Professional Documents
Culture Documents
Timoshenko Beam 18-26
Timoshenko Beam 18-26
Timoshenko Beam 18-26
net/publication/349427879
CITATIONS READS
0 1,392
1 author:
SEE PROFILE
All content following this page was uploaded by Sebastian Maklary Rudbeck-Rønne on 19 February 2021.
Faculty of Engineering
University of Southern Denmark
Author:
Sebastian Maklary Rudbeck-Rønne Supervisors:
(formerly Jørgensen) Assistant. Prof. Andreas
Project period: Mehl
This thesis is part of the master’s degree Structural Engineering at the University of Southern
Denmark and has been written by Sebastian Maklary Jørgensen. During the span of this project, the
author has become a father and the COVID-19 crisis occurred, which extended the project deadline
from the initial date of January 2020 to August 2020. As this extension is due to the events
prohibiting the author from working full-time on the project and thus the extent of the project is
equivalent to that of 10 ECTS over a 4 month period plus 30 ECTS over a 4 month period.
___________________________________
Signature
__________________________________
Date 03/08 - 2020
i
Changelog
- V1.0, 3rd of August 2020: Finished report at the date of hand in.
- V2.0, 11th of August 2020: Errors related to multilinear plasticity modeling located and
some results of the simulations added in sections 5.2.6, 5.2.7 and 5.3.
Rules changes regarding loophole added in section 3.1.
- V2.1, 17th of August 2020: Added a comment on the stiffness of multilinear simulation in
section 5.3 along with graph comparisons.
Added comment in section 5.3 on the stiffness of the simulation of the 3PB specimen with
10mm core using the full model with finer mesh.
- V2.2, 20th of August 2020: Added analysis of the PSS test data and comments on an
empirical model to section 5.4
- V2.3, 23rd of September 2020: Name changed on front page due to marriage. Error in
conclusion corrected relating to core material.
ii
Resumé (Danish)
I dette projekt behandles problemstillingen med at modellere laminerede kompositplader med tykke
skumkerner belastet i 3-punkts bøjning. Formålet med modelleringen er at give studerende på
bachelor-niveau de nødvendige værktøjer til at kunne lave foreløbige designvalg i udformningen af
en sikker og regelret monocoque til deltagelse i ingeniørkonkurrencen Formula Student.
Et eksakt bjælkeelement for laminerede bjælker er udledt med inddragelse af
forskydningsdeformation ved antagelse af konstant forskydningstøjning og en korrektionsfaktor
baseret på elasticitetsteori. En 2D bjælkemodel er også benyttet med disse bjælkeelementer som
kompositlagene og QM6 plane elementer som kernemateriale således at kompression af
kernematerialet kan modelleres. Det kommercielle software ANSYS er brugt til at evaluere
bjælkeantagelsen ved sammenligning med shell-elementer og til at lave en 3-dimensionel model
hvor nonlineariteter inkluderes med en bi-lineær plasticitetsmodel. For at validere disse modeller er
fysiske tests af 3-punkts bøjning udført med 5 forskellige lamineringer, så indflydelsen af antallet af
kompositlag, kernetykkelse og kernestivhed kan redegøres for. Materialetests er udført for
kompositmaterialet og for kernematerialet og tests er udført for at udregne stivheden af
testopstillingen så der korrigeres for denne i sammenligningen mellem simulerede og
eksperimentelle data.
De eksperimentelle data af 3-punkts bøjningsemnerne har en spredning på mindre end 2,3% for
stivheden og under 10% for energi absorption og styrker. Sammenlignet med middelværdierne fra
de eksperimentelle resultater er fejlen for stivhed 18,3 til 35,9% for den 1-dimensionelle
bjælkemodel, 2D bjælkemodellen har en fejl på 11,5 til 27,4% og 3D modellen har en fejl på 0,2 til
12,9%. Bjælkemodellerne er ikke generelt retvisende for udregning af styrker og energi absorption
da nonlineriteter ikke medregnes. For 3D modellen er fejlen for energi absorption 6,7 til 55,9%
mens fejlen for styrken er 3 til 35,1% og uoverensstemmelser skyldes antageligvis primært at den
bi-lineære plasticitetsmodel er utilstrækkeligt præcis til at være generelt anvendelig.
Disse modeller er brugt til at finde det optimale laminat til side beskyttelses-strukturen med en vægt
på 6,3217 kg per kvadratmeter, hvilket sandsynligvis resulterer i en monocoque med en vægt under
20 kg.
Penetrationstests som er påkrævede af Formula Student reglerne er også udført, men det er ikke
fundet muligt at modellere styrken af disse emner.
iii
Abstract
In this project, the problem of modelling 3-point bending of laminated composite plates with thick
foam cores are treated with the purpose of enabling bachelor-level students to make preliminary
design choices in regards to the construction of a safe and rules-compliant composite monocoque
for the Formula Student engineering competition. An exact beam element for laminated beams are
developed including axial degree of freedom and bending-elongation coupling due to asymmetry
and shear deformation are accounted for by fist-order shear deformation theory and a correction
factor. A beam model using the mentioned beam elements as skin and QM6 plane elements for the
core is also developed for the inclusion of compressive deformation of the core and the commercial
software ANSYS is used to evaluate the beam assumption by comparison of beam elements to shell
elements and ANSYS is used to create a 3-dimensional nonlinear model for simulation of the full
load-displacement curve with a bilinear plasticity model. To validate these models, physical tests
are conducted with 5 different layups to evaluate the influence of skin thickness, core thickness and
core stiffness. Material tests of the components of the laminate is also conducted to validate the
model against experimental values without relying fully on data sheet data and tests of the testing
machine are conducted to correct for the flexibility of the testing setup.
The experimental data of the 3-point bending specimens are found to have a spread of less than
2,3% for the stiffness and less than 10% for the strengths and energy absorptions. Compared to the
mean of the experimental results, the error of the stiffness is 18,3 to 35,9% for the 1D beam model,
the 2D beam model has errors of 11,5 to 27,4% and the 3D model has an error of 0,2 to 12,9%. The
error for the strengths and energy absorption is great for the 1D and 2D models due to not
modelling nonlinear behavior. For the 3D model, the error of energy absorption is 6,7 to 55,9% and
the error of strength is 3 to 35,1%, with the inconsistencies presumably due to the bilinear plasticity
model being insufficiently accurate.
The models are used to perform a brute-force optimization with regards to mass of the side impact
structure and the lightest rules-compliant laminate has a mass of 6,3217 kg per square meter, which
would likely result in a monocoque mass of less than 20 kg.
Perimeter shear stress tests have also been conducted but attempts at modeling the failures have
failed.
iv
Contents
Preface .................................................................................................................................................. 1
Changelog ............................................................................................................................................ 2
Resumé (Danish) .................................................................................................................................. 3
Abstract ................................................................................................................................................ 4
Contents ............................................................................................................................................... 5
Nomenclature ....................................................................................................................................... 7
List of figures ..................................................................................................................................... 10
List of tables ....................................................................................................................................... 14
1 Introduction .................................................................................................................................. 1
2 Problem statement ........................................................................................................................ 2
3 Theory .......................................................................................................................................... 3
3.1 Rules & regulations ............................................................................................................... 3
3.2 Composite materials .............................................................................................................. 7
3.3 Laminate theory ..................................................................................................................... 9
3.4 Laminated beams ................................................................................................................. 14
3.5 The beam assumption for bending of a plate ...................................................................... 15
3.6 Bernoulli-Euler beams (Classical beam theory) .................................................................. 17
3.7 Timoshenko beams (First-order shear deformation theory) ................................................ 18
3.7.1 Shear correction factor ................................................................................................. 24
3.7.2 Stress calculations ........................................................................................................ 26
3.8 Other beam theories............................................................................................................. 31
3.8.1 Sandwich theory ........................................................................................................... 31
3.8.2 R. P. Shimpi single-variable shear deformation theory ............................................... 32
3.9 2D plane stress finite elements ............................................................................................ 35
3.9.1 Validity of element formulations ................................................................................. 41
3.10 Other 2D and 3D elements for laminate analysis ............................................................ 44
3.11 Nonlinear analysis using ANSYS .................................................................................... 46
3.11.1 Geometric nonlinearities .............................................................................................. 46
3.11.2 Material nonlinearities ................................................................................................. 50
3.11.3 Boundary condition nonlinearities ............................................................................... 54
3.12 Matlab FEM procedure – Boundary conditions .............................................................. 55
v
3.13 Matlab FEM procedure – Solver ..................................................................................... 58
3.14 Statistical analysis ............................................................................................................ 59
4 Method ....................................................................................................................................... 61
4.1 Material choice .................................................................................................................... 61
4.2 Specimen manufacturing ..................................................................................................... 62
4.3 3 point bending test ............................................................................................................. 66
4.3.1 Test setup stiffness correction and baseline tube test .................................................. 68
4.3.2 Simulation method – Matlab ........................................................................................ 71
4.3.3 Simulation method – ANSYS ...................................................................................... 74
4.4 Perimeter shear stress test.................................................................................................... 76
4.4.1 Simulation method – ANSYS ...................................................................................... 77
4.5 Tensile tests ......................................................................................................................... 78
4.6 Compressive tests ................................................................................................................ 78
4.7 Fiber volume fraction analysis ............................................................................................ 79
5 Results ........................................................................................................................................ 82
5.1 Defects ................................................................................................................................. 82
5.1.1 Contaminants ............................................................................................................... 82
5.1.2 Voids ............................................................................................................................ 83
5.1.3 Large-scale surface unevenness ................................................................................... 85
5.2 Material properties .............................................................................................................. 86
5.2.1 Fiber and void volume fraction of SE48LV-HEC ....................................................... 87
5.2.2 Composite tensile test – longitudinal direction ............................................................ 90
5.2.3 Composite tensile test – transverse direction ............................................................... 93
5.2.4 Composite compressive test – longitudinal direction .................................................. 93
5.2.5 Composite compressive test – transverse direction ..................................................... 94
5.2.6 Gurit Corecell M80 tensile test .................................................................................... 96
5.2.7 Gurit Corecell M130 tensile test ................................................................................ 102
5.3 3-point bending.................................................................................................................. 104
5.4 Perimeter shear strength .................................................................................................... 118
5.5 Optimization and recommended layup.............................................................................. 122
6 Conclusions .............................................................................................................................. 126
6.1 Discussion ......................................................................................................................... 127
7 Biblography .............................................................................................................................. 129
vi
Nomenclature
Unless otherwise noted, subscripts denote instances of the variable.
{ } used to denote a column vector
⌊ ⌋ Used to denote a row vector
[ ] used to denote a matrix
[𝑄] constitutive matrix with the components 𝑄𝑖𝑗 , MPa
vii
[𝑄̅ ] constitutive matrix in the global coordinate system with components 𝑄̅𝑖𝑗 , MPa
[𝑇] Transformational rotation matrix for rotating lamina coordinate system 𝜃 degrees into the global
coordinate system, unitless
𝜃 rotation angle from lamina coordinate system into global coordinate system, degrees or radians
𝑢 displacement into the x-direction, mm
𝑣 displacement into the y-direction, mm
𝑤 displacement into the z-direction, mm
𝑢0 , 𝑣0 , 𝑤0 displacements of laminate midplane, mm
𝜙 rotation of cross-section under small angles assumption, equal to the derivative of the deflection
due to bending. Unitless
(0)
{𝜀 (0) } Vector of midplane strain with components 𝜀𝑖
(1)
{𝜀 (1) } Vector of bending curvature with components 𝜀𝑖
{𝑁} Vector of in-plane forces per unit width at the laminate midplane, N/mm
{𝑀} Vector of bending and twisting moments per unit width, Nmm/mm
[𝐴] Lamina midplane elongation stiffness matrix, MPa ⋅ mm
[𝐵] Lamina bending-elongation coupling stiffness matrix, MPa ⋅ mm2
[𝐷] Lamina bending stiffness matrix, MPa ⋅ mm3
[𝐴̅] Lamina midplane elongation stiffness matrix for inverse constitutive relations, (MPa ⋅ mm)-1
[𝐵̅ ] Lamina bending-elongation coupling stiffness matrix for inverse constitutive relations, (MPa ⋅
mm2)-1
̅ ] Lamina bending stiffness matrix for inverse constitutive relations, (MPa ⋅ mm3)-1
[𝐷
𝒬 Through-thickness shear force per unit width, N/mm
[𝒟] Bending stiffness matrix for homogeneous, isotropic plate, MPa ⋅ mm3
𝐷 Flexural stiffness of plate
𝑀 bending moment, Nm
𝑉 Through-thickness shear force, N
Φ Shear stiffness component in Timoshenko beam element, unitless
𝜅 Shear stiffness correction factor
𝐶𝑖 integration constant
viii
𝑙𝑒 Element length, mm
[𝑘] Element stiffness matrix with coefficients 𝑘𝑖𝑗
𝑒 Green-lagrange strains
The Voight-Kelvin notation will be used. 𝜎1 = 𝜎11 , 𝜎2 = 𝜎22 , 𝜎3 = 𝜎33 , 𝜎4 = 𝜎23 , 𝜎5 = 𝜎13 , 𝜎6 =
𝜎12
𝜀1 = 𝜀11 , 𝜀2 = 𝜀22 , 𝜀3 = 𝜀33 , 𝜀4 = 2𝜀23 , 𝜀5 = 2𝜀13 , 𝜀6 = 2𝜀12
At times, the 1-, 2-, and 3-directions are referred to as the x-, y- and z- directions.
ix
List of figures
Figure 1: Different parts of a spaceframe chassis, Viing X taken as example. ................................... 3
Figure 2: Monocoque design based off of the Viking X spaceframe chassis ...................................... 5
Figure 3: Beam with offset coordinate system ................................................................................... 15
Figure 4: Anticlastic curvature seen underneath bent foam plate ...................................................... 16
Figure 5: Anticlastic curvature seen at edges of bent foam plate ...................................................... 17
Figure 6: Timoshenko beam cross-section rotation and centerline slope .......................................... 19
Figure 7: Boundary conditions for timoshenko beam with arbitrary axial displacement at one end. 21
Figure 8: Beam element with an arbitrary transverse displacement at one end ................................. 22
Figure 9: Beam element with an arbitrary rotation of the cross-sectionat one end ........................... 22
Figure 10: Sketch of Timoshenko beam in 3 point bending .............................................................. 26
Figure 11: Through-thickness shear stress of an asymmetric beam .................................................. 31
Figure 12: Comparison of shear stiffness of sandwich and timoshenko theories .............................. 32
Figure 13: Comparison of timoshenko and shimpi beam elements ................................................... 35
Figure 14: Patch test of plane elements ............................................................................................. 41
Figure 15: Plane element beam in 3 point bending ............................................................................ 42
Figure 16: Comparison of plane elements to analytical beam solutions. Stiffness in Newton per
millimeter of vertical deflection of the loaded central node .............................................................. 43
Figure 17: Zoomed-in view of Comparison of plane elements to analytical beam solutions ............ 43
Figure 18: Convergence of refinement of plane element beam ......................................................... 43
Figure 19: Stiffness of plate elements compared to theoretical values. [10] ..................................... 45
Figure 20: Force convergence plot in ANSYS .................................................................................. 49
Figure 21: Schematic view of a bilinear model ................................................................................. 51
Figure 22: Possible bilinear models to fit a certain dataset................................................................ 52
Figure 23: Multilinear hardening model ............................................................................................ 52
Figure 24: Idealized bilinear graph of isotropic and kinematic hardening. ....................................... 53
Figure 25: Calculated stiffness versus penalty stiffness .................................................................... 56
Figure 26: Calcualted stiffness versus penalty stiffness, zoomed in.................................................. 56
Figure 27: Calculated runtime versus element count for different boundary condition methods ...... 57
Figure 28: Runtimes normalized against penalty method runtime versus element count for different
boundary condition methods .............................................................................................................. 57
Figure 29: Runtime versus element count for solving method .......................................................... 58
Figure 30: Ratio of runtime between inversion and mldivide(K,F) versus element count ................ 58
Figure 31: Runtimes versus element count for full matrix and sparse matrix ................................... 59
Figure 32: Comparison of runtime versus element count of full matrix and sparse matrix .............. 59
Figure 33: Comparison of foam core materials from Gurit. Specific shear stiffness has the units
GPa/(kg/m3) while density has the units kg/m3. ................................................................................ 62
Figure 34: Glass plate with tacky-tape and one layer of prepreg ....................................................... 64
Figure 35: Schematic view of application of tacy-tape at a corner .................................................... 64
Figure 36: Laminate within a vacuum bag on a glass plate ............................................................... 65
Figure 37: Drawing of the 3-point bending test fixture ..................................................................... 67
Figure 38: 3-point bending test setup seen from an angle ................................................................. 67
Figure 39: 3-point bending test setup seen from the end ................................................................... 67
Figure 40: Baseline tubes in 3-point bending .................................................................................... 69
x
Figure 41: Local deformation of baseline tubes................................................................................. 69
Figure 42: Baseline tubes 3PB load-displacement graph................................................................... 70
Figure 43: Baseline tubes 3PB stiffness graph................................................................................... 70
Figure 44: Load-displacement graph of 3PB of 10 mm steel plate ................................................... 70
Figure 45: Stiffness graph of 3PB of 10 mm steel plate .................................................................... 70
Figure 46: Singularity in laminate stiffness for SES.......................................................................... 71
Figure 47: Shear stress in core of beam with a 2D plane element core. The two green elements in
the yellow part are a plotting error caused by all of the nodes having the same calculated shear
stress. .................................................................................................................................................. 73
Figure 48: Von Mises equivalent stresses in the core of a beam with 2D plane elements for core ... 74
Figure 49: Boundary conditions of a shell element model (full plate shown) ................................... 75
Figure 50: Full model boundary conditions. Quarter plate shown. Symmetry boundary conditions
are on the faces and edges to simulate a full plate. ............................................................................ 75
Figure 51: Perimeter shear stress test ................................................................................................. 76
Figure 52: Perimeter shear stress test peak detection ........................................................................ 77
Figure 53: ANSYS model of PSS test with solid-shell top skins ...................................................... 78
Figure 54: ASTM D695 compressive test fixture .............................................................................. 79
Figure 55: Cut surface showing the cross-section of a skin consisting of 4 layers of CFRP in
alternating 0 and 90-degree directions ............................................................................................... 82
Figure 56: Contaminants catching the light when seen a bit from the side ....................................... 83
Figure 57: Contaminants in circles. Imprints of wrinkles of the release film seen as horizontal lies
that catches the light ........................................................................................................................... 83
Figure 58: Voids seen when looking parallel to the fibers. The rulers are 2mm and 0,5mm long. ... 84
Figure 59: Voids photographed with microscope built-in camera, looking onto the ends of the
fibers. Voids marked from the left with three 20 micrometers voids and two 10 micrometer voids. 84
Figure 60: Voids looking onto the ends of the fibers, photographed with an external camera through
the eye-piece. Note that the uneven top surface is filled with fibers and are not just pooled resin. . 84
Figure 61: Large void seen parallel to the fibers direction. ............................................................... 85
Figure 62: Large void seen when looking onto the 90-degree direction. .......................................... 85
Figure 63: Large-scale surface unevenness visible on a 3PB specimen ............................................ 86
Figure 64: Void volume fraction plotted against fiber volume fraction ............................................ 89
Figure 65: Nail polish weight plotted against fiber volume fraction ................................................. 89
Figure 66: Specimen density plotted against fiber volume fraction .................................................. 89
Figure 67: Stress-strain diagram of SE84LV-HEC, longitudinal direction ....................................... 91
Figure 68: Stiffness plot of SE84LV-HEC, longitudinal direction .................................................... 92
Figure 69: End-failure of SE84LV-HEC compressive specimen ...................................................... 94
Figure 70: Stress-strain diagram of SE84LV-HEC in compression, transverse direction ................. 95
Figure 71: Stiffness diagram of SE84LV-HEC in compression, transverse direction....................... 95
Figure 72: Fracture of compression specimen under a microscope ................................................... 96
Figure 73: Engineering stress-strain diagram for Corecell M80 tensile test...................................... 97
Figure 74: Stiffness diagram for Corecell M80 tensile test, engineering stress/strain ...................... 97
Figure 75: Zoomed in view of the stiffness diagram of M80 from engineering stress/strain ............ 97
Figure 76: Zoomed in view of the stiffness diagram of M80 from true stress/strain ........................ 97
Figure 77: True stress versus logarithmic strain, M80 tensile test..................................................... 99
Figure 78: True stress versus logarithmic plastic strain, M80 tensile test ......................................... 99
xi
Figure 79: Fit of Ludwik's model onto specimen 2 of M80 tensile test ............................................ 99
Figure 80: Comparison of exprimental and simulated data of 3PB of M80 foam using bilinear
isotropic hardening with experimentally derived material data, Load-Displacement graph ........... 100
Figure 81: Comparison of exprimental and simulated data of 3PB of M80 foam using bilinear
isotropic hardening with experimentally derived material data, Stiffness graph ............................. 100
Figure 82: Load-displacement graph of 3PB of M80 foam. Comparison of experimental data and
simulated with multilinear plasticity model ..................................................................................... 101
Figure 83: True stress-strain diagram of M130 tensile tests ............................................................ 102
Figure 84: Stiffness diagram of M130 tensile test, from true stress and logarithmic strain ............ 102
Figure 85: Load-Displacement graph of 3PB of M130 foam. Comparison of experimental results to
simulated with multilinear plasticity. ............................................................................................... 104
Figure 86: Compression of core seen at the middle of a 3PB test specimen ................................... 104
Figure 87: Load-displacement graph of specimen 1 ........................................................................ 105
Figure 88: Stiffness graph of specimen 1......................................................................................... 105
Figure 89: 3PB load-displacement curve, 4 layer skins, 25mm M80 core ...................................... 106
Figure 90: 3PB stiffness graph, 4 layer skins, 25mm M80 core ...................................................... 106
Figure 91:3PB load-displacement curve, 4 layer skins, 25mm M130 core ..................................... 107
Figure 92:3PB stiffness graph, 4 layer skins, 25mm M130 core ..................................................... 107
Figure 93:3PB load-displacement curve, 4 layer skins, 10 mm M80 core ...................................... 107
Figure 94:3PB stiffness graph, 4 layer skins, 10 mm M80 core ...................................................... 107
Figure 95: 3PB load-displacement graph, 8 layer skins, 25mm M80 core ...................................... 108
Figure 96: 3PB stiffness graph, 8 layer skins, 25mm M80 core ...................................................... 108
Figure 97: 3PB Load-displacement graph, 14 layer skins, 25mm M80 core................................... 108
Figure 98: 3PB stiffness graph, 14 layer skins, 25mm M80 core .................................................... 108
Figure 99: Load-Displacement curve comparison between simualted and tested data, 3PB, specimen
1 ........................................................................................................................................................ 109
Figure 100: Stiffness graph comparison between simulated and tested data, 3PB, specimen 1 ...... 109
Figure 101: Legend for comparison graphs. .................................................................................... 109
Figure 102: Laminate stiffness versus skin thickness ...................................................................... 110
Figure 103: Laminate stiffness versus core thickness ...................................................................... 110
Figure 104: Laminate stiffness versus core density. The densities here are considered the
identification of the foams in the Gurit Corecell M-series .............................................................. 110
Figure 105: Energy absorption versus skin thickness ...................................................................... 110
Figure 106: Energy absorption versus core thickness ...................................................................... 110
Figure 107: Energy absorption versus core density ......................................................................... 110
Figure 108: Yield load versus skin thickness................................................................................... 111
Figure 109: Yield load versus core thickness .................................................................................. 111
Figure 110: Yield load versus core density ...................................................................................... 111
Figure 111: Maximum load versus skin thickness ........................................................................... 111
Figure 112: Maximum load versus core thickness ........................................................................... 111
Figure 113: Maximum load versus core density .............................................................................. 111
Figure 114: Typical failure at first peak of PSS test ........................................................................ 118
Figure 115: Cross-section of PSS test of only M80 foam................................................................ 119
Figure 116: Load-displacement graph of PSS test, 4 layer skin, 25mm M80 core ......................... 119
Figure 117: Load-displacement graph of PSS test, 4 layer skin, 25mm M130 core ....................... 119
xii
Figure 118: Load-displacement graph of PSS test, 4 layer skin, 10 mm M80 core ........................ 120
Figure 119: Load-displacement graph of PSS test, 8 layer skin, 25mm M80 core ......................... 120
Figure 120: Load-displacement graph of PSS test, 14 layer skin, 25mm M80 core ....................... 120
Figure 121: Stiffness of laminates by beam model for optimization purposes ................................ 123
Figure 122: Specific stiffness of laminates by beam model for optimization purposes .................. 123
Figure 123: Stiffness of laminates by 2D beam model for optimization purposes .......................... 124
Figure 124: Specific stiffness of laminates by 2D beam model for optimization purposes ............ 124
Figure 125: Simulation of optimum layup in ANSYS. Elastic shear strain through the core is plotted
.......................................................................................................................................................... 125
Figure 126: Load-displacement graph of optimum layup, simulated by ANSYS full model ......... 125
Figure 127: Stiffness graph of optimum layup, simulated by ANSYS full model .......................... 125
xiii
List of tables
Table 1 Minimum dimensional requirements for baseline chassis. ..................................................... 4
Table 2: Stiffening effect for beams of mediocre width. [7] ............................................................. 15
Table 3: Shape functions for timoshenko beam element ................................................................... 27
Table 4: Strain-displacement functions for laminated Timoshenko beam element ........................... 28
Table 5: Shape functions for quadrilateral elements. [10] ................................................................. 37
Table 6: Derivatives of shape functions from Table 5 ....................................................................... 38
Table 7: Gauss Quadrature weight factors. [5] .................................................................................. 39
Table 8: Material properties of SE84LV-HEC from data sheet......................................................... 86
Table 9: Material properties for Gurit Corecell M80 and M130 from data sheet .............................. 87
Table 10: Material properties of Gurit SE84LV-HEC derived from tests ......................................... 87
Table 11: Material properties of Gurit Corecell M80 and M130 derived from tests ......................... 87
Table 12: Fiber and void volume fraction of SE84LV-HEC ............................................................. 88
Table 13: Hypothesis testing of fiber and void volume fractions ...................................................... 90
Table 14: Stiffness and strength of SE84LV-HEC, longitudinal direction........................................ 93
Table 15: Stiffness and strength of SE84LV-HEC, transverse direction ........................................... 93
Table 16: Stiffness and strength of SE84LV-HEC in compression, transverse direction ................. 96
Table 17: Stiffnesses and strengths of Gurit Corecell M80 foam ...................................................... 98
Table 18: Fitting coefficients for Ludwik's model and bilinear tangent modulus for M80 tensile test
.......................................................................................................................................................... 100
Table 19: Stress levels for multilinear model of M80 foam ............................................................ 101
Table 20: Stiffness and strength of M130 tensile test ...................................................................... 103
Table 21: Coefficients of Ludwik's model and the calculated tangent modulus for M130 foam
tensile tests ....................................................................................................................................... 103
Table 22: Stress and strain values for a multilinear hardening model, M130 tensile tests .............. 103
Table 23: Summary of experimental data, 3-point bending tests .................................................... 112
Table 24: Simulation error, stiffness ................................................................................................ 113
Table 25: Simulation error, energy absorption ................................................................................ 115
Table 26: Simulation error, Yield load ............................................................................................ 115
Table 27: Simulation error, maximum load ..................................................................................... 116
Table 28: Stiffening effect of laminated beam width ...................................................................... 117
Table 29: mean forces for first and second peak of the PSS tests along with % spread .................. 121
xiv
1 Introduction
University students at bachelor and master levels from around the world compete in the engineering
competition FSAE/Formula Student, in which they are required to design and manufacture a single-
seater race car according to a common ruleset. The ruleset regarding chassis structure is built
around the concept of a baseline steel spaceframe but allows for alternative constructions made
from metals or composites if the safety of the alternative structure can be proven to be equivalent to
the baseline.
The team at the University of Southern Denmark, SDU-Vikings, have existed since 2006 and have
built ten cars which have all had a traditional spaceframe but now the team wishes to look into the
possible weight savings of using composites. A monocoque is a chassis structure in which the skin
is the load-bearing structure instead of distinct frame members and such it allows not only for a
reduction of the weight of the load-bearing structure but also for a reduction in the needed
bodywork, which otherwise forms a shell around the frame of a spaceframe chassis. Monocoques
are used both in aerospace, high-level racecars, boats and also roadgoing cars because it poses a
more efficient use of the material and such it has also been used by many Formula Student teams.
Monocoques are often formed as a laminate with a stiff skin on either side of a lightweight and
relatively flexible core in order to raise the bending stiffness and more efficiently use the materials.
The flexibility of the core means that shear deformation cannot be neglected when assessing the
bending stiffness of the plate. Shear deformation of such sandwich plates are not currently taught to
the bachelor-level students whom the team mostly consist of and therefore, this thesis seeks to
present an accurate way of simulating this bending in a way that can be used by the team.
Page 1 af 125
2 Problem statement
It is sought to build a tool which can calculate the stiffness, strength and energy absorption of
laminated composite plates with a light-weight core under 3-point bending. The tool may consist of
simple FEM code in combination with commercial software to gain the capability of both analyzing
large amounts of layups and also get the necessary accuracy to recommend optimum layups to be
manufactured for testing. The tool should be sufficiently simple that it is usable by bachelor-level
engineering students as the team SDU-Vikings primarily consists of these. To verify the accuracy of
the calculations, physical tests of laminates must be conducted.
A method to determine the load required to penetrate the composite is also sought to enable
simulation of the perimeter shear stress test as well.
Objectives:
To develop a finite element model using beam elements, capable of approximately calculating the
stiffness, strength and energy absorption of laminated composite plates with a light-weight core.
To set up a finite element model using commercial software, capable of accurately calculating the
stiffness, strength and energy absorption of laminated composite plates with a light-weight core.
To validate the finite element models by physical tests in which multiple laminate parameters are
assessed. The parameters varied are the amount of carbon fiber layers, the thickness of the core and
the core stiffness.
To use the developed model as a preliminary design tool to search for the lightest rules-compliant
laminates that can be used to construct a composite monocoque.
Page 2 af 125
3 Theory
Page 3 af 125
Item or application Minimum Minimum Minimum area
wall cross- moment of inertia
thickness sectional area
Main hoop (1) 2,0 [𝑚𝑚] 175 [𝑚𝑚2 ] 11320 [𝑚𝑚4 ]
Front hoop (2)
Shoulder harness mounting bar (8)
Side impact structure (6) 1,2 [𝑚𝑚] 119 [𝑚𝑚2 ] 8509 [𝑚𝑚4 ]
Front bulkhead (4)
Roll hoop bracings (9) (3)
Driver’s harness attachment (except shoulder
harness bar)
Accumulator container protection structure (7)
Tractive system protection structure (7)
Front bulkhead support (5) 1,2 [𝑚𝑚] 91 [𝑚𝑚2 ] 6695 [𝑚𝑚4 ]
Main hoop bracing support (7)
Table 1 Minimum dimensional requirements for baseline chassis.
The main roll hoop and its bracing must be made from steel tubing and the front hoop must be made
from metallic tube but the rest of the chassis is free in its design and materials as long as
equivalency is proven using the structural equivalency spreadsheet. If the spaceframe chassis of the
Viking X is made into a monocoque by removing the chassis tubes, making planes between the
chassis nodes and assigning a laminate thickness to those planes, the monocoque might look
something like Figure 2.
Page 4 af 125
Figure 2: Monocoque design based off of the Viking X spaceframe chassis
For all laminated parts of the monocoque, the cross-sectional area of the skins must exceed the
minimum cross-sectional area stated in Table 1 for the tubes that the laminated part replaces
compared to a baseline chassis. Also, for all laminated parts, the bending stiffness must exceed that
of the tubes that it replaces. In this regard, the bending stiffness is calculated as the second moment
of area, 𝐼𝑐 , of the composite skins multiplied by an equivalent skin stiffness 𝐸𝑒𝑞 which is calculated
from 3 point bending of a representative test specimen. The SES calculates the equivalent skin
stiffness as in equation (3-1) where 𝐾𝑙𝑎𝑚 is the corrected stiffness of the test specimen, 𝑙 is the test
span and 𝐼𝑙𝑎𝑚 is the second moment of area of the tested specimen. The second moment of area are
calculated per equation (3-2) in which, 𝑏 is the specimen width and 𝑡 is the thickness of each part of
the laminate.
−3
𝐾𝑙𝑎𝑚 ⋅ 𝑙 3 (3-1)
𝐸𝑒𝑞 = 10 ⋅
48 ⋅ 𝐼𝑙𝑎𝑚
1 (3-2)
𝐼𝑙𝑎𝑚 = 𝑏 ⋅ ((𝑡𝑐𝑜𝑟𝑒 + 𝑡𝑖𝑛𝑛𝑒𝑟𝑠𝑘𝑖𝑛 + 𝑡𝑜𝑢𝑡𝑒𝑟𝑠𝑘𝑖𝑛 )3 − 𝑡𝑐𝑜𝑟𝑒
3 )⋅
12
The second moment of area of the skins of the actual laminate, 𝐼𝑐 , is calculated with the parallel
axis theorem under the assumption that the laminate is a flat panel and with these values, the SES
calculates the laminate as a beam consisting only of the skins and without shearing deformation. For
all sections of the chassis but the side impact structure, the calculated values 𝐸𝑐 𝐼𝑐 and the strength
Page 5 af 125
results from the 3-point bending test are used to calculate theoretical values of tensile strength, max
load at midspan of a 1m beam, max deflection of a 1m beam and the energy absorption of elastic
deformation of a beam until the ultimate tensile strength.
The Side Impact Structure, abbreviated SIS, is the sides and floor of the cockpit between the two
roll hoops and is subject to the strictest requirements as the entire driver, except the legs, is
protected by this section of chassis. The side of the SIS must be equivalent to two chassis tubes
while half the floor must be equivalent to a third tube. This equivalency is only calculated
theoretically for the bending stiffness and the cross-sectional area, which are both calculated under
a flat panel assumption regardless of the curvature of the structure.
The remaining equivalency requirements are taken directly from the tested specimen. The highest
tested force and the absorbed energy of the tested specimen for the side must exceed the values
from the test of the two baseline tubes it replaces regardless of the dimensions of the actual
structure. Additionally, the laminate used for the SIS requires that the highest peak of a perimeter
shear stress test to be at least 7,5 kN.
When proving equivalency of the main hoop bracing support (MHBS), the front hoop bracing
(FHB) and the front bulkhead support structure (FBHS), the actual shape of the car can be taken
into account to show equivalency if the flat-panel calculations show between 50% and 100%
equivalency. The way the actual shape of the car is taken into account for the MHBS and the BFHS
is to compare the second moment of area with respect to the car centerline against a corresponding
value for a baseline chassis. The second moment of area of the laminate is found by making a cross-
section in the CAD drawing and calculating the combined second moment of area of the skins of
one half of the car. The corresponding value for a baseline chassis is such that with 𝑑, 𝑑𝑖 and 𝑑𝑜
being respectively half the chassis width and the inner and outer diameter of the 𝑛𝑡 baseline tubes
then the second moment of area, 𝐼, of the baseline chassis is calculated by the parallel axis theorem
in equation (3-3).
𝜋 4 𝜋 (3-3)
𝐼 = 𝑛𝑡 ⋅ (𝑑𝑜 − 𝑑𝑖4 ) + 𝑛𝑡 ⋅ (𝑑𝑜2 − 𝑑𝑖2 ) ⋅ 𝑑 2
64 4
For accounting for the actual form of the front hoop bracings, the second moment of area of the
baseline tubes and of the composite structure are used without referring them to the car centerline
using the parallel axis theorem.
The front bulkhead support structure must be constructed of laminate which has been tested to at
least 4 kN in the perimeter shear strength test.
If bolted joints are used in the assembly of the chassis or attachment of the harness to it, rule T 3.16
requires a load carrying capability of 30 kN, which is to be ensured by extrapolating the area from
the first peak of the perimeter shear strength tests.
Other significant rules include a requirement that at most 50% of the fiber weight are within +/- 10
degrees and that it is permitted to use one laminate for testing and another for the construction of
the monocoque only if nothing but the thickness of the core differs between the laminates. The
latter rule can be considered a loophole as the calculated 𝐸𝑐 will be much higher with a thin core
Page 6 af 125
than with a thick core due to the assumption of it being a beam with no shear. This loophole will not
be exploited in this paper.
The following in this subsection is added in V2.0 of this report, the 11th of august 2020
The loophole regarding the impact of core thickness on the calculated 𝐸𝑐 has been closed in the
2020 SES version 1.1.2.
A guidance note has been added which specifies that the rule about testing with one core thickness
and using another is only applicable if the tested laminate from which 𝐸𝑐 is derived has a thicker
core than that which is actually being used.
𝐸𝑓 𝐸𝑚 (3-5)
𝐸2 = 𝐸3 =
𝐸𝑓 𝓋𝑚 + 𝐸𝑚 𝓋𝑓
𝐺𝑓 𝐺𝑚 (3-6)
𝐺6 = 𝐺5 =
𝐺𝑓 𝓋𝑚 + 𝐺𝑚 𝓋𝑓
𝐸2 (3-7)
𝐺4 =
2(1 + 𝜈23 )
𝜈12 = 𝜈13 = 𝜈𝑓 𝓋𝑓 + 𝜈𝑚 𝓋𝑚 (3-8)
𝐸2 (3-9)
𝜈21 = 𝜈31 = 𝜈12
𝐸1
1 − 𝜈21 (3-10)
𝜈23 = 𝜈32 = 𝜈12
1 − 𝜈12
Page 7 af 125
Like the stiffness is orthotropic, so is the strength as this also depends on the orientation. Because of
the nonhomogeneous nature of the composite, the loads to failure vary greatly with the direction of
loading such that the tensile strength is often much higher than the compressive strength because
the failure modes are different. With tension along the fiber direction, the failure is often caused by
fiber fracture as the fibers have a much lower elongation-to-failure than the matrix but in
compression along this direction, the failure is often caused by buckling of the fibers within the
matrix. In the transverse direction, the strength is limited by the strength of the matrix but is further
reduced from this due to voids and that the fibers act as solid inclusions, both of which are creating
stress concentrations within the matrix.
The in-plane shear failure is caused by a failure of the matrix or the matrix-fiber interface and while
each lamina expresses transverse isotropy, the transverse shear strength is often limited by the
interlaminar shear strength. The interlaminar shear strength is the shear strength of the interface
between layers where there may be a lower fiber volume fraction due to fibers crossing at an angle
or there may be a higher concentration of voids as discussed in the defects chapter of this report.
Failure prediction is done either by failure theories, by means of fracture mechanical analysis or by
a combination. To limit the scope of this project, fracture mechanical analysis is omitted. Unlike
ductile homogeneous materials where Von Mises yield criterion is considered almost universally
applicable, there is no single widely accepted failure criterion. Instead, many competing theories of
varying complexity are assessed. Most common failure theories are maximum strain theory, max
stress theory, Azzi-Tsai-Hill failure criterion, Tsai-Wu failure criterion and Hashin failure criterion.
The maximum stress theory does not account for combination of stresses but state that failure will
occur if the stresses in any of the principal material directions exceed the strength in that direction.
Maximum strain theory is similar but predicts failure if the strain exceeds the failure strain in any of
those directions.
The Azzi-Tsai-Hill theory accounts for combination of stresses by predicting failure of an
orthotropic lamina in plane stress condition if equation (3-11) is fulfilled. 𝑆1 and 𝑆2 are taken as
either the tensile or compressive strength depending on the sign of the corresponding stress. [2]
𝜎12 𝜎1 𝜎2 𝜎22 𝜎62 (3-11)
− 2 + 2+ 2=1
𝑆12 𝑆1 𝑆2 𝑆6
The Tsai-Wu failure criterion states that under plane stress, the orthotropic lamina will fail if
equation (3-12) is true. [2]
2
𝑓1 𝜎1 + 𝑓2 𝜎2 + 𝑓6 𝜎6 + 𝑓11 𝜎11 + 𝑓22 𝜎22 + 𝑓66 𝜎62 + 2𝑓12 𝜎1 𝜎2 = 1 (3-12)
Where the factors except 𝑓12 are given in equation (3-13). The factor 𝑓12 must be determined by a
biaxial test which are rarely done and thus the value is often approximated as the lower limit of
equation (3-14). The Tsai-Wu criterion is criticized for not having the value of 𝑓12 uniquely defined
from biaxial tests as it may be found to be different depending on whether the biaxial test is done
with compressive or tensile stresses and also, the criterion is criticized for having the prediction of a
tensile failure depend on the compressive strength. [3]
Page 8 af 125
1 1 1 1
𝑓1 = − , 𝑓2 = − ,𝑓 = 0
𝑆1𝑡 𝑆1𝑐 𝑆2𝑡 𝑆2𝑐 6 (3-13)
1 1 1
𝑓11 = , 𝑓22 = , 𝑓66 = 2
𝑆1𝑡 𝑆1𝑐 𝑆2𝑡 𝑆2𝑐 𝑆6
1 1 (3-14)
− (𝑓11 𝑓22 )2 ≤ 𝑓12 ≤ 0
2
The Hashin failure criterion is a proposed solution to the criticism of the Tsai-Wu theory and does
not claim that the failure can be predicted by a single smooth curve. The Hashin failure criterion is a
set of four equations, each accounting for a failure mode such that the failure of fibers and matrix in
both tension and compression are considered separately. In plane stress, equations (3-15) to (3-18)
describe the failure criterion.
𝜎1 2 𝜎6 2 (3-15)
( ) + ( ) = 1 𝑓𝑜𝑟 𝜎1 > 0
𝑆1𝑡 𝑆6
𝜎1 2 (3-16)
( ) = 1 𝑓𝑜𝑟 𝜎1 < 0
𝑆1𝑐
𝜎2 2 𝜎6 2 (3-17)
( ) + ( ) = 1 𝑓𝑜𝑟 𝜎2 > 0
𝑆2𝑡 𝑆6
𝜎2 2 𝑆2𝑐 2 𝜎2 𝜎6 2 (3-18)
( ) + (( ) − 1) + ( ) = 1 𝑓𝑜𝑟 𝜎2 < 0
2𝑆4 𝑆4 𝑆2𝑐 𝑆6
These failure criteria also exist for 3-dimensional stress states but due to the availability of material
properties from data sheets and simple tests, these are not included in the present paper.
Page 9 af 125
where 1 − 𝜈23 𝜈32 𝜈21 + 𝜈31 𝜈23 𝜈12 + 𝜈32 𝜈13 𝜈31 + 𝜈21 𝜈32 (3-20)
𝑄11 = , 𝑄12 = = , 𝑄13 =
𝐸2 𝐸3 ∆ 𝐸2 𝐸3 ∆ 𝐸1 𝐸3 ∆ 𝐸2 𝐸3 ∆
𝜈13 + 𝜈12 𝜈23
=
𝐸1 𝐸2 ∆
1 − 𝜈13 𝜈31 𝜈32 + 𝜈12 𝜈31 𝜈23 + 𝜈21 𝜈3 1 − 𝜈12 𝜈21
𝑄22 = , 𝑄23 = = , 𝑄33 =
𝐸1 𝐸3 ∆ 𝐸1 𝐸3 ∆ 𝐸1 𝐸2 ∆ 𝐸2 𝐸2 ∆
𝑄44 = 𝐺4 , 𝑄55 = 𝐺5 , 𝑄66 = 𝐺6
1 − 𝜈12 𝜈21 − 𝜈23 𝜈32 − 𝜈31 𝜈13 − 2𝜈21 𝜈32 𝜈13
∆=
𝐸1 𝐸2 𝐸3
As the lamina may be rotated some angle 𝜃 in the plane of the lamina, the stiffness matrix for the
generally orthotropic is made by the use of the specially orthotropic stiffness matrix [𝑄] and a
transformation matrix [𝑇]
[𝑄̅ ] = [𝑇][𝑄][𝑇]𝑇 (3-21)
When normal stress in the 3-direction is neglected, which it often is for plates and beams, the
constitutive relation reduces to equation (3-23).
𝜎1 𝑄11 𝑄12 0 0 0 𝜀1 (3-23)
𝜎2 𝑄21 𝑄22 0 0 0 𝜀2
𝜎4 = 0 0 𝑄44 0 0 𝜀4
𝜎5 0 0 0 𝑄55 0 𝜀5
𝜎
{ 6} [ 0 0 0 0 𝑄66 ] {𝜀6 }
Where
Page 10 af 125
𝐸11 𝐸22 (3-24)
𝑄11 = , 𝑄22 =
1 − 𝜈12 𝜈21 1 − 𝜈21 𝜈12
𝜈12 𝐸22 𝜈21 𝐸11
𝑄12 = 𝑄21 = =
1 − 𝜈21 𝜈12 1 − 𝜈12 𝜈21
𝑄66 = 𝐺12 , 𝑄44 = 𝐺13 , 𝑄55 = 𝐺23
The transformation matrix reduces in a similar manner as the rows and columns related to the 3-
direction are removed.
When the transformed lamina constitutive matrix is formed, the formation of a laminate constitutive
matrix requires displacement assumptions. The transverse displacement of the plate is commonly
assumed to not vary through the thickness and such, the transverse displacement of any point within
the laminate is equal to that of the midplane. The in-plane displacements are assumed differently
depending on the used plate theory but generally, they exist as the midplane displacements and one
or more terms which vary through the thickness. For the classical plate theory, also called Kirchoff-
Love plate theory, the variation through the thickness is linearly related to the first derivative of the
transverse deflection such that it is given by equation (3-25) while for the first-order shear
deformation theory (FSDT), the variation through the thickness are linearly related to the rotation of
the cross-sections about the x- and y-axes and the deformation field is given by (3-26). [5]
𝜕𝑤0 (3-25)
𝑢(𝑥, 𝑦, 𝑧) = 𝑢0 (𝑥, 𝑦) − 𝑧
𝜕𝑥
𝜕𝑤0
𝑣(𝑥, 𝑦, 𝑧) = 𝑣0 (𝑥, 𝑦) − 𝑧
𝜕𝑦
𝑤(𝑥, 𝑦, 𝑧) = 𝑤0 (𝑥, 𝑦)
𝑤(𝑥, 𝑦, 𝑧) = 𝑤0 (𝑥, 𝑦)
Kirchoff-Love plates are analogous to Bernoulli-Euler beams in that shear deformation is assumed
negligible while first-order shear deformation theory is analogous to Timoshenko beams in that
shear deformation is included by a shear strain that is constant through the thickness and shear
correction coefficients are used to correct the shear stiffness. The Kirchoff-love plate theory exist as
a special case of the first-order shear deformation theory as the shear deformation becomes
𝜕𝑤0 𝜕𝑤0
negligible for small thicknesses, leading to 𝜙𝑥 (𝑥, 𝑦) = − and 𝜙𝑦 (𝑥, 𝑦) = − . In this project,
𝜕𝑥 𝜕𝑦
shear deformation is not negligible due to the thick core, so the following equations refer to the first
order shear deformation theory.
The strains are defined in [6] by derivatives of the displacement field and thus, the strains are
Page 11 af 125
𝑑𝑢 𝑑𝑢0 𝑑𝜙𝑥 (3-27)
𝜀1 = = +𝑧
𝑑𝑥 𝑑𝑥 𝑑𝑥
𝑑𝑣 𝑑𝑣0 𝑑𝜙𝑦
𝜀2 = = +𝑧
𝑑𝑦 𝑑𝑦 𝑑𝑦
𝑑𝑤
𝜀3 = =0
𝑑𝑧
𝑑𝑣 𝑑𝑤 𝑑𝑤0
𝜀4 = + = 𝜙𝑦 (𝑥, 𝑦) +
𝑑𝑧 𝑑𝑦 𝑑𝑦
𝑑𝑢 𝑑𝑤 𝑑𝑤0
𝜀5 = + = 𝜙𝑥 (𝑥, 𝑦) +
𝑑𝑧 𝑑𝑥 𝑑𝑥
𝑑𝑢 𝑑𝑣 𝑑𝑢0 𝑑𝑣0 𝑑𝜙𝑥 𝑑𝜙𝑦
𝜀6 = + = + +𝑧 ( + )
𝑑𝑦 𝑑𝑥 𝑑𝑦 𝑑𝑥 𝑑𝑦 𝑑𝑥
The strains 𝜀1 , 𝜀2 and 𝜀6 contain both constant terms and terms which vary linearly through the
thickness so the strains can be expressed as equation (3-28) or in short (3-29).
𝑑𝑢0 𝑑𝜙𝑥 (3-28)
𝜀1
(0)
𝜀1
(1) 𝑑𝑥 𝑑𝑥
𝜀1 𝑑𝑣0 𝑑𝜙 𝑦
{𝜀2 } = {𝜀2(0) } + 𝑧 {𝜀2(1) } = +𝑧
𝜀6 𝑑𝑦 𝑑𝑦
(0) (1)
𝜀6 𝜀6 𝑑𝑢0 𝑑𝑣0 𝑑𝜙𝑥 𝑑𝜙𝑦
+ +
{ 𝑑𝑦 𝑑𝑥 } { 𝑑𝑦 𝑑𝑥 }
{𝜀} = {𝜀 (0) } + 𝑧{𝜀 (1) } (3-29)
By integrating the stresses over the cross-section, the in-plane forces per unit width are calculated in
(3-30) and the bending moment per unit width are calculated in (3-31).
ℎ ℎ (3-30)
𝑧= 𝑧=
2 2
{𝑁} = ∫ 𝜎 𝑑𝑧 =∫ [𝑄̅ ]({𝜀 (0)
} + 𝑧{𝜀 (1)
})𝑑𝑧
ℎ 𝑥𝑥 ℎ
𝑧=− 𝑧=−
2 2
ℎ ℎ (3-31)
𝑧= 𝑧=
2 2
{𝑀} = ∫ 𝑧𝜎𝑥𝑥 𝑑𝑧 = ∫ [𝑄̅ ]𝑧({𝜀 (0) } + 𝑧{𝜀 (1) })𝑑𝑧
ℎ ℎ
𝑧=− 𝑧=−
2 2
To write the laminate constitutive equations, the A-, B- and D-matrices are defined by equation
(3-32), (3-33) and (3-34) and with these, the coupling between strains and applied loads are given
by (3-40) where only the 1-, 2- and 6-directions are included. The coupling between shear strains
and shear deformations are also from the A-matrix because the shear strain doesn’t vary through the
thickness but it is given separately in (3-43) because they don’t interact with the other strains and
(3-40) are how the constitutive relations are given for plates in [2] where shear deformations are
neglected.
Page 12 af 125
ℎ 𝑘 (3-32)
𝑧=
2
[𝐴] = ∫ [𝑄̅ ]𝑑𝑧 = ∑[𝑄̅ ]𝑘 ⋅ (𝑧𝑘 − 𝑧𝑘−1 )
ℎ
𝑧=−
2
ℎ 𝑘 (3-33)
𝑧=
2 1
[𝐵] = ∫ [𝑄̅ ]𝑧𝑑𝑧 = ∑[𝑄̅ ]𝑘 ⋅ (𝑧𝑘2 − 𝑧𝑘−1
2
)
𝑧=−
ℎ 2
2
ℎ 𝑘 (3-34)
𝑧=
2 1
[𝐷] = ∫ [𝑄̅ ]𝑧 2 𝑑𝑧 = ∑[𝑄̅ ]𝑘 ⋅ (𝑧𝑘3 − 𝑧𝑘−1
3
)
𝑧=−
ℎ 3
2
𝒬𝑦 𝐴 𝐴45 𝜀4 (3-36)
{ } = [ 44 ]{ }
𝒬𝑥 𝐴45 𝐴55 𝜀5
The strains can also be calculated from the applied loads by inversion of the constitutive matrix
using equations (3-44) and (3-47).
here
[𝐴̅] = [𝐴−1 ] + [𝐴−1 ][𝐵][(𝐷⋆ )−1][𝐵][𝐴−1 ] (3-39)
Within this inversion, the two-dimensional stiffening effect is removed which is most easily shown
by assessing the stiffness of a homogeneous plate.
For a homogeneous, isotropic plate, the bending stiffness matrix is given by equation (3-40).
𝐷 𝐷𝜈 0 (3-40)
𝐷𝜈 𝐷 0
[𝒟] = [ (1 − 𝜈)𝐷]
0 0
2
Page 13 af 125
𝐸𝑡 3 𝐸
With flexural stiffness 𝐷 = 12(1−𝜈2). Here, (1−𝜈2) is the stiffness from the [𝑄]-matrix where the
𝐼
division by (1 − 𝜈 2 ) is a stiffening effect of the 2-dimensional stress state and the rest is simply 𝑏
because plates are normalized to forces/moments per unit width.
When the inverse of this bending stiffness matrix is taken, the first element of the inverted matrix
−1 1 1 12 𝐸𝐼 −1
becomes[𝒟]11 = − 𝐷(𝜈2−1) = 𝐷(1−𝜈2) = 𝐸𝑡 3 = ( 𝑏 ) . As is seen the stiffening effect is removed
−1 )−1
from the first term by the inversion and 𝑏 ⋅ ([𝒟]11 can be used in place of 𝐸𝐼 in beam
calculations.
Page 14 af 125
Figure 3: Beam with offset coordinate system
Value of 𝑏 2 𝑑 2 𝑣𝑏
𝜈 𝑡 𝑑𝑥 2
𝟎, 𝟐𝟓 𝟏 𝟒 𝟏𝟔 𝟓𝟎 𝟐𝟎𝟎 𝟖𝟎𝟎
𝟎, 𝟏 1 1,0003 1,0033 1,0073 1,0085 1,0093 1,0097
𝟎, 𝟐 1,0001 1,0013 1,0135 1,0300 1,0349 1,0383 1,0400
𝟎, 𝟑 1,0002 1,0029 1,0311 1,0710 1,0826 1,0907 1,0948
𝟎, 𝟑𝟑𝟑𝟑 1,0002 1,0036 1,0387 1,0895 1,1042 1,1146 1,1198
𝟎, 𝟒 1,0003 1,0052 1,0569 1,1357 1,1584 1,1744 1,1825
𝟎, 𝟓 1,0005 1,0081 1,0923 1,2351 1,2755 1,3045 1,3189
Table 2: Stiffening effect for beams of mediocre width. [7]
These figures come from an analytical analysis of the anticlastic curvature of beams of varying
widths [8]. The anticlastic curvature of a beam is a curvature in the lateral direction which forms the
Page 15 af 125
beam into the shape of a saddle instead of a cylinder and is caused by the Poisson’s ratio which
makes the beam wider on the compression side of the beam and narrower at the tension side of the
beam. For beams that are wide, this anticlastic curvature is suppressed, thus leading to a stiffening
1
effect which for an infinitely wide beam equals 1−𝜈2 as the width goes to infinity. During the
bending of a 10 mm thick, 275mm wide plate, anticlastic curvature can be seen on Figure 4 and
Figure 5 where the curvature is seen respectively at the edges underneath the specimen and is
visible as an air gap between the load applicator and the specimen along the upper side of the edges
of the specimen.
To approximate the values of the stiffening effect, the foam plate shown above and a steel plate in
𝑏 2 𝑑2 𝑣𝑏
the same dimensions are used to calculate values of to compare with Table 2.
𝑡 𝑑𝑥 2
At a load of 16000 Newton, tested at a span of 400 mm, the steel plate is approximately at the end
𝑑2 𝑣𝑏
of its linearly elastic region and the curvature is = 6,982 ⋅ 10−4 𝑚𝑚−1 , making the value of
𝑑𝑥 2
𝑏 2 𝑑2 𝑣𝑏
equal to 5,28 which makes the stiffening effect 1,0351 if the poissons ratio is taken as 0,3
𝑡 𝑑𝑥 2
and the value of the stiffening effect is interpolated from the values in Table 2.
𝑑2 𝑣𝑏
Similarly, for a foam plate at a load of 50 Newton, the curvature is = 6,146 ⋅ 10−6 𝑚𝑚−1 and
𝑑𝑥 2
𝑏 2 𝑑2 𝑣𝑏
the corresponding value of equal to 0,0465 𝑚𝑚−1 , which makes the stiffening effect
𝑡 𝑑𝑥 2
approximately 1 regardless of the poissons ratio.
It should be noted that these values of the stiffening effect is calculated from the moment at the
center of the beam and thus the stiffening effect of preventing anticlastic curvature reduces linearly
to 1 at the supports so for the overall stiffness of the beam, the stiffening effect is even lower.
Within this project, it is not attempted to reproduce [8] for laminated plates to evaluate the
assumption that 3-point bending of specimen for formula student can be evaluated as beams.
Instead, beam models for laminated beams are produced and compared to experimental results.
Page 16 af 125
Figure 5: Anticlastic curvature seen at edges of bent foam plate
Page 17 af 125
Bernoulli-Euler beams, this theory will overestimate the stiffness as the shearing deformation is
neglected.
𝑑𝑢 𝑑𝑢0 𝑑 2 𝑣0 (3-42)
𝜀1 = = −𝑦
𝑑𝑥 𝑑𝑥 𝑑𝑥 2
𝑑𝑣
𝜀2 = =0
𝑑𝑦
𝑑𝑢 𝑑𝑣 𝑑𝑣0 𝑑𝑣0
𝜀6 = + =− + =0
𝑑𝑦 𝑑𝑥 𝑑𝑥 𝑑𝑥
By rule of thumb, the Bernoulli-Euler beam theory can be used for homogeneous, isotropic beams
which are ten times as long as they are thick as the shear deformation becomes small compared to
the deformation caused by the bending moment. For anisotropic or laminated beams, this rule of
thumb does not apply as the shearing deformation can be of much larger magnitude and especially
so when the beam is laminated as a sandwich with a flexible core. For a homogeneous beam, the
deflection curve can be solved from the differential equation in (3-43), which is found by
integrating the stresses over the cross-section where the coordinate system coincides with the
geometrical center of the cross-section.
𝑑 2 𝑣0 (3-43)
𝐸𝐼 ⋅ = 𝑀(𝑥)
𝑑𝑥 2
For a laminate in which 𝑀(𝑥) = 𝑀𝑥𝑥 ⋅ 𝑏 is the only nonzero moment, equation (3-44) provides an
analog for the bending stiffness, as laminate theory gives this as the inverse of the factor that
couples the bending moment to the curvature.
𝑏 (3-44)
(𝐸𝐼)𝑐 =
̅11
𝐷
The Bernoulli-Euler beam exists as a special case of the Timoshenko beam in which the shear
stiffness is infinite and thus the stiffness matrix for the finite element can be taken from equation
(3-74) by setting Φ = 0. Doing this, the degree of freedom corresponding to the rotation of the
cross-section equals the slope of the beam centerline and the element is said to have 𝐶 2 -continuity
which means that the slope of the centerline is continuous across elements, even if they have
different stiffnesses or if the node has a load on it.
Page 18 af 125
𝑢(𝑥, 𝑦) = 𝑢0 (𝑥) − 𝑦𝜙𝑥 (3-45)
𝑣(𝑥, 𝑦) = 𝑣0 (𝑥)
This implies that cross-sections remain plane but are not restricted to be normal to the centerline
and as shown in equation (3-46)(3-47), the shear strain does not vary through the thickness of the
beam.
𝑑𝑢 𝑑𝑢0 𝑑𝜙𝑥 (3-46)
𝜀1 = = −𝑦
𝑑𝑥 𝑑𝑥 𝑑𝑥
𝑑𝑣
𝜀2 = =0
𝑑𝑦
𝑑𝑢 𝑑𝑣 𝑑𝑣0
𝜀6 = + = − 𝜙𝑥
𝑑𝑦 𝑑𝑥 𝑑𝑥
The rotation of the cross-section is shown in Figure 6, where it can be seen that the slope of the
centerline is greater than the rotation of the cross-section as the difference between these is the
shearing strain. The rotation of the cross-section can be interpreted as the part of the centerline
slope that is caused by bending deformations as shearing deformations will cause a slope of the
centerline without any rotation of the cross-section.
The development of a shear deformable finite element can be done by shape functions but unless
the chosen shape functions describe the deformation exactly, shear stiffening effects can occur
which requires reduced integration to get rid of. [9].
To develop an exact finite element for laminated beams in a way that is likely to be understood by
the students who will be using the element, it is derived using the direct stiffness method similarly
to how the element for homogeneous beams are developed in [4].
Page 19 af 125
When only an axial force and a bending moment is allowed to act upon the beam, the axial strain is
given by equation (3-47) which is then integrated once to yield the displacement field by equation
(3-48).
̅
𝑑𝑢 𝐴11 𝐵̅11 (3-47)
= 𝑁(𝑥) − 𝑀(𝑥)
𝑑𝑥 𝑏 𝑏
𝐴̅11 𝐵̅11 (3-48)
𝑢(𝑥) = ∫ 𝑁(𝑥)𝑑𝑥 − ∫ 𝑀(𝑥)𝑑𝑥
𝑏 𝑏
Where 𝑁(𝑥) = 𝑏𝑁𝑥𝑥 is an axial tensile load and 𝑀(𝑥) = 𝑏𝑀𝑥𝑥 is a bending moment.
The transverse displacement in equation (3-49) is split into two components, such that one is the
contribution from axial strain due to bending and the other is the contribution from shearing strain.
𝑣 = 𝑣𝑏 + 𝑣𝑠 (3-49)
The displacements due to shearing strain is assumed such that the derivative of the displacement
equals the shearing strain as shown in equation (3-50) while the bending curvature is defined from
laminate theory and are given in equation (3-51). By integrating these and adding them, the
transverse deflection of the beam can be described by equation (3-52).
𝑑𝑣𝑠 𝑉(𝑥) (3-50)
=−
𝑑𝑥 𝑏𝜅𝐴55
𝑑2 𝑣𝑏 𝐵̅11 ̅11
𝐷 (3-51)
= − 𝑁(𝑥) + 𝑀(𝑥)
𝑑𝑥 2 𝑏 𝑏
𝐵̅11 ̅11
𝐷 ∫ 𝑉(𝑥)𝑑𝑥 (3-52)
𝑣(𝑥) = − ∫ ∫ 𝑁(𝑥)𝑑𝑥 𝑑𝑥 + ∫ ∫ 𝑀(𝑥)𝑑𝑥 𝑑𝑥 −
𝑏 𝑏 𝑏𝜅𝐴66
The beam element is loaded as shown on Figure 7 and by taking a cross-section at an arbitrary
position, it can be shown that the loads in equations (3-47) to (3-52) can be defined from the end
loadings as equation (3-53).
𝑁(𝑥) = −𝐹1 , 𝑉(𝑥) = 𝐹2 , 𝑀(𝑥) = 𝐹2 𝑥 − 𝐹3 (3-53)
Page 20 af 125
Figure 7: Boundary conditions for timoshenko beam with arbitrary axial displacement at one end.
Under these loads, the axial displacement, the transverse displacement and the derivative of the
transverse displacement become as shown in equations (3-54), (3-55) and (3-56) respectively. The
components of the stiffness matrix can then be derived by application of boundary conditions
followed by solving for the values of each of the degrees of freedom at the nodes and dividing these
by the loads. To show the process, this is shown for the boundary condition in Figure 7 where the
only non-zero degree of freedom is an arbitrary displacement at the leftmost end of the beam.
̅
𝐴11 𝐵̅11 𝐹2 𝑥 2 (3-54)
𝑢(𝑥) = − 𝐹1 𝑥 − ( − 𝐹3 𝑥) + 𝐶1
𝑏 𝑏 2
𝑑𝑣 𝐵̅11 ̅11 𝐹2 𝑥 2
𝐷 𝐹2 (3-56)
= 𝐹1 𝑥 + ( − 𝐹3 𝑥) − + 𝐶2
𝑑𝑥 𝑏 𝑏 2 𝑏𝜅𝐴66
By application of the boundary conditions at 𝑥 = 0, the constants of integration are solved for.
𝑢(0) = 𝐶1 = 𝑢1 (3-57)
𝑣(0) = 𝐶3 = 0 (3-58)
𝑑𝑣 𝐹2 𝐹2 (3-59)
(0) = − + 𝐶2 = − ⇒ 𝐶2 = 0
𝑑𝑥 𝑏𝜅𝐴55 𝑏𝜅𝐴55
𝑑𝑢
Worth noting here is that the built-in boundary condition is taken as 𝑑𝑧 = 0 at 𝑦 = 0, which implies
𝑑𝑣𝑏 𝑑𝑣 𝑑𝑣𝑠 𝐹
𝜙𝑥 = = 0 as the cross-section is prevented from rotating and thus 𝑑𝑥 = = − 𝑏𝜅𝐴2 . From the
𝑑𝑥 𝑑𝑥 66
boundary conditions at 𝑥 = 𝑙𝑒 , the forces and finally an expression for the arbitrary displacement
are derived in equations (3-60) through (3-65).
Page 21 af 125
𝑑𝑣 𝐵̅11 ̅11 𝐹2 𝑙𝑒2
𝐷 𝐹2 𝐹2 (3-60)
(𝑙𝑒 ) = 𝐹1 𝑙𝑒 + ( − 𝐹3 𝑙𝑒 ) − =−
𝑑𝑥 𝑏 𝑏 2 𝑏𝜅𝐴66 𝑏𝜅𝐴66
𝐵̅11 𝐹2 𝑙𝑒 (3-61)
⇒ 𝐹3 = 𝐹1 +
𝐷̅11 2
𝐵̅11 𝐹1 𝑙𝑒2 𝐷̅11 𝐹2 𝑙𝑒3 𝐹3 𝑙𝑒2 𝐹2 𝑙𝑒 (3-62)
𝑣(𝑙𝑒 ) = + ( − )− =0
𝑏 2 𝑏 6 2 𝑏𝜅𝐴66
⇒ 𝐹2 = 0 (3-63)
The forces and moment at the other end of the beam are expressed in terms of 𝐹1 , 𝐹2 and 𝐹3 , derived
from static equilibrium of the beam and thus the column of the stiffness matrix relating the forces to
𝑢1 can be found similarly to how the first coefficient is shown in equation (3-66).
−1 (3-66)
̅ (𝐵̅11 )2
𝑏 (𝐴11 − ̅ )
𝐹1 𝐷11
𝑘11 = =
𝑢1 𝑙𝑒
By switching which end is allowed to have an arbitrary axial displacement, the coefficients relating
to 𝑢2 are found. Similarly, the remaining coefficients of the stiffness matrix are found by applying
boundary conditions of an arbitrary transverse displacement or an arbitrary rotation of the cross-
section at one end, as shown in Figure 8 and Figure 9.
Figure 8: Beam element with an arbitrary transverse Figure 9: Beam element with an arbitrary rotation of the cross-
displacement at one end sectionat one end
To simplify the equations and keep the contribution from shear deformation clear, the constant Φ is
12𝐸𝐼
introduced, analog to the constant equal to 𝐺𝜅𝐴𝑙2 for isotropic beams which is presented in [4].
𝑒
̅11
12𝑏𝐷 −1 (3-67)
Φ=
𝑏𝜅𝐴66 𝑙𝑒2
All of the coefficients of the stiffness matrix in equation (3-74) is given in equations (3-68) through
(3-73).
Page 22 af 125
−1 (3-68)
̅ − (𝐵̅11 )2
𝑏 (𝐴11 ̅11 )
𝐷
𝑘11 = 𝑘44 = −𝑘14 =
𝑙𝑒
𝑏𝐵̅11 (3-69)
𝑘13 = 𝑘46 = −𝑘16 = −𝑘34 =
̅ 𝐷
𝑙𝑒 (𝐴11 ̅11 − (𝐵̅11 )2 )
12𝑏𝐷 ̅11
−1 (3-70)
𝑘22 = 𝑘55 = −𝑘25 =
𝑙𝑒3 (1 + Φ)
6𝑏𝐷̅11
−1 (3-71)
𝑘23 = 𝑘26 = −𝑘35 = −𝑘56 =
(1 + Φ)𝑙𝑒2
̅ 𝐷
𝐴11 ̅11 (4 + Φ) − 3(𝐵̅11 )2 (3-72)
𝑘33 = 𝑘66 =𝑏
̅11 (1 + Φ)𝑙𝑒 (𝐷
𝐷 ̅11 𝐴11
̅ − (𝐵̅11 )2 )
̅ 𝐷
𝑏(𝐴11 ̅11 (2 − Φ) − 3(𝐵̅11 )2 ) (3-73)
𝑘36 =
̅11 (1 + Φ)(𝐷
𝑙𝑒 𝐷 ̅11 𝐴11
̅ − (𝐵̅11 )2 )
Because of the coupling between extension and bending, an element without extensional DoF
cannot be constructed simply by omitting rows and columns of the stiffness matrix as this would be
equivalent to locking the axial displacement of all nodes, thus creating axial loads due to bending
which has an effect on the bending stiffness. Instead, the development of the element is done while
setting 𝐹1 = 𝐹4 = 0 and only solving in the equations for the displacement and its derivative. The
̅11
stiffness matrix then takes a form similar to that in [2] and [4], just with 𝐸𝐼 replaced by 𝑏𝐷 −1
and
with 𝐴𝐺 replaced by 𝑏𝐴66 . The use of 𝐴66 rather than 𝐴55 is due to the beam having the y-direction
as the through-thickness direction.
12 6𝑙𝑒 −12 6𝑙𝑒 𝑣1 𝐹2 (3-75)
̅11
𝑏𝐷 −1
6𝑙𝑒 𝑙𝑒2 (4+ Φ) −6𝑙𝑒 𝑙𝑒2 (2
− Φ) 𝜙 1 𝐹
{ 𝑣 } = { 3}
3
𝐿𝑒 (1 + Φ) −12 −6𝑙𝑒 12 −6𝑙𝑒 2 𝐹5
[ 6𝑙𝑒 2
𝑙𝑒 (2 − Φ) −6𝑙𝑒 6𝑙𝑒 ] 𝜙2 𝐹6
Page 23 af 125
3.7.1 Shear correction factor
The Timoshenko beam theory employs a physically incorrect assumption of the shear strain and
makes use of a shear correction factor which reduces the shear stiffness to the same value as a beam
with a physically correct shear stress distribution.
The shear stress correction factor is calculated as described in [5], by taking the ratio of the strain
energy of the assumed shear stress distribution to the strain energy of the actual shear stress
distribution. For a homogeneous beam, the shear stress given by elasticity theory is 𝜎6𝑐 =
3𝑉𝑥 2𝑦 2 𝑉
(1 − ( ) ) while the shear stress from first order shear deformation theory is 𝜎6𝑓 = 𝑥 . The
2𝑏ℎ ℎ 𝑏ℎ
shear strain energies are then as in
3𝑉𝑥2 (3-76)
𝑈𝑠𝑐=
5𝐺𝑏ℎ
𝑓 𝑉𝑥2 (3-77)
𝑈𝑠 =
2𝜅𝐺𝑏ℎ
5
So the shear correction factor for the rectangular cross-section homogeneous beam are 𝜅 = 6. The
actual shear stress distribution for a laminated beam is not equal to the shear stress distribution of a
homogeneous beam, so the commonly used value is incorrect.
The axial stress caused by bending and stretching the laminate can be used to calculate the actual
shear stress by solving for it in equation (3-78) which forms part of the 2D elastic equilibrium
equations. Once the actual shear stress through the lamina is calculated, the potential shear strain
energy can be calculated and set equal to the potential energy of the assumed strain distribution so
that the shear correction factor can be solved for.
𝜕𝜎1 𝜕𝜎6 (3-78)
+ =0
𝜕𝑥 𝜕𝑦
𝜕𝜎6 𝜕𝜎2 (3-79)
+ =0
𝜕𝑥 𝜕𝑦
Using this method, the algebraic expression for the shear stress correction factor in equation (3-80)
is developed by [11] but is here given with the y being the through-thickness direction. This shear
stress correction factor is calculated from the assumption that the bending moment whose derivative
equals the shear force are the only loads. Due to the complexity of the expression in (3-80),
attempts at adapting it to include normal forces has not been successful in this project but since the
coupling between axial force and transverse deflection of an unsymmetric beam are due to bending,
axial loads are assumed not to have any coupling to shear deformation.
Page 24 af 125
−1 (3-80)
(𝐴56 )2
𝜅 = (𝐴55 − )
𝐴66
𝑁
1
⋅ ∑ 2
(𝑘)
𝑘=1 (𝑘)
(𝑄̅56 )
𝑄̅55 − (𝑘)
( ( 𝑄̅66 )
−1
𝑅𝑘 2 𝑉𝑘
𝑃𝑘 (𝑦𝑘 − 𝑦𝑘−1 ) + 2
(𝑦𝑘 − 𝑦𝑘−41 ) + (𝑦𝑘3 − 𝑦𝑘−1
3
)
⋅( 2 3 )
𝑊𝑘 4 4
𝑋𝑘 5 5
+ (𝑦 − 𝑦𝑘−1 ) + (𝑦𝑘 − 𝑦𝑘−1 )
4 𝑘 5
)
Where
𝐽𝑘2 𝑦𝑘4
𝑃𝑘 = 𝑇𝑘2 + 𝐻𝑘2 𝑦𝑘2 − 2𝑇𝑘 𝐻𝑘 𝑦𝑘 + 𝑈𝑘2 + − 𝑈𝑘 𝐽𝑘 𝑦𝑘2 + 2𝑇𝑘 𝑈𝑘 − 𝑇𝑘 𝐽𝑘 𝑦𝑘2 − 2𝐻𝑘 𝑈𝑘 𝑦𝑘 + 𝐻𝑘 𝐽𝑘 𝑦𝑘3
4
𝑅𝑘 = 2𝑇𝑘 𝐻𝑘 − 2𝐻𝑘2 𝑦𝑘 + 2𝐻𝑘 𝑈𝑘 − 𝐻𝑘 𝐽𝑘 𝑦𝑘2
𝐽𝑘2 𝑦𝑘2
𝑉𝑘 = 𝐻𝑘2 − + 𝑈𝑘 𝐽𝑘 + 𝑇𝑘 𝐽𝑘 − 𝑦𝑘 𝐻𝑘 𝐽𝑘
2
𝑊𝑘 = 𝐻𝑘 𝐽𝑘
𝐽𝑘2
𝑋𝑘 =
4
𝑘−1
𝑇𝑘 = ∑ 𝐻𝑚 (𝑦𝑚 − 𝑦𝑚−1 )
𝑚=1
𝑘−1
𝐽𝑚 2 2
𝑈𝑘 = ∑ (𝑦 − 𝑦𝑚−1 )
2 𝑚
𝑚=1
(𝑘)
𝐻𝑘 = ∑ 𝑄̅1𝑖 𝐵̅1𝑖
𝑖=1,2,6
(𝑘) ̅
𝐽𝑘 = ∑ 𝑄̅1𝑖 𝐷 1𝑖
𝑖=1,2,6
By use of the correction factor, the stiffness of the Timoshenko beam equals that of a beam with the
correct shear stress distribution, but it does not correct the fact that the boundary conditions
influence the stiffness. The concentrated load at the center of the beam in 3-point bending will mean
that the slope of the shear deformation has opposite sign on either side of the load and thus even
Page 25 af 125
though the slope of the bending deflection is continuous, the slope of the centerline is not
continuous and thus the beam is said to have only 𝐶 0 -continuity because none of the derivatives are
continuous. As shown on Figure 10, the calculated shape of the beams has a sharp bend which
would not be physically possible without fracturing the composite fibers. Another consequence of
the assumed boundary conditions and degree of continuity is that any beam that overhang the
supports have no stresses and thus no influence on the calculated stiffness. The influence of slightly
incorrect boundary conditions is argued by Timoshenko to only be of little influence to the overall
stiffness due to Saint-Venant’s principle. [6] This principle states that small changes to the
distribution of a load in a way that is statically equivalent only produces localized changes to the
stress distribution.
Further discussion of the boundary conditions exist in section 3.8.2 where the Timoshenko beam
model is compared to another beam model for which the only difference in the stiffness of the beam
element exist in the boundary conditions.
3.7.2 Stress calculations
In order to calculate the stresses of a finite element, the strain displacement matrix derived from
shape functions are usually used but since shape functions have not been used to derive the beam
elements in this project, it is necessary to calculate them now in order to calculate the stresses.
The shape functions are found from the equations used to derive the stiffness matrix by diving the
function for the displacements and rotation by the degrees of freedom at the nodes. When the values
of the degrees of freedom at the nodes are calculated by the stiffness matrix and external loads,
these can be multiplied by the shape functions 𝑁𝑖𝑗 to yield the values of the degrees of freedom
within the element as shown in equation (3-81) and (3-95). The shape functions are given in Table
3.
Page 26 af 125
𝑢1 (3-81)
𝑣1
𝒩11 𝒩12 𝒩13 𝒩14 𝒩15 𝒩16 𝑢(𝑥)
𝜙1
[𝒩21 𝒩22 𝒩23 𝒩24 𝒩25 𝒩26 ] 𝑢 = { 𝑣(𝑥) }
2
𝒩31 𝒩32 𝒩33 𝒩34 𝒩35 𝒩36 𝑣 𝜙(𝑥)
2
{𝜙2 }
𝝓𝟐 𝐵̅ 𝒩26 = (𝜙 − 2)𝑙𝑒 𝑥 + 3𝑥 2
3 ̅11 (𝑙𝑒 − 𝑥)𝑥 𝒩36 =
𝐷11 𝑙𝑒2 (1 + 𝜙)
𝒩16 = −𝑙𝑒3 𝜙 + 𝑥 2 𝑙𝑒 (Φ − 2) + 2𝑥 3
(1 + Φ)𝑙𝑒2
2𝑙𝑒2 (1 + 𝜙)
Table 3: Shape functions for timoshenko beam element
To find the axial strain and the bending curvature, only the shape functions related to the axial and
rotational DoF, in equation (3-82) and (3-83) are used.
⌊𝒩1 ⌋ = ⌊𝒩11 𝒩12 𝒩13 𝒩14 𝒩15 𝒩16 ⌋ (3-82)
These are then differentiated with respect to x to gain the strain-displacement functions in equation
(3-84) and (3-85) with the coefficients given in table
⌊𝔅1 ⌋ = ⌊𝔅11 𝔅12 𝔅13 𝔅14 𝔅15 𝔅16 ⌋ (3-84)
Page 27 af 125
𝒅𝒖 𝒅𝝓
𝒅𝒙 𝒅𝒙
𝒖𝟏 1 𝔅31 = 0
𝔅11 = −
𝑙𝑒
𝒗𝟏 𝐵̅ 12𝑥 − 6𝑙𝑒
6 ̅11 (−2𝑥 + 𝑙𝑒 ) 𝔅32 =
𝐷11 𝑙𝑒3 (1 + Φ)
𝔅12 =
𝑙𝑒3 (1 + Φ)
𝝓𝟏 𝐵̅ (−Φ − 4)𝑙𝑒 + 6𝑥
3 ̅11 (𝑙𝑒 − 2𝑥) 𝔅33 =
𝐷11 𝔅12 (1 + Φ)𝑙𝑒2
𝔅13 = 2
=
𝑙𝑒 (1 + Φ) 2
𝒖𝟐 1 𝔅34 = 0
𝔅14 =
𝑙𝑒
𝒗𝟐 𝐵̅ −12𝑥 + 6𝑙𝑒
6 ̅11 (−2𝑥 + 𝑙𝑒 ) 𝔅35 = = −𝔅32
𝐷11 𝑙𝑒3 (1 + Φ)
𝔅15 =− = −𝔅12
𝑙𝑒3 (1 + Φ)
𝝓𝟐 𝐵̅ (Φ − 2)𝑙𝑒 + 6𝑥
3 ̅11 (𝑙𝑒 − 2𝑥) 𝔅36 =
𝐷11 𝔅12 (1 + Φ)𝑙𝑒2
𝔅16 = 2
=
𝑙𝑒 (1 + Φ) 2
Table 4: Strain-displacement functions for laminated Timoshenko beam element
It is worth noting that as a special case, the strain-displacement functions in equation (3-98) reduce
to those of a bar element for a symmetric laminate or homogeneous beam where 𝐵̅11 = 0 and that
those in (3-85) reduce to those of a Bernoulli-Euler beam when Φ = 0. With these strain-
displacement functions, the axial strains are calculated per equation (3-86).
(
𝜀1 0) ⌊𝔅1 ⌋ (3-86)
{ }=[ ] {𝑑}
(1)
𝜀1 ⌊𝔅3 ⌋
To find the in-plane stresses the beam is now assumed to be a plate strip under these strains along
with other to-be-determined in-plane strains. A bending moment and an in-plane force only in the x-
direction are assumed to be the only external loads. The axial force and the bending moment can be
calculated from (3-35) but as the transverse and in-plane shear strains aren’t explicitly calculated in
the beam calculations, calculating by (3-87) and (3-88) are incorrect for this element. For the beam
element where the axial degree of freedom is omitted, calculating the bending moment from (3-88)
̅11
gives the correct result by omitting the terms relating to midplane strain and using 𝐷 −1
rather than
the stiffnesses 𝐷11 , while 𝑁𝑥 = 0.
𝑑𝑢 𝑑𝜙 (3-87)
𝑁𝑥 = 𝑏𝐴11 − 𝑏𝐵11 = 𝑏(𝐴11 ⌊𝔅1 ⌋ − 𝐵11 ⌊𝔅3 ⌋){𝑑}
𝑑𝑥 𝑑𝑥
Page 28 af 125
𝑑𝑢 𝑑𝜙 (3-88)
𝑀𝑥 = −𝑏𝐵11 + 𝑏𝐷11 = 𝑏(−𝐵11 ⌊𝔅1 ⌋ + 𝐷11 ⌊𝔅3 ⌋){𝑑}
𝑑𝑥 𝑑𝑥
For the beam element with axial displacement where (3-87) and (3-88) are incorrect, the external
loads on an element are instead calculated by multiplying the element stiffness matrix in (3-74) by
the nodal displacements and then calculating the in-plane force and bending moment from (3-53).
Using the calculated force and moment, the in-plane strains and bending curvatures can now be
calculated.
(0) (3-89)
𝜀1 ̅
𝐴11 𝐵̅11
(0) [𝐴̅] 𝑁𝑥 [𝐵̅ ] 𝑀𝑥 𝑁𝑥 𝑀𝑥
{𝜀3 } = { 0}− { 0 } = {𝐴̅31 } ⋅ − {𝐵̅31 } ⋅
𝑏 𝑏 𝑏 𝑏
𝜀
(0) 0 0 𝐴̅41 𝐵̅41
4
(1) (3-90)
𝜀1 𝐵̅11 ̅11
𝐷
[𝐵̅ ]𝑇 𝑁𝑥 ̅ ] 𝑀𝑥
[𝐷 𝑁𝑥 𝑀
(1)
{𝜀3 } = − {0}+ { 0 } = − {𝐵̅13 } ⋅ ̅31 } ⋅ 𝑥
+ {𝐷
𝑏 𝑏 𝑏 𝑏
𝜀
(1) 0 0 𝐵̅14 ̅41
𝐷
4
Thus, the in-plane stresses in the 𝑖 ′ 𝑡ℎ lamina at a distance 𝑧𝑖 from the midplane, in the global
coordinate system are calculated per (3-91).
(0) (1) (3-91)
𝜎̅1 𝜀1 𝜀1
{𝜎̅3 } = [𝑄̅ ]𝑖 ⋅ ({𝜀3(0) } − 𝑧 {𝜀3(1) })
𝜎̅4 𝑖 (0) (1)
𝜀4 𝜀4
This stress is then transformed into the coordinate system of the lamina by a transformation matrix
[𝑇], which is similar to the one in (3-22) but rearranged so that it rotates around the y-axis instead
of the z-axis.
𝜎1 𝜎̅1 (3-92)
𝜎
{ 3} = [𝑇] −1 𝜎
{ ̅3 }
𝜎6 𝑖 𝜎̅6 𝑖
With the lamina stresses in the plane of the coordinate system, plane failure criteria can be applied.
While the through-thickness shear stress can be found directly from the equations of the
Timoshenko beam, it is not very useful, as it would be calculated to be a constant value and for a
Bernoulli-Euler beam, it cannot be calculated from the beam theory at all. Similarly, through-
thickness compressive stresses are not calculated by any of these beam models. Instead, the
through-thickness stresses may be computed from the equilibrium equations of static 3D elasticity
in equation (3-93).
𝜕𝜎1 𝜕𝜎6 𝜕𝜎4 (3-93)
+ + =0
𝜕𝑥 𝜕𝑦 𝜕𝑧
Page 29 af 125
𝜕𝜎6 𝜕𝜎2 𝜕𝜎5
+ + =0
𝜕𝑥 𝜕𝑦 𝜕𝑧
𝜕𝜎4 𝜕𝜎5 𝜕𝜎3
+ + =0
𝜕𝑥 𝜕𝑦 𝜕𝑧
Due to the beam assumption, the derivatives with respect to the transverse direction are zero and
thus (3-93) reduces to the plane stress equilibrium equations in (3-78) and (3-79). When the bending
moment and shear force are the only loads, the transverse normal stress is a multiplication of the
derivative of the shear force which is zero in these beam elements as all loads are applied as point
loads at the nodes.
The transverse shear stresses at any point through the thickness is calculated in the k’th layer as in
equation (3-94) by integrating up until the coordinate, y, at which the stress is wanted known. The
integration constant 𝐺 (𝑘) equals the stress at the bottom surface of the k’th layer at 𝑦𝑘−1 .
𝑦 (3-94)
(𝑘) 𝑉𝑥 (𝑘)
𝜎6 = −∫ ( ∑ 𝑄̅1𝑖 (−𝐵̅1𝑖 + 𝑦𝐷
̅1𝑖 )) 𝑑𝑦
ℎ 𝑏
2 𝑖=1,3,4
On Figure 11, the shear stresses of an asymmetric beam with 14 layers in the top skin, 4 layers in
the bottom skin and a 25mm thick core are shown. The stresses in Figure 11 are calculated at
10.000 equidistant points and vary parabolically through each layer even though it doesn’t look like
it. The apparent linear variation of shear stress through each layer are due to the laminated structure
with large differences in shear stiffness.
Page 30 af 125
Figure 11: Through-thickness shear stress of an asymmetric beam
Page 31 af 125
The shear deflection is then added to the deflection due to bending. When the face sheets are very
thin or if they are not sufficiently stiff in shear, this theory will overestimate the stiffness as the
shear stress tends towards the distribution of a homogeneous beam. On the other hand, when the
face sheets are thick, the stiffness is underestimated due to not including the potential shear strain
energy in the face sheets. On Figure 12, the shear deflection under a unit load is compared to the
solution of a Timoshenko beam as derived in section 3.7 with a shear correction factor as described
in section 3.7.1. The beam is assumed to have a 25mm thick core of M80 foam and face sheets
consisting of Gurit SE84-LV HEC prepreg with the fibers in 0-degree direction.
When doing a similar comparison where the face sheets are composed of alternating 0 degree and
90 degree layers, the error rises more slowly than when it consists of only 0 degree layers but is still
significant with an error of 41.2% when the face sheets have a thickness of 4,256mm with 14 layers.
With these materials, the shear stiffness will be underestimated for all practical thicknesses of face
sheets but since the both the sign and magnitude of the error depends on the material stiffnesses and
thicknesses, the sandwich theory cannot be thought of as a generally applicable theory to predict the
stiffness of a laminated beam.
Page 32 af 125
applies to homogeneous beams and is called a single-variable theory because the transverse
deflection of the centerline can be expressed only in terms of the bending deflection.
Initially, the transverse deflection is assumed to comprise of a bending component and a shear
component as in equation (3-49). The assumed shear strain distribution is parabolic and given in
equation (3-96).
5 𝑦 2 𝜕𝑣𝑠 (3-96)
𝜀6 = ( − 5 ( ) )
4 ℎ 𝜕𝑥
Within the derivation, the static equilibrium equations for 2D elasticity, equation (3-78) and (3-79),
are applied and solved for to obtain the gross equilibrium equations of which equation (3-97) is
solved to yield the shear deflection in terms of the second derivative of the bending deflection as in
equation (3-98).
𝜕𝑀𝑥 (3-97)
− 𝑉𝑥 = 0
𝜕𝑥
𝜕 3 𝑣𝑏 5𝑏ℎ 𝜕𝑣𝑠
= 𝐸𝐼 − 𝐺
𝜕𝑥 3 6 𝜕𝑥
6 𝜕 2 𝑣𝑏 (3-98)
𝑣𝑠 = ( 𝐸𝐼 )
5𝐺𝐴 𝜕𝑥 2
Within the solution of equation (3-97) a term in equation (3-99) exist which has the third derivative
of the shear deflection in it but this equals zero for homogeneous beams, thus leading to the
dependency of only one variable.
ℎ (3-99)
𝑦=
2 1 𝑧 5 𝑧 3 𝜕 3 𝑤𝑠
∫ 𝑏ℎ𝑦𝐸 ( ( ) − ( ) ) ⋅ 𝑑𝑧 = 0
𝑦=−
ℎ 4 ℎ 3 ℎ 𝜕𝑥 3
2
When this theory is applied to a laminated beam in which the stiffness of each layer is weighted
differently by equation (3-100), the equivalent of equation (3-99) is given as equation (3-101)
which is not zero. This leads to it not being possible to express the shear deflection only in terms of
the bending deflection as shown in equation (3-102).
𝑛 𝑦=𝑦𝑘 (3-100)
(𝐴𝑖𝑗 , 𝐵𝑖𝑗 , 𝐷𝑖𝑗 , 𝐸𝑖𝑗 , 𝐹𝑖𝑗 , 𝐻𝑖𝑗 ) = ∑ ∫ 𝑄̅𝑖𝑗
𝑘
(1, 𝑦, 𝑦 2 , 𝑦 3 , 𝑦 4 , 𝑦 6 ) 𝑑𝑧
𝑘=1 𝑦=𝑦𝑘−1
𝑛 𝑧=𝑦𝑘 (3-101)
1 𝑦 5 𝑦 3 𝜕 3 𝑣𝑠
∑∫ 𝑏ℎ𝑦𝑄̅11
𝑘
( − ( ) ) 3 𝑑𝑧
𝑧=𝑦𝑘−1 4ℎ 3 ℎ 𝜕𝑥
𝑘=1
Page 33 af 125
𝑏 𝜕 3 𝑣𝑠 5𝑏 𝜕 3 𝑣𝑠
= 𝐷11 − 𝐹
4 𝜕𝑥 3 3ℎ2 11 𝜕𝑥 3
ℎ2 𝜕 2 𝑣𝑏 5 1 𝜕 2 𝑣𝑠 (3-102)
𝑣𝑠 = (−𝐷11 − ( 𝐹 − 𝐷 ) )
10𝐷55 𝑑𝑥 2 3ℎ2 11 4 11 𝜕𝑥 2
The main implication is that for a laminated beam, this theory does not lead to a single-variable
formulation but rather a higher-order theory which requires more degrees of freedom to be defined
at the boundary conditions to be solvable. Furthermore, the assumed displacement fields are not
correct for laminated beams and thus, a shear correction factor would be needed to display the
correct potential shear strain energy. For these reasons, no more work is done within this project to
develop this theory into a usable model for laminated beams, although the existence of a two
variable plate theory for orthotropic plates suggest that it may be possible to produce a usable beam
theory with two variables. [14]
By developing a finite element for a homogeneous beam by the direct stiffness method as in section
𝑑𝑢 𝑑𝑢
3.7, with the built-in boundary condition as 𝜙1 = (𝑑𝑦)𝑦=0 and 𝜙2 = (𝑑𝑦 )𝑦=0 and defining the
𝑥=0 𝑥=𝐿
18𝐸𝐼
shear constant as Φ = 𝐺𝐴𝑙2 , the elemental stiffness matrix takes the form in equation (3-103), which
𝑒
is similar to that which is derived for a Timoshenko beam without axial forces.
12 6𝑙𝑒 −12 6𝑙𝑒 (3-103)
2
𝐸𝐼 6𝑙𝑒 𝑙𝑒 (4 + Φ) −6𝑙𝑒 𝑙𝑒2 (2
− Φ)
3
𝐿𝑒 (1 + Φ) −12 −6𝑙𝑒 12 −6𝑙𝑒
2
[ 6𝑙𝑒 𝑙𝑒 (2 − Φ) −6𝑙𝑒 6𝑙𝑒 ]
In this theory, the expression for the shear force, when evaluated as the integral of shear stresses
over the cross-sectional area, equals that of the Timoshenko theory when the shear correction factor
4
is that of a homogeneous beam. This proves that the difference of a factor 3 in the effect of shear
deformation in the finite element theories is due to the definition of the built-in boundary
conditions, making this theory slightly more flexible. When compared to Timoshenko and exact
theory in [13], this theory more accurately predicts transverse deflection in the presence of built-in
boundary conditions but predicts the same deflection as the Timoshenko beam when the beam is
simply supported. The difference lies in the fact that both theories fix the vertical slope of the
middle of the cross-section as their built-in boundary condition but that a parabolic shear stress
distribution will have a higher rotation at the middle than an equivalent constant shear stress
distribution.
With the finite element models built with the proposed beam elements, the case of a simply
supported beam with a central loading is equivalent of each half of the beam being a cantilever
beam which is built-in at the middle and thus the case of 3 point bending does not yield the same
stiffnesses for these two theories. When comparing steel beams with a span of 400 mm, it can be
seen on Figure 13 that the calculated difference in stiffness is still relatively small but is rising with
Page 34 af 125
increasing importance of the shear deflection as the rotation of the cross-section at the beam
centerline increases.
This doesn’t in itself tell much about how severely a laminated Timoshenko beam with a flexible
core will be affected by the assumptions relating to the boundary conditions. Looking at the actual
shear strains of a laminated Timoshenko beam calculated from the 2D equilibrium equations, it is
seen that the shear strain is close to constant through the core which accounts for most of the
thickness of the beam. The shear correction factors calculated by equation (3-80) are also found to
be an order of magnitude smaller for the laminates in this project than the correction factor for a
homogeneous beam, thus indicating that the cross-section is approximately in a state of constant
shear strain. These arguments make it plausible that very little additional accuracy would come
from using a beam theory which more accurately represents the shear strain at the boundary
conditions and thus the Timoshenko beam theory will be used.
Page 35 af 125
as beams with negligible width, plane stress elements will be developed as isoparametric elements
as in [10]. A parametric element has the shape functions developed for a regularly shaped master
element and then the coordinates in the actual element, which may not have a regular shape, are
calculated as a multiplication of the shape functions by the nodal coordinates as shown in equation
(3-104). The internal displacements are mapped by shape functions as well as shown in equation
(3-105) and as the same shape functions are used for both, the element is said to be isoparametric.
𝑥 ∑𝒩 𝑥 (3-104)
{𝑦} = { 𝑖 𝑖 } = [𝒩]{𝑐}
∑𝒩𝑖 𝑦𝑖
𝑢 ∑𝒩 𝑢 (3-105)
{ } = { 𝑖 𝑖 } = [𝒩]{𝑑}
𝑣 ∑𝒩𝑖 𝑣𝑖
The shape functions are developed for the master element in the (𝜉, 𝜂)-coordinate system so as the
nodal displacements, {𝑑}, are given in the global coordinates, the strain-displacement matrix in
equation (3-106) includes transformations between these coordinate systems. To write these
equations more compact, derivatives are signified by subscript comma followed by the coordinate
which the derivation is with respect to.
Γ11 Γ12 0 0 (3-106)
1 0 0 0 Γ Γ22 0 0
[𝔅] = [0 0 0 1] [ 021 ]
0 Γ11 Γ12
0 1 1 0
0 0 Γ21 Γ22
𝒩1,𝜉 0 𝒩2,𝜉 0 𝒩3,𝜉 0 𝒩4,𝜉 0
𝒩1,𝜂 0 𝒩2,𝜂 0 𝒩3,𝜂 0 𝒩4,𝜂 0
⋅
0 𝒩1,𝜉 0 𝒩2,𝜉 0 𝒩3,𝜉 0 𝒩4,𝜉
[ 0 𝒩1,𝜂 0 𝒩2,𝜂 0 𝒩3,𝜂 0 𝒩4,𝜂 ]
Wherein the Jacobian matrix is given by equation (3-107), the Jacobian is given by (3-108) and the
inverse of the Jacobian matrix is given by (3-109), all replicated from [10]. The shape functions for
the plane quadrilateral elements Q4, Q8 and Q9 are given in Table 5 and the derivatives are given in
Table 6. The Q4 element is called a bilinear element with four nodes, the Q8 and Q9 elements are
quadratic elements with midside nodes and where the Q9 element also has an internal central node.
The Q8 element is called a Serendipity element and the Q9 element is called a Lagrange element.
𝑥,𝜉 𝑦,𝜉 ∑𝒩𝑖,𝜉 𝑥𝑖 ∑𝒩𝑖,𝜉 𝑦𝑖 (3-107)
[𝐽] = [𝑥 𝑦,𝜂 ] = [∑𝒩𝑖,𝜂 𝑥𝑖 ]
,𝜂 ∑𝒩𝑖,𝜂 𝑦𝑖
Page 36 af 125
Include only if node 𝑖 is present
𝑖=5 𝑖=6 𝑖=7 𝑖=8 𝑖=9
𝒩1 1 1 1 1
(1 − 𝜉)(1 − 𝜂) − 𝒩5 − 𝒩8 − 𝒩
4 2 2 4 9
𝒩2 1 1 1 1
(1 + 𝜉)(1 − 𝜂) − 𝒩5 − 𝒩6 − 𝒩
4 2 2 4 9
𝒩3 1 1 1 1
(1 + 𝜉)(1 + 𝜂) − 𝒩6 − 𝒩7 − 𝒩
4 2 2 4 9
𝒩4 1 1 1 1
(1 − 𝜉)(1 + 𝜂) − 𝒩7 − 𝒩8 − 𝒩
4 2 2 4 9
𝒩5 1 1
(1 − 𝜉 2 )(1 − 𝜂) − 𝒩
2 2 9
𝒩6 1 1
(1 + 𝜉)(1 − 𝜂2 ) − 𝒩
2 2 9
𝒩7 1 1
(1 − 𝜉 2 )(1 + 𝜂) − 𝒩
2 2 9
𝒩8 1 1
(1 − 𝜉)(1 − 𝜂2 ) − 𝒩
2 2 9
𝒩9 (1 − 𝜉 2 )(1 − 𝜂2 )
Table 5: Shape functions for quadrilateral elements. [10]
Page 37 af 125
𝒩7 −𝜉(1 + 𝜂) 1 1
(1 − 𝜉 2 ) − 𝜕𝒩9
2 2
𝒩8 1 −(1 − 𝜉)𝜂 1
− (1 − 𝜂2 ) − 𝜕𝒩9
2 2
𝒩9 −2𝜉(1 − 𝜂2 ) −2(1 − 𝜉 2 )𝜂
Table 6: Derivatives of shape functions from Table 5
*The derivative of these additional terms are to be taken with respect to the same of either 𝜉 or 𝜂 as
the first term is.
With these and the width 𝑏, the stiffness matrix is calculated as the integral over the master element
given in equation (3-110).
1 1 (3-110)
[𝑘] = ∫ ∫ [𝔅]𝑇 [𝐸][𝔅] 𝑡 𝐽 𝑑𝜉 𝑑𝜂
−1 −1
Rather than evaluating this integral analytically, it is evaluated numerically using a quadrature rule.
The most commonly used rule are Gauss quadrature which samples internal points and multiplies a
weighing factor. Other quadrature rules also exist and also some exist which sample at the surface,
which is useful for beams and plates where the maximum bending stresses are at the surfaces.
The Gauss quadrature is evaluated at 𝑁 x 𝑀 points such that
𝑁 𝑀
[𝑘] = ∑ ∑ 𝜙(𝜉𝑖 , 𝜂𝑗 ) ⋅ 𝑊𝑖 ⋅ 𝑊𝑗
𝑖 𝑗
Where 𝜙(𝜉𝑗 , 𝜂𝑗 ) is [𝔅]𝑇 [𝐸][𝔅] evaluated at point (𝜉𝑖 , 𝜂𝑗 ) and 𝑊𝑖 and 𝑊𝑗 are weight factors. The
coordiantes of the evaluated points and the weighing factors are taken from Table 7.
𝑁 or 𝑀 Points 𝜉𝑖 or 𝜂𝑗 Weights 𝑊𝑖 or 𝑊𝑗
1 0 2
2 1
1
±√ ≈ ±0,5773502692
3
3 0 8
≈ 0,8888888889
9
3 5
±√ ≈ ±0,7745966692 ≈ 0,5555555555
5 9
4 18 + √30
3 2 6 ≈ 0,6521451548
±√ − √ ≈ ±0,3399810435 36
7 7 5
Page 38 af 125
3 2 6
±√ + √ ≈ ±0,8611363116 18 − √30
7 7 5 ≈ 0,3478548451
36
5 0 128
≈ 0,5688888889
255
10
±√5 − 2√ ≈ ±0,5384693101
7 322 + 13√70
900
≈ 0,4786286705
10
±√5 + 2√ ≈ ±0,9061798459
7
322 − 13√70
900
≈ 0,2369268850
6 ≈ ±0,2386191861 ≈ 0,4679139346
≈ ±0,6612093865 ≈ 0,3607615730
≈ ±0,9324695142 ≈ 0,1713244924
Table 7: Gauss Quadrature weight factors. [5]
For most purposes, 𝑀 = 𝑁 as the same degree of interpolation is used in both dimensions of the
plane element. A polynomial of degree 𝑝 is integrated exactly by having 𝑁 being the smallest
1
integer greater than 2 (𝑝 + 1) so to show a linear variation in strain (2nd degree polynomial of
displacements), a 2 by 2 quadrature is needed. The accuracy of integration rises by order of
integration but a too high order may overly stiffen higher-order displacement modes and thus the
accuracy of element behavior is not the same as accuracy of integration. For the Q8 and Q9
elements, full integration is 3 by 3 and reduced integration is 2 by 2.
The Q4 element is incapable of displaying pure bending due to the lack of quadratic terms in the
shape functions and will be overly stiff due to “shear locking”. To overcome this, the Q6 element is
formed by adding four generalized/nodeless DoF such that the displacement field is interpolated by
(3-111).
𝑢 ∑𝒩 𝑢 + (1 − 𝜉 2 )𝑎1 + (1 − 𝜂2 )𝑎2 (3-111)
{ }={ 𝑖 𝑖 }
𝑣 ∑𝒩𝑖 𝑣𝑖 + (1 − 𝜉 2 )𝑎3 + (1 − 𝜂2 )𝑎4
The added degrees of freedom in the Q6 element are responsible for incompatible modes which will
leave gaps or overlaps where elements meet and the element is made overly flexible but in such a
way that a refined mesh will converge towards the exact result. To generate the strain-displacement
Page 39 af 125
matrix of the Q6 element, that of the Q4 element is simply expanded with a matrix, [𝔅𝑎 ], which
operates on the nodeless DoF such that the strain-displacement matrix is given by (3-112).
[𝔅] = [[𝔅𝑑 ] [𝔅𝑎 ]] (3-112)
This Q6 element is however incapable of displaying constant stress if the initial shape is distorted
from a rectangular shape. To overcome this, the modified element named QM6 is formed by using
the values of [𝐽], 𝐽 and 𝑡 calculated at 𝜉 = 𝜂 = 0 for the calculation of [𝔅𝑎 ] but calculating [𝔅𝑑 ] as
usual. The stiffness matrix is then calculated as described in [15] by equation (3-113).
[𝑘𝑑 ] [𝑘𝑑𝑎 ]𝑇 (3-113)
[𝑘] = [ ]
[𝑘𝑑𝑎 ] [𝑘𝑎 ]
Where
𝑁 𝑀
𝑁 𝑀
𝑁 𝑀
If an element with internal nodes or nodeless DoF is used directly as-is, all DoF are transferred into
the global system to be solved for, even though some DoF are not acted upon by other elements. It
is common to condense the element into an equivalent stiffness matrix, acting only on the nodal
DoF on the boundary such that fewer DoF are transferred into the global system. The stiffness
matrix, DoF and reaction vectors are split into a retained part with subscript r and a condensed part
with subscript c as shown in (3-114).
[𝑘𝑟𝑟 ] [𝑘𝑟𝑐 ] {𝑑𝑟 } {𝑟 } (3-114)
[ ]{ }={ 𝑟 }
[𝑘𝑐𝑟 ] [𝑘𝑐𝑐 ] {𝑑𝑐 } {𝑟𝑐 }
By solving for {𝑑𝑐 } in the calculation of {𝑟𝑐 } and inserting into the calculation of {𝑟𝑟 }, an equation
of the equivalent stiffness matrix is found.
([𝑘𝑟𝑟 ] − [𝑘𝑟𝑐 ][𝑘𝑐𝑐 ]−1 [𝑘𝑐𝑟 ]){𝑑𝑟 } = {𝑟𝑟 } − [𝑘𝑟𝑐 ][𝑘𝑐𝑐 ]−1 {𝑟𝑐 } (3-115)
Thus, the equivalent stiffness matrix working only on the DoF at the boundary is given by (3-116).
Page 40 af 125
3.9.1 Validity of element formulations
The validity of the elements are tested by two methods. One is the patch test and the other is
modeling a beam with the elements and comparing against an analytical result.
To ensure that the elements function properly, they are subjected to a patch test as described in [10],
which is a numerical experiment in which the element has to be able to express a state of constant
stress, preferably also in a deformed shape. For elements where the patch test is not passed in a
deformed state, they are said to pass the weak patch test if a refinement of the mesh results in
convergence to a state of constant stress. Refinements are not tested for the patch test here.
To test the deformed element, a square is subdivided into four elements and the central node is
displaced. On Figure 14 below, the original shape is drawn in blue and the deformed shape is drawn
in red, exaggerated by a factor of 1000. The lower left node is fixed and the two other left nodes are
constrained horizontally while a force is applied at each node to the right, consistent with a constant
stress of 10 MPa.
The Q4 and QM6 elements passed this patch test by showing a uniform stress while the author of
this project was not able to program a functional Q8 or Q9 element.
To test the performance of these elements to show bending and shearing, these elements are
assembled to a steel beam with a length of 400 mm, a width of 275 mm and a thickness of 10 mm.
Page 41 af 125
The beam is supported at each end on the lower side while a central force is applied at the upper
side. This is shown on Figure 15 where the undeformed element is shown with blue stars at the
unloaded nodes and red stars at the loaded or supported nodes while the deformed shape are shown
with red circles at the nodes.
The number of elements lengthwise and through the thickness are adjusted such that the aspect ratio
are 1 and then more elements are added in both dimensions to add more through the thickness.
Having the elements be longer than they are thick would be more computationally efficient as less
elements would be needed but this distortion would also affect the accuracy of the computation so
to limit the scope of this project from also needing to investigate this, the aspect ratio of 1 are
chosen.
The stiffness of the beam is calculated and compared to the analytical result calculated by (3-117)
which is based on a Bernoulli-Euler beam and also against a Timoshenko beam calculated by
(3-118) which is given in [5].
48𝐸𝐼 𝑁 (3-117)
𝐾0 = = 3437,5 [ ]
𝑙3 𝑚𝑚
48𝐸𝐼 𝑁 (3-118)
𝐾1 = = 3430,8 [ ]
𝐸 ℎ 2 𝑚𝑚
𝑙 3 (1 + 𝜅𝐺 ( ) )
𝑙
As can be seen from the comparisons presented in Figure 16 and Figure 17, the QM6 element is
superior to the Q4 element in terms of displaying a stiffness that is closer to the analytical result but
both converge towards a higher stiffness than the theoretical values for beams. This was not
expected as the transverse compression and the supports being concentrated at the lower corners
was expected to yield a lower stress than the beams. This over-prediction of stiffness is assumed to
Page 42 af 125
be due to the elements being linear but this cannot be tested when the programming of functional
Q8 and Q9 elements have failed.
Figure 16: Comparison of plane elements to analytical beam Figure 17: Zoomed-in view of Comparison of plane elements to
solutions. Stiffness in Newton per millimeter of vertical analytical beam solutions
deflection of the loaded central node
As a compromise between accuracy and computation time, QM6 with 6 elements through the
thickness of the beam is chosen as it can be seen on Figure 18 that the QM6 element seems
reasonably converged while the Q4 element will require a much higher number of elements to
converge to a similar level.
Page 43 af 125
3.10 Other 2D and 3D elements for laminate analysis
Besides the elements used here, there are other notable elements which can be used for analyzing
laminates.
Plate elements are two-dimensional elements which are formed by applying constraints to a three-
dimensional solid similarly to how beams are formed by applying constraints to a two-dimensional
plane stress region. These assumptions mean that the strain field through the solid can be calculated
from the deflections of the mid-surface, similarly to how the two-dimensional strains in a beam are
calculated from the deflections of the centerline. As the assumptions of a Bernoulli-Euler or
Timoshenko beam are based on the cross-sections being straight and are assumed perpendicular to
the centerline for the former but not the latter, the assumptions of plate elements relate to vectors
that are normal to the mid-surface. Kirchoff-Love plate theory, also called classical plate theory, are
analogous to Bernoulli-Euler beams in that these normals are assumed to remain perpendicular to
the midsurface after deformation, which imposes the constraint that the plate must be free of
through-thickness shear. Shear deformable plate theories are abundant like they are for shear
deformable beams but the most common theories assume constant shear strain through the thickness
and such the normals will remain straight but not necessarily perpendicular to the midsurface in the
deformed state. Often, this theory is called the Mindlin theory or the Mindlin-Reissner theory. [10]
The latter is a misunderstanding as the theory proposed by Reissner did not assume through-
thickness normal strain to be negligible as Mindlin did and thus the results differ between the
Mindlin and the Reissner plate theories even though they are very similar and were developed
around the same time. [16]
Plate elements are formulated as parametric elements similarly to how plane elements are formed
but there are three or five degrees of freedom per node depending on whether in-plane deformations
of the mid-surface are included. Apart from displacements, these are rotation of the normals about
the x and y axis, similar to how the rotation of the cross-section at the centerline is a degree of
freedom for beams. The similarity to beams is also found in that Kirchoff-Love plate elements are
𝐶 1 continuous as the slope of the mid-surface is equal between elements while Mindlin plate
elements are 𝐶 0 continuous as the slope of the mid-surface is not continuous across elements.
Mindlin plate elements are derived as the addition of a stiffness matrix for bending and a stiffness
matrix such that seperate integration schemes can be applied to the bending part than the shear part.
This is called selective integration and is used to combat a shear stiffening effect that these elements
have when the thickness is small. [10]
[𝑘] = [𝑘𝑏 ] + [𝑘𝑠 ] (3-119)
Page 44 af 125
Figure 19. Only the Lagrange element under reduced or selective integration and the Heterosis
element match the theoretical stiffness for a large range of thicknesses. The Heterosis element is a
nine-node element where the transverse deflection degree of freedom are omitted at the centernode
and selective integration is used. Because the programming of functional plane Serendipity and
Lagrange elements failed in this project, plate elements are not developed either. As the shear strain
is assumed constant through the thickness of Mindlin plates and shells, a shear correction factor is
used which for laminated sections are calculated similarly to how [11] calculates it for beams but
just in two dimensions.
Shell elements are another category of elements which can be used for modeling laminates. Shells
have a lot in common with plates but may have initial curvature and even double-curvature which
the plates does not. This curvature couples bending and membrane stresses. Shell elements are
developed as isoparametric elements with plane stress but pose more challenges than plane and
plate elements as they can have both shear locking and membrane locking effects, which result in a
too-large stiffness. To counteract these locking phenomena, remedies such as reduced integration,
drilling DoF, incompatible modes and penalty functions are employed. [10][17]
Plate elements which include in-plane stresses are often referred to as flat shells. [10] This makes
shell elements more generally applicable and thus shell elements are used in commercial software
such as ANSYS whereas plate elements are not. [17]
Solid-Shell elements are solid elements which have been imposed some of the constraints
applicable to shells but have not been condensed into nodes on the midplane. Thus, the solid-shell
element have nodes with only translational degrees of freedom and is not assumed to have plane
stress and can display a three-dimensional state of stress. In ANSYS, this element is called
SOLSH190 and uses incompatible modes to increase accuracy and overcome thickness locking in
bending and it uses special kinematics to overcome locking when the element is very thin. [17]
While solid-shell elements can be formulated as a laminated section, ANSYS recommends against
using a single element for laminated structures as the stiffness can be highly overestimated and
instead, it is recommended to stack multiple layers of solid-shell elements. Solid-shell elements can
do much of the same as a regular solid element can but have the advantage of being able to model
both relatively thick and very thin plates and it is also much faster than solid elements. While a
proper comparison of computation times has not been conducted in this project, it is the experience
of the author that using SOLSH190 elements to model the core material of a sandwich plate is often
3-4 times faster than with SOLID185 with the same mesh size and approximately the same result.
Among the drawbacks for solid-shell element are limitations to the shape as the top and bottom
surfaces must be parallel and that the element is less developed in terms of nice features. The nice
features that ANSYS has for solid elements which are not applied to solid-shell elements include
automatic h-refinement to ensure convergence of results and non-linear adaptive remeshing.
An element type which is specially developed for analyzing laminates but does not currently exist
in ANSYS is the layerwise theory by Reddy [5]. In this theory, the in-plane properties are modeled
Page 45 af 125
by two-dimensional interpolations like the plane and plate elements while the through-thickness is
modeled through separate one-dimensional interpolations. This allows the in-plane mesh and the
through-thickness mesh to be refined separately and also, the calculations of the stiffness matrix of
a layer-wise element includes significantly fewer multiplications, additions and assignments than a
comparable 3D element. The layerwise theory does not apply an explicit assumption of the strain
through the thickness and instead, the through-thickness elements are only continuous in the
displacement (𝐶 0 -continuous) so that the strains may be discontinuous while the stresses are
continuous. This allows the theory to accurately display the “zigzag” stresses predicted by exact 3D
elasticity which also result in the bending stresses not varying linearly through the thickness of the
beam as it is assumed by equivalent single layer (ESL) beam and plate models. The model is said to
be partial layerwise when the discretization through the thickness is coarser than the amount of
lamina, in which case the layers that span multiple lamina are assumed homogeneous similarly to
equivalent single layer theories. The model is said to be full layerwise when the discretization
includes one or multiple layers through each lamina. Another feature of the full layerwise theory is
the ability to assume multiple displacement fields such that the boundary conditions can be forced
to be those of ESL theories to enable direct coupling to plate and shell elements or delaminations by
imposing a displacement that only applies to the layers above a delamination. While the
presentation of the layerwise theory in [5] is to solve a general 3D stress field, the layerwise theory
can also be applied to beams. [18]
As the implementation of layerwise theory as described in [5] requires successful implementation of
2D elements, it is omitted in this project but is noted as an area with a lot of potential for future
work.
Page 46 af 125
forms the angle 𝜃 with the horizontal axis, the load at the center will form axial stresses along the
cable which resists further deflection. Thus it is shown that by including large deformations, the
calculation of infinite displacements can be avoided. In this case, the addition of a pretension of the
cable would add stiffness to the bending deflection, even when the cable is horizontal and the
addition of a compressive force would decrease the bending stiffness. The complete example is
given in lecture 8 of [19] to which the text book is [21].
To calculate these geometric nonlinearities in finite element analysis, there are generally two
methods that can be used.
The first method is the Total Lagrangian method in which a nonlinear strain measure, the Green-
Lagrange strain, is used along with a strain measure called the 2nd Piola-Kirchoff stress tensor to
describe the configuration at time t with respect to the original geometry such that the tangent
stiffness at time t can be calculated from the principal of virtual work which is then integrated over
the original volume. Within the derivations of the development of the Total Lagrangian method,
there are linearized equations which means that the solution at time 𝑡 + ∆𝑡 must be found in an
iterative manner within each load step to reduce the error that the linearization brings. The solution
at 𝑡 + ∆𝑡 are then used as a starting point for the iterative solution of the following load step, which
means that errors can accumulate if the solution did not converge before the next load step is taken.
The Total Lagrangian method involves transformations of stress concepts and terms which are
nonlinear in the displacement to make it possible to integrate over the original geometry and these
things make this method less used as it is simpler to refer to the geometry in time 𝑡 as is done in the
Updated Lagrangian method. ANSYS uses the Updated Lagrangian method so to keep the
relevancy, the Total Lagrangian method are not further described in detail.
The Updated Lagrangian method refers to the most recent calculated configuration which is at time
t. In the following, superscripts and subscripts to the left are employed to keep track of which times
the quantities are calculated and in with respect to which geometry such that 0𝑡𝜀𝑗𝑘 are the strain at
time t with the respect to the geometry at time 0. When no upper left subscript is written, it is
implied that this is the increment ∆𝑡. Also, the comma notation for derivatives is employed such
that a right subscript of comma followed by a letter means the derivative with respect to that letter.
The principal of virtual work at time 𝑡 + ∆𝑡 can be written in terms of the Cauchy stresses and
Cauchy strain, assuming small strains, and integrated over the unknown volume at time 𝑡 + ∆𝑡. It
can also be written in terms of the 2nd Piola-Kirchoff stress tensor and the Green-Lagrange strains
integrated over the volume of any known configuration but here it is chosen to integrate over the
known volume at time t.
(3-120)
𝑡+∆𝑡
∫ 𝑡+∆𝑡
𝜎𝑖𝑗 𝛿𝑡+∆𝑡 𝜀𝑖𝑗 𝑡+∆𝑡
𝑑𝒱 = ∫ 𝑡𝒮𝑖𝑗 𝛿 𝑡+∆𝑡𝑡𝑒𝑖𝑗 𝑡𝑑𝒱 = 𝑡+∆𝑡
ℛ
𝑡+∆𝑡𝒱 𝑡𝒱
Where 𝒱 is the volume, 𝜎𝑖𝑗 is the Cauchy stresses, 𝜀𝑖𝑗 is the Cauchy strains, 𝒮𝑖𝑗 is the components
of the 2nd Piola-Kirchoff stress tensor, 𝑒𝑖𝑗 is the green-lagrange strains and 𝑡+∆𝑡ℛ is the virtual
work. Working with the principal of virtual work of the volume at time t and by defining the
nonlinear strain increment as (3-121) such that the Green-Lagrange strain increment can be
expressed by the addition of the Cauchy stress increment being linear and the nonlinear term
increment by (3-122). With these, the principal of virtual work becomes as in equation (3-123)
Page 47 af 125
where the right-hand side consists of the external virtual work and the internal virtual work
corresponding to the stresses at time t. [19]
3 (3-121)
1
𝑡𝜀̅𝑖𝑗 = ∑ 𝑡𝑢𝑘,𝑖 𝑡𝑢𝑘,𝑗
2
𝑘=1
(3-123)
∫ 𝑡𝒮𝑖𝑗 𝛿 𝑡𝑒𝑖𝑗 𝑡𝑑𝒱 + ∫ 𝑡𝜎𝑖𝑗 𝛿 𝑡𝜀̅𝑖𝑗 𝑡𝑑𝒱 = 𝑡+∆𝑡
ℛ − ∫ 𝑡𝜎𝑖𝑗 𝛿 𝑡𝜀𝑖𝑗 𝑡𝑑𝒱
𝑡𝒱 𝑡𝑉 𝑡𝑉
This is so far valid for any amount of deformation but includes nonlinearities in the left-hand side
so for creating an approximation that is more easily calculated, the principal of virtual work is
linearized and is given in (3-124) where 𝐶𝑖𝑗𝑟𝑠 = 𝜆𝛿𝑖𝑗 𝛿𝑟𝑠 + 𝜇(𝛿𝑖𝑟 𝛿𝑗𝑠 + 𝛿𝑖𝑠 𝛿𝑗𝑟 ) is the fourth order
𝐸𝜈 𝐸
material tensor with 𝜆 = (1+𝜈)(1−2𝜈) and 𝜇 = 2(1+𝜈) for an isotropic material. For anisotropic,
laminated or nonlinear material laws, 𝐶𝑖𝑗𝑟𝑠 is entered appropriately.
(3-124)
∫ 𝑡𝐶𝑖𝑗𝑟𝑠 𝑡𝜀𝑟𝑠 𝛿 𝑡𝜀𝑖𝑗 𝑡𝑑𝒱 + ∫ 𝑡𝜎𝑖𝑗 𝛿 𝑡𝜀̅𝑖𝑗 𝑡𝑑𝒱
𝑡𝒱 𝑡𝒱
𝑡+∆𝑡
= ℛ − ∫ 𝑡𝜎𝑖𝑗 𝛿 𝑡𝜀𝑖𝑗 𝑡𝑑𝒱
𝑡𝒱
When (3-124) is discretized using interpolation functions by the finite element method, the left hand
side equals the virtual displacement increment of the nodes times the tangent stiffness matrix times
the actual displacement increment of the nodes. The right hand side equals the virtual displacement
increment of the nodes times the externally applied loads minus the force vector corresponding to
the internal stresses of the element. By dividing by the virtual displacement increment at both sides,
the result is equation (3-125) which must be solved iteratively to make the right hand side zero.
𝑡
𝑡𝐾 ∆𝑑 = 𝑡+∆𝑡
𝑅 − 𝑡𝑡𝐹 (3-125)
The right hand side can be thought of as residual of loads which in ANSYS Mechanical can be
viewed under the solution object as the force convergence in order to monitor how the solution
converges to a result within that load step. An example of the force convergence plot within
ANSYS is shown in Figure 20 where it can be seen that in the first iteration of each timestep (what
ANSYS calls substeps), the force imbalance is generally very high and it then iterates within a
timestep until the force imbalance is below a chosen criterion, after which the substep is converged
and the solution moves on to the next.
Page 48 af 125
Figure 20: Force convergence plot in ANSYS
Another convergence criterion which can also be seen in the solution object in ANSYS Mechanical
is the displacement convergence. The displacement is iterated in (3-126) such that at the 𝑘′𝑡ℎ
iteration of the 𝑖′𝑡ℎ loadstep, it is equal to the displacement at the previous iteration plus an
increment. The displacement convergence is then reached when the incremental displacement is
less than a value determined by the software. In equation (3-126), 𝑠 is a line search parameter that
ANSYS applies to stabilize the result by forcing smaller steps. 𝑠 = 1 by default and is reduced if
ANSYS notices convergence difficulties.
𝑡+∆𝑡 (𝑘) 𝑡+∆𝑡 (𝑘−1) (𝑘) (3-126)
𝑑𝑖 = 𝑑𝑖 + 𝑠∆𝑑𝑖
To solve (3-125), an iterative method such as the Newton-Raphson iterations are used. Many other
methods can be used, of which a modified Newton-Raphson method and an arc-length method are
available in ANSYS. The modified Newton-Raphson method is like the Newton-Raphson but
where the tangent stiffness matrix is only updated at the beginning of each load step. This requires
more iterations to solve but each iteration is faster as the tangent stiffness matrix does not have to be
recalculated. The arc-length method is a form of Newton-Raphson where instead of iterating by
going up a straight line to a certain load in each iteration, the straight line is followed up to an arc
some distance from the starting point of the load step such that the load-displacement curve can also
be followed downwards and into post-buckling regimes. [10][20]
By looking through the ANSYS documentation, it is not clear which iterative method is applied by
default in ANSYS Mechanical but it is likely that the program starts with the modified Newton-
Raphson and switches to the arc-length method if needed.
Within ANSYS Mechanical, enabling large deformations and large strains are simply done in the
analysis settings object by toggling the Large Deflection setting to “on”.
Large strains can be included in the updated lagrangian method by replacing the Cauchy
(engineering) strain by the Hencky (logarithmic) strain as these are identical for infinitesimal strains
[21 p. 615]. The engineering strain is not suitable for iterative calculations as adding multiple
incremental strains does not compute the same strain as going directly to the last strain level.
𝐿0 + ∆𝐿 𝐿𝑖 + ∆𝐿𝑖 (3-127)
𝜀𝐸 = ≠∑
𝐿0 𝐿𝑖
Page 49 af 125
More useful is the logarithmic strain, also sometimes called the true strain, which is defined as the
integral of incremental strains as shown in (3-128) for a one-dimensional strain and relates to the
engineering strain as in (3-129).
𝐿𝑖
𝑑𝐿 𝐿𝑖 (3-128)
𝜀𝐿 ∫ = ln(𝐿𝑖 ) − ln(𝐿0 ) = ln ( )
𝐿0 𝐿 𝐿0
𝜀𝑇 = ln(1 + 𝜀𝐸 ) (3-129)
When looking through the documentation of ANSYS, it is found that the elements are developed for
logarithmic strain but since this is equal to the engineering strain at infinitesimal strain levels, these
are automatically also useful for solutions where strains are assumed small.
3.11.2 Material nonlinearities
Many types of material nonlinearities exist but, in this project, elastoplasticity and laminate failure
are the most important and are the only material nonlinearities which will be treated.
Laminate failure is treated similarly as the fracture of an elastic body in the sense that if failure is
calculated at a load step, the stiffness corresponding to the failure mode of the failing lamina is
reduced or removed completely. After reducing the stiffness, the load step is recalculated to see if
any more failures occur and if not, the simulation continues with the reduced stiffness. ANSYS
supports fiber and matrix failures for laminated shells.
Elastoplasticity requires more sophisticated methods as it requires a yielding criteria, a flow rule
and a hardening rule. The yielding criteria is commonly set as the Von Mises criterion where a yield
strength is entered and this is also the criterion used for the bi- and multilinear hardening models
which can be chosen in ANSYS Workbench. The flow rule determines in which direction the
plastic strain occurs and with the bi- and multilinear hardening models in ANSYS, this is an
associative rule and the flow occurs normal to the yield surface.
Within ANSYS Workbench, there are several available choices for hardening models but only the
bi- and multilinear isotropic and kinematic hardening models will be treated here as these are the
simplest and are generally applicable to most homogeneous isotropic materials which the foam core
is assumed to be.
To have a bilinear model means that the stress-strain curve is assumed to consist of two straight
lines. The first line goes from zero and up to the yield stress with a slope that is the Young’s
modulus 𝐸 while the second line goes from the yield stress to infinity with a slope that is the
tangent modulus 𝐸𝑡 . The total strain is assumed to consist of an elastic strain 𝜀𝑒 and a plastic strain
𝜀𝑝 as shown on Figure 21.
Page 50 af 125
Figure 21: Schematic view of a bilinear model
The yield strength can be taken from data of the engineering stress since the strain at yield is often
small, meaning that the reduction of the cross-section due to the poissons ratio is also small. The
tangent modulus however must be estimated from a plot of the true stress versus the logarithmic
strain as the plastic strain will generally not be small. The true stress is the Cauchy stress evaluated
over the actual cross-section and not the initial cross-section as the engineering stress is. The actual
cross-section can be estimated during a tensile test by observing the deformations on the surface
using digital image correlation. [22] Often, such sophisticated measurement techniques are not
applied and the actual cross-section is calculated by assuming the elastic strains to be small and
assuming the plastic strain to occur under a constant volume, in which case, the true stress is
calculated from the engineering stress and the engineering strain by (3-130) [23][24].
𝐹 (3-130)
𝜎𝑡 = = 𝜎𝑒 (1 + 𝜀𝑒 )
𝐴
Since most materials have a gradual transition from the region of the stress-strain plot dominated by
elastic strains to the region dominated by plastic strain, the use of a bilinear model will have some
error. On Figure 22, some stress-strain curve is shown in black and possible bilinear models are
shown in green, red and blue. If the yield strength is chosen as the end of the linear region or the
stress at 0,2% plastic strain while the tangent modulus is taken as the slope within the region
dominated by plastic strain, the green graph is produced. With the green graph, the load-bearing
capacity is underestimated and the contained strain energy can be massively underestimated. If the
red graph is chosen to more accurately calculate the strain energy with a small overestimate during
the transition region, the yield stress may be highly overestimated. As the impact of each of these
Page 51 af 125
models depend heavily on how the actual stress-strain curve of a material looks like, the best fit of a
bilinear model is likely somewhere in between but a compromise could also be something like the
blue graph where the tangent stiffness is overestimated. As the complete stress-strain curve is rarely
available before purchasing a material and conducting tests on it, a graph like the blue graph could
be formed as a coarse approximation by looking at the yield strength, ultimate strength and strain-
to-failure which are often given in data sheets. By assuming that the maximum stress occurs at the
failure strain which is then calculated to be expressed as the logarithmic strain, the tangent modulus
can be found by connecting the yield stress and the maximum stress by a line. Depending on the
material, this would likely be an improvement over the green graph, but only for strains smaller
than the used failure strain. Limiting oneself to the failure strain given in data sheets may lead to a
very conservative estimate of the load-bearing capacity for ductile materials that exhibit necking
before failure in a tensile test. By using digital image correlation, to map the strains within the
necking band of steel specimen, it has been shown that the strain and stresses at failure can be much
higher than what is found by using an extensiometer only [24].
A multilinear model solves many of the issues that a bilinear model has but requires the stress and
strain to be known at multiple points after yielding and thus cannot be created from the data
commonly given in data sheets. The multilinear model, shown in Figure 23, employs multiple
sections with a constant tangent modulus and thus allows for the stress-strain curve to be replicated
at an arbitrarily fine resolution. The data that must be entered in ANSYS Workbench to create the
bilinear model are the stresses and the corresponding plastic logarithmic strain so the elastic strain
must be subtracted from the total strain to find the data that should be entered. After the last entered
datapoint in the multilinear model, ANSYS assumes a tangent stiffness of zero so it is important
that the last datapoint is at a higher plastic strain level than is expected to be seen in the simulation.
As the strain is often measured by an extensiometer as the average strain between two points, the
stress-strain curve at larger strain levels must be extrapolated from the tensile test data if there are
necking or if there are other reasons to believe that the material can be capable of exhibiting larger
Page 52 af 125
strain values than is observed in the tensile test. A suitable model could be Ludwik’s model (3-131)
which states that the stresses in the plastic region can be described as the yield stress plus a strength
coefficient 𝐾 multiplied by the plastic strain with the exponential of the strain hardening coefficient
𝑛. Fitting this equation to the plot of true stress versus plastic logarithmic strain before the
maximum stress allows for extrapolations into higher strain levels.
𝜎 = 𝜎𝑦 + 𝒦𝜀𝑝𝓃 (3-131)
Besides the choice of a bi- or multilinear model, a choice must also be made between isotropic or
kinematic hardening. The difference between these two models are how yielding alters the yielding
criterion for future yielding. For both hardening models, if the material is loaded until yielding at 𝜎𝑌
and then further loaded until 𝜎𝐵 before being unloaded, and then loaded back up, it will behave
elastically up to 𝜎𝐵 before any further yielding will occur. That is, the yield surface has changed but
the way that it changes is different for these two models. For isotropic hardening, the yield criterion
has expanded such that yielding in any direction now requires a yield stress of 𝜎𝐵 and the yield
criterion continues being isotropic under multiaxial loading. For kinematic hardening, the yield
criterion has moved to make the yield strength 𝜎𝐵 in the original direction but it has not expanded
so if the material is loaded in the opposite direction, the yield stress in that direction will be −(𝜎𝐵 −
2𝜎𝑌 ) so that the difference between the yield strength in two opposite directions are 2𝜎𝑌 . This
makes the yield behavior anisotropic in multiaxial loading after yielding. In reality, the material
behavior may be a combination of isotropic and kinematic hardening which can be included by use
of the Chaboche model but this is not further detailed in this project. [10] [20]
In this project, the loading is not reversed and unloading of the specimen only occurs due to
laminate failures so the choice between isotropic and kinematic hardening is of lower importance.
Page 53 af 125
While the comparison between isotropic and kinematic hardening in Figure 24 is shown for a
bilinear model, it is similarly applicable to multilinear hardening models.
ANSYS supports bi- and multilinear isotropic and kinematic hardening for solids, solid-shells and
unlaminated shells but not for laminated shells.
Page 54 af 125
penetration or initial gaps even when they are modeled to touch in the CAD software. This is solved
by changing the contact setting “interface treatment” to “adjust to touch”.
Another boundary condition nonlinearity is when a load is applied normal to the surface and the
structure is analyzed with large deflections. The direction of the load will then change with changes
to the orientation of the element.
After the DoF has been calculated the reaction force is calculated by multiplying 𝑐 to the difference
in calculated and prescribed DoF value as shown in equation (3-139)
𝑅𝑖 = 𝑐(𝑑𝑖 − ∆𝑖 ) (3-133)
The magnitude of 𝑐 is commonly chosen to be 104 times the largest value found in the stiffness
matrix. The effect is similar to if the boundary was connected to ground by a very stiff spring and is
thus only an approximation as unwanted flexibility is introduced and choosing a very high value of
𝑐 may make the stiffness matrix ill-conditioned.
The penalty approach is not recommended for this problem for reasons which will become apparent
by running a parametric study of this method with 10, 100, 1000 and 5000 beam elements and with
penalty stiffnesses from 10 to 1019 . By varying the value of 𝑐 in the calculation of a steel beam with
a rectangular 275mm by 10 mm cross-section, lying flat, it is seen in Figure 25 and Figure 26 that
the choice of penalty stiffness is influential to the calculated stiffness. The calculated values for all
amount of elements are made in the same color as all the curves except those for the 500 element
model coincide until instability arises at 𝑐 = 1014 𝑁/𝑚𝑚.
The stiffness is calculated both directly from the prescribed displacement and reactionary load and
by correcting for the penalty by assuming the penalty stiffnesses to be springs in series with the
beam. By comparing the calculation using Bernoulli-Euler beam elements to the theoretical
48𝐸𝐼
stiffness , it is seen that the stiffness is only correctly calculated within 5 significant digits in the
𝐿3
range of 𝑐 = 1010 to 𝑐 = 1014 but that the correct stiffness can be calculated from the assumption
of linear springs in series at lower values of 𝑐 but only when the amount of elements are low as
instability is seen at both low and high values of 𝑐 with 1000 beam elements and that the theoretical
value is never calculated with 5000 elements. This calculation also shows that calculating 𝑐 as a
fixed factor times the highest value in [𝐾] is dangerous as refinement will then raise 𝑐 and possibly
introduce instability.
Page 55 af 125
Figure 25: Calculated stiffness versus penalty stiffness Figure 26: Calcualted stiffness versus penalty stiffness, zoomed
in
In these calculations, the displacement was solved by Matlab’s inbuilt matrix solver mldivide(K,F)
which may have affected the stability of the solution but nonetheless, the observations regarding the
penalty approach of boundary conditions makes it unsuitable for the problem in this report as it
introduces an error which is not negligible if 𝑐 is chosen incorrectly. Since a suitable value of 𝑐 is
dependent both on the global stiffness of the problem and of the stiffness of each element, ensuring
suitable values would require running parametric studies of 𝑐 for all laminates and would greatly
increase time spent and computational efforts.
An approach of prescribing the value of a DoF without introducing error as the penalty approach
does is the elimination method described by [25] where the row and column of the prescribed DoF
in the stiffness matrix are deleted along with the row of the prescribed DoF in the force vector. The
load components of all other DoF which are connected by the stiffness matrix to the prescribed DoF
are then subtracted the load corresponding to the prescribed DoF. In [10], without naming it the
elimination method, the same approach is described but in such a way that the size of the stiffness
matrix and force vector remains unchanged with the benefit that the force and DoF vectors are
complete with all of the values in the correct places. This formulation is as shown in (3-134) that if
the 𝑖’th DoF are prescribed 𝑑𝑖 = ∆𝑖 , the rows and columns corresponding to the 𝑖’th DoF are zeroed
out except for 𝐾𝑖𝑖 and in the force vector, other DoF are amended as in [25] and the 𝑖’th value are
set to 𝐹𝑖 = 𝐾𝑖𝑖 ∆𝑖 .
𝐾11 0 𝐾1𝑛 𝑑1 𝐹1 − 𝐾1𝑖 ∆𝑖 (3-134)
[ 0 𝐾𝑖𝑖 0 ] { 𝑑𝑖 } = { 𝐾𝑖𝑖 ∆𝑖 }
𝐾𝑛1 0 𝐾𝑛𝑛 𝑑𝑛 𝐹𝑛 − 𝐾𝑛𝑖 ∆𝑖
After the DoF has been calculated, the reaction forces are calculated from the original stiffness
matrix as in (3-141)
𝑛 (3-135)
𝑅𝑖 = ∑ 𝐾𝑖𝑘 𝑑𝑘
𝑘
Page 56 af 125
Yet another method which imposes boundary conditions without the error seen in the penalty
method is the use of lagrange multipliers where the stiffness matrix and force vector are kept
unedited and equations with additional unknown parameters, the lagrange multipliers, are added to
impose the constraints as shown in equation (3-136).
[𝐾] [𝐶]𝑇 {𝑑} {𝐹} (3-136)
[ ]{ } = { }
[𝐶] [0] {𝜆} {∆}
Where [𝐶], {𝜆} and {∆} are such that a prescription of the 𝑖’th DoF to ∆𝑖 by the 𝑗’th constraint
includes the added equations (3-143) and (3-144)
𝐶𝑗𝑖 ⋅ 𝑑𝑖 = ∆𝑖 (3-137)
𝑛 (3-138)
∑ 𝐾𝑖𝑘 𝑑𝑘 + 𝐶𝑖𝑗 ⋅ 𝜆𝑗 = 𝐹𝑖
𝑘
In the case of a simple prescription of DoF, 𝐶𝑗𝑖 = 1, but this formulation are also simple for
multipoint constraints or when a boundary condition is a roller on a slope. In this formulation, the
reaction force is easily extracted as 𝐶𝑖𝑗 ⋅ 𝜆𝑗 is the force imposed by the constraint.
To justify the choice of application of boundary condition these three presented methods are
compared to each other. The time it takes to apply boundary conditions, solve for the DoF and
calculate the applied load is recorded ten times for each method at various levels of refinement. To
limit the effect that of varying load from other programs on the computer, all three methods are
performed in the same program immediately after one another and timed by the “tic” and “toc”
commands in matlab. The timing may be different if it were to be run in another programming
language than matlab or if the implementation of programming the methods were done differently.
Also, the timing of each method may depend on the problem being analyzed so this is more of a
benchmarking of this particular implementation for use in this particular analysis than it is an
absolute benchmarking of the three methods.
Figure 27: Calculated runtime versus element count for Figure 28: Runtimes normalized against penalty method
different boundary condition methods runtime versus element count for different boundary condition
methods
Page 57 af 125
At an element count of 𝑛 = 10, there are such a large spread in the computation time in Figure 27
that it is difficult to draw any conclusions but at larger matrix sizes, it seems as shown in Figure 28
that the elimination method is faster than the lagrange method. For this reason and the fact that the
penalty approach was found to be inaccurate, the elimination method is chosen for prescribing DoF
at boundaries.
To further reduce the solving time, the stiffness matrix, which consists mostly of zeros can be saved
in the form where only the non-zero terms are saved. This is done by calling Ks=sparse(K) in
matlab where Ks is the resulting sparse stiffness matrix. By comparing the time it takes to create Ks
and solve the problem against solving with the full K matrix, it is seen that converting to a sparse
matrix is by far faster and that the benefit increases with problem size. This comparison, which the
result of is shown in Figure 31 and Figure 32 is made with elimination method boundary conditions
and with direct solving by mldivide().
Page 58 af 125
Figure 31: Runtimes versus element count for full matrix and Figure 32: Comparison of runtime versus element count of full
sparse matrix matrix and sparse matrix
𝑏 (3-140)
1 2 1
𝑠2 = ∑(𝑦𝑗 − 𝑦̅) = ((𝑦 − 𝑦̅)2 + (𝑦𝑎+1 − 𝑦̅)2 + ⋯ + (𝑦𝑏 − 𝑦̅)2 )
𝑛−1 𝑛−1 𝑎
𝑗=𝑎
The sample standard deviation is the square root of the sample variance 𝑠 = √𝑠 2 . To calculate the
interval of the mean with a conficence 𝛾 when the actual variance 𝜎 2 is unknown, the coefficient c
is determined from (3-141) by looking up the corresponding value in table A9 in appendix 5 of [26].
1 (3-141)
𝐹(𝑐) = (1 + 𝛾)
2
With the coefficient c determined, the coefficient k can now be calculated by (3-142) and then
confidence interval of the mean value is calculated by (3-143)
𝑐⋅𝑠 (3-142)
𝑘=
√𝑛
𝑐𝑜𝑛𝑓𝛾 {𝑦̅ − 𝑘 ≤ 𝜇 < 𝑦̅ + 𝑘} (3-143)
Page 59 af 125
Confidence intervals of variance can also be calculated. By looking up the values in the chi-square
distribution tables given as table A10 in [26] corresponding to equation (3-144) the values 𝑐1 and 𝑐2
are found with 𝑛 − 1 degrees of freedom.
1 1 (3-144)
𝐹(𝑐1 ) = (1 − 𝛾), 𝐹(𝑐2 ) = (1 + 𝛾)
2 2
Using the values of 𝑐1 and 𝑐2 , the limits of the confidence interval are calculated using equation
(3-145) and the confidence interval are then as in (3-146).
(𝑛 − 1)𝑠 2 (𝑛 − 1)𝑠 2 (3-145)
𝑘1 = , 𝑘2 =
𝑐1 𝑐2
𝑐𝑜𝑛𝑓𝛾 = {𝑘2 ≤ 𝜎 2 ≤ 𝑘1 } (3-146)
To assess whether there are differences between two specimen groups, hypothesis testing is
employed to test whether the mean of the two grous are are the same. The two dataset are assumed
to have unknown variances which are not necessarily equal and are tested with the matlab function
ttest2. In the ttest2 function, the value of t is calculated by equation (3-147) and compared to the
value in a t-table of the ‘Student’ distribution.
𝑥̅ − 𝑦̅ (3-147)
𝑡=
𝑠𝑥2 𝑠𝑦2
√
𝑛𝑥 + 𝑛𝑦
The effective degrees of freedom when two samples of unequal variance is being analyzed is
approximated by the Welch-Satterthwaite equation, (3-148), presented in [27].
2 (3-148)
𝑠 2 𝑠𝑦2
(𝑛𝑥 + 𝑛 )
𝑥 𝑦
𝜈′ ≈
𝑆𝑥4 𝑆𝑦4
+
𝑛𝑥2 𝜈𝑥 𝑛𝑦2 𝜈𝑦
Where 𝑠 is the standard deviation, 𝑛 is the sample size and 𝜈 = 𝑛 − 1 is the degrees of freedom.
Page 60 af 125
4 Method
Page 61 af 125
Figure 33: Comparison of foam core materials from Gurit. Specific shear stiffness has the units GPa/(kg/m3) while density has the
units kg/m3.
A trend is showing that the foam cores with higher density have higher specific shear stiffness with
the Gurit S-series being the stiffest, followed by the M-series which also span a higher range of
stiffnesses. Of the Gurit Corecell types, only the M-series are available at the regional reseller at the
time of purchase. In order to avoid having to purchase larger quantities than needed, the M80 foam
is chosen in 10 mm and 25 mm thicknesses while the M130 foam is bought in 30 mm thickness
which is then reduced in thickness to 25 mm by a planer.
To be able to test multiple parameters of bending and shear stiffness, five layups are chosen to be
𝑐 𝑐
tested by 3 point bending and perimeter shear stress test. These are [(0,90)2 , 21 ] , [(0,90)4 , 21 ] ,
𝑠 𝑠
𝑐1 𝑐2 𝑐3
[(0,90)7 , 2 ] , [(0,90)2 , 2 ] and [(0,90)2 , 2 ] where 0 represents layer of unidirectional SE84LV-
𝑠 𝑠 𝑠
HEC prepreg with the fibers oriented along the test direction of the 3PB test, 90 represents a layer
of prepreg at 90 degrees to the test direction of the 3PB test, (0,90)𝑘 represent 𝑘 repetitions of the
layup in the parenthesis, 𝑐1 represents a core of M80 foam with a 25mm thickness, 𝑐2 represents a
core of M80 foam with a 10 mm thickness, 𝑐3 represents a core of M130 foam with a 25mm
thickness and the subscript 𝑠 after the square brackets represent that these are all symmetric
lamiantes.
Page 62 af 125
of the prepregs by this method is accurate within 2 mm but relies heavily on the skills and care of
the person cutting.
The core pieces is cut to shape with a table circular saw or a band saw as wood would be cut. The
M130 pieces is reduced from 30 mm to 25 mm by use of a planer meant for wood which leaves the
surface slightly rougher than as delivered. Cutting the core by this method has proven accurate
within one millimeter.
Before layup, a glass plate is cleaned and prepared either by taping a release film to it or by
applying a chemical release agent. At the manufacturing of the first few specimen in this project,
release film has been used which requires that the release film lies perfectly smooth against the
glass plate as any wrinkles or contaminants between the glass plate and the release film will transfer
to the surface finish of the part. When release agent is used, the plate must be cleaned with alcohol
or acetone and the plate must be scraped by the blade of a utility knife to remove any contaminants
or residue of tackytape, wax or resin that might not dissolve. After being thoroughly cleaned, the
glass plate is treated with a suitable release agent per the manufacturers guides. Here,
EasyComposites Easy-Lease release agent was used where a coat is spread with a cloth or paper,
given 10 seconds to start evaporating and then polished in with a second cloth until it has fully
evaporated. Between each coat, there is a waiting time of 15 minutes and after the final coat a 1
hour waiting time is had before the layup can begin.
For the first few specimen, the glass plate was prepared with a release film and the later test
specimen were made with a glass plate that was prepared with release agent. When only curing one
specimen, the release film method may be the most time-efficient although it still takes some time
to reduce the amount of wrinkles. When a larger amount of specimen are being cured, the release
agent is the most time-efficient as 5 or 6 glass plates can be prepared simultaneously so that the
waiting time in between each coating is utilized. Also, the release film would need to be reapplied
for each new layup while one treatment of release agent can be used for multiple cure cycles with a
single coat after demolding. The glass plate should also have tacky-tape applied along the edges for
the vacuum bag, which requires any release agent along the edge of the plate to be cleaned with a
suitable solvent as the tacky-tape will otherwise fail to stick to the glass when hot. The corners need
special care as these are often leaky. In Figure 35, a schematic view of one method of creating
corners are seen with the glass plate marked in green, the tacky tape along the edges marked in blue
and a supporting piece marked in red laid inside the corner. Often, corners are sealed simply by
putting one piece of tacky-tape over another but this can make a “tunnel” for the air to leak and thus
the method in Figure 35 seems superior for these glass plates.
Page 63 af 125
Figure 34: Glass plate with tacky-tape and one layer of prepreg Figure 35: Schematic view of application of
tacy-tape at a corner
The prepreg sheets are stacked on a prepared glass plate by removing the backing paper but leaving
the backing plastic on to keep the prepreg from distorting. A corner is then lined up with the layer
underneath and is lightly pressed stuck while lining up the edge which is parallel to the fiber
direction and lightly pressing that stuck as well. Holding the prepreg lightly tensioned, it is held
almost parallel to the layer underneath to ensure that the rest of the edges line up. If they do not, the
edge that was lightly pressed on can be made to release by quick yanks. If they do line up, the
prepreg is laid down, moving from one edge and across the fiber direction, rolling it on and pressing
out air bubbles by rolling a rigid roller, like the one seen on Figure 34, at a slight angle to the 90-
degree direction. When the whole sheet is attached, it is further pressed to the layer underneath by
rolling along the fiber direction with the roller while the backing plastic is still attached. To remove
the backing plastic, a piece of tape is attached to the backing plastic at a corner and given a fast
yank to get the backing plastic to release, at which point the plastic can be grabbed and pulled off.
As the 90-degree layer consist of one larger and one smaller sheet, the larger sheet should be placed
first to ensure that the fiber direction is as intended and the smaller sheet is laid after. To spread out
the effect of defects caused by small gaps or overlaps between the two 90 degree layers, the end at
which the small piece is laid is alternating. The core material is laid directly on top of the prepreg
and pressed on by hand as the resin within the prepreg will bond to the surface during curing. The
layup of the top sheets are done directly on top of the core in a similar fashion as for the bottom
layers. The bottom layers can be laid reasonably accurately when there are few layers but errors
accumulate with larger number of layers and it is also difficult to line up the top layers with the
bottom layers due to the core so a small angular error is to be expected. As the layers are aligned
across the length of 500 millimeters, the misalignment is assumed negligible since a 1-degree
misalignment would require one end to be off by 9mm compared to the other which was not
observed during layup.
Page 64 af 125
Figure 36: Laminate within a vacuum bag on a glass plate
After layup, a layer of release film is laid on top of the laminate, which may either be perforated or
unperforated. Using unperforated film will leave the surface slightly more shiny than when using
perforated as more resin is kept in the layup while perforated film will allow a small quantity of
resin to be sucked out while also removing any air that might otherwise be trapped between the
release film and the laminate. After the release film, wooden bars are placed along the edges of the
laminate to act as a mold, so the vacuum bag doesn’t pull the edges and corners down as much.
After this, breather cloth is laid over the laminate and down to the vacuum adapter so that air can be
sucked out through the breather before the vacuum bag is attached. While attaching the vacuum
bag, “ears” are added near the corners of the laminate, as can be seen on Figure 36, to make sure
that sufficient vacuum bag is added to form around the part. After closing the bag, a hole is cut for
the bottom part of the vacuum adapter which is left inside the bag, and the top part is screwed on
and vacuum is pulled. For these parts, the vacuum is kept below 100 mBar.
The composite specimens are cured in an oven where the temperature ramp is at most 2 degrees per
minute up to 100 degrees where it is held for 4 hours, after which the parts has cured.
The pieces of foam material that are meant for tensile testing are machined from a piece of foam
using a CNC mill to a shape according shape type 1A of DS/EN ISO 1798:2008. It is possible to
hotwire this material, but the surface partially melts, and the finish is not suitable for tensile testing
specimen.
The composite parts for material testing are cut from oversized cured laminates using a diamond
band-saw to have a section with approximately even thickness and are then wet sanded on the edges
with increasingly fine sandpaper while held in a holding tool to make sure that the sides are parallel
and without sanding marks. The top face of the specimens is also sanded to make the top and
bottom faces parallel and the compressive test specimen were also sanded on the ends while in a
holding tool. The tensile specimen in the longitudinal direction are manufactured according to ISO
527-5 so that they have a length of 250 mm, a width of 15mm and a thickness of 1mm. The tensile
Page 65 af 125
specimen in the transverse direction differ from the standard by being only 130 mm long and 18
mm wide due to the available prepreg strips that are left over from cutting for the 3PB and PSS test
specimen. Onto the tensile specimen, a steel tab with a chamfered end are bonded to both sides at
both ends using structural adhesive.
The compressive specimens are manufactured somewhat in accordance with ASTM D695 but as
rectangular pieces with a nominal width of 12,7 mm and a length of 80 mm.
Page 66 af 125
Figure 37: Drawing of the 3-point bending test fixture
Figure 39: 3-point bending test setup seen from the end
Figure 38: 3-point bending test setup seen from an angle
After conducting a bending test, the stiffness, yield strength, energy absorption and maximum load
within 12,7 mm must be extracted. The energy absorption is found by summing the force times
Page 67 af 125
distance increment of all the datapoints and the maximum load is simply the largest load in the data
where the displacement is less than the limit. Finding “the” stiffness as a single value of something
that may be nonlinear in nature and finding the yield strength of something that may not exhibit
yielding are challenging tasks.
As an interpretation of these requirements, it is assumed that an interval of data exists in which the
slope of the load-displacement diagram is approximately constant and that the yield strength is at
the end of this interval. Intervals beginning in all datapoints are examined and the mean slope and
confidence interval of the mean slope is calculated for this interval before extending the interval by
one datapoint and calculating the confidence interval of the mean slope of the new longer interval.
If the new interval has a smaller value of 𝑘 as calculated by (3-142), this interval is taken as the best
interval with that starting point and if not, the calculated values for this interval are discarded before
extending the interval further one datapoint. If 20 consecutive extensions of intervals have resulted
in increasing values of 𝑘, then it is assumed that the confidence interval will not improve by
extending the interval further and the start point of the interval is moved one datapoint and the
process is repeated. From this, the interval of approximately constant stiffness can be identified as
the interval with the lowest value of 𝑘 within the region that appears linearly elastic. This method of
determining the stiffness is very computationally expensive but is the chosen method for three-point
bending tests.
The rules require that the reference specimen are two baseline tubes as seen on Figure 40 which are
assumed to deflect only due to the bending and the theoretical stiffnesses of the tubes are assumed
to be as per (4-2).
48𝐸𝐼 ⋅ 2 (4-2)
𝐾𝑆𝐼𝑆 𝑡ℎ =
𝑙3
In the bending of baseline tubes shear deformation, cross-sectional deformations and local impacts
are assumed negligible but during bending, a local deformation of the tubes where the load is
applied quickly evolved as seen on Figure 41 and the cross-sections were also seen to deform into
an oval shape. As the deflection of the tubes is measured by the contact on the top of the tubes,
these deformations cannot be neglected as they introduce flexibilities which are counted towards the
test rig when the baseline tubes are used as reference specimen. Furthermore, the load of the tubes
onto the supports are much more concentrated than when plates are tested, which will naturally
result in the test rig being less stiff when testing tubes than when testing plates.
Page 68 af 125
Figure 40: Baseline tubes in 3-point bending Figure 41: Local deformation of baseline tubes
The load-displacement graph of these baseline tubes are shown in Figure 42 where it seems to have
a long approximately linear region but looking at the stiffness graph in Figure 43, it can be seen that
these tubes do not have any distinct region of constant stiffness as the stiffness is initially rising as
dominated by contact nonlinearity before dropping due to the deformations of the cross-section and
plastic nonlinearities.
𝑁
The measured stiffness of the baseline side impact structure tubes is 𝐾𝑡𝑒𝑠𝑡 𝑆𝐼𝑆 = 1911,9 [𝑚𝑚],
calculated as the mean of the interval from 0,5800 mm to 2,0558 mm and the yield load at 2,0558
mm is 3767,7 Newton. The absorbed energy after half an inch of displacement is 85,47 Joule and
the maximum loading is 8735,6 Newton. With this measured stiffness of the tube and the theoretical
stiffness calculated per (4-2), the stiffness of the testing rig is calculated to be 𝐾𝑟𝑖𝑔 𝑆𝐸𝑆 =
𝑁
7089,4 [𝑚𝑚].
Page 69 af 125
Figure 42: Baseline tubes 3PB load-displacement graph Figure 43: Baseline tubes 3PB stiffness graph
For the purpose of comparing models of laminates to experimental result, a more accurate estimate
of the testing rig stiffness is needed. To determine this, a steel plate the same size as the laminates is
used. Due to the availability of materials at the time of testing, this was done by laying a 175x500
plate alongside a 100x500 plate, both with a thickness of 10 millimeters. The measured stiffness is
𝑁
2988,8 [𝑚𝑚], calculated as a mean between 3,324 mm and 6,477 mm. Assuming that the anticlastic
curvature is not restricted due to the width of the plates and assuming that the through-thickness
compression is negligible, the theoretical stiffness can be calculated using the Timoshenko beam
N
theory which gives a theoretical stiffness of 𝐾𝑡𝑖𝑚 𝑡ℎ = 3430,8 [mm] corresponding to a rig stiffness
N
of 𝐾𝑟𝑖𝑔 𝑎𝑐𝑡𝑢𝑎𝑙 = 23199 [mm].
Figure 44: Load-displacement graph of 3PB of 10 mm steel Figure 45: Stiffness graph of 3PB of 10 mm steel plate
plate
When both the actual testing rig stiffness and the stiffness as calculated by the SES are known, a
value for the SES can be calculated from simulated data by using the actual rig stiffness to obtain an
equivalent “measured” value and using the SES rig stiffness to calculate the laminate stiffness
Page 70 af 125
which the SES would use to evaluate compliance as by (4-3). Since the actual stiffness of the testing
rig is much higher than that which is calculated from the baseline tubes as required by the rules, the
stiffness requirement for laminates are effectively reduced.
−1 −1 −1 −1 (4-3)
𝐾𝑙𝑎𝑚 𝑆𝐸𝑆 = (𝐾𝑙𝑎𝑚 𝐹𝐸𝑀 + 𝐾𝑟𝑖𝑔 𝑎𝑐𝑡𝑢𝑎𝑙 − 𝐾𝑟𝑖𝑔 𝑆𝐸𝑆 )
−1 −1
This does have the implication that there’s a singularity when 𝐾𝑙𝑎𝑚 𝐹𝐸𝑀 + 𝐾𝑟𝑖𝑔 𝑎𝑐𝑡𝑢𝑎𝑙 equals
−1
𝐾𝑟𝑖𝑔 𝑆𝐸𝑆 which for these stiffnesses exist when the laminate stiffness is a bit over 10.000 Newton
per millimeter as seen on Figure 46. This stiffness is far beyond what would be useful for a formula
student car as, in order to obtain it, it would be far too heavy to be competitive.
Page 71 af 125
section and the transverse deflection. At the free ends, the transverse displacement is prescribed
zero to make a simply supported boundary condition while at the shared node at the center, the
transverse displacement is prescribed a unit displacement. The boundary conditions are applied by
the elimination method. The stiffness of the laminate is calculated from the force required to apply
the prescribed unit displacement at the center node. The stresses in the laminate and in the core are
calculated and are extrapolated until failure due to the assumption of linear elasticity. If a failure is
calculated to happen before a displacement of 12,7 mm, the strength is taken to be the load when
the failure occurs, and the energy is calculated depending on whether a lamina or the core fails first.
If a lamina is calculated to fail before the core, it is assumed that all stiffness is lost and the only
energy that is absorbed is that of elastic deformation up until first ply failure. If the core is
calculated to reach a Von Mises equivalent stress equal to the level where 0,25% plastic strain is
observed in tensile tests, the beam is assumed to deform until 12,7 mm under constant load from
then on.
The failure of lamina are calculated per the max stress criterion as the criterion is linearly related to
the stresses, which makes it simpler to extrapolate until failure. Using the max stress criterion rather
than the more sophisticated criteria is regarded as having little impact to the accuracy of the
prediction of failure as all failures in the experimental part of this project is observed to happen
after both geometrical and material nonlinearities have become unneglectable. Thus, the stresses
and energy absorptions calculated by these linear models should not be trusted more than as a rough
approximation.
Model 2: Beam model as skin, 2D plane elements as core
In this second model, the skin is modeled by Timoshenko beam elements with both vertical and
horizontal displacement degrees of freedom at the nodes along with rotation of the cross-section.
The core is modeled by QM6 plane elements and the beam is modeled in full, including 50 mm
overhang over the supports in each end. At 50 mm and 450 mm from the left end of the beam, the
nodes on the lower surface are prescribed a zero vertical displacement and at 250 mm, the node on
the upper surface are prescribed a zero horizontal displacement and a unit vertical displacement.
The calculation of the shear coefficient for the beam elements that are used for the skins does not
account for the shear traction being non-zero on the side of the beam element that faces the core and
thus, the shear stiffness of the skin is incorrectly calculated too flexible. The shear stiffness of the
beam as a whole is however mostly determined by the core and as seen on Figure 47, the shear
stresses are approximately constant through the thickness of the core, which is the expected result
when it is laminated with relatively stiff skins. The shear stress in the core is seen to drop with
increasing amounts of layers in the skin which means that this model does calculate the shear stress
as incorrectly as the sandwich model but compared to model 1, the shear stresses in the core are
calculated about 10% higher here.
Page 72 af 125
Figure 47: Shear stress in core of beam with a 2D plane element core. The two green elements in the yellow part are a plotting error
caused by all of the nodes having the same calculated shear stress.
The failure of lamina and yielding of the core are treated similarly as with model 1. To evaluate
whether the core yields, the Von Mises equivalent stresses are calculated across all elements of the
core as seen in Figure 48. The linear-elastic analysis of 2D elements with a point-load includes a
stress singularity which means that the stresses calculated near the loads will never be mesh-
independent although the beam elements for the skin will help distribute the load a bit. As a
consequence of this, the elements in the core near the nodes with prescribed DoF are excluded from
the calculation of stresses and it is assumed that 5 mm or 1 element (whichever is further) away, the
singularity is sufficiently dispersed.
Page 73 af 125
Figure 48: Von Mises equivalent stresses in the core of a beam with 2D plane elements for core
To find the optimum layup for a part of the car, for-loops are used to simulate all possible layup
combinations with the M-series core foams, at chosen increments of core thickness and up to 6-
layer thick skins. Simulating all possible combinations of laminate angles is not considered feasible,
even if the angles are chosen in increments as for 𝑛 layers with 𝑣 possible increments of angles,
there’s 𝑣 𝑛 possible combinations, making for millions of possible layups. Instead, an assumption is
made of the lamina angles in the optimum layup. It is assumed that the highest legal amount of
fibers is in the loaded direction, so the outermost layers are placed at 0 degrees, while the remaining
are placed at alternating +/- 10 degrees.
After calculation of the stiffness, strength and energy absorption, the laminates with sufficient
results are ranked according to their mass so that the laminate with the lowest weight can be chosen
for further analysis.
To do the optimization, the stiffnesses of the core material from the data sheets are used and the
limit for yielding is taken as the tensile strength from the data sheet as the observed 0,25% proof
strengths were above these values for the tested materials.
4.3.3 Simulation method – ANSYS
The plates are simulated by modeling a quarter plate and thus taking advantage of the symmetry
planes by use of symmetry regions in ANSYS. Two models are used, analogous to the beam models
described above.
Page 74 af 125
Figure 49: Boundary conditions of a shell element model (full plate shown)
The first model is using shell elements with a defined layered section as ANSYS calculates the
laminate properties as an equivalent single layer. The plate is modeled without any overhang past
the supports, with a displacement boundary condition at the supports and a remote displacement
boundary condition as the applied load.
In this model, geometrical nonlinearities are included and composite stiffness reduction is included
based on the Hashin failure criteria. The stiffness reduction of a failed lamina is set to 75% as a full
reduction of stiffness has shown to lead to unstable results with very early failure. Yielding of the
core is not included in this model as enabling the setting for layered sections causes ANSYS to
crash.
Figure 50: Full model boundary conditions. Quarter plate shown. Symmetry boundary conditions are on the faces and edges to
simulate a full plate.
The second model in ANSYS is the most sophisticated in this project and is referred to as the full
model. In this, the core is modeled by 8 solid-shell elements through the thickness while the skins
are modeled as equivalent single layers by shell elements. The elements are approximately square
with a width of 4,8mm. A displacement boundary condition is applied to a line on the bottom
surface and a flexible remote displacement is applied to the top surface to simulate the physical
Page 75 af 125
boundary conditions while symmetry regions are applied to the surfaces and edges to enable a
quarter plate to be modeled.
Overhang beyond the supports is included, geometrical nonlinearities are included, laminate
damage of the skins is included by a stiffness reduction of 75% based on the Hashin failure criteria
and yielding of the core is included in the model. The yielding of the core is modeled as bilinear as
attempts to apply a multilinear material model has not been successful within the time period of the
project.
The simulated data of most of the simulations with the full model show a nonlinear stiffening effect
before yielding that is not seen in the experimental results and thus the stiffness is calculated as the
average over the first millimeter of displacement.
The first peak, which is to be used for extrapolating required bracket size is defined as the largest
force before the top skin is fully penetrated, regardless of any damage prior to this peak as shown
on Figure 52. For the sake of the SES, the failure is assumed to be due to pure shear along the
perimeter of the punch, so if the upper skin has a thickness of 𝑡1 which failed at a load 𝑃1 , then the
shear strength which is considered the same in all directions is calculated per equation (4-4).
Page 76 af 125
𝑃1 (4-4)
𝑆𝑠 𝑆𝐸𝑆 =
𝑡1 ⋅ 25 ⋅ 𝜋
Page 77 af 125
Figure 53: ANSYS model of PSS test with solid-shell top skins
Page 78 af 125
The compression of the specimens are measured by the translation of the crosshead of the test
machine and to account for the flexibility of the test machine, the tested specimen and the test
machine are assumed to be two ideal springs in series. To find the stiffness of the test machine, steel
specimen are tested and the stiffness of the machine is calculated similarly as done for the 3 point
bending tests by assuming the tested specimen and the machine to each be an ideal spring in series.
Four steel specimens are tested of which two are discarded due to the ends being not sufficiently
flat and the stiffnesses of the other two are found by searching for a region of approximately
constant stiffness, similarly as done for 3-point bending specimen.
Tested stiffness Theoretical stiffness Test machine stiffness
N/mm N/mm N/mm
Specimen 3 20.009 +/- 115 65.774 28.757 +/- 239
Specimen 4 19.420 +/- 125 65.574 27.592 +/- 252
The stiffness of the machine is not uniquely determined by these two specimens but is assumed to
be the mean of the calculated stiffness and thus a machine stiffness of 28.174 N/mm is used to
correct the calculated stiffnesses of the composite specimen. This is done by calculating the
displacement of the idealized spring that is the machine and subtracting it from each data point.
Page 79 af 125
Three groups of test specimens are manufactured. One group is 3-layer specimens with a nominal
thickness of 1mm cut from excess length of the tensile test specimens, cured with perforated release
film and breather which absorbs some excess resin to improve the fiber-volume ratio. The two other
groups are made from a single layer of prepreg with a nominal thickness of 0,3mm. One group is
cured with perforated release film like the thicker specimens and the other group was cured with
unperforated release film to evaluate the influence of absorbing excess resin.
After cutting, the edges of the specimen are sanded clean and then the specimens are weighed to
find the mass 𝓂𝐿 . The sides are then sealed with clear nail polish and left to dry before the test
specimen are then weighed again to get 𝓂𝑑𝑟𝑦 and then submerged into water to be weighed yet
again to get 𝓂𝑤𝑒𝑡 .
The specimen are then burned in an oven at temperatures between 500 and 600 degrees Celsius to
evaporate the resin, leaving only the fiber behind.
Burn-off tests are often not recommended for carbon or polymer fiber as the fiber can also burn off
but the timing and temperatures used here were taken from a test procedure created by the company
Space Composite Structures Denmark (SCSDK), with whom the university shares the composite
lab, as these have tuned this procedure to yield accurate results. As the test procedure is the
intellectual property of SCSDK, the timings and temperature used are not disclosed but worth
noting is that due to the small weight of the tested specimens, all of them are in the same weight
bracket and burned for the same duration despite one group of specimens weighing about 3 times
that of the others. When burn-off tests are used despite the risk of fiber burn off, it is because it is a
simple, cheap and relatively fast method of measuring the fiber volume fraction. Other methods
include x-ray computed tomography and the matrix digestion method where the matrix is
chemically eroded or dissolved.
From the dry and wet weights including nail polish and the density of water, the volumes of the
specimens are calculated as the difference in weight divided by the density of water as given in
equation (4-5).
𝒱𝑑𝑟𝑦 − 𝒱𝑤𝑒𝑡 (4-5)
𝑉0 =
𝜌𝑤𝑎𝑡𝑒𝑟
The density of the laminate is then calculated by (4-6) from the specimen volume and the weight
excluding nail polish, from the assumption that the volume of nail polish is negligible.
𝓂𝐿 (4-6)
𝜌𝐿 =
𝒱0
Taking the density of the fiber, 𝜌𝑓 , and matrix 𝜌𝑚 from manufacturers data sheets, the fiber volume
fraction is calculated by (4-7) from the weight of the remaining fiber after burning.
𝓂𝑓 𝜌𝐿 (4-7)
𝓋𝑓 = ⋅
𝓂𝐿 𝜌𝑓
The fiber and matrix weight fractions are then calculated from (4-8) and (4-9) respectively.
𝓂𝑓 (4-8)
𝓌𝑓 =
𝓂𝐿
Page 80 af 125
𝓂𝐿 − 𝓂𝑓 (4-9)
𝓌𝑚 =
𝓂𝐿
The void (volume) ratio is calculated by (4-10) using the calculated density and the theoretical
density as calculated by (4-11) with the calculated weight fractions and the material densities from
the data sheets.
𝓌𝑟 𝓌𝑓 (𝜌𝐿,𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 − 𝜌𝐿 ) (4-10)
𝓋𝑣 = 1 − 𝜌𝐿 ⋅ ( + )=
𝜌𝑟 𝜌𝑓 𝜌𝐿,𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙
1 (4-11)
𝜌𝐿,𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 = 𝓌 𝓌𝑚
𝑓
𝜌𝑓 + 𝜌𝑚
When the fiber and void volume fractions have been calculated, the matlab function ttest2 is used to
perform hypothesis testing on whether the means of two dataset are equal.
Page 81 af 125
5 Results
In this section, the main findings of the report are presented. The first result to be presented is the
thickness of the cured laminas. After testing, specimen 1 is cut with a diamond bladed saw to reveal
cross-sections of the laminate which are then looked at under a microscope. Using a digital camera
on the microscope and the accompanying software, digital rulers are placed on the picture and it is
seen that the laminates are consistently very close to 0,3mm in thickness. This also matches that the
total thickness of a 4 layer skin is measured with a caliper to be approximately 1,2-1,25mm.
Figure 55: Cut surface showing the cross-section of a skin consisting of 4 layers of CFRP in alternating 0 and 90-degree directions
5.1 Defects
Defects such as voids and contaminants exist within laminated parts and although care can be taken
to minimize these defects, they cannot be avoided completely.
5.1.1 Contaminants
The layup has not been done in a clean room but in a composite working lab, which has introduced
some contaminants. To assess the amount of contaminants, the first 3-point bending test specimen
has been examined on the side facing the glass plate by shining a bright light onto the surface and
circling and counting all visible contaminants as on Figure 56. This does not tell about the amount
of contaminants within the laminate but it may give insight to the importance of cleanliness.
Page 82 af 125
By circling and counting, 148 pieces of solid contamination has been identified, ranging from
millimeter long fibers to centimeters long fibers. The most abundant type of contaminant is very
short fibers of very low diameter which is curled up so much that they are assumed to be very
flexible and thus probably are fibers from either clothing or breather material. Other contaminants
appear to be human hairs and also what appears to be glass fibers can be seen. As the layup of this
specimen has been done on release film rather than on a glass plate with release agent, it is possible
that some surface imperfections due to wrinkles in the release film have been misidentified as long
fibers. This possible misidentification would be a small amount of the counted defects since the
imperfections due to wrinkling are only seen along the fiber direction and often looks different from
the contaminants as seen on Figure 57.
Figure 56: Contaminants catching the light when seen a bit Figure 57: Contaminants in circles. Imprints of wrinkles of the
from the side release film seen as horizontal lies that catches the light
5.1.2 Voids
To examine for voids, the sides of the tensile test specimens which has been wet sanded with fine
grade paper are being looked at under a microscope. Here, two sizes of voids are seen. Small voids
that are common and are likely to not be possible to prevent and rarer large voids which may be
possible to prevent by exercising extreme care to avoid air traps during layups. The small voids
were seen to have a preference for one direction, such that parallel to the fibers, the length of the
voids are measured to be up to 2mm with 1mm being the most common size. When inspecting by
eye, they do seem to be longer, but when looking under the microscope, it seems likely that this is
due to large number of voids in the interface between layers and also due to the sanding being
parallel to the fibers. Two voids, marked with rulers, are visible on Figure 58 although the
resolution and lighting of the image makes the cracks stand out less than when they are seen
through the eye-piece. When looking onto a plane perpendicular to the fibers, the voids seem
circular with a diameter of 10-20 micrometers and are seen to exist in large numbers in the
boundary between layers as seen on Figure 59 and Figure 60.
Page 83 af 125
Figure 58: Voids seen when looking parallel to the fibers. The rulers are 2mm and 0,5mm long.
Figure 59: Voids photographed with microscope built-in camera, looking onto the ends of the fibers. Voids marked from the left with
three 20 micrometers voids and two 10 micrometer voids.
Figure 60: Voids looking onto the ends of the fibers, photographed with an external camera through the eye-piece. Note that the
uneven top surface is filled with fibers and are not just pooled resin.
Page 84 af 125
The larger voids, where a large portion of air is trapped are rare with observations on only two of
the 25 tensile test specimen and one observation on 15 compressive test specimens. The large voids
seem to be about 4mm long, 0,2mm in the thickness direction and with unknown depth in the
width-direction of the specimen. While it does seem that the thickness-dimension is much smaller
than the other dimensions, it is not clear whether there are any preferences for the extension in the
two in-plane directions. On one of the specimens which had large voids, a larger than usual number
of smaller voids was also observed, which may indicate that this specimen had not been rolled with
sufficient pressure during the lay-up. This is shown on Figure 62.
Figure 62: Large void seen when looking onto the 90-degree direction.
Page 85 af 125
micro voids and a region with large voids was among those that had large-scale surface unevenness,
suggesting that pressing hard with the roller does not alleviate these problems. Large-scale
unevenness is also observed in the 0-degree tensile test specimens, where the valleys are once again
parallel to the fiber direction. Large-scale unevenness is also seen to some extent on the top-surface
of the 3 point bending specimen as can be seen on Figure 63.
Because the depth of this large-scale unevenness is so close to the layer thickness, the whole top
layer was sanded off of some tensile test specimens to level the surface, exposing the interface
between the two topmost layers, which is more porous than the layers themselves.
In the data sheets, the compressive properties in the transverse direction are missing and so are the
in-plane shear properties. In [28], tests have been performed to determine the in-plane shear
strength through the double-notch test method, which is found to be 58,9 [𝑀𝑃𝑎] with a standard
Page 86 af 125
cure in an oven. As the interlaminar shear strength in Table 8 are found from short beam bending
tests under a homogeneity assumption using the theory for long slender beams, this cannot be used
for design [2]. Instead, the in-plane shear strength from [28] is assumed to also be the through-
thickness shear strength.
Tensile Tensile Compressive Compressive Shear Shear Shear
strength modulus strength modulus strength modulus elongation
at break
Corecell 1,62 72 GPa 1,02 MPa 71 GPa 1,09 29 GPa 58%
M80 MPa MPa
Corecell 2,85 176 GPa 2,31 MPa 170 GPa 1,98 59 GPa 43%
M130 MPa MPa
Table 9: Material properties for Gurit Corecell M80 and M130 from data sheet
The material properties derived from tests are given in Table 10 and Table 11.
Fiber volume Tensile Tensile Compressive Compressive
fraction strength modulus strength modulus
The shear moduli in Table 11 are calculated from the tensile modulus, assuming isotropy using the
poissons ratio calculated from the tensile and shear moduli of Table 9. The strengths for the foam
are the true stress, assuming constant volume.
Tensile Yield 0,25% proof Ultimate Shear
Modulus strength strength strength modulus
Corecell M80 92,19 GPa 1,252 MPa 1,974 MPa 2,652 MPa 37,12 GPa
Corecell 176,05 GPa 1,826 MPa 3,759 MPa 4,705 MPa 59,02 GPa
M130
Table 11: Material properties of Gurit Corecell M80 and M130 derived from tests
Page 87 af 125
3 layer specimen 1 layer 1 layer
Perforated release film
Perforated release film Unperforated release film
By plotting the void volume fraction against the fiber volume fraction on Figure 64, it is noted that
the 3-layer specimen are much closer grouped than the 1-layer specimen and also, there is an outlier
among the 1-layer group with unperforated release film.
The main possible errors in the method of the burn-off test are incomplete burning of matrix,
excessive burning of fiber, incorrect volume calculation due to nonnegligible amount of nail polish,
incorrect volume calculation due to unnoticed nonnegligible air bubbles stuck to the surfaces of the
specimen and discrepancies between densities of the actual fiber and matrix and the values listed in
the data sheets. While all of the specimen are found to be fibrous after burning, indicating that the
resin is properly burned off, it is noted that many of the 3 layer specimen has soot left on the sides
of the pot while none of the 1 layer specimen has. This suggests that the 3 layer specimen could
have unburned resin included in the weighing and/or that the 1 layer specimen could have excessive
fiber burn off.
Page 88 af 125
Figure 64: Void volume fraction plotted against fiber volume fraction
To investigate further into the possible errors, the weight of the nail polish and the calculated
densities are plotted against the fiber volume fraction in Figure 65 and Figure 66. It is seen that the
weight of the added nail polish are consistent among all specimen while the calculated densities of
the 1-layer specimen are much more spread out than for the 3-layer specimen and that the outlier
among the 1-layer specimen is also an outlier in terms of calculated density.
Figure 65: Nail polish weight plotted against fiber volume Figure 66: Specimen density plotted against fiber volume
fraction fraction
Page 89 af 125
As multiple of the specimen within each group are cut from the same cured laminate strips, the lack
of clustering among the calculated densities of the 1-layer specimen indicates that errors exist in the
weighing of the specimen. This error may then have to do with the ratio of surface to volume as
submerging the specimen into water has been difficult due to surface tension and some specimen
has had to be submerged multiple times to get rid of air bubbles clinging to the surface.
To test if the thickness of lamination or the choice of perforated versus unperforated release film
can be said to influence the fiber and void volume fractions despite the errors, hypothesis testing is
applied by testing if the means of two dataset are equal or if the hypothesis can be rejected at a 5%
significance level. In Table 13, the result of these hypothesis test are given where the difference of
means are given when the hypothesis is rejected and the p-value are given when the hypothesis is
accepted. The outlier of the group with unperforated release film has been excluded from this
hypothesis testing but the conclusions are the same, just with slightly different p-values. As is seen,
the impact of laminate thickness is not what would be expected if more voids exist in the
interlaminar regions. Also, the choice of release film cannot be proven to impact the fiber and void
volumes, although these p-values are very low, casting doubt on the conclusions.
3 layer specimen 3 layer specimen 1 layer
Perforated release film
Perforated release film Perforated release film
&
& &
1 layer
1 layer 1 layer
Perforated release film Unperforated release film
Unperforated release film
𝑣𝑓 being equal Rejected Rejected Accepted
𝑥̅ − 𝑦̅ = [0,04 ; 0,056] 𝑥̅ − 𝑦̅ = [0,045 ; 0,076] 𝑝 = 0,108
As it is not known for sure which errors exist in these tests, the exact fiber volume fraction or void
content cannot be determined uniquely. An approximate upper bound on the fiber volume fraction
and void ratio of the 3 layer specimen can be established from the data as a lower weight after
burning would reduce both of these calculated values. A lower bound on the hypothesizing that the
3 layer specimen had no voids and reducing the after burn weight to that which would result in no
calculated voids, a lower bound of the fiber volume fraction can be placed in the interval 0,52-0,54
which matches nicely with the 55% fiber volume stated in the data sheet as the layup is unlikely to
be completely free of voids.
5.2.2 Composite tensile test – longitudinal direction
The tensile tests in the longitudinal direction are tested until failure with the stress-strain diagram
shown in Figure 67. When individual fibers fractures during loading, it happens with such violence
that the extensiometer is seen to jump which leads to discontinuities in the strain data. While the
data could have been cleaned up by assuming continuity across these jumps, no attempt to do so
Page 90 af 125
have been conducted in this project, which leads to only the ultimate stress and the stiffness before
the discontinuities being reliable data from these tests. The failure strain is not noted as it would
rely heavily on assumptions and interpretation of the data.
The stiffness data is very noisy, possibly due to an overly slow testing speed. The stiffness plot
shown on Figure 68, where the stiffness is calculated as an average over 100 datapoints indicate a
possible strain stiffening effect which is not accounted for. In this plot, instantaneous jumps in
stiffness, as seen by the vertical lines that extend beyond the range of the plot, are related to fiber
fracture. It is seen that the specimen are free of damage in the first 0,0045 % of strain.
Page 91 af 125
Figure 68: Stiffness plot of SE84LV-HEC, longitudinal direction
The stiffness of the specimen are calculated per iso 527, where the stiffness is assumed to be the
slope of a straight line connecting the stress at 0,0005 % strain and 0,0025% strain. The stiffness
and strength of all specimen and the mean values are given in Table 14 along with the 95%
confidence interval of the stiffnesses and of the mean stiffness and strength. The stiffness is found
to be higher than that stated in the data sheet while the strength is found to be lower.
Specimen Stiffness per iso Ultimate strength
no standard [GPa] [Mpa]
1 155,29 +/- 1,08 2024
2 153,95 +/- 2,88 1960
3 155,34 +/- 0,93 1896
4 150,88 +/- 0,26 1944
5 145,31 +/- 0,32 2056
6 145,83 +/- 0,61 1801
7 141,54 +/- 0,31 2066
8 142,20 +/- 0,15 2104
9 146,42 +/- 0,54 1966
10 143,97 +/- 0,34 1825
11 155,00 +/- 0,54 1980
Page 92 af 125
12 146,99 +/- 0,24 2125
13 136,06 +/- 0,22 1963
14 148,02 +/- 0,21 1989
15 150,00 +/- 0,63 2167
Mean and 147,72 +/- 3,17 1991 +/- 57
95th
percentile
interval
Table 14: Stiffness and strength of SE84LV-HEC, longitudinal direction
Page 93 af 125
specimens are considered invalid and it can be concluded that tabs and strain gages must be used to
accurately conduct compressive test of composites in the longitudinal direction.
Page 94 af 125
Figure 70: Stress-strain diagram of SE84LV-HEC in compression, transverse direction
Page 95 af 125
Specimen Stiffness per statistical Ultimate strength
no method [GPa] [Mpa]
(strain interval in
parentheses)
1 7,871 +/- 0,0,049 102,8
(0,0048 – 0,0081)
2 8,003 +/- 0,043 73,1
(0,0048 – 0,0099)
3 8,711 +/- 0,038 111,3
(0,0047 – 0,0112)
4 8,744 +/- 0,039 128,9
(0,0046 – 0,0126)
5 8,731 +/- 0,040 102,4
(0,0048 – 0,0112)
Combined 8,412 +/- 0,542 103,7 +/- 25,1
mean and
95th
percentile
interval
Table 16: Stiffness and strength of SE84LV-HEC in compression, transverse direction
Inspection of the specimen after failure shows a fracture at an angle to the loaded direction as
shown in Figure 72.
Page 96 af 125
to take effect, thus making it difficult to identify a region of approximately constant stiffness. By
applying the constant volume assumption to plot the true stress against the logarithmic strain, the
stiffness is initially approximately constant for a longer strain interval as seen by the zoomed-in
view of Figure 75 and Figure 76.
Figure 73: Engineering stress-strain diagram for Corecell Figure 74: Stiffness diagram for Corecell M80 tensile test,
M80 tensile test engineering stress/strain
Figure 75: Zoomed in view of the stiffness diagram of M80 Figure 76: Zoomed in view of the stiffness diagram of M80
from engineering stress/strain from true stress/strain
The stiffness is then calculated from the true stress/strain as this stiffness should be equal to the
stiffness with engineering stress/strain at infinitesimal strain levels. The calculated stiffnesses and
strengths are presented in Table 17. To calculate the stiffness, the statistical method is used to
identify the end of the linear region. As this method will place the end of the linear region very
early, using the end of the linear region as the yield strength is not usable for building a bilinear
hardening model. Instead, the stress at which there are 0,25% plastic strain are used.
Twelve specimen has been tested but the first is invalidated due to errors with the extensiometer.
Page 97 af 125
Stiffness per Yield strength, 0,25% Proof Ultimate
statistical method end of linear strength strength
[GPa] region
[MPa] True stress
(strain interval in [MPa]
[MPa]
parentheses)
Specimen 2 89,20 +/-1,07 1,251 1,906 2,551
(0 – 0,0140)
Specimen 3 92,80 +/-1,41 1,420 2,007 2,633
(0 – 0,0152)
Specimen 4 90,71 +/-1,13 1,203 1,944 2,615
(0 – 0,0132)
Specimen 5 92,36 +/-1,23 1,115 1,913 2,530
(0 – 0,0120)
Specimen 6 88,48 +/-1,23 1,480 1,993 2,674
(0 – 0,0166)
Specimen 7 94,87 +/-1,19 1,124 1,975 2,634
(0 – 0,0119)
Specimen 8 92,80 +/- 1,20 1,485 2,034 2,743
(0 – 0,0160)
Specimen 9 96,59 +/- 1,20 0,828 1,969 2,712
(0 – 0,0085)
Specimen 10 94,70 +/- 1,20 1,212 1,976 2,731
(0 – 0,0128)
Specimen 11 91,14 +/- 1,20 1,427 2,005 2,654
(0 – 0,0156)
Specimen 12 90,41 +/- 1,17 1,224 1,990 2,696
(0,0037 – 0,0134)
Mean 92,19 +/- 1,68 1,252 +/-0,132 1,974 +/-0,027 2,652 +/-0,047
Table 17: Stiffnesses and strengths of Gurit Corecell M80 foam
Assuming the material to be isotropic, a poissons ratio of 0,241 are calculated from the data sheet
data which are then used to calculate a shear stiffness of 37,12 MPa using the mean stiffness from
these tests.
To form the bilinear model, the tangent modulus are now to be found. One method could be to draw
a straight line from the yield stress to the highest true stress for each curve in Figure 77 but this
would make the calculated tangent modulus stiffer for the specimen that failed earlier. Instead,
Page 98 af 125
Ludwik’s model in equation (3-131) is fitted to the stress versus plastic strain shown in Figure 78 in
order to extrapolate the true stress-strain curve so that the end-point strain of the tangent modulus is
the same for all specimen. For each specimen, (3-131) is fitted to an interval starting at the 0,25%
proof stress and ending at the highest stress value. This model is generally not a perfect fit as seen
on Figure 79 which may in part be explained by the stress/plastic strain graph swinging slightly into
negative strains due to the elastic modulus being calculated lower than the actual modulus at
infinitesimal strains.
Figure 77: True stress versus logarithmic strain, M80 tensile Figure 78: True stress versus logarithmic plastic strain, M80
test tensile test
Figure 79: Fit of Ludwik's model onto specimen 2 of M80 tensile test
The coefficients for Ludwik’s model and the calculated tangent moduli 𝐾𝑡 are given in Table 18.
The end point for calculating the tangent modulus is chosen to be at 0,7 strain.
Page 99 af 125
Specimen 2 3 4 5 6 7 8 9 10 11 12
no.
𝒦 3,426 3,529 3,466 3,536 3,421 3,622 3,521 3,323 3,579 3,493 3,528
𝓃 0,094 0,091 0,092 0,098 0,087 0,096 0,088 0,096 0,094 0,089 0,091
𝜎𝑌 , 𝑀𝑃𝑎 0 0 0 0 0 0 0 0 0 0 0
𝐾𝑡 , 𝑀𝑃𝑎 8,726 8,833 8,877 9,236 8,354 9,365 8,676 9,381 9,141 8,649 8,837
Table 18: Fitting coefficients for Ludwik's model and bilinear tangent modulus for M80 tensile test
Figure 80: Comparison of exprimental and simulated data of Figure 81: Comparison of exprimental and simulated data of
3PB of M80 foam using bilinear isotropic hardening with 3PB of M80 foam using bilinear isotropic hardening with
experimentally derived material data, Load-Displacement graph experimentally derived material data, Stiffness graph
To get the bilinear model to yield the same force at the end of 50 mm displacement as the
experimental results, the tangent modulus must be set to practically zero but with 10%
overprediction of the load at 50 mm deflection, a tangent modulus of 6 can be used.
To make a multilinear model, stress levels are taken from the graphs in Figure 78 until the
maximum load, after which the extrapolated values are taken. The mean values are given in Table
19.
Within this project, attempts at applying this multilinear model to the 3-point bending of a 25mm
thick M80 foam plate has been unsuccessful as ANSYS fails to converge due to distorted elements,
even when the mesh is very fine.
The following in this subsection is added in V2.0 of this report, the 11th of august 2020
An error in the data entry of ANSYS has been located, as the data in Table 19 was entered as-is in
megapascals while ANSYS per default assumes the stresses to be Pascals. When this is corrected,
the simulation does not fail to converge but using the experimentally derived values for stress
versus plastic strain, the simulation with multilinear isotropic hardening fails to capture the
nonlinear behavior properly as seen on Figure 82. Multiple simulated results are shown, which are
different refinement levels of mesh and timestepping, which does not suggest that the discrepancy is
due to numerical errors. Lowering the stress level at zero plastic strain has also been attempted but
yields practically no difference to the results. To control that the issue is not in the symmetry
boundary conditions, a simulation of a full plate has also been conducted, which gives similar
results to the quarter model simulations.
Figure 82: Load-displacement graph of 3PB of M80 foam. Comparison of experimental data and simulated with multilinear
plasticity model
Figure 83: True stress-strain diagram of M130 tensile tests Figure 84: Stiffness diagram of M130 tensile test, from true
stress and logarithmic strain
The coefficients for Ludwik’s model and the calculated tangent moduli 𝐾𝑡 are given in Table 21.
The end point for calculating the tangent modulus is chosen to be at 0,7 strain.
Specimen 1 2 3 4 5 6 7 8 9
no.
𝒦 5,6960 5,8092 6,0069 5,8361 5,7203 5,8888 5,9094 5,9178 5,7472
𝓃 0,0616 0,0670 0,0682 0,0711 0,0656 0,0682 0,0707 0,0754 0,0659
𝜎𝑌 , 𝑀𝑃𝑎 0 0 0 0 0 0 0 0 0
𝐾𝑡 , 𝑀𝑃𝑎 11,299 12,208 12,778 12,737 11,816 12,516 12,866 13,396 11,962
Table 21: Coefficients of Ludwik's model and the calculated tangent modulus for M130 foam tensile tests
The following in this subsection is added in V2.0 of this report, the 11th of august 2020
As with the M80 foam, a data entry error lead to nonconvergence of multilinear plasticity models in
ANSYS using the data in Table 22. With this error corrected, the simulation does not fail to
converge but the load-carrying capability of the nonlinear region is overestimated as shown on
Figure 85.
Figure 86: Compression of core seen at the middle of a 3PB test specimen
During the testing of specimen 11, delamination occurred between the core and the first layer of
composite while specimen 12 had a delamination of only the top layer of the top skin with the
remaining layers attached to the core.
Figure 87: Load-displacement graph of specimen 1 Figure 88: Stiffness graph of specimen 1
In the following, the load-displacement and stiffness graphs for all specimen are shown, grouped
according to laminate. All of the laminates with M80 core deform beyond the rules required test
range of 12,7 mm before failing and have a yield-like behavior of the laminate as a whole. The
laminates with M130 core fail much earlier and with less yield-like behavior. The strength and
stiffness of having a skin thickness of 4 layers and a core of M130 foam are near those of an M80
core with 8 layers for each skin but the weight of the former are around 990g while the weight of
the latter are around 1300g, showing the importance of choosing the core material.
Figure 93:3PB load-displacement curve, 4 layer skins, 10 mm Figure 94:3PB stiffness graph, 4 layer skins, 10 mm M80 core
M80 core
When simulating the bending of these plates using a model with a solid core in ANSYS to capture
the nonlinear effects, the stiffness of the specimen is seen to rise before yielding, like it was seen for
the steel plate. As the stiffness is observed to drop in the experimental values, it indicates that the
nonlinearity of the core material must be simulated to start very early, which cannot be done with a
bilinear hardening model. Like for bending of the core material by itself, employing a multilinear
hardening model was not successful for the sandwich plates either. The load-displacement curve
and the stiffness curves seen in Figure 99 and Figure 100 compare the simulated results with the
stiffness from the data sheet to the simulated result with the tested stiffness and to the experimental
result of bending specimen 1. The flexibility of the testing rig has been added to the simulated data
to compare against the experimental data. The importance of modelling plastic behavior can be seen
very clearly as the maximum load of the simulated data is sensitive to the yield strength but that the
curve cannot be made to follow completely by just a bilinear model.
The material for the core that best fit the experimental results of specimen 1 are a yield strength of
0,95 MPa and a tangent modulus of 2 MPa but this vastly underestimate the strength of specimen 7
which has 8 layer thick skins.
In the following, the stiffness of the test specimen are plotted along with the simulated data in
Figure 102 to Figure 113. The legend is the same for all of these figures and are given in Figure
101.
Figure 105: Energy absorption versus Figure 106: Energy absorption versus Figure 107: Energy absorption versus
skin thickness core thickness core density
Figure 111: Maximum load versus skin Figure 112: Maximum load versus core Figure 113: Maximum load versus core
thickness thickness density
In order to use the computational methods to search for optimum layups in a preliminary design
phase prior to conducting tests, it is necessary to know the percent of error of the simulations
relative to the experiments but also how much the individual tests differ from the mean values. As
there are too few experimental results for each layup to give a meaningful 95th percentile confidence
The yield load data has a higher spread than the other data for each layup which is caused by the
way that this is calculated as the end of the linear region determined by the statistical method. This
method is more sensitive than if it was determined by other methods such as a fixed ratio in drop of
stiffness or a fixed percent of nonlinear displacement which are not chosen as the fixed values
would be chosen somewhat arbitrary and could cause either a system that is very sensitive to noise
or where the range of linear displacement is overestimated, leading to lower calculated stiffnesses
and higher calculated yield loads.
The high spread in the calculated yield load of the specimen with 4 layer skins and a 25mm M80
core are due to specimen 11 having a much earlier calculated end of the linear region than the
others. By examining the stiffness graph in Figure 90, it is seen the stiffness graph follows the other
two specimen pretty well beyond the calculated yield load and thus it is assumed that it is an artifact
of the chosen method as discussed above. If the yield load of specimen 11 is considered an outlier
and are omitted, the mean yield load of that layup are 4788,3 N/mm with a spread of +/-4,4%. As
the spread with specimen 11 omitted is very similar to that of the other layups, this mean yield load
is chosen instead for the evaluation of the simulation errors.
The simulation error are shown in Table 24, Table 25, Table 26 and Table 27 for the stiffness,
energy absorption, yield load and maximum load respectively. From these tables and the figures
Stiffness error 4-layer skins 8-layer skins 14-layer 4-layer skins 4-layer skins
skins
25mm core 25mm core 25mm core 10 mm core
M80 foam M80 foam 25mm core M130 foam M80 foam
M80 foam
1D beam, tested 35,9% 27,8% 18,3% 19,3% 22,0%
data
1D beam, 13,4% 3,7% -5,3% 19,3% 7,3%
datasheet data
2D beam, tested 26,6% 25,4% 27,4% 11,5% 21,1%
data
2D beam, 6,3% 3,9% 6,2% 11,5% 7,2%
datasheet data
Shell model, 38,2% 30,3% 20,7% 18,5% 23,1%
tested data
Shell model, 15,5% 5,9% -3,3% 18,5% 8,4%
datasheet data
Full model, 0,3% -8,2% 12,9% -0,2% -44,9%
tested data 16,1% 15,3% 10,5% 4,4% 20,4%
Full model, -16,4% -11,0% -6,3% 0,3% -22,7%
datasheet data -1,2% -3,2% -6,7% 4,4% 6,9%
Table 24: Simulation error, stiffness
The relatively large error in the stiffness full model with 10 mm core could be due to a meshing
issue as the discretization through the thickness is set to a certain number of elements while the face
Yield load error 4 layer skins 8 layer skins 14 layer skins 4 layer skins 4 layer skins
25mm core 25mm core 25mm core 25mm core 10 mm core
M80 foam M80 foam M80 foam M130 foam M80 foam
1D beam, tested 184% 108% 68,0% 151% 50,0%
data
1D beam, 184% 108% 68,0% 151% 50,0%
datasheet data
2D beam, tested 65,9% 63,3% 69,2% 32,1% 49,9%
data
2D beam, 74,9% 78,3% 78,0% 32,1% 50,0%
datasheet data
Shell model, 447% 391% 359% 265% 116%
tested data*
Shell model, 357% 299% 267% 265% 90,9%
datasheet data*
Full model, 7,7% -1,2% -7,5% 37,0% -40,2%
tested data -42,6% -51,7% -75,9% -73,2% -54,2%
Full model, 26,9% 13,1% -6,4% 22,3% -20,1%
datasheet data -58,4% -53,5% -73,5% -73,2% -68,8%
Table 26: Simulation error, Yield load
*The Shell model in ANSYS does not include core yielding and also fail to predict any the failure
of the skins, meaning that the deformation is approximately linear until failure.
Max load error 4 layer skins 8 layer skins 14 layer skins 4 layer skins 4 layer skins
25mm core 25mm core 25mm core 25mm core 10 mm core
M80 foam M80 foam M80 foam M130 foam M80 foam
The shell model calculates a slightly larger stiffness than the beam model, which indicates that there
are some stiffening effect from prevention of anticlastic curvature. The stiffening effect is found by
dividing the stiffness of the shell model by the stiffness of the beam model and are given in Table
28 and while these are not generally applicable to all laminated structures and are only applicable to
these layups when the beam width is 275mm, the values are used to assess the magnitude of the
stiffening effect. The stiffening effect on these specimen are relatively small compared to the
overprediction of stiffness, the possible variance in the stiffness of the core material and compared
to the spread in the experimental stiffnesses. The stiffening effect of completely preventing
anticlastic curvature of a homogeneous isotropic beam with a Poisson’s ratio of 0,3 is 9,89% so
compared to this, the stiffening effect on these specimen is also considered small. It is unexpected
that the layup with the thinner core exhibits a smaller stiffening effect than the similar layup with a
thicker core since the opposite is the case for homogeneous isotropic beams as described in section
3.5. As no attempt is made to generally quantify this stiffening effect on laminated beams, this is
not further investigated. That the stiffening effect seems negative on the layup with M130 core is
assumed to be due to numerical errors in ANSYS.
4 layer skins 8 layer skins 14 layer skins 4 layer skins 4 layer skins
25mm core 25mm core 25mm core 25mm core 10 mm core
M80 foam M80 foam M80 foam M130 foam M80 foam
Stiffening 1,71% 1,98% 2,02% -0,66% 0,84%
effect of
The shell model is not considered to bring significantly more value to the problem of 3 point
bending than a beam model and are thus not considered valuable for the optimization problem.
The following in this subsection is added in V2.0 of this report, the 11th of august 2020. Expanded in
V2.1, the 17th of August 2020.
With the multilinear isotropic hardening model for modeling plasticity of the core, the model does
not converge, even with very small timesteps and very fine meshes, when the contact between the
skin and core is set to “program controlled” which uses the augmented lagrange formulation. By
setting the contact formulation to pure lagrange, it does converge but similarly as for simulation of
the foam by itself, the nonlinear behavior is not simulated well and the stiffness after core yielding
is overpredicted significantly.
The stiffness prior to yielding is also seen to be much higher than what is measured and what is
simulated with the bilinear model. This is not found to be due to too coarse mesh or timestep size
and also persists when the material nonlinear effects are disabled for the core. The difference is
found to be mostly due to the bilinear models having been meshed with a 5mm face size and 8
through-thickness divisions of the core, giving an aspect ratio of 1,6. Lowering the number of
through-thickness divisions raises the calculated stiffness, as does refining the face size to 3,2mm to
make the aspect ratio close to 1 while there’s 8 through-thickness divisions.
It is found to be connected to the contact formulation between the skins and the core as having the
pure lagrange formulation is about 5% stiffer than the default augmented lagrange formulation
when the mesh is fine and the aspect ratio is close to 1. When the mesh is coarser and the aspect
ratio is close to 1, the difference is about 8% and when the mesh is as used in the simulations in this
report, the difference is 15%. The results with the pure lagrange formulations are in close agreement
with only 1% difference between the fine mesh with a good aspect ratio and the coarser mesh with a
worse aspect ratio.
After penetration of the top skin, the punch travels through the core by propagating a crack
outwards in a conical shape. During this, the stiffness rises both due to the core being compressed
below the punch but also due to the crack moving to a larger diameter so that more crack needs to
be created per distance of downwards movement of the punch. The cross-section of the core after
penetration are seen on Figure 115, which is from a PSS test of only the foam. A major difference is
however that in the test shown in Figure 115, the diameter of the final penetration is that of the
punch, while for the laminated specimen it is the diameter of the hole in the supporting fixture.
After punching through the core, the bottom skin is pressed out through the hole in the supporting
plate. The failure that causes the peak as the bottom skin fails cannot be seen but it is noted that
after the test, all layers have usually delaminated from each other or in pairs of one 0 and one 90
degree layer together.
The load-displacement graphs of the tested laminated specimen are shown below, where it is seen
that the behavior are fairly consistent but that there are still differences with some specimen failing
earlier than the others with the same layup. Specimen 30 did not punch through the top skin and
instead, the whole core yielded and the load maxed out the capacity of the test machine. As the
failures of the top skin seems to be more related to bending than to through-thickness shear, it may
be that the failure loads would be higher if the test specimen were larger.
Modelling of the penetration through the core requires modelling the crack front as it propagates.
Such a model requires input data which cannot be extracted from the tests which have been
conducted in this project and such, it has been deemed out of scope of this project and thus only the
modeling of the first peak is attempted.
Production of a model capable of accurately predicting the failure load during perimeter shear stress
testing has not been possible.
Using solid-shell elements for the core results in a distorted mesh which cannot be refined and thus,
that model is not of any use.
Using shell elements for the skins does not result in failures with clear drops in load but just very
large deformations and remeshing of the core. Remeshing of the core when the mesh is highly
distorted will come with a drop in load which can easily be misinterpreted as the first peak but this
is an artifact of the modeling.
Without a model that simulates the physics of the PSS test, some predictive capability can be gained
by building an empirical model by fitting some mathematical function to the test data, similarly to
how the Pacejka “magic formula” is used to model tire friction without being based on any physics.
Building such a model would require a large dataset and modelling how the different parameters
interact would require an immensely large dataset.
As there are only two different core thicknesses and two different core densities tested, these are
assumed to be linear scalings of the function for the skin thickness, as two datapoints can’t be used
to uniquely define very many functions. Linear functions are however only approximately correct at
interpolation or very low levels of extrapolation as physical phenomena are never truly linear.
The modelling of the load as a function of the skin thickness should have a value of zero at 0 layers,
it should be monotonically increasing but with a decreasing first derivative which goes to 0 at
infinity. This is assumed from the fact that adding one more layer increases the total thickness a lot
when there are few layers and increases it only a little when there are many. The necessary
asymptotic nature is explained by the fact that the compressive stress will at some point limit the
load carrying capacity. Sigmoid functions are a class of functions which could yield this capability
but at the current date (20th of August), no suitable function has been determined.
The fact that the 2nd peak of the 4-layer M80 core laminate schemes overlap means that the linear
relationship that can be formed by this data should not be considered reliable. Dependency on the
core thickness is omitted for the 2nd peak, also as a reduction in strength as the core thickness goes
to zero does not intuitively make sense.
The lightest usable laminate is found to have 2 layers of SE84LV-HEC on either side of a 27,5mm
thick core of M100 foam and a total mass of 4,837 kg per m2. Among the 50 lightest usable
laminates are no layups with M60 or M200 foams, there are none with a core thickness of less than
17,5mm, and there are no layups with more than 8 layers.
Removing the laminate components that were not found among the 50 lightest usable laminates
from the 1D beam model reduces the layups to be analyzed to 450 combinations, which takes 10
minutes to calculate as a beam with a 2D core. Of these, only 88 are usable as most of the laminates
have insufficient stiffness. Of the usable lamiantes, most are with an M130 core, as opposed to
when the beam model are used for optimization where only very few of the lightest usable layups
had M130 cores. This seems to indicate that the core stiffness is a limiting factor to the optimization
and thus the M200 core are added back in, the M80 core is removed and the minimum core
thickness is reduced to 15mm to allow for layups with a thinner but stiffer core.
525 laminates are calculated as beams with a 2D core in 11 minutes but the 6 lightest layups are still
with M130 cores before M200 and M100 cores are among the usable layups. The lightest layup is 3
layers for top skin and two layers for bottom skin with a 30 mm thick M130 core and a mass of
6,551 kg per m2.
The optimum layup from the 2D beam model is calculated to have a stiffness of 2986 N/mm and is
brought on into ANSYS to be calculated at a higher level of detail. It is first analyzed linearly with a
unit displacement load to determine if the stiffness is sufficient. At a mesh with a 5mm face size
and 8 elements through the thickness of the core, the stiffness is 2976,3 N/mm and when refined to
a face size of 3mm, the stiffness rises to 3119,9 N/mm which is a 5% rise from the coarser mesh.
Refined to a face size of 2mm and 15 elements through the thickness of the core, the calculation
time is unacceptably long and the previous result is taken as having a suitable level of refinement.
As the stiffness is calculated approximately equal to that of the 2D beam simulation which is almost
30% above the rules requirement even before equation (4-3) is used to correct the stiffness,
requirement in the 2D beam optimization simulation is lowered to be only 10% higher than the
amount required by the rules. The result of the optimization search is shown in Figure 123 and
Figure 124 where it can be seen by the magenta dots that insufficient energy absorption becomes as
Figure 123: Stiffness of laminates by 2D beam model for Figure 124: Specific stiffness of laminates by 2D beam model
optimization purposes for optimization purposes
The next iteration of simulation have the 4 layers as the top skin with the layup
[0°, 0°, 25𝑚𝑚 𝑀80 𝑐𝑜𝑟𝑒, 10°, −10°, 10°, 0°]. The mass is 6,3217 kg/m2 and the stiffness is
calculated by the 2D beam model to be 2678,7 N/mm and by linear analysis in ANSYS to be
2470,8 N/mm. When corrected by (4-3), the stiffness to be used by the SES is 3259,7 N/mm and the
calculated equivalent bending stiffness, 𝐸𝑒𝑞 𝐼𝑙𝑎𝑚 , are 1,9% above the rules requirement. Rather than
iterating further, the height of the side impact structure is raised from 250 mm to 258mm so that the
equivalent bending stiffness is 5% above the rules requirement. Raising the height results in a
lighter structure than if the next lightest laminate with a higher stiffness than this is used.
The calculated energy absorption is 150,2 Joule, the yield load is 11922 N and the highest measured
load is 16808 Newton, all well beyond the rules required limits. The stiffness in the nonlinear
analysis is slightly higher than in the linear analysis and is found to be 2487,5 N/mm in average
over the first millimeter. The value chosen as the yield load is the onset of plastic deformation,
which is also marked in Figure 126 and Figure 127. On Figure 125, the deformed shape and the
shear stress through the core are shown and a notable feature is that the shear strain seems to be a
little more complex than the near-constant value which the beam model calculates. The distribution
of the shear strain is slightly different than what is seen in simulations of the symmetric laminates
but this is likely caused by the stiffer top skin as the plastic strain is much less concentrated below
the applied load at the center.
The total area on the monocoque that needs to be laminated is 2,7835 square meters, which means
that if all of the monocoque were to be made with this laminate, the weight would be 17,6kg plus
about 5,5 kg of steel tubing for the roll hoops and the main hoop bracing, not including attachment
brackets. If structurally unregulated areas such as the firewall behind the driver’s back and the floor
beneath the driver’s legs are subtracted due to being able to be made from lighter laminates, there
are 2,175 square meters of chassis, resulting in 13,74 kg plus the steel tubing. The whole chassis
can’t however be made from this laminate as the requirements are different for different parts of the
car, such as with the accumulator protection structure and tractive system protection structure
requiring equivalency to 3 of the tubes that the side of the SIS only requires equivalency to two of.
The requirements for the front bulkhead support structure is however lower, so the masses
mentioned here are likely to be a reasonable estimate.
The process of searching for optimum layups for the rest of the monocoque is done in a similar
fashion as shown for the side impact structure but due to time constraints in this project, it cannot be
included in this report.
[27] Welch, B., (1947) “The Generalization of `Student's' Problem when Several Different
Population Variances are Involved”, Biometrika, Volume 34, pages 28-35
[28] Grower, M., Shaw, R., Broughton, W. (2008). ”NPL Report Mat 24”. National Physical
Laboratory. ISSN 1754-2979