You are on page 1of 10

Control of Stresses in a Multi-Span Bridge Nicholas Shinder

1
Control of Stresses in a Multi-Span Bridge Nicholas Shinder

1. Introduction

The Britannia Bridge is to be built to span the Menai Straits. The specification of the design is that
the bridge must be 420m long with supports 70m in from each end and one more in the middle as
shown in Figure 1. It should be built to support a uniformly supported load of 99.6kN using a box
structure with a Youngs modulus of 210Gpa and a second moment of area of 12.07m^4.

Figure 1 shows a diagram of the Britannia Bridge (diagram taken from project handout)
Edwin Clark (1819-1894), using a method of controlled settlement, already attempted a design.
This method consisted of angling each newly added bridge span up a little to be eventually lowered
again creating a hogging moment over the pier. This meant that with the right controlled settlement
distances being calculated, an optimum moment distribution could be achieved where the
maximum hogging moment equals the maximum sagging moment. Clark attempted these
calculations by hand and unfortunately made a few errors. This report outlines how to carry out the
method using MATLAB, hopefully ensuring no errors in the calculations.

A few simpler scripts were written first to calculate systems with easily calculated results in order
to check for errors in the method before designing and implementing the final script.

2. Basic Method

All of the scripts start from cutting into the beam and looking left or right to calculate the moment
at the cut. This will then give a moment distribution along the beam. Using the equation

M=EI
d
2
y
dx
2
(where M is the moment distribution, E is Youngs modulus, I is second moment of
area, x is displacement along the beam and y is the vertical deflection of the beam at point x), the
deflection distribution along the beam can be calculated by a double integration, with respect to x,
by hand. The scripts will be written to then calculate the integration constants and any unknown
reaction forces.

3. Preliminary Scripts

As mentioned in the introduction, a few scripts were written before attempting the main problem
in order to check the correct method was being used and to get to grips with the software.




Control of Stresses in a Multi-Span Bridge Nicholas Shinder
2
i) Simply Supported Beam

The first scenario to consider was a beam supported at both ends with a uniformly distributed load
of 1N/m. L and EI were also both equal to unity. Using the system mentioned in the basic method,
the moment distribution was found:


M=V a x
w x
2
2

This was then integrated once to find the slope and then integrated again to find the deflection
distribution:

y=
1
E I
(
V a x
3
6

w x
4
2 4
+c1x+c2)

Two methods were used to find the integration constants:

The first method was a hand calculation where boundary conditions were subbed in (i.e. deflection
at both ends equals zero) from which it was fairly straightforward to find c1 and c2.

The second method used a function in MATLAB called fminsearch. To use this function, a
function with N independent variables and N different independent terms must be input into the
fminsearch. The fminsearch then iterates through values from a starting point, defined as
another input, until it finds values for the variables that best fit the independent terms. It does this
using the Nelder-Mead method, which is a direct method that uses function evaluations to find a
minimum. This makes fminsearch a powerful and useful function for this project; the moment and
deflection distributions are written as functions of x and x, c1 and c2 respectively and then the
boundary conditions are put in to a separate function called OB:

M = @(x) Va*x - w*(x.^2)/2; %Moment Distribution

slope = @(x,c1) (-1/EI)*((Va*(x.^2)/2)-(w*(x.^3)/6) + c1); %slope distribution

deflection = @(x,c1,c2) (-1/EI)*((Va*(x.^3)/6) - (w*(x.^4)/24) + c1*x +c2);
%deflection distribution

%Introduce boundary condition function. Values of independent terms will be
minimised by fminsearch. Square each term to ensure each values is positive and so
there is not a large negative value cancelling a large positive value out

OB =@(c1,c2)(deflection(0,c1,c2)).^2 + (deflection(L,c1,c2)).^2;

This OB function now has to be minimised using an fminsearch, making sure the optimum values
of the constants of integration can be found to correspond to the boundary conditions of the
scenario:

[C]=fminsearch(@(c)OB(c(1),c(2)),[0 2]);
C1 = C(1);
C2 = C(2);

The fminsearch outputs a vector, C, with all of the optimum values. Therefore the lines
immediately after the fminsearch consist of picking the right values out of the vector (e.g. C1 is
equal to the first value of vector C).

Control of Stresses in a Multi-Span Bridge Nicholas Shinder
3
The integration constants calculated by the two methods were then compared and found to be the
same. This confirmed that the fminsearch method is a valid method of calculating integration
constants.

ii) Simply Supported Beam With Central Support

The second system was to look at a beam supported at both ends with a central support. The
dimensions and strength of the bridge are assumed to be the same as in the first script.

The main difference between this problem and the first one is seen when setting up the moment
equation. Due to the reaction force in the middle of the beam there is a discontinuity in the moment
equation. By hand this could be dealt with by considering each half of the beam separately if the
reactions are known but in MATLAB the heaviside function can be made use of.

Before explaining the heaviside function, the principle of Macaulay brackets must be understood.
A Macaulay bracket is in the form

(x-a

) where the bracket is equal to 0 when x<a and equal to x-a
when x>a. This would mean the moment distribution for a beam with a central strut can be
represented using a Macaulay bracket:

M=V a x +V b (x
L
2
)
w x
2
2

The rules for integrating Macaulay brackets slightly differ from integrating brackets normally:

(xa)d x =
(xa)
2
2
}
+C
m

Using these rules for integration, a deflection equation can be formed as well:

y=
1
E I
(
V a x
3
6
+V b
(x
L
2
)
3
6

w x
4
2 4
+c1x+c2)
Now these distributions have been derived, the challenge is modelling the Macaulay bracket on
MATLAB. The way to do this is using the heaviside function as mentioned before. heaviside (x-
a) is equal to zero when x<a and is equal to one when x>a. Therefore the Macaulay bracket can be
represented using a heaviside function:

Vb*(x-L/2).*heaviside(x-L/2)

This function equals 0 when x<L/2 and Vb(x-L/2) when x>L/2. This means moment and deflection
distributions can be modelled in a script:

%Expressions for moment, deflection and slope all calculated by hand.
%Constants 'c1', 'c2', 'c3' and 'c4' to be calculated using MATLAB
%c1 = Va = Vc (reactions at the ends)
%c2 = F (prop reaction)
%c3 and c4 are integration constants

M = @(x,c1,c2) - w*(x.^2)/2 + c1*x + c2*(x-L/2).*heaviside(x-L/2);

deflection = @(x,c1,c2,c3,c4) (-1/EI)*((-w*(x.^4)/24)+(c1*(x.^3)/6)+((c2*((x-
L/2).^3)/6).*heaviside(x-L/2))+c3*x+c4);

%Introduce boundary conditions: minimise the deflections at x=0,L/2,L and Moment at
x=0 to ensure they are all close to zero. Square to ensure that a big negative value
isnt cancelling out a big positive value.
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
4

OB = @(c1,c2,c3,c4) ((deflection(0,c1,c2,c3,c4)).^2) +
((deflection(L,c1,c2,c3,c4)).^2) + ((deflection(L/2,c1,c2,c3,c4)).^2) +
((M(L,c1,c2)).^2);

In this case there are 4 unknowns and 4 boundary conditions; deflections at the three supports
equal zero and moment at the end of the beam equals zero. The OB function is put through an
fminsearch and the unknowns are found.

iii) The Britannia Bridge Moment Distribution

This script uses the techniques developed in the first two problems to find the moment distribution
on the Britannia Bridge. The dimensions and properties of the bridge can be found in the
introduction and are the first entries into the script. The moment and deflection distributions of the
bridge are formed as in the previous problem. The difference is that there are a few more terms for
the reactions corresponding to the supports 70m in from each end:

%c1 = Va = Ve (reactions at the ends)
%c2 = Fb = Fd (prop reaction)
%c3 = Fc (prop reaction)
%c4 and c5 are integration constants

M = @(x,c1,c2,c3) - w*(x.^2)/2 + c1*x + c2*(x-A).*heaviside(x-A) + c3*(x-
B).*heaviside(x-B) + c2*(x-C).*heaviside(x-C);

slope = @(x,c1,c2,c3,c4) (-1/EI)*((-w*(x.^3)/6) + (c1*(x.^2)/2) + ((c2*((x-
A).^2)/2).*heaviside(x-A)) + ((c3*((x-B).^2)/2).*heaviside(x-B)) + ((c2*((x-
C).^2)/2).*heaviside(x-C)) + c4);

deflection = @(x,c1,c2,c3,c4,c5) (-1/EI)*((-w*(x.^4)/24)+(c1*(x.^3)/6)+((c2*((x-
A).^3)/6).*heaviside(x-A)) + ((c3*((x-B).^3)/6).*heaviside(x-B)) + ((c2*((x-
C).^3)/6).*heaviside(x-C)) + c4*x + c5);

The OB function then uses the 5 boundary conditions of zero displacement at all of the supports to
calculate the 5 reactions.

One problem with this script that was not a issue in the last one was that the fminsearch was not
properly converging to the solution due to the function finding a local minimum rather than the
global one. To solve this, another fminsearch using the outputs of the first fminsearch as inputs
was implemented giving an accurate answer. The resulting moment diagram is shown in Figure 2.












Figure 2 shows the bending moment diagram of the Britannia Bridge
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
5
iv) Finding The Optimum Bending Moment Diagram Of A Beam With A Central Support

This problem is based on the same beam as in the second system but in this case the middle strut
force must be found to optimise the bending moment diagram. Optimising the diagram means
adjusting the reaction forces so that a distribution with the lowest possible maximum absolute
moment along the bridge can be achieved. This will happen when the maximum hogging moment
equals the maximum sagging moment.

In this script, the method used was to try a range of support reaction values and to pick the one that
gives the lowest magnitude of the maximum moment along the beam. This meant that a moment
distribution had to be set up for each force. This was achieved using a for loop:

for k1 = [linspace(1,lengthf,lengthf)];

This shows that the script loops through all forces in a force vector of length lengthf. The best way
to explain this loop is to go through one iteration, so take for example the first iteration:

- k1 is equal to 1 and so the first value of the force vector is picked out.
- Moment distribution is expressed only as a function of the middle strut force and x (the
reaction at A can be expressed in terms of the middle force by making use of equilibrium and
the reaction at C is equal to reaction at A due to symmetry). Therefore

M=
w
2
x
2
+
wL F
2
x+F(xL/2)
where F is the middle strut force.
- The moment distribution for every value of x is then found for each value of F and put into a
matrix where each row corresponds to a different force and each column corresponds to a
different value of x. Therefore a value in the matrix gives the moment at a particular value of
x for a given F.
- To investigate the behaviour of norms, the 2, 4, 8, 16 and infinity norms were taken for all of
the moment distributions. This was achieved by putting in a nested for loop so that, still
looking at the first F iteration, the 2-norm is found and plotted and then the 4-norm and so
on. The results are shown in Figure 3. The 2-norm is the highest line and follows the shape of
a parabola. The infinity-norm is the lowest line and tends to a pair of intersecting lines.
These two lines represent lines of sagging and lines of hogging; if the strut force is zero, the
bridge will sag completely; as the force tends to infinity, the bridge will completely hog. This
means there is an optimum force where the maximum hogging moment equals the
maximum sagging moment. This is the force to be picked out and used to calculate the
optimum moment distribution and is located where the two lines intersect.

Figure 3 shows how the different norms vary with strut force
- A matrix of norms was then formed with the different norms of the moment distribution for
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
6
each force.
- This ends the loop.

As the infinity-norm of a vector gives its maximum absolute value, this is the norm that is important
in this problem. Therefore the relevant column was picked out of the norm matrix giving the
infinity-norm for each F. The force corresponding to the minimum value of the vector was found
using the find function in MATLAB to find the position of the minimum of value norm, which will
be the same as the position of the corresponding force. The force found was then plugged back into
the moment equation and the distribution was plotted and is shown in Figure 4.

Figure 4 shows the optimum moment distribution over the simply supported beam with a central strut
It is seen that the ratio of maximum hogging moment to maximum sagging moment is 1.0031 when
looping through 1000 values of F. This ratio will approach unity as the number of forces looped
approaches infinity where the maximum moments will be equal.

v) Finding The Optimum Bending Moment Diagram Of A Beam With Five Supports

This scenario uses a similar method to the one in the last script but in this case the beam has five
supports, like the Britannia Bridge, and so the reactions at A and B have to be looped through in
order to get to an optimum diagram and the reaction at C can be expressed in terms of the others by
making use of equilibrium (reactions at A and B are equal to reactions at D and E respectively due
to symmetry).

For reasons explained in part vi), the beam had unity values for the load and EI and it was 3m long
with supports m in from each end and one in the middle.

A similar loop was used to the single support system except the moment distribution was 3
dimensional. Each row represented a different value of the reaction at A (Fa), each column
represented a different value of the reaction at B (Fb) and the moment distributions for each
combination of those two forces go into the page for all x. This was achieved by nesting another
for loop to iterate through all values of Fb for each value of Fa.

The infinity-norm of each distribution for each combination of forces is then taken forming a 2D
matrix of infinity-norms. The min function returns the value of the minimum and so, as a shortcut,
it can also produce the index of the minimum and then the function ind2sub changes the index
into row and column numbers, which correspond to the appropriate Fa and Fb respectively.

[B2,I] = min(Ninf(:)); %This function returns the value of the minimum of Ninf (B2)
and it's index (I)
[i,j] = ind2sub(size(Ninf),I); %I is just an index and ind2sub converts the
information into a row (i) and column (j)

As a check, a contour plot of the matrix was produced in order to see if there was indeed a
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
7
minimum. It is seen in Figure 5 that this is the case.


Figure 5 shows how the infinity norm varies with the reaction forces at A and B
The optimum reactions are then subbed into the moment equation and the moment distribution is
plotted. In this case the ratio of the maximum hogging moment to maximum sagging moment is
1.069 when looping through 150 values of each F. Immediately it is observed that the calculation is
not as accurate as in the previous case, mainly as the number of iterations has gone down by nearly
a factor of 7. This is because, due to the use of a 3D matrix, the script takes a lot longer to run. In the
next scenario accuracy is more important and so a method to hone in on the right forces was
implemented.

4. Final Scripts

vi) Finding The Optimum Bending Moment Diagram Of The Britannia Bridge

This problem is very similar to the previous example. The difference is that all of the properties of
the real bridge are used.

The new properties were subbed in and the script was run. It was observed that the diagram was
quite far from accurate. The new values made the new diagram fluctuate a lot before converging
eventually when the script looped though a large amount of forces. Looking at the optimum
bending moment diagram, the moment at B and the moment at C must be equal and also equal to
the negative of the maximum moment. An Accuracy Index was calculated by dividing the
maximum moment by the average of the moments at B and C.
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
8

Figure 6 shows how the accuracy of the optimum moment distribution varies with number of iterations for both scenarios
Figure 6 shows how the accuracy index changes with the number of loops for both the problem
with unity values (blue) and the problem with real values (red) and it can be seen that the
behaviour of the real value model is much more erratic than the unity value problem until a high
number of loops is reached.

Unfortunately, using a high number of loops is impractical as the time to run the script becomes a
big factor. Therefore a different solution had to be found. This solution came in the form of an
fminsearch that could find the optimum forces to minimise the maximum moment along the beam:

%Define Norm as a function of just Fa and Fb. Mnew is the moment function along the
beam.
Nnew = @(Fka,Fkb) norm(Mnew(x,Fka,Fkb),inf);
%Find the minimum of the Norm function using outputs from the for loop as inputs to
the search.
[C] =fminsearch(@(c)Nnew(c(1),c(2)),[Faoptimum,Fboptimum]);

The fminsearch can not be used from the start as it needs inputs that are reasonably close to the
output. Therefore the outputs from the loop method are used as the inputs to the fminsearch.

The optimum forces were found and the optimum moment diagram (red) was plotted along with
the initial estimate from the loop method (blue) in Figure 7.

Figure 7 shows the optimum moment distribution found using an fminsearch and its improvement on the loop method
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
9
To check that the distribution was accurate the moment values at the five central peaks were
plotted and they were all found to be the same, indicating that the distribution was indeed correct.

The next part of the problem was to find the settlement distances at the supports. This was done by
a further fminsearch, similar to the earlier scripts. The deflection this time was only a function of x
and the integration constants (using the values of the optimum reaction forces) and the boundary
conditions were that the deflection at both ends had to be zero. The deflection at the supports was
found to be 0.118m at B and E and 0.275m at C. These distances are the deflections that would be
required in order to obtain the optimum moment diagram but raising the supports is impractical.

vii) Finding The Controlled Settlement Distances In Order To Obtain The Optimum Bending
Moment Distribution On The Britannia Bridge

After the optimum moment distribution had been found, the last problem to solve was how to
obtain this distribution in practice.

The method for obtaining an optimum distribution was outlined in the introduction and involves
attaching each section to the bridge at a slight angle and then lowering it to the horizontal this has
the effect of creating a hogging moment over the support connecting the new section to the rest of
the bridge. Therefore the distance that each section was to be raised to find the optimum moment
distribution had to be calculated. This was achieved by working backwards from the optimum
distribution. A general method could be developed to eliminate each section one at a time by
finding the controlled settlement to make the moment at the central support zero. The method is
shown below along with how it would work for a particular section AB in square brackets to help
illustrate the calculations.
- Optimal moment distribution is to be superimposed onto a point loaded beam with identical
dimensions and properties as the Britannia Bridge but with no distributed load and no
deflection boundary condition at the support to be removed. [A is not fixed.]
- Force at the free end on the point loaded beam is calculated by cutting at the support with
the angle and taking moments so that the moment at that point is equal to the moment at the
same point on the optimised diagram. [Take moments at B (in this case, Fa = Mb/70, where
Mb is the moment at B in the optimum moment distribution).]
- Using the fminsearch method from previous scripts, the deflection distribution and hence
the deflection at the free end of the point loaded beam can be generated. As this deflection at
the free end will be superimposed with a zero deflection, this deflection is the controlled
settlement distance in order to create the optimum bending moment distribution. [Generate
a deflection distribution for the point loaded beam and read off deflection at A. This is the
amount A has to be raised when AB is attached to the beam.]
- The point loaded moment distribution is then subtracted from the optimum moment
distribution giving a new distribution. This distribution takes the place of the optimum
distribution when removing the next section. [Distributions added and then DE is removed.
To check this, the new distribution can be plotted and the moment at B can be confirmed to
be zero].
The order of the sections being removed was AB, DE and then CD. The hardest part of the
calculation was with the removal of DE and CD. It was necessary to cut and look right instead of left,
meaning that when the distributions were superimposed, if left all in terms of a single variable x,
moments at opposite ends of the bridge would be added (eg moment at 0 on the original
distribution being added to moment at 420 on the point loaded beam). To get round this a second x
vector called x2 was introduced. This vector was the same length as x except that it went from L to
Control of Stresses in a Multi-Span Bridge Nicholas Shinder
10
0 instead of 0 to L. This meant that when adding a distribution as a function of x to a distribution as
a function of x2, the new distribution was a function of x and x2:

Mnew2 = @(x,x2) Mnew(x) - MpointDE(x2,FC,FD);

This meant that distributions could be added even if they were defined from different sides. The
moment diagram after each removal is plotted in Figure 8. The point loaded distribution in each
diagram is subtracted from the actual moment distribution on the same diagram. This new distribution
becomes the actual moment distribution on the next diagram. The diagram illustrates how the point
loaded distribution was made to have an equal moment to the actual distribution (ie the peak values
where the two lines intersect).



Raise A by .2144m



Raise E by .1966m




Raise D by .9435m






Figure 8 shows the moment distributions after the removal of each section along with the distributions to be superimposed
4. Conclusion

To conclude MATLAB was a very useful tool in this project due to its accuracy and speed of
calculations. A script that could find the reaction forces to make the maximum hogging and
maximum sagging moments equal was written, as well as a script to find the controlled settlement
distances to obtain the distribution, with both scripts took a relatively short time to run. This is
much quicker than doing the calculations by hand as well as the risk of human error being lower.
Also, because of the way the script is written, it is simple to change the values for any other bridge
of similar design and the same number of supports.

In this project, it was assumed that the forces from wind loading and the mass of trains were
negligible compared to the self-weight of the bridge. To further this project, an investigation into
what effect these factors have on the optimum moment distribution could be undertaken. The
worst-case scenarios, with a heavy loaded train over the middle of the long span of the bridge for
example, could be developed as well as looking into what wind loading would cause failure. These
results could be useful to alert operators to conditions in which the bridge should be closed.

You might also like