Basic non-linear analysis with MIDAS

Q1) What is this article for?

A1) Very basic discussion about non-linear analysis with MIDAS.

Q2) What is linear and non-linear?

A2) One of the best examples is the stress-strain relationship.

Figure 1 Stress-strain relationship (from Wikipedia)

The stress-strain relationship can be considered as linear (only when the stress and/or strain is low). If
the stress is higher than some level (in the figure, yield strength), or the strain is larger than some value
(in the figure, the strain corresponds to yield strength), the stress-strain relationship is non-linear.

This is normally classified as material non-linearity.

Q3) Is this all?

A4) No. There are more. Many engineers have been and are considering two more non-linearity. The
second is geometric non-linearity.

1
Figure 2 Deflections of a cantilever beam (from D. Singhal, et al, Large and small deflections analysis of a
cantilever beam, Journal of the institution of engineers)

Figure (a) represent the linear theory (also called small deflection/deformation theory).

The bending moment at fixed support is M = PL and no axial force in the beam, shear is constant along
the beam. This is true (or the error is small) only if the deflection u (or the rotation φ) is small (this is
why be called small deflection/deformation theory).

Figure (b) is the same as Figure (a), but somewhat exaggerated. There is horizontal displacement ux and
the moment at fixed support is M = P(L-ux). There is axial force in the beam, Psin(φ0) at the tip. The
shear is not constant along the beam and the shear direction also varies.

In the linear theory (Figure (a)), the deflection uy is linearly proportioned to the loading P. In other
words, if P becomes 2P, the deflection is 2uy, if P becomes 3P, the deflection is 3uy, etc.

In the non-linear theory (Figure (b)), this is not true anymore. The relationship between P and uy is not
linear, it is non-linear.

Q4) What is the other one?

A4) It is call boundary non-linear.

Figure 3 Cantilever with gapped support (from Nam-Ho Kim, Finite element analysis of contact problem)

2
If the tip deflection due to load is less than δ, the structural system is a simple cantilever. Along with the
load increase, the tip deflection increases, and at some point, the cantilever tip will hit the rigid block
and the structural system is not a simple cantilever anymore. The structural system is like a cantilever
with support at the end.

Q5) All these are independent?

A5) No. all these are dependent. In most cases, large deflection (or rotation) occurs with large moments,
and the section would yield or be cracked. In this case, material non-linearity also be considered.

Q6) We live in 21 century with high-end computers. Why not consider all these effects together?

A6) We can not and need not.

Do we need to consider each gravel size and reinforcement details to calculated the reinforced concrete
beam moment?

Probably not, but we may need to consider the crack effects if needed.

The same thing for the non-linear analysis. The exact (whatever it means) solution for non-linear analysis
has not been formalized yet and is not needed. What we need is the approximates within allowable
error.

Q7) In what cases should we consider the non-linearity?

A7) As a minimum,

- Columns
- Arch bridges
- Suspension bridges
- Cable-stayed bridge

Q8) Any simple examples?

A8) Example from Goto, Shigeo, Tangent stiffness equation of flexible cable and some considerations,
JSCE, Vol 270, September, 1978.

For given cable system, calculate the vertical deflection at the loading point. Cable self-weight is 1.0 t/m
and cable axial stiffness is EA = 2 550 000 ton.

The cable sag is calculated as 16.871 m (refer to a separated calculation).

3
 Figure 4 Cable structure

30

25

20
Displacement (m)

15

10

0
0 10 20 30 40 50 60 70 80 90 100
Load (ton)

Catenary Element Truss Element

4
 Figure 5 Vertical deflection vs load increase

It can be seen that the deflection vs load relationship is not linear, it is non-linear. In other words, the
stiffness of the system is dependent to the loading.

Why does the stiffness vary along the load/deflection?

5
Q9) In the bridge design, why we need to worry about geometric non-linear analysis?

A9) Say, we are finding the top displacement and base moment for the column shown below, which is a
typical case in the column design (with vertical superstructure reaction and braking force at the fixed
bearing location).

Calculate the base moment with P = 9726 kips, V = 112.5 kips, L = 148.708 ft. The column is d=10’-6”
reinforced concrete circular section with f’c = 4.5 ksi. (from Manual for refined analysis in bridge design
and evaluation, FHA)

Figure 6 Cantilever column, undeflected and deflected shape (from What is p-delta analysis? SkyCiv)

It is easy, M = VL isn’t it?

M = VL = (112.5 kips)(148.708ft) = 16730 ft-kips.

E = 1820√f’c = 1820√4.5ksi = 3861 ksi

Ig = πD4/64 = π(10.5ft)4/64 = 596.7 ft4

Δ = VL3/3EI = (112.5 kips)(148.708 ft)3/[3(3861 ksi)(596.7 ft4)] = 0.372 ft

6
Q10) Can we check this with MIDAS?

A10) Yes, refer to “EX6 Linear”.

-9726.0
112.5

Figure 7-1 Loading (kips)

0.372

0.000

Figure 7-2 Deflection (ft)

16730

Figure 7-3 Moment (ft-kips)

7
 Table 1 Linear Analysis

Top Deflection Base Moment Remark

(ft) (ft-kips)
Hand Calculation 0.372 16730 Q9
EX6 Linear 0.372 16730 Q10

Nice match, isn’t it?

8
Q11) The result from Q9) and Q10) matches very well. Again, why we need to worry about non-linear?

A11) These results are correct only for the linear case.

If P and/or Δ is large, we may need more complex analysis.

In the non-linear analysis, M = VL +PΔ.

First, we need to calculated top horizontal deflection Δ.

Assume top horizontal deflection Δ is the summation of Δ1 (top horizontal deflection due to horizontal
load V) and Δ2 (top horizontal deflection due to vertical load P). This is not correct because the rule of
superposition is not valid in the non-linear analysis.

EIΔ1 (top horizontal deflection due to VL) = (1/2)(L)(M=VL)(2L/3) = VL3/3

Assume deflection curve due to P and V is cubic (3rd order). This is not correct because deflection curve
can be assumed as cubic (without any contradiction) only if there is no axial force within the element
and the moment varies linearly within an element. (Also, the deflection curves with axial compression or
axial tension are not identical.)

EIΔ2 (top horizontal deflection due to PΔ) = (3/4)(L)(M=PΔ)(3L/5) = 9PΔL2/20

EIΔ = EA(Δ1+Δ2) = VL3/3 + 9PΔL2/20

(3861 ksi)(596.7 ft4)Δ = (112.5 kips)(148.708 ft)3/3 + 9(9726 kips)Δ(148.708 ft)2/20

Δ = 0.525 ft

M = VL + PΔ = (112.5 kips)(148.708 ft) + (9726 kips)(0.525 ft) = 21835 ft-kips, from simplified P-delta
(non-linear) analysis.

Even for this medium sized, practical case, the effects of non-linearity cannot be ignored (21835/16730 =
1.3).

Computers do what we let them do without any judgements. It is totally our responsibility to select
correct option to do the analysis.

9
Q12) Can we do this calculation with MIDAS?

A12) Absolutely, refer to “EX6 Large Displacement”.

0.474

0.000

Figure 8-1 Deflection (ft)

21343

Figure 8-2 Moment (ft-kips)

10
Q13) Why the results in Q11) and Q12) are not identical?

A13) There are at least three reasons.

Before starting long and/or boring discussions, let’s divide the element in Q12) into ten elements. Refer
to “EX6 Large Displacement 10 Division”.

0.502

0.426

0.352

0.281

0.215

0.155

0.102

0.059

0.027

0.007

0.000

Figure 9-1 Deflection (ft)

MAX: 21613

Figure 9-2 Moment (ft-kips)

11
 Table 2 Non-linear analysis

Top Deflection Base Moment Remark

(ft) (ft-kips)
Hand Calculation 0.525 21835 Q11
EX6 Large Displacement 0.474 21343 Q12
(1 element)
EX6 Large Displacement 0.502 21613 Q13
(10 elements)
EX6 P-delta 0.502 21614 Q14
EX6 Linearized P-delta 0.502 21614 Q15

As mentioned in A11), the deflection curve under axial loads (either tension or compress, which are not
identical each other) is not cubical. However, most commercial programs, include MIDAS, apply cubical
deflection curve for beam elements, which gives exact (or without any contradiction) results only if the
moment varies linearly between nodes and no axial force within the element.

In the beam elements, the results are independent to the number of divisions if the stiffness is constant,
no axial force, and no loads within the element.

Our cantilever beam has axial force P and the deflection curve is not cubical, but MIDAS assumes the
deflection curve as cubical, so the results are dependent to number of divisions. In other words, a single
cubical curve cannot represent the more complex real deflection curve.

If we perform non-linear analysis with cubical deflection curve, we do need more elements to get the
better results.

Finally, hand calculation results (Q11) show close match with those from non-linear analysis (Q13).

The difference between Q11) and Q13) will be discussed later.

12
Q14) Is this the P-delta analysis?

A14) One of the confusing parts in non-linear analysis is the terminology. P-delta analysis is a part of
non-linear analysis, however, can be interpreted in different ways to each engineer. The author prefers
to interpret P-delta analysis as finite displacement analysis, in contrast to large displacement analysis
(Q13).

In short, P-delta analysis constitute equilibrium equations for initial shape but consider the change of
axial force (and stiffness matrix) in each iteration. This means we do need iteration and the rule of
superposition is not valid.

Detailed discussion about P-delta analysis can be found from the MIDAS Analysis Manual.

The results from the P-delta analysis is as shown below. Refer to “EX6 P-delta”.

0.502

0.000

Figure 10-1 Deflection (ft)

21614.0

Figure 10-2 Moment (ft-kips)

13
Q15) Can you make me more confusing?

A15) Yes. There is one more method to be mentioned.

It is called linearized finite displacement analysis. In this method, the axial force in each element remains
constant and this means no iteration is needed. In other words, the rule of supposition is valid and our
life is much easier. This method can be applied if the change (not the magnitude) of the axial force can
be ignored and works for most structures.

The results from the linearized finite displacement analysis is as shown below. Refer to “EX6 linearized
P-delta”.
0.502

0.000

Figure 11-1 Deflection (ft)

21614

Figure 11-2 Moment (ft-kips)

We can check this result with hand calculation. Actually, linearized finite displacement method is just
considering stiffness reduction (with axial compression) and/or stiffness increase (with axial tension).
The stiffness in this example can be estimated as

K = 3EI/L3+ (6/5)N/L

= 3(3861 ksi)(596.7 ft4)/(148.708 ft)3 + (6/5)(-9726 kips)/(148.705 ft)

14
= 302.6 kips/ft – 78.5kips/ft

= 224.1 kips/ft

Δ = P/K = (112.5 kips)/(224.1 kips/ft) = 0.502 ft

M = VL +PΔ = (112.5 kips)(148.708 ft) + (9726 kips)(0.502 ft) = 21612 ft-kips

Both well match with computer output.

This example is very typical for medium sized bridges. Even in this case, the stuffiness reduction due to
axial force is 26% (=78.5/302.6). These effects must be considered in the design.

15
Q16) What is the findings up to this point?

A16)

1. If there is axial force within the elements, the effects of axial force must be checked even in the
medium sized columns.
2. There are at least three levels of analysis methods in the geometric non-linear analysis, large
displacement, finite displacement, linearized finite displacement.
3. In the large and finite displacement, the rule of superposition is not valid.
4. In the practical structures, linearized finite displacement analysis gives satisfactory results in
many cases.
5. In the non-linear analysis, even the terminologies are somewhat confusing and we must clearly
understand what the differences are.

16
Q17) What makes suspension bridge analysis unique/difficult?

A17) Let’s go back to Q9) but consider P as tension force.

Figure 12 Cantilever beam with axial tension (from Mechanics of materials by Goodno)

From linear analysis, the end moment is M = 16730 ft-kips.

However, from non-linear analysis, the end moment is somewhat reduced (Refer to “EX-6 P-delta 10
division Tension”).

0.000 -0.004 -0.017 -0.037 -0.063 -0.094 -0.129 -0.168 -0.209 -0.252 -0.295

Figure 13-1 Deflection (ft)

MAX: -13857

Figure 13-2 Moment (ft-kips)

17
 Figure 14 Cantilever beam with spring support (from Shegg study)

The structural system in Figure 12 is equivalent with that in Figure 14.

These computer output can be checked by hand calculation.

K = 3EI/L3+ (6/5)N/L

= 3(3861 ksi)(596.7 ft4)/(148.708 ft)3 + (6/5)(9726 kips)/(148.705 ft)

= 302.6 kips/ft + 78.5kips/ft

= 381.1 kips/ft

Δ = P/K = (112.5 kips)/(381.1 kips/ft) = 0.295 ft

M = VL +PΔ = (112.5 kips)(148.708 ft) – (9726 kips)(0.295 ft) = 13859 ft-kips

Both well match with computer output.

This example shows axial tension decreases beam moments and deflections. In other words, the
bending moments can be reduced 17% considering the axial tension force and this is called tension
stiffening effects.

We can calculate bending moments from simple linear analysis (M = 16730 ft-kips), but the output is
somewhat larger than that from non-linear analysis (M = 13857 ft-kips). In this short (L=148.708 ft), stiff
structures these effects may not be significant (17%), however, this effect is huge in the long span
suspension bridges.

If the stiffness of the beam is nominal like truss/cable elements, the cantilever in Figure 12 is unstable in
the linear analysis and the vertical deflection would be infinite. In the non-linear analysis, the cantilever
has vertical stiffness as the function of N/L and can resist load V.

This is the key concept for the suspension bridge analysis.

18
 Figure 15 Suspension bridge (from Science photo library)

In the suspension bridge, the girder is supported vertically by main cable tension force through hangers
which can be assumed not be extensible. Within the linear analysis, we have to deal with (incorrect)
huge moments, our section would be huge, and finally the self-weights would be too heavy and finally
we cannot design long span suspension bridges. In the column/arch design, the non-linear analysis is
safety issue, but in the suspension bridges, it is rather economy or feasibility issue. We do need non-
linear analysis to design a long span suspension bridge.

19
Q18) Can you show me any example for real suspension bridge analysis using MIDAS?

A18)

Figure 16 Example bridge (from Steel bridge III by Hirai)

Figure 16 is so called A-Bridge which was selected by Dr. Hirai as an example. This A-Bridge is medium
sized and main span is 808 m (2650 ft).

For the modeling methodology refer to the MIDAS manual.

40000 0.000

-10.000
30000

-20.000
20000
-30.000

10000
-40.000

0 -50.000

-60.000
-10000

-70.000
-20000
-80.000

-30000
-90.000

-40000 -100.000

Linearized Finite Displacement Linearized Finite Displacement

Linear Linear

Figure 17 Girder moment comparison, Linearized finite displacement analysis vs Linear analysis

20
In Figure 17, blue line represents the bridge shape. The solid red line is live load moments from the
linearized finite displacement analysis which is the simplest methods which can consider the tension
stiffening effects from the cable. The dashed red line is live load moments from the ordinary linear
analysis. The differences are just dramatic, and this is why we couldn’t design/build long span
suspension bridges with through linear analysis.

In the linear analysis, we cannot consider the cable tension stiffening effects, and the girder is supported
by the vertical portions of cable elongation and girder stiffness, both are nominal. Recall that the
transverse stiffness component of truss element is zero in the linear analysis. If the cable is horizontal,
the cable cannot contribute any stiffness to the girder deflection in the linear analysis.

This concept was realized by Melan and the non-linear equation itself was formularized in 1888,
however, we had to wait sometime until brilliant engineer Moisseiff solved this equation and designed
the Manhattan bridge using his own solution. The Manhattan bridge was opened to traffic in 1909 and
bridge engineers call it the 1st “modern” suspension bridge.

Figure 18 Manhattan Bridge (from Wikipedia)

21
Q19) Why cable-stayed bridge is difficult to analyze?

A19) Cable-stayed bridges, includes so-called Extradosed bridges, do not have any classical solutions.
Computer based displacement method is the only way to analyze cable-stayed bridges and it is hard to
check the output.

Also, in the cable-stayed bridge analysis, following three non-linearity should be considered.

The first one is cable sag effects.

Figure 19 Cable sag (from Cable supported bridge, 3rd ed., by Gimsing)

Cables may look straights but cannot be, except vertical cables, due to cable self-weights.

22
 Figure 20 Cables with different sag

Two cables in Figure 20 have the same chord length L, same elastic modules E and same area A. If we
ignore the sag effect, the cable axial stiffness is identical for each case as K=EA/L, however, we can
catch/feel that the axial stiffness shall not be identical, and the second case shall be more flexible. How
can we catch the stiffness differences in this case? We have some options.

A. Catenary elements: Catenary elements give theoretically exact solutions and

intermediate nodes are not needed expect loading points. For detailed element
formulation, refer to MIDAS manual.
B. Non-linear (or linearized) truss elements as shown in Figure 21. Basically, divide a single
cable into multiple non-linear or linearized truss elements to consider the cable shape
and internal forces.

23
 Figure 21 Single cable modelling with multiple non-linear (or linearized) truss elements

Actually, this is the way we model the cables in a suspension bridge. We have to divide the main cables
into many numbers of straight truss element due to the hanger connection. We can model each cable
segment using catenary element, but little benefit can be expected.

C. Equivalent truss elements: With the sag increases, the cable stiffness decreases. We can
consider these effects as reduced area A effect. Here, w is cable weight per unit length and
T is cable tension force.
A
Aeffect = 2 2
w L
1+ 3
EA
12 T
T is not constant, and this method is also kind of non-linear analysis. However, if we can
assume T as constant, the calculation is a lot easier.

There are pros and cons for each method.

Catenary elements give the most exact solutions, but iterative analysis is essential, and the rule of
superposition is not valid. This means we cannot use influence line analysis and the calculation would be
very voluminous and may not practical in many cases.

Non-linear truss elements had been used before the age of catenary elements (which was not that long
ago) but looks like not much benefit comparing to other methods in this age. If we need to check the
displacements within a single cable element, the linearized truss elements may be a good option, but we
have to calculate each nodal coordinate and internal forces. The equivalent truss element has been used
long time and still gives very reasonable results with minimum efforts.

The second is P-delta effects.

In the cable-stayed bridges, not only the tower, but also the girder resist axial compression and we need
to consider the P-delta effects for girder design.

The last one is also parts of P-delta effects.

24
In the cable-stayed bridges, the loading between cable anchoring points are transformed into equivalent
nodal point loads at the nodes. This converting equation for a beam with axial force is different from
that of conventional beam elements. Strictly speaking, we have to calculate the equivalent loads using
more complex equations and which is also non-linear. However, it is known that conventional methods
give reasonable results in most cases, one of the reasons is the beam element length is restricted to
relatively short due to the cables.

25
Q20) What is the Extradosed bridges?

A20) This is a very interesting topic.

The strand that used for the prestressed concrete and for the cable-stayed bridges are just the same.
The only difference is the test item and the test is more severe for the cable-stayed bridges. Again, the
manufacture never makes anything less qualified for the prestressed concrete strands, just the test
items are different.

However, the allowable stresses are different for each case. It is 0.6fpu for the prestressed concrete
strands and 0.45fpu (or 0.40fpu according to Specifications) for the cable-stayed bridge strands.

It is understood that this 15% - 20% differences are mainly from the stress variations (not the values).
For the prestressed concrete bridge strands, the stress variation is relatively small, and it is rather large
for cable-stayed bridges.

Figure 22 Allowable stress vs stress variation (from Kasuga, Extradosed bridges in Japan, 2006)

From this, Kasuga proposed to use 0.60fpu if the stress variation is less than 14.5 ksi, 0.40fpu if the
stress variation is larger than 19 ksi and interpolate between these two limits for prefabricated wires.

Extradosed bridges are something between prestressed concrete bridge and long-span cable-stayed
bridges and we can increase the allowable stress according to the stress variation in the cables and
finally the design would be economical.

26
 Factored/Weighted Cable Weight
900

822
Factored/Weighted Cable Weight (kips)

800

755 752

701 697
700
664 657
653

619 623 625 618

596 594 590
600 586
573
565 560
554
543 535 542
524

500
40 50 60 70 80 90

Tower Height (ft)

2.0 2.3 2.7 3.0

Figure 23 Minimum cost tower height for 500ft span railway bridges.

Figure 23 is preliminary cost comparison to find the most economical tower height for 280ft + 500ft
+280ft, three span extradosed railway bridges. Figure 23 shows tower height 50ft is the most
economical and the optimal H/L ratio (= 50/100 = 1/10) is quite low comparing to conventional cable-
stayed bridges.

27
Q21) Simple example

To understand the differences between each non-linear theory, let’s consider a simple example as
shown below. (Lee, Design and Analysis of Suspension Bridges, 2001)

Consider a truss element, L = 1 m and the self-weight is Pv = 100 ton. Top end is supported by a pin
support. Let’s calculate the displacement and axial force for given horizontal force Ph = 100 ton. Assume
EA = ∞ to ignore the axial elongation.

1) Linear methods

In the linear methods, this structure is unstable. We cannot analyze this structure.

2) Linearized finite displacement methods

The horizontal stiffness is Kh = Pv/L = 100 ton/m and horizontal displacement is Δh = Ph/Kh = (100 ton) /
(100 ton/m) = 1m. Truss axial force is 100 ton. Equilibrium equation is formed for before-displaced
condition.

Deformed Shape for Ph = 100 ton

3) Finite displacement methods

28
For this example, same as linearized finite displacement methods.

4) Large displacement methods

The truss axial force is 100√2 = 141.4 ton from force equilibrium after displaced. The displacements are
Δh = 1/√2 = 0.7071 m and Δv = 1 – 1/√2 = 0.2929 m.

For given load Ph = ∞, Δh = Δv = 1 m ignoring axial elongation as expected. In other words, the truss
shape is changed from vertical to horizontal due to infinite horizontal loads. This is a very simple
example, but one of the extreme non-linear problems.

Deformed Shape for Ph = 100 ton

Deformed Shape for Ph = ∞

Discussion

In the (linearized) finite displacement methods, the force equilibrium is considered for before-displaced
condition and we cannot consider the effects of axial stiffness, EA, and this is why the horizontal
displacement Δh is larger than those from large displacement methods. Also, we cannot calculate
vertical displacement Δv and the after-displaced element length is not correct.

29
In the large displacement methods, the total force Ph is applied without any divided load steps. In other
words, total load is assumed to be applied at once. This is applicable only to simple cases (and with good
solver) and normally we have to divide the loads into small load increase, like 0.1Ph.

The (linearized) finite displacement has been very well formulized we can expect the same solution from
any commercial programs. However, the large displacement method has not, and every calculation
procedure has its own assumptions and limitations. We may not need to fully understand the detailed
calculation procedures, however, we have to understand the basic assumptions and limitations
otherwise we may misunderstand the output.

Finally, in most bridge structures, the linearized finite displacement method gives very satisfactory
results because the displacements are limited for servicebility.

30
21) How about arch bridges?

A21) Many engineers classify arch bridges into three types.

The first one is the Tied arch.

Figure 24 Tied arch

31
=================

below is for next time, not this time.

Q13) So, is it enough to select P-delta option to analysis the example structure?

A14) Unfortunately not.

From the formula,

M = VL + PΔ

EIΔ = EA(Δ1+Δ2) = VL3/3 + 9PΔL2/20

we can catch that the moment is a function of top horizontal deflection Δ, and Δ is a function of flexible
stiffness of the column, EI.

In the statically determinate structures like simple beams or cantilevers, moment is independent to the
flexible stiffness EI. However, this is valid only for the linear analysis.

In the non-linear analysis, the moment is dependent to the flexible stiffness EI.

Along with the moment increases, concrete cracks. With the crack develops, the deflection increases,
and the moment increases. This is a kind of ponding problem and iterative calculation is unavoidable.

Q14) How to do this calculation?

A14) It is tricky, however AASHTO and ACI give simplified and organized way to do this calculation.

AASHTO gives M = 108505 ft-kips (108505/16730 = 6.5, not a typo), and ACI gives M = 27930 ft-kips
(29730/16730 = 1.8).

The ratio 6.5 and 1.8 are somewhat surprising. Actually, the column diameter D = 10’-6” was
intentionally selected to maximize these ratios, to make the readers to be surprised. Application with a
little larger diameter will reduce these ratios dramatically.

Q15) Any detailed calculation?

A14) Refer to “Stability Analysis of a Tall Concrete Pier Column”.

32
33