On the Microstructure, Morphology, and Mechanics of Thin Wall

Laser Powder Bed Fusion Structures

by

Paul Paradise

A Dissertation Presented in Partial Fulfillment

of the Requirements for the Degree
Doctor of Philosophy

Approved October 2022 by the

Graduate Supervisory Committee:

Dhruv Bhate, Chair

Bruno Azeredo
Yang Jiao
Nikhilesh Chawla

ARIZONA STATE UNIVERSITY

December 2022
 ABSTRACT

Applications such as heat exchangers, surface-based cellular structures, rotating

blades, and waveguides rely on thin metal walls as crucial constituent elements of the

structure. The design freedom enabled by laser powder bed fusion has led to an interest in

exploiting this technology to further the performance of these components, many of

which retain their as-built surface morphologies on account of their design complexity.

However, there is limited understanding of how and why mechanical properties vary by

wall thickness for specimens that are additively manufactured and maintain an as-printed

surface finish. Critically, the contributions of microstructure and morphology to the

mechanical behavior of thin wall laser powder bed fusion structures have yet to be

systematically identified and decoupled. This work focuses on elucidating the room

temperature quasi-static tensile and high cycle fatigue properties of as-printed, thin-wall

Inconel 718 fabricated using laser powder bed fusion, with the aim of addressing this

critical gap in the literature. Wall thicknesses studied range from 0.3 - 2.0 mm, and the

effects of Hot Isostatic Pressing are also examined, with sheet metal specimens used as a

baseline for comparison. Statistical analyses are conducted to identify the significance of

the dependence of properties on wall thickness and Hot Isostatic Pressing, as well as to

examine correlations of these properties to section area, porosity, and surface roughness.

A thorough microstructural study is complemented with a first-of-its-kind study of

surface morphology to decouple their contributions and identify underlying causes for

observed changes in mechanical properties. This thesis finds that mechanical properties

i
in the quasi-static and fatigue framework do not see appreciable declines until specimen

thickness is under 0.75 mm in thickness. The added Hot Isostatic Pressing heat treatment

effectively closed pores, recrystallized the grain structure, and provided a more

homogenous microstructure that benefits the modulus, tensile strength, elongation, and

fatigue performance at higher stresses. Stress heterogeneities, primarily caused by surface

defects, negatively affected the thinner specimens disproportionately. Without the use of

the Hot Isostatic Pressing, the grain structure remained much more refined and benefitted

the yield strength and fatigue endurance limit.

ii
DEDICATION

David Paradise

Drew Burns

iii
 ACKNOWLEDGMENTS

The greatest appreciation is reserved for my advisor, Professor Dhruv Bhate, for

his unrelenting patience, time, and positive spirit – thank you.

A lot of work went into this project over the course of 4 years from many people

within and external to my research group. From our research group, I want to personally

thank Shawn Clonts for his dedication in the blue light scanning of AM specimens,

Anushree Saxena and Andrew Sarrasin for their hard work supporting fatigue testing,

Samuel Temes for rapidly finishing the high temperature testing, Mandar Shinde for the

design of all the specimens, and Cameron Noe for facilitating a dozen of successful and

insightful 3D specimen builds.

I would also like to extend my thanks to the members of my thesis committee, Profs.

Bruno Azeredo, Nikhilesh Chawla and Yang Jiao for their time and support of my doctoral

work. Additional thanks are extended to Prof. Nikhilesh Chawla and Dr. Sridhar Niverty

for their help with XCT scanning and to Prof. Bruno Azeredo and Dr. Stanislau Niauzorau

for assistance in striation measurements of fatigue specimens. I want to thank America

Makes and Air Force Research Laboratory for supporting this work. The guidance and

assistance from Dr. Thomas Broderick and Dr. Mark Benedict of the AFRL was essential

for a successful project with America Makes.

iv
 TABLE OF CONTENTS

Page

LIST OF TABLES ................................................................................................................ viii

LIST OF FIGURES .................................................................................................................ix

CHAPTER

1 INTRODUCTION .................................................................................................. 1

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

1.2 Literature Review .............................................................................................. 3

1.3 Research Gaps.................................................................................................... 6

1.4 Thesis Objectives ............................................................................................... 7

2 RESEARCH METHODOLOGY ........................................................................... 9

2.1 Overall Approach............................................................................................... 9

2.2 Preliminary Builds and Lessons Learned ....................................................... 11

2.3 Specimen Design ............................................................................................. 13

2.4 Heat Treatment ................................................................................................ 18

3 THE QUANTIFICATION AND QUALIFICATION OF POROSITY IN

METAL ADDITIVE MANUFACTURING AS A FUNCTION OF WALL

THICKNESS ......................................................................................................... 21

3.1 Introduction ...................................................................................................... 21

3.2 Density Metrologies ........................................................................................ 22

3.3 Measurement Capability Analysis of the Archimedes Method ..................... 31

3.4 Process Sensitivity of the Archimedes Method .............................................. 39

v
CHAPTER Page

3.5 The Results of the Archimedes Density Method on the Final Test Specimens

................................................................................................................................ 43

3.6 Conclusions ...................................................................................................... 45

4 THE CRITICAL ROLE OF MEASURING CROSS SECTIONAL AREA OF

LASER POWDER BED FUSION THIN WALL STRUCTURES WITH AN

AS-PRINTED SURFACE .................................................................................... 48

4.1 Introduction ...................................................................................................... 48

4.2 Methods............................................................................................................ 52

4.3 Results .............................................................................................................. 66

4.4 Discussion ........................................................................................................ 72

4.5 Conclusions ...................................................................................................... 78

5 DECOUPLING THE CONTRIBUTIONS OF MORPHOLOGY AND

MICROSTRUCTURE: A MESO-SCALE SIZE EFFECT STUDY .................. 81

5.1 Introduction ...................................................................................................... 81

5.2 Methods............................................................................................................ 86

5.3 Results .............................................................................................................. 93

5.4 Analysis: Effects of Thickness and HIP .......................................................112

5.5 Discussion ......................................................................................................124

5.6 Conclusions ....................................................................................................127

6 THE EVOLVING BEHAVIOR OF FATIGUE WITH CHANGING

MICROSTRUCTURE AND SECTION AREA OF THIN-WALL

ADDITIVELY MANUFACTURED METAL .................................................. 129

vi
CHAPTER Page

6.1 Introduction ....................................................................................................129

6.2 Methods..........................................................................................................131

6.3 Results ............................................................................................................133

6.4 Discussion ......................................................................................................158

6.5 Conclusions ....................................................................................................163

7 SUMMARY AND CONCLUSIONS................................................................. 165

REFERENCES ....................................................................................................................169

vii
 LIST OF TABLES

Table Page

1. The Chemical Composition of Inconel Alloy 718 ....................................................... 6

2. Measurement Capability Analysis Summary ............................................................. 39

3. A Table to Qualitatively Grade Each of the 4 Metrologies for Purposes of

Calculating Cross-Sectional Area. The Greater the Circle is Filled in, the Better it

Achieves the Respective Metric. ................................................................................ 79

4. The Average Values for Modulus, Yield, UTS, True UTS, and Elongation by

Specimen Manufacturing, Heat Treatment, and Nominal Thickness with the

Standard Deviation in Parentheses. ..........................................................................104

5. Parameters for a Thickness-Dependent Bilinear Elastic-Plastic Material Model of

LPBF IN-718 for HIP and No-HIP Conditions .......................................................127

viii
 LIST OF FIGURES

Figure Page

1. Intricate AM Microstructure with Lattices of Varying Thickness. ............................... 2

2. A Schematic Diagram of a Thin-wall Plate Heat Exchanger. ....................................... 2

3. (A) The Workflow of Designing, Manufacturing, and the Necessary Heat Treatments

Required to Achieve Optimal Microstructure and Performance in the Test Specimens. (B)

The Primary Characterization Tools Used in this Study. (C) The Three Mechanical Testing

Machines Used in this Study for the Room Temperature Quasi-static Testing, High-

Temperature Quasi-static Testing, and the High Cycle Fatigue Testing. ............................. 11

4. Part Cross-Sections (A) Were Used to Determine the Scan Strategy that Provided the

Part With the Greatest Dimensional Correlation to the Cad Model (B) ............................... 12

5. Finite Element Analysis Simulation of the Quasi-static Specimen Under Tensile

Loading, Showing Localization of Stress Concentration, which was Ensured to be Under

Acceptable Limits .................................................................................................................. 14

6. The Room Temperature Quasi-Static Specimen Profile ............................................ 14

7. The High Cycle Fatigue Test Specimen Profile ......................................................... 14

8. (A) The Build Setup Before Sending the CAD File to the Concept Laser LPBF

Printer. (B) The First Build While the Specimens are Still on the Build Plate and Prior to

Receiving a Stress Relief Cycle.............................................................................................. 16

9. Alongside the AM Specimens, Sheet Metal Specimens were Waterjet Cut out of

Inconel 718 Sheet Metal Stock. ............................................................................................. 17

10. A Graph of the Four Major Heat Treatments that were Used in the Study Plotted on

a Temp. vs. Time Graph. ....................................................................................................... 19

ix
Figure Page

11. Test Data Showing the Effects of the Different Heat Treatments on the Stress-strain

Response of the AM Parts as They Proceed Through the Different Heat Treat Cycles. SA is

Shortened for Solutionize and Age. The Stress Relief and the HIP Steps Reduce the

Strength and Increase the Ductility, While the ageing step is Increases Strength. .............. 20

12. A Polished Cross Section of One of the Thin Walls Manufactured as Part of this

Study with the Build Direction Pointing into the Page(A), the Purple Arrows Pointing to

Pores that Show a Lack of Fusion or Partially Melted Powder Spheres While Most of the

Pores Show Keyholing Porosity. After the Image has been Trimmed Down to Remove the

Edges (B), ImageJ is used to Threshold and Quantify the amount of porosity. ................... 24

13. A 2mm Thick AM Cross-section with the Build Direction Pointing up Showing a

Large Number of Pores Generated just under the Sample Surface (the Left and Right sides

of the Micrograph). When the Laser Scan Path is Overlaid on top of the Micrograph, it is

Shown that the Majority of the Generated Porosity Takes Place near the Interior Raster

Turns and/or in Between the Contour Boundary and the Interior Raster Scan Lines. ........ 25

14. A 3D Reconstruction of the 1.5 mm Specimen that was CT Scanned for this Study.

The Generated Porosity is Unevenly Distributed with more Forming just under the As-

printed Surface. ...................................................................................................................... 26

15. X-ray Tomography was used to Quantify the Distribution of Porosity in the Non-

HIP’ed Specimens Compared to the Average Value. It was Shown that most of the Porosity

is Found near the Specimen Edge. ........................................................................................ 27

x
Figure Page

15. X-ray Tomography was Used to Quantify the Distribution of Porosity in the Non-

HIP’ed Specimens Compared to the Average Value. It was Shown that Most of the

Porosity is Found Near the Specimen Edge. ......................................................................... 27

16. Two Selected .tif Images from the Resultant .tif Stack Collected Through the X-ray

Tomography that show a Relatively Greater Amount of Porosity than Average. A 1.5 mm

(A) and a .035 (mm) Specimen are Shown with the As-printed Surface at the Top and

Bottom of the Image. .............................................................................................................. 28

17. The Archimedes Density Measurement Setup for the Collection of Relative

Densities of the 10 mm DOE Cubes as well as the Dogbone Specimens. ............................ 29

18. LPBF Specimens Prior to Removal from Build Plate: (A) 10 mm Cubes

Commonly used in Process Development and for Archimedes Density Estimation, (B) Thin

Wall Tensile test Specimens used for Quantifying Mechanical Properties. ......................... 31

19. Density Computed for a Standard 10g Mass (Shown in inset) from 16 Successive

Measurements. ........................................................................................................................ 33

20. Mean (left) and Standard Deviation (Right) from 8 Measurements made of a Single

Specimen at each of 7 Different Thicknesses. ....................................................................... 34

21. Repeatability Data from 30 Successive Calculations of Density of (A) a Single 10

mm Cube, and (B) a Single 0.3 mm Thin Wall Specimen Showing Short Term ................ 35

22. Repeatability Data from 30 Successive Calculations of Density of (A) a Single 10

mm Cube, and (B) a Single 0.3 mm Thin Wall Showing Short Term Stability ................... 37

xi
Figure Page

23. Archimedes Density Estimated for Thin Wall Specimens Ranging in Thickness

from 0.3 mm to 2 mm. Statistically Significant Differences were Detected Between Thick

Specimens (>0.75mm) and Thin ones (0.3-0.5mm). ............................................................. 41

24. Effect of Hot Isostatic Pressing (HIP) on Archimedes Density Across Seven

Different Specimen Thicknesses, Showing (A) Higher means after HIP, as well as (B)

Significant Reduction in Standard Deviation at Higher Thicknesses. .................................. 42

25. Density Estimates for 100 as-printed 2 mm Thick Specimens Across the Build

Platform, Suggesting Low Densities at the Front and Rear................................................... 43

26. The Absolute Density Values of the Room Temperature, Quasi-static Tensile Test

Specimens for the 2 Different Heat Treatments Across all the Different Thicknesses Tested.

The HIP Heat Treatment Significantly Reduces the Standard Deviation of the test

Specimens for the Thicker Parts. ............................................................................................ 44

27. The Absolute Density Values of the Fatigue Tensile Test Specimens for the 2

Different Heat Treatments Across all the Different Thicknesses Tested. The HIP Heat

Treatment Significantly Reduces the Standard Deviation of the Test Specimens ................ 45

28. An SEM Image of the Surface of an Am Part Shows the Rough Surface Due to

Partially Melted Powder Spheres and Melt Pool Paths. (B) a Cross-section of a Separate,

Polished Am Specimen Reveals Rounded Edges Despite Sharp Corners in the Cad File

Used for Printing - This Specimen Was Sectioned Such That into the Page Is Parallel to the

Build Direction... ..................................................................................................................... 51

29. The Final Specimen Design for this Study with 7 Specimen Thicknesses Ranging

from 0.3 mm to 2.0 mm (all Dimensions in mm) .................................................................. 54

xii
Figure Page

29. The Final Specimen Design for this Study with 7 Specimen Thicknesses Ranging

from 0.3 mm to 2.0 mm (all Dimensions in mm) .................................................................. 54

30. (A) Build Layout of 140 Specimens; (B) with the Final Print Before Receiving a

Stress Relief Heat Treat Cycle................................................................................................ 56

31. (A) Cross-sectional Measurement Using a Scanning Microscope to Compare

Measured Value of Thickness Against Nominal Design, Shown Here for a Specimen with a

6mm X 1.5mm Cross Section, and (B) Dependence of Error in Thickness on Contour-edge

Offset for Two Different Raster Angles ................................................................................ 56

32. The Four Metrologies Used in This Study, Pictured with the Setup Used to Study

the Specimen: (A) Standard Micrometer with a 6 Mm Platen Diameter, (B) Point

Micrometer with a 15-degree Measuring Point, (C) Blue Light 3d Scanner, and (D) X-ray

Tomography. ........................................................................................................................... 59

33. A Graphical Representation of the Workflow of Analyzing the Blue Light Data of

a 1.5 Mm Thick Specimen: (A) the Specimen Held in the Specimen Holder, (B) a 30 Μm

Slice of the Blue Light Data Is First Projected onto (C) a Plane Containing the Width and

Thickness (D) a Bounded Area Is Defined about the Points at Every 30 Μm Slice to Obtain

(E) a Plot of Cross-sectional Area Vs. Specimen Length. ..................................................... 63

34. (A) Raw .Tiff Image of the Ct Data, (B) a Binary Image Is Created in Imagej after

Thresholding and a Text File of the Binary Image Is Exported to Matlab, (C) Matlab Code

Was Written to Find the Edges of the Cross-section from the Binary File, (D) Iterating

Through the Entire Sample Yields a Full Point Cloud of the Surface Data.......................... 64

xiii
Figure Page

35. Cross-sectional Area Percent Differences from CT- exact, as a Function of

Specimen Thickness. .............................................................................................................. 67

36. The Cross-sectional Areas of the Middle of the Gauge Section for the (A) .3 mm

and (B) 2.0 mm Specimen for the Blue Light and Ct-30 Data. The Average Values of

Thickness and Width Measured with the Standard Micrometer Are Also Shown for the

Respective Specimens. Images Not to Scale.......................................................................... 68

37. Stress vs. Strain for the (A) 0.3 mm and (B) 2.0 mm Thick Specimens for the 6

Measurement Techniques; (C) Magnified Version of 2.0 mm Specimens’ Stress-strain

Response…. ............................................................................................................................ 70

38. The (a) Elastic Modulus, (B) Engineering Yield Strength, and (C) Engineering

Ultimate Tensile Strength for the 7 Different Specimens Vs Thickness for Each of the 6

Different Measurement Techniques. ...................................................................................... 72

39. Schematic Showing Contact Area Comparisons Between (A) Standard Micrometer

Platens, and (B) Point Micrometer Tips. ............................................................................... 73

40. Two SEM Images of the Surface of an As-printed LPBF Specimen After the use of

a Point Micrometer to Measure Thickness. Red Circles Showing Approximate Contact

Area of Point Micrometer Tip. .............................................................................................. 74

41. (A) Comparison of Differences Between the Blue Light Scan and the Point

Micrometer Relative to Ct-exact, as a Function of Nominal Thickness, (B) Correlation Plot

of Blue Light and Point Micrometer Section Area Computations, and (C) Delta Chart

Showing Deviations from Nominal for Both Metrologies .................................................... 75

xiv
Figure Page

42. (A) the Aligned Point Cloud Data of the Ceramic Thickness Calibrant Before and

after Scanning – the Additional Data in the Width Direction Is an Artifact of Data

Trimming; (B) the Projection of the Thickness with and Without Spray, Clearly Showing a

Slight Increase Due to Spray. …. ........................................................................................... 77

43. Stack of Blue Light Scan Measurements for Three 30 µm Slices ............................ 78

44. Coefficient Variation (%) for the Section Area Estimation for All Slices as a

Function of Nominal Thickness ............................................................................................. 78

45. A Flow Chart Describing the Different Potential Variables That Will Affect the

Thickness Dependance on the Meso-scale Am Parts. The Main Contributing Factors Can

Come from Printing and Process Artifacts, the External Structure of the Specimen, and the

Internal Structure of the Specimen. ........................................................................................ 82

46. A Literature Review of Additively Manufactured Inconel 718 for LPBF, DLD

(Direct Laser Deposition), and EBM (Electron Beam Melting). All References Used

Specimens That Were Machined ............................................................................................ 85

47. (A) OBJ Surface Body with a Closeup and (B) 3D Solid Body with a Closeup. .... 83

48. Applied Boundary Conditions. .................................................................................. 91

49. The Absolute Density Values of the AM Parts for the 2 Different Heat Treatments

Across all of the Different Thicknesses Tested. The HIP Heat Treatment Significantly

Reduces the Standard Deviation of the Test Specimens for the Thicker Parts. ................... 92

50. The Sa Values for the HIP and Non-HIP specimens. All Surface Roughness Values

are in Close Approximation, but in General, the HIP Parts Have a Greater Surface

Roughness. .............................................................................................................................. 95

xv
Figure Page

50. The Sa Values for the HIP and Non-HIP specimens. All Surface Roughness Values

are in Close Approximation, but in General, the HIP Parts have a Greater Roughness. ...... 95

51. (A,B,C) the As-built, Non-hip, and Hip Grain Structure, Respectively, for the Build

Direction Pointing Up. (E,F,G) the As-built, Non-hip, and Hip Grain Structure,

Respectively, for the Build Direction Pointing into the Page. Significant Grain

Recrystallization Can Be Seen after the Hip Cycle, While the Non-hip Specimens Retain

More of the As-built Porosity and Grain Size (Distribution and Aspect Ratio). ................. 98

52. Micrographs for the Non-hip and Hip Specimens That Are .3 Mm in Thickness. (A

and B) Show the Grain Sizes and Aspect Ratio for the Non-hip Specimens with the Build

Direction Pointing up and into the Page, Respectively. (C and D) Show the Grain Sizes and

Aspect Ratio for the Hip Specimens with the Build Direction Pointing up and into the Page,

Respectively. The Porosity Generated in the Non-hip Specimen Is Typically Closer to the

Center of While the Porosity Is Significantly Reduced/Eliminated in the Hip Specimen.

Significant Annealing and Homogenization Can Be Seen in the Hip Specimen. ............... 99

53. For All Images, the Build Direction Is Pointing to the Right. The Non-hip

Microstructure for Increasing Magnification (A,B,C). The Hip Microstructure for

Increasing (E,F,G). The Hip Step Shows a Significant Amount of Precipitate Dissolution

along the Grain Boundaries and an Increase in Annealing Twins. .....................................100

xvi
Figure Page

54. For All Images, the Build Direction Is into the Page. The Non-HIP Microstructure

for Increasing Magnification (A,B,C). The HIP Microstructure for Increasing (E,F,G). The

HIP Step Shows a Significant Amount of Precipitate Dissolution along the Grain

Boundaries and an Increase in Annealing Twins. A Tighter, Cellular Structure Can Be Seen

in (C) Which Can Be Attributed to Increase Strengths in AM Parts...................................101

55. (A) the Stress-strain Curve for 28 Non-HIP and HIP Specimens (2 Curves per

Thickness and Heat Treatment) with the Red Solid Line Representing the HIP Specimens

and the Black, Dashed Lines Representing the Non- HIP Specimens. The Black Inset Box

Represents the Stress-strain Location for (B) the Enveloping Stress-strain Flow Curve

Bounds Are Plotted up to 6% Strain.. ..................................................................................103

56. The Macroscopic Fracture Surface for A .3 mm Non- HIP and HIP Specimen (A)

and (B), Respectively. A Closeup of the Fracture Surface for the Non- HIP and HIP

Specimen (C) and (D), Respectively. ...................................................................................107

57. A Series of Fractographic Images for a 2 mm Non- HIP ((A), (C), (E), and (G)) and

HIP Specimen ((B), (D), (G), and (H)). Series (a) and (B) Show the Macroscopic Fracture

Surface Where the Blue, Dashed Line Indicates Approximate Location of the Fracture

Profile Shown in (E) and (F). Series (C) and (D) Show a Closeup of the Fracture Surface.

Series (G) and (H) Shows a Closeup of the Fracture Profile. ..............................................108

58. (A) the FEA Analysis for the (a) 2.0, (B) .75 and the (C) .3 mm Thick Specimen.

Stress Heterogeneities Arise as the Specimen Thickness Decreases Due to Surface

Morphology.. .........................................................................................................................112

xvii
Figure Page

58. (A) The FEA analysis for the (A) 2.0, (B) 0.75 and the (C) 0.3 mm thick specimen.

Stress Heterogeneities Arise as the Specimen Thickness Decreases Due to Surface

Morphology.. .........................................................................................................................112

59. The Mechanical Behavior of the AM Specimens for the Four Metrics Examined in

This Study, Shown with Connecting Letters Indicating Statistically Equivalency Between

Means: (A) Engineering Yield Strength, (B) Engineering Ultimate Tensile Strength, (C)

Young’s Modulus, and (D) Elongation ................................................................................114

60. The Coefficient of Variation as a Function of Thickness and HIP Condition for

each of the Four Metrics Examined......................................................................................114

61. The Correlation of Young's Modulus (GPa) Vs Density (g/cc) for the Non-hip

Specimens Only... .................................................................................................................116

62. The Calculated Modulus Values as a Function of Thickness for the AM and Sheet

Metal Parts.............................................................................................................................116

63. The Correlation of Engineering Yield (MPa) vs Density (g/cc) for the Non-HIP

Specimens Only. No Strong Correlation or Trend Can Be Across the Thicknesses,

Implying That the Amount of Porosity Generated in the Parts of This Study (>98% Dense)

Has No Discernable Impact on the Final Yield Value.........................................................119

64. The Correlation Between UTS (Mpa) and Density (g/cc) Is Plotted for the Non-HIP

Specimens. There Is a Strong, Positive Signal Between the Two Variables for the Thinner

Specimens (<.75 Mm), Demonstrating That the Thickness Effects for the UTS Are Partly

Attributed to the Higher Density Parts for Thicker Specimens. ..........................................120

xviii
Figure Page

65. The DIC Results Showing the Evolution of Necking Behavior in the Specimens

Across Different Thicknesses. As the Specimen Thickness Is Increased, the Specimen Can

Achieve Greater Necking and Ductility Partially Due to Specimen Geometry. .................122

66. The Percent Elongation Vs. The Percent Necking for the Non- HIP and HIP

Specimens Show a Strong Positive Trend. Samples That Are Capable of a Greater Overall

Elongation Also Demonstrate a Greater Amount of Necking. The HIP Specimens

Generally Have a Greater Amount of Total Elongation and Necking. ...............................123

67. The Correlation Between Specimen Elongation and Density Is Plotted for the Non-

HIP Specimens. There Is a Strong, Positive Signal Between the Two Variables for the

Thicker Specimens (>.35 mm), Demonstrating That the Thickness Effects for the

Elongation Are, in Part, Due to the Greater Density Achieved by the Thicker. .................123

68. The Total Elongation Until Failure vs. Specimen Thickness for the AM and Sheet

Metal Specimens. ..................................................................................................................124

69. Model Fits Relative to Experimental Data for (A) Yield Strength, (B) Ultimate

Tensile Strength, and (C) Plastic Strain at Failure ...............................................................126

70. An Overview of the Literature Found for AM Inconel 718 Parts ...........................130

71. The Absolute Density and Standard Deviation Values of the AM Parts for the 2

Different Heat Treatments Across All of the Different Thicknesses Tested. The HIP Heat

Treatment Significantly Reduces the Standard Deviation of the Test Specimens for the

Thicker Parts........ .................................................................................................................134

72. The Sa and Sv Values for the HIP and Non-HIP Specimens ..................................136

xix
Figure Page

73. The non-HIP and HIP HCF Fatigue Results. Increasing Marker Size and Line

Thickness Corresponds to Greater Nominal Thickness of the Specimen. The Lines are the

Basquin Slopes of the Fatigue Data Set. .............................................................................137

74. The Non-HIP Set of Data for all Thicknesses Separated by Color. .......................138

75. The HIP Set of Data for all Thicknesses Separated by Color. ................................139

76. The Log-Log Plot of Max Stress vs the Total Number of Cycles to Failure for the 2

Extreme Thicknesses for the AM and the Sheet Metal Parts. .............................................141

77. A Closeup of the Grain Structure for a Non-HIP (A) and HIP (B) Specimen. The

Non-HIP Specimen Shows a Very Tight Grain Structure at the Surface of the Specimen,

While the HIP Specimen Shows an Increase of Annealing Twins with Larger and More

Uniform Sized Grains ...........................................................................................................142

78. An Overview of the Main Sections of a Fatigue Specimens Fractographic Surface

with a Sheet Metal Specimen Used as an Example. ............................................................143

79. A Macroscopic SEM Image of a 2 mm AM No-HIP Specimen that Failed at ~171k

Cycles. An Appreciable Amount of Porosity can be Seen Across the Sample Surface

(Specimen N15). ….. ............................................................................................................144

80. A Closeup of Figure 79 Showing the Locations of the Different Areas that Were

Found Through SEM Imaging (Specimen N15)..................................................................145

81. A Closeup of the Fracture Initiation Site (Specimen N15). ...................................145

xx
Figure Page

82. One of the Potential Locations of a Fatigue Crack Initiation Site. Notably, There Is

a Relatively Large Pore Just Under the Surface, and Surface Defects at the Location of the

Cleavage Patterns. Location of Image Can Be Seen by the Red Square in Figure 81

(specimen N15). ….. .............................................................................................................146

83. One of the Potential Locations of a Fatigue Crack Initiation Site. No Immediate

Sub-Surface Porosity Can Be Seen in the Image, but the Cleavage Patterns Look to Be near

a Partially Melted Powder Sphere Located on the Surface. Location of Image Can Be Seen

by the Blue Square in Figure 81 (Specimen N15). .............................................................146

84. One of the Potential Locations of a Fatigue Crack Initiation Site. No Immediate

Sub-Surface Porosity Can Be Seen in the Image, but the Cleavage Patterns Look to Be

Near a Partially Melted Powder Sphere Located on the Surface with Sharp Surface

Topology Indicated by the Red Arrow (specimen N15). ....................................................147

85. Near the Middle of the Specimen where Striations Form, However, They are a

Little Sparse in this Section (Specimen N15). .....................................................................147

86. The Location of This Image Can Be Seen by the Inset Red Square from Figure 85.

The Striations Point Back Towards the Predicted Fracture Locations Depicted above

(Specimen N15). ….. ............................................................................................................148

87. An Example of the Fracture Surface in the Overload Zone, Marked with a Variety

of Different Sized Dimples Across the Sample Surface (Specimen N15). .........................148

87. An Example of the Fracture Surface in the Overload Zone, Marked with a Variety

of Different Sized Dimples Across the Sample Surface (Specimen N15). .........................148

xxi
Figure Page

88. A Macroscopic SEM Image of a 2 mm AM HIP Specimen that Failed at ~131k

Cycles. An Appreciable Amount of Porosity Can be Seen Across the Sample Surface

(Specimen N5). …… ............................................................................................................150

89. A Closeup of the Approximate Failure Location Where the Inset Colored Squares

Correspond to Figure 90, Figure 91, and Figure 92 (Specimen N5). ..................................150

90. One of the Potential Crack Initiation Sites Located at the Corner of the Specimen

(Specimen N5). .....................................................................................................................151

91. One of the Potential Crack Initiation Sites Located Just Below the Corner of the

Specimen (Specimen N5). ....................................................................................................151

92. A Closeup of Figure 91 Showing River like Patterns and Cleavage Fracture.

(Specimen N5). .....................................................................................................................152

93. A Macroscopic View of the Striation Region just Inside of the Taper (Red Arrow)

where the Overload Region will Occur Further to the Right. The Red Inset Square Marks

the Location of Figure 94 (Specimen N5). ...........................................................................152

94. Striations Mark that the Crack Propagated from Left to Right. (Specimen N5). ..153

95. A Macroscopic View of the Striation Region Just Inside of the Taper Where the

Overload Region Will Occur Further to the Right. The Yellow Inset Square Marks the

Location of Figure 96 (Specimen N5). .................................................................................153

96. Defined Striations Indicate that the Crack Propagated from the Top-Left to Bottom-

Right of the Image (Specimen N5). ......................................................................................154

97. The Overload Region of the Specimen on the Far-right Side. The Red, Inset Square

Marks the Location of Figure 98 (Specimen N5). ...............................................................154

xxii
Figure Page

98. Minimal Porosity Can Be Seen from the Specimen after the Hip Cycle, the Non-

uniform Edges Indicate That the Pore Surface Has Been Deformed Through the High

Isobaric Pressure Applied During the Hip Cycle, as Compared to the Smooth Pore Walls

Found in Figure 82 (Specimen N5). .....................................................................................155

99. The Overload Zone of the Specimen, Marked by Relatively Even Dimples Across

the Surface (Specimen N5). ..................................................................................................155

100. A 1.5 Mm Thick Am Hip Specimen Failing after ~100k Cycles. Large Faceting

Can Be Seen near One of the Corners of the Specimen Indicative of a Crack Initiation Site

(Specimen M3). .....................................................................................................................156

101. The Final 3.7 Seconds of Infrared Camera Imaging of a Fatigue Test. ................157

102. A Closeup Image of the Striation Region of a HIP’ed Sheet Metal Specimen that

was used During the IR Imaging Fatigue Test. ....................................................................157

103. An AFM was used to Measure the Topology of the Striations of the Sheet Metal

Specimen. It was Found that the Striations were Spaced Almost Exactly 1 μm Apart. .....158

104. The Values of the Basquin Slope for the Data for a Respective Thickness.........159

105. The Fatigue Limit (the Max Stress at 107 Cycles or more) is Plotted vs. the Room

Temperature Quasi-Static Mean UTS Value for Each Thickness. ......................................160

106. The Stress Difference from the Basquin Slope vs. Relative Density for the Non-

HIP Specimens ......................................................................................................................162

107. An SEM Image of a HIP Specimen with a Number of Annealing Twins (Green

Arrows) that Have Formed During the High Temperatures of the HIP Cycle. ..................163

xxiii
 CHAPTER 1

INTRODUCTION

1.1. Motivation
With the advancing maturation of metal Additive Manufacturing (AM) technologies,

and in particular Laser Powder Bed Fusion (LPBF), part designers have been enabled with

a higher level of design freedom relative to traditional subtractive machining operations

and castings [1],[2]. Intricate lattices (Figure 1), surface-based cellular structures, and thin

walls can now be readily incorporated into part design to optimize performance. Metal AM

technologies do however introduce process and material specific manufacturability and

metallurgical challenges with regard to metrics such as surface roughness, internal porosity

and dimensional accuracy [3],[4], all of which have potentially significant effects on

mechanical performance. Printing parts such as heat exchangers (Figure 2) [5], acoustic

liners [6], energy absorbers [7], and Radio Frequency (RF) antennas [8] with integrated

thin-wall structures often result in geometries where it is not feasible to use finishing tools

to reach into the part’s interior and machine down surfaces to reduce the surface roughness

or eliminate sub-surface porosity.

1
 Figure 1: Intricate AM microstructure with lattices of varying thickness. Credit: By Robin Zitt -
Own work, CC BY-SA 4.0, https://commons.wikimedia.org/w/index.php?curid=84635269

Figure 2: A schematic diagram of a thin-wall plate heat exchanger. Credit: By

Plate_frame_1.png: Ubderivative work: Malyszkz (talk) - Plate_frame_1.png, CC BY-SA 3.0,
https://commons.wikimedia.org/w/index.php?curid=15033819

A key concern of the designer therefore is how these as-printed LPBF structures behave

when they are on the order of a few hundred microns in thickness, which for most LPBF

2
systems is near the limit of their capabilities. Can an analyst assume that these thin walls

have the same mechanical properties as those obtained from bulk specimens? If not, what

debits should they consider? From a process development standpoint, this raises additional

questions: why do properties change as thickness reduces? Can process modifications

either during the process, or post-processing minimize any detrimental effects? To answer

these questions, one needs both an empirical understanding of how properties change as a

function of thickness, and a deep investigation of the causes for these changes. It is

therefore end-use applications of thin wall metal AM structures, and questions similar to

the ones just posed that need addressing prior to their successful implementation, that

ultimately motivate this work.

1.2. Literature Review

This thesis examines the mechanical behavior of thin-wall metal additive

manufacturing structures, focusing its scope on Inconel 718 fabricated with laser powder

bed fusion, for quasistatic tensile and high cycle fatigue conditions. It also examines the

role of Hot Isostatic Pressing as a potential solution for mitigating any debits seen in

mechanical properties as wall thickness reduces. Subsequent chapters explore each of these

concepts in detail and are each preceded by an independent literature review specific to

that chapter. In this opening chapter, however, some key aspects of prior work are

summarized to set the stage for the thesis objectives.

3
Properties of Thin Wall LPBF Structures

Most of the prior work in developing properties for metal AM has focused on machined

specimens even led to process-specific industry standards [9], [10], [11]. There is however

also a need for estimating properties in as-built parts to better understand how AM can

make its way into the design space where machining surfaces is not an option. In this work,

“as-built” describes the state of the AM specimen without any mechanical post-processing

such as machining, honing, or polishing conducted to the surface – independent of heat

treatment. At low thicknesses (defined here as under 2 mm), it is not always apparent that

mechanical properties are the same as bulk properties measured from specimens with larger

dimensions.

Prior literature on thickness effects on metal AM properties in an as-built condition has

been limited to a few alloys studied under monotonic quasistatic tension and has reported

trends such as reductions in Ultimate Tensile Strength (UTS) with reduction in specimen

thickness [12] [13] [14] [15] [16]. These differences are typically attributed to some

combination of surface morphology, porosity, and microstructural differences.

Metal Additive Manufacturing of Inconel 718

The Inconel 718 alloy was designed for its strength, corrosion resistance, and good

fatigue life at elevated temperatures (up to 650°C) [17]. Inconel 718 has a been a popular

alloy for AM manufacturing for many reasons. Since Inconel 718 has good weldability, it

makes it an applicable alloy for the AM process since it won’t have as many issues with

cracking during the printing process [18]. Inconel 718 parts makeup ~30% of the total

weight of a modern aircraft engine making it an industry standard alloy for several

4
applications [17], [19], [20]. The high material and machining cost of conventionally

wrought Inconel 718 gives an advantage to AM since parts can be manufactured with a

near net shape to part design with minimal surfaces that need to be machined [21]. The

high machining costs typically come from the alloy’s high hardness and low thermal

conductivity.

Inconel 718 is a nickel based super alloy that is precipitation hardened through heat

treatments. The nominal composition of Inconel 718 can be found below in Table 1.

Inconel 718 has a Face-Centered Cubic (FCC) γ matrix with several strengthening phases.

The 2 most important strengthening phases are the FCC γ’ (Ni3(Al,Ti,Nb) and the ordered

tetragonal γ’’ (Ni3Nb). Alongside the strengthening precipitates, there are several

undesirable precipitates that can form in the material. The hexagonal Laves phase

((Ni,Fe,Cr)2(Nb,Mo,Ti)) is relatively brittle and hard and will impact the strength and

ductility of the alloy [22]. Meanwhile, the needle-like δ phase (Ni3(Nb,Ti)) precipitate does

not actively diminish the alloy’s strength, but it does rob the needed niobium from the

matrix (typically forming along the grain boundaries) in order to create the beneficial γ’

and γ’’ phases [23], [24].

5
 Table 1: The chemical composition of alloy Inconel 718 [25]

1.3. Research Gaps

The gaps in the literature that inform this thesis can be classified in three areas:

• Appropriateness of Research Methods: The majority of publications that do discuss

LPBF thin-wall behavior use machined parts and surfaces, which are not realistic
in the industrial design space of having blind walls and features. Studies that do
examine as-printed thin wall LPBF data have done so for very limited sample sizes.
Further, these studies leverage traditional methods of mechanical testing for
specimens that do not always lend themselves well to these studies, particularly
regarding the estimation of a section area for calculating stress values.
• Lack of Validated Causes: While several causes have been hypothesized in the
literature for the observations of reduction in properties to causes, few of these have
been validated, with no attempts to decouple these causes from each other.
• Unavailable Data: There is no published data in the following areas that this thesis
aims to provide:
o Thickness effects in the High Cycle Fatigue of as-printed Thin Wall LPBF
specimens (for any alloy)
o The role of Hot Isostatic Pressing in modulating property dependence on
thickness

6
 o None of the prior works examining as-printed, thin wall behavior in LPBF
specimens has specifically obtained data for Inconel 718

1.4. Thesis Objectives

This thesis seeks to address some of the gaps identified. Specifically, the three primary

research objectives of this work are:

• To elucidate the size effects of thin walls in the context of quasi-static and high

cycle fatigue environments, and

• To determine the effects of an added Hot Isostatic Pressing (HIP) heat treat cycle

to the industry standard solutionize and double aging heat treatment for Inconel

718.

• To identify the underlying microstructural and morphological reasons that drive

the observed size effects for quasistatic and high cycle fatigue behavior

A key distinction of this work relative to the published literature is that the specimens

have an entirely as-printed microstructure. Additionally, the specimens were printed

vertically without supports, and this is a statistically robust set of data which is important

to the inherently stochastic nature of AM parts compared to their wrought and machined

counterparts.

The subsequent chapters in this thesis address each of these three research objectives.

Chapter 2 lays out the overall research methodology used in this work, as well as discusses

some of the preliminary findings that informed subsequent work but are not included in the

final results. It also provides information on the thermal treatments used in all post-

processing, including for HIP which was used only for half the specimens in the study.

7
Chapter 3 examines porosity in the context of thin walls from three different perspectives:

measurement capability of the method used to obtain porosity, the correlation of these

properties to process parameters, and finally the observed differences of porosity with

thickness and Hot Isostatic Pressing (HIP). Chapter 4 anticipates mechanical testing by

examining morphology of the specimens used while estimating a valid and accurate section

area for use in computing stress. Chapter 5 collects all the quasistatic tensile testing data

with an emphasis on using a combination of morphology and microstructure studies to

decouple their contributions to the observed effects of property changes with thickness and

critically examine several hypotheses posed in the literature. Chapter 6 focuses on high

cycle fatigue of thin-wall as-printed LPBF specimens and represents the first such

publication of this data. Chapter 7 finally summarizes the thesis and provides directions for

future work.

8
CHAPTER 2

RESEARCH METHODOLOGY

This chapter discusses the overall approach used in this research, while saving the

details of each individual method for the respective chapter that it appears in. Also included

are findings from preliminary builds and tests that were conducted which do not appear in

the final results reported in Chapters 3-6 but were necessary prior work that was needed to

inform them. Finally, the temperature profiles for all thermal steps used in post-processing

is also provided here.

2.1. Overall Approach

The overall approach of this thesis involved three main elements, each of which shown

in Figure 3: (a) specimen design, manufacturing and post-processing; (b) characterization;

and (c) mechanical testing.

Specimens were designed using SolidWorks CAD software and analyzed using

ABAQUS Finite Element Analysis to ensure stress concentrations were in line with

standard recommendations, since these specimens were outside the dimensional range

recommended by standards for testing. These specimens were then fabricated on a Concept

Laser M2 with a 400W laser and then stress relieved in a vacuum furnace. The specimens

were then removed from the build platform using wire Electrical Discharge Machining

(EDM) and then half the specimens were processed through a HIP process, the other half

skipped this step. Final solutionizing and aging heat treatments were also included for all

specimens. Characterization was carried out prior to and after testing and involved

Archimedes testing for density estimation, optical scanning microscopy for roughness, blue
9
light scanning for part geometry, Scanning Electron Microscopy (SEM) for microstructural

studies and occasional X-ray Computed Tomography (XCT) for studies of porosity and

specimen morphology. With the exception of SEM and X-ray tomography, all other data

was collected on 100% of specimens used in this study. Finally, specimens were tested

under quasistatic monotonic tensile test conditions as well as under High Cycle Fatigue

(HCF). High-temperature testing was also conducted as part of this work but is not in scope

of this dissertation. The results from the High-temperature quasi-static test can be found

within the appendix.

(a)

10
 (b)

(c)
Figure 3: (a) The workflow of designing, manufacturing, and the necessary heat treatments
required to achieve optimal microstructure and performance in the test specimens. (b) The
primary characterization tools used in this study. (c) The 3 mechanical testing machines used in
this study for the room temperature quasi-static testing, high-temperature quasi-static testing, and
the high cycle fatigue testing.

2.2. Preliminary Builds and Lessons Learned

Before specimens could be fabricated at large sample sizes, some preliminary builds

were conducted to optimize the process and the specimen design. Three different DOE

(Design of Experiments) studies were conducted to optimize printing of thin-walled parts.

The first two DOEs were used to evaluate the use of supports (often needed in complex 3D

printed parts to reduce warpage), laser raster patterns, and the variation of laser trace width;

the effects of these variables were quantified with density measurements and a visual

11
examination. It was established that the thinnest specimen that could be printed reliably

was 0.3 mm in thickness without printing supports that would attach to the surface. Being

able to print free-standing specimens without supports was strongly desired so that a

completely as-built surface would be around the specimen. Printing thinner than 0.3 mm

produced incomplete specimens, or specimens with a variety of printing artifacts.

The final DOE further explored a variety of scan strategies to further optimize printing

reliability, reduction in surface roughness (although, almost all parts had identical surface

roughness since this metric is primarily associated with laser power and velocity), increase

in part density, and greater dimensional accuracy to the Computer-Aided Drafting (CAD)

model (Figure 4(b)). After all the DOE experiments, it was established that the final

printing parameters would have a laser power of 130 W, laser velocity of 481 mm/s, while

using a single contour boundary, a 45° raster angle for the interior scan paths, and a printing

layer of 30 μm.

(a) (b)

Figure 4: Part cross-sections (a) were used to determine the scan strategy that provided the part
with the greatest dimensional correlation to the CAD model (b).

In another study involving increasing LPBF printing layer thickness, several DOEs

were used incrementally to increase the printing layer thickness from 30 μm to 60 μm and

12
finally to 80 μm to help reduce total printing time and costs. The primary variables of these

DOEs were laser power and velocity. More of this study can be found in the appendix of

this document.

2.3. Specimen Design

Three different specimens had to be designed for each of the different mechanical

tests. The ASTM E8 [26] standard was used as a baseline for creating the quasi-static

tensile specimens, however, deviations from the standard had to be made to ensure parts

could be printed in the limited build volume of the LPBF printer and to consider testing

constraints. Likewise, ASTM E466[27] was used as a baseline for the fatigue specimens.

The plate type specimen from each of the standards was used since that is the most

representative geometry to study thin wall behavior. Finite element analysis was used to

ensure the changed specimen geometries had a stress concentration factor under 1.1 as

shown in Figure 5. Figure 6 and Figure 7 show the final specimen profile for the room

temperature quasi-static test, and the fatigue test, respectively. For fatigue testing, there

were 2 options for rectangular specimen design. The first option was to use a specimen

with a continuous radius and no gauge length, while the second option used a radius and a

reduced gauge length. The second option was chosen since it would allow the fatigue crack

to grow and develop based on the internal/external defects naturally as opposed to a

geometrically forced fatigue crack initiation in the center of the specimen.

13
 Figure 5: Finite Element Analysis simulation of the quasi-static specimen under tensile loading,
showing localization of stress concentration, which was ensured to be under acceptable limits

Figure 6: The room temperature quasi-static specimen profile

Figure 7: The high cycle fatigue test specimen profile

Once the specimen geometries and printing parameters were all established final

specimens could be manufactured. Figure 8(a) shows the layout of 140 specimens being

printed simultaneously. The 140 specimens include 7 different thicknesses and 20

replicates per build plate. For the room temperature quasi-static build, these 7 thicknesses

are 0.3, 0.35, 0.5, 0.75, 1.0, 1.5, and 2 mm. The specimens are arranged randomly across

the build plate to reduce any sort of mechanical property dependance on build plate

location. The specimens were oriented such that the coater blade moved across the
14
specimen perpendicular to the thickness to decrease the chance of the coater blade

perturbing the printing process. Six builds were printed in total with 2 builds belonging to

each of the three tests. The first set of test specimens that were printed used virgin powder.

Every subsequent build used the powder recycled from the previous build after it was

filtered with additional virgin powder being added as needed. The build order was: 2 builds

of the room temperature quasi-static specimens and 2 builds of the HCF specimens.

Although, it is less than ideal to use recycled powder due to a potential decrease in

mechanical properties, an increase in surface roughness, or an increase in pores generated

[28], some authors have found no decrease in microstructure, mechanical properties, or

porosity for powder that has been reused 14 times [29]. Figure 8(b) shows a final,

successful print of the quasi-static tensile specimens before the stress relief and subsequent

EDM removal from the base plate.

15
 (a)

(b)

Figure 8: (a) The build setup before sending the CAD file to the Concept Laser LPBF printer. (b)
The first build while the specimens are still on the build plate and prior to receiving a stress relief
cycle.

16
 Sheet metal specimens were also manufactured alongside the AM parts to be used

as a baseline, testing validation, and for hypothesis building from parts with minimal

surface roughness and internal defects. Seven Inconel 718 sheets were purchased with the

intention of getting thicknesses as close to the AM printed parts. The 7 thickness were 0.41,

0.51, 0.64, 0.81, 1.02, 1.60, and 2.03 mm; the 6 thickest sheets came from the same metal

manufacturer, but the 0.41 mm sheet came from a second metal manufacturer. The sheet

metal specimens were cut from the sheet using a waterjet cutter with the long axis parallel

to the sheet rolling direction using the specimen profiles found in Figure 6 and Figure 7.

All sheet metal specimens went through the same exact heat treatment process in the same

batches as their AM counterparts. The purpose of the sheet metal specimens is to help

validate the specimen geometry, to be used as a baseline for Inconel 718 parts to compare

to the AM specimens, and to help elucidate any thickness and surface roughness effects in

the testing. All sheets are in compliance with AMS5596 and were machined from the sheet

metal stock using a waterjet cutter as seen in Figure 9.

Figure 9: Alongside the AM specimens, sheet metal specimens were waterjet cut out of Inconel
718 sheet metal stock.

17
2.4. Heat Treatment
Heat treatments are essential for the AM Inconel 718 alloy. All specimens reported

in this work did receive a series of heat treatments after the parts were printed, however, a

series of mechanical tests on as-printed parts were conducted as a baseline to compare

subsequent heat-treated specimens against.

While all the specimens are still on the build plate, they were taken to Phoenix Heat

Treating Inc., an offsite heat treatment provider, to first receive a stress relief heat treat

cycle. The stress relief cycle is used to mitigate internal stresses that can form during the

rapid heating and cooling cycles of the additive manufacturing process. Since the tensile

specimens printed are thin and have only one point of contact to the build plate, it is to be

expected that these parts would not have severe internal stresses when compared to parts

with a much larger printed mass and several different contact points on the plate.

Regardless, it was important to perform the stress relief cycle to best match the industrial

standards for heat treating AM parts. The stress relief cycle takes place in a vacuum furnace

with an initial ramp to 760°C (1400°F) for 2 hours followed by a ramp up to 1066°C

(1950°F) for 90 minutes.

After the stress relief cycle, the specimens are then removed from the build plated

with wire electrical discharge machining (EDM). Half of the test specimens were sent to

Quintus to receive a hot isostatic pressing (HIP) heat treat cycle following the stress relief

cycle. The HIP step is used to close the internal porosity generated by the AM process by

heating the specimen while simultaneously applying an isobaric pressure with argon gas.

The HIP cycle consists of a temperature of 1120°C (2048°F) for 4 hours with 100 MPa of

pressure per the recommendation in ASTM F3055-14a [30].

18
 As part of the final heat treat steps, all specimens received a solutionize and double

age heat treatment, as per the recommendations of AMS2774E [31]. The solutionizing

temperature and time is 954°C (1750°F) with a 1 hour hold time and the double aging step

starts at 718°C (1325°F) for 8 hours followed by 621°C (1150°F) for a total precipitation

time of 18 hours. Figure 10 shows the 4 main heat treatment temperatures and times. Figure

11 shows the stress-strain curves for the 5 different heat treat conditions as they proceed

through the different heat treat cycles. SA is an abbreviation for Solutionize and Age. It is

to be noted that the HIP + SA and SA steps of Figure 11 are the final heat treat conditions

for the specimens within this study. The sheet metal specimens originally have a vastly

different grain structure, texture, and heat treat condition when compared to the AM parts.

To help unify the 2 distinct specimens (AM and sheet), all sheet metal specimens went

through the same heat treat cycles alongside their AM counterparts.

Figure 10: A graph of the 4 major heat treatments that were used in the study plotted on a Temp.
vs. time graph.

19
 Figure 11: Test data showing the effects of the different heat treatments on the stress-strain
response of the AM parts as they proceed through the different heat treat cycles. SA is shortened
for solutionize and age. The stress relief and the HIP steps reduce the strength and increase the
ductility, while the ageing step is crucial for high strength parts.

20
 CHAPTER 3

THE QUANTIFICATION AND QUALIFICATION OF POROSITY IN METAL

ADDITIVE MANUFACTURING AS A FUNCTION OF WALL THICKNESS

3.1. Introduction
One of the primary goals of establishing LPBF printing parameters is the qualification

and quantification of the pores generated through the relatively chaotic melting and

remelting process. There are a number of ways that porosity is formed through the LPBF

process, and this typically comes from having too much energy applied to the melt pool

(keyhole porosity [32]–[34]), or having too little (lack-of-fusion defects [35], [36]).

Keyhole porosity is typically more spherical in shape, while lack-of-fusion defects can take

on a much sharper topography. Both forms are detrimental to mechanical strength, as pores

can add internal stress risers to the flow stress and remove key material to ensure a strong

part. In achieving the goal of fully dense parts, it is typical to print an array of 10 mm cubes

with varying laser speed and power and measuring the density of the cubes to ascertain the

ideal applied volumetric energy density (a combination of the laser power (W), the laser

speed (mm/s), laser hatch spacing (μm), and the build layer height (μm)). A number of

DOEs were previously discussed in Section 2.2 where the density of the parts was

evaluated using cross-sections and the Archimedes density method, but a more thorough

examination of measuring the density, specifically in the context of thin wall specimens,

which is central to this work, will be discussed herein. There has been no prior work

examining the appropriateness of using the Archimedes density for thin wall specimens

and since this thesis relies heavily on the quantification of porosity for the purposes of

21
correlation to mechanical properties, it was crucial to examine the appropriateness of this

metrology for the specimens in question in this work. As a result, a thorough investigation

of the Archimedes method was conducted and is included in this chapter.

3.2. Density Metrologies

The density of additively manufactured parts is typically evaluated in one of three

different methods:

1. The use of microscopy to image and analyze the relative amounts of dense
material from pores in a cross-section of the material
2. The use of X-ray tomography to probe the entire specimen to measure the amount
of dense material, and
3. The Archimedes density method that uses the relative mass of an object in two
different mediums of known density

Each of the above three methods has its own pros and cons and together they can contribute

to the total understanding of the porosity generated through the AM process.

Examining cross-sections under the microscope can be very time consuming, and

the density information ascertained is not representative of the entire specimen, but rather

a single slice. This leads to the necessity of polishing and measuring a multitude of cross-

sections to begin to have a statistically valid measure of the overall density of the part. On

the other hand, the benefit of examining the cross-sections of the material is gaining an

understanding of the porosity distribution throughout the part and the types of pores that

are generated. The ability to see if keyhole or lack-of-fusion pores are being generated can

give the user a greater insight to help optimize process parameters. For example, the

22
location at which the porosity is being generated can lead to laser scan optimization through

choosing better scan strategies to mitigate the generation of pores.

An example of using an image processing software to threshold and ascertain the

density of a section is shown in Figure 12, which shows a thin wall section with the build

direction pointing into the page, having an estimated density of 98.68%. To better visualize

how the location of the generated porosity can elucidate the cause, the micrograph of Figure

13 has been overlaid with the laser scan pattern obtained from the Concept Laser software

that controls the path of the laser in the LPBF machine. Figure 13 shows the cross-section

of a 2mm thick specimen with most of the porosity being just under the as-printed surface;

the build direction is pointing up. This sub-surface porosity is likely generated from an

increased amount of laser power being directed into the melt pool due to the laser velocity

slowing down to make the tight turn at the corners of the interior raster scan path. In a study

evaluating pore formation with in situ x-ray imaging by Martin et al. [37], it was found that

greater than 80% of pores are generated within 200 μm of a raster turn. The laser slows

down slightly as it’s making a turn which creates the keyhole pores. This phenomenon is

less evident from the micrograph in Figure 12 but can be clearly seen in Figure 13 as most

of the porosity is found at the corners of the laser scan path.

23
 (a) (b)
Figure 12: A polished cross section of one of the thin walls manufactured as part of this study
with the build direction pointing into the page(a), the purple arrows pointing to pores that show a
lack of fusion or partially melted powder spheres while most of the pores show keyholing
porosity. After the image has been trimmed down to remove the edges (b), ImageJ is used to
threshold and quantify the amount of porosity.

24
Figure 13: A 2mm thick AM cross-section with the build direction pointing up showing a large
number of pores generated just under the sample surface (the left and right sides of the
micrograph). When the laser scan path is overlaid on top of the micrograph, it is shown that the
majority of the generated porosity takes place near the interior raster turns and/or in between the
contour boundary and the interior raster scan lines.

X-ray tomography is a powerful tool that uses x-rays and detectors to measure the

amount of absorbed light through the test specimen which can then be processed to

understand the location and size of pores distributed throughout the material [38],[39]. The

primary issues with X-ray scanning are the cost of running the CT (Computed

Tomography) scanner, the acquisition time, and the inability to scan very thick, highly

dense parts. Since Inconel 718 is a relatively dense alloy, parts scanned for this work were

machined down to be under 1.5 mm in width (while having a 1.5 mm printed thickness) to

avoid longer scanning times and the mitigation of the X-ray absorption. A Zeiss Xradia

520 was used for the X-ray Computerized Tomography (XCT) scanning of the parts.

Voltage, power, and target current were set to 140 kV, 10 W, and 71 μA, respectively. The

25
XCT resolution was 3.6 μm on each voxel side, therefore pores appreciably smaller than

this could not be captured with the XCT scanner with confidence. Figure 14 shows the 3D

reconstruction of the CT data with a higher density of porosity generated just underneath

the as-printed surface, while the machined face (representative of the interior of the

specimen) shows much less porosity. Binning the CT data by the amount of porosity in 100

μm slices, Figure 15 shows the strong bimodal distribution of porosity found across the

thickness of a 1.5 mm specimen. Near the edges, the porosity increases to approximately

3x greater than the specimen average density calculated from the entire scanned volume.

Figure 14: A 3D reconstruction of the 1.5 mm specimen that was CT scanned for this study. The
generated porosity is unevenly distributed with more forming just under the as-printed surface.

26
Figure 15: X-ray tomography was used to quantify the distribution of porosity in the non-HIP’ed
specimens compared to the average value. It was shown that most of the porosity is found near
the specimen edge.

Using a similar approach as before with the optical micrographs, scan patterns can

be overlaid onto the resultant CT .tiff images. Figure 16 shows 2 images selected for having

greater than average porosity, when compared to other images found in the .tiff stack – the

as-printed surface is located at the top and bottom of the image, while the left and right

show the machined surfaces. Figure 16(a) shows that the pores generated are typically

closer to the surface of the printed specimen for the nominally thick 1.5 mm part. Again,

as is shown before in Figure 15, most of the porosity is generated approximately 200 µm

from the sample surface. As the specimen thickness decreases, there is no longer a denser

center that is sufficiently away from the sides. Instead, the porosity is distributed more

27
evenly across the cross section of the specimen as shown in Figure 16(b) for a specimen

that is 0.35 mm in thickness.

(a) (b)

Figure 16: Two selected .tif images from the resultant .tif stack collected through the X-ray
tomography that show a relatively greater amount of porosity than average. A 1.5 mm (a) and a
.035 (mm) specimen are shown with the as-printed surface at the top and bottom of the image.

The Archimedes density approach is the fastest, most simplistic, and offers relative
density for the entire specimen [38]. The use of an Archimedes density measurement kit
was the chosen tool for measuring part density as it leverages reliable data while being
cheap and fast. The Archimedes density kit used for this study is shown in Figure 17 with
an analytical balance with a resolution of 0.1 mg, and a dual weighing pan to conduct
measurements in air and a fluid of a known density. To calculate the absolute density of a
part, 4 things need to be measured: the temperature of the distilled water (to determine the
fluid density from a table), the temp of the air (to determine the air density from a table),
the mass of the specimen in air (measured and averaged 3 times), and the mass of the
specimen submerged in the fluid (measured and averaged 3 times). Density is calculated
28
using Equation 1, where ρ is the calculated relative density, md is the average dry mass, mw
is the average wet mass, ρa is the density of air, and ρw is the density of water at a given
temperature.
𝑚𝑑
𝜌=𝑚 ∗ (𝜌𝑤 − 𝜌𝑎 ) + 𝜌𝑎 (1)
𝑑 −𝑚𝑤

Figure 17: The Archimedes density measurement setup for the collection of relative densities of
the 10 mm DOE cubes as well as the dogbone specimens.

Two ASTM standards detail the process for the Archimedes method as it may be

applied to compacted or sintered powder metallurgy products which may be leveraged for

LPBF manufactured parts as well. The most general of these standards is ASTM B962

[40]. The ASTM B311 standard is used for parts that have porosity that is less than 2%

[41], which will typically be the case for most optimized AM parts. However, due to the

uncertainty of using the Archimedes density method further analysis of the measurement

system had to be first accomplished before applying density results to the final tensile test

specimens. Unlike the micrograph cross-sections where the porosity can be directly seen

29
and separated via thresholding, or like the CT scan with micron resolution, the Archimedes

density approach involves several measurements each with its own uncertainty.

Despite applicable ASTM density measurement standards for the Archimedes density,

there is still a need to better understand the metrology limitations and accuracy in general

and as it pertains to thin wall specimens such as those shown in Figure 18(b). While

studying density in cubes is useful for process selection, it does not guarantee that the same

density will hold as specimens get thinner and the surface area of the specimen increases

relative to its volume. This is primarily because sub-surface porosity contributes to a larger

proportion of the overall volumetric density as thickness of the specimens reduces [42],

[43]. Prior work has demonstrated that thin wall specimens in the LPBF process do not

have identical properties to those as determined from using large section “bulk” specimens,

with reductions in strength reported with decreasing thickness [44]–[48]. As a result, there

is a need to understand how thin wall structures behave independent of the bulk material

property, and the underlying reasons for any differences in observed behavior. A key aspect

of enabling this understanding is being able to measure the density in thin wall geometries

and examine how they relate to process variables and measured properties, potentially

serving as a mechanistic explanation connecting the two. If the Archimedes density method

can be shown to be capable, sensitive, and correlative, it can increasingly be used for

purposes beyond just process development.

30
 (a) (b)

Figure 18: LPBF specimens prior to removal from build plate: (a) 10 mm cubes commonly used
in process development and for Archimedes density estimation, (b) thin wall tensile test
specimens used for quantifying mechanical properties.

3.3. Measurement Capability Analysis of the Archimedes Method

The purpose of an MCA is to assess the capability of a metrology for measuring a

metric of interest on a specific component or specimen. The method generally consists of

three specific evaluations, conducted sequentially: (i) Accuracy – how close the predicted

result is to the best known “true” value, (ii) Repeatability – how stable the measurement

method is in the short term under controlled circumstances such as with successive

measurements made by a single operator, and (iii) Reproducibility – how stable the

metrology is in real-world conditions, assessed across different specimens, operators, and

days. MCAs can also help make assessments of the appropriateness of using a metrology

for a specific customer against pre-determined specification limits. Ideally, the metrology

does not cast a dominant influence on overall measured process variation and is an

acceptably small component of it [49]. This work seeks to apply the MCA approach to
31
Archimedes density measurement – while the intent is to assess this for thin wall

specimens, a baseline is first established for this method’s suitability for measuring density

in 10 mm cubes, the standard specimen geometry used when developing the LPBF process.

The Inconel 718 10 mm cubes were manufactured on a Concept Laser M2 with a wide

range of parameters and a 30 µm layer thickness, as part of a process optimization DOE as

a precursor to this work [50]. Three of these cubes were selected for use in this study, along

with a 10 g standard calibrated weight, which were then used for the MCA. The results of

the MCA are first reported in turn for each of the three main components of the study:

accuracy, repeatability, and reproducibility, after which the implications are assessed in the

context of the LPBF process.

3.3.1. Accuracy

The first task in an MCA is to assess the accuracy of the metrology of interest,

which typically requires a standard of known value. For this study, a newly obtained 10 g

Troemner precision class mass was used, backed by an NVLAP accreditation certificate.

While technically a mass standard used for calibration purposes, the certificate also

specifies the density of the mass as 8.03g/cm3, though this is not a density standard. The

10 g mass is close to the measured mass of the 10 mm Inconel 718 cubes and similarly

sized. A separate, calibrated 200 g mass was first used to calibrate the mass balance. After

this, the 10 g mass was measured 16 times successively, carefully dried following each wet

measurement before making the next dry measurement. The results of the density

calculations due to these measurements is shown in Figure 19.

32
 Figure 19: Density computed for a standard 10g mass (shown in inset) from 16 successive
measurements.

The mean for the accuracy study with the 10 g mass was 7.991 g/cm3, with a bias

from the standard value of -0.039 g/cm3, amounting to 0.49% in the context of relative

density. Given that this is a mass, and not a density standard, this bias must be viewed

with caution, but nonetheless is a good indication of the accuracy of the setup, and

indicates that given a well trusted standard, subsequent readings for the Inconel 718

cubes be increased by 0.039 g/cm3 for greater accuracy. Finally, these 16 measurements

demonstrated a standard deviation of 0.002 g/cm3, which will be of interest when

comparing to those derived for cubes and tensile test specimens.

3.3.2. Repeatability

Repeatability was assessed by measuring the density of a one cube and one thin-

wall specimen 30 times in succession by a single operator on a single day. To replicate

actual testing conditions, the specimen was weighed in dry conditions, followed by wet,
33
and then completely dried before retesting. The balance was tared back to 0 after each mass

measurement, all 30 measurements were made within a 4-hour window of time. Since this

is a time-consuming activity, a preliminary study was conducted on thin wall specimens

with seven different thicknesses, but with only 8 repetitions to see what effect thickness

had on repeatability. As shown in Figure 20, there was a clear reduction in mean density

calculated from 8 repeated measurements on one specimen at each thickness, but more

importantly for the current discussion, the standard deviation increased as thickness

reduced, with the 0.3 mm specimen having the highest repeatability sigma (σRPT). As a

result, the 0.3 mm was selected for the full 30-repetitions study.

Figure 20: Mean (left) and standard deviation (right) from 8 measurements made of a single
specimen at each of 7 different thicknesses.

All 30 datapoints are shown in Figure 21 (a) and (b) for the 10 mm cube and the 0.3

mm specimen, showing short term stability for both, based on statistically calculated upper

and lower control limits, labeled as UCL and LCL respectively. σRPT of the 10 mm cube

was 0.0013 g/cm3, which was even lower than that obtained from the prior accuracy study.
34
On the other hand, σRPT for the 0.3 mm specimen was estimated at 0.0057 g/cm3,

approximately four times what was obtained for the cube – both values are listed in Table

2. In any case, repeatability only assesses short-term stability and is not representative of

the metrology’s use in real-world conditions, where many operators over different days are

measuring different specimens. To assess the stability of the metrology in these conditions,

a reproducibility study is required.

(a)

(b)

Figure 21: Repeatability data from 30 successive calculations of density of (a) a single 10 mm
cube, and (b) a single 0.3 mm thin wall specimen showing short term stability for both

35
 3.3.3. Reproducibility

Reproducibility quantifies the variation in the average of the measurements made

by various operators on different days using the same measurement equipment and

technique on the same parts. A statement of work that clearly identifies each step is

provided to each operator to encourage use of the same procedure. In this work, three

different 10 mm cubes and thin wall specimens at three different thicknesses (0.3 mm,

0.75mm and 2 mm) were selected, and three operators were asked to estimate the density

of each specimen on each of three different days.

The results from this study for the cubes are shown in Figure 22(a), which shows

the measurement values centered by the mean of the three cubes, to allow for varying

means, indicated in different colors by each of three different operators (labelled A, B and

C), on three days (1, 2 and 3). The entire range of variation is captured within

approximately 0.02 g/cm3 and the standard deviation associated with this reproducibility

study (σRPD) was 0.0050 g/cm3, which is approximately four times that estimated in the

repeatability study, which suggests that the metrology has good discrimination since it can

divide the tolerance band into more than five distinct data categories [49]. The same results

are shown in Figure 22(b) for the three thin wall specimens, clearly showing a wider spread

of about 0.08 g/cm3, excluding one outlier from operator A on day 1. The resulting σRPD

for thin walls was 0.0156 g/cm3, which is three times what was obtained for the cubes.

36
 (a)

(b)

Figure 22: Repeatability data from 30 successive calculations of density of (a) a single 10 mm
cube, and (b) a single 0.3 mm thin wall specimen showing short term stability for both.

3.3.4. Precision-to-Tolerance (P/T) Ratio

Accuracy, repeatability, and reproducibility studies generate a quantifiable

assessment of a particular metrology’s capability but say little about the appropriateness of
37
that metrology for the manufacturing process of interest. The Precision-to-Tolerance, or

P/T ratio, is typically used to address this. Precision-to-Tolerance assesses the variation

introduced by a metrology within the context of overall process variation. Strictly speaking,

the P/T ratio is only meaningful when specification limits relevant to a customer are known.

In the context of relative density in the LPBF process, one may specify a Lower

Specification Limit (LSL) of 98%, with no Upper Specification Limit (USL) – in other

words, the assumption is made that a customer will only accept parts with as-printed

relative densities in excess of 98%. This is reasonable since densities for LPBF in excess

of 98% can be subsequently improved to 99.5% and higher after Hot Isostatic Pressing

(HIP) [50]. P/T ratios can be specified for both repeatability and reproducibility, with the

latter enveloping the former. Since there is no USL for this process (100% being theoretical

maximum), P/T can be calculated for a one-sided specification in terms of the measured

mean, standard deviation, and the LSL as [49],

𝑃 3𝜎 RPD
(𝑇 ) = Process Mean × 100 (2)
𝑅𝑃𝐷 - LSL

A process mean for the Archimedes density of as-printed Inconel 718 without any heat

treatment on this machine was established as 8.196 g/cm3 in a separate study[50] , which

was set as the process mean in equation 2, after reducing it by the bias established

previously. Fully dense parts in that study resulted in a LSL (98%) of 8.0321 g/cm3. For

the Archimedes density study on cubes, the P/T for reproducibility was thus estimated as

12.4%, indicating that the process is capable since only 12.4% of the overall process

variation can be attributed to the metrology. A value less than 30% is typically sought after

38
in industrial applications. On the other hand, due to the higher σRPD for thin wall specimens,

the P/T for thin walls was estimated as 28%, marginally within acceptable limits.

Summarizing the findings of the MCA of the Archimedes density method, the

metrology is stable, has adequate discrimination, and is capable of measuring the density

of both 10 mm LPBF cubes and thin wall specimens. As a result, it may reliably be used

in process development for thin wall specimens, with the caveat that measurement error

increases with reduction in wall thickness. The values obtained in this section are

summarized in Table 2 for the 10 mm cube and the thin walls, bearing in mind repeatability

sigma for the thin walls comes from the 0.3 mm wall, representing the worst case in this

study.

Table 2: Measurement Capability Analysis Summary

10 mm Cube Thin walls
(0.3-2 mm)
Repeatability Sigma (σRPT) 0.0013 g/cm3 0.0057 g/cm3
3
Reproducibility Sigma 0.0050 g/cm 0.0156 g/cm3
P/T Ratio 12.4% 28.0%

3.4. Process Sensitivity of the Archimedes Method

This section attempts to answer the question: is the Archimedes density method

sensitive enough to detect changes in design and process? The next section deals with

answering the question in the context of its usefulness for evaluating correlations in density

to measured mechanical properties. Both sections examine these questions only for thin

wall, rectangular section LPBF specimens.

39
 3.4.1. Specimen Thickness

To assess if the Archimedes density could detect differences in density as thickness

reduces, 14 tensile test specimens (manufactured with identical process parameters) at each

of seven thicknesses ranging from 0.3 mm to 2 mm were characterized with the

Archimedes method. These 14 specimens received standard heat treatment including HIP.

As the thickness of the specimen reduces, one would expect density of the part to reduce

as subsurface volume increasingly dominates overall behavior. This is borne out in the

Archimedes density estimations as shown in Figure 23, which shows that at and above

0.75mm, there is no statistical change in density (following the Tukey-Kramer method of

comparison) but a trend is nonetheless suggested. However, there is a statistically

significant reduction in density below 0.75mm. This confirms expectations from the prior

metrology capability analysis that variation increases as the section gets thinner, at least

part of which is attributable to measurement error increasing. Despite this increase in

measurement error, it is evident that the method is sensitive enough to detect changes in

density with reducing wall thickness.

40
 Figure 23: Archimedes density estimated for thin wall specimens ranging in thickness from 0.3
mm to 2 mm. Statistically significant differences were detected between thick specimens
(>0.75mm) and thin ones (0.3-0.5mm).

3.4.2. Effect of Hot Isostatic Pressing

Hot Isostatic Pressing, or HIP, is well known as a method for increasing post-print part

densities. In total, 168 specimens were studied where 84 specimens across the 7 thicknesses

mentioned previously were HIP’d, and an additional 84 went through identical heat

treatment processes except they did not get HIP (referred to henceforth as the “NO-HIP”

condition), thus retaining most of the post-print porosity. Of interest to this work is whether

the Archimedes density method is sensitive enough to detect differences in HIP condition

for thin wall specimens, and this comparison is shown in Figure 24(a) for both conditions

across all 7 thicknesses, clearly showing a discernible difference between HIP and no-HIP

specimens. Figure 24(b) shows variation in standard deviation. For specimens with

thickness above 0.75mm, HIP greatly reduces the sigma relative to NO-HIP specimens,

while at lower thicknesses it is likely that the increase in standard deviation is driven by

the higher measurement error expected for thin specimens.

41
 (a)

(b)

Figure 24: Effect of Hot Isostatic Pressing (HIP) on Archimedes density across seven different
specimen thicknesses, showing (a) higher means after HIP, as well as (b) significant reduction in
standard deviation at higher thicknesses.

3.4.3. Build Location

In a separate study [50], 100 2 mm thick rectangular-section tensile test Inconel 718

specimens were fabricated in a single, uninterrupted build using virgin powder and

identical laser process parameters. The X and Y-coordinates of all 100 specimens on the

build platform was tracked and density was estimated using the Archimedes method with

the specimens having received no heat treatment of any kind (including HIP). The resulting

42
density values are plotted on an X-Y plot shown in Figure 25, representing the build

platform, to assess if the method is sensitive enough to detect location dependence in the

build, for the identical specimen design. The plot does suggest some spatial dependence,

with lower values of density (indicated by lighter shades of blue) at the front (low y value)

and rear (high y value) ends, with higher values (indicated by darker shades of blue)

concentrated in the middle.

Figure 25: Density estimates for 100 as-printed 2 mm thick specimens across the build platform,
suggesting low densities at the front and rear, and higher values in the middle

3.5. The Results of the Archimedes Density Method on the Final Test Specimens
It was shown that the Archimedes method for measuring density on thin-wall parts is

sufficiently accurate and reliable for this work. The use of polishing and microscopy to

qualify the location of the porosity is still useful, but for statistical purposes and uses for

1-1 correlations of density and mechanical properties, the Archimedes method will be used.

Figure 26 and Figure 27 show the results of the Archimedes density measurements for all

the specimens used in this work for the room temperature quasistatic and the HCF
43
mechanical tests, respectively. For both sets of specimens it was found that the HIP heat

treatment had a strong increase to the overall part density. After the densification of the

HIP process, the test specimens show a much stronger trend with thickness, but this

thickness dependence can still be found amongst the non-HIP set of data. Figure 26 and

Figure 27 shows that the HIP step greatly decreases the specimen density standard

deviation and creates parts with much more consistent densities.

Figure 26: The absolute density values of the room temperature, quasi-static tensile test
specimens for the 2 different heat treatments across all the different thicknesses tested. The HIP
heat treatment significantly reduces the standard deviation of the test specimens for the thicker
parts – this is especially evident after the 0.5 mm thick specimen.

44
Figure 27: The absolute density values of the fatigue tensile test specimens for the 2 different heat
treatments across all the different thicknesses tested. The HIP heat treatment significantly reduces
the standard deviation of the test specimens for the thicker parts.

3.6. Conclusions
This work examined 3 different methods for characterizing the density of additively

manufactured parts: cross-sections of AM specimens, X-ray CT data, and the Archimedes

density method. The findings from the cross-sections and X-ray CT data are as follows:

45
 • The use of examining cross-sections and X-ray CT data proved to be pivotal in
understanding the location dependence of the porosity generated via the laser
powder bed fusion process.
• With the process parameters and LPBF machine used in this study, most of the
pores will be located close to the surface of the part near the turn of the interior scan
path.
• As the part thickness decreases, the amount of sub-surface porosity stays
consistent, but the thinner specimens will have a reduction of more dense internal
material which will decrease the part density.
• Most of the porosity formed is spherical in shape, but lack of fusion and sharp
internal topography of pores is still seen.

For high-throughput density measurements, it was found that the Archimedes density

method is reliable, cost-effective, and accurate for the purpose of this work in the context

of thin-wall LPBF specimens in the range of 0.3-2 mm thickness possessing densities in

excess of 98%. The method was evaluated in two different areas: it’s capability as a

metrology for these specimens and its sensitivity to design and process conditions. The

main conclusions of the Archimedes density measurement metrology capability, applicable

to the scope of material, process, and thickness range evaluated here, are as follows:

• The Archimedes method is a capable metrology for thin wall LPBF specimens in
the thickness range studied here (as low as 0.3 mm).
• Repeatability and reproducibility sigma values for thin walls are about three times
higher than those obtained for standard 10 mm cubes.
• Measurement error increases with reducing specimen thickness.
• The Archimedes method is sensitive enough to detect differences in thin wall
thickness, with sigma values increasing as thickness reduces, requiring higher
sample sizes to establish differences in means.
• Density values obtained detect increases associated with the Hot Isostatic Process.
46
• Variations in density across the build platform were detectable for 2 mm thick
specimens.

47
 CHAPTER 4

THE CRITICAL ROLE OF MEASURING CROSS SECTIONAL AREA OF LASER

POWDER BED FUSION THIN WALL STRUCTURES WITH AN AS-PRINTED

SURFACE

4.1. Introduction
An important challenge in the mechanical testing of as-built thin wall parts is

establishing an accurate cross-sectional area to calculate stresses from the measured forces,

as conducted in a standard tensile test. At first glance, this may appear to be a trivial

problem, but as the specimen being examined approaches smaller dimensions, ordinarily

negligible errors in calculating this area can have a large impact on the accuracy of the

reported stress. This is particularly true for metal AM, with its potential for higher

dimensional variation, surface roughness, and specimen warpage, when compared to

machined specimens of the same geometry. This questions the accuracy of commonly used

measuring techniques for the estimation of section area in metal AM specimens: either

leveraging nominal design assumptions or using contact-based measurement techniques

such as calipers and micrometers, as is typically recommended in mechanical testing

standards [51]. As shown in Figure 28(a), surfaces from powder bed fusion processes

typically have partially melted powder spheres adhered to the sides, with visible melt pool

traces. Further, parts fabricated with raster techniques typically possess rounded corners as

shown in Figure 28(b).

Metal AM processes may be broadly classified into powder bed fusion, directed energy

deposition, binder jetting, and material extrusion. Of these different processes, it is the

48
powder bed fusion process and specifically LPBF, that has dominated adoption for

engineering applications. Further, the resolution of LPBF makes it possible to fabricate

thin-walls in the hundreds of microns in components that can be as large as tens of

centimeters, without needing to resort to post-process machining of these walls. Surface

roughness (Sa) values for parts printed with the LPBF process range from 5-15 µm

depending on material, laser parameters and orientation [11], [12], while specimens

manufactured via the EBM process, which is also a powder bed fusion technology, are

typically higher [13], [14]. The ASTM E8 standard, which is the most commonly used

guideline in the tensile testing of metals, provides recommendations for micrometer and

caliper accuracies based on minimum specimen dimensions [52]. However, the use of such

single-point measuring devices for rectangular section specimens needs to be re-examined

when the surface finish is rough, the specimen topology is non-planar, and the corners are

rounded, as is the case for LPBF specimens. An examination of the literature dealing with

the measurement of mechanical properties in thin wall specimens shows that the choice of

metrology used for assessing the section area is dependent on the specimen size. For micron

and submicron size specimens, it is well recognized that inaccuracies in specimen

dimension can have significant effects on the estimated stress values [53]–[55]. Sharp et

al. used an ellipsometer to measure the sample thickness (in their work, ranging between

1.5-3.5 µm in thickness), and optical images to measure the specimen width (6-600 µm

wide) [56]. For specimens in the mesoscale, that are more in line with the geometries of

interest in this work, researchers may easily employ contact metrology techniques such as

calipers or micrometers, if the parts are machined and have section shapes that are

49
consistent with the nominal design. In the literature specific to metal AM, calipers have

been used to estimate thickness of as-built specimens as thin as 0.35 mm for LPBF

specimens [48], and 0.5 mm for EBM specimens [47]. Optical microscopy has also been

used as a means of non-destructive 2D measurement of specimen thickness and width, prior

to estimation of mechanical properties [44]. The aim of this work is to compare different

metrologies that may be used in the measurement of section area for as-built thin wall laser

powder bed fusion specimens, and to assess how detrimental any deviations in these

metrology measurements are to the estimation of mechanical properties, something that, to

the best of our knowledge, has not been systematically studied and described before. At a

practical level, the work seeks to make data-based recommendations for measuring

thickness in thin wall, as-built LPBF specimens.

(a)

50
 (b)

Figure 28: An SEM image of the surface of an AM part shows the rough surface due to partially
melted powder spheres and melt pool paths. (b) A cross-section of a separate, polished AM
specimen reveals rounded edges despite sharp corners in the CAD file used for printing - this
specimen was sectioned such that into the page is parallel to the build direction.

Four different measurement techniques were examined in this study: x-ray

microtomography, blue light optical profilometry, point micrometer, and standard

micrometer. Optical profilometry, commonly called 3D scanning, has proven to be one of

the most promising non-contact systems for measuring the 3D surface of objects with

relatively quick speed, adjustable resolution and sensitivity, and being able to produce are

large number of data points of the surface [57]–[60]. Optical profilometry is a relatively

mature technology that has found great applicability in rapid part examination and reverse

engineering [61]. The resolution and accuracy of the scanned part is dependent on the

hardware and measurement technique, but researchers are typically capable of achieving

lateral and vertical resolutions of 10 and 3 µm, respectively [62]. Wi et al. have shown the

utility of 3D structured light scanning as a non-contact, non-destructive metrology for

51
characterizing print quality including roughness and distortion for 3D printed clay objects

[63]. X-ray Computed Tomography (CT) is also often used in characterizing metal AM

parts, though most commonly it finds use for the purpose of identifying and quantifying

internal porosity [64], [65]. Optical profilometry and x-ray tomography both provide a non-

contact, full 3D representation of the gauge section in the specimen of interest.

Specimens used in this study were all manufactured out of Inconel 718 on a laser powder

bed fusion system and were all built normal to the build plate, typically referred to as the

“z” orientation, as shown in Figure 30. A total of seven different specimen thicknesses

ranging from 0.3 to 2 mm were studied, the former chosen since it represented the thinnest

printable vertical section for the particular design, process parameters, material, and

machine combination used in this work. Section 2 provides the methods used in the study

in more detail. Section 3 presents results of section measurement from each of the four

metrologies and how each influences the estimation of mechanical properties. Section 4

discusses the underlying reasons for the differences between the metrologies, and the final

section 5 summarizes conclusions from the study and closes with a recommendation.

4.2. Methods
The overarching plan followed in this study was as follows: seven specimens ranging

in thickness from 0.3 mm to 2.0 mm were evaluated on each of the four measurement

instruments described above to estimate a cross-sectional area value to be used in the

calculation of stress and key mechanical property metrics. All specimens had an as-built

surface finish and received a stress-relief, HIP, solutionizing, and double-aging heat

treatment. Specimens were first scanned using the CT scanner before being scanned with

52
the blue light scanner. Specimen thicknesses were then measured with a standard

micrometer and a point micrometer six times across the gauge section each to establish an

average; a rectangular cross-section was assumed to calculate a sample cross-sectional

area, with calipers used to estimate the width. Tensile testing was performed, and the data

was analyzed using the different cross-sectional areas obtained with the different

measurement techniques to identify the significance of measurement on the estimated

mechanical properties. In addition to the seven specimens that were part of this round-robin

study, an additional 28 specimens were used for correlative purposes, discussed later.

These 28 specimens (4 at each of the 7 thicknesses) were scanned with the blue light

scanner and measured with the point micrometer only.

4.2.1. Manufacturing

A tensile test specimen geometry was selected based on the ASTM E8 standard [66].

The standard provides various geometries of plate and round specimens. In this study, a

plate type specimen design was selected since it is expected to be more representative of

thin wall behavior (as opposed to a cylindrical specimen) and was designed as shown in

Figure 29. Finite Element Analysis was conducted on this specimen design to ensure stress

concentration factors were under 1.1, which was obtained for a corner radius of 18 mm

between gauge and grip sections of the specimen. Specimen designs were then created for

each of the seven different thicknesses under study (0.3, 0.35, 0.5, 0.75, 1.0, 1.5, 2.0 mm).

53
Figure 29: The final specimen design for this study with 7 specimen thicknesses ranging from 0.3
mm to 2.0 mm (all dimensions in mm)

All specimens used in this study were manufactured on a Concept Laser M2 400W

single laser LPBF machine with Argon as the inert gas. This build had a total of seven

different thicknesses, with 20 replications per thickness, randomized across the build plate.

These seven specimens were pulled out at random from a build of 140 specimens shown

in Figure 30. Virgin, commercially available Inconel 718 powder from Praxair was used

for this study. The laser parameters were previously optimized for maximum density and

minimum surface roughness, with a laser power of 130 W, velocity of 481 mm/s, a powder

bed layer thickness of 30 µm, while using a single contour and a continuous raster scan

strategy (without islanding). Since the emphasis of this work was on measuring cross-

sectional area, preliminary builds were used to optimize laser scan parameters to obtain a

printed section that matched the nominal thickness as closely as possible. Two parameters

were modified towards this objective: trace width and the contour-edge offset. The trace

width parameter is the assumed thickness of the melt pool line and will affect the overall

raster line spacing. The contour-edge offset dictates how close the outer contour is from

the raster turns. A Keyence VR-3200 scanning microscope was used to measure cross-

sectioned specimens, as shown in Figure 31(a). The trace width and contour-edge offset

were chosen to be 130 and 65 µm, respectively to minimize the error between the design

54
and measured values. Figure 31(b) shows the divergence from nominal CAD dimensions

for the different contour-edge offsets evaluated for two different raster angles. This study

selected the 45-degree raster.

All specimens received standard heat treatment including HIP. The samples were first

exposed to a stress relief cycle in a vacuum furnace of an initial ramp to 760°C (1400°F)

for 2 hours followed by a ramp up to 1066°C (1950°F) for 90 minutes. The specimens were

then removed from the build plate with wire EDM before receiving a HIP cycle to 1120°C

(2048°F) for 4 hours with 100 MPa of pressure per the recommendation in ASTM F3055-

14a[30]. The solutionizing and double aging steps were performed in a vacuum furnace

based on the recommendations of AMS2774E [31], where the solutionizing temperature is

954°C (1750°F) with a 1 hour hold time and the double aging step starts at 718°C (1325°F)

for 8 hours followed by 621°C (1150°F) for a total precipitation time of 18 hours.

Seven of the original 140 specimens were randomly selected to conduct a metrology

“round-robin” study, followed by uniaxial tensile testing. All 7 specimens retained an as-

built surface finish with no machining or finishing outside of wire EDM to remove them

from the build plate and a standard HIP, solution treatment, and a double age heat treat

described above.

55
 (a) (b)

Figure 30: (a) Build layout of 140 specimens; (b) with the final print before receiving a stress
relief heat treat cycle.

(a)

(b)
Figure 31: (a) Cross-sectional measurement using a scanning microscope to compare measured
value of thickness against nominal design, shown here for a specimen with a 6mm x 1.5mm cross

56
 section, and (b) Dependence of error in thickness on contour-edge offset for two different raster
angles

4.2.2. Measurement

This work centered around the measurement of the cross-sectional area of as-built thin

wall test specimens for use in stress calculations. Four different metrologies were used for

sample measurement and are discussed here in more detail: standard micrometer, point

micrometer, a blue-light 3D optical scanner, and an x-ray CT scanner (Figure 32).

A. Standard Micrometer

Micrometers are a commonly used metrology for estimating section area for tensile

testing, as they provide reasonably accurate measurements on specimens with a uniform

surface finish. In ASTM E8-16a, recommendations for measurement accuracy can be

found based upon the specimen’s least dimension. For this study, since the thinnest

specimens had a nominal thickness of 0.3 mm, the ASTM E8-16a recommendation is to

have a measurement accuracy to the nearest 0.002 mm. Two types of micrometers were

used in this study: a standard micrometer and a point micrometer. The standard micrometer

is manufactured by Mitutoyo and has a digital readout, with a resolution of 0.001 mm, and

a platen outer diameter of 6 mm with a flat face (Figure 32a). For this study, three

measurements were made with the standard micrometer for each specimen at three points

in the gauge section. The average of these three measurements is reported in the data

discussed here.

B. Point Micrometer

The point micrometer, also manufactured by Mitutoyo, has a digital readout, and a

resolution of 0.001 mm. It features a hardened carbide tip with a 15° measuring point,

57
corresponding to a tip radius of approximately 0.3 mm (Figure 32b). As with the standard

micrometer, three measurements were averaged from the gauge section to obtain the values

used in this analysis.

C. Blue Light 3D Scanning

Blue light scanning has several advantages over conventional contact metrologies such

as full specimen surface data that can be used to detect warpage and morphological defects

when compared to a CAD file, as well as a detailed examination of the cross-sectional area

showing true specimen section shape without the assumptions of perfectly rectangular

cross-sections as calculated with micrometers.

(a) (b)

58
 (c) (d)

Figure 32: The four metrologies used in this study, pictured with the setup used to study the
specimen: (a) standard micrometer with a 6 mm platen diameter, (b) point micrometer with a 15-
degree measuring point, (c) blue light 3D scanner, and (d) x-ray tomography.

The specimens in this study were scanned using a Zeiss Comet L3D 5MP blue light

scanner fitted with a 75 mm lens, paired with the Comet rotary table (Figure 32c). This

scanner uses a projector to cast a fringe pattern of light onto the surface of the part, while

the camera measures the distortions in the lines. A thin film of matte white spray (Hellings

laser scan spray) was used before scanning to reduce the surface reflections, increase point

cloud count, and increase data fidelity – the effect of the thin film on measurement was

also studied and is discussed later. For the 0.3 and 0.35 mm thick specimens, 19 different

scans of the sample were completed while the specimen was held vertically in a specially

fabricated finger vise on the rotary table (Figure 33a). Thicker specimens were scanned in

using 15 different scan angles. For each specimen, about 700,000 data points were obtained

from one grip section through the gauge up to the other grip end. No post-processing of the

blue light data was performed after the scan - instead, the file was exported as a raw data

file containing the x-y-z positions of all surface data points.

59
 Once a point cloud was generated, it was aligned in the Geomagic Control X software

[67] to its respective CAD file for comparison purposes and ease of analysis in Matlab

[68]. A custom script was written in Matlab to systematically analyze the specimen’s gauge

section to determine the exact 3D cross-sectional area in 30 µm increments (Figure 33b

and Figure 33c). The use of a 30 µm section was determined by a study of varying scan

section thicknesses: if the section thickness was too thin, there were insufficient data points

to achieve a strong outline of the cross-section boundary and if the section was too thick,

granularity and accuracy of cross-sectional area estimation compromised. The 30 µm

section was then projected onto a plane containing only the width and thickness

measurements where a 2D boundary function in Matlab was employed to generate a

boundary around the points and calculate the area circumscribed. The Matlab boundary

function has a “shrink factor” input on a scale of 0 to 1 where 0 gives the convex hull and

1 provides a boundary that encompasses all points. A value of 0.8 was chosen as this was

the most sensitive shrink factor input without providing irregularities in the final bounded

area (Figure 33d). For stress calculations, the average cross-sectional area was calculated

through the gauge, spanning 12.5 mm on either side of the center point established from

the nominal CAD design (Figure 33e).

D. X-Ray Tomography
The Zeiss Xradia 520 Versa was used for x-ray Computerized Tomography (XCT)

scanning of the parts. Voltage, power, and target current were set to 140 kV, 10 W, and 71

µA, respectively. The voxel size was 7.14 µm on each side. Approximately seven scans

were needed of each specimen that were then stitched together.

60
 The first step in analyzing the XCT scanning data was to create an x-y-z point cloud of

the specimen surface from the XCT .tiff files (Figure 34a). To do this, a thresholding value

had to be established for each specimen. Using ImageJ [69], average intensities of the

specimen and the surrounding air were measured, and the thresholding value was the

average of those. This was completed 10 times throughout the gauge section of each

specimen to establish a thresholding intensity where anything greater than the threshold

was established as solid material and anything below was considered outside of the

specimen. Other line detection algorithms such as the Sobel operator were explored, but it

was determined that the gradient operation of the Sobel operator acting on the CT image

did a poor job of defining a pixel-exact boundary between part and vacuum. Since the CT

image has a bimodal intensity distribution (between solid material and vacuum) a simple

thresholding value was employed (Figure 34b). A macro was used in ImageJ to

systematically go through the .tiff stack and create a binary text file delineating solid

material from vacuum for each .tiff image. Matlab was used to import the binary text files

and create a 3-dimensional point cloud for each of the 7 specimens (Figure 34c and Figure

34d).

61
(a) (b)

(c)

(d)

62
 (e)
Figure 33: A graphical representation of the workflow of analyzing the blue light data of a 1.5
mm thick specimen: (a) The specimen held in the specimen holder, (b) A 30 µm slice of the blue
light data is first projected onto (c) a plane containing the width and thickness (d) a bounded area
is defined about the points at every 30 µm slice to obtain (e) a plot of cross-sectional area vs.
specimen length.

Two methods were explored for the extraction of the cross-sectional area from the CT

data. The first method employed the same approach as was seen in Figure 33 for the blue

light data (herein referred to as “CT-30” due to the use of 30 µm incrementation throughout

the gauge length). The second method used the binary files created after thresholding to

determine the exact number of voxels that contained “solid” material. This was then

multiplied by the square of the voxel size to determine a cross-sectional area every 7.14

µm along the specimen long axis (herein referred to as “CT-exact”). The second approach

will be the most accurate since it calculates area using the exact number of “solid” pixels

in a slice and it is not determined by the boundary function (which uses a single conforming

2-D boundary that encompasses all data points) over a 30 micron slice of data.

63
 (a)

(b)

(c)

(d)
Figure 34: (a) Raw .tiff image of the CT data, (b) a binary image is created in ImageJ after
thresholding and a text file of the binary image is exported to Matlab, (c) Matlab code was written
to find the edges of the cross-section from the binary file, (d) Iterating through the entire sample
yields a full point cloud of the surface data (small section within gauge shown).

64
 4.2.3. Mechanical Testing

Quasi-static tensile tests were performed on the seven tensile specimens following the

guidelines of ASTM E8 [66] on an Instron 5985 with a 250kN load cell. The tests were run

in displacement control with a cross-head speed of 1.5 mm/min, corresponding to a strain

rate of 1x10-3 s-1. A GOM Aramis 12 MP adjustable dual camera system was used to

optically measure strain using the Digital Image Correlation (DIC) method along the

specimen longitudinal axis. The raw data from the Instron coupled with the DIC strain data

was imported into Matlab for analysis of specimen modulus, engineering yield (0.2%

offset), and Ultimate Tensile Strength (UTS). Young’s modulus was calculated by first

removing all stress values below 100 MPa in the toe region of the test, and then removing

data from the tail until a linear regression could achieve a minimum of 0.99 for the squared

residual (R2).

Data acquisition was divided into two stages. First, images were taken at a rate of 25Hz

for the elastic region followed by a rate of 1 Hz for the plastic region. The greater capture

rate in the elastic region was chosen to ensure a more accurate estimation of elastic modulus

and engineering yield. These images were used to compute full field displacements and

strains using GOM’s ARAMIS software for DIC [70] . Force data was extracted from the

Instron 5985 using a BNC connector. For strain calculation using DIC, the center point of

each specimen was first identified, and a gauge section defined either side of it wherefrom

average strain values could be calculated and reported.

65
4.3. Results
4.3.1. Comparison of Measurement Methods for Section Area Estimation

With the four different measurement instruments, six different methods for calculating

cross-sectional area were defined, including the nominal dimensions. These were:

1. Nominal CAD dimensions

2. Standard micrometer
3. Point micrometer
4. Blue light scanning with 30 µm segmentation
5. CT-30: CT scanning with 30 µm segmentation (to match blue light scanning)
6. CT-exact: CT scanning analyzing each image file (.tiff) as a separate measurable
cross-sectional area

Of these six methods, the CT-exact approach is expected to be the most accurate since

it has the highest resolution, does not pre-suppose a rectangular section (as the micrometers

and nominal dimensions do), and does not rely on the boundary method to circumscribe

point cloud data to measure an area. With the CT-exact approach as the baseline, the other

methods were compared to it and are plotted in Figure 35 as a percent difference in the

cross-sectional area computed against this baseline for the seven different specimen

thicknesses studied in this work.

66
 Figure 35: Cross-sectional area percent differences from CT- exact, as a function of specimen
thickness.

The first observation from Figure 35 is that the standard micrometer greatly

overestimates the cross-section area, particularly at lower specimen thicknesses. The CT-

30 has, unsurprisingly, the greatest agreement with the CT-exact data, with the point

micrometer and the BL scanning data demonstrating deviations under 5% for all but the

thinnest specimen, where the BL scan has a 13.5% deviation in the calculated section area.

The nominal CAD data is provided for reference only, since it is strongly a function of how

laser parameters were selected and optimized, as discussed in section 2.1.

Figure 36 shows two different cross-sections from a 0.3 mm and 2.0 mm thick specimen

for 3 different measurement methods: CT-30, blue light scanning data, and a standard

micrometer (axes not to scale). Figure 36(a) strongly illustrates the shortcomings that the

standard micrometer has for thin, as-built specimens. The standard micrometer effectively

measures the maximum thickness from surface-to-surface between the two micrometer

platens and is strongly affected by warpage and surface roughness, which is why it greatly

67
overestimates section area. Remarkably, in all cases, the point micrometer deviated from

the CT-exact estimation by 5% or less. The blue light scanner much more accurately

captures the warpage as well as larger surface roughness peaks when compared to the CT

data and on average has a lower error from CT-exact than the point micrometer, but also

overestimates section area for the thinnest sections, for reasons that are discussed in section

4.

(a)

(b)

Figure 36: The cross-sectional areas of the middle of the gauge section for the (a) 0.3 mm and (b)
2.0 mm specimen for the blue light and CT-30 data. The average values of thickness and width
measured with the standard micrometer are also shown for the respective specimens. Images not
to scale.

4.3.2. Effects on Mechanical Property Estimation

The goal of defining an accurate section area is ultimately to extract accurate

stresses during mechanical testing. With this in mind, stress-strain graphs were created

for all seven thicknesses for each of the six different cross-sectional area definitions. The

stress-strain curves of the thinnest (0.3 mm) and thickest (2.0 mm) specimens are shown

68
in Figure 37(a) and Figure 37(b), respectively, with Figure 37(c) showing a magnified

version of Figure 37(b) for clarity. The large spread in stress vs. strain data for the 0.3

mm thick specimen in Figure 37(a), relative to the closer agreement for the 2.0 mm thick

specimen in Figure 37(b) is to be expected from the information in Figure 35, since the

errors from the CT-Exact estimations are largest for the thinner specimens. Once again,

for the 0.3 mm specimen, the point micrometer is in remarkable agreement with the CT-

30 data, which is the closest to the CT-Exact data. In general, as thickness reduces there

is a tendency for estimated stresses to be lower than actual, for the micrometer and blue

light scan estimates, since they all tend to overestimate section area.

(a)

69
 (b)

(c)

Figure 37: Stress vs. strain for the (a) 0.3 mm and (b) 2.0 mm thick specimens for the 6
measurement techniques; (c) magnified version of 2.0 mm specimens’ stress-strain response

Figure 38 shows the calculated Young’s modulus, engineering yield, and ultimate

tensile strength for each of the 7 thicknesses and each of the 6 different methods for

calculating cross-sectional area as a function of thickness. The use of a standard

micrometer and caliper for section area estimation greatly underestimates modulus, yield

strength and UTS, in particular as specimen thickness reduces. Blue light scanning and

70
point micrometer estimated results are for the most part in line with CT data. These findings

have implications for studies of the influence of specimen thickness on mechanical

properties, since they show that the metrology used to calculate the section area can distort

the attribution of thickness dependence in observed behavior, and in general make the

dependence appear more significant than it is.

(a)

(b)

71
 (c)

Figure 38: The (a) elastic modulus, (b) engineering yield strength, and (c) engineering Ultimate
Tensile Strength for the 7 different specimens vs thickness for each of the 6 different
measurement techniques.

4.4. Discussion
The results in the previous section are indicative of the fact that while standard

micrometer measurements are significantly inaccurate for thin, as-built LPBF specimens,

point micrometer and blue light scanning perform reasonably well relative to CT data. In

this section, these latter two metrologies are discussed in the context of their limitations

and benefits.

4.4.1. Point and Standard Micrometer

The larger errors in the standard micrometer can be attributed to the contact area of the

platen faces which effectively measure the local maximum thickness between the 2 platen

faces and overestimate specimen thickness, as shown in Figure 39(a). The fine carbide tip

of the point micrometer on the other hand, greatly reduces the point of contact with the

specimen and therefore is more representative of local thickness, as shown in Figure 39(b).

To assess how large the contact area with the point micrometer was, SEM images were

taken of the surface of the Inconel 718 AM part after its use and are shown in Figure 40.
72
The point micrometer contact region is clearly visible, and in fact is strong enough to crush

powder spheres on the surface of the material, but is approximately 100 µm in diameter,

which is an order of magnitude less than the contact area of the standard micrometer. This

difference in contact area explains why the point micrometer approaches CT-exact data

better than the standard micrometer, and why both tend to overpredict area estimation.

While the point micrometer is efficient at estimating accurate section areas, it does so at

the expense of full surface information, which is where blue light scanning is beneficial.

(a) (b)

Figure 39: Schematic showing contact area comparisons between (a) standard micrometer
platens, and (b) point micrometer tips.

73
 (a) (b)

Figure 40: Two SEM images of the surface of an as-printed LPBF specimen after the use of a
point micrometer to measure thickness. Red circles showing approximate contact area of point
micrometer tip.

4.4.2. Blue Light Scanning Deviations from Point Micrometer

Results from blue light scan data show that it is comparable to the point micrometer for

all specimen thicknesses except for the thinnest specimen (Figure 41a), where percentage

deviations from CT-exact are plotted for each of the seven thicknesses for point micrometer

and blue light scanning. To investigate whether the 13% error for blue light scanning for

the 0.3 mm specimen was a systematic error, a matching study was conducted with four

additional replications at each of the 7 thicknesses, this time only measured with the point

micrometer and blue light scanning. The correlation between the cross-section areas

estimated using these two metrology methods is shown in Figure 41(b).

74
 (a)

(b)

(c)
Figure 41: (a) Comparison of differences between the blue light scan and the point micrometer
relative to CT-exact, as a function of nominal thickness, (b) correlation plot of blue light and
point micrometer section area computations, and (c) delta chart showing deviations from nominal
for both metrologies

While there is general agreement (relative to the dotted 1:1 line), the blue light

estimates tend to be higher. A delta chart shown in Figure 41(c) plots the deviation in
75
thickness estimates from their nominal values to estimate a mean difference between point

micrometer and blue light for all thicknesses studied, and it shows that the blue light

measurements are on average, 28.5 µm thicker than the point micrometer. This amounts to

a 1.4% deviation for the thickest 2 mm specimen but is as much as 9.5% for the 0.3 mm

specimen. As such the higher error at low thicknesses is not particularly attributable to a

thickness-specific physical behavior. The overestimation of section area for blue light

scanning is partly attributable to the use of the matte white powder that is sprayed onto the

sample surface to improve data capture, without which large losses in data capture were

observed. A study was done on measuring the thickness increase of blue light scanned

surfaces before and after applying the surface spray on a 0.5 mm thick ceramic calibration

gauge block. The block was scanned before and after applying the typical spray coating.

The scans were then aligned to each other, as shown in Figure 42(a). Once this alignment

was completed, the resulting surfaces were analyzed relative to each other as shown in

Figure 42(b). It was found that by spraying the surface, the thickness increased by 9.2 µm

on average, which represents a third of the difference between point and blue light scanning

data. Further, it is likely that the spray could be thicker for a rough surface with peaks and

valleys as is the case for as-built LPBF specimens, in comparison to a relatively smooth

ceramic block. It may thus be speculated that at least half the difference in thickness seen

in Figure 41(c) results from the spray. A second source of error in the blue light scanning

data is in the boundary method used in Matlab that defines the cross-sectional area (Figure

33d) for a given point cloud (Figure 33c). Finally, the slight plastic deformation noticed

76
for the point micrometer (Figure 40) could explain the final reason for the difference

between the two measurements.

(a)

(b)

Figure 42: (a) The aligned point cloud data of the ceramic thickness calibrant before and after
scanning – the additional data in the width direction is an artifact of data trimming; (b) The
projection of the thickness with and without spray, clearly showing a slight increase due to spray.

4.4.3. Section Area Variation Within Gauge

An advantage of scanning the gauge section of thin wall specimens is that it enables

the calculation of areas for multiple sections. A stack of these sections is shown in Figure

43, obtained by collapsing point cloud data in 30 µm slices onto a single section plane.

This now enables the calculation of 833 areas over the 25 mm gauge length under study.

Calculation of the mean and standard deviation for these areas enables the estimation of a

Coefficient of Variation (CV), which is the ratio of standard deviation over mean. CV can

then be plotted in terms of percentage, as a function of nominal thickness as shown in

Figure 44. This result demonstrates that the CV greatly increases as the thickness drops
77
below 0.75 mm, and is a remarkable result enabled by the use of full 3D scanning of the

gauge section and one that would be harder to estimate with the point micrometer.

Figure 43: Stack of blue light scan measurements for three 30 µm slices in the gauge

section.

Figure 44: Coefficient variation (%) for the section area estimation for all slices as a function of
nominal thickness

4.5. Conclusions
This work evaluated four metrologies for the determination of the cross-sectional area

to be used in stress calculations used in estimating mechanical properties for Inconel 718

78
LPBF AM, thin-walled, as-built parts. This work focused on the accuracy of each

metrology, measuring deviations from the highest resolution CT scan data available, but

speed and cost are also important considerations for practical implementation. While the

CT data has the greatest resolution amongst all the metrologies evaluated, it has the longest

scan times and is cost-prohibitive for 100% sampling. Blue light scanning, while accurate

for all but very thin sections, also takes approximately 20 minutes per specimen to collect

the data necessary, followed by offline data processing to estimate an average section area.

The use of a micrometer is the quickest and cheapest way to obtain specimen thickness but

the standard micrometer is inadequate for the task, and it is the point micrometer that had

the best performance. Finally, a key element in understanding the contribution of specimen

geometry to any observed differences in mechanical behavior (particularly strength) is to

obtain a full scan of the gauge section, for which blue light scanning has been shown to be

capable since it models the entire surface in a non-contact manner. A qualitative grading

of the 4 different metrologies along the four factors of interest can be found in Table 3.

Table 3: A table to qualitatively grade each of the 4 metrologies for purposes of calculating cross-
sectional area. The greater the circle is filled in, the better it achieves the respective metric.

In summary, this work shows that the point micrometer is the compelling choice of

metrology for measuring section area for use in stress computations for as-built LPBF

79
thin wall specimens up to 0.3 mm in thickness. Blue light scanning is valuable when the

intent is to model the entire surface to try and correlate differences in area to observed

failure behavior, but care must be taken at the thinnest sections when artifacts in the

scanning and data analysis can introduce errors in the measurement.

80
 CHAPTER 5

DECOUPLING THE CONTRIBUTIONS OF MORPHOLOGY AND

MICROSTRUCTURE: A MESO-SCALE SIZE EFFECT STUDY

5.1. Introduction
5.1.1. Causes of Size-Effects

The central research objective in this thesis is to examine how and why mechanical

properties change for as-printed LPBF specimens as a function of thickness. This chapter

aims to answer this question. Before examining the data itself, it is useful to consider all

the different reasons one may expect to see size effects in these specimens over the range

of interest in this thesis (0.3mm to 2mm). Figure 45 identifies the different factors that

might influence the thickness effects in AM properties, dividing it into three aspects:

artifacts, morphology, and microstructure.

81
 Figure 45: A flow chart describing the different potential variables that will affect the thickness
dependance on the meso-scale AM parts. The main contributing factors can come from printing
and process artifacts, the external structure of the specimen, and the internal structure of the
specimen.

Artifacts here addresses those aspects of the measured data that are not attributable to

a true physical cause but derived from the manufacturing processes and measurement

methods used to estimate properties – which can be themselves influenced by specimen

size. The objective of the prior chapters was to quantify and minimize the effects of these

artifacts, by addressing specimen design, optimizing process parameters (Chapter 2), and

developing metrology methods for accurate section area calculation (Chapter 4). This

chapter will be examining the influence of the two other main aspects of size effects:

morphology and microstructure. The morphology represents the geometry of the as-printed

surface, and the internal microstructure addresses aspects such as grain size and
82
distribution, phases and porosity. This chapter will also examine how HIP influences each

of these, and in turn the mechanical behavior of the specimens. All these effects are

examined in the context of quasistatic tensile behavior, the following chapter addresses

high cycle fatigue.

5.1.2. Prior Work on Size Effects in Metal AM

While there is significant data on the properties of machined metal AM specimens [9],

[10], [11], there is also a need for estimating properties in as-built parts to better understand

how AM can make its way into the design space where machining surfaces is not an option

(in this work, “as-built” describes the state of the AM specimen without any mechanical

post-processing such as machining, honing, or polishing conducted to the surface –

independent of heat treatment). At low thicknesses (defined here as under 2 mm), it is not

evident that mechanical properties are the same as bulk properties measured from

specimens with larger dimensions. Literature on thickness effects of metal AM properties

in an as-built condition has reported trends such as reductions in Ultimate Tensile Strength

(UTS) with reduction in specimen thickness [17] [13] [14] [15] [16]. None of these prior

studies have looked at the effect of thickness specifically in Inconel 718, nor have they

isolated the effect of HIP in modulating any thickness dependencies. Most studies do not

adequately address the accurate computation of section area that is used in the estimation

of stress quantities, as discussed in the previous chapter. Finally, most studies involve

limited sample sizes, and do not conduct statistical analyses to identify significance in

variations observed, nor seek to develop mathematical models relating properties to

thickness.

83
 Specific to AM Inconel 718, Figure 46 compiles a literature review of 20 publications

that examined mechanical properties, specifically reporting UTS, which is plotted against

the cross-sectional area of the sample. Other papers have been published that report

mechanical testing data for AM Inconel 718, but they were not added to the list if the

specimen dimensions and the UTS value were not reported. Note, when collecting data

from sources with multiple values of UTS based on different test parameters (e.g., different

print orientations, or slightly varying heat treatment temperatures, etc.), the value that was

selected for this literature review was the one that matched our data as closely as possible

(stress relief + solution + age and vertical orientation). All references used specimens that

were machined – no as-printed Inconel 718 parts found in literature. The cross-sectional

area studied in this work is indicated in red. None of this prior work performed a size effect

study, all specimens were machined to size and as such did not consider the effects of as-

printed surfaces – and only one publication (see Ref. C in Figure 46) tested specimens that

had a smaller cross-sectional area than considered for this work.

84
 Figure 46: A literature review of additively manufactured Inconel 718 for LPBF, DLD (Direct
Laser Deposition), and EBM (Electron Beam Melting). All references used specimens that were
machined. [71]–[89]

This work studies the effects of changing specimen thickness (for seven different

thickness values ranging from 0.3-2 mm) for two different heat treatment conditions (with

and without a HIP step) for LPBF Inconel 718. A total of 12 specimens per thickness and

heat treatment condition are tested (a total of 168 specimens), and statistical analyses are

conducted to identify the significance of any differences observed. Finally, this study also

considers how Hot Isostatic Pressing (HIP) influences these differences, asking if HIP can

serve as a way to ameliorate mechanical debits in thin wall structures. HIP is a common

industry heat treat practice for Inconel 718 LPBF parts and is conducted before the

solutionize and double age cycle for maximization of part strength [90][91]. The HIP heat

treatment is used to close pores within the specimen that are potentially detrimental to

mechanical performance such as fatigue strength and ductility. Due to the high temperature

of the HIP cycle, it is common to see changes in microstructure from before and after the

85
HIP step; grain recrystallization and greater precipitate solutionizing can also be seen [92].

This data forms the basis for a mathematical model that is proposed to relate properties to

thickness and HIP condition.

5.2. Methods
5.2.1. Specimen Design and Manufacturing

All specimens were manufactured using the Laser Powder Bed Fusion (LPBF)

process [93], on a Concept Laser M2 machine (400W, single laser) over the course of two

consecutive builds. Virgin powder was used for the first build, and the powder was recycled

for use in the second build, with argon used as the inerting gas. LPBF process parameters

were optimized for maximum density using standard 10mm cube experiments beforehand.

Special care was taken to optimize contour offsets to obtain the best dimensional accuracy

in thin wall structures, as well as establish a range of printable thicknesses by conducting

a series of experiments, summarized in the Preliminary Builds and Lessons Learned (page

11). At the end of these experiments, a laser power of 130 W, scan velocity of 481 mm/s,

and a build height of 30 µm were selected as the primary printing parameters. The scan

strategy consisted of a 45° raster, relative to the thin wall edges, rotating 90° every

successive layer, with a single contour pass along the edges.

Tensile specimens were designed based on the ASTM E8 standard [66] but needed

to be adapted to accommodate both the printer build volume as well as the range of

thicknesses evaluated. A rectangular section geometry was selected for this study to better

represent thin wall behavior. Finite Element Analysis (FEA) was used to ensure any

deviation from the standard still produced a tensile test specimen with stress concentration

86
factors under 1.1, see Preliminary Builds and Lessons Learned . The sample geometry can

be found in Figure 6. The tensile specimens were rectangular with varying thicknesses and

a constant width of 6 mm in the gauge section. Seven different thicknesses were printed

with nominal dimensions of 0.3, 0.35, 0.5, 0.75, 1.0, 1.5, and 2.0 mm, spread out over two

builds. Specimens of different thicknesses were randomized by thickness across the build

plate as shown in Figure 8(a), and emerged from the printer without any gross defects, as

shown in Figure 8(b). To minimize the influence of build-to-build variation on the data,

each of the two builds were split in half to accommodate the two different heat treatment

cycles that the specimens were subjected to.

All information on the specimen heat treatment cycles can be found in Section 2.4.

Heat Treatment. To recap, all specimens tested have undergone 1 of 2 different heat

treatment cycles – no specimens were tested with the weaker as-printed microstructure.

The 2 different heat treatment cycles are either called “HIP” or “Non-HIP/No-HIP”; the

HIP set of specimens received a stress relief, HIP cycle, solution treatment, and a double

age (SR+HIP+SA), while the Non-HIP set of specimens received a stress relief, solution

treatment, and a double age (SR+ SA).

5.2.2. Characterization and Mechanical Testing

Density

Density measurements were made with an analytical scale balance and a density

measurement kit that allows the user to make mass measurements of the specimen in the

air and submerged in a fluid of known density. All the details of the measurement setup

87
and the results of a measurement capability analysis on AM cube specimens and AM thin

wall specimens have been published by Bruce et al. [94], and can be found in Chapter 3.

Surface Roughness

The surface roughness of the specimens was measured with a Keyence VR-3200

scanning microscope. All surface roughness measurements were taken in the grip section

of the specimen since the gauge section was not wide enough to support solid data

collection without edge effects distorting the surface roughness results. The surface

roughness evaluation area comprised of a 5 mm diameter circle. A Gaussian filter and a

short pass filter were used on the data before obtaining the results. Sa, Sq, Sv, and Sz were

all collected for each specimen, but for this work, only the results of Sa will be shown.

Polishing and Microscopy

Sample polishing followed a routine procedure of using silicon carbide papers up

to 600 grit (250 RPM), then using a polycrystalline diamond suspension of 9 μm, 6 μm,

and 1 μm (125 RPM), and finishing with a 0.05 μm alumina slurry (125 RPM). Specimens

were etched for grain relief with a variant of Kalling's 2 waterless reagent with a mixture

of 2 mL methanol, 6 mL HCl, and 0.3 g of CuCl2. SEM images were collected with a Zeiss

Auriga scanning electron microscope.

Mechanical Testing

Mechanical testing was performed on an Instron 5985 universal testing machine

with a 250 kN load cell. Tensile tests were run in displacement control with a crosshead

speed of 1.5 mm/min, resulting in an effective strain rate of ~0.8 x 10-3 s-1 over the 32 mm

reduced gauge section. Strain data was collected with an Advanced Video Extensometer

88
(AVE) that tracks the relative displacement of two dots placed in the gauge section 25 mm

apart. For calculating stress, specimen dimensions were measured three times and averaged

with a digital point micrometer for the thickness and a digital caliper for the width, per

findings in a separate study (see Chapter 4) assessing the appropriateness of point

micrometer usage for thin wall LPBF specimens relative to estimates obtained from X-ray

tomography scans [95]. Use of surface or edge contact metrologies for the measurement of

as-built wall thickness, as has been the case in some of the literature, inflates the estimation

of section area. This in turn results in significantly lower reported stresses at low

thicknesses, where measurement errors dominate.

All data analysis was performed on the output load-displacement data in Matlab using

a custom script to generate curves and calculate key material metrics. The modulus was

calculated by first removing any data below 100 MPa in stress, and then removing data

starting at the toe until a linear regression of 0.98 was achieved. The yield strength was

then calculated by using the 0.2% offset method leveraging the calculated modulus.

Blue Light Scanning

Blue light scanning data was collected for every specimen with a Zeiss Comet L3D

5MP blue light scanner fitted with a 75 mm lens. The blue light scanning process generates

~700,000 data points per specimen which can be used to analyze and inspect the surface

topography of each specimen systematically using scripts in Matlab. All the data generated

from the blue light scanner is kept as raw with no post processing outside of aligning the

scans to themselves and orienting them to the CAD model of the specimens for ease of

analysis. Once the raw blue light data is aligned, the sample cross-section is examined in

89
30 μm intervals throughout the gauge section to measure the sample mean, minimum, and

coefficient of variation (CV) of the cross-sectional area. A comprehensive examination of

the blue light scanning process to measure cross-sectional area for thin-wall parts with an

as-printed surface can be found in the paper by Paradise et al. [96]. A more detailed

examination of the process of collecting and analyzing the blue light scanned data can be

found in Chapter 4.

5.2.3. Finite Element Analysis

ANSYS Mechanical was used to conduct a Finite Element Analysis of the quasi-

static tensile tests for 0.3mm, 0.35mm, 0.5mm, 0.75mm, 1mm, 1.5mm, and 2mm

specimens. The geometry for each specimen was created using Blue-Light scan data of the

specimen. The Blue-Light scan data was imported into ANSYS Spaceclaim as an .OBJ file

(Figure 47a). SpaceClaim was used to convert the surface data from the .OBJ file into a 3D

solid body (Figure 47b). This was done using the “Shrinkwrap” feature in SpaceClaim. The

“Shrinkwrap” feature takes an object and generates a surface from its exterior. This process

helped remove any holes and spurious points in the scan data. The “Shrinkwrap” surface

was then converted into a 3D solid object and imported into ANSYS

Mechanical.Experimental Plan.

90
 (a)

(b)
Figure 47: (a) OBJ surface body with a closeup and (b) 3D solid body with a closeup.

For each specimen thickness, a custom material model was created from the uniaxial

tensile test data from that thickness. The custom material model was created by taking the

uniaxial test data and applying it to an Isotropic Elasticity Model for the elastic regime and

a Multilinear Isotropic Hardening model for the plastic regime. The specimen was

constrained with two boundary conditions at the locations where the specimen is fixtured

into the tensile test (Figure 48). A “Fixed Support” boundary condition was used at the

base of the specimen and a “Displacement” boundary condition was used at the top. The
91
“Displacement” boundary condition was used to apply the tensile load on the specimen of

1.5mm/s until ultimate strength was reached.

Figure 48: Applied boundary conditions.

The Finite Element mesh was created using linear tetrahedral elements. A mesh

refinement study was conducted to refine the mesh to a body element size of 0.1mm which

resulted in a total mesh size of 1,201,694 elements and 1,876,072 nodes for the 0.3mm

thickness specimen.

5.2.4. Test and Post-test Protocol

A total of 168 AM specimens were examined and tested in this work spanning across

7 different thicknesses (0.3, 0.35, 0.5, 0.75, 1.0, 1.5, 2.0 mm) and 2 different heat treatment

schedules (specimens that received a full standard heat treatment with and without a HIP

cycle, herein referred to as “HIP” and “non-HIP”, respectively). A total of 12 specimens

per thickness and heat treat condition were examined and tested for robust statistical

analysis. Before the tensile test, all specimens were examined for their relative density,

surface roughness, and were blue light scanned to obtain a full 3d model of the surface

92
topology. Outside of the 168 test specimens, several other specimens were sectioned and

polished to obtain information on pore distribution, grain size, and precipitates. All 168

specimens were then tensile tested. Post-mortem fractography was done on select samples

after. Alongside the AM parts, 168 sheet metal specimens were also tested across 7

different thicknesses (0.41, 0.51, 0.64, 0.82, 1.02, 1.6, and 2.03 mm) and the 2 different

heat treatment schedules (HIP and non-HIP).

The scope of this work is to elucidate the thickness effects on the mechanical properties

and performance of thin-walled, heat treated, AM, Inconel 718 with an as-printed surface.

Secondly, this work is set to establish if a HIP cycle before the final aging treatment can

help mitigate thickness dependance and increase the part performance by closing the

internal pores of the thin-walled specimens.

5.3. Results
5.3.1. Archimedes Density

The absolute density values found through the Archimedes density testing method are

shown in Figure 49 (these were previously shown in Figure 26, but have been included

here for ease of reading). The non-HIP specimens show a small amount of thickness

dependance, while the HIP specimens show a strong increase in the part density on a per

nominal thickness basis. While the non-HIP specimens have a large degree of variability

to the part densities, the HIP process helps to eliminate the density variation for the parts

and increase the part density overall. The standard deviation of density for the HIP parts

drops significantly for thicknesses of 0.75 mm and greater, as seen in Figure 49. The greater

variation in part densities for the thinner specimens is partly on account of increase in

93
measurement uncertainty in the metrology as specimens. This is attributed to the increase

in surface area to volume ratio as the specimen thickness reduces. However, greater

variation in part density for the thinnest specimens is also expected since the melt pool

dynamics in thin wall structures result in more chaotic, and unpredictable keyhole porosity

as the laser is constantly accelerating and decelerating in rapid succession within a

narrower confine.

Figure 49: The absolute density values of the AM parts for the 2 different heat treatments across
all of the different thicknesses tested. The HIP heat treatment significantly reduces the standard
deviation of the test specimens for the thicker parts.

5.3.2. Surface Roughness

The surface roughness values (Sa) for all 168 specimens can be found in Figure 50. It

was found that the printing parameters that were optimized for this study produced parts

94
with minimal surface roughness that were in good control. Only the thinnest specimens

showed a minor increase in the Sa values. Although the reason is not entirely clear, the HIP

process tends to increase the Sa values on a nominal thickness basis. One hypothesis for

the increase in HIP part roughness is due to the majority of porosity being formed within

~200 μm from the surface (more on this in the next section). As the HIP process closes the

sub-surface porosity it changes the surface morphology which will affect the Sa values.

Figure 50: The Sa values for the HIP and non-HIP specimens. All surface roughness values are in
close approximation, but in general, the HIP parts have a greater surface roughness.

95
 5.3.3. Microstructure

The microstructure evolves significantly from the as-built AM parts to the two unique

heat treatments of the HIP and non-HIP specimens that are tested in this work. Figure 51

(a & d) shows the initial microstructure of the as-printed AM material with no heat

treatment, the non-HIP specimen microstructure (b & e), and the HIP specimen

microstructure (c & f)). The printing orientation for Figure 51 (a, b, & c) is up, while the

printing direction for (d, e, & f) is into the page.

The classic “fish scale” melt pool boundary pattern can be found in Figure 51(a) along

with long grains that are pointing the in the build direction. Figure 51(d) shows the melt

pool boundaries with the 45° raster trace going across the specimen. For the fully aged non-

HIP specimens, the grains are reminiscent of the as-built specimen with the long, high-

aspect ratio orientation. Figure 51(e) shows a duplex grain arrangement that is an artifact

of the raster scan path with large grains (that formed in the center of the melt pool path)

that are surrounded by very fine grains created via rapid solidification on the edge of the

melt pool. The non-HIP specimens also show a large degree of grain refinement near the

surface from the contour scan that takes place after the interior of the specimen is cooled

and crystalized. Figure 51(c) and Figure 51(d) shows the HIP specimen grain structure. In

both sets of images, significant recrystallization has taken place as the grains are much

more equiaxed and uniform as there is an elimination of the high aspect ratio grains and

the duplex grain structure. An almost complete elimination of the pores can be seen in the

HIP specimens. Most of the pores that are generated in the parts examined can be found

either close to the bottom of a melt pool and/or at the end of the raster scan path as it makes

96
a turn. Most of the porosity is found to be ~200 μm from the sample surface which is

coincidence of the raster turn in the print; this is a phenomenon that others have studied

[37].

Figure 52 shows the microstructure for the 0.3 mm specimen (the thinnest specimen

examined). Figure 52 (a and b) shows that the non-HIP specimen has a significant amount

of porosity forming along the center of the sample, and shows a distribution of very long,

high-aspect ratio grains forming in the build direction - due to the melting-remelting and

heat conduction through the bottom of the specimen - as well as very fine grains in between.

The HIP specimen Figure 52(c and d) shows that the porosity is significantly reduced or

eliminated with a much more uniform grain size throughout.

SEM imaging was used to better understand the precipitates formed under the two

separate heat treat conditions. Figure 53 and Figure 54 shows images of increasing

magnification with the build direction oriented to the right and into the page, respectively,

for the 2 different sets of heat treat specimens. Figure 53(a-c) show that the non-HIP

specimen has a lot of generated precipitates along the grain boundary and that the internal

porosity does not affect the microstructure. Figure 53(d-f) shows that the higher

temperature HIP step helps to solutionize some of the precipitates found along the grain

boundary back into the matrix and creates more annealing twins in the specimen. Figure

54(a-c) again show the large number of precipitates that form along the grain boundary

while Figure 54(c) shows a finer sub-micron cellular structure that is typically presumed

to be left over from the dendrites formed in the printing process which has been found to

act as a strengthening mechanism [24]. Figure 54(d-e) again shows that the HIP step helps

97
to mitigate the number and/or size of precipitates formed along the grain boundaries. The

higher temperature HIP step also eliminates the sub-micron cellular structure and increases

the number of annealing twins.

Figure 51: (a,b,c) The as-built, non-HIP, and HIP grain structure, respectively, for the build
direction pointing up. (e,f,g) The as-built, non-HIP, and HIP grain structure, respectively, for the
build direction pointing into the page. Significant grain recrystallization can be seen after the HIP
cycle, while the non-HIP specimens retain more of the as-built porosity and grain size
(distribution and aspect ratio).

98
 (a) (b) (c) (d)

Figure 52: Micrographs for the non-HIP and HIP specimens that are 0.3 mm in thickness. (a and
b) show the grain sizes and aspect ratio for the non-HIP specimens with the build direction
pointing up and into the page, respectively. (c and d) show the grain sizes and aspect ratio for the
HIP specimens with the build direction pointing up and into the page, respectively. The porosity
generated in the non-HIP specimen is typically closer to the center of while the porosity is
significantly reduced/eliminated in the HIP specimen. Significant annealing and homogenization
can be seen in the HIP specimen.

99
Figure 53: For all images, the build direction is pointing to the right. The non-HIP microstructure
for increasing magnification (a,b,c). The HIP microstructure for increasing (e,f,g). The HIP step
shows a significant amount of precipitate dissolution along the grain boundaries and an increase
in annealing twins.

100
 Figure 54: For all images, the build direction is into the page. The non-HIP microstructure for
increasing magnification (a,b,c). The HIP microstructure for increasing (e,f,g). The HIP step
shows a significant amount of precipitate dissolution along the grain boundaries and an increase
in annealing twins. A tighter, cellular structure can be seen in (c) which can be attributed to
increase strengths in AM parts

101
 5.3.4. Mechanical Behavior

Figure 55(a) plots a general representation of the stress-strain response for the non-

HIP and HIP specimens across the 7 different thicknesses; shown are 28 of the 168 tested

AM specimens with 2 curves per thickness and heat treat condition (black, dashed line

represents the non-HIP specimens, while the red line represents the HIP specimens). The

HIP specimens will typically have a greater UTS and elongation value over the non-HIP

counterpart. Qualitative observations of mechanical behavior for the two extreme

thicknesses for both HIP conditions immediately suggest differences in stress-strain

response by thickness and HIP. The stress-strain flow curves above 600 MPa have been

plotted with the shaded area representing the envelope of the results for 12 different tests

run for each of the 2.0 and 0.3 mm specimens up to 6% strain for each of the two HIP

conditions shown in Figure 55b. The non-HIP specimens appear to have a greater yield

value and a larger disparity between the two specimen thicknesses, suggesting that HIP

reduces thickness dependency. The average values of the 4 main metrics examined in the

work (modulus, yield, UTS, and elongation) with the true UTS are tabulated in Table 4

with the standard deviation of the values in parentheses. All values are broken up by

manufacturing method, heat treatment, and nominal part thickness. All averages and

standard deviations are based on 12 tensile tests for each metric.

102
 (a)

(b)
Figure 55: (a) The stress-strain curve for 28 non-HIP and HIP specimens (2 curves per thickness
and heat treatment) with the red solid line representing the HIP specimens and the black, dashed
lines representing the non-HIP specimens. The black inset box represents the stress-strain
location for (b) the enveloping stress-strain flow curve bounds are plotted up to 6% strain

103
 Table 4: The average values for modulus, yield, UTS, true UTS, and elongation by specimen
manufacturing, heat treatment, and nominal thickness with the standard deviation in parentheses.

Nominal
Heat Thickness Modulus Yield UTS True UTS Elongation
Manufacturing Treat (mm) (GPa) (MPa) (MPa) (MPa) (%)
0.3 167.78 (8.48) 1094.50 (18.44) 1306.97 (15.82) 1443.00 (30.17) 10.65 (2.23)
0.35 164.45 (6.68) 1098.05 (16.85) 1326.83 (18.40) 1544.88 (28.78) 16.84 (1.86)
0.5 161.69 (6.24) 1104.36 (17.62) 1342.02 (12.37) 1599.95 (18.59) 20.86 (1.38)
Additive Manufactured

HIP

0.75 160.58 (7.33) 1111.99 (18.34) 1346.90 (12.45) 1614.10 (12.37) 24.92 (0.29)
1.0 159.80 (4.75) 1107.51 (17.81) 1343.07 (11.32) 1613.67 (9.87) 25.20 (0.99)
1.5 159.08 (3.15) 1107.68 (17.47) 1340.00 (12.04) 1607.19 (9.37) 25.76 (0.90)
2.0 166.02 (6.20) 1115.48 (10.97) 1349.97 (7.04) 1616.64 (6.51) 26.23 (0.60)
0.3 141.53 (6.31) 1109.80 (11.85) 1287.14 (17.35) 1419.37 (45.72) 10.40 (2.74)
0.35 147.09 (6.17) 1113.55 (10.09) 1299.22 (11.94) 1468.57 (29.72) 13.39 (1.90)
Non-HIP

0.5 146.25 (4.59) 1136.73 (10.62) 1324.26 (10.71) 1533.05 (28.06) 17.19 (2.29)
0.75 134.08 (4.53) 1159.51 (11.21) 1329.75 (10.67) 1553.59 (16.48) 20.61 (0.98)
1.0 138.19 (4.19) 1158.05 (6.51) 1318.97 (6.54) 1527.26 (14.46) 20.16 (1.17)
1.5 141.80 (7.33) 1174.81 (8.78) 1332.05 (6.13) 1544.80 (21.24) 21.34 (1.91)
2.0 153.03 (8.18) 1183.93 (15.58) 1337.02 (9.59) 1539.61 (8.89) 19.97 (1.34)
0.41 168.21 (8.19) 1102.85 (5.42) 1346.21 (5.75) 1596.28 (16.94) 19.36 (1.45)
0.51 166.70 (8.85) 1102.99 (13.06) 1305.09 (7.03) 1554.01 (20.38) 20.95 (2.06)
0.64 162.52 (5.65) 1077.65 (8.13) 1281.87 (5.33) 1554.66 (9.07) 23.97 (1.33)
HIP

0.81 164.78 (6.50) 1086.18 (12.49) 1296.22 (15.39) 1581.11 (17.00) 25.70 (1.12)
1.02 163.01 (4.14) 1092.25 (5.85) 1298.18 (5.40) 1588.77 (7.88) 27.31 (0.86)
Sheet Metal

1.60 165.76 (5.00) 1091.32 (7.45) 1288.25 (4.74) 1573.88 (6.97) 27.10 (1.04)
2.03 166.07 (4.61) 1042.32 (4.91) 1250.33 (3.60) 1554.75 (8.56) 30.29 (1.14)
0.41 176.25 (20.84) 1073.10 (16.54) 1296.89 (9.91) 1492.49 (30.07) 15.44 (2.00)
0.51 167.11 (8.46) 1032.05 (8.31) 1231.47 (9.10) 1444.73 (23.87) 17.87 (1.45)
Non-HIP

0.64 161.36 (3.88) 1017.68 (7.11) 1196.50 (7.74) 1380.08 (30.42) 15.73 (2.16)
0.81 163.51 (6.55) 1013.56 (8.03) 1211.07 (9.14) 1427.16 (33.92) 18.41 (2.35)
1.02 168.34 (7.76) 1027.88 (7.46) 1224.46 (7.54) 1478.99 (28.75) 21.98 (2.15)
1.60 168.02 (6.44) 1007.28 (9.45) 1211.33 (6.24) 1502.39 (16.10) 26.14 (1.56)
2.03 167.16 (6.17) 989.11 (6.55) 1189.81 (6.19) 1479.53 (21.38) 26.24 (2.00)

5.3.5. Fractography

A series of fractographic images were taken to better understand the fracture mechanics

for the different specimens across thickness and heat treat conditions. Figure 56 shows the

fracture surface of the 0.3 mm non-HIP and HIP specimens. Figure 56(a) shows that the

majority of the pores generated for the thinnest specimens can be found near the center as

the laser raster for melting the Inconel 718 powder is constantly being accelerated and

decelerated in rapid succession which can lead to an increase of keyholing porosity

104
generation. Figure 56(b) shows most of the porosity is removed through the HIP process.

Figure 56(c) shows a that the non-HIP specimen has areas of a shallow dimpled surface –

indicative of a typical transgranular failure mode – and areas that exhibit signs of a brittle

cleavage fracture, specifically in between the internal porosity of the sample. Figure 56(d)

shows far more areas of a dimple like surface which indicates greater ductility over the

non-HIP specimen.

Figure 57 shows a series of fractographic images for a 2 mm non-HIP ((a), (c), (e),

and (g)) and HIP ((b), (d), (g), and (h)) specimen. Figure 57(a) shows that the non-HIP

specimen generates the majority of the porosity near the surface while the overall fracture

topography is relatively flat when compared to the HIP specimen. Figure 57(b) shows high

angle fracture behavior at the top and left-hand side of the specimen, indicative of a

specimen with higher ductility and elongation. The large pores have been collapsed

revealing a much denser fracture surface. For Figure 57(a) and (b), the red square shows

the inset image location for Figure 57(c) and (d), while the blue, dashed line shows the

approximate location of the fracture profile shown in Figure 57(e) and (f). The fracture

profile was exposed by grinding the surface to the interior of the specimen. Figure 57(c)

and (d) show a dimple surface indicating significant elongation until specimen failure by

nucleation of voids. Figure 57(d) shows that the HIP specimen has a greater dimple density

with deeper lips indicating that the HIP specimen has greater ductility over the non-HIP

specimen. Figure 57(e) and (f) are macroscopic images of the fracture profile with the non-

HIP specimen showing a significant amount of elongated porosity since the section is taken

near the surface where most of the porosity is formed. The left and right-hand side of Figure

105
57(f) shows that the fracture surface is approximately at a 45° angle to the pulling direction

which is typical of a ductile specimen. Figure 57(g) and (h) show the closeups of the

fracture profile. A significant number of voids are nucleating around the hard precipitates

that reside along the grain boundary for the non-HIP specimen; in fact, complete

decohesion can be see along some of the grain boundaries. Intergranular and transgranular

fracture behavior can be found in both non-HIP and HIP specimens, but it is to be expected

that more intergranular behavior will be found with the non-HIP specimen with the weak

grain boundaries.

106
Figure 56: The macroscopic fracture surface for a 0.3 mm non-HIP and HIP specimen (a) and
(b), respectively. A closeup of the fracture surface for the non-HIP and HIP specimen (c) and (d),
respectively.

107
Figure 57: A series of fractographic images for a 2 mm non-HIP ((a), (c), (e), and (g)) and HIP
specimen ((b), (d), (g), and (h)). Series (a) and (b) show the macroscopic fracture surface where the
blue, dashed line indicates approximate location of the fracture profile shown in (e) and (f). Series
(c) and (d) show a closeup of the fracture surface. Series (g) and (h) shows a closeup of the fracture
profile.

5.3.6. Finite Element Analysis

Finite Element Analysis was used to examine the role of the part’s surface

morphology on the internal stress fields. By using the blue light scanned surface
108
topography data and converting it into a meshable solid, FEA can then be run on the 3D

model while imparting correct boundary conditions and load requirements to simulate the

quasi-static tensile test. Figure 58 shows the results of the FEA analysis for a (a) 2 mm, (b)

0.75 mm, and (c) a 0.3 mm nominal thick specimen, that has been loaded to the model’s

UTS. FEA analysis was not performed on both HIP and non-HIP specimens, since there is

no strong reason to believe the results will be vastly different despite the surface roughness

values being slightly higher for the HIP set of specimens. Regardless, the FEA was

performed on a HIP set of specimens. Figure 58(a) shows that the equivalent total strain

(left-hand side) has large areas of equivalently high strain, while the right-hand side shows

that the cross-section with the highest maximum principal stress is very uniform throughout

with no strong stress localization. However, as the specimen thickness decreases to the

thinnest tested specimen (0.3 mm), the strain field and the stress field become much less

uniform, and localization is more pronounced. Figure 58(c) shows that most of the surface

of the model is relatively unstrained, while the max localized strain is more than double

the value of the surrounding part. The right-hand side of Figure 58(c) depicts a similar

result for the stress fields; the stress fields throughout the bulk of the cross-section is

appreciably lower than the very localized maximum stress value that is found on the surface

near the corner.

From the FEA results, it is apparent that the rough surface morphology can strongly

affect the stress and strain fields within the specimen. These localized stresses and strains,

that increase as the part thickness decreases, are expected to influence the value of the

specimen’s elongation and UTS value negatively and will therefore increase the thickness

109
dependence for these two metrics. However, it is not expected that these created stress

heterogeneities will influence the yield and modulus values.

110
 (a)

(b)

111
 (c)

Figure 58: (a) The FEA analysis for the (a) 2.0, (b) 0.75 and the (c) 0.3 mm thick specimen.
Stress heterogeneities arise as the specimen thickness decreases due to surface morphology.

5.4. Analysis: Effects of Thickness and HIP

The primary metrics of interest for this study are the engineering yield strength, the

ultimate tensile strength (UTS), the Young’s modulus, and the total specimen elongation

to failure, all plotted in Figure 59(a-d) with connecting letters indicating statistical

equivalency of the respective means.

Figure 59a shows a remarkable result for yield strength, clearly demonstrating that

while the non-HIP specimens have a strong thickness dependency, this is eliminated for

the specimens that received a HIP cycle. UTS on the other hand shows strong thickness

dependency for both heat treat conditions (Figure 59b). Young’s modulus values for both

HIP and no-HIP show no strong thickness dependency (Figure 59c). Lastly,

112
 Figure 59d shows strong thickness dependance for total elongation to failure for both

HIP and non-HIP specimens. In addition to examining the means, the Coefficient of

Variation (CV) was also compared, as shown in Figure 60. While yield strength, UTS and

Young’s modulus do not show significant differences in CV with thickness, the CV of

elongation shows a significant increase with reducing thickness. HIP does not significantly

alter these trends in CV. In sections 5.5.1-5.5.4, correlation studies of the mechanical

properties and the specimen density are examined only for the non-HIP specimens since

variation of density of HIP specimens is reduced so significantly no strong correlations can

be established between density and corresponding mechanical metrics. The correlations

need to be separated by the specimen nominal thickness, since the density values are also

a function of specimen thickness (Figure 49a).

(a) (b)

113
 (c) (d)
Figure 59: The mechanical behavior of the AM specimens for the four metrics examined in this
study, shown with connecting letters indicating statistically equivalency between means: (a)
engineering yield strength, (b) engineering ultimate tensile strength, (c) Young’s modulus, and
(d) elongation

Figure 60: The coefficient of variation as a function of thickness and HIP condition for each of
the four metrics examined

5.4.1. Modulus

The calculated Young’s Modulus for the non-HIP and HIP specimens shows no strong

thickness effects as the connecting letters report shows significant crossover of mean

values across the different thicknesses. Figure 59(c) shows that the non-HIP specimens

have less predictable modulus value, while the added HIP step reduces part variability with
114
modulus. The HIP specimens have an appreciably greater modulus value, and this is

believed to be from the recrystallization and annealing of the grains due to the high HIP

temperature. Typically, as-built LPBF specimens have a large degree of (001) texture in

the build direction for the FCC Inconel 718 [97], [98]. This texture can be reduced via the

HIP step which will increase the modulus value since isotropic Inconel 718 has a modulus

value that’s approximately twice over that of perfectly textured specimen that is being

pulled in the (001) direction [99]. Examining the correlation between modulus and density

for the non-HIP specimen yields interesting results as Figure 61 shows. For the thinnest

specimens, there is no correlation for modulus and density, but surprisingly strong

correlations can be found for the thicker specimens (≥ 1.0 mm). The cause of these strong

inverse relationships with modulus and density for the thickest specimens is currently

unknown.

Figure 62 compares the calculated modulus values for the AM and sheet metal

parts. No thickness dependence is found for the sheet metal specimens for the two

different heat treatments. In general, it is fundamentally expected that modulus would not

have a thickness dependency since elastic interactions should show no size effects.

115
 Figure 61: The correlation of Young's Modulus (GPa) vs Density (g/cc) for the non-HIP
specimens only. No strong correlations are found for the thinnest specimens, but for specimens of
≥ 1.0 mm we see a strong negative correlation with modulus and density.

Figure 62: The calculated Modulus values as a function of thickness for the AM and sheet metal
parts.

116
 5.4.2. Yield

A strong thickness dependency for yield can be found in Figure 59(a) for the non-HIP

specimens. The cause of this thickness dependency is not entirely clear. The HIP set of

specimens shows no thickness dependency, and this is believed to be caused by the much

more uniform microstructure that is created after the HIP cycle with a much more isotropic

grain orientation and precipitate distribution. The non-HIP set of data has an appreciably

greater yield value, which is potentially attributed to the remnants of the dendrites that are

observed as a fine cellular structure (Figure 54c) whereas the HIP cycle removes all

artifacts of the dendrites formed during the printing process. Through heat treat temperature

studies, other authors have attributed this sub-micron cellular structure as a potential

strengthening mechanism to impede dislocation motion[24]. However, the HIP step

reaches temperatures that are high enough to allow for the dissolution of the cellular

structure which can reduce the yield strength. No strong correlations exist between the

yield of the non-HIP specimens and the densities as shown in (Figure 63), illustrating that

the amount of porosity in a specimen does not affect the yield strength in a detrimental

way. Within this study, the only benefit received from a HIP cycle on the engineering yield

is a more predictable and consistent value across all the thicknesses, however, this comes

at the expense of lowering the yield strength.

To help better understand the thickness dependence of the non-HIP specimens a

cursory literature review was conducted on specimens that are below the thickness of this

study in the microscale. For many alloys it was found that yield strength has little

dependence on size in the microscale and in some cases, authors show a strong increase in

117
the yield strength with decreasing specimen thickness. Zhang et al. show an almost

perfectly monotonic increase in yield strength with reducing copper thin films that is

primarily attributed to a slightly decreasing mean grain diameter size with reducing

thickness[100]. While in another study on Ni, that is closer to the mesoscale size range that

was studied here (specimen size is ~200x200 μm), showed a highly dependent yield

strength that was based on the processing of the material which influenced the grain

sizes[101]. It appears that specimen geometry is not an influencing factor in the reduction

of yield stress. In fracture mechanics, fracture toughness can be dependent on specimen

size with a reduction in size leading to an increase in fracture toughness as the specimen

approaches a state of plane stress. However, in the context of yield strength of ductile

materials, it appears that boundaries to dislocation movement are far more pervasive in

influencing the yield strength in the mesoscale and microscale.

118
 Figure 63: The correlation of Engineering Yield (MPa) vs Density (g/cc) for the non-HIP
specimens only. No strong correlation or trend can be across the thicknesses, implying that the
amount of porosity generated in the parts of this study (>98% dense) has no discernable impact
on the final yield value.

5.4.3. Ultimate Tensile Strength

The ultimate tensile strength shows significant thickness dependance for both the non-

HIP and HIP specimens for a nominal thickness below 0.75 mm where both sets show

almost a monotonic rise in values for increasing thickness. From 0.75 mm and thicker, the

UTS values are approximately the same, and show a significant amount of overlap in the

connecting letters report (Figure 59b). This strong thickness dependance is believed to

come from the stress heterogeneities that are seen in the as-printed specimens, where the

localized stresses become much more egregious as the thickness is reduced (Figure 58),

and will lead to a lower apparent tensile strength. The HIP specimens have an increased

UTS value over the non-HIP set of data. Figure 64 shows that the thinnest specimens (≤

0.75 mm) have a decent correlation with specimen density and the amount of porosity in

the specimen. This implies that by adding a HIP step in the heat treatment, an increase of
119
UTS can be achieved for thin wall specimens by removing the porosity which will act as a

stress riser and lead to an early onset failure. The HIP step has an added benefit of reducing

the number of precipitates that form along the grain boundary that can act as a void

nucleation site and reduce the measured UTS. Figure 57g shows that below the fracture

surface, there are several voids forming along the grain boundaries while Figure 57h still

shows a strong grain boundary structure.

Figure 64: The correlation between UTS (MPa) and density (g/cc) is plotted for the non-HIP
specimens. There is a strong, positive signal between the two variables for the thinner specimens
(<0.75 mm), demonstrating that the thickness effects for the UTS are partly attributed to the
higher density parts for thicker specimens.

5.4.4. Elongation and Necking

There is a strong thickness dependance with elongation and specimen thickness for

both non-HIP and HIP specimens (Figure 59d). The decrease in sample total elongation to

failure with decreasing sample thickness is primarily attributed to the sample’s changing

aspect ratio as the thickness is reduced – the geometric constraints dictate that a thinner

120
specimen will not facilitate as much ductility and necking until final failure. Figure 65

shows the results of a series of tensile tests that used Digital Image Correlation (DIC) on a

set of HIP specimens to visualize the evolving strain fields as a function of specimen

thickness just before and after specimen failure. The thinner specimens have a much more

horizontal fracture with minimal amount of specimen necking, while the thicker specimen

portray much greater ductility with a greater amount of necking. Secondly, due to the stress

heterogeneities that cause a lower UTS, it is also expected that they would reduce sample

ductility and total elongation. The HIP heat treat cycle significantly increases sample

elongation for the thicker specimens, but both the HIP and non-HIP specimens converge

on similar values as thickness reduces and as the previously stated causes start to dominate.

The fracture surfaces correspond well with the mechanical behavior as the HIP specimens

have a much greater number of deep voids across the surface with minimal zones of flat

cleavage. In this work, necking is represented as the percent of strain between the UTS and

the total amount of elongation to failure. It’s believed that the increase of elongation and

necking (Figure 66) in the HIP set of specimens is dictated by the strong dissolution of

precipitates along the sample grain boundaries and within the matrix that nucleate voids.

The benefits of an increased elongation for the HIP specimens can also attributed to the

closure of the porosity generated via the additive manufacturing process. Figure 67 shows

a positive trend of increasing elongation for increasing printed part density for all

thicknesses for the non-HIP specimens with the strongest correlations for specimens greater

than 0.5 mm in nominal thickness. Lastly, Figure 68 shows the elongation for both the AM

and sheet metal specimens. Both the AM and sheet metal parts show a strong amount of

121
thickness dependence which furthers the argument that the amount of elongation that a

specimen can withstand is tied to the geometry of the specimen.

Figure 65: The DIC results showing the evolution of necking behavior in the specimens across
different thicknesses. As the specimen thickness is increased, the specimen can achieve greater
necking and ductility partially due to specimen geometry.

122
 Figure 66: The percent elongation vs. the percent necking for the non-HIP and HIP specimens
show a strong positive trend. Samples that are capable of a greater overall elongation also
demonstrate a greater amount of necking. The HIP specimens generally have a greater amount of
total elongation and necking.

Figure 67: The correlation between specimen elongation and density is plotted for the non-HIP
specimens. There is a strong, positive signal between the two variables for the thicker specimens
(>0.35 mm), demonstrating that the thickness effects for the elongation are, in part, due to the
greater density achieved by the thicker specimens.

123
 Figure 68: The total elongation until failure vs. specimen thickness for the AM and sheet metal
specimens.

5.5. Discussion
5.5.1 Modeling

The prior section clearly demonstrated thickness dependence for most parameters of

interest when designing components with as-built thin walls, particularly under 1 mm

thickness. The assumption of bulk properties in these instances will result in under-

designing these structures and making them prone to early failure. From a designer’s

perspective, it is important to define a material model to capture the change in material

properties with thickness, as has been proposed before for laser sintered Nylon 12, where

a Weibull cumulative distribution function was shown to model the relationship between

modulus and thickness [102]. For a given function F(t), where t is the wall thickness, the

Weibull function takes the exponential form:

𝑐
𝐹(𝑡) = 𝑎(1 − 𝑒 −𝑏𝑡 ) (1)

124
where the parameter a represents the bulk property and has the same units as that of the

material property in question, and its value is given by the asymptote. The shape parameters

b and c are determined by fitting Equation 1 to the experimentally obtained property data.

In this work, four metrics of relevance to the development of a bilinear elastic-plastic

material model were considered: the elastic modulus, the yield strength, the UTS and the

plastic strain at failure. These metrics enable a designer to establish a bilinear elastic-plastic

material model for implementation into analysis tools. The MATLAB curve fitting toolbox

was used to derive the parameters for the Inconel 718 using mean values at each tested

thickness reported before. The goodness of fit against the entire experimental dataset is

shown in Figure 69. The solid continuous line shows an equation with the obtained

parameters listed in Table 5. These parameters listed apply to HIP and No HIP Inconel 718

for thickness greater than 0.3 mm. A key insight from these models is how HIP and No

HIP predictions converge at low thicknesses, suggesting a shift from microstructural

differences (attributed to the HIP process) to geometric differences that are independent of

HIP condition. More work is in progress to validate this hypothesis.

(a)

125
 (b)

(c)

Figure 69: Model fits relative to experimental data for (a) Yield Strength, (b) Ultimate Tensile
Strength, and (c) Plastic Strain at Failure

126
Table 5: Parameters for a thickness-dependent bilinear elastic-plastic material model of LPBF IN-
718 for HIP and no-HIP conditions

HIP No-HIP
Young's Modulus 162.171 GPa 143.192 GPa
a=1121 MPa, b=6.725,
Yield Strength c=0.4048 a=1211 MPa, b=3.233, c=0.1299
a=1611 MPa, b=31.39,
Ultimate Strength c=2.183 a=1540 MPa, b=11.48, c=1.262
Plastic Strain at
Failure a=0.1736, b=690.1, c=2.183 a=0.139, b=62.95, c=3.434

5.6. Conclusions
This work sought to establish thickness dependence of quasistatic mechanical behavior
in thin wall Inconel 718 specimens manufactured with the Laser Powder Bed Fusion
process, as well as examine how this dependence changes with Hot Isostatic Pressing
(HIP). The following conclusions can be drawn from this study:
1. Yield strength for specimens without HIP show clear debits under 0.75mm wall
thickness, but the use of HIP removes this dependence.
o The reductions in yield strength (for no-HIP specimens) with thickness is
attributable to differences in grain size distributions as the specimens get
thinner.
o HIP has the effect of homogenizing grain size across the specimen and
eliminating the above dependence.
2. Ultimate tensile strength and elongation to failure both show reductions under
0.75mm wall thickness, independent of HIP condition.
o Stress heterogeneities due to surface topology are a leading factor for
thinner specimens.
o Density of non-HIP specimens show strong correlation with UTS (for
thinner specimens) and elongation (for thicker specimens).
o DIC measurements show that specimen geometry influences ductility in
thin parts.
3. Young’s modulus does not show any systematic reductions with thickness.

127
 o Sample internal/external defects and phase fraction changes don’t influence
modulus across thicknesses in the mesoscale size frame.
4. An added HIP step significantly improves part density while increasing the UTS
and total elongation.
o The higher HIP temperatures allow recrystallization and homogenization of
the microstructure while eliminating brittle precipitates that form along the
grain boundary which act as sites for void nucleation.
5. A mathematical model based on the Weibull cumulative distribution function was
shown to model thickness dependence well and demonstrates convergence of HIP
and NO-HIP data for most properties of interest at low thicknesses.

128
 CHAPTER 6

THE EVOLVING BEHAVIOR OF FATIGUE WITH CHANGING

MICROSTRUCTURE AND SECTION AREA OF THIN-WALL ADDITIVELY

MANUFACTURED METAL

6.1. Introduction
A comprehensive look at the Inconel 718 alloy has been examined in previous sections

while characterizing the size effects that have been seen for the room temperature quasi-

static specimens. AM parts will typically perform as well or better as their wrought

counterparts in tensile testing scenarios; however, it has been shown that the fatigue

performance of AM parts will typically show a strong decline that is exacerbated as testing

approaches the endurance limit [21], [103]. Due to the constant melting and remelting

process of the relatively violent AM process, internal stresses that are not sufficiently

purged from the part can influence the fatigue properties [104]. The internal porosity

(especially the sub-surface porosity), microstructural twins, and a rough surface have all

been shown to be contributing components to the decreased fatigue life in AM parts [103],

[105].

Figure 70 illustrates that there are many authors that are examining Inconel 718 in the

role of fatigue, but no authors were found to study fatigue performance based on a changing

wall thickness coupled with an additional HIP cycle to gain a more complete understanding

on the size dependence for the AM parts. Additionally, due to the number of variables that

go into designing a fatigue experiment (specimen geometry, R-ratio, LCF/HCF, different

129
heat treatments, adding a notch, etc.) it becomes increasingly more difficult to compare

and extrapolate information from other authors.

Figure 70: An overview of the literature found for AM Inconel 718 parts [106]–[112].

This work studies the effects of changing specimen thickness (for seven different

thickness values ranging from 0.3-2 mm) for two different heat treatment conditions

(with and without a HIP step) for LPBF Inconel 718 The HCF testing range for ideal

cycles to failure fore this work begins at 105 and extends to the specimen runout at 107.

An examination of the role of the internal (porosity and grain sizes) and external (surface

roughness) is revisited and applied to the results of the HCF test. Finally, this study also

considers how HIP influences these differences, asking if HIP can serve as a way to

ameliorate mechanical debits in thin wall structures at any point along the S-N curve.

130
6.2. Methods
6.2.1. Specimen Design and Manufacturing

Details on the LPBF process parameters, build setup, specimen design, and heat

treatment can be found within Chapter 2 and are not repeated here.

6.2.2. Specimen Characterization

The relative density, surface roughness, and the part morphology for each fatigue

specimen was measured, including untested specimens. The details on the measurement of

density using the Archimedes method can be found in Chapter 3. Since all the additively

manufactured parts maintained an as-printed surface finish (no machining) surface

roughness measurements were made on every specimen. Surface Roughness measurements

were taken with a Keyence VR3200 optical profilometer within the grip section of the

specimen since the gauge section width (6 mm) was not wide enough for the desired scan

area without edge effects distorting the measurements. A gaussian and short pass filter were

first used on the data to remove any specimen warpage from the final surface roughness

values. Sa, Sq, Sv, and Sz were collected for this work, but only Sa and Sq will be shown in

the results section.

Polishing followed standard polishing procedures with guidance of rpm, time and

polishing force taken from Voort et al. [113], starting with silicon carbide grinding papers

up to 600 grit, then transitioning up to 9, 6, and 1 μm diamond, and finishing with 0.05 μm

alumina. Select samples were etched using a mixture of 2 mL methanol, 6 mL HCl, and

0.3 g of CuCl2. SEM images were collected on a Zeiss Auriga scanning electron

microscope.

131
 Due to the rough surface finish of as-printed AM parts, care must be taken to

properly measure specimens’ cross section with thin walls since a 10 μm Sa surface on each

side of the part can lead to errors exceeding 10%. An investigation on measuring thin wall

structures for mechanical testing was previously completed and shown in Chapter 4. It was

found that using a point micrometer to measure the thickness and calipers to measure the

width provides sufficient accuracy to measure the load-bearing cross sectional area for the

fatigue specimens.

6.2.3. Fatigue Testing

Only tension-tension fatigue testing was feasible due to the aspect ratio of these

thin dogbone specimens which made them prone to buckling under any compressive load.

An R-ratio of 0.1 was used for all fatigue testing with a frequency of 40 Hz while the

machine was run in force control. For testing purposes, five target fatigue lives were

selected to span the HCF regime. These five target lives were 100k, 250k, 500k, 1M, and

10M cycles (the last of these was considered runout – i.e., the test was stopped upon

reaching 10m cycles). Additional testing was done on the thickest and thinnest specimens

(2.0 and 0.3 mm) to help define stronger S-N curves. A servo-hydraulic Instron 8801 with

a 50 kN fatigue testing machine was used for the HCF testing

6.2.4. Experimental Plan

In total, 280 specimens were printed across 2 separate AM builds with 7 different

thicknesses (nominal CAD dimensions of 0.3, 0.35, 0.5, 0.75, 1.0, 1.5, 2.0 mm). All

specimens received a full standard heat treatment with half of those specimens receiving

an additional HIP step before the final solutionize and double-aging cycle. Select

132
specimens were selected for fractographic inspection post testing to help ascertain the

location of the crack initiation. The set of specimens that received a full heat treatment plus

a HIP cycle (SR+HIP+SA) will be referred to herein as “HIP”, while the other set of

specimens that only received a stress relief, solutionize, and double age cycle will herein

be referred to as “Non-HIP” (SR+SA). Not all 280 specimens were fatigue tested, but all

specimens were characterized for their surface roughness, density, and blue light surface

morphology information.

6.3. Results
6.3.1. Density

The results of the Archimedes density measurements show that the non-HIP specimens

have a consistently lower absolute density value when compared to the HIP set, Figure 71.

Both the HIP and the non-HIP sets of data have an increasing value of density as the part

thickness increases. The increasing density is primarily attributed to the greater amount of

sub-surface porosity generated at the location of the internal scan raster turns (~200 μm

from the specimen surface) where the laser momentarily slows down and the melt pool

enters the keyhole zone as the applied energy density increases during the turn and

generates more pores. Since all the specimens (regardless of thickness) have an equal

number of laser raster turns throughout the build process, it is to be expected that the thinner

specimens will have a greater number of pores when compared to the overall mass of the

specimen – as the specimen thickness increases there will be a greater amount of internally

dense material that is printed well away from the specimen edges. In addition to increasing

the part absolute density, the HIP cycle also decreases the standard deviation providing

133
parts with much greater consistency of internal voids (Figure 71). Figure 71 has already

been shown in Chapter 3 but is reproduced here for readability.

Figure 71: The absolute density and standard deviation values of the AM parts for the 2 different
heat treatments across all of the different thicknesses tested. The HIP heat treatment significantly
reduces the standard deviation of the test specimens for the thicker parts.

6.3.2. Surface Roughness

Rough surfaces are not ideal for fatigue testing and can lower the effective fatigue life

of the specimen. ASTM E466 recommends a maximum surface roughness of 0.2 μm Ra

(in the longitudinal direction) for machined specimens, but for AM structures, the

machining and polishing of all surfaces is not always practical or possible. The Sa values
134
for the HIP and non-HIP specimens both show no thickness dependency or trends with the

mean values and standard deviations being approximately equal across all the thickness

per heat treat condition (Figure 72 (a)) with all mean values being lower than 3 μm.

However, there was a small increase in Sa for the HIP specimens compared to the non-HIP

specimens, on a per thickness basis. The slight increase of surface roughness for the HIP

specimens is potentially the result of the shrinkage of internal porosity after the HIP cycle

which could slightly change the surface topology of the specimen. The Sv values, Figure

72(b), show similar results and trends as Sa. Despite surface roughness showing no

thickness dependence this does not imply that surface roughness effects will be

proportional across all thicknesses. Surface stress heterogeneities arising at the surface of

a thin specimen will make up a greater portion of the specimen load-bearing area when

compared to a thicker specimen.

135
 (a) (b)

Figure 72: The Sa and Sv values for the HIP and non-HIP specimens respectively (a & b).

6.3.3. High-cycle Fatigue

The results of the HCF test can be seen below in Figure 73 where the results are

qualitatively shown to demonstrate the gross differences between the HIP and non-HIP sets

of data. The size of the markers and lines in Figure 73 scale with the specimen nominal

thickness. In general, the HIP data set has a slightly better HCF performance before 10 6

cycles, while the non-HIP sets of data generally dominate with their respective runout

stresses for the thinner specimens.

136
 Figure 73: The non-HIP and HIP HCF fatigue results. Increasing marker size and line thickness
corresponds to greater nominal thickness of the specimen. The lines are the Basquin slopes of the
fatigue data set.

Figure 74 shows the HCF results for just the non-HIP set of data where the lines

are the Basquin slopes. A thickness dependency can be found, but there is significant cross-

over amongst the thicker specimens (0.75 mm and thicker). The slopes of fit are not

consistent for all sets of data. Figure 75 shows the HCF results for the HIP set of data.

Again, a strong thickness dependency is shown with the thicker specimens having a greater

HCF performance, however for the HIP set of data the thickness dependence is more

apparent. All the Basquin slopes for the HIP set of data are very uniform and consistent

with no strong outliers. The spread of the runout data is greater for the HIP specimens

compared to the non-HIP, but both sets show strong thickness dependence.
137
Figure 74: The non-HIP set of data for all thicknesses separated by color.

138
 Figure 75: The HIP set of data for all thicknesses separated by color.

Inconel 718 sheet metal specimens were used in this study for several reasons: to

ensure that the specimen design and geometry was valid, to ensure the fatigue machine was

functioning as intended, to have a baseline wrought comparison for the 3d printed parts,

and to examine the thickness effects against a specimen with a machined surface and no

internal pores. Figure 76 plots the results from the thickest and thinnest AM and sheet metal

specimen for both sets of heat treat conditions. It was found that the 2 mm HIP AM

specimen had a comparable HCF life to the 2 mm sheet metal specimen that did not see a

HIP heat treatment. The decrease in the fatigue performance between the non-HIP and HIP

sheet metal specimens is likely due to the HIP heat treating process removing the beneficial

139
rolling texture in the sheet metal. Minor caution should be observed with the sheet metal

specimens; despite the faces of the specimens being very smooth as they retained their

purchased surface finish, the edges that were water-jet cut were not polished down. The

edges of the sheet metal are rough and minor improvements to their fatigue life should be

expected if the edges were polished down. Lastly, the AM specimens showed a

significantly greater amount of thickness dependence. Even though the thickest AM

specimens are similar to the thickest sheet metal specimens, the AM 0.3 mm specimens

have a large spread in performance from the 2 mm AM parts. Lastly, the runout for the 2

mm AM specimens is significantly greater than the sheet metal specimens, despite the

fatigue life being comparable for 106 cycles and less. This is likely due to the AM

specimens having a greater UTS over the sheet metal specimens since there is strong

correlation between the quasi-static UTS and the endurance limit[114].

140
Figure 76: The log-log plot of max stress vs the total number of cycles to failure for the 2 extreme
thicknesses for the AM and the sheet metal parts.

6.3.4. Microstructure

Since the HCF specimens are manufactured identically to the room temperature

quasi-static specimens, the microstructural analysis shown in Section 5.3.3 Microstructure

is still applicable. In general, the non-HIP specimens show a strong duplex grain

arrangement with a crossing pattern in the interior of the specimen (~200 microns away

from the surface of the specimen). This duplex grain arrangement reduces in magnitude as

the specimen thickness is reduced as the primary driver of the crossing pattern is the raster

scan pattern of the laser during the manufacturing process. The HIP specimens have a

strong reduction in the porosity generated and show a much more homogenous grain

pattern. Analysis of the γ’ and γ” strengthening precipitates was not conducted in this study
141
since this will typically involve the use of a TEM, but it was found that most of the

precipitates formed in the non-HIP specimens were generated along the grain boundaries.

(a) (b)

Figure 77: A closeup of the grain structure for a non-HIP (a) and HIP (b) specimen. The non-HIP
specimen shows a very tight grain structure at the surface of the specimen, while the HIP
specimen shows an increase of annealing twins with larger and more uniform sized grains.

6.3.5. Fractography and IR imaging

The fractography section will be broken into 2 parts. The first will be examining

the fractography of the LPBF parts, and the second will examine a sheet metal specimen

that was tested while an IR camera was recording the surface temperature of the specimen.

Figure 78 shows an overview of a typical Inconel 718 HCF fracture surface and

will be used as an initial aid to help describe subsequent images. The crack initiated near

the bottom-left corner of the specimen, characterized by cleavage failure which can take

on the appearance of faceted surface, feather like features, “tongues”, or river patterns to

name a few. The crack propagates quickly from left to right as indicated by the striations

just outside of the initiation site. Each line of the striation signifies another fatigue cycle as
142
the crack is now quickly propagating across the sample surface to the final failure region.

The curvature of the striations can be useful as the concentric circles will typically point

back to the initiation of the crack. Finally, the crack has progressed through the specimen

far enough that the part will fail by ductile overload in the final cycle – characterized by a

dimpled surface. A strong taper can also be observed in the specimen halfway through. The

taper does not occur on the portion of the specimen where the crack initiated since there is

no significant global plastic flow at this point; only as the crack progresses to the ductile

overload zone does the taper occur signifying appreciable plastic deformation associated

with the overload zone.

Figure 78: An overview of the main sections of a fatigue specimens fractographic surface with a
sheet metal specimen used as an example.

Figure 79-Figure 87 show the fractographic images for the same AM, Non-HIP

specimen that had failed at ~171k cycles that has a specimen ID of N15. All of the

macroscopic images show a scattered amount of porosity across the specimen surface. This

143
amount of porosity is not entirely representative of a random 2D cross-section that could

be taken of the sample since this is a 3D surface with greater surface area, and since this

surface may have been preferentially created due to an increased amount of porosity that

leads to an earlier fracture. From Figure 79 it is relatively apparent that the crack initiation

site is located somewhere on the left-hand side of the image since a taper can be seen

forming on the right side. Figure 80 shows the rough locations of the different fracture

regimes that were noted during fractography with Figure 81-Figure 84 belonging to images

taken at or near the initiation site, Figure 85 and Figure 86 correspond to images taken in

the striation zone, and Figure 87 shows an example of the ductile overload zone. It appears

that the crack initiated at the surface and typically near large sub-surface porosity (Figure

82) or near sharp surface defects (Figure 83 and Figure 84). Both of these morphological

defects can create stress risers which can lead to earlier part failure with fatigue life.

Figure 79: A macroscopic SEM image of a 2 mm AM No-HIP specimen that failed at ~171k
cycles. An appreciable amount of porosity can be seen across the sample surface (specimen N15).

144
Figure 80: A closeup of Figure 79 showing the locations of the different areas that were found
through SEM imaging (specimen N15).

Figure 81: A closeup of the fracture initiation site (specimen N15).

145
 Figure 82: One of the potential locations of a fatigue crack initiation site. Notably, there is a
relatively large pore just under the surface, and surface defects at the location of the cleavage
patterns. Location of image can be seen by the red square in Figure 81 (specimen N15).

Figure 83: One of the potential locations of a fatigue crack initiation site. No immediate sub-
surface porosity can be seen in the image, but the cleavage patterns look to be near a partially
melted powder sphere located on the surface. Location of image can be seen by the blue square in
Figure 81 (specimen N15).
146
Figure 84: One of the potential locations of a fatigue crack initiation site. No immediate sub-
surface porosity can be seen in the image, but the cleavage patterns look to be near a partially
melted powder sphere located on the surface with sharp surface topology indicated by the red
arrow (specimen N15).

Figure 85: Near the middle of the specimen where striations form, however, they are a little
sparse in this section (specimen N15).

147
Figure 86: The location of this image can be seen by the inset red square from Figure 85. The
striations point back towards the predicted fracture locations depicted above (specimen N15).

Figure 87: An example of the fracture surface in the overload zone, marked with a variety of
different sized dimples across the sample surface (specimen N15).

148
 Figure 88-Figure 99 show the fractographic surface for an AM, HIP specimen that

failed after ~131k cycles. Figure 88 shows a stitched macroscopic of the specimen surface

after failure, with relatively no porosity to be seen. Through the fractographic process, a

couple of locations were observed to be the most likely for the origin of the crack initiation

(Figure 89) with both locations taking place on the surface and at or near the corner of the

specimen. Without the sub-surface porosity acting as an additional stress concentrator,

more of the cracks were observed to initiate near the corner of the HIP specimens due to

geometrically created stress concentrations. Figure 94 and Figure 96 show a much more

defined striation field when compared to the non-HIP specimens, both figures have

striations that appear to have propagated in a direction that originates from the top-left

corner of the specimen. An example of the overload region of the specimen can be seen in

Figure 99 which shows a much more uniformly dense network of dimples across the

sample surface when compared to the non-HIP specimen (Figure 87). The observations on

the size and distribution of the dimples for the non-HIP and HIP parts were previously seen

and explained in the room temperature, quasi-static specimens. The surface is relatively

free of porosity, but minor pores can still be found like in Figure 98. The pore found in

Figure 98 appears to be caused by gas trapped porosity, which can typically be very

difficult to eliminate through HIP since the internal gasses can equilibrate to the external

isobaric pressure. The walls of the pore are much less smooth when compared to the pores

found in the non-HIP specimens (reference Figure 82). Gas trapped pores may also expand

when further heat treatment cycles are applied after the HIP step. Figure 100 shows a 1.5

149
mm thick AM, HIP specimen that failed at ~100k cycles with strong cleavage markings at

the corner of the specimen.

Figure 88: A macroscopic SEM image of a 2 mm AM HIP specimen that failed at ~131k cycles.
An appreciable amount of porosity can be seen across the sample surface (specimen N5).

Figure 89: A closeup of the approximate failure location where the inset colored squares
correspond to Figure 90, Figure 91, and Figure 92 (specimen N5).

150
 Figure 90: One of the potential crack initiation sites located at the corner of the specimen
(specimen N5).

Figure 91: One of the potential crack initiation sites located just below the corner of the specimen
(specimen N5).

151
Figure 92: A closeup of Figure 91 showing river like patterns and cleavage fracture. (specimen
N5).

Figure 93: A macroscopic view of the striation region just inside of the taper (red arrow) where
the overload region will occur further to the right. The red inset square marks the location of
Figure 94 (specimen N5).

152
Figure 94: Defined striations mark that the crack propagated from left to right. (specimen N5).

Figure 95: A macroscopic view of the striation region just inside of the taper where the overload
region will occur further to the right. The yellow inset square marks the location of Figure 96
(specimen N5).

153
Figure 96: Defined striations indicate that the crack propagated from the top-left to bottom-right
of the image (specimen N5).

Figure 97: The overload region of the specimen on the far-right side. The red, inset square marks
the location of Figure 98 (specimen N5).

154
Figure 98: Minimal porosity can be seen from the specimen after the HIP cycle, the non-uniform
edges indicate that the pore surface has been deformed through the high isobaric pressure applied
during the HIP cycle, as compared to the smooth pore walls found in Figure 82 (specimen N5).

Figure 99: The overload zone of the specimen, marked by relatively even dimples across the
surface (specimen N5).

155
Figure 100: A 1.5 mm thick AM HIP specimen failing after ~100k cycles. Large faceting can be
seen near one of the corners of the specimen indicative of a crack initiation site (specimen M3).

In a complementary study to examine the thermal evolution of a HCF specimen to

build an entropy-based fatigue model, infrared photography was used throughout the

entirety of the fatigue test. Figure 101 shows the final 3.7 seconds of the fatigue test, as the

HIP, sheet metal specimen failed at ~85k cycles. These final 3.7 seconds corresponds to

the crack propagating through the striation regime (~150 cycles at 40 Hz) and the following

overload zone of the specimen. The striations for the sheet metal specimen are much more

defined than what was seen for the AM specimens (Figure 102). A WITec AFM (Atomic

Force Microscope) was used to measure the length of the striations to corroborate with the

time scale found from the IR imaging. Figure 103 shows that the striation spacing was

almost exactly 1 μm apart, which implies that the crack propagates at an approximate rate

156
of 40 μm/s and needed ~150 cycles to reach the final overload portion of the fatigue

specimen.

Figure 101: The final 3.7 seconds of infrared camera imaging of a fatigue test.

Figure 102: A closeup image of the striation region of a HIP’ed sheet metal specimen that was
used during the IR imaging fatigue test.

157
 Figure 103: An AFM was used to measure the topology of the striations of the sheet metal
specimen. It was found that the striations were spaced almost exactly 1 μm apart.

6.4. Discussion
Both the non-HIP and HIP specimens showed a significant amount of thickness

dependence with the thickest specimens (0.75 mm thick and thicker) being more tightly

grouped near the top of the fatigue curves while appreciable decreases in fatigue

performance occurred monotonically for the 0.5, 0.35, and 0.3 mm specimens. The slopes

of the linear fits of the log-log plots of the S-N curves can be found below in Figure 104

where the values of the Basquin slope are plotted as a function of specimen thickness,

separated by heat treatment. The HIP specimens all had a steeper Basquin slope, and the

values were much closer together than the non-HIP set of data. No significant thickness

evolution can be observed in Figure 104. The tighter spread of Basquin slope values for

the HIP specimens is believed to be from the much more homogeneous microstructure of

the HIP specimen, as well as the removal of the semi-stochastic porosity (especially the

sub-surface porosity) that has been found to act as a stress concentrator for crack

nucleation. The more negative Basquin slope values for HIP is partially caused by the

greater UTS value found for the HIP specimens (Figure 59(b)) and due to the decreased

runout values, that were achieved by the HIP specimens since the Basquin slope is
158
approximately a straight line running from the UTS value (100 cycles) to the endurance

limit (107 cycles). This is a poor approximation for the Basquin slope but gives partial

insight as to the differences between the HIP and non-HIP sets of data.

Figure 104: The values of the Basquin slope for the fatigue data for a respective thickness.

The fatigue limit is plotted against the mean UTS value in Figure 105. Both the HIP

and non-HIP sets of data show a strong linear trend between the runout stress and the UTS

of the specimen with the HIP specimen have a slope more than twice as great. Linearity

between the two variables is commonly seen [114], as the UTS typically influences the

runout stress of a specimen. The reason for the HIP set having a greater slope is primarily

driven by the greater thickness dependence seen in the runout values for the HIP data

compared to the non-HIP data.

159
 Figure 105: The fatigue limit (the max stress at 107 cycles or more) is plotted vs. the room
temperature quasi-static mean UTS value for each thickness.

HIP specimens therefore have a better HCF performance at the higher stress,

while the non-HIP specimens have a greater endurance limit. This phenomenon is

believed to be caused by the large number of sub-surface pores that are generated during

the LPBF process and are retained in the non-HIP specimens in combination with the HIP

specimens achieving a greater UTS value in the quasi-static tests. In a LCF (low cycle

fatigue) study of DMLS (direct metal laser sintered) Inconel 718, the authors found that

for high strain amplitudes the internal porosity of the specimen led to early part failure

while the HIP specimen with reduced porosity had a greater LCF life [115]. While Yu et

al. found that the reduction of porosity had no influence of fatigue life for VHCF testing

(cycles greater than 106)[116]. A correlation study was performed on the non-HIP

specimens examining how stress values relate to the part’s density to better corroborate

160
the theory that non-HIP specimens have a lower fatigue life at higher stress values. Since

no tests were ran at identical force/stress values for a given thickness, plotted stress

values show the difference from the Basquin slope (a linear fit of data points between 105

and 106 cycles). Although it is not ideal to plot the stress difference from the Basquin

slope, it should be a decent starting point in understanding how density might correlate

with fatigue life. Figure 106 shows the results of the correlation as a function of specimen

thickness; note, there is no data shown for the 1 mm specimen since only 2 data points

were between 105 and 106 cycles. Figure 106 shows no strong evidence that the relative

density of the part strongly influences the HCF life before reaching the endurance limit.

The correlations are weakest for the 0.3 and 2 mm specimens that had the greatest

number of tests. Correlations increase for the rest of the specimens primarily due to a

reduction of data points. Overall, there are no strong trends to the dataset. Although

overall specimen density is not a great predictor for fatigue performance, it is not

believed that this is conclusive evidence that porosity is not a key driver for influencing

fatigue life at higher stresses. Due to the stochastic nature of fatigue testing AM

specimens, only 1 large pore near the surface can be enough to reduce the fatigue life by

creating a stress localization at the surface, while a large amount of internal porosity may

not significantly contribute to decreasing the fatigue life.

161
 Figure 106: The stress difference from the Basquin slope vs. relative density for the non-HIP
specimens.

Additionally, it has been shown that a high content of annealing twins can reduce

the fatigue life of Ni-superalloys [115]. It was shown previously that the HIP

microstructure has a significantly greater number of annealing twins compared to the non-

HIP set of specimens - this can also be seen in Figure 107. There have been a number of

papers published on the fatigue life dependence on grain size. It’s typically seen that

smaller grains improves fatigue life and the endurance limit for polycrystalline alloys

[117], [118]. Due to the coarser grains and the addition of the annealing twins in the HIP

specimen, it’s believed that this is the root cause for the reduction of the endurance limit

when compared to the non-HIP specimen.

When the AM specimens are compared to the sheet metal specimens, it’s shown

that the performance of the AM parts is similar to the wrought specimens. However, a
162
significant reduction in fatigue performance is observed for the AM parts as the porosity,

surface roughness, and stress heterogeneities are more pronounced for the AM specimens.

Figure 107: An SEM image of a HIP specimen with a number of annealing twins (green arrows)
that have formed during the high temperatures of the HIP cycle.

6.5. Conclusions
The primary goal of this work was to elucidate the size effects that could be seen with

an additively manufactured thin-wall specimen, and if a HIP heat treatment would have

any added benefits to the fatigue life. These are the findings of this work:

1. Thin walls in LPBF show clear thickness dependence in High Cycle Fatigue, with
HIP clearly influencing HCF behavior under 0.75mm thickness. Above 0.75 mm
thickness, HIP does not have a significant influence on HCF life.
2. Due to a more homogenous microstructure and a reduction of the sub-surface
porosity, the HIP heat treating process creates a more consistent and predictable
fatigue life.

163
 3. The HIP heat treatment achieved a better HCF performance for the higher stress
values due to the reduction of the porosity that was shown to be a crack initiation
site for the non-HIP parts at high max-stress values.

The non-HIP specimens have an increased runout value due to the significant grain

refinement that can be found at the surface of the specimen while the HIP specimens

have an increased number of annealing twins that have been found in literature to

influence fatigue performance at lower stresses.

164
 CHAPTER 7

SUMMARY AND CONCLUSIONS

This dissertation is centered around developing a better understanding of how to

properly characterize AM thin-walled specimens (density and cross-section), examine the

thickness effects in a quasi-static tensile test and HCF test, and to determine if an added

HIP heat treatment cycle can increase the part performance – all of this while decoupling

the contributions of microstructure and morphology to observed behavior.

Chapter 3 examined the role of porosity while qualifying and quantifying the

Archimedes density method as the main tool for high throughput density measurements.

The following conclusions were obtained:

• The use of examining cross-sections and X-ray CT data proved to be pivotal in

understanding the location dependence of the porosity generated via the laser
powder bed fusion process.
• With the process parameters and LPBF machine used in this study, most of the
pores will be located close to the surface of the part near the turn of the interior scan
path.
• As the part thickness decreases, the amount of sub-surface porosity stays
consistent, but the thinner specimens will have a reduction of more dense internal
material which will decrease the part density.
• Most of the porosity formed is spherical in shape, but lack of fusion and sharp
internal topography of pores is still seen.

Chapter 4 investigated the proper metrology to measure the cross-sectional area of a

specimen and discovered the following:

165
 • The use of metrology can have significant impacts on the calculated
engineering stress values, and the subsequent measured mechanical properties
for thin specimens.
• CT data has the greatest resolution amongst all the metrologies evaluated, it has
the longest scan times and is cost-prohibitive for 100% sampling.
• Blue light scanning, while accurate for all but very thin sections, also takes
approximately 20 minutes per specimen to collect the data necessary, followed
by offline data processing to estimate an average section area. However, it
provides an amplitude of surface topology data throughout the gauge section that
can be processed for further analysis.
• The use of a micrometer is the quickest and cheapest way to obtain specimen
thickness, but the standard micrometer is inadequate for the task, and it is the
point micrometer that had the best performance.

Chapter 5 examined the thickness dependence of AM specimens during a quasi-static

tensile test:

• Yield strength for specimens without HIP show clear debits under 0.75mm wall
thickness, but the use of HIP removes this dependence.
o The reductions in yield strength (for no-HIP specimens) with thickness is
attributable to grain size distributions.
o HIP has the effect of homogenizing grain size across the specimen and
eliminating the above dependence.
• Ultimate tensile strength and elongation to failure both show reductions under
0.75mm wall thickness, independent of HIP condition.
o Stress heterogeneities due to surface topology are a contributing factor for
thinner specimens.
o Density of non-HIP specimens show strong correlation with UTS (for
thinner specimens) and elongation (for thicker specimens).
o DIC measurements show that specimen geometry influences ductility in
thin parts.
166
 • Young’s modulus does not show any systematic reductions with thickness.
o Sample internal/external defects and phase fraction changes don’t influence
modulus across thicknesses in the mesoscale size frame.
• An added HIP step significantly improves part density while increasing the UTS
and total elongation.
o The higher HIP temperatures allow recrystallization and homogenization of
the microstructure while eliminating brittle precipitates that form along the
grain boundary which act as sites for void nucleation.
• A mathematical model based on the Weibull cumulative distribution function was
shown to model thickness dependence well and demonstrates convergence of HIP
and NO-HIP data for most properties of interest at low thicknesses.

Finally, chapter 6 explored the effects of thickness dependence on a HCF test:

• Regardless of heat treatment, thickness dependence was seen for the thinnest
walls manufactured, while thin walls that are 0.75 mm thick and greater show
comparable results in HCF.
• Due to a more homogenous microstructure and a reduction of the sub-surface
porosity, it was shown that the HIP fatigue specimens have a more consistent and
predictable fatigue life.
• The HIP heat treatment achieved a better HCF performance for the higher stress
values due to the reduction of the porosity that was shown to be a crack initiation
site for the non-HIP specimens at high max stress values.
• The non-HIP specimens have an increased runout value due to the significant
grain refinement that can be found at the surface of the specimen while the HIP
specimens have an increased number of annealing twins that have been found in
literature to influence fatigue performance at lower stresses.

In summary, this work clearly highlights that debits in mechanical properties obtained

at thinner sections have different attributions and that while HIP can have some beneficial

effects for thin wall performance, it is not a panacea for mitigating mechanical property
167
debits. It has also demonstrated the criticality of accurately qualifying metrologies in the

context of thin-wall, as-printed AM specimens.

Further work is recommended to develop thin-wall specific process parameters that can

minimize porosity. Additionally, the use of laser rastering for grain size control is another

potential area for further exploration.

168
 REFERENCES

[1] J. Plocher and A. Panesar, “Review on design and structural optimisation in

additive manufacturing: Towards next-generation lightweight structures,” Mater
Des, p. 108164, Aug. 2019, doi: 10.1016/j.matdes.2019.108164.

[2] D. Bhate, Design for Additive Manufacturing: Concepts and Considerations for
the Aerospace Industry. SAE International, 2018.

[3] L. Zhang et al., “Energy absorption characteristics of metallic triply periodic

minimal surface sheet structures under compressive loading,” Addit Manuf, vol.
23, no. July, pp. 505–515, 2018, doi: 10.1016/j.addma.2018.08.007.

[4] R. Hedayati, M. Sadighi, M. Mohammadi Aghdam, and A. A. Zadpoor,

“Mechanical properties of additively manufactured thick honeycombs,” Materials,
vol. 9, no. 8, p. 613, 2016, doi: 10.3390/ma9080613.

[5] S. M. Thompson, Z. S. Aspin, N. Shamsaei, A. Elwany, and L. Bian, “Additive

manufacturing of heat exchangers: A case study on a multi-layered Ti-6Al-4V
oscillating heat pipe,” Addit Manuf, vol. 8, pp. 163–174, Oct. 2015, doi:
10.1016/j.addma.2015.09.003.

[6] K. C. Chuang, J. E. Grady, R. D. Draper, C. Patterson, and T. D. Santelle,

“Additive Manufacturing and Characterization of Ultem Polymers and
Composites,” The Composites and Advanced Materials Expo, p. 15, 2015.

[7] L. Zhang et al., “Energy absorption characteristics of metallic triply periodic

minimal surface sheet structures under compressive loading,” Addit Manuf, vol.
23, pp. 505–515, Oct. 2018, doi: 10.1016/j.addma.2018.08.007.

[8] Optisys, “Design for Additive Manufacturing: Transforming RF antennas through

intelligent optimisation,” Metal Additive Manufacturing, pp. 109–113, 2017.

[9] J. J. Lewandowski and M. Seifi, “Metal Additive Manufacturing: A Review of

Mechanical Properties,” Annu Rev Mater Res, vol. 46, no. 1, pp. 151–186, 2016,
doi: 10.1146/annurev-matsci-070115-032024.

[10] J. A. Gonzalez, J. Mireles, S. W. Stafford, M. A. Perez, C. A. Terrazas, and R. B.

Wicker, “Characterization of Inconel 625 fabricated using powder-bed-based
additive manufacturing technologies,” J Mater Process Technol, 2019, doi:
10.1016/j.jmatprotec.2018.08.031.

169
[11] A. Sterling, N. Shamsaei, B. Torries, and S. M. Thompson, “Fatigue Behaviour of
Additively Manufactured Ti-6Al-4 v,” in Procedia Engineering, 2015, vol. 133,
pp. 576–589. doi: 10.1016/j.proeng.2015.12.632.

[12] A. M. Roach, B. C. White, A. Garland, B. H. Jared, J. D. Carroll, and B. L. Boyce,

“Size-dependent stochastic tensile properties in additively manufactured 316L
stainless steel,” Addit Manuf, vol. 32, no. January, p. 101090, 2020, doi:
10.1016/j.addma.2020.101090.

[13] T. Horn et al., “Thickness dependency of mechanical properties for thin-walled

titanium parts manufactured by Electron Beam Melting (EBM) ®,” Addit Manuf,
2016, doi: 10.1016/j.addma.2016.06.009.

[14] C. Phutela, N. T. Aboulkhair, C. J. Tuck, and I. Ashcroft, “The Effects of Feature

Sizes in Selectively Laser Melted Ti-6Al-4V Parts on the Validity of Optimised
Process Parameters,” Materials, vol. 13, no. 1, p. 117, Dec. 2019, doi:
10.3390/ma13010117.

[15] J. K. Algardh et al., “Thickness dependency of mechanical properties for thin-

walled titanium parts manufactured by Electron Beam Melting (EBM)®,” Addit
Manuf, vol. 12, pp. 45–50, 2016, doi: 10.1016/j.addma.2016.06.009.

[16] B. Brown, W. Everhart, and J. Dinardo, “Characterization of bulk to thin wall

mechanical response transition in powder bed AM,” Proceedings - 26th Annual
International Solid Freeform Fabrication Symposium - An Additive Manufacturing
Conference, SFF 2015, pp. 501–513, 2015.

[17] H. Qi, M. Azer, and A. Ritter, “Studies of standard heat treatment effects on
microstructure and mechanical properties of laser net shape manufactured
INCONEL 718,” Metall Mater Trans A Phys Metall Mater Sci, vol. 40, no. 10, pp.
2410–2422, 2009, doi: 10.1007/s11661-009-9949-3.

[18] L. Lavanya Parimi, M. M. Attallah, J. Gebelin, and R. C. Reed, “DIRECT LASER

FABRICATION OF INCONEL-718: EFFECTS ON DISTORTION AND
MICROSTRUCTURE,” 2012.

[19] D. F. Paulonis and J. J. Schirra, “Alloy 718 at Pratt & Whitney – Historical
Perspective and Future Challenges.”

[20] R. E. Schafrik, D. D. Ward, and J. R. Groh, “Application of Alloy 718 in GE

Aircraft Engines: Past, Present and Next Five Years.”

170
[21] E. Hosseini and V. A. Popovich, “A review of mechanical properties of additively
manufactured Inconel 718,” Additive Manufacturing, vol. 30. Elsevier B.V., Dec.
01, 2019. doi: 10.1016/j.addma.2019.100877.

[22] F. Liu, F. Lyu, F. Liu, X. Lin, and C. Huang, “Laves phase control of inconel 718
superalloy fabricated by laser direct energy deposition via ı aging and solution
treatment,” Journal of Materials Research and Technology, vol. 9, no. 5, pp.
9753–9765, 2020, doi: 10.1016/j.jmrt.2020.06.061.

[23] A. Strondl, R. Fischer, G. Frommeyer, and A. Schneider, “Investigations of MX

and γ′/γ″ precipitates in the nickel-based superalloy 718 produced by electron
beam melting,” Materials Science and Engineering A, vol. 480, no. 1–2, pp. 138–
147, May 2008, doi: 10.1016/j.msea.2007.07.012.

[24] M. E. Aydinöz et al., “On the microstructural and mechanical properties of post-
treated additively manufactured Inconel 718 superalloy under quasi-static and
cyclic loading,” Materials Science and Engineering A, vol. 669, pp. 246–258, Jul.
2016, doi: 10.1016/j.msea.2016.05.089.

[25] “AMS 5662L (AEROSPACE MATERIAL SPECIFICATION),” 1998.

[26] ASTM E8, “ASTM E8/E8M standard test methods for tension testing of metallic
materials 1,” Annual Book of ASTM Standards 4, 2010, doi: 10.1520/E0008.

[27] A. International, “E466: Standard Practice for Conducting Force Controlled

Constant Amplitude Axial Fatigue Tests of Metallic Materials,” vol. i, pp. 1–6,
2018, doi: 10.1520/E0466-15.2.

[28] S. Vock, B. Klöden, A. Kirchner, T. Weißgärber, and B. Kieback, “Powders for

powder bed fusion: a review,” Progress in Additive Manufacturing, vol. 4, no. 4.
Springer, pp. 383–397, Dec. 01, 2019. doi: 10.1007/s40964-019-00078-6.

[29] L. C. Ardila et al., “Effect of IN718 recycled powder reuse on properties of parts
manufactured by means of Selective Laser Melting,” in Physics Procedia, 2014,
vol. 56, no. C, pp. 99–107. doi: 10.1016/j.phpro.2014.08.152.

[30] “Designation: F3055 − 14a Standard Specification for Additive Manufacturing

Nickel Alloy (UNS N07718) with Powder Bed Fusion 1”, doi: 10.1520/F3055-
14A.

[31] “AMS2774E - IN718 Heat Treatment (SAE Standard)”.

171
[32] W. Wang, J. Ning, and S. Y. Liang, “Analytical prediction of keyhole porosity in
laser powder bed fusion,” International Journal of Advanced Manufacturing
Technology, 2022, doi: 10.1007/s00170-021-08276-9.

[33] Y. Chen et al., “Synchrotron Imaging of Keyhole Mode Multi-layer Laser Powder
Bed Fusion Additive Manufacturing.”

[34] W. Wang and S. Y. Liang, “A 3D analytical modeling method for keyhole porosity
prediction in laser powder bed fusion,” International Journal of Advanced
Manufacturing Technology, vol. 120, no. 5–6, pp. 3017–3025, May 2022, doi:
10.1007/s00170-022-08898-7.

[35] A. Ulbricht, G. Mohr, S. J. Altenburg, S. Oster, C. Maierhofer, and G. Bruno,

“Can potential defects in lpbf be healed from the laser exposure of subsequent
layers? A quantitative study,” Metals (Basel), vol. 11, no. 7, Jul. 2021, doi:
10.3390/met11071012.

[36] J. v. Gordon et al., “Defect structure process maps for laser powder bed fusion
additive manufacturing,” Addit Manuf, vol. 36, Dec. 2020, doi:
10.1016/j.addma.2020.101552.

[37] A. A. Martin et al., “Dynamics of pore formation during laser powder bed fusion
additive manufacturing,” Nat Commun, vol. 10, no. 1, Dec. 2019, doi:
10.1038/s41467-019-10009-2.

[38] A. B. Spierings, M. Schneider, and R. Eggenberger, “Comparison of density

measurement techniques for additive manufactured metallic parts,” Rapid Prototyp
J, vol. 17, no. 5, pp. 380–386, 2011, doi: 10.1108/13552541111156504.

[39] A. du Plessis and C. Broeckhoven, “Looking deep into nature: A review of micro-
computed tomography in biomimicry,” Acta Biomater, vol. 85, pp. 27–40, Feb.
2019, doi: 10.1016/j.actbio.2018.12.014.

[40] ASTM International, “Standard Test Methods for Density of Compacted or

Sintered Powder Metallurgy (PM) Products Using Archimedes’ Principle,” Astm
B962-13, vol. i, pp. 1–7, 2013, doi: 10.1520/B0962-17.2.

[41] ASTM B311-17, “Standard Test Method for Density of Powder Metallurgy ( PM )
Materials Containing Less Than Two Percent Porosity,” ASTM International, vol.
i, pp. 1–5, 2017, doi: 10.1520/B0311-17.2.

[42] K. N. Amato et al., “Microstructures and mechanical behavior of Inconel 718

fabricated by selective laser melting,” Acta Mater, vol. 60, no. 5, pp. 2229–2239,
2012, doi: 10.1016/j.actamat.2011.12.032.

172
[43] W. Tillmann, C. Schaak, J. Nellesen, M. Schaper, M. E. Aydinöz, and K. P. Hoyer,
“Hot isostatic pressing of IN718 components manufactured by selective laser
melting,” Addit Manuf, vol. 13, pp. 93–102, 2017, doi:
10.1016/j.addma.2016.11.006.

[44] A. M. Roach, B. C. White, A. Garland, B. H. Jared, J. D. Carroll, and B. L. Boyce,

“Size-dependent stochastic tensile properties in additively manufactured 316L
stainless steel,” Addit Manuf, vol. 32, no. January, p. 101090, 2020, doi:
10.1016/j.addma.2020.101090.

[45] T. Horn et al., “Thickness dependency of mechanical properties for thin-walled

titanium parts manufactured by Electron Beam Melting (EBM) ®,” Addit Manuf,
2016, doi: 10.1016/j.addma.2016.06.009.

[46] C. Phutela, N. T. Aboulkhair, C. J. Tuck, and I. Ashcroft, “The Effects of Feature

Sizes in Selectively Laser Melted Ti-6Al-4V Parts on the Validity of Optimised
Process Parameters,” Materials, vol. 13, no. 1, p. 117, Dec. 2019, doi:
10.3390/ma13010117.

[47] J. K. Algardh et al., “Thickness dependency of mechanical properties for thin-

walled titanium parts manufactured by Electron Beam Melting (EBM)®,” Addit
Manuf, vol. 12, pp. 45–50, 2016, doi: 10.1016/j.addma.2016.06.009.

[48] B. Brown, W. Everhart, and J. Dinardo, “Characterization of bulk to thin wall

mechanical response transition in powder bed AM,” Proceedings - 26th Annual
International Solid Freeform Fabrication Symposium - An Additive Manufacturing
Conference, SFF 2015, pp. 501–513, 2015.

[49] M. L. George, J. Maxey, D. Rowlands, and M. Price, The Lean Six Sigma Pocket
Toolbook: A Quick Reference Guide to 100 Tools for Improving Quality and
Speed, 1st ed. McGraw-Hill, 2004.

[50] P. Paradise, D. Patil, N. van Handel, S. Temes, A. Saxena, and D. Bruce,

“Improving Productivity in the Laser Powder Bed Fusion of Inconel 718 by
Increasing Layer Thickness : Effects on Mechanical Behavior,” UNDER REVIEW.

[51] ASTM International, Standard Test Methods for Tension Testing of Metallic
Materials. pp. 1–27. doi: 10.1520/E0008.

[52] ASTM E8, “ASTM E8/E8M standard test methods for tension testing of metallic
materials 1,” Annual Book of ASTM Standards 4, 2010, doi: 10.1520/E0008.

173
[53] K. J. Hemker and W. N. Sharpe, “Microscale characterization of mechanical
properties,” Annu Rev Mater Res, vol. 37, pp. 92–126, 2007, doi:
10.1146/annurev.matsci.36.062705.134551.

[54] M. D. Uchic, D. M. Dimiduk, J. N. Florando, and W. D. Nix, “Sample dimensions

influence strength and crystal plasticity,” Science (1979), vol. 305, no. 5686, pp.
986–989, 2004, doi: 10.1126/science.1098993.

[55] M. Legros, B. R. Elliott, M. N. Rittner, J. R. Weertman, and K. J. Hemker,

“Microsample tensile testing of nanocrystalline metals,” Philosophical Magazine
A: Physics of Condensed Matter, Structure, Defects and Mechanical Properties,
vol. 80, no. 4, pp. 1017–1026, 2000, doi: 10.1080/01418610008212096.

[56] W. N. Sharpe, K. M. Jackson, K. J. Hemker, and Z. Xie, “Effect of specimen size

on young’s modulus and fracture strength of polysilicon,” Journal of
Microelectromechanical Systems, vol. 10, no. 3, pp. 317–326, 2001, doi:
10.1109/84.946774.

[57] J. Ye and C. Zhou, “Time-resolved coded structured light for 3D measurement,”

Microw Opt Technol Lett, vol. 63, no. 1, pp. 5–12, 2021, doi: 10.1002/mop.32548.

[58] S. Zhang, “High-speed 3D shape measurement with structured light methods: A

review,” Opt Lasers Eng, vol. 106, no. March, pp. 119–131, 2018, doi:
10.1016/j.optlaseng.2018.02.017.

[59] J. Ye, C. Zhou, J. Zhang, and J. Xie, “Miniaturized 3-D fourier transform
profilometry,” Optics InfoBase Conference Papers, no. 390, pp. 3–4, 2009.

[60] J. Geng, “Structured-light 3D surface imaging: a tutorial,” Adv Opt Photonics, vol.
3, no. 2, p. 128, 2011, doi: 10.1364/aop.3.000128.

[61] K. Zhong, Z. Li, X. Zhou, Y. Li, Y. Shi, and C. Wang, “Enhanced phase
measurement profilometry for industrial 3D inspection automation,” International
Journal of Advanced Manufacturing Technology, vol. 76, no. 9–12, pp. 1563–
1574, 2015, doi: 10.1007/s00170-014-6360-z.

[62] “Development of a high-precision surface metrology, Ming june Tsai.pdf.”

[63] K. Wi, V. Suresh, K. Wang, B. Li, and H. Qin, “Quantifying quality of 3D printed
clay objects using a 3D structured light scanning system,” Addit Manuf, vol. 32,
no. December 2019, p. 100987, 2020, doi: 10.1016/j.addma.2019.100987.

174
[64] A. B. Spierings, M. Schneider, and R. Eggenberger, “Comparison of density
measurement techniques for additive manufactured metallic parts,” Rapid Prototyp
J, vol. 17, no. 5, pp. 380–386, 2011, doi: 10.1108/13552541111156504.

[65] A. du Plessis, I. Yadroitsev, I. Yadroitsava, and S. G. Le Roux, “X-Ray

Microcomputed Tomography in Additive Manufacturing: A Review of the Current
Technology and Applications,” 3D Print Addit Manuf, vol. 5, no. 3, p.
3dp.2018.0060, 2018, doi: 10.1089/3dp.2018.0060.

[66] ASTM International, Standard Test Methods for Tension Testing of Metallic
Materials. pp. 1–27. doi: 10.1520/E0008.

[67] 3D Systems, “Geomagic Control X.” 2018.

[68] Inc. The Math Works, “Matlab.” 2020.

[69] W. S. Rasband, “ImageJ.” U.S. National Institutes of Health, Bethesda, MD, 1997.

[70] G. (Zeiss), “Aramis.” 2020.

[71] H. Qi, M. Azer, and A. Ritter, “Studies of standard heat treatment effects on
microstructure and mechanical properties of laser net shape manufactured
INCONEL 718,” Metall Mater Trans A Phys Metall Mater Sci, vol. 40, no. 10, pp.
2410–2422, 2009, doi: 10.1007/s11661-009-9949-3.

[72] F. Brenne et al., “Microstructural design of Ni-base alloys for high-temperature

applications: impact of heat treatment on microstructure and mechanical properties
after selective laser melting,” Progress in Additive Manufacturing, vol. 1, no. 3–4,
pp. 141–151, Dec. 2016, doi: 10.1007/s40964-016-0013-8.

[73] T. Trosch, J. Strößner, R. Völkl, and U. Glatzel, “Microstructure and mechanical

properties of selective laser melted Inconel 718 compared to forging and casting,”
Materials Letters, vol. 164. Elsevier B.V., pp. 428–431, Feb. 01, 2016. doi:
10.1016/j.matlet.2015.10.136.

[74] X. Zhao, J. Chen, X. Lin, and W. Huang, “Study on microstructure and mechanical
properties of laser rapid forming Inconel 718,” Materials Science and Engineering
A, vol. 478, no. 1–2, pp. 119–124, Apr. 2008, doi: 10.1016/j.msea.2007.05.079.

[75] A. Strondl, M. Palm, J. Gnauk, and G. Frommeyer, “Microstructure and

mechanical properties of nickel based superalloy IN718 produced by rapid
prototyping with electron beam melting (EBM),” Materials Science and
Technology, vol. 27, no. 5, pp. 876–883, 2011, doi:
10.1179/026708309X12468927349451.

175
[76] Z. Wang, K. Guan, M. Gao, X. Li, X. Chen, and X. Zeng, “The microstructure and
mechanical properties of deposited-IN718 by selective laser melting,” J Alloys
Compd, vol. 513, pp. 518–523, Feb. 2012, doi: 10.1016/j.jallcom.2011.10.107.

[77] F. Liu et al., “The effect of laser scanning path on microstructures and mechanical
properties of laser solid formed nickel-base superalloy Inconel 718,” J Alloys
Compd, vol. 509, no. 13, pp. 4505–4509, Mar. 2011, doi:
10.1016/j.jallcom.2010.11.176.

[78] S. Raghavan et al., “Effect of different heat treatments on the microstructure and
mechanical properties in selective laser melted INCONEL 718 alloy,” Materials
and Manufacturing Processes, vol. 32, no. 14, pp. 1588–1595, Oct. 2017, doi:
10.1080/10426914.2016.1257805.

[79] D. Zhang, W. Niu, X. Cao, and Z. Liu, “Effect of standard heat treatment on the
microstructure and mechanical properties of selective laser melting manufactured
Inconel 718 superalloy,” Materials Science and Engineering A, vol. 644, pp. 32–
40, Sep. 2015, doi: 10.1016/j.msea.2015.06.021.

[80] E. Chlebus, K. Gruber, B. Kuźnicka, J. Kurzac, and T. Kurzynowski, “Effect of

heat treatment on the microstructure and mechanical properties of Inconel 718
processed by selective laser melting,” Materials Science and Engineering A, vol.
639, pp. 647–655, Jul. 2015, doi: 10.1016/j.msea.2015.05.035.

[81] V. A. Popovich, E. v. Borisov, A. A. Popovich, V. S. Sufiiarov, D. v. Masaylo, and

L. Alzina, “Impact of heat treatment on mechanical behaviour of Inconel 718
processed with tailored microstructure by selective laser melting,” Mater Des, vol.
131, pp. 12–22, Oct. 2017, doi: 10.1016/j.matdes.2017.05.065.

[82] D. Ivanov et al., “Evolution of structure and properties of the nickel-based alloy
EP718 after the SLM growth and after different types of heat and mechanical
treatment,” Addit Manuf, vol. 18, pp. 269–275, Dec. 2017, doi:
10.1016/j.addma.2017.10.015.

[83] M. M. Kirka, F. Medina, R. Dehoff, and A. Okello, “Mechanical behavior of post-

processed Inconel 718 manufactured through the electron beam melting process,”
Materials Science and Engineering A, vol. 680, pp. 338–346, Jan. 2017, doi:
10.1016/j.msea.2016.10.069.

[84] D. Deng, J. Moverare, R. L. Peng, and H. Söderberg, “Microstructure and

anisotropic mechanical properties of EBM manufactured Inconel 718 and effects
of post heat treatments,” Materials Science and Engineering A, vol. 693, pp. 151–
163, May 2017, doi: 10.1016/j.msea.2017.03.085.

176
[85] D. Deng, R. L. Peng, H. Brodin, and J. Moverare, “Microstructure and mechanical
properties of Inconel 718 produced by selective laser melting: Sample orientation
dependence and effects of post heat treatments,” Materials Science and
Engineering A, vol. 713, pp. 294–306, Jan. 2018, doi: 10.1016/j.msea.2017.12.043.

[86] M. M. Kirka, K. A. Unocic, N. Raghavan, F. Medina, R. R. Dehoff, and S. S.

Babu, “Microstructure Development in Electron Beam-Melted Inconel 718 and
Associated Tensile Properties,” JOM, vol. 68, no. 3, pp. 1012–1020, Mar. 2016,
doi: 10.1007/s11837-016-1812-6.

[87] M. Ni et al., “Anisotropic tensile behavior of in situ precipitation strengthened

Inconel 718 fabricated by additive manufacturing,” Materials Science and
Engineering A, vol. 701, pp. 344–351, Jul. 2017, doi: 10.1016/j.msea.2017.06.098.

[88] Y. Lu et al., “Study on the microstructure, mechanical property and residual stress
of SLM Inconel-718 alloy manufactured by differing island scanning strategy,”
Opt Laser Technol, vol. 75, pp. 197–206, Jul. 2015, doi:
10.1016/j.optlastec.2015.07.009.

[89] V. A. Popovich, E. v. Borisov, A. A. Popovich, V. S. Sufiiarov, D. v. Masaylo, and

L. Alzina, “Functionally graded Inconel 718 processed by additive manufacturing:
Crystallographic texture, anisotropy of microstructure and mechanical properties,”
Mater Des, vol. 114, pp. 441–449, Jan. 2017, doi: 10.1016/j.matdes.2016.10.075.

[90] S. H. Chang, “In situ TEM observation of γ′, γ″ and δ precipitations on Inconel
718 superalloy through HIP treatment,” J Alloys Compd, vol. 486, no. 1–2, pp.
716–721, Nov. 2009, doi: 10.1016/j.jallcom.2009.07.046.

[91] “Designation: F3055 − 14a Standard Specification for Additive Manufacturing

Nickel Alloy (UNS N07718) with Powder Bed Fusion 1”, doi: 10.1520/F3055-
14A.

[92] R. Jiang, A. Mostafaei, J. Pauza, C. Kantzos, and A. D. Rollett, “Varied heat

treatments and properties of laser powder bed printed Inconel 718,” Materials
Science and Engineering A, vol. 755, pp. 170–180, May 2019, doi:
10.1016/j.msea.2019.03.103.

[93] I. Yadroitsev, I. Yadroitsava, A. du Plessis, and E. MacDonald, Eds.,

Fundamentals of Laser Powder Bed Fusion of Metals, 1st ed. Elsevier, 2021.

[94] D. Bruce et al., “A critical assessment of the Archimedes density method for thin-
wall specimens in laser powder bed fusion: Measurement capability, process
sensitivity and property correlation,” J Manuf Process, vol. 79, pp. 185–192, Jul.
2022, doi: 10.1016/j.jmapro.2022.04.059.

177
[95] P. Paradise et al., “Section Area Estimation Methods for Determining the
Mechanical Properties of As-Built Laser Powder Bed Fusion Thin Wall
Structures,” Manuf Lett, vol. in press, no. Proceedings of NAMRC, 2022.

[96] P. Paradise et al., “Manufacturing Letters Section Area Estimation Methods for
Determining the Mechanical Properties of Laser Powder Bed Fusion Thin Wall
Structures,” 2022. [Online]. Available: www.sciencedirect.com

[97] Y. Tian, J. A. Muñiz-Lerma, and M. Brochu, “Nickel-based superalloy

microstructure obtained by pulsed laser powder bed fusion,” Mater Charact, vol.
131, no. March, pp. 306–315, 2017, doi: 10.1016/j.matchar.2017.07.024.

[98] V. A. Popovich, E. v. Borisov, A. A. Popovich, V. S. Sufiiarov, D. v. Masaylo, and

L. Alzina, “Functionally graded Inconel 718 processed by additive manufacturing:
Crystallographic texture, anisotropy of microstructure and mechanical properties,”
Mater Des, vol. 114, pp. 441–449, 2017, doi: 10.1016/j.matdes.2016.10.075.

[99] V. A. Popovich, E. v. Borisov, A. A. Popovich, V. S. Sufiiarov, D. v. Masaylo, and

L. Alzina, “Functionally graded Inconel 718 processed by additive manufacturing:
Crystallographic texture, anisotropy of microstructure and mechanical properties,”
Mater Des, vol. 114, pp. 441–449, 2017, doi: 10.1016/j.matdes.2016.10.075.

[100] J. Y. Zhang, X. Zhang, G. Liu, R. H. Wang, G. J. Zhang, and J. Sun, “Length scale
dependent yield strength and fatigue behavior of nanocrystalline Cu thin films,”
Materials Science and Engineering A, vol. 528, no. 25–26, pp. 7774–7780, Sep.
2011, doi: 10.1016/j.msea.2011.06.083.

[101] K. J. Hemker and W. N. Sharpe, “Microscale characterization of mechanical

properties,” Annual Review of Materials Research, vol. 37. pp. 92–126, 2007. doi:
10.1146/annurev.matsci.36.062705.134551.

[102] S. L. Sindinger, D. Marschall, C. Kralovec, and M. Schagerl, “Material modelling

and property mapping for structural FEA of thin-walled additively manufactured
components,” Virtual Phys Prototyp, vol. 16, no. 1, pp. 97–112, 2021, doi:
10.1080/17452759.2020.1824427.

[103] A. S. Johnson, S. Shao, N. Shamsaei, S. M. Thompson, and L. Bian,

“Microstructure, Fatigue Behavior, and Failure Mechanisms of Direct Laser-
Deposited Inconel 718,” JOM, vol. 69, no. 3, pp. 597–603, Mar. 2017, doi:
10.1007/s11837-016-2225-2.

178
[104] R. Konečná, G. Nicoletto, L. Kunz, and A. Bača, “Microstructure and directional
fatigue behavior of Inconel 718 produced by selective laser melting,” Procedia
Structural Integrity, vol. 2, pp. 2381–2388, 2016, doi:
10.1016/j.prostr.2016.06.298.

[105] D. B. Witkin, D. N. Patel, and G. E. Bean, “Notched fatigue testing of Inconel 718
prepared by selective laser melting,” Fatigue Fract Eng Mater Struct, vol. 42, no.
1, pp. 166–177, Jan. 2019, doi: 10.1111/ffe.12880.

[106] A. R. Balachandramurthi, J. Moverare, N. Dixit, and R. Pederson, “Influence of

defects and as-built surface roughness on fatigue properties of additively
manufactured Alloy 718,” Materials Science and Engineering A, vol. 735, pp.
463–474, Sep. 2018, doi: 10.1016/j.msea.2018.08.072.

[107] S. Sui, J. Chen, E. Fan, H. Yang, X. Lin, and W. Huang, “The influence of Laves
phases on the high-cycle fatigue behavior of laser additive manufactured Inconel
718,” Materials Science and Engineering A, vol. 695, pp. 6–13, May 2017, doi:
10.1016/j.msea.2017.03.098.

[108] K. Solberg, J. Torgersen, and F. Berto, “Fatigue Behaviour of Additively

Manufactured Inconel 718 Produced by Selective Laser Melting.,” in Procedia
Structural Integrity, 2018, vol. 13, pp. 1762–1767. doi:
10.1016/j.prostr.2018.12.371.

[109] Y. Yamashita, T. Murakami, R. Mihara, M. Okada, and Y. Murakami, “Defect

Analysis and Fatigue Design Basis for Ni-based Superalloy 718 manufactured by
Additive Manufacturing,” in Procedia Structural Integrity, 2017, vol. 7, pp. 11–
18. doi: 10.1016/j.prostr.2017.11.054.

[110] L. Sheridan, O. E. Scott-Emuakpor, T. George, and J. E. Gockel, “Relating

porosity to fatigue failure in additively manufactured alloy 718,” Materials
Science and Engineering A, vol. 727, no. April, pp. 170–176, 2018, doi:
10.1016/j.msea.2018.04.075.

[111] V. P. Sabelkin et al., “Mitigation of anisotropic fatigue in nickel alloy 718

manufactured via selective laser melting,” Mater Des, vol. 182, p. 108095, Nov.
2019, doi: 10.1016/j.matdes.2019.108095.

[112] H. Y. Wan, Z. J. Zhou, C. P. Li, G. F. Chen, and G. P. Zhang, “Enhancing Fatigue

Strength of Selective Laser Melting-Fabricated Inconel 718 by Tailoring Heat
Treatment Route,” Adv Eng Mater, vol. 20, no. 10, Oct. 2018, doi:
10.1002/adem.201800307.

179
[113] G. F. vander Voort and E. P. Manilova, “Metallographic techniques for
superalloys,” Microscopy and Microanalysis, vol. 10, no. SUPPL. 2, pp. 690–691,
2004, doi: 10.1017/S1431927604883442.

[114] R. Kapoor, “Severe Plastic Deformation of Materials,” in Materials Under

Extreme Conditions: Recent Trends and Future Prospects, Elsevier Inc., 2017, pp.
717–754. doi: 10.1016/B978-0-12-801300-7.00020-6.

[115] S. Gribbin, J. Bicknell, L. Jorgensen, I. Tsukrov, and M. Knezevic, “Low cycle

fatigue behavior of direct metal laser sintered Inconel alloy 718,” Int J Fatigue,
vol. 93, pp. 156–167, Dec. 2016, doi: 10.1016/j.ijfatigue.2016.08.019.

[116] C. Yu, Z. Huang, Z. Zhang, J. Shen, J. Wang, and Z. Xu, “Influence of post-
processing on very high cycle fatigue resistance of Inconel 718 obtained with laser
powder bed fusion,” Int J Fatigue, vol. 153, Dec. 2021, doi:
10.1016/j.ijfatigue.2021.106510.

[117] P. Lukj~ and L. Kunz, “Effect of Grain Size on the High Cycle Fatigue Behaviour
of Polycrystalline Copper,” 1987.

[118] A. W. Thompsonts and W. A. Backopen, “THE EFFECT OF GRAIN SIZE ON

FATIGUE*.”

180