Professional Documents
Culture Documents
Rotational Capacity of Steel I-Beams Under Fire Conditions-Numerical Simulation
Rotational Capacity of Steel I-Beams Under Fire Conditions-Numerical Simulation
www.elsevier.com/locate/engstruct
Received 27 June 2006; received in revised form 18 September 2006; accepted 24 November 2006
Available online 22 January 2007
Abstract
This paper describes the use of finite element models to study the inelastic behaviour of a steel I-beam in terms of its rotational capacity
at elevated temperature. Two main objectives of this study are to investigate the feasibility of applying the finite element method to study the
moment–rotation relationship of steel I-beams at elevated temperature and to investigate the main parameters affecting rotational capacity at
elevated temperatures through parametric study. The finite element (FE) model was validated against published test results [Lukey AF, Adams
PF. Rotation capacity of beams under moment gradient. Journal of the Structural Division, ASCE 1969;95(ST6):1173–88] at ambient temperature
and test results reported in Part I of the paper. It is demonstrated that the finite element analysis gives reasonable accuracy compared to test results,
providing an efficient, economical, and yet accurate tool to study the rotational capacity of beams in fire. An extensive parametric study was then
undertaken using this validated model. Finally, a simple moment–rotation relationship at elevated temperature was developed for use in design.
As a result of this study, the ductility of beams in fire can be based on the concept of member behavioural classes instead of the current EC3:1.2
[European Committee for Standardization (CEN). Eurocode 3: Design of steel structures, part 1.2: General rules — structural fire design. EN
1993-1-2. Brussels (Belgium); 2005] concept of cross-sectional classes, which tend to greatly oversimplify the problem. In addition, this paper
shows that plastic theory exercised with due care can also be applied to fire engineering design when it considers the more limited rotational
capacity of members under fire conditions.
c 2007 Elsevier Ltd. All rights reserved.
Keywords: Steel beams; Rotational capacity; Fire resistance; Moment–rotation; Local buckling; Numerical simulations
Table 1
Summary of validation results at ambient temperature
Source Test d tw b tf Li Mp Mm /M p ra
no. (mm) (mm) (mm) (mm) (mm) (kN m) Test FEA Test FEA
Dharma and Tan S3-1 276.00 8.00 163.00 10.00 1725 146.67 1.33 1.26 7.57 8.01
B-2 189.59 4.45 73.91 5.28 518.16 44.14 1.15 1.14 10.40 7.70
B-3 189.59 4.45 86.11 5.28 627.38 48.82 1.14 1.13 6.70 6.86
B-5 189.59 4.45 96.77 5.28 723.90 52.92 1.04 1.12 3.20 4.53
Lukey and Adams [1]
C-1 239.93 4.60 101.85 5.26 685.80 72.29 1.11 1.10 4.20 4.46
C-4 239.93 4.60 93.47 5.26 647.70 68.26 1.12 1.08 4.20 3.98
C-5 239.93 4.60 89.92 5.26 619.76 66.55 1.14 1.12 6.50 5.75
Table 2
Summary of validation results at elevated temperature
form of initial imperfection will be subsequently used in the exists between the FE predictions and the test results. There
parametric studies. is a slight difference in the non-linear pre-buckling region for
two beams, S1-1 and S3-2, which were tested at 415 ◦ C. This
3.2. Validation of FE model at elevated temperature discrepancy in response may be attributed to the differences
The FE model is also validated with 8 test results at between the stress–strain relationship proposed by EC3:1.2 [2]
elevated temperature conducted by the authors. The material and the actual stress–strain relationship. However, it should
stress–strain curve is taken from the EC3:1.2 [2] formulations be noted that the difference lies only in a certain part of
for steel at elevated temperatures. The initial geometric the stress–strain relationship relating to the proportionality
imperfection used in this FE model is taken from the measured limit or commencement of non-linearity. The initial stiffness
profiles of test specimens. There are two load steps: first, the and maximum capacity still show very good agreement and
temperature is increased to the desired failure temperature, and the predictions of maximum moment are within 2%. This
subsequently the load is applied. It should be noted that there discrepancy does not seem to occur on the beams tested
is no thermally induced compressive stress in the FE model. at 615 ◦ C, which shows that the stress–strain relationship
The rotational capacities and the ratios of maximum moment to proposed by EC3:1.2 [2] is indeed very close to actual the
plastic moment are compared with FE predictions in Table 2. stress–strain relationship of the tested beams. Besides the
Unlike the predictions at ambient temperature, the predicted stress–strain relationship, the excellent agreement between test
maximum moments at elevated temperature are generally less results and FE predictions is also attributed to the accuracy of
than 110% of the plastic moment at that temperature. This the initial geometric imperfection which has been incorporated
is observed in all specimens tested at elevated temperature into the FE models. A numerical study on the effect of initial
(Table 2). This is due to the fact that the effective yield strength imperfection by the authors shows that the initial geometric
at elevated temperature is defined at 2% strain [2] and the imperfection influences the initiation of local buckling, hence
characteristic stress–strain relationship of steel above 400 ◦ C affecting the inelastic rotational capacity [7].
has very little strain hardening above 2% strain. The rotational
Besides the loading–unloading response, the observed
capacity is much lower compared to ambient temperature
because the plastic rotation θ p at elevated temperature is much failure modes between tested specimens and numerical models
greater due to a highly non-linear stress–strain relationship at are also very close, as illustrated in Fig. 8. The model for S3-2
that temperature. shows local buckling failure (flange and web buckling), similar
Comparisons of FE predictions and test results are presented to the local buckling failure (Fig. 8(a)) observed after the tests.
in Figs. 6 and 7 for moment–rotation and load–deflection (mid- In addition, lateral torsional buckling failure, which occurred in
span) response, respectively. It can be seen that close agreement S2-1 is also predicted in the FE model as shown in Fig. 8(b).
2408 R.B. Dharma, K.-H. Tan / Engineering Structures 29 (2007) 2403–2418
In this section, the numerical model described in the In order to study the effects of different temperatures
previous section is used for parametric studies where four on rotational capacity, the nodal temperatures of RB2 and
reference beams, RB1 to RB4, are used. The simulation RB4 are varied from ambient temperature up to 800 ◦ C.
configuration, geometric properties, and restraint positions of Figs. 9 and 10 summarize the moment–rotational response at
these beams are shown in Table 3. These four beams have flange various isothermal conditions for RB2 and RB4, respectively.
and web slenderness ratios of plastic or compact sections. The The moment–rotation response has been plotted in a non-
lateral restraints are provided such that the LTB slenderness dimensional form, with the vertical axis showing the ratio of
ratio is around 13. Steel grade S275 (yield strength and elastic applied moment to plastic moment capacity at the respective
modulus at ambient temperature are 275 MPa and 210 GPa, temperature. The horizontal axis indicates the ratio of rotation
respectively) is used for these reference beams, which are to respective elastic rotation based on elastic theory. Similar
subjected to a uniform nodal temperature of 600 ◦ C. A few responses are observed from both RB2 (Fig. 9) and RB4
variables governing the rotational capacity of steel I-beams specimen (Fig. 10). At ambient temperature, both beams have
are varied, namely, temperature, flange and web slenderness large inelastic rotation with rotational capacity ra of 19 and
ratios, LTB slenderness ratio and steel grade. The influence 17, respectively. Besides, both beams possess plastic rotation
of increasing temperature will first be studied using RB2 and θ p which is very close to elastic rotation θe and relatively
RB4 beams. Secondly, RB1 and RB2 will be used to study high moment capacity of around 1.3M p . After the temperature
the effect of varying flange slenderness ratio. Thirdly, the web is increased to 300 ◦ C, the rotational capacity reduces
slenderness of RB3 and RB4 beams will be varied to study significantly. This is due to a reduction in the modulus of
this effect. Finally, the influence of effective length and steel elasticity (k E = 0.8) and the commencement of non-linearity
grade on the rotational capacity will be investigated using RB1 of the stress–strain relationship at around 0.613 f y (the yield
and RB3. The results of these parametric studies and some strength is still the same as ambient temperature) that cause
important observations are discussed in the following sub- the beams to attain their plastic moment capacity at a higher
sections. strain and non-linearity in the pre-buckling moment–rotation
R.B. Dharma, K.-H. Tan / Engineering Structures 29 (2007) 2403–2418 2409
Fig. 7. Load–deflection (mid-span) comparison of FEA and test results at elevated temperature.
Table 3
Summary of reference beams for parametric studies
T = 600 ◦ C.
Steel grade: S 275.
response. Besides, the value of Mm /M p has also decreased in the values of Mm /M p for the RB2 and RB4 specimens
due to a reduction in the strain hardening modulus. Similarity (ranging between 1.06 and 1.09) can be observed for cases
2410 R.B. Dharma, K.-H. Tan / Engineering Structures 29 (2007) 2403–2418
the moment–rotation response. Thus, it should be noted that buckling occur earlier in beams with a higher steel grade as il-
improvement of the rotational capacity can be achieved by lustrated in Fig. 18 for RB1-S235 and RB1-S450. This is obvi-
reducing the effective length of beams in which the lateral ous as while the elastic modulus for higher steel grade is about
torsional buckling failure mode is still observed. the same, the yield strength is increased substantially, thus forc-
ing steel beams to buckle earlier. This earlier loss of stability for
4.5. Influence of steel grade beams with higher grade steel contributes to limited available
inelastic rotation.
The influence of steel grade on the rotational capacity has
been studied by varying the yield strength of RB1. Four steel 5. Development of design moment–rotation relationship
grades are compared, namely, S235, S275, S355 and S450. It
can be seen in Fig. 17(a) that a higher steel grade results in less In this section, the development of a design moment–
rotational capacity. The main reason is due to the differences in rotational relationship at elevated temperature is described.
the stress–strain relationships among the different steel grades. Since the purpose of current research is to quantify the
Another reason is that by using a higher steel grade, a higher rotational capacity of steel I-beams to be used in plastic design,
plastic moment capacity is obtained; hence plastic rotation will the design moment–rotation relationship is only developed for
be attained at a higher rotation as shown in Fig. 17(b). As a beams which are able to achieve their plastic moment capacity.
result, less inelastic rotation is available for beams using higher Slender beams which lose their capacity prematurely are not
steel grades. Unlike Fig. 17(a) which shows the normalised considered. The moment–rotation relationship comprises three
rotation θ/θe in the x-axis, Fig. 17(b) shows the rotation θ in parts: a non-linear pre-peak curve, a horizontal plateau at the
the x-axis. plastic moment capacity, and an unloading curve. The proposed
A closer examination of lateral web and flange displace- design equations offer a quick, accurate, and practical method
ments indicates that both lateral torsional buckling and flange for structural engineers.
2414 R.B. Dharma, K.-H. Tan / Engineering Structures 29 (2007) 2403–2418
Table 4
Statistical analysis results of the ultimate rotation regression model
Galambos [10] and Kemp [11]. However, the complexity of Finally, the best regression model is selected based on the
yield-line plastic collapse mechanisms and the inability of the following criteria:
torsional buckling of restrained plate approach to plot the post-
• R 2 value, which describes the proportion of variation in the
buckling moment–rotation response (rotation capacity is only
response data explained by the predictors in the regression
determined up to the occurrence of buckling), led the authors
model;
to quantify the unloading portion of the moment–rotation 2 value, which is a modified version of R 2 value which
• Radj
response using a multi-linear regression fit to the finite
takes into account the number of predictors in the model;
element predictions. Both the ultimate rotation θu , which
• C– p value for assessing how well the model fits the data;
determines the start of the unloading curve, and the unloading
and
moment–rotation equation need to be determined.
• s value, which is the error of standard deviation.
In order to cover a wider range of parameters and meet
The best model, which has high R 2 and Radj 2 values, small
statistical requirements, additional simulations have been
conducted. A total of 71 simulations have been performed in s value, and C– p value close to the number of predictors, is
this study with a range of parameters as below: selected.
The best regression model to predict the ultimate rotation θu
4.0 ≤ c/t f ≤ 13.0; 24.0 ≤ d/tw ≤ 81.0; as a result of the analyses is
11.0 ≤ λ L T ≤ 29.0; 275 MPa ≤ f y ≤ 355 MPa; s s
θu ky
r
c d
400 ◦ C ≤ T ≤ 800 ◦ C (7) = 8.21 − 1.26 − 0.14 γf γw
θe kE tf tw
in which λ L T is the lateral torsional buckling (LTB) slenderness s
ratio and T is temperature. θp
− 0.0518λ L T ≥ (8)
The LTB slenderness ratio λ L T is used as one of the main θe
parameters in this study since the authors find that this is
an important factor (supported by statistical and experimental in which γ f andpγw are flange and web yield strength factors,
evidence) which not only accounts for effective length, but respectively (= f y /275). The recommended range of use for
also represents many other minor factors that may affect this regression model is the same as the range of the data
the rotational capacity, such as the type of loading (bending (Eq. (7)). The statistical analysis results for the regression
moment diagram), and the ratio of beam width to depth [12]. model are shown in Table 4. It can be observed that the p-values
As a result, a high accuracy can be obtained even with a few for the constant and the three predictors are zero, which means
regression parameters. that the association between the response and all four terms are
Two most commonly used regression analyses, namely, statistically significant. The high R 2 value of 95.9% and very
stepwise regression analysis and best subsets regression low s value of 0.1 show that the model fits the FE data well.
In addition, the Radj2 value of 95.7%, which is very close to
analysis are employed to find the best regression model from
possible parameters/variables for the purpose of predicting the the R 2 value, indicates that excessive number of terms is not
ultimate rotation θu . Firstly, stepwise regression analysis is the reason why the model fits the data well. This conclusion is
used to add and remove potential variables to the regression supported by the C– p value of 4.0, which is close to the number
model for the purpose of identifying useful subsets of the of predictors in the model.
predictors. This regression analysis is effective in reducing the In addition to this statistical analysis, the normal probability
number of potential variables. Subsequently, a few important plot of the residuals and the plot of residuals versus fitted
parameters that have been identified are analysed using best values are shown in Fig. 20. It can be observed that the normal
subsets regression. This best subsets analysis generates possible probability plot of the residuals roughly follow a straight line,
regression models using the maximum R-value criterion by first which indicates that the residuals are normally distributed. In
examining one-predictor regression models, and then selecting addition, the residuals are scattered randomly about zero in
the two models giving the largest R-value. This process Fig. 20(b), which means that the variance is approximately
continues until the model contains all the essential predictors. constant. Further investigation of this plot indicates that there
2416 R.B. Dharma, K.-H. Tan / Engineering Structures 29 (2007) 2403–2418
(a) Normal probability plot of the residuals. (b) Plot of the residuals versus fitted values.
Fig. 22. Validation of design model with FEA and test results.
[4] MSC. Marc Volume A: Theory and user information. 2005. [9] Gioncu V, Petcu D. Available rotation capacity of wide-flange beams and
[5] Tan KH, Ting SK, Huang ZF. Visco-elasto-plastic analysis of steel frames beam–columns. Part 1. Theoretical approaches. Part 2. Experimental and
in fire. Journal of Structural Engineering ASCE 2002;128(1):105–14. numerical tests. Journal of Constructional Steel Research 1997;43(1–3):
[6] Huang ZF, Tan KH. Structural responses of axially restrained steel 161–217; 219–44.
beams with semirigid moment connection in fire. Journal of Structural [10] Lay MG, Galambos TV. Inelastic beams under moment gradient. Journal
Engineering ASCE 2005;131(4):541–51. of the Structural Division ASCE 1967;93(ST1):381–99.
[7] Dharma RB. Buckling behaviour of steel and composite beams at elevated [11] Kemp AR. Interaction of plastic local and lateral buckling. Journal of
temperatures. Ph.D. thesis. Singapore: Sch. of Civ. and Env. Engrg., Structural Engineering ASCE 1985;111(10):2181–96.
Nanyang Technological University; 2006. [12] Dharma RB, Tan KH. A numerical study of rotational capacity of steel
[8] Climenhaga JJ, Johnson RP. Moment–rotation curves for locally buckling beams in fire. In: Proceedings of the fourth international conference on
beams. Journal of the Structural Division ASCE 1972;98(ST6):1239–53. advances in steel structures, vol. 2. 2005. p. 981–9.