Beruflich Dokumente
Kultur Dokumente
00
Printed in Great Britain. Pqamon Rem plc
Department of Civil Engineering, Technical University of Poznali ul. Piotrowo $60-965 Poznari, Poland
Abstract-A new concept of explanation of the shear locking phenomenon occurring in the Timoshenko
beam is presented. Instead of element matrix analysis the global expression of the equilibrium equation
for a whole beam is considered. Based on the element stiffness matrix the equilibrium conditions for a
regular discretized beam are derived in the form of one difference equation that converges to the analytical
differential formulation for the continuous beam. It is clearly shown that application of the linear shape
functions leads to an overly stiff element in the Mindlin elements because this approximation converges
to the solution of a different equilibrium equation. From the stiffness matrices of exactly and reduced
integrated linear and quadratic elements the difference equilibrium conditions are derived and they are
compared with the exact equations of the bending beam problem. For each type of element the improved
one is elaborated that assures the exact solution for beams irrespective of beam thickness and discretization
(number of elements). The numerical analyses for various boundary conditions, beam load, discretization
and thickness ratios are given.
769
770 J. RAKOWSKI
The solution of eqn (3), using (1) and (2), for unit
boundary conditions (t$ = 1, wi = 1, i = 1,2) leads
to the exact shape functions for the two-node
Timoshenko beam element, which are presented in
Table 1 [{ =x/a, D = l/(d + l)].
Fig. 1. Beam element description. By neglecting the shear deflection (d = 0) one gets
the standard Hermitan polynomials for the classical
associated with application of these low-order el- Euler-Bernoulli beam. Having found the shape func-
ements. Also presented are the results of calculations tions, the element stiffness matrix is computed by the
for beam with various thickness ratios, boundary application of the virtual work principle. The matrix
conditions, discretization and loads. corresponding to the vector of nodal displacements
and rotations {w,, 01, w2, t9,} can be expressed as
fo11ows:
2. TIMOSHENKO BEAM-EXACT SOLUTION
6a [3 - (d + l)*]a* -6a
1
Y(X)= qg) - e(x),
12 6a -12 6a
where w(x) is the transverse displacement, e(x) the 6a 3a2 -6a 3a2
rotation, and y(x) the shear strain of the cross- K,=
-12 -6a 12 -6a ’
section; M(x) is the bending moment, Q(x) the shear
force; E and G are the Young’s and shear modulus, 6a 3a2 -6a 3a2
respectively; I and A are the moment of inertia and I
cross-section area, respectively; K is the shear correc- Evaluation of (4) leads to the following form of the
tion factor. element stiffness matrix:
The interpolations of displacements and rotations
are assumed as follows:
KC- ~&P,+d*K,l, (5)
w(x) = t: c,xi
i-0 where now Kl and K, include only numbers
where
EI
KB=>‘d+,-
1 K
1
lime
a4
O-+O
=$,lim
a-0
[
a2.$
1 =O
&=$&K2. (7)
lim
n-0 [ 2 1
!!.A2-6Ef d2
-KG/~ ‘da2
1
expression for strain energy [4,8]
(A’ + 6) - 4 A2 Cp,- 3(E - E-‘)w, = 3B(d + l)m,
(E
-E-')$,
- 2 * A2w,= B(u’ + l)P,, (9)
712 J. RAKOWSKI
1.0
3.1. Linear element
First, we assume the element shape functions w(x) 0.6
and 6(x) to be independent and linear
0.6
w(x)=w1+(w2--,).x/a, 0.4
0.0
Substituting (13) into (12) after exact integration we
get the stiffness matrix in the form
Fig. 3. Convergence of normalized node displacements for
linear element beam vs thickness ratio (L/h) and number of
(14) elements (R).
- 38 . (E - E-‘)m,
1 . (15) with Ki and K2 identical to the number matrices
in (5).
Comparing the above equation with the exact The difference equilibrium equation for a beam
one, eqn (lo), we see that the application of consisting of an arbitrary number of identical el-
linear elements leads to fully different equation de- ements with stiffness matrices given in (16) can be
scribing the bending problem of the Timoshenko expressed as follows:
beam. For d-r0 (h/a +O) the quotient d/(d + 1)
converges rapidly to zero. It explains the essence of
the locking phenomenon. Fortunately, eqn (15)
differs from the exact eqn (10) on its right hand
side only on the factor d/(d + l), so it is not necessary -3B*(E-E-‘)m,. (17)
to perform numerical computations to estimate the
error caused by shear locking. Moreover, one can The form of the equilibrium condition (17) is nearly
see here that the error is independent of the load to the exact formulation of (10). One can see that
acting on beam and boundary conditions and the the ‘underintegrated’ linear elements do not lock
results of computations converge to the exact sol- but the solution for a finite number of elements
ution as well as the quotient d/(d + 1) converges to will never reach the exact result for any value of
1 (see Fig. 3). d, however in the element length limit (a-0,
It is possible to improve this linear beam element i.e. R + co) eqn (17) converges to the exact differential
in a simple way: one must only introduce to the equation for the continuous beam, eqn (11). It can be
exactly integrated model modified physical par- seen more precisely by looking at the results that were
ameters, namely, instead of E and G, E * d/(d + 1) performed for beams with various discretization,
and G * d/(d + l), respectively. boundary conditions and load distribution. The
: 0.8571
0.7895 09643
0.9474 0.9911
0.9868 0.9978
0.9967
4 0.7612 0.9403 0.9851 0.9963
8 0.1529 0.9382 0.9846 0.9961
16 0.7507 0.9377 0.9844 0.9961
32 0.7502 0.9375 0.9844 0.9961
64 0.7500 0.9375 0.9844 0.9961
etc. 0.7500 0.9375 0.9844 0.9961
Table 6. Normalized node displacements for beams due to various beam loads
Boundary conditions
Simply supported Cantilever beam
Number jmiddle node) (end node)
of elements L/h II I II
1 0.9375 0.9375 0.9643 0.9615
4 0.7895 0.7895 0.9403 0.9322
R=2 8 0.7612 0.7612 0.9382 0.9295
32 0.7507 0.7507 0.9375 0.9286
128 0.7500 0.7500 0.9375 0.9286
central node ~spla~rn~~ are determined from the The stifl’ness matrix co~s~nding to the nodal
analytical solution of eqns (10) and (17) by means of element displacements and loads
the eigenfunction method described in 1131and they
are presented in Tables 2-5. The results refer to the 9T=(W*,~I,W2,~2rW~,(P~}
end load for the cantilever beam and to the central
node for the other cases of boundary conditions. The p={p,,m,,P,,m,,P,,m,},
reduced integration for beams of small numbers
of elements is not effective-two-element clamped- with c#+= ~0, and mi= M,/u, (i = 1,2,3) can be
clamped beam locks (see Table 2). written now in the following form:
The convergence to the exact solution is much
better for cantilever beams (Table 3).
The results of computations for simply supported
and simply suppo~ed~lam~ beams are presented
in Tables 4 and 5. Assuming that in each element P2 = 0 and m2 = 0 we
In Table 6 one can find the displacement results can express the middle node displacements as
due to various beam loads:
(I) one node of the beam is loaded by a unit force
wz = (WI + wj )/2 + (9, - 43 )I8
(the middle node for the simply supported, the
end node for the cantilever beam) 1
(II) all nodes of the beam are loaded by uniform - ~ * [3O(w, - w,) + (10d - 3)(43 + #,)I.
(62- 4(6 + 5d)
forces.
(21)
The underintegrated linear element can be improved
by substituting into eqn (16) (d + I) instead of d. This
Using the difference technique presented in Section 2
modification is implied by the eqn (l?), where intro-
we get from the relations (19x21) the equilib~um
ducing expression (d + 1) (instead of d) transforms it conditions for regular beam
to the exact eqn (10).
0 00 00 0 -3B*(E-E-‘)m,. (23)
c;(;,;;,;_,
,!“:‘,,,,
KY and KY. Indeed, if we replace d in eqn (20) by
d,,, = 0.1 . [Sd - 1 + J(25d* + 10d + 21)], where d, is
the root of the equation
1
d,,,-- = d,
10 10' 10' 10' 10'
1 1 + 5d,,,
Fig. 4. Comparison of central node displacement for beams
with reduced integrated linear (L) and exactly integrated
quadratic elements (Q). R is the number of beam elements. we get the exact solution of the Timoshenko beam
bending problem.
the beam thickness, as well as in the case of linear Now we estimate the accuracy of the reduced
elements, the displacements attain a certain limit of integrated quadratic element. Applying two-point
accuracy different of course for each value of R. quadrature to the shear term of strain energy (12) we
Table 9. Comparison of the normalized displacements for the middle and the end
nodes of a cantilever beam
Number of elements
R=2 R=4 R=8
Middle End Middle End Middle End
Llh node node node node node node
1 0.9993 0.9994 1.oooo 1.0000 1.0000 1.0000
2 0.9952 0.9967 0.9997 0.9998 1.0000 1.0000
4 0.9804 0.9874 0.9985 0.999 1 0.9999 0.9999
8 0.9493 0.968 1 0.9948 0.9967 0.9996 0.9998
16 0.9194 0.9495 0.9872 0.9920 0.9987 0.9992
32 0.9056 0.9410 0.9798 0.9874 0.9968 0.9980
64 0.9015 0.9384 0.9764 0.9852 0.9949 0.9968
128 0.9004 0.9377 0.9754 0.9846 0.9941 0.9963
256 0.9001 0.9376 0.975 1 0.9844 0.9938 0.9962
512 0.9000 0.9375 0.9750 0.9844 0.9938 0.9961
776 J. RAKO~SKI
cause an alteration of nine numbers in the KF matrix shape functions used in the variational formulation
(19), which has now the following form: - of the bending problem leads to the differential
equilibrium equation which does not cover the
KT=
exact one. The presented method allows determi-
nation of exact errors that also occur in reduced
integrated elements. Owing to this method it was
28 6 -32 8 4 -2
possible to elaborate corrections to each type of
6 413 -8 413 2 -213 element in order to get the exact solutions. The
important topic of the numerical examples was that
-32 -8 64 0 -32 8
all of them were carried out analytically, so there was
8 413 0 16/3 -8 413 no loss of precision and any disorders did not appear
during computations, e.g. for large values of the
4 2 -32 -8 28 -6
thickness ratio L/h.
-2 -213 8 413 -6 413 The author is of the opinion that this technique can
be extended on the analysis of the C” and C’ plate
(25) and shell elements.
Let us express the middle node displacements by
REFERENCES
external. From (19) where KY is given now by (25) we
have 1. A. Kamoulakos, Understanding and improving the
reduced integration of Mindlin shell elements. Znt. J.
Nruner. Me&. Engng 26, 2009-2029 (1988).
~2 = (w, + ~3 112 + (4, - #3 I/‘8
2. I. Kreja and Z. Cywinski, Is reduced integration just a
nurne~~l trick? Commrt. Struet. 2X491-496 (1988).
#2= -. l
4(1 +d)
Ww3- WI> + W - lN#s+ #i 11. 3. D. Briassoulis, The co shell, plate and beam elements
freed from their deficiencies. Comp. Meth. appl. Meek
(26) Engng 72, 243-266 (1989).
4. T. J. R. Hughes, R. L. Taylor and W. Kanoknukulchai,
A simple and efficient finite element for plate bending.
Substituting (26) into the element stiffness matrix (20) Znt. J.-Numer. Meth. Engng 11, 1529-1543 (1977).
in accordance with the difference equation formu- 5. 0. C. Zienkiewicz. The Finite Element Method. 3rd Bdn.
lation for a regular beam we get finally McGraw-Hill, London (1977).
6. T. Belytschko, C. S. Tsay and W. K. Liu, A stabilization
matrix for the bilinear Mindlin plate element. Comput.
A4w,=B+ P,-3B~(E--E-‘)m, Struct. 29, 313-327 (1981).
I. E. Hinton and D. R. J. Gwen, An Zntroduction to Finite
(27) Element Computations. Pineridge Press, Swansea, U.K.
(1979).
8. G, Prathap and G. R. Bhashyam, Reduced integration
in which we recognize the exact ~uilib~um equation
and the shear-flexible beam element. Znt. J. Numer.
of bending problem for the Timoshenko beam. So the Meth. Engng 18, 195-210 (1982).
reduced integrated quadratic element gives the exact 9. H. Stolarski. N. Carnenter and T. Belytschko. A
results in numerical calculations. Kirchhoff-mode method for Co bilinear and-serendipity
plate elements. Comput. Meth. appl. Meth. Engng SO,
121-145 (1985).
4. CONCLUDING REMARKS 10. R. H. MacNeal, A simple quadrilateral shell element.
Comput. Struct. 8, 175-183 (1978).
The problem of shear locking has been examined 11. N. Carpenter, T. Belytschko and H. Stolarski, Locking
in a different way. A more general method to analyze and shear scaling factors in Co bending elements. Com-
the convergence of finite element scheme using the put. Struct. 22, 39-52 (1986)
12. J. M. Ball, Existence of solutions in finite elasticity.
difference equation formulation of FEM was pre- Proc. of the ZUTAM Symposium on Finite Elasticity,
sented. It has been shown that knowing the element Martinus Nijhoff, The Hague, The Netherlands (1981).
stiffness matrix it is possible to create an equivalent 13. J. Rakowski and R. Switka, On Some generalization of
difference equation for a whole regular system. In- the E~genf~ction Method Used to the EZastic Discrete
stead of the usually applied element matrix consider- Systems (in Polish), Rozprawy nr 116, Polit. Pozn.,
Poznali ( 1980).
ation the study of one difference equation gives a 14. A. K. Chugh, Stiffness matrix for a beam element
clear explanation of the shear locking essence. It was including transverse shear and axial force effects. Znt. J.
proved that locking appears because the linear Numer. Meth. Engng 11, 1681-1697 (1977).