Prager 1974
Prager 1974
Prager 1974
William PRAGER
Savogprin, Swirzerland
While achieving the greatest possible economy of material, Michell structures with their infinity of bars are not
practical. This note indicates a way of designing trusses with finite numbers of bars that achieve nearly the same
e~nomy of material as the corresponding Michell structures.
I. Introduction
Culmann [ 1 ] appears to have been the first to attack the problem of the optimal layout of a
truss that is to transmit specified loads to given points of support. Stipulating that buckling could
be disregarded, he defined the characteristic number (“Werth”) of a truss as the sum ci lSj / Zj,
where Si and Jj respectively are the axial force in bar i and the length of this bar. When all bars
are designed for the same allowable stress, this number is proportional to the volume of material
used for the bars of the truss. Culmann therefore proposed to position the joints of the truss so
as to minimize its characteristic number.
A simple problem of this kind is shown in fig. 1. A load P with given line of action is to be
transmitted to the given joints A and B by two bars AC and BC. Where should the joint C be posi-
tioned on the line of action of the foad? If D is the point of intersection of this line with the line
AB, Culmann showed that the length of the segment CD sjould be the geometric mean of the
lengths of the segments AD and DB. Using this and similar basic results, and employing grap~cal
techniques, Culmann determined minimal characteristic numbers of widely used types of trusses
and the ranges of optimality of these types.
Whereas Culmann treated realistic trusses with finite numbers of bars, Michell 125 discussed a
trusslike continuum with infinitely many bars. While achieving the greatest possible economy of
material, Micheil structures are not practical. The present paper is concerned with the layout of
trusses with finite numbers of bars that achieve nearly the same economy of materiai as Michell
structures.
Michell minimized the total weight of the bars of his structural continuum subject to the con-
straint that a given allowable stress should nowhere be exceeded. Since Michell structures are
* Thisnote is cordially dedicated to Professor 1u.N. Rabotnov on the occasion of the sixtieth anniversary of his birth.
A Russian translation of this Note appeared in the anniversary volume “Mechanicsof Deformable Solids and Structures”
dedicated to Professor Rabotnov.
350 W. Prager, A note on discretized Michell structures
A_-___--- D s
-r-----
/’
/
l’
c;
Y P
statically determinate, this is equivalent to minimizing the weight of the bars subject to the con-
straint that the given loads should not exceed the 1oad”carrying capacity of the structure, whose
bars are assumed to be rigid, perfectly plastic and to have identical yield limits in tension and
compression. It is this formulation that is adopted in the following, except that the total weight
of the bars and their connections is to be minimized instead of only the total weight of the bars.
It is worth noting, however, that Hegemier and Prager 13 J have shown that Michell structures are
also optimal under certain other constraints, for instance the constraint of prescribed elastic com-
pliance.
Since the force transmitted by bar i of a statically determinate truss is proportional to the cross-
sectional area Ai, it appears reasonable to assume that the contribution of this bar to the weight
of connections is also proportional to A i. The total weight of bars and connections will therefore
be proportional to an expression of the form
W = CAi(li +I,) )
i
where cl0 is the yield limit and P, the load vector at joint Q[,and the center dot indicates the scalar
product.
Consider next an alternative design of the basic truss whose load-carrying capacity is not exceed-
ed by the given loads. If A; is the cross-sectional area of bar i in this design, it fol.lows from the
kinematic theorem of limit analysis that
W.Bager, A note on discretized Michell structures 351
(3)
C(Af-Aj)ljlq 2 0. (4)
i
Finally, if the design Ai is to be optimal, the value of the expression (1) cannot decrease when
Ai is replaced by Ai*. Thus,
C(Ai’-A,)((.+l,)> 0. (6)
i
According to a theorem of Farkas [4f, the inequality (6) for the differences Af - A i follows from
the inequalities (4) and (5) for these differences only if it is a nonnegative linear combination of
the latter. Let l/e0 and pi (with pi = 0 when Ai > 0) be the coefficients of this linear combination.
Farkas’ theorem then furnishes l,le,l + pie, = (fi+Ec,)eO, or
as a necessary and sufficient condition for the optimality of the design with cross sections Aj.
Since the differences Af - Aj in the preceding proof need not be small, the condition (7) assures
global optimality.
3. First example
In this section the application of the optimality condition (7) will be illustrated by the simple
example shown in fig. 2. The horizontal load P with the point of application 0 is to be transmitted
by a two-bar truss to the horizontal surface S-S of a rigid foundation. Choosing a sufficiently
dense sequence of joints on S-S, we form the basic truss by connecting each of these joints by a
bar to the joint 0. Since the optimal truss will be- symmetric with respect to the vertical through 0,
this joint will have a horizontal rightward velocity v in the collapse mechanism of the optimal truss
under the load P. As the time rate of collapse is immaterial, v may be assigned a numerical value
equal to that of the distance h of the point 0 from the surface S-S. If bar i of the truss forms the*
angle 8, with the vertical (fig. 2), its rate of extension Ej in this collapse mechanism has the abso-
lute value
It follows that the bars of the optimal truss form angles Bi = f 8 with the vertical that maximize
the left side of (9). This furnishes
from which the angle 8 between the bars of the optimal truss and the vertical can be determined
by iteration when 1,/h is given. Table 1, which has the value of 1,/h as entry, shows the correspond-
ing values of 0 and W/(Ph) and also the values of W*/(Ph) when 8 = 45” is used instead of the
optimizing value of 8. It is seen that even for the value to/h = 0.5, for which bars forming an angle
of 4.5” with the vertical contribute an amount equal to 35% of their weight to the weight of con-
nections, W * exceeds the optimum W only by about 1%.
Table 1
4. Second example
This example differs from the one in the preceding section by the fact that the foundation has
a limited width 2 b. If the optimal two-bar truss is then ruled out by the given value of h, optimal
trusses with finite numbers of bars have layouts of the kind shown in figs. 3 and 4.
To prove that a specific layout L, say the one in fig. 3, is optimal for a given value of 1, /h,
W.Pmger, A note on ~iscretiz~ M~c~eilstru~?~res 353
_o
” ;;
.c
L
B A
BI /A
i------ 2b - E-------2b d
Fig. 3. Nearly optimal truss with 11 joints. Fig. 4. Nearly optimal truss with 18 joints.
one might proceed as follows. First, choose a value of e0 and assign to the bar i of this truss the
rate of elongation
Starting from the rigid foundation and working upwards, determine one by one the velocity
vectors of the joints. In each triangular or q~drilateral mesh formed by bars, determine respec-
tively a linear or bilinear velocity field based on the velocity vectors at the vertices of the mesh,
and continue the resulting velocity field suitably beyond the contour AOB of the truss.
Consider now an alternative layout L’ of the same type obtained from L by displacing its
joints while maintaining symmetry and avoiding any crossing of bars. For each bar of L*, deter-
mine the rate of elongation e; as the sum of the rates of elongation of its elements in the con-
sidered velocity field. The layout L will be optimal of these rates of elongation satisfy
(13)
for each bar of L” and for all possible positions of its joints. It will obviously be most difficult to
verify this. In the case investigated by Michell, I, = 0, and after division by Zj the condition (13)
shows that the bars of the optimal structure must follow the principal lines of extension of a
strain rate field with the principal rates of extension f eO.
354 W. Prager, A note on discretized Michell structures
The proximity of the values of W and W * in table 1 indicates a low response of the structural
weight to deviations from the geometry of the optimal structure. This suggests that it may be
worthwhile to bypass the difficulty mentioned at the end of the last section by discussing nearly
optimal layouts obtained by appropriate discretization of properties of the Michell layout. This
section is concerned with a discretization of this kind.
Consider the bars along, say, the left half OB of the boundary of the truss in fig. 4. The forces
in the corresponding bars of the Michell structure have the same intensity, while the forces in the
other bars at boundary joints vanish. For the truss in fig. 4, we may stipulate the first property
but not the second. Instead, we shall require the forces in the bars LH. KG, and IF to have the
same intensity, though not zero intensity. These stipulations have the following consequences:
(1) the angle (Yformed by the two boundary bars at a boundary joint is bisected by the third
bar at this joint, and
(2) the angle (Yhas the same value at each boundary joint.
We shall take these properties as the appropriate modification of the orthogonality between the
boundary bars of the Michell structure and the other bars at boundary joints, and shall stipulate
this kind of modified orthogonality throughout our truss. For example, any two consecutive bars
of the chain BFGHL’ form the same angle (Yand these angles are bisected by the bars FD, GE,
and HG’.
When the symmetry of the truss is taken into account, it is seen that in each quadrilateral mesh
formed by bars opposite sides form the angle 71- Q. This corresponds to a well-known property
of the Hencky-Prandtl nets in the theory of plane plastic flow (see, for instance, [5]), which are
identical with the nets of principal lines in strain rate fields with the principal rates of extension
teO [3].
In figs. 3 and 4, the numbers near the bars of the trusses indicate the angles these bars form
with the horizontal. They will enable the reader to verify that these trusses possess the geometri-
cal properties stipulated above. It should be noted that the length 2b of the base and the value of
the angle (Y.which is 15 6” in fig. 3 and 164” in fig. 4, completely specify the layout of this kind
of truss.
It remains to be shown that the trusses in figs. 3 and 4 are nearly optimal. As a preparation for
this, we point out a remarkable property of these trusses. When the scale of forces is chosen so
that the load P is represented by the vector AB, the truss and its Maxwell-Cremona diagram
coincide as is shown by the Bowes notation in figs. 3 and 4. This fact greatly facilitates the evalua-
tion of the characteristic numbers of these trusses. For example, because BC = BD = BF and
CD = DF, the characteristic number of the truss in fig. 3 is obtained by multiplying the value of
6. Conclusions
In the computation of the efficiencies at the end of the preceding section, no allowance was
made for the weight of connections, but then, this weight was primarily introduced as a device to
enforce a finite number of bars.
By increasing the number of joints from 11 in fig. 3 to 18 in fig. 4, an efficiency increase of
only 1.6% is achieved. If, for simplicity, all connections are assumed to have the same weight, the
increased number of joints would be worthwhile only if the weight of the additional seven joints
did not cancel the small gain in the weight of the bars.
While it cannot be asserted that the discretization of Michell structures proposed in this note
furnishes optimal trusses even when the weight of connections is neglected, the high efficiencies
of the example trusses suggest that this kind of discretization does not entail significant losses
in efficiency.
References