CAESARIICAUx2015 9/28/2015

Failure to Launch

Resolving nonlinear convergence issues

in CAESAR II

Intergraph 2015

Do you ever see this screen in CAESAR II?

Intergraph 2015

1
CAESARIICAUx2015 9/28/2015

Heres the model that produced those results:

Intergraph 2015

BACKGROUND

2
CAESARIICAUx2015 9/28/2015

Nonlinear boundary conditions and CAESAR II

CAESAR II has always accommodated nonlinear restraint conditions

Modeling (e.g., +Y, gaps, friction, rod swing)
Analysis
Codes: T1 (EXP)
CAESAR II: L1-L2 (EXP)
The piping codes say little about nonlinear support conditions in piping systems.
Nonlinear references in B31.3:
Appendix P - ALTERNATIVE RULES FOR EVALUATING STRESS RANGE: P300 GENERAL (a)
This Appendix provides alternative rules for evaluating the stress range in piping systems. The
method is more comprehensive than that provided in Chapter II and is more suitable for computer
analysis of piping systems, including nonlinear effects such as pipes lifting off supports.
Appendix S - PIPING SYSTEM STRESS ANALYSIS EXAMPLES:S302 Example 2: Anticipated
Sustained Conditions Considering Pipe Lift-off
302.3.5(c) Stresses Due to Sustained Loads, SL. :edited in 2010
The sum of the longitudinal stresses due to sustained loads, S , e.g., the pressure and weight in
L

any component in a piping system (see para. 320), shall not exceed S , where S is taken from
h h

Table A-1 at the metal temperature of the operating condition being considered.
302.3.5(d) Definition of SL used in (1b): SA=f [ 1.25 (Sc + Sh) - SL ] :edited in 2014
Stress due to sustained loads; in systems where supports may be active in some conditions and
inactive in others, the maximum value of sustained stress, considering all support conditions, shall
be used.
Intergraph 2015

What is a nonlinear restraint or condition

in CAESAR II?

CAESAR II uses F=KX to determine the position of every node in the piping system
F is the load vector under evaluation
K is the system and boundary condition stiffness
K is constant

Intergraph 2015

3
CAESARIICAUx2015 9/28/2015

What is a nonlinear restraint or condition

in CAESAR II?

Not all supports respond in a linear fashion. Examples include:

A resting support (+Y) that may lift off
A gap on a guide
Friction on a support is doubly nonlinear (magnitude and direction)
Friction magnitude may have the support stick or slip
With slip, the friction force is added to F and opposite the slip direction
CAESAR II tests the solution to assure that all nonlinear conditions are satisfied for each
load case
CAESAR II will iterate until all nonlinear conditions are consistent using a linear K (&
adjusted F)
Some programs just stop
CAESAR II has tolerances for some of these checks

Intergraph 2015

Heres the stiffness matrix for a pipe element

Here is the pipe

stiffness in the Y
direction for the
From node

Intergraph 2015

4
CAESARIICAUx2015 9/28/2015

Now, include the Y restraint

The stiffness matrix now

includes the added
stiffness of the restraint
in the Y direction (KT).

Intergraph 2015

This added stiffness is constant

The same restraint stiffness (KT) is effective for any Y position of this node.
A constant value for stiffness K models a linear response.

A Linear
Restraint

Intergraph 2015

5
CAESARIICAUx2015 9/28/2015

But it need not be constant

Here, the restraint stiffness (K) added to the system stiffness matrix is a function of the
nodes position.
This changing stiffness is a nonlinear boundary condition.
Nonlinear restraints require special treatment in the stiffness matrix.

A Nonlinear Restraint

Intergraph 2015

CAESAR II will adjust the stiffness matrix to

suit the current load conditions

Depending on the calculated, final position of the restraint node, the program will check the
restraint stiffness defined at that position.
If the stiffness used in the analysis matches the defined stiffness, the solution for this
restraint is correct. If the defined stiffness for that position is incorrect, the program will
update the systems global stiffness matrix with the proper restraint stiffness, and the
process continues.
CAESAR II will continue until all the nonlinear conditions applied in the global stiffness
matrix are consistent with the calculated position.
This process restarts for each basic load case in the analysis.

Intergraph 2015

6
CAESARIICAUx2015 9/28/2015

CAESAR II will adjust the stiffness matrix

(assume and test)

Here is a common resting support,

a +Y restraint in CAESAR II terms:

Initially, the restraint will be included in the model so the

stiffness matrix will reflect this:

If the solution shows a -Y load on that restraint, this support is

modeled correctly for this load case. If not, the program
resets the stiffness matrix to appear as this:

If this updated solution shows a +Y deflection, this support is

modeled correctly for this load case. If not, the program
resets the stiffness matrix to appear as this:

and this iteration continues for all the nonlinear restraints

and for each load case analyzed until the stiffness matrix is
consistent with the final position. ?

Intergraph 2015

Nonlinear solution in CAESAR II

lets look at the model in the opening slide

Intergraph 2015

7
CAESARIICAUx2015 9/28/2015

Nonlinear solution in CAESAR II

lets look at the model in the opening slide

Initial program guess friction is sufficient to hold the pipe from horizontal movement
stick (include rigid restraint in Y and friction stiffness in X & Z at 40)

(Although stick, note deflection at 40)

+Y load @ 40 indicates liftoff
Update the linear assumption(s) and try again

Intergraph 2015

Nonlinear solution in CAESAR II

lets look at the model in the opening slide

Second try liftoff will be modeled with no restraint (or friction) at 40

-Y deflection indicates that the Y restraint should be included

Back to square one! No convergence here.

Intergraph 2015

8
CAESARIICAUx2015 9/28/2015

TOOLS TO RESOLVE THE ISSUE IN CAESAR II

Its your model

Are your supports properly located? (Do you want liftoff?)

If you are using friction, is your normal load dependable?
Do you always have deadweight in your load case?
Do you want friction on guides?
Is there full friction on gapped restraints?
Do you need those construction gaps on guides?

Intergraph 2015

9
CAESARIICAUx2015 9/28/2015

Interaction of nonlinear conditions may cause

looping

Guide with a gap on a vertical run over a

horizontal with a +Y and friction:
Thermal strain on horizontal run
overcomes friction (slides in X)
Closing gap puts a bending moment on
horizontal which increases normal load on
+Y
Increased friction reduces slide and gap
does not close
Moment is eliminated and +Y normal load
drops; original strain returns
and the cycle continues

Intergraph 2015

A few approaches to resolving nonconvergence

Its usually a tipping point a small movement or a small load change may
activate/deactivate the support model
Take what you get if load case 8 does not converge, you can reset load cases to generate
output for the first 7 cases

Intergraph 2015

10
CAESARIICAUx2015 9/28/2015

A few approaches to resolving nonconvergence

Focus on the load case that is giving the trouble
Identify a non-converging, nonlinear condition and make it linear
Simple active/inactive supports are the best candidates (e.g., +Y)
Choose one and make one linear
If that converges, now you can check the error introduced by that change
(remember, it is usually small)
If that does not converge linearize another nonlinear condition
Document these model adjustments

Intergraph 2015

A few approaches to resolving nonconvergence

When friction is a cause:
Remove or reduce friction for that load case (using the Load Case Editor) by changing the Friction
Multiplier (Friction Load = Friction Multiplier * Mu * Normal Load)
You could also remove friction from a single restraint but that will not provide the same strain
distribution.
Change friction tolerance either in the program Configuration or on the fly in the analysis
screen

Intergraph 2015

11
CAESARIICAUx2015 9/28/2015

Convergence tolerance in the Configuration file

Friction angle variation

If the intended movement is within
15 degrees of the existing friction
load vector, the condition is
converged
Friction normal force variation
If the normal load does not change
more than 15% between the
current and previous iteration, the
condition is converged

Intergraph 2015

Convergence tolerance in the Configuration file

Friction stiffness
Sets the stiffness used for
restraints modeling sticking friction
Not the same as rigid restraint
stiffness
Rod (e.g., +YROD) controls
Rod increment sets the maximum
change in rod angle between
iterations
If the rod angle changes less than
rod tolerance between iterations,
the condition is converged

Intergraph 2015

12
CAESARIICAUx2015 9/28/2015

The non-convergence window

Solution Statistics

This window appears for each CAESAR II static analysis

but flashes so quickly, you would not notice.
Non-convergence, however, has this screen remain.

Equations and Bandwidth characterize the size of the

stiffness matrix (number of nodes and closed loops)
Keep an eye on the Current Case as that is your static
load case that is causing the difficulties here
Iteration shows how many times the program has tried to
resolve all the nonlinear boundary conditions for this load
case using a set of linear assumptions
Here, after 1270 attempts and no acceptable solution, this
model will NOT converge

Intergraph 2015

The non-convergence window

Interactive Control

Controls are available for the convergence process

Press F2 or click on the F2 button to Single Step Thru

Restraints. This action will have the program display
the restraints that remain incorrect and pause after each
iteration. Use this to learn more about your model.
Click on Continue at the bottom of the window to move to
the next iteration
Clicking F2 again will toggle off the single step pause

Intergraph 2015

13
CAESARIICAUx2015 9/28/2015

The non-convergence window

Interactive Control

Controls are available for the convergence process

You can Change the current Friction Tolerances by

pressing F3.
Adjust acceptable Normal Load Variation (default = 15%)
and Friction Angle (Default = 15 degrees).
This may be useful in ending the current iteration
I discourage using this for a final analysis. Results may
change based on what iteration you instigate the change.
(Change the program configuration instead.)
Print Restraint Status (F4) may consume a lot of paper.
Refer to the window in the lower right instead.

Intergraph 2015

The non-convergence window

Status

The status area displays the density of the nonlinear

conditions in this model
No. of Nonlinear Restraints shows the total number of
nonlinear conditions in the model
Non-converged last iteration shows the conditions that
remain unresolved
Here there is a single condition
Oftentimes you will see this cycle (e.g., 4, then 3, then 5,
and repeat) where several conditions interact.

Intergraph 2015

14
CAESARIICAUx2015 9/28/2015

The non-convergence window

NonLinear Restraint Changes

This window displays all the unresolved nonlinear

conditions
There is much to learn from this window

Node number of the uncertain nonlinear condition is listed

Followed by the nonlinear direction (X,Y,Z)
[here: DIR: 0.000 1.000 0.000 indicates a Y restraint]

Intergraph 2015

The non-convergence window

NonLinear Restraint Changes

There is much to learn from this window

After the line identifying the restraint, two lines display the
model change that requires another iteration:
OLD STATE / NEW STATE
OPEN / CLOSED OPEN indicates the linear restraint is
excluded from the iteration, CLOSED indicates that the
linear restraint is included in the iteration
POS / NEG shows the direction of the current restraint
action
SLIDING / NOT SLIDING determines the influence of friction
SLIDING can be viewed as applying a friction force opposite the
direction of sliding
NOT SLIDING adds two restraints orthogonal to the calculated
normal load
SLIDING may be followed by ERR=, this indicates that the
change in friction between iterations is excessive (15% on load,
by default)

Intergraph 2015

15
CAESARIICAUx2015 9/28/2015

The non-convergence window

Stop

Use Continue to move on to the next iteration

Identify a pattern (e.g., Nonconverged = 5,6,3,5,6,3,5,6,3,)
Identify the iteration with the smallest un-converged set (in
this example: 3) that has a simple OPEN/CLOSE change
Click Cancel to abort the analysis
No results will be displayed
(That is why we are talking about this you want results.)
Path forward:
You can remove the non-converging Load Case from
analysis and review other results
Also, by reducing some nonlinearity in the input processor,
this load case may produce results for review
Note: Clicking F2 again will toggle off the single step.

Intergraph 2015

My approach

Get results to evaluate the trouble

Focus only on the Load Case that is presently not converging

Find a simple Open/Close situation and make it linear NOTE THE CHANGE
Rerun
Remove more of these simple Open/Close situations if necessary to get convergence
Check the error in the results for this load case
If still no solution and only friction remains, select a friction support and either 1) lock down
the node or 2) remove mu at that node (again, simplify, make it more linear all you want is
a look into the results to decide)
Note that if you have to remove several nonlinearities, you may be able to reinsert some of
them when you finally pick the right one

So, linearize and review the impact

Intergraph 2015

16
CAESARIICAUx2015 9/28/2015

LETS LOOK AT SOME EXAMPLES

SmallDTwith+Y

40

Intergraph 2015

17
CAESARIICAUx2015 9/28/2015

Multibranch

57
50

Intergraph 2015

Load Case Blowout

830

920

Intergraph 2015

18
CAESARIICAUx2015 9/28/2015

Dummy Leg

741

Intergraph 2015

LP Stream

String
of +Ys

String
of
gapped
guides

Intergraph 2015

19
CAESARIICAUx2015 9/28/2015

CHANGING FRICTION PARAMETERS

Friction parameters under your control

Coefficient of friction (model input)

(Default) input value
Global adjustment as a Load Case Option
Friction stiffness (Configuration file)
Sets the stiffness for restraints CAESAR II adds to the model to make the pipe stick (rather than
slip).
Current default value is much less than rigid
Many years ago, we used 50,000 lbf/in (8.76E4 N/cm)
To speed up convergence (on much slower PCs)
These sliding supports (that were supposedly sticking) interacted with other changing nonlinear conditions
to cause odd and unacceptable results
Convergence tolerance (Configuration file and analysis interaction)
Change in normal load magnitude
Change in sliding direction

Intergraph 2015

20
CAESARIICAUx2015 9/28/2015

SmallDTwith+Y

This is not a friction issue that can be addressed by:

Changing mu
Y load is too sensitive to any horizontal load
Operating mu = 0 in Load Case Options
Tolerance
Friction is never tested against a tolerance its a liftoff issue
Friction stiffness
Works here but with little effect
Using default of 1.75E6 N/cm causes liftoff
Using 8.76E4 N/cm is very similar to no friction

1.75E6 N/cm=1E6 lbf/in

8.76E4 N/cm=50,000 lbf/in

Intergraph 2015

SmallDTwith+Y

No friction in Operating Case:

Friction eliminated from Load Case 1

Friction included in Load Case 2

Friction stiffness = 8.76E4 N/cm:

Friction included in both Load Cases

Sticking friction changed from
1.75E6 to 8.76E4 N/cm

Intergraph 2015

21
CAESARIICAUx2015 9/28/2015

Multibranch

Here, friction error cannot be resolved

CAESAR II requires that the vector representing friction does not vary more than 15% in
magnitude and does not change orientation by more than 15 degrees between iterations (default)
In the earlier look at this model, the +Y restraint at 50 was made linear. This time we will
focus on the friction tolerance

Intergraph 2015

Multibranch

What tolerance is required to accept the friction vector applied at the guide at node 57?
Note the minimum sliding error at 57 is 0.408
Press (or click on) F3 to adjust the tolerance during the analysis

Intergraph 2015

22
CAESARIICAUx2015 9/28/2015

Multibranch

Reset the current tolerance

A new Normal Load Variation of even 0.4081 will accept this
friction error
Sliding Angle Variation plays no role at the +Y & Guide support
(a value of 1 works just like default 15)
But the model does not converge due to the instability of the
+Y restraint at node 50
Returning the tolerance to 0.15 will return to the previous
condition
Note that the program will return to the default tolerance
settings for the next load case

Intergraph 2015

Multibranch

But consider what we have done here

We are adjusting tolerance on Load Case 33
The previous load cases used default tolerance settings
Additional load cases may have a different tolerance as well
Interactive changes to friction tolerance may produce results but
Tolerances may change between load cases and
These results may not be easily reproduced in the next analysis

Intergraph 2015

23
CAESARIICAUx2015 9/28/2015

Remaining models introduced earlier

Models
Load Case Blowout
Dummy Leg
LP Stream

Adjusting friction tolerance will overlook friction errors

So, opening tolerance on friction will eliminate convergence issues with friction
But these models still do not converge because of the underlying active/inactive restraint
issues
Note, too, that when these other unstable (active/inactive) restraints are addressed by
model simplification, the friction troubles disappear.

Intergraph 2015

CHANGING FRICTION TOLERANCE:

PRESET VERSUS INTERACTIVE

24
CAESARIICAUx2015 9/28/2015

Convergence tolerance in the Configuration file

Friction angle variation

If the intended movement is within 15
degrees of the existing friction load
vector, the condition is converged
Friction normal force variation
If the normal load does not change
more than 15% between the current
and previous iteration, the condition is
converged

Intergraph 2015

The non-convergence window

Interactive Control

You can Change the current Friction Tolerances by

pressing F3.
Adjust acceptable Normal Load Variation (default = 15%)
and Friction Angle (Default = 15 degrees).
This may be useful in ending the current iteration
I discourage using this for a final analysis. Results may
change based on what iteration you instigate the change.
(Change the program configuration instead.)

Intergraph 2015

25
CAESARIICAUx2015 9/28/2015

One more model with friction - Manifold

Node 40:
+Y &
Node 240: Limit Stop
+Y &
Limit Stop

Intergraph 2015

Four operating states:

All hot and then each pump spared

Node 40:
+Y
Node 240:
+Y &
Limit Stop

L3 L4 L5 L6

Intergraph 2015

26
CAESARIICAUx2015 9/28/2015

Comparing results from two approaches:

Interactive & Preset

Node 40:
+Y
Node 240:
+Y &
Limit Stop

Excel

Intergraph 2015

Changing friction tolerances

Changing tolerances during analysis:

May be helpful in getting results to review
Results are sensitive to when you make the change (which iteration?)
Tolerances return to default (configuration) values with the start of each new load case
Changing default tolerances:
Provide a consistent solution (see second bullet above)
Changing friction tolerances does not resolve those Active/Inactive support issues described
earlier
Individually resolving non-converging supports (Active/Inactive supports) should be settled
before addressing friction tolerances

Intergraph 2015

27
CAESARIICAUx2015 9/28/2015

IN CONCLUSION

When CAESAR II does not converge

Dont panic
Make sure your load cases are sensible for nonlinear evaluation
Include deadweight
Do not evaluate occasional loads alone
Reduce the nonlinearity to get results
Evaluate the (possible) error introduced by the added linearity

Intergraph 2015

28
CAESARIICAUx2015 9/28/2015

Conclusions

Manipulate the model to produce output

Evaluate the impact of that change
In many cases, there is no adverse effect on the accuracy of the model
Be transparent in your change
Single changes are easily controlled
Suggested order
Single direction to double acting
Gapped restraints to no gap or totally free (closed gaps require a displacement set)
Friction
At a single restraint
For a Load Case
Tolerance as a Configuration setting
Tolerance in the analysis processor
Document your model adjustment

Intergraph 2015

FAILURE TO LAUNCH
Questions / Comments?

29
CAESARIICAUx2015 9/28/2015

FAILURE TO LAUNCH
Thank You

30