You are on page 1of 8

Stiffness properties in timber structure analysis and design

Kolbein Bell
Professor
Norwegian University of Science and Technology, NTNU
Trondheim, Norway

Summary
The importance of the timber stiffness is investigated by means of an effective analysis and design
program for 3D frame type structures with geometrical nonlinear capabilities. The emphasis is on
ultimate load design. Results obtained by both linear and nonlinear analyses of simple columns and
beams, as well as of a timber bridge structure, are presented and discussed with reference to the new
Eurocode 5 and its specifications. While not conclusive, the results indicate discrepancies in the
code that should be further investigated, and the need for more work on geometrical imperfections in
connection with nonlinear analysis as basis for ultimate limit state design is pointed out.

1. Introduction
Most of the current timber codes assume the design of a typical timber structure to be based on linear
(1st order) elastic analysis. While not excluding nonlinear analyses, the codes contain few and rather
vague guidelines for the use of these methods. Even the recently published Eurocode 5 [1],
henceforth referred to as EC5, is fairly sketchy on this issue. In view of the rapid development of
powerful PC tools with nonlinear analysis capabilities, this situation leaves something to be desired.
We now have the analysis capability to do away with some of the cumbersome coefficients used to
smuggle higher order effects into linear analysis results.
The various factors influencing a static analysis depend on the structure itself - statically determinate
or indeterminate, and the number of different types of material used - to what purpose the results of
the analysis are to be used - serviceability limit state or ultimate limit state - and the type of analysis
- linear or nonlinear. Stiffness properties, of both members and joints, are always important, except
perhaps in the determination of section forces of statically determinate structures by linear methods.
For a nonlinear analysis aimed at providing section forces for an ultimate load design, the shape and
size of the geometrical imperfections are also important factors.
In order to limit its scope, this paper concentrates on the role of the member stiffness in ultimate load
design. Furthermore, we limit our discussion to solid timber and glulaminated timber (glulam) of
softwoods.
We start with a brief review of the material stiffness parameters, including some comments on what
various codes have to say about these parameters and their use in design. We then present some
results obtained using both linear and nonlinear analysis in combination with EC5, for both simple
column and beam problems and also for a real life structure (a small glulam arch bridge). Finally we
try to draw some conclusions from these results.

2. Timber stiffness parameters


In a computational model the stiffness of a member is determined by its geometry and its material
properties. The material parameters influencing the stiffness are the modulus of elasticity (E) and, to
a lesser degree, the shear modulus (G). In most codes the ratio E/G is taken to be constant, for solid
timber in the range 15 to 16, for all strength classes. Hence, we narrow the discussion to E.
The two most quoted values for E are its mean value, E mean and its fifth percentile value, E 0,05 . For
all strength classes EC5 specifies a constant ratio for E 0,05 / E mean , which for solid softwood is 0,67
and for glulam 0,81. For the most common strength classes, the characteristic values for E mean vary
from 7000 Mpa (C14) to 14000 Mpa (C40) for solid timber, and from 12600 Mpa (GL24c) to 14700
Mpa (GL36c) for glulam.
As a material parameter, E is less affected by load duration and moisture content than the strength
parameters. However, in a typical static analysis, E is the only parameter that can account for creep
effects, and most codes have fairly detailed and specific provisions for calculating displacements in
the serviceability limit state. This is solved by either multiplying displacements obtained with non-
modified E-values with a creep factor (that depends on load duration and moisture content) or, more
commonly, by modifying the E-values themselves with such creep factors.
For the ultimate limit state design, however, most codes seem to be less precise when it comes to
which value of E should be used in the determination of internal force distribution. For instance, the
Australian, Canadian and New Zealand timber codes do not seem to distinguish between linear and
nonlinear analysis, and for strength purposes (ultimate limit state) they seem to suggest that the static
analyses should be based on the mean value of E.
For a linear analysis of a statically determinate structure the value of E is of no importance for the
distribution of internal section forces and moments. The same can be said for an indeterminate
structure if all its members are made of the same timber material. If more than one material is
involved the value of E will influence the internal force distribution, but probably not a great deal.
A design based on linear analysis will have to resolve the problem of stability of members through
the use of factors compensating for the higher order moments. In EC5 for instance, columns
subjected to combined compression and bending need to satisfy the expressions
σ c,0,d σ m,y,d σ m,z,d
----------------
- + ----------- -≤1
+ k m ---------- (1)
k c,y f c,0,d f m,y,d f m,z,d
σ c,0,d σ m,y,d σ m,z,d
----------------
- + k m ----------- + ----------- ≤ 1 (2)
k c,z f c,0,d f m,y,d f m,z,d
Here σ designates design stress, index c for compression and m for bending, f designates strength, km
is a stress reduction factor (equal to 0,7 for a rectangular cross section), while k c,y and k c,z are the
“instability factors” accounting for higher order effects (moments). The instability factors, which are
defined by fairly complicated formulas, depend on the slenderness ratios, i.e. the buckling lengths
and the radius of gyration, and the member stiffness. The background for the instability factors of
EC5 is given by Blass [2]. It is interesting to note that the stiffness parameter appearing in the
formulas for the instability factors is E 0,05 , and that there is no mention of any effect of load duration
or moisture content. In other words, the instability factors are independent of load duration and
moisture content. Is that reasonable?
The internal force distribution obtained by a nonlinear analysis is, regardless of the type of structure,
dependent on the member stiffnesses, and hence their E-values and in some cases also their G-
values. The simplest illustration of this is the standard simply supported column subjected to a
centric axial compression. Assuming a geometric imperfection, in the shape of a half sine-wave with
an amplitude in the range of L/300 to L/500, this deviation from straightness will give rise to a
bending moment which in turn will increase the lateral deflection, and more so the smaller E is. This
increase in deflection in turn gives rise to an increase in the moment and so on until equilibrium
between internal and external forces is reached. We shall see in the next section that the role of E in
this simple example is quite significant, as is of course also the magnitude of the imperfection.
Eurocode 5 [1] is quite specific on the issue of stiffness parameters. For ultimate limit state design it
states in clause 1 of section 2.2.2:
The analysis of structures shall be carried out using the following values for stiffness properties:
- for a first order linear elastic analysis of a structure, whose distribution of internal forces is not
affected by the stiffness distribution within the structure (e.g. all members have the same time-
dependent properties), mean values shall be used;
- for a first order linear elastic analysis of a structure, whose distribution of internal forces is
affected by the stiffness distribution within the structure (e.g. composite members containing
materials having different time-dependent properties), final mean values adjusted to the load
component causing the largest stress in relation to strength shall be used;
- for a second order linear elastic analysis of a structure, design values, not adjusted for duration
of load, shall be used.
The final mean value referred to is defined as
E mean,fin = E mean ⁄ ( 1 + ψ 2 k def ) (3)
and the design value is defined as
E d = E mean ⁄ γ M (4)
Here ψ 2 is “the factor for the quasi-permanent value of the action causing the largest stress in
relation to the strength (if this action is a permanent action, ψ 2 should be replaced by 1)”, k def is “a
factor for the evaluation of creep deformation taking into account the relevant service class”, and γ M
is “the partial factor for a material property” (equal to 1,3 for solid timber, and 1,25 for glulam). It
should be noted that for a permanent loading in service class 3, E mean,fin = E mean ⁄ 3 .
The difference between the design value E d and the fifth percentile value, E 0,05 , is very small (about
13% for solid timber and only 1% for glulam), and an obvious question then is why not use E 0,05
instead of E d ? After all, the instability factors that compensate for the higher order effects, when
using a linear static analysis model, are based on E 0,05 . There are other questions too that could be
raised already, but we await some results.

3. Some simple column and beam examples


In this section we present some results obtained for simple structural members with a computer
program, FrameIT [3], developed in the course of a doctorate study [4]. This is a fully nonlinear
(large displacement) analysis program for 3D frame type structures with a built-in timber code check,
according to either the Norwegian code or Eurocode 5. All results presented here are based on EC5.
Shear deformations are included in all analyses and for glulam the ratio E/G is taken to be 17,5.

3.1 Glulam column


Figure 1 shows a simply supported glulam column with a rectangular
P cross section. Two different lengths are considered, L = 4000 mm
(medium slenderness) and L = 6000 mm (large slenderness). The
column is loaded by 1) a single, centric compressive force P, and 2) the
same plus a uniformly distributed transverse force p = 1 kN/m (causing
cross section bending about the strong axis z).
by
Glulam of strength class GL36c is used with the following characteristic
(Norwegian) strength parameters:
bz
f m,k = 36 Mpa, f t,0,k = 22,5 Mpa, f c,0,k = 29 Mpa, f v,k = 3,0 Mpa
by = 180 mm
The shear strength is somewhat smaller than specified by EN 1194 (3,8),
bz = 140 mm but otherwise there is no difference. We also have:
L p E mean = 14700 Mpa and E 0,05 = 11900 Mpa.
The capacity is, in all cases, checked by expressions (1) and (2). If a
nonlinear analysis is used as the basis for design, the instability factors
are “neutralized” by setting them equal to unity.

3.1.1 Axial compression only (p = 0)


For the sake of argument we assume the following (extreme) conditions:
service class 3 (“wet condition”) and P is a long term action. Hence:
E d = 11760 Mpa and E mean,fin = 4900 Mpa (conservative value)
A linear analysis in this case means a straightforward use of (1) and (2)
with both bending stresses equal to zero. We find that the column can at
Figure 1 Glulam column most carry a load of P max = 122,7 kN for L = 4000 mm, and
P max = 56,5 kN for L = 6000 mm, in order to satisfy (1) and (2).
Next we consider a design based on a nonlinear analysis with a geometric imperfection in the shape
of the first buckling mode (a half sine wave in the z-direction) and with a maximum amplitude of 10
mm for L = 4000 mm and 15 mm for L = 6000 mm (corresponding to L/400). Results are shown in
tables 1 and 2 for different values of E. Pmax is the largest load that will satisfy expressions 1 and 2
(found by trial and error), M y is the higher order moment about the y-axis, and δ is the additional
lateral displacement (in z-direction). e is the maximum geometrical imperfection.

Table 1: Column capacity for axial compression - nonlinear analysis with L = 4 m and e = 10 mm

E [Mpa] 5000 6000 8000 10000 12000 15000

Pmax [kN] 106,8 122,3 147,6 166,4 180,0 194,4

M y [kNm] 6,86 6,36 5,54 4,95 4,50 4,06

δ [mm] 54,3 42,0 27,6 19,8 15,0 10,9

Table 2: Column capacity for axial compression - nonlinear analysis with L = 6 m and e = 15 mm

E [Mpa] 5000 6000 8000 10000 12000 15000

Pmax [kN] 51,4 60,3 76,8 91,4 104,2 120,5

M y [kNm] 8,87 8,34 7,81 7,33 6,92 6,40

δ [mm] 153,6 123,5 86,8 65,3 51,4 38,1

It should be emphasized that E is the only variable parameter here. The strength parameters are the
same for all columns of the tables. Clearly something is amiss. Using the specified stiffness for a
nonlinear analysis, that is E d which is very close to 12000, we see that the nonlinear approach gives
us almost 50% higher load than the (standard) approach with instability factors for the shorter
column, and almost 85% higher for the longer column. If we halve the stiffness (to 6000 Mpa) we
see that the two approaches give very similar results. One can always argue that the imperfections of
10 and 15 mm, respectively, are too small, although they seem to be in line with other imperfections
specified by EC5. In order for the nonlinear approach to give the same answer as the “linear”, with
an E of 12000 Mpa, we would need an imperfection amplitude e of 30,5 mm for the shorter column
(equivalent to L/130), which seems excessive. These results do not necessarily demonstrate that
EC5 is conservative for this column case. They do, however, demonstrate a discrepancy between the
two methods, both of which are allowed. And it is a bit worrying that the current EC5 will produce
higher order moments that are independent of load duration and moisture content, irrespective of a
linear or nonlinear approach.
Before we leave this example it should be pointed out that the results are not quite as dramatic as
they may seem. EC5’s standard approach, based on instability factors ( k c,y and k c,z ), implies a
linearization; if we halve the load, 50% of the capacity is used. Not so for the nonlinear approach;
here a small increment in the load around full capacity will cause a significant change in utilization.

3.1.2 Combined axial force and bending ( p = 1 kN/m)


We now repeat the analyses for the longest column ( L = 6 m), the only difference being that we now
have a constant uniformly distributed load p = 1 kN/m acting in the y-direction (causing bending
about the strong axis) in addition to the axial load P. The “nonlinear results” are shown in table 3.

Table 3: Column capacity for combined bending and axial compression for L = 6m;
nonlinear analysis with a maximum imperfection of 15 mm in the z-direction

E [Mpa] 5000 6000 8000 10000 12000 15000

Pmax [kN] 44,0 50,6 62,3 72,1 80,5 90,7


p [kN/m] 1,0 1,0 1,0 1,0 1,0 1,0
With a linear analysis we find that, in order to satisfy formulas (1) and (2), the column can at most
carry an axial force of P max = 43,0 kN in addition to the transverse load p. We see the same
discrepancy here between the two approaches. It should be mentioned that an imperfection in the
form of the 2nd buckling mode, which is half a sine wave in the y-direction, gives a higher axial
force P than shown in the table, for all values of E.

3.2 Glulam beam


Figure 2 shows a 6 m long, simply supported
glulam beam with a rectangular cross section.
b = 90 mm It is subjected to a uniformly distributed load p
h = 450 mm applied at the (neutral) beam axis. At the
z supports, the beam is free to rotate about the y-
b and z-axis, but it is completely prevented from
p rotation about its own (x-) axis.
y x The material quality, load duration and service
h L class are as for the column in the previous
section, that is GL36c, long term and 3.
L = 6000 mm
This beam is prone to lateral torsional
buckling, and EC5 [1] has provisions for this.
Without going into detail (the interested reader
Figure 2 Simply supported glulam beam may find them in section 6.3.3) the code
defines a parameter ( k crit ), similar to the
instability factors ( k c,y and k c,z ) of the previous section, to account for the higher order effects. For
bending, without presence of axial force, the code requires the design bending stress to satisfy
σ m,d ≤ k crit f m,d (5)
k crit depends on the relative slenderness for bending ( λ rel,m ) which in turn depends on several cross
sectional parameters, stiffness parameters, taken as the 5 percentile values of E and G, and the so-
called effective length of the beam. The latter depends on the support conditions and the load
configuration, and (according to EC5) is equal to 0,9 × L = 5400 mm in our case.
Assuming first that lateral buckling of the beam is fully prevented (through bracing), that is k crit = 1,
we find that it can at most carry a load of pmax = 11,0 kN/m. If there is no bracing at all, formula (5)
suggests a maximum load of p max = 8,26 kN/m .
Next we consider a design based on a nonlinear analysis with a geometric imperfection taken to have
the shape of the first lateral torsional buckling mode, with a maximum amplitude of 15 mm
(equivalent to L/400). Care must be taken to use the correct mode since the eigenvalue analysis will
produce eigenmodes in pairs, with eigenvalues of opposite sign, see figure 3 in which the first two
modes (for E = 12000 Mpa and G = 686 Mpa) are shown. It should be noted that G is now an
important parameter; as stated earlier it is taken to be equal to E/17,5 in all cases.

mode # 1 mode # 2

buckling load factor = -1,9215 buckling load factor = 1,9215

Figure 3 The two lowest buckling modes of the beam - E = 12000 Mpa and p = 10 kN/m

The capacity is now checked by setting k crit equal to unity in expression (5). However, we now have
bending about both cross sectional axes, and we therefore need to use expressions (1) and (2), without
the axial stress term. Table 4 shows the maximum load pmax that the beam can carry according to a
nonlinear static analysis and expressions (1) and (2), for various values of E. Again the code’s
approach, based on a linear analysis and expression (5), seems to be conservative compared to the
nonlinear based design. However, the stiffness does not seem to be quite as important as in the case
of the column. In order for a nonlinear based design to give the same result as expression (5), the

Table 4: Beam capacity - nonlinear analysis with a maximum imperfection of 15 mm (mode # 1)

E [Mpa] 5000 6000 8000 10000 12000 15000

pmax [kN/m] 7,31 8,18 9,20 9,62 9,84 10,01

imperfection amplitude needs to be 50 mm, for E = 12000 Mpa.

4. A real life structure

T
11,70 m
710 × 800 mm
F T
hanger C
cross-
beam
deck-beams
E A
m 3 2 1
526
D = 26, z
305
3 B
5 × 5, cross- 273 kN
2 1 beam
y x
deck-beams 2,35 3,95 3,00 2,40

Figure 4 A glulam arch bridge for ordinary road traffic

Figure 4 shows a glulam arch bridge with a span of about 26,5 m. The bridge is very similar, but not
identical to one of the side spans of a bridge recently built in Norway (Tynset bridge). It has two
lanes for vehicle traffic (deck-beams 1 and 2), and one for pedestrians and cyclists (deck-beam 3).
The traffic lanes consist of an asphalt covered stress laminated timber deck on steel cross-beams.
The cross-beams are carried by two 3-hinge glulam arches, via vertical hangers. The four middle
hangers are steel columns rigidly connected to the cross-beams, forming two U-shaped frames (one
of which is C-D-E-F). These two frames provide the sidewise stiffening of the arches. The 4 end
hangers (like A-B) consist of round steel with little or no bending stiffness. The arches have hinges
at the top (point T) that permit free rotation about the y-axis. Each arch has a radius of curvature of
18,07 m and a massive rectangular cross section with a width of 710 mm and a height of 800 mm.
The main focus of the analyses is the timber arches. A previous investigation with a 2D frame
analysis and design program demonstrated that the critical section of the arch is at the fastening of
the outer hanger, for instance point A in figure 4. This is reflected in the modelling of the deck and
the positioning of the live load. The two deck-beams representing the traffic lanes have been moved
as close to the relevant arch as the guard rail permits, and the traffic loading has been chosen so as to
produce the maximum bending moment M y (about the y- axis) at section A. It consists of 4
concentrated loads, each of which represents an imaginary vehicle axle and has a magnitude of 273
kN (including load factor). In addition a uniform traffic load of 11,7 kN/m is placed in each traffic
lane, and 7,8 kN/m in the pedestrian lane, in the first two spans of the bridge deck.
An important detail of the model is the arch support. At all four points the three displacement
components are suppressed, whereas the rotation about a 750 mm long bolt in the y-direction is
unrestrained. The other two rotations, about the x- and z-axis, are resisted by rotational springs with
an estimated stiffness of 300 kNm/deg. The two lowest buckling modes, and the corresponding
buckling load factors, obtained from a linearized buckling analysis with these boundary conditions
are shown in figure 5. With no rotational springs at the supports (i.e. free rotation about all axes), the
buckling load factors are 5,35 and 8,10, respectively (and the second mode is now similar to the first

mode # 1 buckling load factor = 10,75 mode # 2 buckling load factor = 13,98

Figure 5 Linearized buckling - dead + traffic load - E = 12000 Mpa and kθ = 300 kNm/deg
mode of figure 5, but with buckling of the left-hand arch). If we go to the other extreme and
suppress the rotations about the x- and z-axis, the buckling analysis gives 14,03 and 18,50 for the two
lowest modes, and the first mode now looks like the mode to the right in figure 5, and the second like
the one to the left.
Each arch is modelled by 44 straight beam elements, and the total model has about 2200 degrees of
freedom.

4.1 Linear static analysis and linearized buckling analysis


Linear analyses are carried out for different stiffness properties (E and G) of the glulam arches, and
some characteristic results are shown in table 5. It should be pointed out that the only parameters

Table 5: Linear static analysis of the arch bridge in figure 4

E [Mpa] (G = E/17,5) 6000 8000 10000 12000 15000

M y at section A [kNm] 1237,6 1308,4 1360,0 1392,7 1429,2

1st buckling load factor 6,88 8,25 9,53 10,75 12,50


2nd buckling load factor 7,75 9,85 11,93 13,98 17,02

that are varied are E and G of the arches (the deck beam parameters are constant throughout). The
axial compressive force at section A is not included in the table since it varies very little, from
1376,0 to 1378,6 kN. The variation in the bending moment, however, is perhaps more noticeable
than expected, whereas the influence of the stiffness on the buckling factors is probably as expected.

4.2 Design check based on both linear and nonlinear static analyses
The arches are made of GL36c glulam (the same as the columns and the beam). Short time loading
and service class 3 (high moisture content) are assumed. The capacity (κ) of section A of the arch
(see figure 4) in combined axial compression and bending is determined on the basis of EC5
specifications for both linear and nonlinear static analyses, for different stiffness parameters. For the
linear analyses, buckling lengths of 13700 mm (for in-plane buckling) and 11000 mm (for out-of-
plane buckling) are assumed (based on simplified plane models).
Results are shown in table 6. For the nonlinear analyses, e is the magnitude of the largest
geometrical imperfection taken in the shape of the first or second buckling mode, whichever gives
the highest κ. Note that κ is given in per cent of permissible capacity, that is, κ = 100 indicates a
fully utilized section. It should also be noted that the deck stiffness is the same in all cases, and the
results are therefore more applicable to a non-timber deck structure.
The shaded cells of tables 5 and 6 apply to the stiffness values closest to those specified by EC5. We
notice that in this particular case, EC5’s “linear approach” seems to underestimate the higher order
effects. We also notice that, in this case, the increase of arch displacements, due to a lower stiffness,
does not produce higher order moments that counteract the effect of a reduced load sharing of the
Table 6: Capacity κ (in % ) of arch section A in combined axial compression and bending

E [Mpa] (G = E/17,5) 6000 8000 10000 12000 15000

κ - linear analysis 98,8 103,3 106,3 108,5 110,8


κ - nonlinear analysis - e = 0 100,4 104,2 106,7 108,5 110,3
κ - nonlinear analysis - e = 25 mm 101,8 105,6 108,1 109,8 111,7
κ - nonlinear analysis - e = 50 mm 103,3 107,0 109,4 111,2 113,0

arch compared to the deck (which maintains an unchanged stiffness). In other words, we see the
opposite effects here, compared to the earlier column and beam cases.

5. Concluding remarks
The results presented show that stiffness is an important parameter for the response analysis of a
timber structure, and hence its design. However, the results do not all point in the same direction. In
the case of simple columns and beams, the results seem to indicate that the standard “linear
approach” of EC5 is conservative compared with a “nonlinear approach”. For the glulam arch
bridge we see the opposite, although not to the same extent.
The results also show that the magnitude of the stiffness is important, but more results are obviously
necessary for conclusive statements on the issue. However, having seen the effects of E (and G) on
the column and beam (tables 1, 2, 3 and 4), it is hard to understand why EC5 does not distinguish
between short term loading in service class 1 conditions on the one hand, and long term loading in
service class 3 on the other, when it comes to determining the higher order effects. The code
specifies the same stiffness ( E d = E mean ⁄ γ M ) for both cases. Would it not be more appropriate to use
E mean,fin of equation (3) also in the case of an ultimate limit state design based on nonlinear static
analysis? This question leads to another: are the “instability factors” k c,y , k c,z and k crit really
independent of load duration and moisture content? It appears that there is some lack of logic here,
and these questions ought to be looked into more closely.
This can also be said for the geometrical imperfection. In order for nonlinear analyses to take their
rightful place in the design process, it is important to agree upon both shapes and magnitudes of
geometric imperfections. Easy implementation into the computational tools suggests that buckling
modes are good candidates for the shape, but in certain cases these shapes may (locally) need
additional imperfection. Again, more results are required before definite answers can be given.

6. References
[1] EN 1995-1-1 (2004) Eurocode 5: Design of timber structures - Part 1-1: General - Common
rules and rules for buildings. 123 pp.
[2] Blass, H.J. (1995) Columns. Lecture B6 of Timber Engineering STEP 1, Centrum Hout, The
Netherlands. 8 pp.
[3] Wollebæk, L. and Bell, K.: (2005) FrameIT - A Windows-based program for analysis of 3D
frame type structures - User’s Manual, Department of Structural Engineering, Norwegian
University of Science and Technology, Trondheim. 48 pp.
[4] Wollebæk, L. (2005) Analysis of geometrical nonlinearities with application to timber
structures, Doctoral Theses 2005:74, Norwegian University of Science and Technology,
Trondheim. 212 pp.

You might also like