Magazine of Concrete Research

doi: 10.1680/macr.2007.00005

Automated lower bound analysis of concrete
C. J. Burgoyne* and A. L. Smith†

University of Cambridge; Steel Construction Institute

A method is developed that allows automated lower bound analyses of concrete slabs to be carried out, with the
specific aim of producing a technique that can be applied to existing structures which are being assessed after many
years in use and for which the original calculations, and hence the assumed methods of load distribution, have been
lost. It is shown that any load distribution within the slab can be expressed in terms of an equilibrium state and a
set of states of self-stress. Optimisation techniques are developed that allow the highest lower bound to be found. It
is shown that an analysis based on the original Hillerborg two-strip model fails to find a reasonable lower bound
when the slab is subjected to point loads and in the corners of slabs but a refinement using four sets of strips at 458
to each other can give very good agreement with both exact analytical solutions and upper bound methods.

Notation RL , RR reactions at left- and right-hand ends
of strip
A equilibrium equations matrix S matrix of states of self-stress
a, b, u, v directions of four-strip systems (Fig. wi distributed load acting on length Li of
19) strip
a–j letters defining individual strips x, y directions
b column vector of load terms from Æ proportion of q carried in x-direction
equilibrium equations strips
e column vector of equilibrium state º vector of magnitudes of states of
L length of side of slab self-stress
Li length of distributed load on a strip
(Fig. 3)
Ma , Mb , Mu , Mv moments in strips defined by a, b, Introduction
u, v
ML, MR moments at left- and right-hand ends The problems caused by using different methods for
of strip the design of concrete slabs, and their subsequent re-
mp flexural moment capacity analysis for assessment, have been highlighted in an
Mx , M y , Mxy applied moments (defined in Fig. 1) earlier paper.1 If the original calculations have been
p, pi self-equilibrating inter-strip force lost, it is common for structures to be condemned
p* column vector of self-equilibrating because the checker does not know the original de-
forces signer’s logic.
q distributed load In many cases the design would have been carried
Q, Q max total distributed load out by hand calculation, often using the Hillerborg strip
qi distributed load (dimensionless) method2,3 or code rules that were derived from it. That
method assumes a distribution of load between two sets
of strips, usually orthogonal, for which the reinforce-
* Department of Engineering, University of Cambridge, Trumpington ment can then be designed independently. The engi-
Street, Cambridge, CB2 1PZ, UK neer’s choice of load distribution is arbitrary, and the
† Steel Construction Institute, Silwood Park, Ascot, Berkshire, SL5
resulting moments may differ from those in the real
structure, but the lower bound theorem of plasticity4
(MACR-D-07-00005) Paper received 20 December 2006; last revised means that the structure will be safe. As long as the
21 February 2008; accepted 13 March 2008 engineer has provided sufficient reinforcement for the
1 1751-763X (Online) 0024-9831 (Print) # 2008 Thomas Telford Ltd

@2 M x ¼ Æq Sign convention @x 2 Figure 1 shows the sign convention used for @2 M y moments. The worst that could happen is that the structure would crack prematurely in the under-strength direction. Even though neither method gives applied load q could be shared between the two strips the true collapse load. Many structures are being wrongly condemned. they can be used as the basis for an assessment of structural safety. if the two bounds differ by only a with Æq applied to one strip and (1Æ)q applied to the small amount the checker could have confidence in the other strip. that the analyst is tempted to believe that the results are precise. primarily aimed at the assessment of convention is not required. which allowed a that the highest lower bound can be compared with the simpler equilibrium system to be set up in which the lowest upper bound. The rent paper presents a method to address that problem. and to longitudinal mechanisms whose parameters are altered to obtain the strips (which run in the x-direction) and transverse lowest upper bound. Mxy moments is also more costly in reinforcement. by definition. As defined. positive mo- ments relate to hogging moments and are thus resisted The reinforcement in each strip can be designed on the by the top steel in a slab and Mx refers to moments basis of beam theory: if the strips are designed in resisted by steel running in the x-direction and not to accordance with normal code rules the strips will be 2 Magazine of Concrete Research . The Hillerborg strip method has been automated they will be taken here to be orthogonal but the exten- before by O’Dwyer and O’Brien. Mxy ence between the actual and assumed moments does My not matter. and give such pretty Mx pictures as output. despite being an upper bound. at least in their linear elastic incarnations. rigour of sign line method.Burgoyne and Smith moments that have been calculated. but it would be strong enough in the other Mx direction and the structure would be safe. Nevertheless fears are still expressed that The standard Hillerborg strip method assumes that the results are on the unsafe side. so Hillerborg chose to make Mxy zero. the slab is made up of two sets of intersecting strips. y The problem is that the checker does not know how the original designer chose to distribute the loads.  2 þ ¼q (1) What is needed is a method that can be used to @x 2 @x@ y @ y2 automate the lower bound method for use in analysis. and reference will be made bridge decks. Fig. equilibrium equations for the two strips could then be separated. He has conducted many tests to strips (which run in the y-direction). Mid- dleton5 has developed a technique to automate the yield moments about the x-axis. The cur. He was almost certainly aware that resisting judgement that is made of the slab’s capacity. which is consistent with that used for most ¼ (1  Æ)q (2) @ y2 reinforced concrete slab work. It has also been observed that membrane forces are frequently developed which cause Hillerborg strip method the final collapse load to be above the upper bound prediction. May and Mann (personal commu- My nication) have recently shown that a fully non-linear Mxy finite element analysis does allow the post-cracking and post-yield behaviour of the slab to be followed. but the use of such programs requires expert knowledge and is not undertaken in routine assessment. but the Hillerborg approach @2 M x @ 2 M xy @ 2 M y itself has to be modified for use as an analysis tool. He uses a library of possible failure to sagging and hogging moments. Modern x finite element packages are now so easy to use. The differs from the one presented here. In most cases.6 using a technique that sion to non-orthogonal strips is straightforward. easily be adapted to any other self-consistent sign con- tions are sufficiently close to the true collapse load that vention. Sign convention for moments There have been attempts to automate the use of the plasticity bound theorems for design and analysis. his predic. Both techniques are basic equilibrium equation for flexure of a plate is satisfactory as design tools. 1. The method can show that. and has complied with normal code rules to ensure ductility. the differ. or Mxy Mxy even what type of analysis was performed.

if the right-hand end is an existing structure can be analysed. dicted maximum loading for the structure. 3. ML and MR can @ y2 take any values and it will still be possible to determine As described above. The method This can be expressed in matrix form as Consider a rectangular slab divided into orthogonal regions. By considering every strip that makes up the com- plete slab and replacing w i with the loads given by equation (3) a set of simultaneous equations can be derived that defines an equilibrium system for the slab. whereas the inverse problem ment equilibrium so that all the end reactions and requires searching for optimal values of p. q i . A general strip (Fig. will have a fixed sing p. As originally proposed. column vector that contain the terms from the simul- tely defined by the values of pi assigned to each rectan. Qmax q i . since many strips will be unloaded. but conditions. The applied load and moment capacities are assumed to be constant within each region. The moments in the strips can be comple. depending on the type of support @x 2 some of these may be zero or there will be additional @2 M y equilibrium equations that have to be satisfied. 3) has up to four end @ Mx ¼ qþ p reactions or moments. A holds the values gular region. but a set of self-equilibrating forces p can be introduced Symmetrical edge between the two sets of strips whose total external effect is zero. Instead of choo. that relate to the entries in the p* vector and b holds The objective is to determine a distributed load. Automated lower bound analysis of concrete slabs Simply supported edge Clamped edge under-reinforced and will fail in flexure before they fail Free edge in shear. and the p* is a column vector of internal loads. in either hogging or sagging bending. This For a strip that is free at both ends the applied loads is much more complex. as shown in Fig. in order for over- cities m x and m y . The The shape of the matrix A depends on the degree of loads applied to the longitudinal strips are thus Qmax (q i statical indeterminacy. the method is suitable for design. In practice. relationship to the applied loads. Qmax the values that relate to the applied loads. values for a slab with a known moment capacity. free end will be set to zero. Allowance has to be made for the support conditions 2 for a strip. cannot be exceeded. which defines two sets of A p ¼ b (4) strips. meters in the normal yield line analysis. the problem can be each rectangular region simplified by setting Æ to unity (thus applying all the load to the strips in the x-direction) and using p to specify how the load is transferred within the slab. to give the pre. the design problem leads themselves will have to satisfy both vertical and mo- directly to a solution. A and b are the matrix and set of strips. free. conditions will always be satisfied. and the boundary conditions for the strips are constant for each non-zero edge moments. codes often specify how the load is to be distributed. 2. In both cases the edge moment for a are many more of these than there are pattern para. Fig. but these requirements are designed to prevent early cracking of the slab and are not required for safety. If the system is determinate the + pi ) and those on the transverse strips Qmax pi . w i . The lower bound theorem of plasticity applies and the structure will be safe whatever value of Æ has been chosen. and there moments are zero. RL and RR using the two equations of overall equili- The problem now is to reverse the usual process so that brium for the strip. the left-hand edge moment. q i . the method cannot deal with patch loads. The L1 L2 L3 L4 values of q i are known but the pi have to be determined ML w1 w2 w3 w4 RR MR RL subject to the condition that the moment capacity in each strip. ML . This would leave the designer with two Fig. ¼ p (3) For a strip that is fixed at both ends. and by using dimensionless load distribution. However. Equation (dimensions FL22 ) that can be multiplied by the (4) defines the equilibrium of the system. Moment capacities and loads are constant within without any loss of generality. Typical general slab showing notation for boundary parameters Æ and p that can be chosen at will. taneous equations described above. which describes values for p* that satisfy this equation the support how the load is applied to the slab. and then calculating the required moment capa. Notation for loading on strip Magazine of Concrete Research 3 . 2. a method is needed to find the p all equilibrium to be satisfied. using ‘column strips’ and Hole ‘middle strips’. pi .

There are ð q5 þ p5 Þ þ ð q6 þ p6 Þ þ ð q7 þ p7 Þ simple supports at edges a. any value of the applied loads and edge moments as ple is slightly artificial but it shows all aspects of the the reaction forces at both ends of both strips can be way the method has been implemented. and clamped 2 2 2 (7) supports at edges j and k. but details are not given here. the support reactions Transverse strips a–j and b–i will be in equilibrium for and moments in all strips can be calculated. The slab shown in Fig. To establish  p3 L þ  p7 L ¼ 0 (8) equilibrium for this strip the moment applied by the 2 2 four loads. pi . q i + pi . stage. about edge m must be zero. and m. 4 is divided into eight regions. adjusted. In addition to the eight 7L2 internal forces (p1 to p8 ) there will be two edge mo. significantly speeding up the process. þ ð q8 þ p8 Þ ¼ Mk 2 ments (M j and M k ). the matrices A and b have to be found. Once these are determined. each of which is a state of self-stress: º is a column vector a b c d of variables which add differing amounts of the states of self-stress to the equilibrium state. so the vector p* has ten compo- nents. cross-hatched edges represent This reduces the number of variables that have to be clamped supports. The exam. j i h g reflecting statical indeterminacy. Moment equilibrium of strip k–f gives a similar equa- The method can most easily be demonstrated by tion means of an example. each of which can be loaded L2 3L2 5L2 independently by choosing values of q i . Fig. 4. so these need not be taken into account at this To find p*. Thus the objec. holes and lines of symmetry.Burgoyne and Smith L L L L matrix will be square and the moment distribution is fixed. Hatched edges tive is to choose the values of º that maximise Qmax. Example used to demonstrate method. plain edges represent free edges optimised later. Consider first the longitudinal strip m–e. brium. b. giving  p3 L 3 p7 L þ ¼0 (9) 2 2 L2 3L2 5L2  p4 L þ  p8 L ¼ 0 (10) ð q1 þ p1 Þ þ ð q2 þ p2 Þ þ ð q3 þ p3 Þ 2 2 2 2 2  p4 L 3 p8 L (6) þ ¼0 (11) 7L 2 2 2 þ ð q4 þ p4 Þ ¼0 2 There are thus six equations defining ten variables 2 3 L2 3L2 5L2 7L2 6 6 2 0 0 0 0 0 0 7 2 7 3 2 3 6 2 2 2 7 p1 (q1 þ 3q2 þ 5q3 þ 7q4 )L2 6 7 6 p2 7 6 7 6 6 0 L2 3L2 5L2 7L2 7 6 7 7 6 6 2 7 7 1 7 6 7 6 (q5 þ 3q6 þ 5q7 þ 7q8 ÞL 7 6 0 0 0 0 p 2 7 6 3 7 6 2 2 2 2 6 7 6 7 6 7 6 p4 7 66 2 7 7 6 0 0 L 0 0 0 L 0 0 7 0 7 6 p5 7 6 7 6 6 736 6 p 7¼6 7 6 0 7 (12) 7 6 L2 3L2 7 6 6 7 6 7 6 0 0 0 0 0 0 0 0 7 6 p 7 6 7 6 2 2 7 6 7 7 6 0 7 6 7 6 p 7 6 7 6 7 6 8 7 6 7 6 0 0 0 L 0 0 0 L 0 0 7 4 M 5 4 0 5 6 7 j 6 7 M 4 L2 3L2 5 k 0 0 0 0 0 0 0 0 0 2 2 4 Magazine of Concrete Research . i. The relationship between the variables needs to be will need to satisfy both force and moment equili- determined so that the edge conditions are satisfied. For strips c–h and d–g the applied loads. represent simple support. and the solution will take the form L k q5 q6 q7 q8 f p ¼ e þ Sº (5) Here e is an equilibrium state of the system while S is L m q1 q2 q3 q4 e a matrix composed of a series of column vectors. Normally A will have more columns than rows. The method can be extended to deal with columns.

and if the equa- tions had been reordered differently. . as defined by the 6 0:333 0 0 0 :667 74 º3 7 6 5 final column of the S matrix. q1 ¼ q2 ¼ . and the remaining four variables are introduced d–g.M k ). p4 . which can be rearranged by Gaussian elimi- nation. An increase in M k 6 0 0 0 0 7 6 7 corresponds to a decrease in the load on strip b–i. and other equation. defined by º3 . ¼ q8 ¼ q. . four variables could back to the natural order. as would four equations. Automated lower bound analysis of concrete slabs All the loading terms are on the right-hand side of this equation. . For a uniform load over the entire slab. . other variables p ¼ e þ Sº (14) might have been involved in the equilibrium state. Each refers to a state of self-stress in one of the 2 3 longitudinal strips which allows the load to be 0 1 0 0 shifted between strip a–j and strip b–i. 6 7 6 1 0 0 0 76 º2 7 þ6 76 7 (e) The fourth state of self-stress. the result is have been eliminated. If this had been recognised before one state of self-stress. as a free-free beam with two regions cannot via the states of self-stress. without nullify- 6 0 0 7 6 0 0 7 º1 ing the equilibrium of the system. there are six equa. Strip a–j carries no load in the 6 7 6 7 6 7 6 7 initial equilibrium position as defined by the e 6 p7 7 6 0 7 6 7 6 7 vector. . satisfy both moment and force equilibrium unless tions in ten variables so the solution will involve four both regions carry no load. shows that any 6 7 6 7 6 0 0 0 0 7 º4 change to the edge moment M k is balanced by a 6 7 change in the internal load p6 . pivoting reorders the variables. Here. The 2 3 2 3 result would still have needed four states of self- p1 0 stress. equation (12) becomes 2 3 p8 6 7 2 36 p4 7 7 2 1:898q 3 1 0:068 0:593 0:356 0 0 0:119 0 0 0:237 66 p7 7 60 0:047 0:028 0:690 0:414 0:009 0:138 0:019 76 7 6 : 7 6 1 0 76 p6 7 6 2 058q 7 60 0 1 0:175 0:399 0:070 0:058 0:023 0 : 7 0 117 76 p3 7 6 6 1:306q 7 6 7¼6 7 (13) 60 0:005 0:397 0:333 0:132 0:667 76 7 6 : 7 6 0 0 1 0 76 p2 7 6 7 449q 7 40 0 0 0 1 0:202 0 0:067 0 0 56 7 4 1 075q 5 : 6 p5 7 0 0 0 0 0 1 0 0:333 0 0 6 7 5:333q 6 p1 7 4 Mj 5 Mk Equation (13) can be partitioned into a square matrix It is worth considering the physical meaning of this relating to the first six unknowns (p8 . 6 7 6 7 6 p5 7 6 0 7 and these are balanced by forces from the trans- 6 7 ¼6 7 6 p6 7 6 5:333q 7 verse strip b–i. p3 . 6 6 0 0:333 0 0 7 7 6 7 (d) The third column of S shows that the edge moment 6 7 M j is entirely independent of the applied loads and 6 0 0 0 0 7 6 72 3 can take any value. Magazine of Concrete Research 5 . . 6 p8 7 6 0 7 6 7 6 7 (c) º1 and º2 are multipliers for the first two states of 6 7 6 7 4 Mj 5 4 0 5 self-stress. and with L ¼ 1 m. p7 and p8 must unknown values º i .p2 ). 6 7 4 0 0 1 0 5 0 0 0 1 In more complex cases the physical interpretation of or the states of self-stress is less obvious. as defined by the first two columns of Mk 0 S. each of which is a multiplier for always be zero. 6 p2 7 6 7 6 6 : 7 7 6 7 6 5 333q 7 (b) The internal forces on regions 2 and 6 are used to 6 p 7 6 0 7 balance the forces applied on the remaining 6 3 7 6 7 6 7 6 7 regions for the longitudinal strips: in other words 6 p4 7 6 0 7 6 7 6 7 the entire load is applied to the longitudinal strips. After reordering the variables the equations were set up. The solution of the square matrix gives an equilibrium (a) There is no load on the transverse strips c–h and state. terms relating to the remaining unknowns (p5 .

If the function is smooth the running from NE–SW.Burgoyne and Smith 1 1 1 1 Optimisation The problem has now been reduced to its simplest form. It is thus very difficult to find simple expressions for bounds on the values ⫺3 ⫺3 of º i . because it can lead directly to the optimal so. By enforcing the two lines of symmetry the problem can be reduced to two variables. imagine a climber going up a mountain in fog defined. 5. and finishing points depending on how the problem was set up. the highest lower bound has to be Q found. Two degree-of-freedom problem value of Q will be a lower bound on the collapse load of the slab. the climber is on a ridge. To be useful. The maximum distributed load that the slab can carry for any value of p1 and p2 is plotted in Fig. If a point is reached where any step takes the directions to find the value of ºi that gives the largest climber downhill. Any combination of the various states of self. Many optimisation techniques have been devised and there is a widespread literature. all lie on one or other of the ridge lines. of the strips. By definition. The multiplier Q that can be applied to all the loads q i without exceeding the moment capacity for any com- bination of º i can thus be found. and simple to program. 6. that Fig. 1 p1 p2 stress. from different starting yielded strips (some contours have been omitted for clarity) 6 Magazine of Concrete Research . Strips yielding in hogging The surface consists of a series of smooth regions. Slab capacity against p1 and p2 times referred to as the Hill Climbing method. If. An alternative is Powell’s method. will satisfy equilibrium (and all possible equilibrium states can be expressed by equation (14)). as defined by the values of º i . The Simplex algorithm (or a more sophisticated ver- sion thereof) is frequently used in cases where the 0 limiting conditions can be expressed as simple func. 7. it may be possible to go higher. 5. the states of self-stress points. By starting from the equilibrium state (defined who is constrained to walk only in N–S or E–W direc- by the vector e) a search is made along one of the tions. The technique conducts a series of one-dimensional optimi- sations along various vectors. it will be concluded that the summit failure load. provided that the slab is sufficiently under- reinforced and thus ductile. 6. Press et al. The circles correspond to results where the numerical Strips yielding in sagging procedure stopped before reaching the highest point. 2 p2 0 lution. which shows a simply supported slab divided into two longitudinal strips and four transverse strips under a uniformly dis- tributed load. procedure will zigzag its way to a (possibly local) but the climber does not realise this. but it is not suitable here because yield could p1 0 occur at any point along a strip. separated by discontinuities in slope. The finishing Fig. As an are a convenient set of vectors that have already been analogy. Fig. maximum of the function and is reasonably efficient Figure 7 shows the same surface as a contour plot. The moments anywhere in the strips can be calculated in the normal way and 1 compared with the moment capacity in each strip. which allows the value of Q to be visualised for all combina- tions of pi . another search vector. Slope dis- simple slab problems. the solution reached different continuities occur where these surfaces intersect. An example is given in Fig. however. Each of the smooth regions corresponds to yield being When attempts were made to apply the technique to reached in some.7 give a good discussion of the techniques and pitfalls that can arise. but not all. The process is repeated by moving along has been reached. 2 tions. which is some. Contour plot of the two-dimensional surface showing points of the optimisation.

or the structure tain holes. (b) shows the load proportions determined by the analysis Magazine of Concrete Research 7 . if that were an issue. ‘turn a sow’s ear into a directions as the strips. It is the inability to do that with a finite always possible for all the strips to yield. but the This example shows that the technique is capable of optimum solution will still be the one where the largest recovering the assumptions that were made by the ori- possible number of strips are yielding. The optimisation is slab. and after a few iterations. been implemented. which repeated and it is likely that the new optimum will also leads to the required moment capacities in sagging be on the same ridge. column supports and lines of symmetry.5. When the optimisation routine type of slab to analyse. be capable of determining the tion of uniform moment capacity throughout the slab. design load has changed significantly. however. so it is less easy to visualise the process. Slab designed by strip theory a single global maximum. In the used as the starting point. the mountaineering analogy the climber would be allowed program determined that the load had been distributed to walk along the ridge. A modern assessor might be faced with this made to the procedure. if the original design was inadequate. or near. the ridge line power of the method. and can take account of free. In practice. If the iteration technique does not allow movement in these directions. so in many cases it is found L/4 L L/4 L/4 mp/8 0·5 0 0·5 y mp/2 x 1·0 0 1·0 L/2 L/4 mp/8 0·5 0 0·5 Proportion of load applied to x-direction strips mp/8 mp mp/8 (a) (b) Fig. The has deteriorated with time. even if it is not ginal designer. unnecessarily. as shown in Fig. If the original design was adequate before. Reynolds away is made by randomly choosing a small º vector to and Steedman8 give a strip design procedure for such a obtain a revised starting point. and it is still element program that leads to structures being repaired possible to become stuck on a ridge line. Automated lower bound analysis of concrete slabs some of these discontinuities form ridges. silk purse’. This is precisely the distribution In more realistic problems. the reinforce- or omitted reinforcement. so to reach the summit the climber has to proceed along the ridge lines. of aspect To counteract this effect a modification has been ratio 1. The highest arrangements. and slabs that con. subject to a uniformly distributed load. but it is a trivial extension to vary the moment capacity The method has been checked against other design in different regions to allow for material deterioration cases. It cannot. 8. ables. or the fixed and simply supported edges. The software is capable of adequate. since the checker is unable to find the The whole procedure has been automated in a com. analysing rectangular slabs. 8(a). however. stop- ping on a ridge can occur even if the function has only Example 1. Exten. The method results presented here are all carried out on the assump. would still. and more complex reinforcement assumptions of capacity. Note that this is not the normal prob. maximum available capacity. there will be more vari. bending as shown in Fig. Layout of slab used in example 1: (a) shows the input moment capacities. would be straightforward but have not point is reached when several of the ridge lines meet. the analysis will stop prematurely. ment detailer would usually have made conservative sions to other shapes. Figure 8 shows a simply supported slab. suggested by Reynolds and Steedman. 8(b). To demonstrate the mates of the optimum will be on. load distribution that the original designer found to be puter program in C++. these moment capacities were and can be added to the list of search vectors. but without prior knowledge of has reached its best estimate of the solution. The line joining the two esti. subdivided into rectangular this method should show that the structure remains regions with moment capacities specified in the same adequate. it may still fail. Examples lem associated with reaching a local maximum. a jump the technique used by the original designer. with similar results.

There are six different types of longitudinal No. The slab has been re- failure patterns for such structures involve partial width analysed using the methods described in this paper. structures are currently being reassessed owing to the requirements for heavier truck loads. Fig. albeit by a different route. yield line patterns with fans. 9. where the fan mechanism developed. The mo- be important for point loads applied to slabs where ments vary along the length but always remain within compressive arching action can be resisted by ring the hogging and sagging moment capacities. slab and obtained the lowest upper bound of 25. 11. Strip layout used to analyse Collins’ slab Fig. however. but one slab. Final crack pattern on underside of Collins’ slab (dimensions in mm). Yield line pattern analysed by Collins Example 2. Short single span with concentrated load The second example is of point loads applied to a tension in the reinforcement away from the loading simply supported. She attribu.6 kN The program converged very slowly to a solution. Many such positions. 12. which show that the method. Collins carried out yield line analyses of the material tests were used. 12. which means there are were applied symmetrically on 30 mm square pads as 60 different pi values to be found. the situation Load positions (30 mm ⫻ 30 mm) 55 70 65 80 80 80 80 80 80 80 Fig. two point loads of transverse (vertical) strips.6 kN. 3. which were much 0·571 kNm/m (hogging) stronger in sagging bending than in hogging bending. for the mechanism shown in Fig. 10. strips (horizontal in the figure) and ten different type The slab layout is shown in Fig. Fig. The maximum load carried by the slab was was increased in size by adding one half of the slab 31. one quarter of of such slabs were tested by Collins. 11. and indeed a result such as this could have been achieved. One quarter only of slab shown 8 Magazine of Concrete Research . The loading patch shown. and as predicted by the yield line method. however.3 kN (shared between the two loading points). had top and bottom steel in both directions. some good match to her observed failure mode. which is clearly a and the highest lower bound found was 17. has limitations. and thickness all round to allow for local load spreading. which are shown as the top and bottom of the boxes containing 1500 mm the bending moments. Fig. 33% lower than the lowest upper bound. The top bound capacity to strain-hardening in the steel and the half of the figure shows the predicted bending moments development of membrane actions which are known to in the six relatively weak longitudinal strips. The predicted Lower bound analysis. In the area in 0·720 kNm/m (hogging) 5·57 kNm/m (sagging) between. 0·490 kNm/m (sagging) The dominance of the sagging bending in the transverse 560 mm 90 mm direction under the load is clear. other problems that need to be tackled when assessing existing structures. 13 shows ted the fact that the slab carried more than its upper the moment distributions in the various strips. single-span bridge deck. using the techni- ques of O’Dwyer and O’Brien. 10. as is the absence of moment in the right hand half of the slab. It would be possible to stop at this point and be satisfied with a tool that can verify existing good de- sign. 9. using the strip layout shown in had limited amounts of reinforcement. The lower half shows the varia- tion in the ten transverse strips. Collins’ slab 3 50 50 50 50 45 35 Load Fig.6 There are.9 Two of the slabs the slab was analysed. Several physical models To minimise the number of variables. the final crack pattern on the underside is shown in The moment capacities determined by Collins from Fig. as explained so far.Burgoyne and Smith 128 120 that the structure can carry more load than the original designer intended.

Layout of strips used for lower bound analysis of upper and lower bounds have been established is the example 3. showing strip divisions clamped square slab (Fig. Only one quarter of the slab is shown for clarity L/6 L2 is far from simple. Two necessarily complex.85. which was analysed by Fox. 13.10 Both the distribution of moments for his lower bound and his upper bound collapse mechanism are was performed on a slab nominally 10 m square. T1 T2 T3 Example 3. is not good at determining when yielding takes place at L/6 L1 an angle to the line of action of the strip. Fully fixed square slab under uniformly distributed load. involving in the one case para. (b) strips between supports. The lower bound analysis predicts a maximum load Lower bound analysis results. Automated lower bound analysis of concrete slabs Free edge L Simple support Symmetry line Symmetry Loaded area line L CL CL (a) Hogging Hogging yield Sagging Fig. Moment distributions in strip analysis of Collins’ slab: (a) strips parallel to the support. showing eight-fold symmetry L/6 L/6 L/6 Sagging yield (b) L/6 L3 Fig. although it was expected that this would and hogging bending. 5 m 3 5 m slab divided into three longitudinal strips ners and in the other complex corner fans. giving nine values of pi to be found. It is clear that the method. The details (L1. which ments of º i that generate the same maximum. but there are several different arrange- the layout used for the lower bound analysis. Figure 15 shows of QL2 /m p ¼32. To see why that might occur. 14). but his result is important. L2 and L3) and 3 transverse strips (T1. Fully fixed square slab One of the few non-trivial cases for which identical Fig. T2 and are of no concern here. He T3). and no clear pattern can be dis- cerned. which could reduced the number of is the moment capacity per unit width in both sagging variables. The pro- showed that the true collapse load of the slab Q was gram does not have the ability to enforce the diag- given by QL2 /m p ¼ 42. 14. lines of symmetry are enforced. emerge naturally from the answer. a case is needed for which good upper and lower bounds are known. where L is the span and m p onal symmetry. as presently constituted. and the resulting bolic and hyperbolic moment distributions in the cor. 15. In each Magazine of Concrete Research 9 .

half the load goes in each direction giving w ¼ 32m p /L2 . although the bending moment distributions differ between strips as well as between the different º i distributions. This can be im- proved if load is shared between the strips in a way that Fig. an upper bound of QL2 /m p ¼48 which is much closer to In an attempt to overcome this problem. Doubling (or even quadrupling) the number of strips in each direction does not alter the solution that is found. It consists of diagonal sagging reason that example 2 does not work well for the point lines with corner fans. Simplified corner fan yield line pattern Consideration of the failure patterns shows why this might be the case. Incompatible ‘yield line pattern’ predicted by lower bound analysis Fig. will not have strips and since it is ˇ2 longer than the side. 17.Burgoyne and Smith case. The analysis has found the solution corresponding to a uniformly distributed load w applied to a fixed-ended beam with moment capacity m p . all the strips are at yield in hogging at the sup- ports and in sagging at midspan. 18. In the present case. Diagonal strips layout used for example 3 10 Magazine of Concrete Research . 16. which would be prohibitively costly in compu- ter time. this gives loads. Because the lower bound does not cor. with strips two sets of strips. the lower bound perpendicular to most of the yield lines. for the same satisfy compatibility. This reflects the fact that there is no advantage in altering the moment distribution away from the yielding region. 17 shows the unsurprising that the lower bound obtained by the strip classic yield line pattern for a clamped slab that does method differs from the true solution. It is possible that the routine has found a local maxi- mum for Q. shown in Fig. 18. It is also analysed using strips parallel to the diagonals. a similar phenomenon was observed in the previous example in the unloaded half of the slab. and that a higher global maximum exists but this cannot be checked without an exhaustive search. the slab was the true value than the lower bound produced here. Even without the fans. 16 shows the ‘yield line pattern’ that would be suggested by the lower bound that is predicted here. so cannot be load will be halved to QL2 /m p ¼ 16. Fig. It is thus satisfy the compatibility conditions. If load is shared equally between the It is also obvious that a strip analysis. as immediately obvious that the two patterns differ widely. Fig. Fig. expected to have a stress field that is close to that respond to the true collapse load this pattern does not which follows from the yield line pattern. which would give a collapse load of w ¼ 16m p /L2 . then the longest diagonal will govern running horizontally and vertically.

Denton and Burgoyne15 presented a method of Analysis with four sets of strips assessing the reserve moment capacity of a slab sub- One way of introducing Mxy moments is to add jected to a given set of moments (Mx . Strips a–a and b–b are parallel to discrete points. Deterioration.16 showed that. M y . In the Hillerborg strip analysis it is assumed circumstances. lead to structures that can safely carry the desired loads. It would also be suitable yield function so that the safe load multiplier possible to use an Ilyushin17 or von Mises yield surface can be determined. and similar studies that followed. which are intended done in the previous analysis. but instead of one set of for design of slabs and which include an optimality self-equilibrating inter-strip forces (pi above) it will condition that is not required when performing an now be necessary to have three. the moments from a–a and b–b to give the total All of this work. moment field. To these must be added use of diagonal strips spanning across the corners. as strips only satisfies the moment capacity conditions at shown in Fig. and set of inter-strip forces pi it will be necessary to deter. When the Magazine of Concrete Research 11 . M y . towards the concentrated load. (M v þ M u ) mental problem is that the strip method is poor at M y ¼ Mb þ 2 distributing loads in directions that run at an angle to both sets of strips. state at all points. That strips at 458 to the original strips. Satisfactory designs can are be produced without worrying about optimality condi. could also easily be checked mine the corresponding moment triad (Mx . For example. This overcomes the problem that axisymmetric moment patterns around a the x. and finite element methods. The funda. tributes to the moment capacity in one set of strips. With four sets of strips that will no longer be yield condition should be substituted for the normal the case. but usually in connection with point loads. such as those (Mv ⫹ Mu)/2 imparted to flat slabs by supporting columns. (Mv ⫺ Mu)/2 Analogous effects have been noted before. The highest lower bound that has been found this way is QL2 /m p ¼25. high.and y-axes respectively. Rozvany also showed that moments in the 458 elements are rotated to the 0–908 the optimal design of slabs with corners required the direction using Mohr’s circle. and have not ment capacity described above. The resulting equations been widely used in practice. (M v þ M u ) tions. each of which con. while u–u and v–v are point load necessarily involve torsion. in certain addressed. Mxy ) values at one point It should be noted that this analysis with four sets of the values from four strips have to be combined. But that is not (M v  M u ) M xy ¼ (15) the problem that is being addressed here where a good 2 lower bound analysis technique is required. it is possible to check whether May’s nored. The lower part of the figure shows how the allowed in Hillerborg strips. as has been mally built into computer programs. Because the method presented here checks the yield interactions between the strips can be conveniently ig. Mxy ) at provided that the program could be told how the re- any point of interest. 19. which must be compared with the mo- were concerned with the design problem. Mv Mu (Mv ⫹ Mu)/2 Hillerborg11 himself recognised the problem and intro- (Mv ⫺ Mu)/2 duced the idea of a rectangular corner-supported ele- ment which distributed the concentrated forces from the column to conventional strips elsewhere. To determine the (Mx . For any given combination of the expanded condition at any particular point. 19. analysis. Other appropriate yield functions could be The question of the moment capacity has to be used. Load can then be shed from long strips to short a Ma strips in the corners. thus reduced capacity. This can then be compared with a duced capacity would be distributed. Combining moments from four strips. especially when Mxy moments were that there are two layers of steel. for analysing steel structures. Mxy ). which is not at 458. M y . The applied load method differs from the Wood–Armer equations nor- can still be applied to one set of strips. the Johanson yield criterion is unconservative. The method can be applied to any suitable yield func- tion. This is well below the collapse load predicted by having strips par- allel to the sides.13. May et al. combined with the Mx ¼ Ma þ 2 use of the Wood–Armer14 equations. Automated lower bound analysis of concrete slabs Mb is linearly (although not proportionally) related to the b v u relative stiffnesses of the two strips that cross at any point. which need to be predefined. Rozvany12 and Morley13 introduced discontinuities in the moment field in two orthogonal directions. which allow the loads to be carried by diagonal shear forces Fig.

19. The mo. moments in all four strips are known. but is bound on the collapse load of the slab. and uses the Denton and An analysis of a square slab has also been carried Burgoyne approach. as Other analyses of square slab they are when only two orthogonal strips are used. so the results should be directly out using a non-linear finite element program (Aba- comparable with Fox’s exact solution and normal yield- qus). The slab was modelled as a homoge- when only two orthogonal strips are used. Although linear finite elements. using the pattern shown in Fig. the capacity can be checked at any point. It is possible that this is QL2 /m p ¼ 47.880. 21. A revised version of the program has been written for the particular case of a square clamped plate under Finite element analysis uniformly distributed load. at the expense of significantly increased computation time. but when four There have been many upper bound analyses of the strips are used. This requires specific dimensions and material line analyses. but it does represent a valid lower than the loads given by the upper bound solution. 21. they are too complicated for has to be applied to a regular square grid so that routine use. at which point both concrete crushing not the highest lower bound that could be achieved with and steel yielding are taking place. the steel starts to yield at QL2 /m p ¼ remarkably similar to the exact form found by Fox.895 and an improved version by strips are not perpendicular to the supports. several of which were summarised ment capacity has to be checked at locations where the by Fox. Predicted yield line pattern for example 3 from Fig. as would normally be undertaken directions of yielding are not shown. layer of reinforcement at mid-depth (E ¼ 28 GPa.10 Wood’s analysis18 gave QL2 /m p as 43.Burgoyne and Smith moment capacities of two strips are independent. which shows how uninformative such programs can be. If analysed using is yielding at the highest lower bound found. The refined method thus simpler than that of Fox. so tries in the slab so that only one quarter has to be the results will not be exactly comparable to the other analysed. 6. and each neous material with an elastic modulus and moment calculation is more complicated. The load–deflection curve for the finite ele- Sagging yield Hogging yield Fig. The diagonal Mansfield19 gave 42.84. m p ¼ 20 kNm). The analysis of the slab using a fully which is much closer to Fox’s value than that obtained non-linear procedure gave an ultimate collapse load of with only two sets of strips. Analysis using an automated yield line moments in each strip are calculated at the same point. The program exploits the various symme- properties. Yielding locations in highest lower bound COBRAS5 12 Magazine of Concrete Research . square slab problem.852. 20. so the Morley20 gave 42. the pattern is for assessment.63. obtained by increasing the number of strips. but even so there are far more variables than results quoted. and although these patterns are length of each strip varies. enhanced by the membrane action which both upper Figure 20 shows the nodal locations where the slab and lower bound methods ignore. program (COBRAS)5 gives an upper bound of QL2 / It also means that a more accurate solution will be m p ¼ 43. the highest lower bound found is QL2 /m p ¼ 41.99. by di- capacity equivalent to a concrete slab with a single viding one quarter of a fixed plate into a 6 3 6 grid. Nevertheless. various complications arise. and will also model membrane effects. This load is higher this layout of strips.

the significant advantages Cobras YL that could be offered for the assessment of existing Fox (exact) structures if the techniques described here could be LB with 4 strips 8·0 generalised. to analyse slabs using the Hillerborg strip method. and it is mechanism can happen. but the method converges very slowly. This explains why. and the lower bound method is 4% low.85 1 tions made in the original design. This technique starts by are shown in Table 1. It also shows. than when simple elastic analyses were performed.586 dure to prevent the solution stopping at a non- Lower bound with two strips parallel to sides 32 0. and are likely to become stuck on a local maxi- mum. but this must include a proce- Lower bound with simple diagonal strips 16 0.88 1.026 Fully plastic finite element 47. real problems the exact solution is not known. 22.112 the method to be used for assessment of existing Classic yield line analysis 48 1. and then slowly reducing is 3% high. in the bounds shown for comparison.e. 22. (d) A valid lower bound can be found from each of Method QL2 /mp Factor these equilibrium states. In the jump size as the solution is approached. is much harder to apply. 22. (b) It is necessary to include self-equilibrating inter- 0 strip forces to allow load to be transferred into 0 0·05 0·10 0·15 0·20 strips that are otherwise unloaded. same as that used when there are only two sets of they only condemn the structure when a global failure strips.e.120 structures. As expected the upper and lower allowing quite large random jumps in the various para- bounds bracket the exact solution. Comparison of predictions for square slab under of states of self-stress.001 calculations are no longer available.13 0. all of the simple lower bounds significantly underestimate the load capacity. For more complex problems it would probably better be Discussion better to use an alternative technique. all valid combinations forces and moments that satisfy equi- Table 1. when yield is reached anywhere in the structure. so much less unnecessary repair would be carried out. Values for QL2 /mp using various analysis methods librium can be described. Central deflection: m (c) The load distribution within the slab can be ex- pressed in terms of an equilibrium state and a set Fig. as observed by no means clear that the true optimum has been in Fig. some bounds on the degree of uncertainty about the because the structure is deemed to be at its capacity slab’s capacity. Upper The optimisation method used for this analysis is the bound methods are not concerned with local failure. even if those Best yield line: Mansfield19 and Morley20 42. whereas even the worst upper bound is quite close to the true plastic collapse 10·0 Classic yield line load.63 1. (g) Comparison with published values shows that low- Magazine of Concrete Research 13 .373 Lower bound with optimised diagonal strips 25. Automated lower bound analysis of concrete slabs ment analysis is shown in Fig. Far fewer structures would be condemned. however. The COBRAS result meters (here the pi values). with the other reached. although ‘safe’. Lower bound with four strips 41. Hill-climbing techniques are greedy.7 which has been used to solve the notorious The results of all the different analyses and theories ‘travelling salesman’ problem. (e) An automated procedure can be used to find the Linear elastic finite element 6. (a) It is possible to construct an automated procedure 2·0 Linear f.19 0.99 1.747 optimal position. By varying the amount of uniformly distributed load each state of self-stress that is added.160 highest lower bound. The following conclusions can be drawn from this LB (diagonal) study. sense that they walk as fast as possible up the nearest hill. so this It is clear from the present work that the lower bound method offers the real possibility of giving the checker approach. or at a point from which they cannot escape. which allows COBRAS yield line 43. which is normally used only for design. LB (2 normal strips) 6·0 Load: kN/m2 LB (optimised diagonal) Conclusions 4·0 Non-linear f.84 0.961 ( f ) The method can be used to recover the assump- Fox’s exact solution 10 42. such as simulated annealing.

tions. O’Dwyer D. Chichester. N.. 76. C. N. P. 146. 4. T. 2001. T. Betong. Collins E. Wood R. 22–26. (k) Lower bound methods are governed by local ef. 121–155. and Burgoyne C. 1226. 9. London. H. No. Cambridge University Press. 1976.. concrete slabs using a modified strip method. Mansfield E. 2. Concrete Slabs. T. Proceedings by global effects. Viewpoint. 2004. bridges. 1982. Samad A. 82. Montague P. 147–152. 2002. Reinforced Concrete Designers Handbook. T. No.. Proceedings of Insti- 1976. A. 1981. of Institution of Civil Engineers. 2nd edn. No. Press W. 6. Concrete. 18. 1957. Numerical Recipes in C++. Denton S. bridge. Hillerborg A. 277. subject to bending and twisting moments. This Cambridge. 8. macr. 17. E. Exact Ilyushin yield 5. 3. No. 2. 4. 175– 181. 9. J. 41. Strip Method of Design. and Brennan M. forced concrete slabs and bridge decks. No. Limit analysis for plates: the exact solution for a is because the Mxy moments are being ignored. 1986. Horwood. 74. Teukolsky S. Morley C. but they may be well Low Transverse Reinforcement. 1965. Generalised collapse analysis of concrete surface. The reinforcement of slabs in accordance with a References predetermined field of moments. 319–320. S. Flannery B. I. Burgoyne C. Oxford. Calladine C. Fraser A. 1974. 121.Burgoyne and Smith er bounds can be calculated. Concrete. Cam. Wood R. 34. 2. Rozvany G. and 182. The assessment of rein- The Structural Engineer. 9th edn. Hillerborg A.. The Art of Scien. 1997. The behaviour of reinforced concrete elements 3. 447– 460. Slough. Reynolds C. The Structural Engineer. and Steedman J. No. Magazine of Concrete Research. Are structures being repaired unnecessarily? 15. Optimal Design of Flexural Systems. plates with square yield diagram. Fox E. R. 1968. University of Cambridge. London.00091. 1. 13.2008.1680/ 30. 10. tution of Civil Engineers. H. 241. Perga- ( j) The method is not restricted to one yield function. 17. MEng Thesis. A. J. Strength Assessment of Concrete Bridge Slabs with Discussion contributions on this paper are welcomed by the editor 14 Magazine of Concrete Research . W. and of London. University of below the true collapse load of the structure. Jämviksteori för armerade betongplatter (Equili. Design and analysis of Thames and Hudson. H. 171– 16. No. doi: 10. Lodi S. R. Structures and Buildings. closer to the true collapse load. 329–333. Spon Press. S. mon. Middleton C. 1996. 1998. No. H. 19. J. Magazine of Concrete Research. 12. 20. 1968. 1961. 1993. M. 161–171. No. whereas upper bound methods are governed lages or hillerborg slabs under concentrated loads. 311–338. 2000. Plasticity for Engineers. H. Hillerborg A. Series A. R. Equilibrium design solutions for torsionless gril- fects. clamped square plate of isotropic homogeneous material obey- (h) By adding two additional sets of strips at 458 a ing the square yield criterion and loaded by uniform pressure. moment capacities are no longer independent. Morley C.. higher lower bound can be obtained. and O’Brien E. 14. Burgoyne C. The advanced strip method: a simple design applied moments in four sets of strips since the tool. 2008. Plastic and Elastic Design of Slabs and Plates. G. PhD Dissertation. (i) More care is needed when combining together the 11. The Ultimate bending Strength of Reinforced tific Computing. 1956. Proceedings of Royal Society 7. Part 2. Series A. 2. J. 8. A. Theory and Applica. 81. No. Studies in collapse analysis of rigid-plastic neer. 69–76 and correspondence by Armer G. brium theory for concrete slabs). May I. 1. 1113–1131. Vetterling W. International Journal of Solids and Structures. Structural Engi. 1265. which is much Philosophical Transactions of the Royal Society of London.