ARTICLE IN PRESS

Journal of Constructional Steel Research ( )

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

journal homepage: www.elsevier.com/locate/jcsr

A design model for stainless steel box columns in fire

B. Uppfeldt a, , T. Ala Outinen b , M. Veljkovic c
a b c

The Swedish Institute of Steel Construction, Vasagatan 52, SE-111 20 Stockholm, Sweden VTT, Finland Lule University of Technology, Sweden

article

info

a b s t r a c t
A study of stainless steel cold-rolled box columns at elevated temperatures is presented, which is a part of an on-going RFCS project Stainless Steel in Fire, 20042007. Experimental results of six, class 4, stub columns at elevated temperature, tested by Ala-Outinen [Members with Class 4 crosssections in fire: Work package 3, ECSC project stainless steel in fire. Contract no. RFS-CR-04048, Espoo, Finland; 2005], were used to evaluate the finite element (FE) model. The FE analysis obtained using the commercially available software, ABAQUS, shows that the critical temperature was closely predicted. Further, a parametric study was performed using the same numerical model. This was a basis to check the quality of prediction of a newly proposed improvement for design rules of class 4 cross-sections in fire according to EN 1993-1-4 [Eurocode 3 - Design of steel structures Part 1-4: General rules Supplementary rules for stainless steels, CEN; 2006] and EN 1993-1-2 [Eurocode 3: Design of steel structures Part 1.2: General rules Structural fire design, CEN; 2005]. 2008 Elsevier Ltd. All rights reserved.

Article history: Received 21 November 2007 Accepted 13 May 2008 Keywords: Stainless steel Buckling Column Fire

1. Introduction Several recent research projects and published results in Karlstrm [14], and Ng and Gardner [15], have shown that stainless steel performs better than carbon steel at elevated temperatures. The improved behaviour is mainly explained by the enhanced material properties and a favourable relationship between strength and stiffness that makes stainless steel less prone to buckling in fire. This behaviour is taken into account in EN 1993-1-2 [9] for flexural buckling but not for local buckling. This makes the Eurocode treatment inconsistent and leads to conservative results, especially for slender cross-sections. Ng and Gardner [15] have presented recommendations for design guidance that include the relationship between strength and E -modulus for both the local slenderness and the global slenderness, as well as for the cross-section classification. All available test results have then been used to propose a revised semi-empirical buckling curve. A series of tests has been performed on rectangular hollow sections, steel grade EN 1.4301, in order to investigate the effect of local buckling. A small length of the test specimens (the modified slenderness is 0.2) was used to avoid global flexural buckling. The centric compression tests were performed for two different

rectangular hollow sections, RHS 200 200 5 and RHS 150 150 3. 2. Experiments 2.1. Material properties at ambient temperature The material properties at ambient temperature were determined and tests were performed for both hollow sections. Four tensile tests for both types of cross-section were carried out. The test pieces were taken from the web face (adjacent to the welded face) and the flange face (opposite the welded face), not including the welded seam. All tensile tests were repeated twice. The material properties were determined from the coupon tests. The measured strength values of the flange and the web of both crosssections are higher than the nominal values of yield strength according to the steel mill certificates; see Table 1. Global imperfections and the flatness of faces of all specimens were measured. The small differences in dimensions and small deviations in straightness were also measured. 2.2. Tests at ambient temperature The RHS specimens were tested at ambient temperature with the same load equipment as at elevated temperatures. The tests were repeated twice for each cross-section. The specimens and test arrangement are shown in Fig. 1. The strain was measured at midlength with four strain gauges. The longitudinal displacements at

Corresponding address: The Swedish Institute of Steel Construction, Vasagatan 52, 4 tr, 111 20 Stockholm, SE, Sweden. Tel.: +46 8 6610254; fax: +46 8 24 54 64. E-mail address: bjorn.uppfeldt@sbi.se (B. Uppfeldt).
0143-974X/$ see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.jcsr.2008.05.003

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

Journal of Constructional Steel Research (2008),

ARTICLE IN PRESS
2 Table 1 Test results coupon tests at ambient temperature Specimen Flange face Web face Flange face Web face
a

B. Uppfeldt et al. / Journal of Constructional Steel Research (

Profile RHS 200 200 5 RHS 200 200 5 RHS 150 150 3 RHS 150 150 3

Yield strength (0.2% strain) (MPa) 341 286 397 329

Yield strength, virgin sheet (0.2% strain) (MPa) 314a 363a

Average of two tests.

Table 2 Test results at ambient temperature Specimen A13 A16 B13 B16 Profile RHS 200 200 5 RHS 200 200 5 RHS 150 150 3 RHS 150 150 3 Length (mm) 900 900 900 900 Load (kN) 1129 1118 398 393

Fig. 2. Test set-up in furnace tests.

Fig. 1. Specimen types and test arrangement.

the front and back side were measured. The loss of load-bearing resistance at ambient temperature occurred very suddenly as a result of local buckling in the middle of the column. The test results at ambient temperature are presented in Table 2. 2.3. Tests at elevated temperatures A total of 6 tests was performed at the VTT for columns of class 4 cross-section. The columns were the same cross-sections, RHS 200 200 5 and RHS 150 150 3, as for the ambient temperature tests. The length (with end plates) of the columns was 900 mm. The columns were tested with fixed ends and centric compression load. Details of tests are summarised in Table 3, where the load level is the ratio between the maximum load at critical temperature and the column resistance at ambient temperature. 2.3.1. Test set-up The steel columns were heated in a model furnace specially built for testing loaded columns and beams. The test furnace is designed to simulate conditions to which a member might be exposed during a fire, i.e. temperatures, structural loads, and heat transfer. It comprises a furnace chamber located within a steel framework. The internal dimensions of the furnace chamber are: width 1500 mm, height 1300 mm and depth 1500 mm. The interior faces of the chamber are lined with fire-resistant bricks. Inside are four oil burners, arranged on the two walls of the furnace, containing two burners each.

The RHS cross-sections were tested at three different load levels. During the tests the axial load was kept constant and the furnace temperature was controlled to rise subsequently by 10 C/min. This rate was chosen to give a slow steel temperature rise in the failure phase of the columns. The furnace temperature was not controlled according to standard the EN 1363-1: 1999 (ISO 834-1) [7] standard fire curve as this would have given too high a temperature rise for unprotected steel columns. Measurements of the failure temperature would be very imprecise in such conditions. Unprotected columns were studied so the measurements could be accurately made. Loading was applied by a hydraulic jack, with capacity 2 MN, located outside the furnace chamber; see Fig. 2. A steel unit with circulating water was placed between the column and the loading jack. Sideways support of the steel unit was achieved by the furnace roof. The longitudinal displacement of the test specimen was measured using transducers attached to the water-cooled steel unit. The load was controlled and measured using pressure transducers. The temperatures of each column were measured at three cross-sections using a total of 12 chromelalumel thermocouples. The location and numbering of thermocouples of a specimen are shown in Fig. 3. The temperatures measured by furnace thermocouples were averaged automatically, and the average was used for controlling the furnace temperature. 3. FE model There are several important aspects of FE modelling, addressed in the modelling of experimental results both at ambient and elevated temperature. A suitable element type and its size is analysed in a so-called sensibility analysis. The finite element analysis (FEA) capability to predict the failure mode, and the maximum load and stiffness of the column, are prerequisites
Journal of Constructional Steel Research (2008),

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

ARTICLE IN PRESS
B. Uppfeldt et al. / Journal of Constructional Steel Research ( Table 3 Summary of tests for rectangular hollow sections RHS 200 200 5 and RHS 150 100 3 Specimen A11 A12 A15 B15 B11 B14 Cross-section RHS 200 200 5 RHS 200 200 5 RHS 200 200 5 RHS 150 150 3 RHS 150 150 3 RHS 150 150 3 Length (mm) 900 900 900 900 900 900 Material EN 1.4301 EN 1.4301 EN 1.4301 EN 1.4301 EN 1.4301 EN 1.4301 Load (kN) 694 567 463 248 203 165 Load level 0,62 0,5 0,41 0,63 0,51 0,42 Test date 24.5.2005 19.5.2005 7.6.2005 10.5.2005 9.5.2005 12.5.2005 ) 3

Table 4 The FE mesh sensitivity study Element type S4R S4R S4R Element size (mm) 25 12.5 6.125 Element sizea Nu b 396 387 385 CPU time (s) 127 474 2227

b/6 b/12 b/24

Analysis was performed at ambient temperature and on specimen B13. a Relative to side length of the cross-section, b = side length. b Failure load from non-linear analysis.

Fig. 4. Corner regions.

results in significantly higher 0.2% proof strengths in the corner regions compared to the flat faces. Ashraf et al. [5] have proposed a formula, Eq. (1), to predict the strength of cold-formed corner regions, 0.2,c . This equation is independent of the production route, and it can be used both for roll-formed and press-braked columns. Prediction is based on 0.2% proof strength of a virgin sheet, 0.2,v , inner corner radius of the cross-section, ri , and sheet thickness, t ; see also Fig. 4.

0.2,c =
Fig. 3. Column specimen details, and location and number of thermocouples.

1.8810.2,v
ri 0.194 t

(1)

for a reliable model. The input data are of great importance, including the material properties. The influence of different parameters was studied and expected results were confirmed. Increasing the slenderness of the cross-section increases the size of the local imperfections, which reduces the failure load and the critical temperature significantly. Global imperfections are more important for slender columns. 3.1. Elements A general-purpose shell element, called S4R, within Abaqus/Standard was used in order to avoid limitations of plate thickness chosen for the modelling of the experiments. S4R is a four-node element with six degrees of freedom per node, and it is suitable for thick and thin shell element applications, ABAQUS/Standard manual [1]. An element size of b/6 was chosen based on the results from the sensitivity study, Table 4. 3.2. Material It is well established that the mechanical properties of stainless steel are strongly influenced by the amount of cold-working. This

The extent of the corner properties, in addition to the importance of the material properties in the corner regions, is essential for the FE models to predict the critical temperature. Gardner and Nethercot [11] have found that extending the corner properties to 2t beyond the curved portions of the cross section, Fig. 4, gives the best agreement with test results. Zhao and Blanguernon [17] have defined reduction factors for cold-worked material and concluded that for temperatures below 700 C the use of the reduction factors of the annealed material lead to conservative results. For instance, at 600 C the 0.2% proof strength differs by more than 20% for EN 1.4571 C850 compared to the annealed grade. These large differences for cold-worked material indicate that cold-forming affects the material properties at elevated temperatures. Ala-Outinen [2] tested both virgin sheet and corner material from cold-rolled square hollow sections made of EN 1.4301. It was concluded that the cold-formed material performs better at elevated temperatures compared to annealed material at temperatures below 700 C; see Table 5. A comparison of experimental results and results of FE predictions with different corner properties is presented in Table 6. For the FE model the reduction factors from EN 1993-1-2 [9] were used for the material in the flat faces and the reduction factors
Journal of Constructional Steel Research (2008),

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

ARTICLE IN PRESS
4 Table 5 Comparison of reduction factors Temperature ( C) 20 100 200 300 400 500 600 700 800 900
a

B. Uppfeldt et al. / Journal of Constructional Steel Research (

Table 8 Sensitivity analysis of different imperfections at 800 C k0.2,p, b 1.00 0.83 0.74 0.69 0.66 0.59 0.52 0.43 0.30 0.18 k0.2,p, c 1.00 0.82 0.68 0.64 0.60 0.54 0.49 0.40 0.27 0.18 Local imperfection b/100 b/200 None Global imperfection L/500 n/a 125 kN 151 kN L/1000 114 kN 126 kN 156 kN None 114 kN 126 kN 172 kN

k0.2,p, a 1.00 0.91 0.88 0.83 0.80 0.70 0.64 0.42 0.28 0.10

Analysis performed on specimen A11. Table 9 Validation of the FE model Specimen B11 B14 B15 A11 A12 A15 Cross-section 150 150 3 150 150 3 150 150 3 200 200 5 200 200 5 200 200 5 Test temp ( C) 676 720 588 609 685 764 FEA temp ( C) 716 758 593 482 645 732 TempFEA /TempTest 1.06 1.05 1.01 0.79 0.94 0.96

Reduction factor for 0.2% proof strength of cold-formed corner regions of EN 1.4301 according to Ala- Outinen [2]. b Reduction factor for 0.2% proof strength of virgin sheet of EN 1.4301 according to Ala-Outinen [2]. c Reduction factor for 0.2% proof strength of annealed material according to EN 1993-1-2 [9]. Table 6 Comparison between test and FEA with different assumptions for material properties in the corner regions Section Test Nfi 200 200 5 567 FE analysis Temp 685 Tempa 520 Tempb 625 Tempc 645

Table 10 Results from FEA at ambient temperature compared to predicted failure loads according to EN 1993-1-4 [10] Cross-section Failure load EN 1993/FEA

= 0.5
200 200 4 200 200 5 300 300 5 1.06 1.05 1.08

= 0.8
1.06 1.10 1.04

= 1.2
0.96 1.00 0.95

Analysis performed on specimen A12, applied load N = 567 kN. a No strength enhancement of corner regions. b Assumed corner material properties from cold-forming, reduction factors the same as for annealed material according to EN1993-1-2 [9]. c Assumed corner material properties from cold-forming, reduction factors for cold-formed material according to Ala-Outinen [2]. Table 7 Sensitivity analysis of different imperfections at ambient temperature Local imperfection b/100 b/200 None Global imperfection (L = Length) L/500 n/a 398 kN 486 kN L/1000 382 kN 401 kN 489 kN None 382 kN 401 kN 610 kN

Analysis performed on specimen B13, Nu = 398 kN.

the lowest eigenmode and the measured magnitude of the local imperfections for modelling plated structures, good predictions of the failure load are obtained. The same approach is used in Karlstrm [14] for carbon steel columns in fire, and the small variations in failure load with different magnitudes of imperfections seen in Table 8 imply that this conclusion also holds for stainless steel at the elevated temperature. For the modelling of the tested stub columns only the measured local imperfections were used, and global imperfections were ignored. 3.4. Residual stresses Gardner and Nethercot [11] concluded that residual stresses cause a small reduction in stiffness but have little influence on the overall behaviour or the ultimate load-carrying capacity of stub columns. Therefore, no residual stresses were introduced in the modelling of the tested columns. 3.5. Validation The results obtained from FE predictions were compared to the results from the experiments; see Table 9. It is concluded that the FE model predicts the failure temperatures accurately for all tests except for specimen A11. The most probable explanation for the poor prediction of specimen A11 is that the measured local imperfections used in the FE model were different from the real local imperfections. Table 10 shows the important influence of the level of local imperfection on the ultimate load. The general conclusion is that the FE model is reliable for parametric study. 4. Parametric study The more general behaviour of thin walled stainless steel columns was investigated by performing a parametric study. The
Journal of Constructional Steel Research (2008),

derived from tests by Ala-Outinen [2] were used for the corner regions. The temperature-dependent thermal elongation given in Annex C in EN 1993-1-2 [9] was implemented in the FE model. Engineering stresses and strains were used to calculate the true stresses and logarithmic strains necessary for ABAQUS input data. 3.3. Imperfections The two types of geometrical imperfection that have to be considered are global imperfections and local imperfections. The most common way to model geometrical imperfections is by using a shape obtained from eigenvalue buckling analysis, adjusting the amplitude of the relevant imperfection, and implementing it in a non-linear analysis. The eigenvalue analysis gives a shape corresponding to global and local instability mode. It is assumed that a non-linear analysis using a combination of eigenmodes to model the imperfections will lead to conservative results. A sensitivity analysis was performed in order to evaluate the influence of geometrical imperfections on the maximum load. The magnitudes of both the local and global imperfections were varied. Tables 7 and 8 present the results of the sensitivity analysis. The level of the global imperfections on the maximum load is negligible because of the small differences in maximum load computed as long as the magnitudes of the local imperfections exist. Johansson and Veljkovic [13] have shown that by using

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

ARTICLE IN PRESS
B. Uppfeldt et al. / Journal of Constructional Steel Research ( Table 11 Results from FEA compared to predicted failure loads according to EN 1993-1-2 [9] for different slenderness Cross-section Failure load EN 1993/FEA ) 5 Table 13 Influence of load level on results for cross-section 300 300 5 for the Ng and Gardner [15] approach Load level (%) Failure load Ng and Gardner [15]/FEA

= 0.5
200 200 4 200 200 5 300 300 5 0.76 0.81 0.71

= 0.8
0.74 0.82 0.69

= 1.2
0.73 0.79 0.67 30 40 50

= 0.5
0.97 0.93 0.91

= 0.8
0.96 0.91 0.88

= 1.2
0.96 0.91 0.88

Load level 30% of ultimate load at the ambient temperature. Table 12 Results from FEA compared to predicted failure loads according to the design model proposed by Ng and Gardner [15] for different slenderness Cross-section Failure load Ng and Gardner [15]/FEA Table 14 Failure temperature from FEA with load level equal to 0.3 and failure times calculated with the standard fire curve (ISO 834) Cross-section Failure temperature ( C) 810 790 816 Failure time (min) 28.1 27.0 30.5

= 0.5
200 200 4 200 200 5 300 300 5 1.02 1.00 0.97

= 0.8
1.02 1.03 0.96

= 1.2
1.02 1.01 0.96

200 200 4 200 200 5 300 300 5

Load level 30% of ultimate load at the ambient temperature.

applied load levels, as well as the global slenderness and the local slenderness, were varied and the results were compared to the predicted strengths according to the EN 1993-1-2 [9] design model. To investigate a possible practical application of class 4 stainless steel columns, the parametric study was extended to include columns of length L = 3100 mm for all cross-sections and load levels. 4.1. FE model The validated FE model was used for the parametric study. However, due to the greater slenderness simulated compared with the experiments, global imperfections have to be taken into account. The local imperfection was taken as b/200, [18], and the global imperfection was taken as L/1000 in accordance with the permitted tolerances in prEN1090-2 [16]. The failure loads from the FE simulations at room temperature were compared with the ultimate loads calculated in accordance with EN1993-1-4 [10]; see Table 10. Nominal material properties, corner properties, and cross-sectional dimensions were used in the FE model. It can be seen that the FE model gives good agreement with the design method in EN1993-1-4 [10] for class 4 sections at room temperature. Ng and Gardner [15] have shown that using a global imperfection of L/2000 in the FE analysis gives the best prediction of all available test results. However, the use of a smaller imperfection than the permitted tolerance in prEN 1090-2 [16] seems unjustified in a parametric study. The end constraints were pinned for all columns, both at ambient temperature and at elevated temperature. It was assumed that the temperature distribution was uniform across and along the column. The failure loads from the FE simulations at ambient temperature were used to calculate the appropriate loads for each load level, cross-section, and slenderness used in the simulations at elevated temperatures. 4.2. Results The results from the parametric study were compared to the design model in EN 1993-1-2 [9], Table 11, as well as the design model proposed by Ng and Gardner [15], Table 12. It is clear that the design model according to the EN 19931-2 [9] predicts the failure load at elevated temperature with varying results, depending on the cross-section slenderness. The larger local slenderness leads to more conservative results. This is fully consistent with the conclusion by Ng and Gardner [15] and is a result of the Eurocode method neglecting the more

favourable relationship between strength and stiffness at elevated temperatures for local buckling. Results obtained using the Ng and Gardner [15] approach show that this model predicts the failure load better for various values of the local slenderness. When comparing the results from different load levels it is clear that it is important to account for this influence; see Table 13. A higher load level gives lower critical temperatures. From Table 13 it is obvious that the results obtained by the Ng and Gardner [15] design model become increasingly more conservative than the FE results as the load level increases. Furthermore, FEA of the columns with L = 3100 mm was considered to get an indication of the possiblity to use unprotected stainless steel class 4 columns in buildings. The results are shown in Table 14. It is clear that it is possible to use unprotected stainless steel class 4 columns for fire resistance class R30 if the load level is low. 5. Development of improved design model for class 4 crosssections The intention of the design model proposed for elevated temperatures is that it is valid even for ambient temperature. Therefore the buckling curve with imperfection factor, = 0.49, 0 = 0.4, is used at ambient and the limiting slenderness, temperature as it is given in EN 1993-1-4 [10]. The results from the parametric study clearly indicated the importance of taking the temperature-dependent relationship between strength and stiffness into account for local buckling as well as for global buckling. The basic form of the buckling curve given in EN 1993-14 [10], Eqs. (2a) and (2b), is used to further improve the design model. Apart from the local slenderness and the global slenderness being temperature dependent as proposed by Ng and Gardner [15], Eqs. (3) and (4), the limiting slenderness are suggested to depend on the strengthstiffness ratio according to Eq. (5).

2 +
2

0.5

(2a) (2b)

0 + 2 = 0.5 1 +

is the non-dimensional where is a imperfection factor; 0 is the non-dimensional slenderness where the slenderness; and reduction of the strength starts due to the slenderness. p, =
b/t 28.4

(3a)
Journal of Constructional Steel Research (2008),

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

ARTICLE IN PRESS
6 B. Uppfeldt et al. / Journal of Constructional Steel Research ( )

Fig. 5. Comparison of the proposed design model with FE simulations. Table 15 Results from FEA compared to the proposed design model at different load levels Cross-section Load level (%) Failure load proposed/FEA Fig. 6. Comparison of the SBI design model and FEA at elevated temperature, material EN 1.4571.

= 0.5
200 200 4 200 200 5 300 300 5 200 200 4 200 200 5 300 300 5 200 200 4 200 200 5 300 300 5 30 30 30 40 40 40 50 50 50 1.06 1.04 1.01 1.01 1.04 0.98 1.00 1.02 0.97

= 0.8
1.06 1.08 1.00 1.01 1.06 0.95 1.00 1.06 0.94

= 1.2
1.07 1.06 1.01 1.00 1.02 0.96 0.94 1.00 0.93

is the relevant width; t is the relevant thickness; and k is where b the buckling factor. =
kE , k0.2p ,
0.5

into the design model at the ambient temperature. This approach leads to more consistent agreement with results available from the tests. The mean values and the values of the coefficient of variation, COV, shown in Table 18, confirm the validity of the model. Experimental results obtained using stainless steel EN 1.4301 are considered in the paper, as shown in Table 17. However, 18 numerical simulations have been made to confirm that the proposed design model may be used for other grades as well. The simulations were made on cross-sections 200 200 4 and 300 300 5 at 400, 600 and 800 C with three different global slendernesses, = 0.5, = 0.8, and = 1.2. Material parameters that correspond to austenitic steel grade EN 1.4571 were chosen for the numerical analysis. The results presented in Fig. 6 give a mean value of Proposed/FEA = 0.97 and COV = 0.08. 6. Conclusions a/ Comparison between experiments at the elevated temperature and results obtained from FEA indicates that - assumptions for the influence of material properties in the corners according to Eq. (1) and Fig. 1 are realistic and necessary for a reliable FE parametric study; - assumptions for the shape and level of the local imperfection, b/200, and global imperfections, L/1000, are consistent with assumptions according to the established routines for non-linear analysis of columns at ambient temperature. b/ The analysis of 3.1 m long pinned columns in a standard ISO834 fire shows that it is possible to use unprotected stainless steel columns and fulfil requirement for resistance, R30. c/ Design recommendations for class 4 cross-sections made of austenitic stainless steel presented are fully coherent with EN 1993-1-1 [8] and EN 1993-1-4 [10] at ambient temperature. The comparison with other design models leads to the following conclusions. - The SBI model better predicts the retention of strength and stiffness of stainless steel than other design models, with a mean and COV of 0.95 and 0.24, respectively, compared to available tests [3,4,12]. - The same buckling curve used at ambient temperatures = 0.49 and 0 = 0.4 is recommended for use at elevated temperatures. 0 is dependent on the relation between the strength and stiffness at elevated temperature. - The relationship between strength and stiffness in the case of local buckling is taken into account in a rather realistic way by Eq. (3b). d/ The SBI design model gives a consistent approach to buckling at ambient temperature. The failure loads at elevated temperatures
Journal of Constructional Steel Research (2008),

(3b)

where is the material factor; kE , is the reduction factor for Youngs modulus; and k0.2p, is the reduction factor for 0.2 proof stress.

= 0,

k0.2p , kE ,

0.5

(4)
0.5

= 0

k0.2p , kE ,

(5)

The results for the proposed revised design model, called here the SBI model, are given in Table 15 and in Fig. 5. The results give a mean value of SBI/FEA = 1.01 with a coefficient of variation (COV) of 0.08. Compared to the design model in EN 1993-1-2 [9] and the Ng and Gardner [15] design model that give a mean of EN 1993/FEA = 0.73 with COV = 0.09 and Ng and Gardner/FEA = 0.96 with COV = 0.08, respectively, it is clear that the proposed design model gives improved predictions of the failure loads. In order to further prove the validity of the SBI design model, comparison was made with more test results available mainly in cross-section class 1, [3,4,12]. The results are compared with other available design models, [9,10,15]. The SBI design model uses the strength at 2% strain for cross-section classes 12 and the 0.2 plastic strain limit for cross-sections in classes 34. The summary of available test results is given in Table 16. The predicted resistances obtained by various design models are normalised by the maximum applied load obtained in the tests; see Table 17. The results clearly indicate that the SBI design model gives better predictions compared to the ultimate load obtained in the tests. The SBI design approach contains all parameters included

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

ARTICLE IN PRESS
B. Uppfeldt et al. / Journal of Constructional Steel Research ( Table 16 Cross-sectional dimensions, slenderness, applied load, and critical temperature for the tested columns Cross-section Measured section size (mm) b1 RHS 150 100 6a RHS 150 75 6a RHS 100 75 6a SHS 100 100 4a SHS 200 200 4a SHS 40 40 4 (T1)b SHS 40 40 4 (T2)b SHS 40 40 4 (T3)b SHS 40 40 4 (T4)b SHS 40 40 4 (T5)b SHS 40 40 4 (T7)b SHS 200 200 5 (A11)c SHS 200 200 5 (A12)c SHS 200 200 5 (A15)c SHS 150 150 3 (B15)c SHS 150 150 3 (B11)c SHS 150 150 3 (B14)a
a b c d e

Area, A (mm2 ) ri 4.5 4.5 4.5 4.0d 4.0d 4.0d 4.0d 4.0d 4.0d 5.0d 5.0d 5.0d 3.0d 3.0d 3.0d 2852 2556 1973 1467e 3067e 535 535 535 535 535 535 3835 3837 3835 1742 1743 1746

Slenderness,
0.49 0.65 0.65 1.27 0.51 1.17 1.17 1.17 1.17 1.17 1.17 0.13 0.13 0.13 0.07 0.07 0.07

Load, N (kN)

Critical temp, T ( C)

b2 99.0 74.2 78.6 100 200 40d 40d 40d 40d 40d 40d 200.0 200.2 199.9 150.5 150.6 150.7

t 6.23 6.23 6.2 4.0 4.0 4.0d 4.0d 4.0d 4.0d 4.0d 4.0d 5.0d 5.0d 5.0d 3.0d 3.0d 3.0d

148.9 149.9 99.4 100 200 40d 40d 40d 40d 40d 40d 199.9 199.9 200.0 149.7 149.8 150.1

268 140 156 80 230 45 129 114 95 55 75 694 567 463 248 203 165

801 883 806 835 820 873 579 649 710 832 766 609 685 764 588 676 720

Tests reported in [12]. Tests reported in [3]. Tests reported in [4]. Nominal values, no measurements taken. Area calculated with ri = 2t .

Table 17 Predicted resistance at critical temperature normalised by the applied load in tests for different design models Cross-section Cross-section classa Measured material properties (MPa) f0,2 fu 262 262 262 298 298 595 595 595 595 595 595 314 314 314 363 363 363 540b 540b 540b 540b 540b 736 736 736 736 736 736 623 623 623 654 654 654 Predicted resistance at critical temperature normalised by the applied load in test EN 1993 [9,10] Euro Inox [6] Ng and Gardner [15] 0.84 0.87 0.91 0.90 0.57 0.95 0.75 0.78 0.84 0.97 0.92 0.66 0.69 0.65 0.81 0.84 0.92 0.75 0.78 0.86 0.90 0.68 1.00 0.77 0.81 0.88 1.03 0.97 0.68 0.72 0.67 0.82 0.85 0.93 0.95 0.99 1.03 1.00 0.85 0.96 0.74 0.77 0.83 0.97 0.91 0.79 0.85 0.82 0.96 1.04 1.16

E 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b 200 000b

SBI 0.99 1.02 1.09 1.06 0.88 1.01 0.81 0.84 0.90 1.03 0.98 0.79 0.85 0.82 0.96 1.04 1.16

RHS 150 100 6 RHS 150 75 6 RHS 100 75 6 SHS 100 100 4 SHS 200 200 4 SHS 40 40 4(T1) SHS 40 40 4(T2) SHS 40 40 4(T3) SHS 40 40 4(T4) SHS 40 40 4(T5) SHS 40 40 4(T7) SHS 200 200 5(A11) SHS 200 200 5(A12) SHS 200 200 5(A15) SHS 150 150 3(B15) SHS 150 150 3(B11) SHS 150 150 3(B14)
a b

1 1 1 2 4 1 1 1 1 1 1 4 4 4 4 4 4

At ambient temperature. Nominal values, no measurements taken.

Table 18 The mean values and coefficients of variation for different design models EN 1993 [9,10] Mean All tested cross-sections Cross-section class 1-3 Cross-section class 4 0.82 0.87 0.74 COV 0.24 0.12 0.19 Euro Inox [6] Mean 0.83 0.87 0.77 COV 0.20 0.15 0.16 Ng and Gardner [15] Mean 0.92 0.92 0.93 COV 0.24 0.18 0.24 SBI Mean 0.95 0.97 0.93 COV 0.21 0.17 0.23

for class 4 cross-sections are predicted with a mean value of 1.01 and a coefficient of variation of 0.08 according to the parametric study. This is an improvement on the design model that is based on EN 1993-1-2 [9] and EN 1993-1-1 [8] as it is proposed by Ng and Gardner [15]. The difference between Ng and Gardner and the SBI model is only in the shape of the buckling curve. When 0,Ng and Gardner and 0,SBI the two models give the same results. Compared to available tests the SBI design model also shows better predictions of failure loads for cross-section classes 13.

Acknowledgement The authors are grateful to Dr. Leroy Gardner for his valuable help. References
[1] ABAQUS/standard manual. ABAQUS/standard users manual volumes I-III, version 6.2. Pawtucket: Hibbitt, Karlsson & Sorenssen, Inc.; 2001. [2] Ala-Outinen T. Fire resistance of austenitic stainless steels Polarit 725 (EN 1.4301) and Polarit 761 (EN 1.4571). VTT research notes 1760. Espoo, Finland; 1996. Journal of Constructional Steel Research (2008),

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

ARTICLE IN PRESS
8 B. Uppfeldt et al. / Journal of Constructional Steel Research ( ) [3] Ala-Outinen T, Oksanen T. Stainless steel compression members exposed to fire. VTT research notes 1864. Espoo, Finland; 1997. [4] Ala-Outinen T. Members with Class 4 cross-sections in fire: Work package 3, ECSC project stainless steel in fire. Contract no. RFS-CR-04048, Espoo, Finland; 2005. [5] Ashraf M, Gardner L, Nethercot DA. Strength enhancement of the corner regions of stainless steel cross sections. Journal of Constructional Steel Research 2005;61(1):3752. [6] Design manual for structural stainless steel. 3rd ed. Building series, vol. 11. Euro Inox and the Steel Construction Institute; 2006. [7] EN 1363-1: 1999 Fire resistance tests Part 1: General requirements, CEN European Committee for Standardization, Brussels, Belgium, 1999. [8] EN 1993-1-1. Eurocode 3: Design of steel structures Part 1-1: General rules and rules for buildings, CEN; 2005. [9] EN 1993-1-2. Eurocode 3: Design of steel structures Part 1.2: General rules Structural fire design, CEN; 2005. [10] EN 1993-1-4. Eurocode 3 - Design of steel structures Part 1-4: General rules Supplementary rules for stainless steels, CEN; 2006. [11] Gardner L, Nethercot DA. Numerical modelling of stainless steel structural [12] [13] [14] [15] [16] components A consistent approach. Journal of Structural engineering, ASCE 2004;130(10):1586601. Gardner L, Baddoo NR. Fire testing and design of stainless steel structures. Journal of Constructional Steel Research 2006;62:53243. Johansson B, Veljkovic M. Steel plated structures. Progress in Structural Engineering and Materials 2001;3(1):1327. Karlstrm P. Thin-walled steel studs in fire: Analysis and design recommendations, Licentiat Thesis, Lule University of Technology, Sweden; 2004. p. 73. Ng KT, Gardner L. Buckling of stainless steel columns and beams in fire. Engineering Structures 2007;29(7):71730. prEN 1090-2. Execution of steel structures and aluminium structures Part 2: Technical requirements for the execution of steel structures Stage 34, CEN; 2005. Zhao B, Blanguernon A. Member tests in fire and structural fire design guidance. Work package 6, ECSC project Structural design of cold-worked austenitic stainless steel. Contract No 7210-PR-318. The Steel Construction Institute, London, UK; 2004. EN 1993-1-5. Eurocode 3 rd -Design of steel structures Part 15: Plated structural elements, CEN; 2006.

[17]

[18]

Please cite this article in press as: Uppfeldt B, et al. A design model for stainless steel box columns in fire. doi:10.1016/j.jcsr.2008.05.003

Journal of Constructional Steel Research (2008),