦ F) and slab and soil temperatures begin to change accordingly.

and

If possible for the software being tested, the simulation should be run long enough
that there is ≤ 0.1% variation between the floor slab conduction for the last hour
of the last year of the simulation and the last hour of the preceding year of the
simulation; i.e., ≤ 0.1% variation over an interval of 8760 hours.
 88
Table 4.4: BESTEST steady-state test case results: Slab heat loss [W]

GC10a GC30a GC30b GC30c GC60b GC65b

Analytical 2,433 – – – – –
TRNSYS 2,427 2,642 2,533 2,137 2,113 1,994
FLUENT 2,425 2,585 2,504 2,123 2,104 1,991
MATLAB 2,432 2,695 2,570 2,154 2,128 2,004
Kiva: Steady-State 2,435 2,667 2,554 2,153 2,129 2,007
Kiva: ADE 2,435 2,667 2,554 2,153 2,129 2,007
Kiva: Explicit 2,420 2,631 2,528 2,137 2,118 2,000
Kiva: ADI 2,422 2,572 2,497 2,123 2,108 1,997
Kiva: Implicit 2,435 2,667 2,554 2,153 2,129 2,007
Kiva: Crank-Nicolson 2,435 2,667 2,554 2,153 2,129 2,007

105
Simulation wall time [s]

104

103

102

101

100
GC10a GC30a GC30b GC30c GC60b GC65b
Kiva: Steady-State Kiva: Explicit Kiva: Implicit
Kiva: ADE Kiva: ADI Kiva: Crank-Nicolson

Figure 4.8: BESTEST steady-state test case simulation wall times (note logarithmic y-axis)

Table 4.5: BESTEST steady-state test case simulation wall times [s]

GC10a GC30a GC30b GC30c GC60b GC65b

Kiva: Steady-State 8 5 6 6 7 6
Kiva: ADE 1,999 1,743 801 1,730 804 804
Kiva: Explicit 8,231 7,699 6,189 5,196 6,169 6,101
Kiva: ADI 2,180 1,891 1,807 1,628 1,809 1,802
Kiva: Implicit 13,108 12,103 11,852 9,273 11,995 12,080
Kiva: Crank-Nicolson 11,726 10,907 10,833 8,822 10,714 10,862
 89

meaning that, as specified, the steady-state test cases must use a constant initialization method,

simulated over as many years as required to reach the ≤ 0.1% criteria. However, the user’s manual

provides an exception (in section 1.2.1.6) if the software has other initialization methods:

If the program being tested allows for preconditioning (iterative simulation of an

initial time period until temperatures or fluxes, or both, stabilize at initial values),
use that capability.

Each of the unsteady solution methods were initialized with the steady-state solution. The

slab heat loss reported is the final value after responding to the defined boundary conditions after

a year’s time of simulation, requiring significantly more time than the purely steady-state solution.

Important observations of the resulting simulation wall times follow:

• There is relatively little difference in wall time among BESTEST cases for the same solution.

The ADE solution is the one exception as 30 minute time-steps were required in some cases

(roughly doubling the wall time).

• Steady-state simulations are orders of magnitude faster since there is only a single calcula-

tion of the domain.

• ADE simulations are the fastest, though when the time-step is reduced to 30 minutes, they

have nearly the same simulation time as the ADI simulations.

• Explicit simulations, due to their increased number of time-steps, have simulation times

near those of the solutions that require an iterative solver (i.e., Implicit and Crank-Nicolson).

• Given the relatively small change in accuracy among each of the solution methods, the ADE

and ADI methods show the most promise for typical annual energy analysis practices.
 90

4.1.3 Harmonic (Unsteady) Test Cases

Table 4.6 shows the specification of each of the harmonic test cases. For each harmonic test

case, the outdoor air dry-bulb temperature, Tdb,a is given by:

Nday − 15 Nhour − 4
Tdb,a = (10 ◦ C) − (6 ◦ C) · cos · 2 · π − (2 ◦ C) · cos ·2·π (4.3)
365 24

where Nday is the N th day of the calendar year, and Nhour is the N th hour of the day (0 =

midnight).

Table 4.6: BESTEST harmonic test cases specifications

Slab dimensions hint hext Ground depth Far-field width Soil conductivity
Case
[m x m] [W/m2 ·K] [W/m2 ·K] [m] [m] [W/m·K]
GC40a 12 x 12 ∞ ∞ 30 20 1.9
GC40b 12 x 12 100 100 15 15 1.9
GC40c 12 x 12 7.95 ∞ 15 8 1.9
GC45b 36 x 4 100 100 15 15 1.9
GC45c 36 x 4 7.95 ∞ 15 8 1.9
GC50b 80 x 80 100 100 15 15 1.9
GC55b 12 x 12 100 100 2 15 1.9
GC55c 12 x 12 7.95 ∞ 5 8 1.9
GC70b 12 x 12 7.95 11.95 15 15 1.9
GC80b 12 x 12 100 100 15 15 0.5
GC80c 12 x 12 7.95 ∞ 15 8 0.85

Other characteristics similar to all harmonic test cases include the following:

• Interior air temperature = 30 ◦ C,

• Foundation wall thickness = 0.24 m,

• Deep ground temperature = 10 ◦ C,

• Soil density = 1490 kg/m3 , and

• Soil specific heat = 1800 J/kg·K.

4.1.3.1 Initialization Method Analysis

As specified by the IEA BESTEST User’s Manual (Part I of Neymark and Judkoff [90]) for

harmonic test cases,

 91

If possible for the software being tested, the simulation should be run long enough
that there is <= 0.1% variation between the annual floor slab conduction for the
last year of the simulation and the preceding year of the simulation.

Reducing the number of years required to initialize steady periodic conditions can significantly

decrease computation time. BESTEST specifies that all temperatures in the ground domain be

initialized to a spatially-constant temperature by default, but it allows room for other initialization

methods as well:

If the program being tested allows for preconditioning (iterative simulation of an

initial time period until temperatures or fluxes, or both, stabilize at initial values),
use that capability.

This section explores the advantages and disadvantages of different preconditioning (referred

to here as initialization) methods. There are three traditional initialization methods implemented

in Kiva:

(1) Constant temperature: Spatially-constant initial temperature.

(2) Kusuda: A one-dimensional analytical solution developed by [75] that provides tempera-

ture variation as a function of depth driven by an annual harmonic temperature fluctuation.

There is no temperature variation in either of the horizontal dimensions.

(3) Steady-state: A steady-state solution scheme initializes the temperatures with the first

time-step’s boundary conditions. This provides an initial condition temperature variation

in all dimensions.

Kiva’s other initialization method, called the accelerated method, simulates the ground do-

main prior to the first time-step, t = 0. This presimulation period is itself initialized using a

steady-state solution. The presimualtion then commences with the following:

(1) A period of long time-step (on the order of days to months), implicit calculations. Since

the fully implicit method is unconditionally stable and physically realistic, these long time-
 92

steps can be used to coarsely capture some of the thermal history driven by the annual

weather cycle.

(2) A subsequent period of normal time-step calculations prior to capture some of the shorter

term dynamic effects.

The accelerated initialization method begins with a coarse timestep discretization and tran-

sitions to a finer discretization near the beginning of the simulation, similar to the way the spatial

discretization becomes finer as it approaches the wall boundary.

Figure 4.9 shows the GC40a slab heat loss for the first year of simulation. Four traditional

initialization methods (Kusuda, steady-state, and two different constant temperatures) are com-

pared with an accelerated method and the “ideal” result, which is the steady periodic solution

after several years of simulation (i.e., the net result is within 0.1% of the previous year’s heat loss).

Figure 4.10 shows how the annual heat loss for each initialization method approaches the “ideal”

steady periodic value over seven years (the percent difference is shown in Table 4.7).

5000
Kusuda
4500 Constant temperature (10 ± C)
Steady-state
Slab heat loss [W]

4000 Accelerated
Ideal
3500 Constant temperature (30 ± C)

3000

2500

2000

1500

1000
0 1000 2000 3000 4000 5000 6000 7000 8000
Hour of year

Figure 4.9: BESTEST GC40a comparison of initialization methods: First year slab heat loss

Important observations from these results are summarized below:

 93

30

Annual slab heat loss [MWh]

28 20%

26 10%
24
0%
22
-10%
20
Kusuda
Constant temperature (10 ± C) -20%
18 Steady-state
Accelerated
16 -30%
Ideal
Constant temperature (30 ± C)
14 -40%
1 2 3 4 5 6 7
Year

Figure 4.10: BESTEST GC40a comparison of initialization methods: Annual slab heat loss for
seven years after initialization

• The accelerated method is nearly identical to the “ideal” result, with only a slight deviation

near the beginning of the year.

• Of the four traditional methods, the steady-state initialization provides the best approxima-

tion of the “ideal” case for the first year, but the constant (10◦ C) and Kusuda initializations

respond faster to the dynamics and provide a better approximation in subsequent years.

• By the second year, the constant (10◦ C) method is the best approximation among the

traditional methods. However, the performance of this method depends on an appropriately

selected initial temperature. Initializing to 30◦ C instead produces the worst approximation.

• Although the Kusuda method allows for spatially varying initial temperatures and may

provide a reasonable approximation of the depth varying temperature in the far-field, it

does a poor job of approximating the initial conditions near the building.

The acceleration method exemplified above is one of many possible variants. Other variants

of the presimulation period can be defined using three parameters:

 94
Table 4.7: BESTEST GC40a initialization method percent difference from “ideal” for seven years
of simulation

Constant Constant
Kusuda Steady-state Accelerated
temperature (10◦ C) temperature (30◦ C)
Year
1 22.1% 18.7% 7.49% 0.01% -32.3%
2 2.09% 2.07% 3.91% -0.19% -19.0%
3 1.02% 1.02% 2.69% -0.14% -14.5%
4 0.65% 0.65% 2.00% -0.10% -11.4%
5 0.46% 0.46% 1.54% -0.06% -8.96%
6 0.35% 0.36% 1.21% -0.02% -7.03%
7 0.28% 0.29% 0.96% 0.00% -5.50%

• Δtaccel : The time-step duration of the acceleration period (1 day, 1 week, 1 month, etc.)

• tpre : The total duration of the presimulation period

• twarm-up : The total duration of the warm-up period (directly before primary simulation on

same time-step as primary simulation)

A wide range of potential accelerations were simulated to compare accuracy and relative

simulation time. This range represents all combinations of the following parameter values (for the

sake of consistency the duration of a month is defined as 4 weeks):

• Acceleration time-step, Δtaccel : 1 day, 1 week, 1 month

• Presimulation duration, tpre : 3 months, 6 months, 12 months, 24 months

• Warm-up period, twarm-up : none, 1 week, 1 month

In Kiva, the presimulation duration is not input by the user but is, instead, calculated

internally as follows:

tpre = Δtaccel · Naccel + twarm-up , (4.4)

where Naccel is the number of acceleration time-steps used during the presimulation period and is

an input into Kiva.

 95

Figure 4.11 shows a comparison of the relative accuracy and wall time of all the initialization

methods tested. In this figure, the relative difference in annual heat loss is the percent difference

between the first year of simulation and the “ideal” steady periodic results. The additional sim-

ulation wall time represents the difference in time relative to the fastest simulation. The initial

results revealed that the warm-up period, twarm-up , has negligible impact on the accuracy of the

results and added considerable time to the simulation. Variations of the warm-up period are not

illustrated in Figure 4.11 to maximize legibility.

Kusuda 3 months init. Daily time-steps

Constant temp. (10 ± C) 6 months init. Weekly time-steps
Constant temp. (30 ± C) 12 months init. Monthly time-steps
Steady-state 24 months init.

30%
annual slab heat loss
Relative difference in

20%

10%

0%

-10%

-20%

-30%

0 200 400 600 800 1000 1200

Additional simulation wall time [s]

Figure 4.11: BESTEST GC40a comparison of first year slab heat loss for traditional and accelerated
initialization methods

From Figure 4.11 we can infer the following conclusions:

• The acceleration method is consistently more accurate than the traditional initialization

methods.

• Warm-up periods do not improve first year accuracy.

 96

• There is not an acceleration time-step that is clearly more accurate than the others, though

daily acceleration time-steps showed significant additional wall time without a consistent

improvement in accuracy.

• There is no clear trend between the duration of the presimulation period and accuracy. By

three months a majority of the thermal history has been captured by the presimulation.

• The most accurate presimulation initialization was found at weekly acceleration time-steps

for a three month period. This may not be the best set of presimulation parameters for all

cases as it could change depending on the thermal characteristics of the domain.

4.1.4 Harmonic (Unsteady) Test Case Results

Each of the Kiva solution methods was used to simulate the BESTEST harmonic test cases.

Each simulation leveraged an accelerated initialization followed by a full year of warm-up to sat-

isfy the output requirements for allowable variation in slab heat loss between consecutive years.

While the accelerated initialization method allows Kiva to simulate only two years, the secondary

mathematical truth standard solutions required an average of seven years to initialize.

The results for the harmonic BESTEST cases are shown in Figure 4.12 and in Table 4.8. It

is evident that each of the solution methods in Kiva produce results within the statistical variation

of the secondary truth standards.

Table 4.8: BESTEST harmonic (unsteady) test case results: Annual slab heat loss [MWh]

GC40a GC40b GC40c GC45b GC45c GC50b GC55b GC55c GC70b GC80b GC80c
TRNSYS 23.0 22.1 18.6 32.8 27.0 277.9 35.1 20.8 17.4 6.0 9.2
FLUENT 22.8 21.9 18.6 32.5 26.9 278.0 34.9 20.7 17.4 5.9 9.1
MATLAB 23.6 22.5 18.9 33.5 27.4 281.4 35.5 21.0 17.6 6.2 9.3
Kiva: ADE 23.3 22.4 18.9 33.4 27.5 279.5 35.2 20.9 17.6 6.1 9.3
Kiva: Explicit 23.0 22.2 18.8 33.0 27.3 278.4 35.0 20.8 17.6 6.0 9.2
Kiva: ADI 22.5 21.9 18.6 32.5 27.1 276.2 34.7 20.6 17.5 6.0 9.2
Kiva: Implicit 23.3 22.4 18.9 33.4 27.5 279.5 35.2 20.9 17.6 6.1 9.3
Kiva: Crank-Nicolson 23.3 22.4 18.9 33.4 27.5 279.5 35.2 20.9 17.6 6.1 9.3

IEA BESTEST procedures specify that annual time-series results must be provided for each

of the GC40 cases (a, b, and c). In a plot of annual results (Figure 4.13), it can be difficult to
 97

40

Annual slab heat loss [MWh]

35

30

25

20

15

10

0
GC40a GC40b GC40c GC45b GC45c GC50b GC55b GC55c GC70b GC80b GC80c
TRNSYS MATLAB Kiva: Explicit Kiva: Implicit
FLUENT Kiva: ADE Kiva: ADI Kiva: Crank-Nicolson

Figure 4.12: BESTEST harmonic (unsteady) test case annual slab heat loss (results for GC50b are
divided by 10 to maintain a consistent scale)

notice some of the more subtle differences among the solution results at the time-step level. It is

more informative to compare the annual and daily time-scale characteristics of the solutions. As

the time-series solutions are driven by the sinusoidal variation in exterior temperature (Equation

4.3), the solution for slab heat loss takes a similar form:

¯ + Q̇ Nday − Φyear − 15 Nhour − Φday − 4

Q̇ = Q̇ year · cos · 2 · π + Q̇day · cos ·2·π (4.5)
365 24

where each solution is represented by the following five harmonic paramters:

¯ The average slab heat loss

• Q̇:

• Q̇year : The annual amplitude of the slab heat loss

• Φyear : The annual phase shift of the slab heat loss (relative to the phase of the annual

temperature variation)

• Q̇day : The daily amplitude of the slab heat loss

 98

• Φday : The daily phase shift of the slab heat loss (relative to the phase of the daily temper-

ature variation)

3400

3200
Slab heat loss [W]

3000

2800

2600

2400

2200

2000
0 1000 2000 3000 4000 5000 6000 7000 8000
Hour of year
TRNSYS MATLAB Kiva: Explicit Kiva: Implicit
FLUENT Kiva: ADE Kiva: ADI Kiva: Crank-Nicolson

Figure 4.13: BESTEST GC40a harmonic (unsteady) test case hourly results

The values of each of the five harmonic parameters for the three GC40 cases are shown in

Table 4.9. The annual slab heat loss never exceeds a 5% difference between solutions for a given

case. For each case any deviation from the median parameter values is indicated by the degree

of background shading (darker indicates more deviation). The only outliers occur for the daily

parameters, Q̇day and Φday , in the ADE solutions, which appears be less accurate at shorter time

scales.

Similar to the stability issues mentioned for the steady-state tests, the explicit solution

method required the same restrictions on time-step and cell size (1 minute time-steps and 20

mm minimum cell size) to ensure stability. The ADE solution method required shorter time-steps

for several cases involving infinite convective coefficients (the a and c cases). The stable time-steps

used in verification for the ADE solutions are listed in Table 4.10.
 99
Table 4.9: BESTEST GC40 harmonic parameters for each solution (shading indicates deviation
from the median value for each case)
¯
Q̇ Q̇year Φyear Q̇day Φday
Case/Solution
[W] [W] [days] [W] [hours]
GC40a:
TRNSYS 2,629 447 16.6 10.2 8.3
FLUENT 2,598 435 17.2 8.2 8.1
MATLAB 2,695 465 16.4 13.4 8.2
Kiva: ADE 2,664 453 16.4 13.7 2.3
Kiva: Explicit 2,631 443 16.9 10.6 9.4
Kiva: ADI 2,565 425 17.5 7.6 9.7
Kiva: Implicit 2,660 452 16.4 11.6 8.3
Kiva: Crank-Nicolson 2,660 453 16.4 12.6 8.9
GC40b:
TRNSYS 2,523 412 17.9 6.4 9.3
FLUENT 2,504 405 18.3 5.6 8.9
MATLAB 2,570 424 18.0 8.1 9.3
Kiva: ADE 2,554 416 18.1 14.2 7.7
Kiva: Explicit 2,532 409 18.4 6.8 10.4
Kiva: ADI 2,497 399 18.7 5.2 10.4
Kiva: Implicit 2,554 415 17.9 6.9 9.3
Kiva: Crank-Nicolson 2,554 415 17.9 7.6 10.0
GC40c:
TRNSYS 2,129 323 21.9 2.4 10.1
FLUENT 2,123 319 22.2 2.2 9.8
MATLAB 2,154 329 22.0 3.0 10.1
Kiva: ADE 2,156 325 22.5 1.1 8.8
Kiva: Explicit 2,141 321 22.4 2.4 11.4
Kiva: ADI 2,124 316 22.7 1.9 11.3
Kiva: Implicit 2,153 325 22.0 2.6 10.4
Kiva: Crank-Nicolson 2,154 325 22.0 3.0 11.1

Table 4.10: ADE time-step size for harmonic BESTEST cases

Case Time-step [minutes]

GC40a 30
GC40b 60
GC40c 20
GC45b 60
GC45c 20
GC50b 60
GC55b 60
GC55c 20
GC70b 60
GC80b 60
GC80c 30

Figure 4.14 and Table 4.11 show the total simulation time required for each solution method

and for each case. The following observations can be taken from these results:
 100

• Variation among cases is caused mostly by changes in the domain size.

• ADE simulations are again the fastest, though when the time-step is reduced, there is a

proportional increase in wall time.

• Explicit simulations, due to their increased number of time-steps, have simulation times

near those of the solutions, which require a linear algebra solver (i.e., Implicit and Crank-

Nicolson).

• Given the relatively small change in accuracy between each of the solution methods, the

ADE and ADI methods remain the most promising for typical annual energy analysis

practices.

30000
Simulation wall time [s]

25000

20000

15000

10000

5000

0
GC40a GC40b GC40c GC45b GC45c GC50b GC55b GC55c GC70b GC80b GC80c
Kiva: ADE Kiva: Explicit Kiva: ADI Kiva: Implicit Kiva: Crank-Nicolson

Figure 4.14: BESTEST harmonic (unsteady) test case simulation wall times

4.2 Summary and Conclusions

The results of the BESTEST comparison cases indicate that there are no statistical outliers

with regard to total slab heat loss. Kiva’s capability to utilize several different finite difference
 101
Table 4.11: BESTEST harmonic (unsteady) test case results: Simulation wall time [s]

GC40a GC40b GC40c GC45b GC45c GC50b GC55b GC55c GC70b GC80b GC80c
Kiva: ADE 3,488 1,618 4,599 1,610 4,283 1,963 1,286 3,637 1,630 1,607 3,180
Kiva: Explicit 15,017 12,233 10,528 13,243 10,411 17,063 7,527 8,242 12,376 12,201 10,387
Kiva: ADI 3,832 3,617 3,249 3,547 3,178 4,350 2,442 2,632 3,609 3,595 3,251
Kiva: Implicit 17,991 21,863 16,488 22,152 15,984 26,358 16,875 13,878 22,969 17,593 14,572
Kiva: Crank-Nicolson 17,771 20,344 15,169 20,051 15,184 24,728 15,018 13,314 21,623 17,567 13,712

solution methods makes it the first tool to directly compare multiple methods under the same

computational framework, thus ensuring that differences among the solutions are isolated from other

programming differences. Each method was evaluated regarding numerical accuracy, computation

time, and stability.

In general, a fully implicit solution provides the most accurate results as it is not subject

to some of the stability issues associated with the explicit schemes or the sub-time-step approxi-

mations of the ADI scheme. The Crank-Nicolson scheme provides a better approximation of time

differencing than the fully implicit scheme though it is still subject to unrealistic oscillations.

Although the ADI solution tended to underapproximate the slab heat loss relative to the

other solutions, it required less than one quarter of the computation time of the other implicit

schemes. It is also superior to the ADE scheme in that it has less stringent stability criteria.

Finally, the investigation into initialization methods revealed that each of the traditional

methods (Kusuda, constant temperature, and steady-state) required seven or more years to arrive

at a steady harmonic response to the driving exterior temperature variation. Several variations of

accelerated initializations were tested and were all were found to provide a more accurate approxi-

mation of the first year heat loss than the traditional methods. Of these accelerated initializations,

those with a short (three month) initialization period and relatively long acceleration time-steps

(one week) provided a favorable trade-off between accuracy and the additional required simulation

time.

On the whole, the simulation time required to perform foundation heat transfer calculations

can be significantly reduced, relative to traditional approaches, without a significant loss in accuracy.
 102

For the GC40a test case, a two-year simulation using the ADI solution method with accelerated

initialization required 56.5 minutes in contrast to the projected 18.25 hours required for a seven-

year simulation using the fully implicit method with a traditional initialization method (a nearly

95% reduction in simulation wall time).

 Chapter 5

Two-Dimensional Approximation Methods

Chapter 4 demonstrated how computation time is reduced through the selection of the Alter-

nating Direction Implicit (ADI) solution scheme and the accelerated initialization method, without

significantly affecting the veracity of Kiva’s calculations. Chapter 4 also established discretization

independence for the solutions within the context of the IEA BESTEST procedures.

Further reductions in computation time are possible, but they require a more delicate balance

with computational accuracy. Reducing the problem to two-dimensions is a common approach and

can reduce the number of calculations by several orders of magnitude.

Many efforts have been taken to approximate three-dimensional effects using an equivalent

two-dimensional geometry. Walton [114] noted that improper application of two-dimensional calcu-

lations can result in up to 50% error, relative to three-dimensional calculations. Only a few authors

have compared their two-dimensional approximations to a three-dimensional reference solution.

It is generally accepted that the heat transfer from building foundations is strongly two-

dimensional near the perimeter of the foundation. However, there is no consensus on how the

two-dimensional simulation results should be scaled to represent the heat transfer within the ac-

tual, three-dimensional context. This chapter aims to resolve the disparity between two- and

three-dimensional simulation results by exploring the implications of different two dimensional ap-

proximation methods using the Kiva numerical framework.

 104

5.1 Current Approximation Methods

The sections below describe several established methods of scaling two-dimensional simulation

results. Each approximation method uses the same vertical dimensions as the three-dimensional

definition, but differs in how it defines the lateral dimension of the foundation floor, Δxf , in Figure

5.1.



z

Δxf

Figure 5.1: Typical context for two-dimensional approximations

Many authors refer to Δxf /2 as the “half-width,” which is the dimension used for the foun-

dation in a symmetric domain [39, 24].

5.1.1 Perimeter

Many authors provide results normalized by foundation perimeter. Scaling by the perimeter

is a logical first choice since the magnitude of the foundation heat transfer is stronger near the

perimeter. The total representative three-dimensional heat transfer is calculated as follows:

Q̇3D = Δxf P3D q̇cart,f + Δzw P q̇cart,w , (5.1)

where
 105

• Q̇3D is the three-dimensional heat transfer rate [W],

• P3D is the perimeter [m] of the foundation shape in three-dimensions,

• q̇cart,f is the two-dimensional heat flux [W/m2 ] through the foundation floor calculated in

a Cartesian coordinate system,

• Δzw is the height [m] of the foundation wall, and

• q̇cart,w is the two-dimensional heat flux [W/m2 ] through the foundation wall calculated in

a Cartesian coordinate system.

Beausoleil-Morrison et al. [16] developed a modified perimeter-scaled approximation called

the “Corner Correction Method.” This method introduces corner correction factors that relate the

heat transfer at the corners of the foundation to the heat transfer of the center portions of the

foundation perimeter, where heat transfer is more two-dimensional:

−1
q̇corner q̇center
Fc = (5.2)
Pcorner Pcenter

Corner correction factors were calculated from a regression based on 3,024 numerical simu-

lations of rectangular basements. The result is a table of 19 regression coefficients that correlate

the correction factors to six independent variables related to insulation configuration, soil thermal

properties, and foundation geometry.

Scaling by the perimeter does not strictly define the value of foundation floor width, Δxf ,

needed to create the two-dimensional domain. Both Bazjanac et al. [13] and Deru [39] defined the

foundation width as the narrow dimension of a rectangular foundation. However, instead of using

the relationship defined by Equation 5.1, Bazjanac et al. [13] weighted the floor results by the area

of discrete rectangular annuli corresponding to different locations along the floor width.

5.1.2 Area-to-Perimeter Ratio

Bahnfleth [9] and Bahnfleth and Pedersen [11] simulated a series of three-dimensional square,

rectangular, and L-shaped slabs with a range of foundation areas and perimeter lengths. They
 106

demonstrated that heat transfer does not vary linearly with the perimeter of the foundation, and

indicated that the area of the foundation also plays an important role. They concluded that the

average heat flux through rectangular and L-shaped slabs could be calculated as a function of the

area-to-perimeter ratio A/P :

q̇ = c [A/P ]−d , (5.3)

where

• q̇ is the average slab heat flux [W/m2 ],

• c is a correlation constant [W/m2+d ],

• A is the slab area [m2 ],

• P is the slab perimeter [m], and

• d is a dimensionless correlation constant.

The area-to-perimeter ratio also appears in the study of fluid flow—another diffusion driven

phenomenon. Fluid flowing through channels is often characterized by a single characteristic length,

the hydraulic diameter, defined by the following:

A
Dh = 4 , (5.4)
P

where A is the cross-sectional area of the channel and P is the wetted perimeter.

The hydraulic diameter defines the effective diameter of a circular channel, allowing any

channel to use similar flow calculations. It is logical that that the same relationship applies to heat

diffusion through building foundations.

Extrapolating Equation 5.3 beyond rectangular and L-shaped slabs implies that foundation

heat transfer can be calculated by scaling heat fluxes by foundation area. Therefore, heat trans-

fer from geometrically complicated shapes can be calculated from simpler shapes with the same

characteristic length.
 107

Although Bahnfleth and Pedersen [11] never developed a two-dimensional approximation

method themselves, a number of authors have utilized the A/P characteristic length for scaling

two-dimensional results [95, 24, 25]. The A/P characteristic length leads to two possible shapes

that can be simulated in two dimensions: one for an axisymmetric cylindrical coordinate system

and another for a Cartesian coordinate system.

In axisymmetric cylindrical coordinates, any foundation shape can be represented by a circle

of radius
A
r=2 (5.5)
P

as shown in Figure 5.2.

Figure 5.2: Characteristic length of a circle in a cylindrical coordinate system

With Δxf = 2r, the total representative three-dimensional heat transfer for the circular

method is as follows:

Q̇3D = A3D q̇cyl,f + P3D Δzw q̇cyl,w , (5.6)

where

• A3D is the area [m2 ] of the three-dimensional foundation shape,

• q̇cyl,f is the two-dimensional heat flux [W/m2 ] through the foundation floor calculated in a

cylindrical coordinate system, and

 108

• q̇cyl,w is the two-dimensional heat flux [W/m2 ] through the foundation wall calculated in a

cylindrical coordinate system.

In Cartesian coordinates, any foundation shape can be represented by an infinite rectangle

of length, l, and width, w. The area-to-perimeter ratio can be expressed as follows:

A lw w
= = , (5.7)
P 2l + 2w 2 + 2w
l

and as l → ∞,
A w
→ , (5.8)
P 2

which means any foundation shape can be represented by an infinite rectangle of width

A
w=2 . (5.9)
P

as shown in Figure 5.3.

Figure 5.3: Characteristic length of a infinite rectangle in a Cartesian coordinate system

With Δxf = w, the total representative three-dimensional heat transfer for the infinite

rectangle method is as follows:

Q̇3D = A3D q̇cart,f + P3D Δzw q̇cart,w , (5.10)

where
 109

• q̇cart,f is the two-dimensional heat flux [W/m2 ] through the foundation floor calculated in

a Cartesian coordinate system, and

• q̇cart,w is the two-dimensional heat flux [W/m2 ] through the foundation wall calculated in

a Cartesian coordinate system.

Chuangchid [24] and Chuangchid and Krarti [25] verified that both circular and infinite

rectangular shapes in two-dimensional coordinates provide reasonable approximations for insulated

circular, rectangular, and square slabs in three dimensions.

5.1.3 Rounded Rectangle

Walton [114] developed an approximation method that produced two-dimensional results

within 2% of a three-dimensional reference. This method, called the “rounded rectangle,” or RR

method, transforms any three-dimensional foundation geometry into a rounded rectangle foundation

of the same perimeter and area. The rounded rectangle allows entire problems to be solved using

the superposition of heat transfer calculated using a Cartesian coordinate system (for the straight

sides) and a cylindrical coordinate system (for the rounded ends). From the area, A, and perimeter,

P , the dimensions of the rounded rectangle can be calculated as follows:

 
a = P − P 2 − 4πA /π, (5.11)

and

b = (P − πa) /2. (5.12)

as shown in Figure 5.4.

With Δxf = a for both coordinate systems, the total representative three-dimensional heat

transfer is as follows:

Q̇3D = abq̇cart,f + 2bΔzw q̇cart,w + πa2 q̇cyl,f + 2πaΔzw q̇cyl,w . (5.13)

 110

a
b

Figure 5.4: Dimensions of a rounded rectangle with equivalent area and perimeter

Walton verified his rounded rectangle approximations for uninsulated rectangular basements

and slabs under a steady-state solution, while results for X-shaped and H-shaped basements showed

up to a 6.3% difference in some cases.

The rounded rectangle approximation method uses two characteristic lengths, a and b, which

allow for an additional degree of freedom in determining the representative heat transfer. In this

case, the additional degree of freedom requires two independent two-dimensional simulations, one for

each coordinate system. However, because the simulations are independent, they can be calculated

on parallel processors without significantly more computation time than a single simulation.

5.2 Heat Transfer from Concave Foundation Shapes

Bahnfleth and Pedersen [11] and Walton [114] explain that two-dimensional results can better

approximate three-dimensional results when scaled by both area and perimeter of the shape, not just

perimeter. Bahnfleth and Pedersen [11] also introduced the characteristic length, A/P , which can

be used to generate representative heat fluxes in two-dimensional geometry. However, no established

approximation method developed thus far accounts for concave foundation shapes. Even the corner-

correction method developed by Beausoleil-Morrison et al. [16] only provides adjustments for convex

rectangular corners.
 111

Shape cavities are defined relative to the shape’s convex hull. The convex hull is the smallest

convex polygon that envelops the original polygon. The spatial difference between the original

shape (Figure 5.5(a)) and the convex hull (Figure 5.5(b)) defines the set of cavities (Figure 5.5(c)).

(a) Original shapes (b) Convex hulls (c) Cavities

Figure 5.5: Determination of cavities for various shapes

While heat transferred from convex edges is largely unaffected by other edges, the ground

within cavities is warmer because of the additive heat flow from neighboring concave edges. This

results in reduced heat transfer from the foundation. The influence of neighboring concave edges

on heat transfer can be observed in Figure 5.6 for three-dimensional simulation results. Heat flux

is higher through the ground near the concave slab corner, relative to heat fluxes further from the

corner. Consequently, heat flux is lower through the slab in proximity to the warmer ground.

The degree of heat transfer reduction within cavities depends on several factors beyond the

shape itself. These influential factors are illustrated by isolating the concave effects. The following

sections describe two separate types of interactive effects that occur between concave foundation

edges.

5.2.1 Opposing Parallel Edges

The first example of interaction between concave edges occurs when two parallel edges come

within close proximity of each other. This effect can be isolated by simulating heat transfer from
 112

Figure 5.6: Plan view cross-section at grade height of heat flux near a concave slab corner (exterior
grade in the upper left)

two infinitely long rectangles separated by a distance, p (Figure 5.7). This arrangement can be

simulated in two-dimensions with two planes of symmetry. This exercise demonstrates how the

heat flux is reduced as the distance p decreases.

The extent of the impact of opposing parallel edges will depend on the actual perimeter of

opposing edges. In the example of infinite rectangles, heat transfer reduces to zero as the distance

between the edges closes. Actual foundation shapes are enclosed, which means that the effect

applies to less than half of the perimeter, at most (in the case of long U-shaped foundations).

Parallel concave edges are simulated for a combination of area-to-perimeter ratios, soil con-

ductivity, and insulation levels. All simulations are steady-state with an indoor-outdoor tempera-

ture difference of 20◦ C. Unless stated otherwise, the foundations are slabs-on-grade with exterior

horizontal insulation extending roughly one meter from the perimeter. The results presented in
 113

A/P p A/P

Figure 5.7: Plan view of parallel concave edges

Figures 5.8–5.10 demonstrate the sensitivity of opposing parallel edges for a single example foun-

dation.

100%
Percent of convex heat loss

80%

60%

40%

A=P [m]
1.5
20% 4.6
7.6
10.7
13.7
0%
1 2 3 4 5
Distance between parallel boundaries [m]

Figure 5.8: Reduction in heat loss between two parallel concave edges over a range of area-to-
perimeter ratios (ksoil = 1.0 W/m·K, R-10 insulation)
 114

As the area-to-perimeter ratio increases, the heat loss reduction relative to a convex (i.e.,

with no opposing edge) infinite rectangle decreases. Therefore the interaction effect has a higher

impact for larger foundations. Smaller foundations see a less gradual reduction and are relatively

unaffected until the two edges are very close, at which point heat transfer drops steeply.

100%
Percent of convex heat loss

80%

60%

40%
ksoil [W/m ¢K]
0.25
0.50
20% 0.75
1.00
1.25
1.50
0%
1 2 3 4 5
Distance between parallel boundaries [m]

Figure 5.9: Reduction in heat loss between two parallel concave edges over a range of soil conduc-
tivity (A/P = 4.6 m, R-10 insulation)

As the soil conductivity increases, the interaction between the two edges becomes stronger.

Higher soil conductivity allows heat to flow further from the foundation, causing a greater impact

on the opposing edge.

With increased levels of insulation, the path of heat flow is driven further from foundation

perimeter, where there is a higher potential for interaction with the opposing edge. As a result,

the heat transfer is reduced more at higher levels of insulation.

The shape of the curves in Figures 5.8–5.10 represents reduced heat loss as the two edges

grow closer together. The curves show a noticeable change where the distance between the two

edges is less than the width of the insulation. At long distances there is little effect, but within
 115
100%

Percent of convex heat loss

80%

60%

40%

R-value
R-0
20% R-5
R-10
R-15
R-20
0%
1 2 3 4 5
Distance between parallel boundaries [m]

Figure 5.10: Reduction in heat loss between two parallel concave edges over a range of insulation
R-values [hr·ft2 ·R/Btu] (ksoil = 1.0 W/m·K, A/P = 4.6 m)

the last several meters the heat transfer reduction drops drastically toward zero. The heat transfer

reduction is related to the magnitude of heat flux from the foundation, which is strong near the

perimeter and drops quickly to zero beyond a few meters. Though the distance where the magnitude

drops to zero is difficult to quantify precisely, it is similar to the concept of a boundary layer in

fluid mechanics. The boundary of heat flow near the perimeter of the foundation progressively

transitions to a constant value sufficiently far away from the foundation.

The profile of this boundary layer can be calculated for any vertical cross-section extending

from the perimeter of the foundation. The boundary layer is defined by the vertical heat flux along

the grade plane. Figure 5.11 illustrates the boundary layer profile for a slab with exterior horizontal

insulation.

The integral of the boundary layer profile from the foundation perimeter yields the function,

δ (x). This function is normalized to represent the fraction of total heat transfer within any distance,
 116

z
. x
qz(x)

Figure 5.11: Illustration of a heat flux boundary layer profile for a slab with exterior horizontal
insulation

x, of the perimeter. In the limit as x approaches infinity, the value of δ (x) approaches unity:

lim δ (x) = 1. (5.14)

x→∞

Figure 5.12 shows examples of the normalized boundary layer profile, q̇z (x), and the inte-

grated boundary layer profile, δ (x).

The boundary layer interaction of the two opposing edges causes reduced heat transfer pro-

portional to the value of the integrated boundary layer function at a given distance. For any shape

cavity, the degree of interaction is related to the integrated boundary layer profile. The next section

demonstrates this relationship for concave corners.

5.2.2 Concave Corners

Depending on the angle of a corner, the boundary layer will be more or less constrained.

Sharper angles will cause a higher level of interaction within a cavity. As Kiva is only capable of
 117
100%

80%
q_ z (x)

60%
Percent of total boundary heat loss
40%

20%

0%

100%
Z x
80% ±(x) = q_ z (x)dx
0
60%

40%

20%

0%
0 1 2 3 4 5
Distance from perimeter, x [m]

Figure 5.12: Example boundary layer profile for a slab with exterior horizontal insulation (ksoil =
1.0 W/m·K, A/P = 4.6 m, R-10 insulation)

simulating rectilinear foundation shapes, the effects in concave corners can only be demonstrated

for right angles.

The effect of corners can be isolated by defining shapes with the same area and perimeter,

but with a different number of corners. The range of shapes with corners is somewhat limited

as complex shapes with many corners create several instances of near-field mesh refinements that

can drastically increase the computational requirements. For this reason, only four shapes are

selected to demonstrate the effect of concave corners on foundation heat transfer. These shapes are

illustrated in Figures 5.13.

Heat transfer from slabs of each of these shapes was simulated under the same range of area-

to-perimeter ratios, soil conductivities, and insulation levels that were simulated for the parallel
 118
13d
d

3d
d
6d
d

d
4d A = 13d2
P = 28d

Figure 5.13: Shapes with equal area and perimeter but different number of concave corners

concave edge example. The proportion of the shapes remains constant even with different area-

to-perimeter ratios. For a given area-to-perimeter ratio, A/P , the shape thickness, d, is defined

as
28
d= [A/P ] . (5.15)
13

Figures 5.14–5.16 show the sensitivity of heat loss reduction to changes in area-to-perimeter

ratio, soil conductivity, and insulation R-value. Results are presented for each concave shape

relative to the convex rectangle.

The jump from zero corners (the rectangle shape at 100%) to a single corner (the L shape) is

much larger than the difference between one corner and two corners (the T shape). By four corners

(the + shape) the impact of additional corners continues to drop. Although the fraction of the

perimeter affected by the corners increases proportionally to the number of corners, the heat flux

does not also decrease proportionally.

Unlike the opposing parallel edges, the reduction effect for corners increases as the area-

to-perimeter ratio decreases. For lower area-to-perimeter ratios, the interaction effect for corners
 119
100%

Percent of convex heat loss

98%

96%

94%

92%

90%
0 2 4 6 8 10 12 14
A=P [m]

Figure 5.14: Reduction in heat loss due to corners over a range of area-to-perimeter ratios (ksoil =
1.0 W/m·K, R-10 insulation)

100%
Percent of convex heat loss

98%

96%

94%

92%

90%
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
2
ksoil [W/m ¢K]

Figure 5.15: Reduction in heat loss due to corners over a range of soil conductivity (A/P = 4.6 m,
R-10 insulation)
 120
100%

Percent of convex heat loss

98%

96%

94%

92%

90%
0 5 10 15 20
2
Insulation R-value [hr ¢ft ¢R/Btu]

Figure 5.16: Reduction in heat loss due to corners over a range of insulation R-values (ksoil = 1.0
W/m·K, A/P = 4.6 m)

between shorter edges impacts a larger portion of the perimeter causing a larger heat transfer

reduction. For larger area-to-perimeter ratios, the corners represent a smaller portion of the total

perimeter, and therefore the heat transfer reduction is less impactful. In the limit as the area-to-

perimeter ratio approaches infinity, the impacts of corners become negligible.

The impact of increasing soil conductivity and insulation R-value for corners is similar to the

impact for parallel edges as shown in Figures 5.15 and 5.16, respectively.

The boundary layer profile is amplified near the corner due to interactions between the two

intersecting concave edges. Figure 5.17 illustrates example cross-sections for both corner bisector

and unaffected edge profiles.

The boundary layer heat flux profile along a corner bisector, q̇z,b (x), can be estimated from

an unaffected edge profile, q̇z (x) as

q̇z,b (x) = q̇z (x sin (θ/2)) [1 + cos (θ/2)] (5.16)

where
 121

  

  

Figure 5.17: Illustration of corner bisector and unaffected edge boundary profile cross-sections

• x is the distance from the perimeter [m],

• θ is the angle of the corner,

• the sin (θ/2) accounts for the change in dimensions of the foundation elements along the

bisector,

• the 1 + cos (θ/2) accounts for increased heat flux in the corner due to interacting boundary

layer profiles.

Kiva can only simulate right angles, and cannot confirm this relationship for other angles. Fig-

ure 5.18 validates the estimated bisector boundary layer profile and the calculated two-dimensional,

unaffected boundary layer profile against three-dimensional simulation results. The discrete points

in Figure 5.18 represent the three-dimensional heat flux calculated along each cross-section. The

curves represent the generated unaffected boundary layer profile, and the calculated boundary layer

profile using the relationship in Equation 5.16. There is near-perfect agreement between the three

dimensional profiles and those estimated based on two dimensional simulations.

 122

Percent of boundary layer heat loss

140% 3D unaffected
3D corner
120% 2D unaffected
2D corner approximation
100%

80%

60%

40%

20%

0%
0 5 10 15 20
Distance from perimeter [m]

Figure 5.18: Comparison of straight boundary profile and corner bisector

Integrating the bisector boundary layer profile yields

! x
1 + cos (θ/2)
δb (x) = q̇z,b (x) dx = δ (x sin (θ/2)) . (5.17)
0 sin (θ/2)

The heat transfer reduction near concave corners is related to differences between the bisector

integrated boundary layer profile and the integrated boundary layer profile of an unaffected edge.

5.2.3 Compounding Cavity Effects

Many shapes exhibit both opposing parallel edges and corners. The interactive effects com-

bine as illustrated in Figure 5.19 for an H-shaped slab. The cavities of the H-shape are adjusted

by varying the dimensions of the web (the connector) and flanges (the parallel sides), while main-

taining constant area and perimeter. Figure 5.19 demonstrates the progression of heat transfer

reduction at discrete intervals between convex extremes where the shape has no flanges or no web.

The flange and web thickness, d, is set to 10 meters, and the area-to-perimeter ratio is

13d2
A/P = = 4.64 m. (5.18)
28d
 123
R-20, k = 1.50 W/m ¢K R-20, k = 0.25 W/m ¢K
Unins., k = 1.50 W/m ¢K Unins., k = 0.25 W/m ¢K
100%

95%

Percent of convex heat loss 90%

85%

80%

75%

70%

65%

60%

55%

Figure 5.19: Heat loss reduction from H-shaped slabs

The four curves in Figure 5.19 represent bounding values of soil conductivity and insulation level

for exterior vertical insulation.

On the left-hand side of Figure 5.19, there is relatively little heat transfer reduction where

the flanges are not long enough to form an unaffected boundary layer. For the cases with narrow

boundary layers (e.g., uninsulated with low soil conductivity), the relative heat transfer settles at

a constant value for shorter flanges, while shapes with wide boundary layers require longer flanges

before an unaffected boundary layer is formed. In the case of a high insulation value and high soil

conductivity, the unaffected boundary layer is not established before the interaction between the

opposing parallel edges begins to further reduce heat transfer.

The established approximation methods described in Section 5.1 fail to account for heat

transfer reduction in cavities and thus predict the same heat loss for each shape on a given curve in

Figure 5.19. In the extreme case with no web, there is a potential 40% error in the approximations
 124

relative to the three-dimensional solution. It is not coincidence that the opposing parallel edges in

this extreme comprise 40% of the total foundation perimeter.

5.3 Boundary Layer Adjustment Method

The integrated boundary layer profile can be used to estimate heat transfer reduction for a

foundation shape with cavities. The profile can be calculated easily within the existing computa-

tional framework, and it contains all of the information about the domain which influences heat

transfer within shape cavities. There is no need to develop regressions for all of the variables which

impact the heat transfer reduction within cavities because the profile is specific to the combination

of foundation design characteristics and soil thermal properties defined for a given model.

Bahnfleth and Pedersen [11] demonstrated that as the area-to-perimeter ratio of a foundation

increases, the average heat flux from the foundation decreases. The concept behind the boundary

layer adjustment method is to adjust the dimensions of parallel edges and concave corners, based

on the integrated boundary layer profile, to produce a new adjusted area-to-perimeter ratio. The

adjusted area-to-perimeter ratio defines the dimensions for a representative two-dimensional domain

and is calculated as follows:

A A
= , (5.19)
P adj 
Nv
P− Pv,i
i

where

• A is the original area [m2 ] of the foundation shape,

• P is the original perimeter [m] of the foundation shape,

• Nv is the number of concave features (either parallel edges or corners),

• Pv,i is the perimeter adjustment [m] for concave feature i.

The value of Pv for each concave feature of the foundation shape depends on the integrated

boundary layer profile function. If thermal properties are linear, then the boundary layer profile
 125

is independent of temperature and can be calculated using a single steady-state, two-dimensional

simulation with an arbitrary non-zero temperature difference. The values of vertical heat flux are

integrated along the grade plane to generate a mapping of the percent total heat flux to the distance

from the perimeter at each discrete cell. The map is as coarse as the spatial discretization in the

lateral direction. The value of the integrated boundary layer profile, F , at any distance, x, is eval-

uated using a linear interpolation between mapped values. The inverse of the integrated boundary

layer profile function, δ −1 (F ), also uses the same map to provide the distance, x, corresponding to

a given fraction of total heat loss, F .

5.3.1 Opposing Parallel Edges

For opposing parallel edges, the edge perimeters are adjusted as follows:

Pv = 2a [1 − δ (p)] . (5.20)

where p is the distance between the edges, and a is the common length of the parallel edges as

illustrated in Figure 5.20.

p
a a

Figure 5.20: Variables used to adjust perimeter of opposing parallel edges

 126

5.3.2 Concave Corners

For concave corners, the perimeter is adjusted based on the integrated boundary layer profile

of the bisector described in Equation 5.17, the angle of the corner, θ, and the lengths, A and B, of

the edges intersecting at the corner.

A chamfer at a distance, xc , from the corner along the bisector, as shown in Figure 5.21,

defines the adjustment to the perimeter.

θ b

Figure 5.21: Variables used to adjust perimeter of concave corners

The value xc is selected such that the total heat flux beyond the chamfer is the same as the

total heat flux from an unaffected edge. The total relative heat flux along the bisector is

1 + cos (θ/2)
lim δb (x) = . (5.21)
x→∞ sin (θ/2)

Therefore, the bisector’s integrated boundary profile at xc is

1 + cos (θ/2)
δb (xc ) = − 1. (5.22)
sin (θ/2)

which, when substituting in Equation 5.17, gives

1 + cos (θ/2) 1 + cos (θ/2)

δ (xc sin (θ/2)) = − 1. (5.23)
sin (θ/2) sin (θ/2)

Solving Equation 5.23 for xc gives:

δ −1 (1 − tan (θ/4))
xc = . (5.24)
sin (θ/2)
 127

where δ −1 (x) is the inverse of the integrated boundary profile function.

The distance, d, from the angle to the points where the chamfer intersects the edges is

calculated as
xc
d= . (5.25)
cos (θ/2)
The final dimensions of the edges used depend on the lengths of the edges relative to d:

a = min (A, d) , (5.26)

and

b = min (B, d) . (5.27)

The perimeter adjustment, Pv , for each corner is calculated as follows:


Pv = a + b − a2 + b2 − 2ab cos(θ). (5.28)

5.4 Description of Approximation Method Tests

The verification testing in Chapter 4 was based solely on the IEA BESTEST test case

specifications for slab-on-grade foundations. These test cases are constrained to idealized conditions

and do not cover the breadth of modeling capability in Kiva. The two-dimensional approximation

methods discussed in this chapter are tested in a more realistic context under a wider range of

parameters.

All simulations performed to test approximation methods use the set of conditions described

in Table 5.1. All three-dimensional simulations are modeled using planes of symmetry where they

exist to reduce domain size.

Several other parameters are varied between tests to highlight potential sensitivities in results.

The ranges of these parameters are described in the following sections.

5.4.1 Climates

Approximation methods are tested using realistic weather data representing a range of cli-

mates. The selection of weather locations is based on the range of outdoor dry-bulb temperatures
 128
Table 5.1: Approximation method testing conditions

Description Value
Foundation wall thickness 0.3 m
Slab thickness 0.2 m
Interior temperature 22.2 ◦ C
Soil conductivity 0.864 W/m·K
Soil density 1,510 kg/m3
Soil specific heat 1,260 J/kg·K
Surface absorptivity 0.8
Deep-ground boundary depth 40 m
Deep-ground boundary condition Zero-flux
Far-field boundary width 40 m
Minimum cell dimension 0.006 m
Cell growth coefficient 1.20
Time-step duration 60 minutes
Numerical method ADI
Initialization acceleration time-step duration 1 week
Initialization acceleration period duration 12 weeks
Initialization warm-up duration 1 year

in Typical Meteorological Year (TMY3) data [117]. Figure 5.22 illustrates the ranges of dry-bulb

temperatures selected for the representative hot (Phoenix, Arizona), mixed (Albuquerque, NM),

and cold (Anchorage, Alaska) climates.

5.4.2 Foundation Designs

The approximation method test covers a total of 25 different foundation designs. These

designs represent the three most common foundation types. Insulation configurations are shown

in Figures 5.23, 5.24, and 5.25 for slabs, crawlspaces, and basements, respectively. For each of the

insulated configurations, there are two designs at different insulation levels: R-5 (0.88 m2 ·K/W)

and R-20 (3.52 m2 ·K/W).

All basements are modeled with an 8-foot (2.4 m) foundation depth, crawlspaces are modeled

with a 4-foot (1.2 m) foundation depth, and slabs are modeled 8 inches (0.2 m) above grade. The

foundation wall footer is 5-feet-8-inches (1.7 m) deep for slabs and crawlspaces, and 9-feet-8-inches

(2.9 m) deep for basements.

 129

40

30
Dry-bulb temperature [ ± C]

20

10

−10
Hot Mixed Cold
−20
Time of year

Figure 5.22: Annual outdoor dry-bulb temperatures from weather data files for selected climates.
Hot = Phoenix, Arizona; Mixed = Albuquerque, New Mexico; Cold = Anchorage, Alaska

5.4.3 Foundation Shapes

The foundation shapes selected for testing two-dimensional approximation methods represent

a range of characteristic length, compactness, and concavity.

The characteristic length, defined by the area-to-perimeter ratio, A/P , describes the relative

size of the foundation. The range of characteristic lengths used in the tests are as follows:

• A/P = 1.5 m (5 feet)

• A/P = 4.6 m (15 feet)

• A/P = 13.7 m (45 feet)

 130

(a) Uninsulated (b) Exterior horizontal (c) Exterior vertical

(d) Interior vertical (e) Interior horizontal (f) Interior whole-slab

Figure 5.23: Slab insulation configurations for approximation method tests

Compactness is a nondimensional characteristic that ranges between zero and one, defined

by the following equation:

A
C = 4π . (5.29)
P2

Table 5.2 demonstrates the compactness of several shapes. Each of the shapes shown in Table

5.2 can be simulated in Kiva with the exception of the equilateral triangle, because it is neither

rectilinear nor described by a two-dimensional coordinate system. The equilateral cross and square

shapes are included in the approximation tests. Two variations of the H-shape are also included

to represent different distances between the opposing parallel edges: a narrow-H with 3 meters (10

feet) between flanges, and a wide-H with 122 meters (400 feet) between flanges.

The final set of shapes used in the approximation method tests includes four shapes simu-

lated in three-dimensional Cartesian coordinates (square, cross, narrow-H, and wide-H), a circle in

two-dimensional axisymmetric cylindrical coordinates, and a infinite rectangle in two-dimensional

Cartesian coordinates.
 131

(a) Uninsulated (b) Exterior vertical

(c) Interior vertical (d) Interior horizontal and

vertical

Figure 5.24: Crawlspace insulation configurations for approximation method tests

5.5 Initial Shape Analysis

The shapes simulated in three-dimensional Cartesian coordinates are shown in Figure 5.26

for each value of the characteristic length, A/P .

Initial simulation results for the mixed climate revealed notable differences in foundation heat

flux among shapes with the same characteristic length. Figure 5.27 shows deviation from the mean

heat flux for uninsulated foundation walls and floors, across all shapes.

Figure 5.27 illustrates a number of important shape-related effects on heat transfer:

• The absolute difference in heat transfer is calculated as:

Q̇ = Aq̇f + P Δzw q̇w (5.30)

where q̇f and q̇w are the heat fluxes through the floor and wall, respectively. Although the

crawlspace and foundation heat flux deviates more through floors than through walls, it
 132

(a) Uninsulated (b) Exterior half-wall

(c) Exterior full-wall (d) Interior full-wall

Figure 5.25: Basement insulation configurations for approximation method tests

is also a lower magnitude. Depending on the dimensions of the foundation, this deviation

may have a small impact on the overall heat transfer.

• Heat flux is less sensitive to foundation shape for shallower foundations. The area-to-

perimeter ratio, which Bahnfleth and Pedersen [11] demonstrated characterizes heat trans-

fer through slab foundations, does not extend as effectively to crawlspace and basement

foundations.

• For crawlspaces and basements, there are two independent trends worth noting:

◦ Shapes with higher compactness have proportionally higher heat flux than the infinite

rectangle (where compactness is zero).

◦ As explained for Figure 5.19, shapes with concavities have lower heat fluxes than the

convex shapes due to higher levels of interaction within the concavities.

 133
Table 5.2: Compactness of various shapes

Description Shape Compactness

Circle 1

Square π
4 ≈ 0.785

Equilateral triangle π
√
3 3
≈ 0.605

5π
Equilateral cross 36 ≈ 0.436

7π
Equal-spaced H 64 ≈ 0.344

Infinite
0
rectangle

• The magnitude of deviation decreases proportionally as the characteristic length increases.

When comparing different insulation designs, engineers are often more concerned with rel-

ative differences among designs rather than absolute heat transfer from foundations. Figure 5.28

illustrates the reduction in heat flux for different insulation designs relative to the uninsulated

foundations portrayed in Figure 5.27. Floor heat flux reductions are not shown for crawlspaces and

basements because their magnitudes are substantially lower than heat flux through walls. Similar

heat flux reductions are demonstrated among all shapes, with a slight disparity for shapes with

higher compactness.
 134

A/P =1.5 m Square Cross Narrow-H Wide-H

A = 37 m2 , C = 0.79 A = 67 m2 , C = 0.44 A = 81 m2 , C = 0.36 A = 1200 m2 , C = 0.03

A/P =4.6 m

A = 330 m2 , C = 0.79 A = 600 m2 , C = 0.44 A = 590 m2 , C = 0.45 A = 3800 m2 , C = 0.07

A/P =13.7 m

A = 3000 m2 , C = 0.79 A = 5400 m2 , C = 0.44 A = 4900 m2 , C = 0.48 A = 14000 m2 , C = 0.17

Figure 5.26: Shapes simulated in three dimensions for approximation method testing. Each plot is
200 m × 200 m. A = Area, C = Compactness.

5.6 Approximation Method Accuracy

The four two-dimensional approximation methods tested are listed in Table 5.3.

Table 5.3: List of two-dimensional approximation methods tested

Characteristic Coordinate
Method
Length(s) System(s)
Circle A/P Cylindrical
Infinite rectangle A/P Cartesian
Cartesian and
Rounded rectangle a, b
Cylindrical
Boundary layer adjustment [A/P ]adj Cartesian
 135
Slab Crawlspace Basement
Wall Floor
20%

A=P = 1.5 m
10%

0%

-10%
Deviation from mean heat flux

-20%

20%

A=P = 4.6 m
10%

0%

-10%

-20%

20%

A=P = 13.7 m
10%

0%

-10%

-20%
Circle Semi-Infinite Rectangle Cross
Square Wide-H Narrow-H

Figure 5.27: Comparison of heat flux from a range of uninsulated foundation shapes

The approximation methods are compared to three-dimensional simulation results for a com-

bination of:

• 3 climates,

• 25 foundation designs,

• 3 characteristic lengths, and

• 4 foundation shapes.
 136
Square Narrow-H Circle
Cross Wide-H Semi-Infinite Rectangle

Slab Floor Crawlspace Wall Basement Wall

Reduction in heat flux relative to uninsulated foundations

100%

A=P = 1.5 m
80%

60%

40%

20%

0%
100%

A=P = 4.6 m
80%

60%

40%

20%

0%
100%

A=P = 13.7 m
80%

60%

40%

20%

0%

R-20 int. vert. & hor.

R-5 int. vert. & hor.
R-20 ext. hor.

R-20 int. hor.

R-5 ext. hor.

R-20 half-wall ext.

R-5 int. hor.

R-20 full-wall ext.

R-5 half-wall ext.

R-20 full-wall int.

R-5 full-wall ext.

R-5 full-wall int.

R-20 ext. vert.

R-20 int. whole-slab

R-20 ext. vert.

R-20 int. vert.

R-20 int. vert.

R-5 ext. vert.

R-5 int. whole-slab

R-5 ext. vert.

R-5 int. vert.

R-5 int. vert.

Figure 5.28: Heat flux reduction relative to uninsulated foundations in mixed climate for a range
of foundation shapes

The results presented in Figures 5.29–5.31 represent a total of 4,500 simulations. Each plot

compares approximated two-dimensional annual total heat transfer to three-dimensional annual

total heat transfer. The results demonstrate generally good agreement among each of the approxi-

mation methods with the exception of the circle approximation, which tended to over-predict three

dimensional heat transfer. Mean absolute deviation (MAD), and mean bias deviation (MBD) are

shown for each approximation method. Although the differences between the infinite rectangle,

rounded rectangle, and boundary layer adjustment methods are not clearly visible in the plots,
 137

these metrics show that the former two methods have MAD values of 4–5%, while the boundary

layer adjustment method has MAD values of 2.5–3%.

The boundary layer adjustment method provides a more accurate representation of three-

dimensional heat transfer than previously established approximation methods. This is achieved

with no observable increase in simulation time.

160 160
Circle Infinite Rectangle
140 140
Three-dimensional annual total heat transfer [MWh]

120 120
100 100
80 80
60 60
40 40
20 MAD = 8.9% 20 MAD = 4.8%
MBD = 8.8% MBD = 3.1%
0 0
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160

160 160
Rounded Rectangle Boundary Layer
140 140 Adjustment
120 120
100 100
80 80
60 60
40 40
20 MAD = 4.6% 20 MAD = 2.5%
MBD = 4.5% MBD = -0.8%
0 0
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160

Approximated, two-dimensional annual total heat transfer [MWh]

Figure 5.29: Comparison of annual total heat transfer between three-dimensions and two-
dimensional approximations for mixed climate simulations

More detailed results are presented for mixed, cold, and hot climates in Appendices B, C,

and D, respectively. In these figures, the deviations of each approximation method are compared

for each three-dimensional simulation. Observations from these figures that are not apparent in

Figures 5.29–5.31 include:

 138
600 600
Circle Infinite Rectangle
500 500

Three-dimensional annual total heat transfer [MWh]

400 400

300 300

200 200

100 MAD = 9.7% 100 MAD = 5.0%

MBD = 9.6% MBD = 3.4%
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600

600 600
Rounded Rectangle Boundary Layer
500 500 Adjustment

400 400

300 300

200 200

100 MAD = 4.7% 100 MAD = 2.9%

MBD = 4.6% MBD = -0.5%
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600

Approximated, two-dimensional annual total heat transfer [MWh]

Figure 5.30: Comparison of annual total heat transfer between three-dimensions and two-
dimensional approximations for cold climate simulations

• Methods other than the boundary layer adjustment method tend to over-predict concave

shape heat transfer.

• The approximation methods perform better for shapes with larger area-to-perimeter ratios,

especially for deeper foundations.

While annual total heat transfer is often the most important metric for energy design of

building foundations, variations in heat transfer throughout the year also impact peak heating and

cooling loads in a building used to calculated the appropriate capacity of HVAC equipment. Figure

5.32 demonstrates the time varying accuracy of the boundary layer adjustment method for a cross-
 139
0 0
Circle Infinite Rectangle
−20 −20

Three-dimensional annual total heat transfer [MWh]

−40 −40

−60 −60

−80 −80

−100 −100

−120 −120

−140 MAD = 8.1% −140 MAD = 4.4%

MBD = 8.0% MBD = 2.9%
−160 −160
−160 −140 −120 −100 −80 −60 −40 −20 0 −160 −140 −120 −100 −80 −60 −40 −20 0

0 0
Rounded Rectangle Boundary Layer
−20 −20
Adjustment
−40 −40

−60 −60

−80 −80

−100 −100

−120 −120

−140 MAD = 4.0% −140 MAD = 2.7%

MBD = 3.9% MBD = -0.6%
−160 −160
−160 −140 −120 −100 −80 −60 −40 −20 0 −160 −140 −120 −100 −80 −60 −40 −20 0

Approximated, two-dimensional annual total heat transfer [MWh]

Figure 5.31: Comparison of annual total heat transfer between three-dimensions and two-
dimensional approximations for hot climate simulations

shaped basement with exterior, half-wall R-20 insulation. The three-dimensional results and the

two-dimensional approximation are virtually indistinguishable from each other.

Appendix E contains figures comparing annual minimum and maximum heat transfer rates

for all the tested combinations. The trends of agreement are similar to the comparisons of annual

total heat transfer. The boundary layer adjustment method provides the best approximation of

minimum and maximum heat transfer rates.

 140
7
3D 2D

Foundation heat transfer [kW]

6

−1
l t r r r r
u ary ruaryarch Apri May June July gus mbetobe mbe mbe
n
Ja Feb M Au p te O c N o v e D e c e
Se
Time of year
Figure 5.32: Example comparison of hourly three-dimensional heat transfer and approximated
two-dimensional heat transfer

5.7 Approximation Method Simulation Time

Table 5.4 presents the average simulation wall time for each of the two-dimensional approx-

imation methods. The methods that use only a Cartesian coordinate system require less time

because their domains are typically smaller.

Table 5.4: Average simulation wall time for two-dimensional approximation methods

Average
Method
Simulation Time [s]
Circle 40.9
Infinite rectangle 37.4
Rounded rectangle 40.2
Boundary layer adjustment 34.8

Within a two-dimensional context, the largest driver of simulation time is the foundation

design. Figure 5.33 demonstrates the variability of simulation time across foundation designs for
 141

the boundary layer adjustment method. Simulations with more foundation elements require more

cells to define the near-field discretization region and therefore take longer to simulate.

A=P [m]
1.5 4.6 4.6
13.7 13.7 13.7

Slab Crawlspace Basement

40
Simulation wall time [s]

30

20

10

R-20 int. vert. & hor.

R-5 int. vert. & hor.
R-20 ext. hor.

R-20 int. hor.

R-5 ext. hor.

R-20 half-wall ext.

Uninsulated

Uninsulated

Uninsulated

R-20 full-wall ext.

R-5 int. hor.

R-5 half-wall ext.

R-20 full-wall int.

R-5 full-wall ext.

R-5 full-wall int.

R-20 ext. vert.

R-20 int. whole-slab

R-20 ext. vert.

R-20 int. vert.

R-20 int. vert.

R-5 int. whole-slab
R-5 ext. vert.

R-5 ext. vert.

R-5 int. vert.

R-5 int. vert.

Figure 5.33: Two-dimensional approximation simulation wall times

The average simulation wall time of the boundary layer adjustment method is approximately

35 seconds. This is considerably faster than three-dimensional alternatives, but it still represents

a considerably large portion of the total simulation time for a whole-building energy simulation.

Further reductions in simulation time can be achieved by relaxing the refinement of the spatial

discretization. All simulations performed in approximation method tests used a minimum cell

dimension, Δxmin , of 6 millimeters, and a cell growth factor, fg , of 1.2. Figure 5.34 illustrates

the reduction in simulation wall time and corresponding impact on approximation accuracy for
 142

the foundation design with the longest simulation time, a cross-shaped basement with exterior,

half-wall, R-20 insulation and an area-to-perimeter ratio of 13.7 meters.

Percent deviation in annual heat transfer 2.0%

1.5%

1.0%

0.5%

0.0%

-0.5%
fg = 1.2 ¢xmin = 6 mm
-1.0%
fg = 1.5 ¢xmin = 37 mm
fg = 1.7 ¢xmin = 69 mm
-1.5%
fg = 2.0 ¢xmin = 100 mm

-2.0%
0 10 20 30 40 50
Simulation wall time [s]

Figure 5.34: Deviation from three-dimensional annual heat transfer vs. simulation wall time for
different combinations of discretization parameters

Increasing the cell growth factor results in larger positive deviations in approximation accu-

racy. Increasing the minimum cell dimension results in larger negative deviations in approximation

accuracy. These counteracting results cannot be relied upon to generate faster, more accurate sim-

ulations. However, it is possible to relax either or both of the discretization parameters to reduce

simulation wall time within 10 seconds without introducing deviations greater than 2%.

5.8 Summary and Conclusions

The accuracy and computational performance of the boundary layer adjustment method

demonstrate how the Kiva framework is applied to improve foundation heat transfer calculations.

Testing approximations against a verified, three-dimensional solution bolsters the algorithmic con-

fidence behind the calculations. Each simplification, whether to the spatial discretization, the
 143

numerical scheme, the initialization method, or the number of dimensions, is justified through

comparative analysis.

Foundation heat transfer, which would take several days to simulate in three-dimensions on

a supercomputer, is approximated within five seconds on a laptop with less than 5% deviation

in results. The boundary layer adjustment method, with a suitably coarse spatial discretization,

would account for a suitably small fraction of total simulation time in a whole-building energy

simulation context.
 Chapter 6

Conclusions and Future Work

6.1 Conclusions

Kiva provides a balance of capability, complexity, computational performance and algorithmic

confidence that is not available in other simulation tools or calculation methods.

Generalized descriptions of the foundation construction elements allow for assessment of a

fully continuous range of insulation design options. The boundary layer adjustment method allows

for accurate simulations of almost any foundation shape. Automated discretization of domain

dramatically decreases the complexity of the problem, requiring the user to only input information

about the foundation system, and not details of numerical methods used to solve the problem.

Three different components of the foundation heat transfer calculation were tested within the

Kiva framework to assess their impact on both computational performance and accuracy. These

components include the numerical scheme, the initialization method, and the two-dimensional

approximation of three-dimensional heat transfer. Of the numerical schemes tested, the Alternating

Direction Implicit (ADI) scheme demonstrates the best balance between accuracy, performance, and

numerical stability. The new accelerated initialization approach significantly reduces the required

years of presimulation. The boundary layer adjustment method improves accuracy over other

established two-dimensional approximation methods with a negligible increase in computation time.

This method accounts for reduced heat transfer from concave foundation shapes, which has not

been adequately addressed to date. Within the Kiva framework, three-dimensional heat transfer
 145

that can require several days to simulate is approximated in two-dimensions in a matter of seconds

while maintaining accuracy within 5%.

Two-dimensional approximations are able to capture the strongest directional dependence

of foundation heat transfer providing algorithmic confidence without the computational burden

required of three-dimensional analyses.

The combined recommendations presented in Table 6.1 result in reliably efficient and accurate

foundation heat transfer calculations that can be performed in the context of whole-building energy

analysis.

Table 6.1: Recommended methods and numerical parameters for foundation heat transfer

Description Value
Minimum cell dimension 0.02 m
Cell growth coefficient 1.50
Time-step duration 60 minutes
Numerical method ADI
Initialization method Accelerated
Initialization acceleration time-step duration 1 week
Initialization acceleration period duration 12 weeks
Initialization warm-up duration 1 year
Two-dimensional approximation Boundary layer adjustment method

6.2 Future Work

The flexible framework and open source development positions Kiva for extension into future

studies and further improvements to calculating heat transfer from building foundations.

6.2.1 Whole-Building Energy Simulation Tools

Kiva must be integrated into a whole-building energy simulation engine before its advantages

are fully utilized by energy modeling professionals. The Kiva framework was designed with integra-

tion in mind, by exposing access to important run-time variables that determine the heat balance

between the foundation system and the rest of the building. Once integrated, Kiva will provide
 146

convective heat flux from foundation surfaces to thermal zones in a whole-building energy model,

which, in turn, provides the thermal zone temperatures that determine heat transfer through the

foundation in the following time-step.

In addition to interior zone temperatures, the whole-building simulation engine will provide

information from HVAC systems that interact with the foundation and the surrounding ground.

Kiva will use heat source values to simulate radiant slabs, ground source heat pumps, and snow-melt

systems.

6.2.2 Foundation Design Guidelines

Once Kiva is integrated into a whole-building simulation engine, it can be used to help in-

form changes to foundation design guidelines. Current guidelines for foundation insulation design

include the Builder’s Foundation Handbook [20], which includes insulation recommendations based

on analysis performed in 1988 [77] and ASHRAE Standard 90.1 [8], which defines levels of insu-

lation based on analysis performed in 1991 [12]. The underlying methodology behind both sets

of guidelines are woefully out of date. Both the recommendations in the Builder’s Foundation

Handbook and the standard insulation levels in ASHRAE Standard 90.1 can be updated based on

new, more accurate analyses using Kiva.

6.2.3 Additional Capabilities

There are several capabilities mentioned in the literature review chapter (Chapter 2) that

were not incorporated into Kiva by the time of this thesis. Some suggested capabilities are listed

below in order of priority:

• Multiple foundation types within a single building

• Multiple interior thermal zones

• Spatially varying soil properties

• Spatially varying surface properties

 147

• Coupled heat and moisture transfer calculations with dynamic thermal properties

6.2.4 Sensitivity to Model Parameters

Two types of sensitivity analyses can be performed using the Kiva framework. The first

type of analysis tests computational performance and accuracy impacts of certain computational

routines. There are several routines that are not included in Kiva because of their perceived impact

on computational performance. The goal of these analyses is to quantify the trade-off between

computational performance and accuracy of their respective routines. Potential routines include:

• shading of incident solar on the ground,

• coupled heat and moisture transfer,

• soil freezing and thawing,

• snow cover, and

• evapotranspiration

The other type of analysis involves exploration of propagation of uncertainty in model inputs.

Because there are many relevant inputs to a foundation heat transfer model that are difficult

to measure within reasonable certainty, it is important to understand the implications of this

uncertainty as it propagates into results and, ultimately, recommended foundation insulation levels.

Two important inputs to investigate are the thermal properties of the soil, and the deep ground

boundary condition, which is strongly influenced by the presence of a high water table.

6.2.5 Performance Optimization

The source code of Kiva has not been optimized for memory utilization or CPU performance.

There are likely several areas in the source code that can be improved to increase Kiva’s speed

and accessibility. There are several tools available to profile the execution of Kiva’s source code to

identify potential bottlenecks in the simulation process.

 148

Iterative solvers have also improved drastically over the past several decades. Many authors

cite Algebraic Multigrid (AMG) methods with computational efficiencies comparable to the Al-

ternating Direction Implicit (ADI) method. AMG methods do not require a structured grid, and

can be solved using parallel processes [19]. With the increasing trend of parallel architecture in

computing, AMG methods should be evaluated as a potential replacement for the ADI method.
 Bibliography

[1] M.H. Adjali, M. Davies, and J. Littler. Three-Dimensional Earth-Contact Heat Flows: A
Comparison of Simulated and Measured Data for a Buried Structure. Renewable Energy, 15:
356–359, April 1998.

[2] M.H. Adjali, M. Davies, and J. Littler. Earth-Contact Heat Flows: Review and Application
of Design Guidance Predictions. Building Services Engineering Research and Technology, 19
(3):111–121, 1998.

[3] M.H. Adjali, M. Davies, and J. Littler. A Numerical Simulation of Measured Transient
Temperatures in the Walls, Floor and Surrounding Soil of a Buried Structure. International
Journal of Heat and Mass Transfer, 9(4):405–422, 1999.

[4] M.H. Adjali, M. Davies, S.W. Rees, and J. Littler. Temperatures in and under a Slab-on-
Ground Floor: Two- and Three-Dimensional Numerical Simulations and Comparison with
Experimental Data. Building and Environment, 35:655–662, October 2000.

[5] M.H. Adjali, M. Davies, and S.W. Rees. A Comparative Study of Design Guide Calculations
and Measured Heat Loss through the Ground. Building and Environment, 39(11):1301–1311,
November 2004.

[6] A. Al-Anzi and M. Krarti. Local/global analysis of transient heat transfer from building
foundations. Building and Environment, 39(5):495–504, May 2004.

[7] ASHRAE. ASHRAE Handbook—Fundamentals. ASHRAE, Atlanta, Georgia, 2009.

[8] ASHRAE. ANSI/ASHRAE/IES Standard 90.1-2013: Energy Standard for Buildings Except
Low-Rise Residential Buildings, 2013.

[9] W. Bahnfleth. Three-Dimensional Modeling of Heat Transfer from Slab Floors. Ph.d. thesis,
University of Illinois at Urbana Champaign, 1989.

[10] W. Bahnfleth and J. Amber. Algorithms for Slab-on-Grade Heat Transfer Calculations. Tech-
nical report, US Army Corps of Engineers Construction Engineering Research Laboratory,
Champaign, IL, 1990.

[11] W. Bahnfleth and C. Pedersen. A Three-Dimensional Numerical Study of Slab-on-Grade

Heat Transfer. ASHRAE Transactions, 96(2):61–72, 1990.
 150

[12] D. Baylon and M. Kennedy. Calculating the Impact of Ground Contact on Residential Heat
Loss. In Proceedings of Thermal Performance of the Exterior Envelopes of Buildings X,
Clearwater, Florida, 2007. ASHRAE.

[13] V. Bazjanac, Y. Huang, and F. Winkelmann. DOE-2 Modeling of Two-Dimensional Heat

Flow in Underground Surfaces. Technical report, California Energy Commission, 2000.

[14] I. Beausoleil-Morrison and M. Krarti. Predicting Foundation Heat Losses: Neural Networks
Versus the BASESIMP Correlations. In Proceedings of Building Simulation 1997, Prague,
Czech Republic, 1997. IBPSA.

[15] I. Beausoleil-Morrison and G. Mitalas. BASESIMP: A residential-foundation heat-loss al-

gorithm for incorporating into whole-building energy-analysis programs. In Proceedings of
Building Simulation 1997, Prague, Czech Republic, 1997. IBPSA.

[16] I. Beausoleil-Morrison, G. Mitalas, and H. Chin. Estimating Three-Dimensional Below-Grade

Heat Losses from Houses Using Two-Dimensional Calculations. In Proceedings of Thermal
Performance of the Exterior Envelopes of Buildings VI, pages 95–99, Clearwater, Florida,
1995. ASHRAE.

[17] I. Beausoleil-Morrison, G. Mitalas, and C. McLarnon. BASECALC(TM): New Software for

Modelling Basement and Slab-on-Grade Heat Losses. In Proceedings of Building Simulation
1995, Madison, Wisconsin, 1995. IBPSA.

[18] C. Booten, N. Kruis, and C. Christensen. Identifying and Resolving Issues in EnergyPlus
and DOE-2 Window Heat Transfer Calculations. Technical Report August, NREL, Golden,
Colorado, 2012.

[19] W. Briggs, V. Hensen, and S. McCormick. A Multigrid Tutorial. Society of Industrial and
Applied Mathematics, Philadelphia, Pennsylvania, 2 edition, 2000.

[20] J. Carmody, J. Christian, and K. Labs. Builder’s Foundation Handbook. Oak Ridge National
Laboratory, Oak Ridge, Tennessee, 1991.

[21] M. Chang, L. Chow, and W. Chang. Improved Alternating-Direction Implicit Method for
Solving Transient Three-Dimensional Heat Diffusion Problems. Numerical Heat Transfer:
Part B: Fundamentals, 19:69–84, 1991.

[22] S. Choi and M. Krarti. Heat transfer for slab-on-grade floor with stepped ground. Energy
Conversion and Management, 39(7):691–701, 1998.

[23] S. Choi and M. Krarti. Thermally optimal insulation distribution for underground structures.
Energy and Buildings, 32(3):251–265, September 2000.

[24] P. Chuangchid. Heat Gain from Slab-on-Grade Floor of Refrigerated Structures. Ph.d. thesis,
University of Colorado at Boulder, 2000.

[25] P. Chuangchid and M. Krarti. Steady-State Component of Three-Dimensional Slab-on-Grade

Foundation Heat Transfer. Journal of Solar Energy Engineering, 123(1):18, 2001.

[26] J. Claesson and C.E. Hagentoft. Heat loss to the ground from a building–I. General theory.
Building and Environment, 26(2):195–208, 1991.
 151

[27] D. Claridge. Design Methods for Earth-Contact Heat Transfer. In Karl Böer, editor, Advances
in Solar Energy: An Annual Review of Research and Development, chapter 7, pages 305–351.
American Solar Energy Society, Inc., Boulder, CO, 4 edition, 1988.

[28] J. Clark and M. Korybalski. Algebraic Methods for the Calculation of Radiation Exchange
in an Enclosure. Wärme - und Stoffübertragung, 7(1):31–44, 1974.

[29] E. Clements. Three Dimensional Foundation Heat Transfer Modules for Whole-Building
Energy Analysis. Master’s thesis, Pennsylvania State University, 2004.

[30] C. Cogil. Modeling of Basement Heat Transfer and Parametric Study of Basement Insulation
for Low Energy Housing. Master’s thesis, Pennsylvania State University, 1998.

[31] M. Crowley. Modeler Report for BESTEST Cases GC10a-GC80c, MATLAB 7.0.4.365 (R14)
Service Pack 2. Technical report, Dublin Institute of Technology, Dublin, Ireland, 2007.

[32] J. Cullin, L. Xing, E. Lee, J. Spitler, and D. Fisher. Feasibility of Foundation Heat Ex-
changers for Residential Ground Source Heat Pump Systems in the United States. ASHRAE
Transactions, 118(1):1039–1048, 2012.

[33] M. Davies. Computational and Experimental Three-Dimensional Conductive Heat Flows in

and around Buildings. Ph.d. thesis, University of Westminster, 1994.

[34] M. Davies, S. Zoras, and M.H. Adjali. A Potentially Fast, Flexible and Accurate Earth-
Contact Heat Transfer Simulation Method. In Proceedings of Building Simulation 1999,
number 1987, Kyoto, Japan, 1999. IBPSA.

[35] D. De Vries. Simultaneous Transfer of Heat and Moisture in Porous Media. American
Geophysical Union Transactions, 39(5):909–916, 1958.

[36] D. De Vries. Thermal Properties of Soils. In W Van Wijk, editor, Physics of Plant
Environment, chapter 7, pages 210–235. North-Holland Publishing Co., Amsterdam, 1966.

[37] A. Delsante. The effect of water table depth on steady-state heat transfer through a slab-on-
ground floor. Building and Environment, 28(3):369–372, 1993.

[38] A. Delsante, A. Stokes, and P. Walsh. Application of Fourier Transforms to Periodic Heat
Flow into the Ground Under a Building. International Journal of Heat and Mass Transfer,
26(1):121–132, 1983.

[39] M. Deru. A Model for Ground-Coupled Heat and Moisture Transfer from Buildings A Model
for Ground-Coupled Heat and Moisture Transfer from Buildings. Technical Report June,
National Renewable Energy Laboratory, Golden, Colorado, 2003.

[40] M. Deru, R. Judkoff, and J. Neymark. Whole Building Energy Simulation with a Three-
Dimensional Ground-Coupled Heat Transfer Model. ASHRAE Transactions, 109(1):557–565,
2003.

[41] G.H. dos Santos and N. Mendes. Multidimensional Effects of Ground Heat Transfer on the
Dynamics of Building Thermal Performance. ASHRAE Transactions, 110(2):345–354, 2004.

[42] G.H. dos Santos and N. Mendes. Simultaneous heat and moisture transfer in soils combined
with building simulation. Energy and Buildings, 38(4):303–314, April 2006.
 152

[43] J. Duffie and W. Beckman. Solar Engineering of Thermal Processes. John Wiley & Sons Ltd,
Hoboken, New Jersey, third edit edition, 2006.

[44] D. Fisher and S. Rees. Modeling Ground Source Heat Pump Systems in a Building Energy
Simulation Program (EnergyPlus). In Proceedings of Building Simulation 2005, pages 311–
318, Montreal, Canada, 2005. IBPSA.

[45] P. Fritzson. Principles of Object-Oriented Modeling and Simulation with Modelica 2.1.
Wiley-IEEE Press, 2003. ISBN 0471471631.

[46] K. Fukuyo. Conditional Stability of Larkin Methods with Non-Uniform Grids. Theoretical
and Applied Mechanics, 37(2):139–159, 2010. ISSN 1450-5584. doi: 10.2298/TAM1002139F.

[47] R. Gilpin and B. Wong. ”Heat-Valve” Effects in the Ground Thermal Regime. Journal of
Heat Transfer, 98(4):537–542, 1976.

[48] C.E. Hagentoft. Heat Loss to the Ground from a Building: Slab on the Ground and Cellar.
Ph.d. thesis, Lund Institute of Technology, Sweden, 1988.

[49] C.E. Hagentoft. Heat Losses and Temperature in the Ground under a Building with and with-
out Ground Water Flow—I. Infinite Ground Water Flow Rate. Building and Environment,
31(1):3–11, January 1996.

[50] C.E. Hagentoft. Steady-State Heat Loss for an Edge-Insulated Slab : Part I. Building and
Environment, 37(1):19–25, January 2002.

[51] C.E. Hagentoft and J. Claesson. Heat Loss to the Ground from a BuildingII. Slab on the
Ground. Building and Environment, 26(4):395–403, 1991.

[52] C.E. Hagentoft and J. Claesson. Influence of Rain Water Percolation on Ground Heat
Losses and Temperature for Basement Foundation. In Proceedings of the 3rd International
Conference on Research in Building Physics and Building Engineering, Montreal, Canada,
2006.

[53] H. Hottel and A. Sarofim. Radiative Transfer. McGraw Hill Book Co., 1967.

[54] Y. Huang, L. Shen, J. Bull, and L. Goldberg. Whole-house simulation of foundation heat
flows using the DOE-2.1 C program. ASHRAE Transactions, 94(2):936–958, 1988.

[55] James J. Hirsch & Associates. DOE-2, 2010. URL http://www.doe2.com/.

[56] H. Janssen. The Influence of Soil Moisture Transfer On Building Heat Loss Via the Ground.
Ph.d. thesis, Katholieke Universiteit Leuven, July 2002.

[57] H. Janssen, J. Carmeliet, and H. Hens. The Influence of Soil Moisture Transfer on Building
Heat Loss via the Ground. Building and Environment, 39(7):825–836, July 2004.

[58] N. Jones and D. Greenberg. Fast Computation of Incident Solar Radiation from Preliminary
to Final Building Design. In Proceedings of Building Simulation 2011, pages 14–16, Sydney,
Australia, 2011. IBPSA.
 153

[59] R. Judkoff and J. Neymark. Home Energy Rating System Building Energy Simulation Test
( HERS BESTEST ) Volume 1 Tier 1 and Tier 2 Tests User s Manual. Technical Report
November, National Renewable Energy Laboratory, Golden, Colorado, 1995.

[60] R. Judkoff and J. Neymark. Model Validation and Testing : The Methodological Foundation
of ASHRAE Standard 140. ASHRAE Transactions, 112(2):367–376, 2006.

[61] B.D. Kay and E. Perfect. State of the art: Heat and mass transfer in freezing soils. In
Proceedings of 5th International Symposium on Ground Freezing, Rotterdam, Netherlands,
1988. Balkema, Rotterdam.

[62] M. Krarti. Developments in Ground-Coupled Heat Transfer. Ph.d. thesis, University of

Colorado, 1987.

[63] M. Krarti. Time-Varying Heat Transfer From Slab-on-Grade Floors with Vertical Insulation.
Building and Environment, 29(1):55–61, January 1994.

[64] M. Krarti. Comparison of a Neural Network Model with a Regression Model for Foundation
Heat Loss Calculation. In Proceedings of Thermal Performance of the Exterior Envelopes of
Buildings VI, pages 79–85, Clearwater, Florida, 1995. ASHRAE.

[65] M. Krarti. Effect of Spatial Variation of Soil Thermal Properties on Slab-on-Ground Heat
Transfer. Building and Environment, 31(1):51–57, January 1996.

[66] M. Krarti. Building Foundation Heat Transfer. In Yogi Goswami and Karl Böer, editors,
Advances in Solar Energy: An Annual Review of Research and Development, chapter 6, pages
241–315. American Solar Energy Society, Inc., Boulder, CO, 13 edition, 1999.

[67] M. Krarti. Impact of Above-Grade Walls on Three-Dimensional Building Foundation Heat

Transfer From Slab-On Grade Floors. Journal of Solar Energy Engineering, 136(1), 2013.

[68] M. Krarti and S. Choi. Simplified Method for Foundation Heat Loss Calculation. ASHRAE
Transactions, 102(1):140–152, 1996.

[69] M. Krarti and S. Choi. A Simulation Method for Fluctuating Temperatures in Crawlspace
Foundations. Energy and Buildings, 26(2):183–188, 1997.

[70] M. Krarti, D. Claridge, and J. Kreider. ITPE Technique Applications to Time-Varying Two-
Dimesional Ground-Coupling Problems. International Journal of Heat and Mass Transfer, 31
(9):1899–1911, 1988.

[71] M. Krarti, D. Claridge, and J. Kreider. The ITPE Technique Applied to Steady-State Ground-
Coupling Problems. International Journal of Heat and Mass Transfer, 31(9):1885–1898, 1988.

[72] M. Krarti, J. Kreider, and D. Claridge. ITPE Technique Applications to Time-Varying Three-
Dimensional Ground-Coupling Problems. Journal of Heat Transfer, 112:849–856, 1990.

[73] M. Krarti, P. Chuangchid, and P. Ihm. Foundation Heat Transfer Module for EnergyPlus
Program. In Proceedings of Building Simulation 2001, pages 931–938, Rio de Janeiro, Brazil,
2001. IBPSA.

[74] N. Kruis. Kiva, 2014. URL https://github.com/nealkruis/kiva.

 154

[75] T. Kusuda and P.R. Achenbach. Earth temperature and thermal diffusivity at selected sta-
tions in the United States. ASHRAE Transactions, 73:61–75, 1965.

[76] T. Kusuda and J.W. Bean. Simplified Methods for Determining Seasonal Heat Loss from
Uninsulated Slab-on-Grade Floors. ASHRAE Transactions, 90(1):611–632, 1984.

[77] K. Labs, J. Carmody, R. Sterling, L. Shen, Y. Huang, and D. Parker.

Building Foundation Design Handbook. Oak Ridge National Laboratory, Oak Ridge,
Tennessee, 1988.

[78] A. Lachenbruch. Three-Dimensional Heat Conduction in Permafrost Beneath Heated Build-

ings. In U. S. Geological Survey: Experimental and Theoretical Geophysics: Bulletin 1052,
chapter B, pages 51–69. United States Government Printing Office, Washington D.C., 1959.

[79] J. Latta and G. Boileau. Heat Losses from House Basements. Canadian Building, XIX(10):
39–42, 1969.

[80] Lawrence Berkeley National Laboratory and James J. Hirsch & Associates.
DOE-2.2: Building Energy Use and Cost Analysis Program. Volume 2: Dictionary. James
J. Hirsch & Associates, Camarillo, California, 2009. URL http://www.doe2.com/.

[81] N. Li, F. Chen, B. Xu, and G. Swoboda. Theoretical Modeling Framework for an Unsaturated
Freezing Soil. Cold Regions Science and Technology, 54:19–35, 2008.

[82] X. Liu, S. Rees, and J. Spitler. Modeling Snow Melting on Heated Pavement Surfaces. Part
I: Model Development. Applied Thermal Engineering, 27(5-6):1115–1124, April 2007.

[83] C. Ludeman. Passive Project Foundation and Slab Insulation, 2009. URL http://www.
100khouse.com/2009/06/11/passive-project-foundation-slab-insulation/.

[84] X. Ma, B. Cheng, J. Mao, W. Liu, and D. Zi. Finite Element Modelling of Coupled Heat and
Moisture Transfer in Typical Earth-Sheltered Building Envelope. In Proceedings of Building
Simulation 2009, pages 1850–1856, Glasgow, Scotland, 2009. IBPSA.

[85] J. MacArthur, G. Meixel, and L. Shen. Application of Numerical Methods for Predicting
Energy Transport in Earth Contact Systems. Applied Energy, 13:121–156, 1983.

[86] T. McDowell, J. Thornton, and M. Duffy. Comparison of a Ground-coupling Reference Stan-

dard Model to Simplified Approaches. In Proceedings of Building Simulation 2009, pages
591–598, Glasgow, Scotland, 2009. IBPSA.

[87] G. Mitalas. Calculation of Below-Grade Residential Heat Loss: Low-Rise Residential Build-
ing. ASHRAE Transactions, 93(1):743–783, 1987.

[88] F. Moukalled and Y. Saleh. Heat and Mass Transfer in Moist Soil, Part I. Formulation and
Testing. Numerical Heat Transfer: Part B: Fundamentals, 49(5):467–486, October 2006.

[89] A. Nakhi. Modeler Report for BESTEST Cases GC10a-GC80c, FLUENT Version 6.0.20.
Technical report, Public Authority for Applied Education and Training, Kuwait, 2007.
 155

[90] J. Neymark and R. Judkoff. International Energy Agency Building Energy Simulation Test
and Diagnostic Method ( IEA BESTEST ): In-Depth Diagnostic Cases for Ground Coupled
Heat Transfer Related to Slab-on-Grade Construction. Technical Report September, National
Renewable Energy Laboratory, Golden, Colorado, 2008.

[91] J. Rantala and V. Leivo. Heat Loss into Ground from a Slab-on-Ground Structure in a Floor
Heating System. International Journal of Energy Research, 30(12):929–938, October 2006.

[92] S. Rees, J. Spitler, and X. Xiao. Transient Analysis of Snow-Melting System Performance.
ASHRAE Transactions, 108(2):406–423, 2002.

[93] S.W. Rees and H. Thomas. Two-Dimensional Heat Transfer beneath a Modern Commercial
Building: Comparison of Numerical Prediction with Field Measurement. Building Services
Engineering Research and Technology, 18(3):169–174, 1997.

[94] S.W. Rees, R. Lloyd, and H. Thomas. A Numerical Simulation of Measured Transient Heat
Transfer through a Concrete Ground Floor Slab and Underlying Substrata. International
Journal of Numerical Methods for Heat & Fluid Flow, 5(8):669–683, 1995.

[95] S.W. Rees, H. Thomas, and Z. Zhou. Ground Heat Transfer: Some Further Insights into
the Influence of Three-Dimensional Effects. Building Services Engineering Research and
Technology, 21(4):233–239, 2000.

[96] S.W. Rees, Z. Zhou, and H. Thomas. The Influence of Soil Moisture Content Variations
on Heat Losses from Earth-Contact Structures : An Initial Assessment. Building and
Environment, 36:157–165, 2001.

[97] S.W. Rees, H. Thomas, and Z. Zhou. A Numerical and Experimental Investigation of Three-
Dimensional Ground Heat Transfer. Building Services Engineering Research and Technology,
27(3):195–208, 2006.

[98] S.W. Rees, Z. Zhou, and H. Thomas. Ground Heat Transfer: A Numerical Simulation of a
Full-Scale Experiment. Building and Environment, 42(3):1478–1488, March 2007.

[99] B. Rock. Sensitivity Study of Slab-on-Grade Transient Heat Transfer Model Parameters.
ASHRAE Transactions, 110(1):177–184, 2004.

[100] V. Saul’yev. Integration of Equations of Parabolic Type by the Method of Nets. Pergamon
Press, New York, 1964.

[101] L. Shen. An Investigation of Transient, Two-Dimensional Coupled Heat and Moisture Flow
in Soils. Ph.d. thesis, University of Minnesota, 1986.

[102] L. Shen and J. Ramsey. An Investigation of Transient, Two-Dimensional Coupled Heat and
Moisture Flow in the Soil Surrounding a Basement Wall. International Journal of Heat and
Mass Transfer, 31(7):1517–1527, 1988.

[103] B. Sibbitt, D. Mcclenahan, R. Djebbar, J. Thornton, and B. Wong. The Performance of

a High Solar Fraction Seasonal Storage District Heating System Five Years of Operation.
Energy Procedia, 30(SHC 2012):856–865, 2012.
 156

[104] Solar Energy Laboratory. TRNSYS 17: a TRaNsient SYstem Simulation program. Solar En-
ergy Laboratory, University of Wisconsin-Madison, Madison, Wisconsin, 2010.

[105] R. Sterling, S. Gupta, L. Shen, and L. Goldberg. Assessment of Soil Thermal Conductivity
for Use in Building Design and Analysis. Technical report, ASHRAE, Atlanta, GA, 1993.

[106] The Scalable Software Infrastructure Project. Lis User Guide. The Scalable Software Infras-
tructure Project, Fukuoka, Japan, 2014.

[107] H. Thomas and S.W. Rees. The Thermal Performance of Ground Floor Slabs–a Full Scale
in-situ experiment. Building and Environment, 34:139–164, March 1999.

[108] H. Thomas and S.W. Rees. Measured and Simulated Heat Transfer to Foundation Soils.
Géotechnique, 59(4):365–375, 2009.

[109] J. Thornton. Modeler Report for BESTEST Cases GC10a-GC80c, TRNSYS Version 16.1.
Technical report, Thermal Energy Systems Specialists, Madison, Wisconsin, 2007.

[110] H. Trethowen and A. Delsante. A Four-Year Site Measurement of Heat Flow in Slab-on-
Ground Floors with Wet Soils. In Proceedings of Thermal Performance of the Exterior
Envelopes of Buildings VII, Clearwater, Florida, 1998. ASHRAE.

[111] United States Department Of Energy (DOE). EnergyPlus: Input Output Reference. The
Board of Trustees of the University of Illinois and the Regents of the University of California
through the Ernest Orlando Lawrence Berkeley National Laboratory, Berkeley, California,
version 8. edition, 2014.

[112] B. Usowicz, J. Lipiec, and A. Ferrero. Prediction of soil thermal conductivity based on
penetration resistance and water content or air-filled porosity. International Journal of Heat
and Mass Transfer, 49(25-26):5010–5017, December 2006.

[113] G. Walton. Thermal Analysis Research Program Reference Manual. U.S. Dept. of Com-
merce, National Bureau of Standards, Washington D.C., 1983.

[114] G. Walton. Estimating 3-D heat loss from rectangular basements and slabs using 2-D calcu-
lations. ASHRAE Transactions, 93(1):791–797, 1987.

[115] F. Wang. Mathematical Modeling and Computer Simulation of Insulation Systems in Below
Grade Applications. In Proceedings of Thermal Performance of the Exterior Envelopes of
Buildings I, pages 456–471, Clearwater, Florida, 1979. ASHRAE.

[116] E. Wentzel. Thermal Modeling of Walls, Foundations and Whole Buildings Using Dynamic
Thermal Networks. Ph.d. thesis, Chalmers University of Technology, 2005.

[117] S. Wilcox and W. Marion. Users Manual for TMY3 Data Sets. Technical report, National
Renewable Energy Laboratory, Golden, Colorado, 2008.

[118] F. Winkelmann. Underground Surfaces: How To Get A Better Underground Surface Heat
Transfer Calculation In DOE-2.1E. DOE-2 Articles from the Building Energy Simulation
User News, 23(5):5–14, 2002.
 157

[119] X. Xie, Y. Jiang, and J. Xia. A New Approach to Compute Heat Transfer of Ground-Coupled
Envelope in Building Thermal Simulation Software. Energy and Buildings, 40(4):476–485,
2008.

[120] M. Yazdanian and J. Klems. Measurement of the exterior convective film coefficient for
windows in low-rise buildings. ASHRAE Transactions, 100(1), 1994. URL http://btech.
lbl.gov/papers/34717.pdf.

[121] C. Zhang and G. Ding. A novel thermal response factor method for the dynamic load cal-
culation of buildings. Journal of Asian Architecture and Building Engineering, 1(1):75–79,
2002.
 Appendix A

Finite Difference Discretizations

The discretization equations for each of the finite difference methods implemented in Kiva

are provided for a two-dimensional axisymmetric cylindrical coordinate system.

A.1 Continuous Equation

Heat transfer in the ground domain is solved using the diffusion equation
∂T
CH,T = ∇ [k∇T ] + q̇ (A.1)
∂t

where:

• CH,T is the thermal heat capacitance [J/m3 -K]

• T is the temperature at a given spatial position and point in time [K]

• t is time [s]

• ∇ is the differential spatial gradient operator

• k is the thermal conductivity [W/m-K]

• q̇ is the heat source or sink in [W/m3 ]

The spatial gradient operator creates different equations depending on the coordinate system.

For Cartesian coordinates in three-dimensions, Equation A.74 becomes

∂T ∂ ∂T ∂ ∂T ∂ ∂T
CH,T =+ k + k + k + q̇. (A.2)
∂t ∂x ∂x ∂y ∂y ∂z ∂z
 159

For Cartesian coordinates in two-dimensions, Equation A.74 becomes

∂T ∂ ∂T ∂ ∂T
CH,T =+ k + k + q̇. (A.3)
∂t ∂x ∂x ∂z ∂z

For Axisymmetric cylindrical coordinates in two-dimensions, Equation A.74 becomes

∂T 1 ∂ ∂T ∂ ∂T
CH,T = rk + k + q̇. (A.4)
∂t r ∂r ∂r ∂z ∂z

A.2 Boundary Conditions

There are four types of boundary conditions encountered in Kiva. The formulation of the

boundary conditions are framed generically for a direction w, which is perpendicular to the surface.

Interior flux boundaries in Kiva fall along the foundation floor, the foundation wall, and the

under-side of any partial wall insulation.

∂T
−k = [hc + hr ] [T∞ − T (w)] + q̇sw , (A.5)
∂w

where

• hc is the convection coefficient [W/m2 ·K],

• hr is the long-wave radiation coefficient [W/m2 ·K],

• T∞ is the ambient air temperature [K],

• T (w) is the boundary temperature [K], and

• q̇sw is the incident short-wave radiation flux [W/m2 ].

Exterior flux boundaries are very similar to the interior boundaries except that they include

a separate long-wave radiation term which accounts for the difference between the radiative tem-

perature, T∞ F ∗ 1/4 , and the outdoor ambient temperature, T∞ . This boundary is applied to the

grade surface and the exterior wall surface if it is exposed.

∂T 
−k = hc [T∞ − T (w)] + hr T∞ F ∗ 1/4 − T (w) + q̇sw (A.6)
∂w
 160

where F ∗ is a correction factor to account for the difference between the sky temperature and the

ambient air temperature.

Zero-flux boundaries occur at far-field, deep-ground and symmetry planes as well as the top

of the foundation wall.

∂T
=0 (A.7)
∂w

Constant temperature boundaries are found either at deep-ground or at the wall-top (to

mimic the constraints of some analytical solutions).

T (w) = Tc (A.8)

where Tc is a constant temperature [K] applied to the boundary.

A.3 Discretized Equations

The following sections describe the discretized equations for each of the finite difference

methods implemented in Kiva. The equations presented here correspond to the continuous partial

differential equation for a two-dimensional cylindrical coordinate system (Equation A.75). For-

mulations for Cartesian two- and three-dimensional coordinate systems can be easily extended by

omission or addition of certain terms.

The only difference between the different finite difference schemes is the time-step associated

with the temperature terms. A generic finite difference scheme with temperatures T ∗ will provide

a basis for the more specific formulations:

n+1
Ti,j − Ti,j
n ∗
Ti+1,j ∗
− Ti,j ∗ − T∗
Ti,j
1 Δrm Δrp i−1,j
CH,T = kr,p + kr,m
Δt ri Δrm + Δrp Δrp Δrm + Δrp Δrm
∗
Ti+1,j ∗
− Ti,j ∗ − T∗
Ti,j
2 i−1,j (A.9)
+ kr,p − kr,m
Δrm + Δrp Δrp Δrm
∗
Ti,j+1 ∗
− Ti,j ∗ − T∗
Ti,j
2 i,j−1
+ kz,p − kz,m + q̇.
Δzm + Δzp Δzp Δzm

where

• i and j are spatial indices,

 161

• n is the temporal index,

• Δt is the time-step [s], and

Δri + Δri+1
Δrp = , (A.10)
2

Δri + Δri−1
Δrm = , (A.11)
2

Δzi + Δzi+1
Δzp = , (A.12)
2

Δzi + Δzi−1
Δzm = , (A.13)
2

−1
Δri Δri+1
kr,p = + , (A.14)
2Δrp ki,j 2Δrp ki+1,j

−1
Δri Δri−1
kr,m = + , (A.15)
2Δrm ki,j 2Δrm ki−1,j

−1
Δzj Δzj+1
kz,p = + , (A.16)
2Δzp ki,j 2Δzp ki,j+1

−1
Δzj Δzj−1
kz,m = + , (A.17)
2Δzm ki,j 2Δzm ki,j−1

where

• Δr is the cell width [m], and

• Δz is the cell height [m].

The nontemperature terms can then be collected into constant coefficients to simplify the

formulation of the discretized equations:

Δt
θ= , (A.18)
CH,T

Δrm kr,p
cr,p = θ , (A.19)
ri [Δrm + Δrp ] Δrp
 162
Δrp kr,m
cr,m = θ , (A.20)
ri [Δrm + Δrp ] Δrm

2kr,p
cr,p = θ , (A.21)
[Δrm + Δrp ] Δrp

2kr,m
cr,m = −θ , (A.22)
[Δrm + Δrp ] Δrm

2kz,p
cz,p = θ , (A.23)
[Δzm + Δzp ] Δzp

2kz,m
cz,m = −θ . (A.24)
[Δzm + Δzp ] Δzm

For Cartesian coordinates, the radial terms (cr,p , cr,m , cr,p , and cr,m ) will be omitted (set to

zero), and the x and y terms (cx,p , cx,m , cy,p , and cy,m ) can be defined in a similar fashion as the z

terms.

With the definitions of the constant coefficients, Equation A.9 can be re-written as:

 ∗   ∗   ∗ 
n+1
Ti,j − Ti,j
n
= cr,p Ti+1,j ∗
− Ti,j + cr,m Ti,j ∗
− Ti−1,j + cr,p Ti+1,j ∗
− Ti,j
 ∗ ∗
  ∗ ∗
  ∗ ∗
 (A.25)
+ cr,m Ti,j − Ti−1,j + cz,p Ti,j+1 − Ti,j + cz,m Ti,j − Ti,j−1 + θq̇.

The spatial derivative in the boundary condition relationships can be descretized as follows:
∂T T ∗ − Tl∗
≈ l−1 (A.26)
∂w Δwm

where l is the spatial index in the direction normal to the boundary, and
Δwm = Δwl−1 /2. (A.27)

This results in the following descretized boundary conditions:

∗ − T∗
Tl−1
−kw,m l
= [hc + hr ] [T∞ − Tl∗ ] + q̇sw , (A.28)
Δwm

∗ − T∗
Tl−1 
−kw,m l
= hc [T∞ − Tl∗ ] + hr T∞ F ∗ 1/4 − Tl∗ + q̇sw , (A.29)
Δwm

∗ − T∗
Tl−1 l
= 0, (A.30)
Δwm

Tl∗ = Tc . (A.31)
 163

A.3.1 Explicit Scheme

In the explicit scheme, all the temperatures on the right-hand side of Equation A.9 represent

the previous time-step, n:

 n   n   n 
n+1
Ti,j − Ti,j
n
= cr,p Ti+1,j − Ti,j
n
+ cr,m Ti,j − Ti−1,j
n
+ cr,p Ti+1,j − Ti,j
n
 n   n   n  (A.32)
+ cr,m Ti,j − Ti−1,j
n
+ cz,p Ti,j+1 − Ti,j
n
+ cz,m Ti,j − Ti,j−1
n
+ θq̇.

This equation can be solved explicitly for the temperature at the new time-step, n + 1:

    
n+1
Ti,j n
= Ti,j 1 + cr,m + cr,m + cz,m − cr,p − cr,p − cz,p + Ti−1,j
n
−cr,m − cr,m
n
   n n
(A.33)
+ Ti+1,j cr,p + cr,p + Ti,j−1 [−cz,m ] + Ti,j+1 [cz,p ] + θq̇.

A.3.1.1 Boundary Conditions

For interior flux boundaries, the discretized equation becomes

n − T n+1
Tl−1 l  
−kw,m = [hc + hr ] T∞ − Tln+1 + q̇sw , (A.34)
Δwm
which solves for Tln+1 as
kw,m n
T + [hc + hr ] T∞ − q̇sw
Δwm l−1
Tln+1 = . (A.35)
kw,m
+ [hc + hr ]
Δwm
For exterior flux boundaries, the discretized equation becomes
n − T n+1
Tl−1   
−kw,m l
= hc T∞ − Tln+1 + hr T∞ F ∗ 1/4 − Tln+1 + q̇sw , (A.36)
Δwm
which solves for Tln+1 as
kw,m n 
Tl−1 + hc + hr F ∗ 1/4 T∞ − q̇sw
Δwm
Tln+1 = . (A.37)
kw,m
+ [hc + hr ]
Δwm
For zero-flux boundaries, the discretized equation becomes
n − T n+1
Tl−1 l
= 0, (A.38)
Δwm
which solves for Tln+1 as
Tln+1 = Tl−1
n
. (A.39)

For constant temperature boundaries, the discretized equation becomes

Tln+1 = Tc . (A.40)
 164

A.3.2 Fully Implicit Scheme

In the fully implicit scheme, all the temperatures on the right-hand side of Equation A.9

represent the new time-step, n + 1:

  
n+1
Ti,j − Ti,j
n
= cr,p Ti+1,j
n+1 n+1
− Ti,j + cr,m Ti,j
n+1 n+1
− Ti−1,j n+1
+ cr,p Ti+1,j n+1
− Ti,j
   (A.41)
n+1 n+1 n+1
+ cr,m Ti,j − Ti−1,j + cz,p Ti,j+1 − Ti,j n+1 n+1 n+1
+ cz,m Ti,j − Ti,j−1 + θq̇.

This equation can be rearranged with all the unknown temperatures at the new time-step,

n + 1, on the left-hand side of the equation with constant coefficients and all the constant terms

on the right-hand side:

    
n+1
Ti,j 1 + cr,p + cr,p + cz,p − cr,m − cr,m − cz,m + Ti−1,jn+1
cr,m + cr,m
n+1
   (A.42)
+ Ti+1,j n+1
−cr,p − cr,p + Ti,j−1 n+1
[cz,m ] + Ti,j−1 n
[−cz,p ] = Ti,j + θq̇

A.3.2.1 Boundary Conditions

For interior flux boundaries, the discretized equation becomes

n+1
Tl−1 − Tln+1  
−kw,m = [hc + hr ] T∞ − Tln+1 + q̇sw , (A.43)
Δwm

which can be arranged with the unknown temperatures on the left-hand side as
kw,m n+1 kw,m
Tln+1 + [hc + hr ] − Tl−1 = [hc + hr ] T∞ − q̇sw . (A.44)
Δwm Δwm

For exterior flux boundaries, the discretized equation becomes

n+1
Tl−1 − Tln+1   
−kw,m = hc T∞ − Tln+1 + hr T∞ F ∗ 1/4 − Tln+1 + q̇sw , (A.45)
Δwm

which can be arranged with the unknown temperatures on the left-hand side as
kw,m 
n+1 kw,m
Tln+1 + [hc + hr ] − Tl−1 = hc + hr F ∗ 1/4 T∞ − q̇sw . (A.46)
Δwm Δwm

For zero-flux boundaries, the discretized equation becomes

n+1
Tl−1 − Tln+1
= 0, (A.47)
Δwm
 165

which can be arranged with the unknown temperatures on the left-hand side as

Tln+1 − Tl−1
n+1
= 0. (A.48)

For constant temperature boundaries, the discretized equation becomes

Tln+1 = Tc . (A.49)

A.3.3 Crank-Nicolson Scheme

In the Crank-Nicolson scheme, the right-hand side of the equation is the average of the explicit

and fully implicit schemes: In the fully implicit scheme, all the temperatures on the right-hand side

of Equation A.9 represent the new time-step, n + 1:

1   
n+1
Ti,j − Ti,j
n
= cr,p Ti+1,j
n+1 n+1
− Ti,j + cr,m Ti,j
n+1 n+1
− Ti−1,j n+1
+ cr,p Ti+1,j n+1
− Ti,j
2   
n+1
+ cr,m Ti,j n+1
− Ti−1,j n+1
+ cz,p Ti,j+1 − Ti,jn+1
+ cz,m Ti,jn+1 n+1
− Ti,j−1
  n   n   n   n 
+ cr,p Ti+1,j − Ti,j
n
+ cr,m Ti,j − Ti−1,j
n
+ cr,p Ti+1,j − Ti,j
n
+ cr,m Ti,j − Ti−1,j
n
 n   n 
+ cz,p Ti,j+1 − Ti,j
n
+ cz,m Ti,j − Ti,j−1
n
+ θq̇.
(A.50)

This equation can be rearranged with all the unknown temperatures at the new time-step,

n + 1, on the left-hand side of the equation with constant coefficients and all the constant terms

on the right-hand side:

1   n+1 1
  
n+1
Ti,j 1+ cr,p + cr,p + cz,p − cr,m − cr,m − cz,m + Ti−1,j cr,m + cr,m
2 2
n+1 1
   n+1 1 n+1 1
+ Ti+1,j −cr,p − cr,p + Ti,j−1 [cz,m ] + Ti,j−1 [−cz,p ]
2 2 2 (A.51)
1   1  
n
= Ti,j 1+ cr,m + cr,m + cz,m − cr,p − cr,p − cz,p + Ti−1,j n
−cr,m − cr,m
2 2
1   1 1
n
+ Ti+1,j c + cr,p + Ti,j−1
n n
[−cz,m ] + Ti,j+1 [cz,p ] + θq̇.
2 r,p 2 2

A.3.3.1 Boundary Conditions

The boundary conditions for the Crank-Nicolson scheme are identical to those of the fully

implicit scheme.
 166

A.3.4 Alternating Direction Explicit (ADE) Scheme

In the ADE scheme, two discretized equations are solved independently. In the upward sweep,

all the lower indexed temperature differences on the right-hand side of Equation A.9 represent the

new time-step, n + 1 and all the higher indexed temperature differences represent the previous

time-step, n. The temperatures calculated in the upward sweep are designated by U :

 n    n 
n+1
Ui,j − Ui,j
n
= cr,p Ui+1,j − Ui,j
n
+ cr,m Ui,j
n+1 n+1
− Ui−1,j + cr,p Ui+1,j − Ui,j
n
  n   (A.52)
n+1 n+1
+ cr,m Ui,j − Ui−1,j + cz,p Ui,j+1 − Ui,j + cz,m Ui,j − Ui,j−1 + θq̇.
n n+1 n+1

This equation can be solved explicitly for the temperature at the new time-step, n + 1:

       
n+1
Ui,j n
= Ui,j 1 − cr,p − cr,p − cz,p − Ui−1,j
n+1 n
cr,m + cr,m + Ui+1,j n+1
cr,p + cr,p − Ui,j−1 [cz,m ]
 −1 (A.53)

+ Ui,j+1 [cz,p ] + θq̇ 1 − cr,m − cr,m − cz,m
n

In the downward sweep, all the higher indexed temperature differences on the right-hand

side of Equation A.9 represent the new time-step, n + 1 and all the lower indexed temperature

differences represent the previous time-step, n. The temperatures calculated in the downward

sweep are designated by V :

  n  
n+1
Vi,j − Vi,j
n
= cr,p Vi+1,j
n+1 n+1
− Vi,j + cr,m Vi,j − Vi−1,j
n n+1
+ cr,p Vi+1,j n+1
− Vi,j
 n    n  (A.54)
n n+1
+ cr,m Vi,j − Vi−1,j + cz,p Vi,j+1 − Vi,j n+1
+ cz,m Vi,j − Vi,j−1 + θq̇.
n

This equation can be solved explicitly for the temperature at the new time-step, n + 1:

       
n+1
Vi,j n
= Vi,j 1 + cr,m + cr,m + F − Vi−1,j
n n+1
cr,m + cr,m + Vi+1,j cr,p + cr,p − Vi,j−1
n
[F ]
  (A.55)
n+1  −1
+ Vi,j+1 [cz,p ] + θq̇ 1 + cr,p + cr,p + cz,p

The final temperature is calculated as follows:

n+1 n+1
n+1
Ui,j + Vi,j
Ti,j = . (A.56)
2
 167

A.3.4.1 Boundary Conditions

The ADE boundary conditions follow the same pattern of higher and lower indices. The

equations below provide only the cases where indices are increasing in a direction normal to the

boundary. Cases where indices are decreasing can be derived by inference.

For interior flux boundaries, the discretized equation for the upward sweep becomes
n+1
Ul−1 − Uln+1  
−kw,m = [hc + hr ] T∞ − Uln+1 + q̇sw , (A.57)
Δwm

which solves for Tln+1 as

kw,m n+1
U + [hc + hr ] T∞ − q̇sw
Δwm l−1
Uln+1 = . (A.58)
kw,m
+ [hc + hr ]
Δwm
The discretized equation for the downward sweep becomes
n − V n+1
Vl−1 l  
−kw,m = [hc + hr ] T∞ − Vln+1 + q̇sw , (A.59)
Δwm

which solves for Tln+1 as

kw,m n
V + [hc + hr ] T∞ − q̇sw
Δwm l−1
Vln+1 = . (A.60)
kw,m
+ [hc + hr ]
Δwm
For exterior flux boundaries, the discretized equation for the upward sweep becomes
n+1
Ul−1 − Uln+1   
−kw,m = hc T∞ − Uln+1 + hr T∞ F ∗ 1/4 − Uln+1 + q̇sw , (A.61)
Δwm

which solves for Tln+1 as

kw,m n+1 
Ul−1 + hc + hr F ∗ 1/4 T∞ − q̇sw
Δwm
Uln+1 = . (A.62)
kw,m
+ [hc + hr ]
Δwm
The discretized equation for the downward sweep becomes
n − V n+1
Vl−1   
−kw,m l
= hc T∞ − Vln+1 + hr T∞ F ∗ 1/4 − Vln+1 + q̇sw , (A.63)
Δwm
 168

which solves for Tln+1 as

kw,m n 
Vl−1 + hc + hr F ∗ 1/4 T∞ − q̇sw
Δwm
Vln+1 = . (A.64)
kw,m
+ [hc + hr ]
Δwm
For zero-flux boundaries, the discretized equation for the upward sweep becomes
n+1
Ul−1 − Uln+1
= 0, (A.65)
Δwm

which solves for Tln+1 as

Uln+1 = Ul−1
n+1
. (A.66)

The discretized equation for the downward sweep becomes

n − V n+1
Vl−1 l
= 0, (A.67)
Δwm

which solves for Tln+1 as

Vln+1 = Vl−1
n
. (A.68)

For constant temperature boundaries, the discretized equation for both sweeps becomes

Tln+1 = Tc . (A.69)

A.3.5 Alternating Direction Implicit (ADI) Scheme

In the ADI scheme, each sub-time-step is represented by a different discretized equation. In

two dimensions, there are two sub-time-steps within each time-step, and the value of θ in Equation

A.18 should be adjusted to reflect the sub-time-step size, Δt/2.

In the first sub-time-step, all the temperatures in the r terms represent the new sub-time-step,

n + 1/2, and all the temperatures in the z represent the previous sub-time-step, n:

 
= [2 − f ] cr,p Ti+1,j − Ti,j + cr,m Ti,j
n+1/2 n+1/2 n+1/2 n+1/2 n+1/2
Ti,j − Ti,j
n
− Ti−1,j
 
n+1/2 n+1/2 n+1/2 n+1/2 (A.70)
+ cr,p Ti+1,j − Ti,j + cr,m Ti,j − Ti−1,j
  n   n 
+ f cz,p Ti,j+1 − Ti,j
n
+ cz,m Ti,j − Ti,j−1
n
+ θq̇.

where f is the implicit weighting factor.

 169

This equation can be rearranged with all the unknown temperatures at the new sub-time-

step, n + 1/2, on the left-hand side of the equation with constant coefficients and all the constant

terms on the right-hand side:

n+1/2   
Ti,j 1 + [2 − f ] cr,p + cr,p − cr,m − cr,m
    (A.71)
+ Ti−1,j [2 − f ] cr,m + cr,m + Ti+1,j [2 − f ] −cr,p − cr,p
n+1/2 n+1/2

n
= Ti,j [1 + f [cz,m − cz,p ]] − Ti,j−1
n n
f [cz,m ] + Ti,j−1 f [−cz,p ] + θq̇

In the second sub-time-step, all the temperatures in the z terms represent the new sub-time-

step, n + 1, and all the temperatures in the r represent the previous sub-time-step, n + 1/2:

 
= f cr,p Ti+1,j − Ti,j + cr,m Ti,j
n+1 n+1/2 n+1/2 n+1/2 n+1/2 n+1/2
Ti,j − Ti,j − Ti−1,j
 
n+1/2 n+1/2 n+1/2 n+1/2 (A.72)
+ cr,p Ti+1,j − Ti,j + cr,m Ti,j − Ti−1,j
 
n+1
+ [2 − f ] cz,p Ti,j+1 n+1
− Ti,j n+1
+ cz,m Ti,j n+1
− Ti,j−1 + θq̇.

This equation can be rearranged with all the unknown temperatures at the new sub-time-

step, n + 1, on the left-hand side of the equation with constant coefficients and all the constant

terms on the right-hand side:

n+1
Ti,j n+1
[1 + [2 − f ] [cz,p − cz,m ]] + Ti,j−1 n+1
[2 − f ] [cz,m ] + Ti,j−1 [2 − f ] [−cz,p ]
  
1 + f cr,m + cr,m − cr,p − cr,p
n+1/2 (A.73)
= Ti,j
n+1/2   n+1/2  
− Ti−1,j f cr,m + cr,m + Ti+1,j f cr,p + cr,p + θq̇

A.3.5.1 Boundary Conditions

The boundary conditions for the ADI scheme are identical to those of either the explicit

scheme or the fully implicit scheme depending on the orientation of the boundary relative to the

implicit direction, r or z, for the sub-time-step. When the implicit direction is perpendicular to

the boundary the conditions are fully implicit, and when the implicit direction is parallel to the

boundary the conditions are explicit.

 170

A.3.6 Steady-State Solution

The steady-state finite difference formulation, used for some of the initialization methods in

Kiva, solves a different continuous equation:

∇ [k∇T ] + q̇ = 0, (A.74)

which in an axisymmetric cylindrical coordinate system resolves to

1 ∂ ∂T ∂ ∂T
rk + k + q̇ = 0. (A.75)
r ∂r ∂r ∂z ∂z

This equation is independent of time and is discretized as:

cr,p [Ti+1,j − Ti,j ] + cr,m [Ti,j − Ti−1,j ] + cr,p [Ti+1,j − Ti,j ] (A.76)
+ cr,m [Ti,j − Ti−1,j ] + cz,p [Ti,j+1 − Ti,j ] + cz,m [Ti,j − Ti,j−1 ] + θq̇ = 0.

Each of the constant coefficients contains θ, which can be divided out of the equation, but it

is kept in the equation for consistency with the other finite difference scheme descriptions.

This equation can be rearranged with all the unknown temperatures on the left-hand side of

the equation with constant coefficients and all the constant terms on the right-hand side:

   
Ti,j cr,m + cr,m + cz,m − cr,p − cr,p − cz,p + Ti−1,j −cr,m − cr,m
  (A.77)
+ Ti+1,j cr,p + cr,p + Ti,j−1 [−cz,m ] + Ti,j+1 [cz,p ] = −θq̇

A.3.6.1 Boundary Conditions

The boundary conditions for the steady-state solution are identical to those of the fully

implicit scheme.
 171

A.4 Solving the Linear System of Equations

The system of equations generated for the set of discretized cells is generalized as

a1,1 T1 + a1,2 T2 + . . . + a1,m Tm + . . . + a1,N TN = b1

a2,1 T1 + a2,2 T2 + . . . + a2,m Tm + . . . + a2,N TN = b2

..
.
(A.78)
an,1 T1 + an,2 T2 + . . . + an,m Tm + . . . + an,N TN = bn
..
.

aN,1 T1 + aN,2 T2 + . . . + aN,m Tm + . . . + aN,N TN = bN

where

• n is equation index,

• m is the cell index,

• an,m is the constant coefficients for the temperature of cell m in equation n,

• Tm is the temperature of cell m [K],

• N is the total number of cells in the problem, and

• bn is the constant term on the right-hand side of equation n

Equation A.78 is represented by a linear system in matrix notation (Ax = b):

⎡ ⎤⎡ ⎤ ⎡ ⎤
⎢ a1,1 a1,2 · · · a1,N ⎥ ⎢ T1 ⎥ ⎢ b1 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ a2,1 a2,2 · · · a2,N ⎥ ⎢ T2 ⎥ ⎢ b2 ⎥
⎢ ⎥⎢ ⎥ = ⎢ ⎥. (A.79)
⎢ . .. .. .. ⎥ ⎢ . ⎥ ⎢ . ⎥
⎢ .. . . . ⎥ ⎢ . ⎥ ⎢ . ⎥
⎢ ⎥⎢ . ⎥ ⎢ . ⎥
⎣ ⎦⎣ ⎦ ⎣ ⎦
aN,1 aN,2 · · · aN,N TN bN
If there is a single unknown per equation, then the system is solved explicitly, and no linear

algebra techniques need be applied. For implicit schemes, where there are potentially multiple

unknowns in an equation, the linear system is solved either directly, using Gaussian elimination, or

iteratively. The ADI method uses a simplified form of Gaussian elimination specific to tridiagonal

matrices (each equation has up to three consecutive non-zero coefficients centered at n, n).
 Appendix B

Mixed Climate Two-Dimensional Approximation Method Test Results

The following results were simulated according to the two-dimensional approximation method

test description provided in Section 5.4.

 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

shapes and two-dimensional approximations

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Square

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure B.1: Percent difference in annual heat transfer between three-dimensional square foundation
173
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

foundations and two-dimensional approximations

R-5 ext. vert.
Cross

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure B.2: Percent difference in annual heat transfer between three-dimensional cross-shaped
174
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

dation shapes and two-dimensional approximations

R-5 ext. vert.

R-5 int. vert.

Narrow-H

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m
175

Figure B.3: Percent difference in annual heat transfer between three-dimensional narrow-H foun-
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

shapes and two-dimensional approximations

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Wide-H

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure B.4: Percent difference in annual heat transfer between three-dimensional wide-H foundation
176
 Appendix C

Cold Climate Two-Dimensional Approximation Method Test Results

The following results were simulated according to the two-dimensional approximation method

test description provided in Section 5.4.

 178

Slab Crawlspace Basement

Wall Floor
20%

A=P = 1.5 m
10%

0%

-10%
Deviation from mean heat flux

-20%

20%

A=P = 4.6 m
10%

0%

-10%

-20%

20%

A=P = 13.7 m
10%

0%

-10%

-20%
Circle Semi-Infinite Rectangle Cross
Square Wide-H Narrow-H

Figure C.1: Comparison of heat flux from a range of uninsulated foundation shapes in the cold
climate
 Reduction in heat flux relative to uninsulated foundations

40%
40%
40%

0%
20%
0%
20%
0%
20%

60%
60%
60%

80%
80%
80%

100%
100%
100%
R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

R-5 int. whole-slab

R-20 ext. hor.

Slab Floor
Cross

R-20 ext. vert.

Square

foundation shapes in the cold climate

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Wide-H
Narrow-H

R-5 ext. vert.

R-5 int. vert.

R-5 int. vert. & hor.

Circle

R-20 ext. vert.

R-20 int. vert.

Crawlspace Wall

R-20 int. vert. & hor.

Semi-Infinite Rectangle

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

R-20 half-wall ext.

R-20 full-wall ext.

Basement Wall

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure C.2: Heat flux reduction relative to uninsulated foundations in mixed climate for a range of
179
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Square

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

shapes and two-dimensional approximations in the cold climate

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure C.3: Percent difference in annual heat transfer between three-dimensional square foundation
180
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Cross

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle

foundations and two-dimensional approximations in the cold climate

R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure C.4: Percent difference in annual heat transfer between three-dimensional cross-shaped
181
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

R-5 int. vert.

Narrow-H

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

dation shapes and two-dimensional approximations in the cold climate

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure C.5: Percent difference in annual heat transfer between three-dimensional narrow-H foun-
182
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Wide-H

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

shapes and two-dimensional approximations in the cold climate

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m
183

Figure C.6: Percent difference in annual heat transfer between three-dimensional wide-H foundation
 Appendix D

Hot Climate Two-Dimensional Approximation Method Test Results

The following results were simulated according to the two-dimensional approximation method

test description provided in Section 5.4.

 185

Slab Crawlspace Basement

Wall Floor
20%

A=P = 1.5 m
10%

0%

-10%
Deviation from mean heat flux

-20%

20%

A=P = 4.6 m
10%

0%

-10%

-20%

20%

A=P = 13.7 m
10%

0%

-10%

-20%
Circle Semi-Infinite Rectangle Cross
Square Wide-H Narrow-H

Figure D.1: Comparison of heat flux from a range of uninsulated foundation shapes in the hot
climate
 Reduction in heat flux relative to uninsulated foundations

40%
40%
40%

0%
20%
0%
20%
0%
20%

60%
60%
60%

80%
80%
80%

100%
100%
100%
R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

R-5 int. whole-slab

R-20 ext. hor.

Slab Floor
Cross

R-20 ext. vert.

Square

R-20 int. vert.

of foundation shapes in the hot climate

R-20 int. hor.

R-20 int. whole-slab

Wide-H
Narrow-H

R-5 ext. vert.

R-5 int. vert.

R-5 int. vert. & hor.

Circle

R-20 ext. vert.

R-20 int. vert.

Crawlspace Wall

R-20 int. vert. & hor.

Semi-Infinite Rectangle

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

R-20 half-wall ext.

R-20 full-wall ext.

Basement Wall

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure D.2: Heat flux reduction relative to uninsulated foundations in mixed climate for a range
186
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Square

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

shapes and two-dimensional approximations in the hot climate

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure D.3: Percent difference in annual heat transfer between three-dimensional square foundation
187
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Cross

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle

foundations and two-dimensional approximations in the hot climate

R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m
188

Figure D.4: Percent difference in annual heat transfer between three-dimensional cross-shaped
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

R-5 int. vert.

Narrow-H

R-5 int. vert. & hor.

R-20 ext. vert.

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

dation shapes and two-dimensional approximations in the hot climate

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m
189

Figure D.5: Percent difference in annual heat transfer between three-dimensional narrow-H foun-
 Deviation from three-dimensional annual heat transfer

-5%
0%
5%
-5%

20%
0%
5%
-5%

20%
0%
5%
20%

-20%
-15%
-10%
-20%
-15%
-10%
-20%
-15%
-10%

10%
15%
10%
15%
10%
15%
Uninsulated

R-5 ext. hor.

R-5 ext. vert.

R-5 int. vert.

R-5 int. hor.

Circle
R-5 int. whole-slab
Slab
R-20 ext. hor.

R-20 ext. vert.

R-20 int. vert.

R-20 int. hor.

R-20 int. whole-slab

Infinite Rectangle
Uninsulated

R-5 ext. vert.

Wide-H

R-5 int. vert.

R-5 int. vert. & hor.

R-20 ext. vert.

shapes and two-dimensional approximations in the hot climate

Crawlspace

R-20 int. vert.

Rounded Rectangle
R-20 int. vert. & hor.

Uninsulated

R-5 half-wall ext.

R-5 full-wall ext.

R-5 full-wall int.

Boundary Layer
R-20 half-wall ext.
Basement

R-20 full-wall ext.

R-20 full-wall int.

A=P = 13.7 m A=P = 4.6 m A=P = 1.5 m

Figure D.6: Percent difference in annual heat transfer between three-dimensional wide-H foundation
190
 Appendix E

Comparisons of Minimum and Maximum Foundation Heat Transfer

The following figures compare simulated annual minimum and maximum heat transfer rates

for each two-dimensional approximation method to corresponding three-dimensional simulation

results. Results were simulated according to the two-dimensional approximation method test de-

scription provided in Section 5.4.

 192

5 5
Circle Infinite Rectangle
0 0
Three-dimensional annual minimum heat transfer [kW]

−5 −5

−10 −10

−15 −15

−20 −20

−25 MAD = 44.8% −25 MAD = 48.3%

MBD = 7.2% MBD = 3.6%
−30 −30
−30 −25 −20 −15 −10 −5 0 5 −30 −25 −20 −15 −10 −5 0 5

5 5
Rounded Rectangle Boundary Layer
0 0 Adjustment

−5 −5

−10 −10

−15 −15

−20 −20

−25 MAD = 39.8% −25 MAD = 49.8%

MBD = 2.5% MBD = 2.4%
−30 −30
−30 −25 −20 −15 −10 −5 0 5 −30 −25 −20 −15 −10 −5 0 5

Approximated, two-dimensional annual minimum heat transfer [kW]

Figure E.1: Comparison of annual minimum heat transfer between three-dimensions and two-
dimensional approximations for mixed climate simulations
 193

60 60
Circle Infinite Rectangle
50 50
Three-dimensional annual maximum heat transfer [kW]

40 40

30 30

20 20

10 MAD = 7.2% 10 MAD = 4.1%

MBD = 7.1% MBD = 3.3%
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60

60 60
Rounded Rectangle Boundary Layer
50 50 Adjustment

40 40

30 30

20 20

10 MAD = 4.4% 10 MAD = 2.4%

MBD = 4.4% MBD = -0.4%
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60

Approximated, two-dimensional annual maximum heat transfer [kW]

Figure E.2: Comparison of annual maximum heat transfer between three-dimensions and two-
dimensional approximations for mixed climate simulations
 194

30 30
Circle Infinite Rectangle
25 25
Three-dimensional annual minimum heat transfer [kW]

20 20

15 15

10 10

5 MAD = 11.0% 5 MAD = 5.6%

MBD = 11.0% MBD = 2.5%
0 0
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30

30 30
Rounded Rectangle Boundary Layer
25 25 Adjustment

20 20

15 15

10 10

5 MAD = 5.1% 5 MAD = 4.0%

MBD = 4.1% MBD = -1.2%
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30

Approximated, two-dimensional annual minimum heat transfer [kW]

Figure E.3: Comparison of annual minimum heat transfer between three-dimensions and two-
dimensional approximations for cold climate simulations
 195

120 120
Circle Infinite Rectangle
100 100
Three-dimensional annual maximum heat transfer [kW]

80 80

60 60

40 40

20 MAD = 9.0% 20 MAD = 4.9%

MBD = 9.0% MBD = 3.8%
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120

120 120
Rounded Rectangle Boundary Layer
100 100 Adjustment

80 80

60 60

40 40

20 MAD = 4.8% 20 MAD = 2.7%

MBD = 4.8% MBD = -0.1%
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120

Approximated, two-dimensional annual maximum heat transfer [kW]

Figure E.4: Comparison of annual maximum heat transfer between three-dimensions and two-
dimensional approximations for cold climate simulations
 196

0 0
Circle Infinite Rectangle
−10 −10
Three-dimensional annual minimum heat transfer [kW]

−20 −20

−30 −30

−40 −40

−50 MAD = 7.4% −50 MAD = 4.3%

MBD = 7.3% MBD = 3.8%
−60 −60
−60 −50 −40 −30 −20 −10 0 −60 −50 −40 −30 −20 −10 0

0 0
Rounded Rectangle Boundary Layer
−10 −10 Adjustment

−20 −20

−30 −30

−40 −40

−50 MAD = 4.7% −50 MAD = 2.7%

MBD = 4.7% MBD = 0.1%
−60 −60
−60 −50 −40 −30 −20 −10 0 −60 −50 −40 −30 −20 −10 0

Approximated, two-dimensional annual minimum heat transfer [kW]

Figure E.5: Comparison of annual minimum heat transfer between three-dimensions and two-
dimensional approximations for hot climate simulations
 197

25 25
Circle Infinite Rectangle
20 20
Three-dimensional annual maximum heat transfer [kW]

15 15

10 10

5 5

0 MAD = 55.0% 0 MAD = 45.7%

MBD = -28.3% MBD = -18.4%
−5 −5
−5 0 5 10 15 20 25 −5 0 5 10 15 20 25

25 25
Rounded Rectangle Boundary Layer
20 20 Adjustment

15 15

10 10

5 5

0 MAD = 43.8% 0 MAD = 25.9%

MBD = -17.3% MBD = -3.1%
−5 −5
−5 0 5 10 15 20 25 −5 0 5 10 15 20 25

Approximated, two-dimensional annual maximum heat transfer [kW]

Figure E.6: Comparison of annual maximum heat transfer between three-dimensions and two-
dimensional approximations for hot climate simulations