Design and analysis of a composite beam for infrastructure applications

Part III: Experimental results and nonlinear FE analysis

By

Mario Springolo 1 , Amar Khennane 2* , Gerard van Erp 3

1,2
Faculty of Engineering and Surveying, The University of Southern Queensland,
Toowoomba, Qld 4350, Australia
3
Professor, Executive Director, Fibre Composite Design and Development Centre. The
University of Southern Queensland, Toowoomba, Qld 4350, Australia

Abstract:

Using the analytical approaches developed, the cross section of the new fibre composite beam

described in the prequels to this paper is designed in order to avert secondary failure modes. A

series of specimens have been built and put through a thorough testing regime to establish the

performance of the beam. To gain confidence in the analytical models and achieve further

understanding of the beam behaviour, a rigorous nonlinear finite element analysis is also

presented. It was found that the analytical model agreed relatively well with the experiment

and the FE analysis, thus validating the underlying assumptions.

Keywords: composite beam, nonlinear finite element, Abaqus, Python, bending, shear,

moment shear interaction

*Lecturer, Faculty of Engineering and Surveying, The University of Southern Queensland,

Toowoomba, Qld 4350, Australia
Tel: (+61) (7) 4631 1383, Fax: (+61) (7) 4631 2526, email: [email protected]

1
Biographical notes: Mr Mario Springolo has submitted his PhD to the Faculty of
Engineering and Surveying at USQ. He is currently working as a structural engineer with
Larken Teys consulting Pty Ltd. http://www.larkinteys.com.au

Biographical notes: Dr Amar Khennane holds a BE from Tizi-Ouzou, Algeria, an M.Sc.

from Heriot Watt, UK, and a PhD, from UQ, Australia. Before joining the Faculty of
Engineering and Surveying at the University of Southern Queensland, he was a researcher
with the Cooperative Research Centre for Advanced Composite Structures, Ltd, Australia. His
research interests are in composite materials: micro-mechanics, durability and through life
estimation of composite structures, and computational mechanics.

Biographical notes: Professor Van Erp was born and educated in the Netherlands and
moved to Asutralia in 1989. In 1990 he joined the Faculty of Engineering and Surveying at
USQ. Currently, he is the Executive Director of Fibre Composites Design & Development,
University of Southern Queensland (USQ) responsible of Australia's first fibre composites
bridge.

2
INTRODUCTION

The new fibre composite beam described in the prequels is put through a testing program that

includes tests designed for primary and secondary failure modes. In particular, the behaviours

in bending, shear, combined bending and shear, buckling, and lateral torsional buckling were

investigated. A detailed description of the experimental set up and the complete results are

given in [1], and only the most important results are reported herein.

In particular, it was found that the core material cracks prior to the ultimate failure of the

beam. This was caused by the low failure strain of 0.6% of the core when compared to the

main unidirectional laminates, which have a failure strain around 2%. Associated with the

cracking of the core is localised de-bonding at the interface between the laminates and core.

In addition, it was found that the different constituents of the beam exhibited different

behaviours in tension and compression. The FRP laminates were found to undergo

progressive damage due to matrix cracking, fibre-matrix de-bonding and fibre breakage

before failure. The best approach to model these phenomena appears to be at micro-

mechanical level. However, such an analysis is impossible for large components such as

structural elements. Previously some research efforts have assumed equivalent homogeneity

to make limited progress. In the present paper, instead of ignoring the lack of homogeneity,

attention will be focussed on the internal structure of the material, since it is the latter that

governs the behaviour of the composite system and consequently its failure. The approach

favoured herein considers the damage progression in the constituents.

Using both the scripting ability of the ABAQUS finite element software [9], and user defined

subroutines, the damage phenomena are incorporated in the form of damage variables to

3
model the nonlinear behaviour of an FRP beam. The obtained results will be compared to

both the analytical and experimental results.

FINITE ELEMENT ANALYSIS

Modelling the FRP beam using ABAQUS

ABAQUS is a general finite element package with non-linear capabilities. In particular,

whenever a material definition is not supported, the user is provided with the ability to use

external subroutines. Taking advantage of the ability to script within ABAQUS, using the

Python language [10], an algorithm taking into account the progressive damage within the

constituents is proposed. To account for the different material properties in tension and

compression, a linear elastic analysis with homogeneous properties is first run to identify

stress states within the beam. Material properties are then assigned correspondingly. Because

of changes in the model, caused by modelling the damage at different parts of the beam

throughout the analysis, this process is updated at each iteration, depending on the stress

states obtained at the previous iteration. The cracking in the core material is modelled using a

smeared crack approach. When the strain in the core material exceeds a limiting value, the

elastic modulus is put to zero at the offending Gauss point. The peel-off of the laminate from

the core material is not modelled as the interaction behaviour between core material and

laminates is yet to be quantified, and is the subject of continuing investigation. Damage in the

laminates is modelled by means of an external Fortran subroutine implementing the damage

model of Chang et al. [8]. Furthermore, initial material properties throughout the beam model

are varied to imitate the heterogeneous nature of the materials so that stress concentrations

may form, providing the sites for crack initiation and laminate failure. A flow chart of the

process is shown on Figure 1.

4
To model the progressive matrix, shear, and compressive failure of the laminates, the existing

formulations by Chang et al. [8] were also incorporated within a user defined Fortran

subroutine.

PURE BENDING

Finite element model

Figure 2 shows the loading details and the geometrical details of the cross section used to

simulate the beam under pure bending. Only the shaded area, pure bending part, shown on

Figure 2-a is analysed.

Due to symmetry, only half of the section is analysed as shown on Figure 3 representing the

finite element mesh and boundary conditions. The nodes on the left face were constrained to

remain on a plane to comply with the Bernoulli hypothesis that plane sections remain plane.

The nodes on the right hand cross section are fixed in both the first and third direction but free

to move in the third (vertical) direction except for three nodes in the middle of the section that

are totally restrained. This is necessary to avoid any rigid body movement of the model. The

nodes on the longitudinal face of symmetry are only constrained in the first (transversal)

direction.

The loading is applied through a couple as shown on Figure 4 to induce pure bending. The

core and longitudinal laminates were discretised using 8 noded brick elements as shown on

Figure 5. The RHS laminates were discretised using 4 noded shell elements as shown on

Figure 6.

5
Material properties

Material properties were randomly assigned to the brick and shell elements in the models. The

distribution was based upon the average and standard deviation recorded in testing the

laminates and the Particulate Filled Resin (PFR) shown in Table 1 of Part I of the prequel. To

ensure that results were statistically valid, 30 models, each with a different randomisation

sequence, were analysed for each loading configuration. Both the assigning of tensile and

compressive properties, and the modelling of progressive failure was achieved within

ABAQUS via a user defined Fortran subroutine. The core material cracking and FRP tensile

capacities were based on a maximum strain criterion. To simulate the cracked behaviour of an

element, the tensile and shear moduli, respectively parallel and perpendicular to the direction

of failure, were re-set to one. The 10 different principal strain states possible as represented on

Table 1. For example, if a brick element representing a laminate is under a state of tension (T)

in the direction of ε11, and compression (C) in the directions of ε22 and ε33, then the tensile

modulus E11 will be used in the direction of ε11, while the compressive moduli E22 and E33

will be used in the directions corresponding to ε22 and ε33. Original non-failed value (OV) for

Poisson’s ratios and shear moduli will also be used. However, if it has already failed in any

direction (represented by the letter F), then failed values (FV) for the elastic properties in that

direction will be used. Table 1 is used to model the progressive matrix, shear, and

compressive failure of the laminates using the damage formulation developed by Chang et al.

[8]

Results and discussion

Figure 7 shows the obtained results for the beams tensile and compressive strains at mid-span

(denoted T and C respectively in the figure). It can be seen that there exist an excellent

6
correlation with both the experimental and analytical results. Furthermore, the FEA and

analytical ultimate capacity predictions agree very well with the experimental results.

The difference in the analytical and FEA predictions of tensile strains, between moments of

1E6 to 1.75E6 Nmm on Figure 7, could be attributed to the fact that the FEA model does not

provide for core-laminate de-bonding. Given the current state of knowledge, this de-bonding

behaviour cannot be modelled reliably, regardless of the FEA package chosen. As mentioned

previously, the characterisation of this phenomenon is still under investigation. However, the

FEA results converge to the analytical predictions above a moment of 1.75E6 Nmm, which is

of prime importance as far as prediction of loading capacity is concerned. To further test the

validity of the developed FEA approach a number of models with altered beam geometry are

compared against experimental data. As can be seen on Figure 8 to Figure 11, the FEA

predictions are consistent for all the beam geometries tested.

MODELS UNDER SPECIFIC LOADING REGIMES

The developed approach provides excellent correlation with analytical and experimental

results under a state of pure bending. To establish the overall applicability of the approach, it

is used to model the beam behaviour, and predict the occurrence of following range of failure

modes:

• moment buckling of the webs,

• combined shear and bending,

• shear buckling of the webs, and

• lateral torsional buckling

7
Moment buckling of the webs

Testing for web buckling required the manufacture and testing of separate web specimens.

These specimens are modelled as per the four-point bending test arrangement. The first

buckling mode, as predicted by the model, is shown on Figure 12. Comparison of results, as

shown on Table 2, shows that the analytical and finite element solutions are similar and agree

well with the experimental results.

Combined shear and bending

Six beams loaded in four-point bending were analysed. Due to symmetry, only half spans

were modelled. The lengths of the spans were successively reduced to obtain the interaction

diagrams between moment and shear. Figure 13 shows the two extremes of the model

geometry’s. The obtained failure predictions, together with the analytical and experimental

results, are plotted on Figure 14. It can be seen that the FEA results reproduce the trend of the

experimental results shown as dots on the plot. However, the FE results correlate with the

analytical results only at either high-shear and low moment or low-shear and high-moments.

Furthermore, the general trend shown by the FE results seems to follow occurrences of

cracking in the core material as experimentally observed.

Shear buckling of the webs

To model shear buckling, a web specimen similar to that used to test for moment buckling,

was modelled. The first predicted buckling mode is shown on Figure 15. As with moment

buckling, comparison of the results, as reported on Table 3, shows good agreement between

the analytical, FE and experimental results.

Lateral torsional buckling

8
In line with the experimental set up, a two-meter long cantilevered beam was modelled. The

model was solved to obtain the eigenvalues. Figure 16 shows the first predicted buckling

mode. As shown on Figure 17, there is some variability between the experimental, theoretical

and FEA results. The Eurocomp formulation for FRP beams is substantially lower than the

other methods. The remainder of the predictions lie within 27% of each other. The FEA is

within 11% of the critical moment as observed in the experimentation.

CONCLUSIONS

The behaviour of the developed beam has been well described using advanced nonlinear finite

element analysis that incorporates heterogeneous material property description and damage

progression in the individual constituents of the beam. In particular, the developed FE

approach captured the interaction between shear and bending better than any other method

Both the experimental results and the non linear finite element analysis corroborated that the

addition of the core material to the flanges and webs of the newly developed beam suppresses

most of the premature failures modes known to occur in existing designs. This constitutes an

improvement in overall beam performance. However, it was also revealed that failures from

point loads, and combined moment–shear interaction, are dominated by buckling failure due

to the cracking of the core material. Contrary to existing belief, it was found that the pure

shear capacity of the web laminates is governed by fibre fracture rather than fibre pull-out as

is described in codes and the literature.

9
The simplified analytical formulas described in the sequels show very good agreement with

both the experimental results and FE analysis. These formulas were initially formulated for

designing the experimental program. However, in the light of the observed accuracy, these

formulas could constitute the basis for designing these FRP beams.

References

1. Springolo, M. New fibre-reinforced polymer box beam: investigation of static

behaviour. PhD thesis. Faculty of Engineering, The University of Southern Queensland,

Australia. 2005. http://adt.usq.edu.au/adt-QUSQ/public/adt-QUSQ20050421.122411/

2. Ascione L, Feo G, Mancusi G. On the statical behaviour of fibre-reinforced polymer

thin-walled beams. Composites Part B: Engineering. 2000; 31B(8): 643-653.

3. Barbero EJ, Lopez-Anido R, Davalos JF. On the mechanics of thin-walled laminated

composite beams. Journal of Composite Materials. 1993; 27(8): 806-829.

4. Davalos JF, Qiao P. Analytical and experimental study of lateral distortional buckling

of FRP wide-flange beams. Journal of Composites for Construction. 1997; 1(4): 150-

159.

5. Davalos JF, Salim P, Qiao R., Lopez-Anido R, Barbero E J. Analysis and design of

pultruded FRP shapes under bending. Composites Part B: Engineering. 1996; 27(3/4):

295-306.

6. Haj-Ali R, Kilic H. Nonlinear behavior of pultruded FPR composites. Composites Part

B: Engineering. 2002; 33(3):173-191

7. Taufik A, Barrau JJ, Lorin F. Composite Beam Analysis with Arbitrary Cross Section.

Composite Structures. 1999; 44(2/3): 189-194.

8. Chang F, Lessard L, Chang K, Liu S. Damage tolerance of laminated composites

containing a hole and subjected to tension or compression. 34th International SAMPE

10
 Symposium and Exhibition - Tomorrow's Materials: Today. SAMPE, Covina, CA,

USA, 1989 p:559-568.

9. Hibbitt, Karlsson & Sorensen Inc., 'ABAQUS', ver 6.4, Hibbitt, Karlsson & Sorensen,

Inc., Pawtucket. RI 02860-4847, USA, 2003

10. Van Rossum G. Python Software Foundation. http://www.python.org/

11
 Table 1: Material properties for failure modelling

Principal strain Material properties

state Com. = compression modulus,
C = compression, Ten. = tension modulus,
T = tension, and OV = original non-failed value used, and
F = failed in FV = failure value used (1 for E and G, 0 for ν)
tension
ε11 ε22 ε33 E11 E22 E33 ν12 ν13 ν23 G12 G13 G23

C C C Com. Com. Com. OV OV OV OV OV OV

T C C Ten. Com. Com. OV OV OV OV OV OV

F C C FV Com. Com. FV FV OV FV FV OV

T T C Ten. Ten. Com. OV OV OV OV OV OV

F T C FV Ten. Com. FV FV OV FV FV OV

F F C FV FV Com. FV FV FV FV FV FV

T T T Ten. Ten. Ten. OV OV OV OV OV OV

F T T FV Ten. Ten. FV FV OV FV FV OV

F F T FV FV Ten. FV FV FV FV FV FV

F F F FV FV FV FV FV FV FV FV FV

12
 Table 2: Moment buckling of the web specimen

Analytical FEA Experimental

Buckling moment
(105 N.mm) 9.66 9.68 Buckled at: 8.60
Failed at: 9.72

13
 Table 3: Shear buckling of the web

Analytical FEA Experimental

Shear Load (N) 6871 7023 Buckled at: 7500

Failed at: 9370

14
 Start

Run i = 1
Increment j = 1
Set number of increments, N

Run i = i + 1 For Run number i

Randomise material properties based upon the

average and standard deviation of the material
tests

Assemble model, write ABAQUS input file

Run increment j FEA using ABAQUS with ABAQUS

automatic sub-increments to ensure
convergence
program

From results assign changed tensile and

FORTRAN
subroutine
compressive properties to individual elements

Check for initiation and progression of

damage to core and laminates, then assign
appropriate material properties

Have any of the

Yes element properties
changed?
Increment j = j + 1

No

Has increment, j,
No reached N?

Yes

Access ABAQUS output file

Compile results with previous results

Has runs, i, reached

No 30

Yes

End

Figure 1: Flow chart of Python script

15
Figure 2: Loading details and cross section .

16
Figure 3: Finite element mesh and boundary conditions for pure bending

17
Figure 4: Finite element mesh and loading for pure bending

18
Figure 5: Finite element dicretisation of the core and longitudinal laminates

19
Figure 6: Finite element dicretisation of the RHS laminates

20
 3.50E+06

3.00E+06

2.50E+06
Moment (Nmm)

2.00E+06 Abaqus-C
Abaqus-T
1.50E+06 Analytical-C
Analytical-T
1.00E+06 Experiment-C
Experiment-T
5.00E+05

0.00E+00
0 0.005 0.01 0.015 0.02
Strain (mm/mm)

Figure 7: FE models versus analytical and experimental results

21
 3.00E+06 Compressive strain, L2

2.50E+06 Compressive strain, L3

Compressive strain, L4
2.00E+06
Tensile strain, L14
1.50E+06
Analytical Compressive
strain, L2
1.00E+06 Analytical Compressive
strain, L3
Analytical Compressive
5.00E+05 strain, L4
Analytical Tensile strain, L14
0.00E+00
FE Compressve strain, L2
0.0 0.5 1.0 1.5 2.0
FE Tensile strain, L13
Strain (%)

Figure 8: FE versus analytical and experimental results for changing flange core thickness,

series B6

22
 3.00E+06 Compressive strains, S1
Tensile strains S1
2.50E+06 Compressive strains, S2
Tensile strains, S2
2.00E+06 Compressive strains, S3
Tensile strains, S3
1.50E+06
Compressive strains, S4
Tensile strains, S4
1.00E+06
Compressive strains, S5

5.00E+05 Tensile strains, S5

Analytical Compressive
strains
0.00E+00 Analytical Tensile strains
0.0 0.5 1.0 1.5 2.0 FE Compressive strains
FE Tensile strains
Strain (%)

Figure 9: FE versus analytical and experimental results for changing flange core thickness,

series B6

23
 3.00E+06 Compressive strain, S1

2.50E+06 Tensile strain, S1

2.00E+06 Compressive strain, S2

1.50E+06 Tensile strain, S2

1.00E+06 Analytical Compressive

strain
5.00E+05 Analytical Tensile strain

0.00E+00 FE Compressive strain

0.0 0.5 1.0 1.5 2.0
FE Tensile strain
Strain (%)

Figure 10: FE versus analytical and experimental results for changing flange core thickness,

series B8

24
 4.00E+06
Tensile strains, S1
3.50E+06

3.00E+06 Tensile strains, S2

Moment (Nmm)

2.50E+06 Analytical Compressive

strains
2.00E+06
Analytical Tensile
1.50E+06 strains
FE Compressive strains
1.00E+06

5.00E+05 FE Tensile strains

0.00E+00
0.0 0.5 1.0 1.5 2.0
Strain (%)

Figure 11: FE versus analytical and experimental results for changing flange core
thickness, series B9

25
Figure 12: Moment buckling of the web specimen

26
Figure 13: Combined shear and bending.

27
Figure 14: Results for combined shear and bending

28
Figure 15: Shear buckling of the web

29
Figure 16: Lateral torsional buckling

30
Figure 17: Comparison of lateral torsion buckling predictions

31