Professional Documents
Culture Documents
Laura TepfersASCEpaper
Laura TepfersASCEpaper
net/publication/240504795
CITATIONS READS
246 1,414
2 authors:
Some of the authors of this publication are also working on these related projects:
Reproducing Sneddon's solution using fracture PhF formulations accurately is not easy View project
All content following this page was uploaded by Laura De Lorenzis on 03 April 2014.
Abstract: The use of fiber-reinforced polymer 共FRP兲 composites for strengthening and/or rehabilitation of concrete structures is gaining
increasing popularity in the civil engineering community. One of the most attractive applications of FRP materials is their use as confining
devices for concrete columns, which may result in remarkable increases of strength and ductility as indicated by numerous published
experimental results. Despite a large research effort, a proper analytical tool to predict the behavior of FRP-confined concrete has not yet
been established. Most of the available models are empirical in nature and have been calibrated against their own sets of experimental
data. On the other hand, the experimental results available in the literature encompass a wide range of values of the significant variables.
The objective of this work is a systematic assessment of the performance of the existing models on confinement of concrete columns with
FRP materials. The study is conducted in the following steps: the experimental data on confinement of concrete cylinders with FRP
available in the technical literature are classified according to the values of the significant variables; the existing empirical and analytical
models are reviewed, pointing out their distinct features; the whole set of available experimental results is compared with the whole set
of analytical models; and strengths and weaknesses of the various models are analyzed. Finally, a new equation is proposed to evaluate
the axial strain at peak stress of FRP-confined concrete cylinders.
DOI: 10.1061/共ASCE兲1090-0268共2003兲7:3共219兲
CE Database subject headings: Comparative studies; Columns; Fiber reinforced polymers; Ductility; Concrete structures;
Rehabilitation; Confinement.
Empirical or Analytical Models were calibrated using the experimental values of co . Lateral
strains were not measured, therefore, it was not possible to check
The other models, except the one by Spoelstra and Monti 共1999兲,
are empirically based: they use best fitting of experimental data to whether the FRP hoop strain at failure was equal or less than its
correlate the stress-strain behavior of confined cylinders 共particu- uniaxial ultimate tensile strain. However, it is very likely that
larly the enhancement in strength and ultimate strain兲 to the val- ‘‘premature’’ tensile failure of the FRP occurred in Miyauchi’s
ues of the parameters believed to be significant. The equations are specimens such as in most wrapped specimens of the available
reported in Table 3. literature. If so, the empirical equations incorporate such phenom-
The equation proposed by Miyauchi et al. 共1997兲 to estimate enon in themselves, since they have been calibrated using the
the strengthening effect of the FRP confinement differs from uniaxial tensile strength of the FRP. The complete axial stress-
Richart’s equation 共with the active pressure substituted by the strain relationship of the confined concrete was proposed, based
confining pressure at FRP rupture兲 by the only addition of the on the observed behavior, as consisting of a parabolic branch,
empirical ‘‘effectiveness coefficient’’ k e . Such coefficient was modeled by the same equation of the unconfined concrete, fol-
calibrated by the authors from their own experimental results and lowed by a straight line tangent to the parabola at the intersection
was set equal to 0.85. The experiments were conducted on speci- point.
mens with values of the p u / f ⬘co ratio ranging from 0.1 to 0.5. The Kono et al. 共1998兲 developed two empirical equations for the
two empirical equations suggested for the computation of cc evaluation of peak stress and corresponding strain of confined
concrete cylinders. In these equations, the ratios f ⬘cc / f ⬘co and Saaman et al. adopted and calibrated the four-parameter stress-
cc / co are linearly correlated with the ultimate pressure p u strain relationship by Richard and Abbott 共1975兲 to model the
rather than with the ratio p u / f ⬘co . The equation giving the strain bilinear response of FRP-confined concrete, for more details see
at peak stress was obtained using the experimental values of co . the original paper 共Saaman et al. 1998兲. The expression of the
The experiments from which the previous equations were cali- confined strength was calibrated on experimental results obtained
brated are characterized by p u / f ⬘co ratios ranging from 0.37 to by the authors on FRP-encased specimens with p u / f ⬘co ratios
1.19, that is, rather high compared to those on which the preced- ranging from 0.31 to 0.84. The recorded FRP hoop strains at
ing models were based. The concrete unconfined strength ranged failure were very close to the ultimate tensile strain of the FRP
between 32.3 and 34.8 MPa. The lateral strains were measured tube and, therefore, the model does not incorporate the ‘‘prema-
and results indicated that only a percentage of the FRP ultimate ture’’ failure phenomenon. In this model as in those by Fardis,
strain ranging from 38 to 59% was developed at failure. There- Toutanji, and Saafi, cc depends not only on the ultimate confine-
fore, Kono’s equations incorporate such ‘‘premature’’ failure phe- ment pressure but also on the stiffness of the confining device.
nomenon in themselves. Besides, it is no more related to the unconfined strain at peak co .
‘‘approximate’’
Spoelstra and Monti 113 19.4 17.3 89.2 8 共8a兲 22.4 共22.4a兲 17.0 共17.0a兲 76.0 共76.0a兲 121 19.6 17.2 87.9
‘‘exact’’
Xiao and Wu 139 32.0 26.2 81.8 41 共31a兲 47.5 共24.7a兲 43.7 共16.3a兲 92.0 共66.1a兲 180 35.6 31.6 88.9
Table 5. Evaluation of Models in Prediction of f ⬘cc / f ⬘co with Effective Rupture Strain of FRP
⬘ / f co
Prediction of f cc ⬘ 共wraps兲 ⬘ / f co
Prediction of f cc ⬘ 共tubes兲 ⬘ / f co
Prediction of f cc ⬘ 共overall兲
Sample Average absolute Standard Coefficient of Sample Average absolute Standard Coefficient of Sample Average absolute Standard Coefficient of
Model number error 共%兲 deviation variation 共%兲 number error 共%兲 deviation variation 共%兲 number error 共%兲 deviation variation 共%兲
Fardis and Khalili 105 19.0 11.5 60.4 30 35.0 9.6 27.4 135 22.6 12.9 57.2
共Newman and Newman兲
Saadatmanesh et al. 105 18.7 11.6 62.0 30 22.2 11.2 50.2 135 19.5 11.6 59.3
Saaman et al. 105 19.8 13.6 68.9 30 6.8 5.1 75.7 135 16.9 13.4 79.4
Miyauchi et al. 105 19.7 14.0 71.2 30 22.8 9.1 40.0 135 20.4 13.1 64.3
Kono et al. 105 26.2 15.8 60.4 30 16.1 6.9 43.0 135 23.9 14.9 62.2
Toutanji 105 18.6 11.8 63.6 30 30.7 9.2 30.0 135 21.3 12.3 58.0
Saafi et al. 105 21.8 14.7 67.2 30 5.6 5.1 92.0 135 18.2 14.8 81.2
Spoelstra and Monti 105 32.2 19.5 60.5 30 6.5 5.5 85.1 135 26.5 20.4 77.0
‘‘approximate’’
Spoelstra and Monti 94 19.0 12.2 64.4 7 19.0 7.5 39.6 101 19.0 11.9 62.8
‘‘exact’’
Xiao and Wu 105 21.6 14.0 64.8 30 24.6 10.7 43.2 135 22.3 13.4 59.9
228 / JOURNAL OF COMPOSITES FOR CONSTRUCTION © ASCE / AUGUST 2003
measured values. A 45° line lying completely into the 95% con- with the highest values of the ratio p u / f ⬘co . This may be due to
fidence band would indicate very good agreement between theory the use of formulas that overestimate the effectiveness of high
and experiments. Due to limited space, only a few of these graphs lateral pressures 共Fardis and Khalili 1981兲 or to the empirical
are reported in Fig. 5, for the complete set see De Lorenzis and calibration being conducted on specimens with low ultimate
Tepfers 共2001兲. The accuracy of the models has been also quan- confinement pressures 共Miyauchi et al. 1997; Toutanji 1999;
titatively evaluated computing the average absolute percent error Xiao and Wu 2000兲.
as follows: • For the models previously mentioned, the prediction is less
accurate for the tube-encased cylinders, which typically have
兺 i⫽1
N
兩 E i兩
E av⫽ higher values of the ratio p u / f ⬘co due to the greater thickness
N of the tube compared to the wrap. The experimental set with
(4)
⬘
f cciTHEO ⬘ P
⫺ f cciEX cciTHEO ⫺ cciEX P the highest p u / f co⬘ values is that by La Tegola and Manni, with
E i⫽ •100 or E i⫽ •100 ⬘ ranging from 1.45 to 2.06. If this set is excluded, the
p u / f co
⬘ P
f cciEX cciEX P
average absolute error decreases approaching the value rela-
⬘ or cc relative to
where E i ⫽ percent error in the evaluation of f cc tive to the wrapped specimens.
the ith data point and N⫽ total number of data points. The latter • The remaining models give better predictions of the confined
may differ depending on the parameter ( f ⬘cc or cc ) and on the strength, with no appreciable difference between wrapped and
model under investigation. Average values, standard deviations, encased specimens. It is interesting to note that the models by
and coefficients of variation of the absolute percent errors of the Spoelstra and Monti 共1999兲 ‘‘exact’’ and by Saadatmanesh
various models in the prediction of strengths and strains are re- 共Mander兲 yield very similar results: the ratio between f ⬘cc / f co⬘
ported in Tables 4 –7. computed by the two models has an average value of 0.9637
with a standard deviation of 0.0347 共coefficient of variation
3.6%兲. This supports what was observed in the previous dis-
⬘
Prediction of f cc
cussion.
Fig. 5 illustrates the performance of some of the models in the • The model by Kono is penalized by the fact that it relates the
prediction of f ⬘cc . Fig. 6 summarizes results by showing the av- strength enhancement to the value of p u rather than to the
erage absolute error of the various models in the prediction of f ⬘cc p u / f ⬘co ratio. Since the empirical calibration was conducted on
for specimens confined by FRP wraps and tubes 共the specimens data with concrete unconfined strengths ranging between 32.3
confined by filaments are included in the wrapped ones兲. It ap- and 34.8 MPa, the model tends to overestimate f ⬘cc / f ⬘co for test
pears that: results with higher concrete strengths and to underestimate it
• The models by Fardis and Khalili 共1981兲; Miyauchi et al. for lower concrete strengths. This effect is superposed to the
共1997兲; Toutanji 共1999兲; and Xiao and Wu 共2000兲 largely over- ‘‘premature failure’’ effect. Since the model incorporates the
estimate the strength of most confined cylinders. The error is premature failure phenomenon with a reduction factor 共ratio of
larger for cylinders with heavy confinement, that is, for those the measured tensile strain in the FRP at failure to its uniaxial
ultimate strain兲 of about 0.5, it should be conservative when out considerably affecting the accuracy of the prediction. There-
applied to data with a reduction factor greater than 0.5, and fore it was decided that only the data sets for which co was
unconservative for the data with reduction factors smaller than known were suitable for comparison of the cc values with the
0.5. theoretical ones. This implies that the global data set used to
• The three most accurate models are those by Saaman et al. assess the prediction of cc is narrower than that used for the
共1998兲; Saafi et al. 共1999兲; and Spoelstra and Monti 共1999兲 assessment on f cc . Fig. 7 shows the average absolute error of the
‘‘approximate,’’ whose average absolute errors on the whole various models in the prediction of cc for specimens confined by
set of specimens are 13.4, 13.8, and 15.2%, respectively. FRP wraps 共or filaments兲 and tubes. It appears that:
• The errors in the prediction of cc are much larger than those
in the prediction of f cc . This result was also found from pre-
Prediction of cc
vious researchers assessing the accuracy of models on steel-
In most models, the expression of the axial strain at maximum confined concrete. This may partly be due to the larger scatter
stress, cc , involves the respective unconfined value co . There- of strain data as compared to strength data, but also to the
fore, in order for the experimental–theoretical comparison to be higher accuracy required when modeling the deformation
carried out, the value of co must be known. Furthermore, appli- characteristics of concrete rather than simply its strength prop-
cation of the model by Spoelstra and Monti 共1999兲 also requires erties. As previously discussed, the deformability of concrete
the initial modulus of elasticity of concrete, E co . Since the values under triaxial stresses is path-dependent, and therefore the
of co and E co are not always reported by the authors of experi- stiffness of the confining device has a remarkable influence on
mental studies, an attempt was made to correlate them to the the strain behavior. Furthermore, parameters such as type and
concrete strength 共as traditionally done at least for the elastic size of aggregates, mix proportions, and moisture content may
modulus兲 on the basis of the available data. Such correlation can- exert an influence and yet are very difficult to include in a
not be more than approximated, since other factors such as the model suitable for design purposes.
type of aggregate and its proportion within the mix also play a • In most cases, the error results from overestimating the strain
considerable role. For the elastic modulus, the equation by Ahmad at peak stress. This is unconservative as the ductility of the
and Shah 共1982兲 (E co ⫽3,950冑 f ⬘co ) was found to give the most confined cylinder is overestimated.
accurate prediction among several available ones, whereas the • Unlike the prediction of f ⬘cc , that of cc is less accurate for
others overestimated the concrete elastic modulus. Therefore, it wrapped than for encased cylinders. This is due to the fact
was decided that, when not available, the initial elastic modulus that, in wrapped cylinders, the experimental value of the lat-
of concrete be computed as a function of the unconfined strength eral strain in the FRP at tensile failure is mostly lower than the
according to the formula by Ahmad and Shah. For the value of FRP ultimate strain in uniaxial tension. This discrepancy af-
co , none of the available equations reasonably interpolated the fects the accuracy in the prediction of cc , in fact, the models
experimental data, nor can the usual value of 0.002 be used with- that mostly show a difference in accuracy between wrapped
wrapped specimens. The multiaxial stress state in the FRP con- eters and the reduction factor. They developed a tentative equa-
fining device is common to both wrapped and encased specimens. tion based on the experimental data reported in this study. Such
Even if load is not applied directly on the FRP, part of the axial equation gave an average absolute error of 24.7%, versus the 87%
load acting on the concrete is transmitted to the FRP by means of average absolute error given by a previous equation 共Matthys
bond stresses at the contact surface. However, the bond between 2001兲.
wraps and concrete is ensured by the impregnating resin whereas
that between the tube and the concrete surface is essentially based
⬘ with the Effective Fiber-Reinforced
Prediction of f cc
on friction and might be affected by shrinkage of the inner con-
Polymer Ultimate Strain
crete. Finally, the radius of curvature of the FRP should have an
influence on its tensile strength for both wraps and tubes. The Fig. 8 shows the average absolute error of the various models in
effect of corner radius on the tensile performance of FRP sheets predicting f ⬘cc for wrapped specimens when the experimental
has been experimentally investigated by Yang et al. 共2001兲. Re- value of the FRP hoop strain at failure 共which from now on will
sults indicated that the tensile strength of the laminate decreased be called ‘‘effective ultimate strain’’兲 and the uniaxial ultimate
with decreasing corner radius and was approximately equal to strain 共which will also be termed ‘‘reference ultimate strain’’兲 are
66% of the reference strength for the highest radius investigated, introduced into the equations. However, it should be kept in mind
equal to 51 mm 共diameter equal to 102 mm兲. To the writers’ that the database used in the assessment with the ultimate FRP
knowledge, no analogous studies are currently available for the strain is wider that used in the assessment with the effective
case of FRP tubes. strain, as only part of the authors reported the experimental hoop
The previous considerations seem to indicate that the most strain values. The following conclusions can be drawn:
important parameters affecting the ‘‘premature’’ tensile failure of • For FRP-wrapped specimens, since using the effective FRP
the confining wraps are the number of overlapping layers, the ultimate strain results in lower theoretical values of f ⬘cc , the
elastic modulus of the wrap, and the radius of curvature, that is, average absolute error decreases 共as compared to that obtained
for cylindrical specimens, the radius of the cylinder. All these with the reference ultimate strain兲 for the models that were
parameters are present in the confinement stiffness, E l . The same previously found to be unconservative, that is, those by Fardis,
parameters were identified, although through different thinking, Miyauchi, Toutanji, and Xiao. Conversely, the models that
by Matthys 共2001兲. However, other parameters may be influential were previously found to be the most accurate become too
on the phenomenon: the quality of workmanship, the degree of conservative and their absolute error increases. The data sets
straightness of the FRP fibers in the laminate, the type of laminate that mostly influence results are those by Shahawy et al.
used for wrapping 共for instance, prepregs might be easier to apply 共2000兲 and Rousakis 共2001兲, both characterized by very low
than dry sheets兲, and possibly the quality, thickness, and unifor- values of the reduction factor, and partly those by Harmon and
mity of the resin layer. Unfortunately such parameters are very Slattery 共1992兲 and Kono et al. 共1998兲. The points representa-
difficult to evaluate quantitatively and so is their influence on the tive of these results are clustered on the upper side of the 45°
premature rupture. De Lorenzis and Tepfers 共2001兲 attempted to line and deviate more and more from it as the theoretical
find a quantitative correlation between the three chosen param- f ⬘cc / f ⬘co increases. This is because, for tests with a given FRP
elastic modulus, the FRP axial stiffness and the p u / f ⬘co ratio Prediction of cc with the Effective FRP Ultimate Strain
both increase with the thickness of the confining layer. Ac-
cordingly, f ⬘cc / f ⬘co increases and the reduction factor decreases, Fig. 9 shows the average absolute error of the various models in
making theoretical predictions more and more conservative. predicting cc for wrapped specimens with both the ultimate and
• For FRP-encased specimens, since the reduction factor is close effective FRP hoop strains. Also in this case, it should be kept in
to the unity, no appreciable differences are noticed. The com- mind that the database used in the assessment with the ultimate
parison has to be made excluding the data by La Tegola and FRP strain is wider than that used in the assessment with the
Manni 共1999兲. Since these authors have not reported the val- effective strain. It appears that:
ues of the FRP hoop strain, their data cannot be included in the • For wrapped specimens, the average absolute error of all mod-
models assessment when the effective FRP strains are used. els shows a remarkable decrease when the FRP effective strain
On the other hand, it has been noticed that including such data is inserted in the equations. This confirms what was antici-
共characterized by the highest degree of confinement of the pated previously, that the accuracy in the prediction of cc is
whole database兲 in the assessment with the reference ultimate
influenced by the premature rupture of the FRP. Such influence
strain leads to higher prediction errors for the models that
is stronger than for the prediction of f ⬘cc , because of the higher
overestimate the effect of high lateral pressures. Therefore, if
sensitivity of the equations giving cc / co to the value of p u
the effect of the FRP ultimate strain being used in the compu-
tation has to be evaluated, the database on which the assess- and because of the wider range of percent variation of cc / co
ment is made must be the same. when compared to f ⬘cc / f ⬘co 共see Figs. 3 and 4兲. Since all models
• If the effective rupture strain of the FRP wrap were known in overestimate cc when using the reference ultimate strain of
advance, predictions of the strength enhancement by the exist- the FRP wrap, using the effective one always results in de-
ing models would not necessarily be more accurate than they creasing error.
are using the reference ultimate strain. Based on the overall • For encased specimens, since the reduction factor is close to
average absolute percent error, the most accurate models are the unity, only small differences are noticed due to the differ-
now those by Saaman et al. 共1998兲; Saafi et al. 共1999兲; and ent databases used for the two predictions.
Spoelstra and Monti 共1999兲 ‘‘exact’’ and Saadatmanesh et al. • If the effective rupture strain of the FRP wrap were known in
共1994兲, with average errors 16.9, 18.2, 19.0, and 19.5%, re- advance, the model by Toutanji would be the most accurate in
spectively. predicting cc , with an average absolute error of about 18%.
冉 冊
cc pu 0.80 were significantly larger than those on the prediction of f cc and in
⫽1⫹26.2 E l⫺0.148 for FRP wraps (6a) most cases on the unconservative side. No model was able to
co f ⬘co predict cc with reasonable accuracy, especially so for wrapped
cc
co
⫽1⫹26.2 冉 冊
pu
f ⬘co
0.68
E l⫺0.127 for FRP tubes (6b)
specimens and for models particularly sensitive to the value of
f u , due to the premature failure phenomenon. When the FRP
effective strain was inserted into the equations, the average abso-
Eq. 共6兲 provides a more accurate estimate of the experimental lute error of all models showed a remarkable decrease for
cc / co compared to that of the available models, with an average wrapped specimens, whereas for encased cylinders only small
absolute error of less than 20%. differences were noticed since the reduction factor was close to
the unity.
A new equation was proposed to evaluate the strain at peak
Conclusions stress as a function of the ultimate confinement pressure and of
the confinement stiffness. This equation provides a more accurate
The objective of this work was a systematic assessment of the estimate of the experimental cc / co ratio compared to that of all
performance of the existing models on confinement of concrete available models, with an average absolute error lower than 20%.
cylinders with FRP materials. Results of about 180 tests from 17 Accuracy of the models on FRP confinement of concrete still
different experimental sets were collected and classified. Then, needs to be improved. For this purpose, a standardized test
the existing empirical and analytical models were reviewed and method for the evaluation of the tensile properties of FRP lami-
the whole set of experimental results was compared with the nates is urgently needed. Stiffness and strength characteristics of
whole set of analytical models. The following conclusions were FRP materials are not as obvious to determine as those of homo-
drawn. geneous hysotropic materials, and test procedures are very influ-
It has often been remarked that steel-based models are not able ential on results. Furthermore, the premature failure phenomenon
to capture the behavior of FRP-confined concrete due to the fact of FRP wraps confining concrete cylinders still needs investiga-