Professional Documents
Culture Documents
Modeller User Manual - 6
Modeller User Manual - 6
Chapter 6 Running
An Analysis
LUSAS Analysis Types
LUSAS may be used to numerically model a wide range of engineering problems.
The following sections briefly explain the analysis types, and how LUSAS processes
them.
q Static analysis is, perhaps, the most common analysis carried out by the
engineer, and assumes that the loads are applied instantaneously, hence any
transient effects are ignored. For linear static analysis it is assumed that the
loaded body instantaneously develops a state of internal stress so as to
equilibrate the total applied loads. In a nonlinear static analysis joint elements
and/or ‘slidelines’ may be used to model contact between any two bodies.
q Linear analysis assumes that the overall structural response is linear,
and implies linearity of both the geometric and material response.
213
Chapter 6 Running An Analysis
The following analysis types are also possible but the tabulation of the control
section is not fully supported by LUSAS Modeller:
214
Preparing the Model for Analysis
7. Create the LUSAS datafile. Either click on the Solve button , or use the File,
LUSAS Datafile menu command.
215
Chapter 6 Running An Analysis
216
Frontal Optimisation and LUSAS Solvers
Notes
q For linear problems with multiple load cases, or eigenvalue analyses where
multiple modes are extracted, the iterative technique is inefficient as a
separate iterative process is required for each load case (or mode). The frontal
solution method carries out only one elimination process and its efficiency is
far superior for this type of analysis.
q Problems involving constraint equations cannot be solved using iterative
techniques.
q Conjugate gradient techniques are inappropriate for reduction of matrices and
therefore cannot be used for Guyan reduction or superelement analyses.
It is difficult to recommend a particular solver since which is the most efficient is
highly dependent on the problem being solved. However, in general terms, the
frontal solver will use less memory, and the conjugate gradient solver less CPU time
and disk space. In addition, the frontal solver is less susceptible to problems caused
by ill-conditioned solutions.
217
Chapter 6 Running An Analysis
Static
Dynamic
Creep
Natural Frequency
Eigenvalue Buckling
Spectral Response
Harmonic Response
Fourier Analysis
Field or Thermal
218
About Nonlinear Analysis
q Total Lagrangian
q Updated Lagrangian
q Eulerian
q Co-rotational
All four formulations are valid for arbitrary large deformations. In general, if
rotational degrees of freedom are present, rotations must be small for Total
Lagrangian. An exception to this rule is the Total Lagrangian formulation for thick
shell elements where large rotations may be applied. Large rotations are allowed for
Updated Lagrangian (provided that they are small within each load increment) or
Eulerian. The co-rotational formulation is unconditionally valid for large rotations
and results are generally independent of load step size.
219
Chapter 6 Running An Analysis
All formulations are valid for small strains. For some elements the Updated
Lagrangian formulation is valid for moderately large strains. The Eulerian
formulation is also generally valid for moderate strains. In general, the Total
Lagrangian is a more robust formulation, which is usually able to cope with
substantial load increments. The Updated Lagrangian, and particularly Eulerian,
formulations generally require smaller load increments in order to avoid a divergent
solution.
Standard geometrically nonlinear formulations account for the change in position of
the loading, but not the change in direction relative to the deformed configuration.
Loading is always conservative for the Total Lagrangian geometrically nonlinear
formulations (that is, the load is always applied in the same direction as was initially
prescribed). Using an Updated Lagrangian formulation, the geometry is updated at
the end of each increment and the applied loads may maintain the same relative
orientation as to the original surface (depending on element and load types).
Therefore non-conservative loading can be increment size dependent. True non-
conservative loading may only be achieved by using the Eulerian and co-rotational
formulations.
The choice of particular formulations is both problem and element dependent (the
element formulation determining which strain formulations are available). The
availability of each formulation is given for each element in the LUSAS Element
Library. For further details regarding the geometrically nonlinear formulations refer
to the LUSAS Theory Manual.
220
About Nonlinear Analysis
Distance(x)
221
Chapter 6 Running An Analysis
The iterative corrections are referred to some form of convergence criteria which
indicates to what extent an equilibrate state has been achieved. Such a solution
procedure is therefore commonly referred to as an incremental-iterative (or
predictor-corrector) method shown in the figure above. In LUSAS, the nonlinear
solution is based on the Newton-Raphson procedure. The details of the solution
procedure are controlled using the Nonlinear Control properties assigned to load
case.
For the analysis of nonlinear problems, the solution procedure adopted may be of
significance to the results obtained. In order to reduce this dependence, wherever
possible, nonlinear control properties incorporate a series of generally applicable
default settings, and automatically activatedfacilities.
Iterative Procedures
In LUSAS the incremental-iterative solution is based on Newton-Raphson iterations.
In the Newton-Raphson procedure an initial prediction of the incremental solution is
based on the tangent stiffness from which incremental displacements, and their
iterative corrections may be derived.
222
Nonlinear Solution Procedures
Incremental Procedures
For the Newton-Raphson solution procedures it is assumed that a displacement
solution may be found for a given load increment and that, within each load
increment, the load level remains constant. Such methods are therefore often referred
to as constant load level incrementation procedures.
However, where limit points in the structural response are encountered (for example
in the geometrically nonlinear case of snap-through failure) constant load level
methods will, at best, fail to identify the load shedding portion of the curve and, at
worst, fail to converge at all past the limit point. The solution of limit point problems
therefore leads to the development of alternative methods, including displacement
incrementation and constrained solution methods.
223
Chapter 6 Running An Analysis
224
Nonlinear Solution Procedures
q Interpolation
q Riks semi-direct approach
Two further options for the bracketing procedure exist depending on whether the
material response is elastic (reversible) or plastic/path dependent (irreversible). Only
the first critical point can be processed and a subsequent eigenvalue analysis must be
invoked to determine whether the critical point encountered is a limit or bifurcation
point. A limit point may be defined as the point at which load starts to decrease as
displacements increase (e.g. the snap through of a shallow arch). In this instance the
structure may not have failed completely and could subsequently still be capable of
carrying more load. A bifurcation point indicates that the solution of the nonlinear
differential equations has encountered an alternative unstable solution path or paths
which may be followed instead of the stable equilibrium path. The branch switching
procedure must then be undertaken if an unstable equilibrium path is to be followed.
The branch switching procedure should only be carried out within a restart analysis
after bracketing has been successfully completed. Two options exist for guiding the
solution onto a secondary path
q eigenmode injection
q artificial force and Rheinboldt’s arc-length
Incremental Loading
Incrementation for nonlinear problems may be specified in four ways:
225
Chapter 6 Running An Analysis
Solution termination
When using manual incrementation, the solution will automatically terminate
following execution of one increment. With automatic incrementation, the solution
226
Nonlinear Solution Procedures
progresses one Nonlinear Control chapter at a time. The finish of each Nonlinear
Control chapter is controlled by its Termination parameters.
Termination may be specified in 3 ways:
q Limiting the maximum applied load factor.
q Limiting the maximum number of applied increments.
q Limiting the maximum value of a named freedom.
Where more than one criteria is specified, termination will occur on the first criteria
to be satisfied.
Failure to converge within the specified maximum number of iterations will either
result in a diagnostic message and termination of the solution or, if automatic
incrementation is being used, a reduction of the applied load increment. If required,
the solution may be continued from an unconverged increment (Option 16, 17),
although the consequences of such an action should be appreciated.
In addition, the solution will be terminated if, at the beginning of an increment,
more than two negative pivots are encountered during the frontal elimination phase.
227
Chapter 6 Running An Analysis
The convergence tolerance for each criteria is specified in the Solution parameters
and advanced solution parameters section of the Nonlinear Control properties. The
selection of a convergence criteria, and the associated tolerance is problem
dependent. However, the following points should be considered:
Clearly, the convergence criteria must not be too slack so as to yield an inaccurate
solution, nor too tight so as to waste computer time performing unnecessary
iterations. In general, sensitive geometrically nonlinear problems require a tight
convergence criteria, whereas, with predominantly materially nonlinear problems,
larger local residuals may be tolerated.
Where more than one criteria has been specified, convergence will be assumed only
on the satisfaction of all specified tolerances.
The following considerations apply to individual convergence parameters:
q Absolute Residual Norm is of limited use owing to its dependence upon the
units being used. It is a strict criteria and for some problems, especially those
involving plasticity, it may be very difficult to reduce locally large residuals.
However, in sensitive geometrically nonlinear problems near bifurcation
points, it can sometimes be necessary to ensure that large residuals are
completely eliminated.
q Root Mean Square Residual Norm is the square root of the average of the
squares of the residual forces and is generally more applicable than the above,
but is still dependent upon the units being used.
q Displacement Norm is the sum of the squares of all the iterative
displacements as a percentage of the sum of the squares of the total
displacements and is a useful measure of how much the structure has moved
during an iteration. Being a scaled norm it is not affected by units but
convergence is not guaranteed. Typical values of slack and tight norms are
(5.0 - 1.0) and (0.1 - 0.001) respectively.
q Residual Force Norm is the sum of the squares of all the residual forces as a
percentage of the sum of the squares of all the external forces. This is the
most versatile of the five criteria. Typical slack and tight values are (10.0 -
5.0) and (0.1 - 0.00001) respectively.
q External Work Norm is the work done by all the residuals acting through
the iterative displacements, as a percentage of the work done by the loads on
iteration zero of the increment. Since all freedoms are considered it is very
versatile (the default displacement and force norms consider only the
translational freedoms). However, it should be noted that a minimum detected
potential energy need not necessarily coincide with the equilibrate state.
Typical values of slack and tight norms are (0.1 - 0.001) and (10-6 - 10-9)
respectively.
228
Eigenvalue Analysis
Eigenvalue Analysis
What is an Eigenvalue Analysis?
An Eigenvalue extraction analysis is the extraction of the natural modes of
vibration of a structure, or a natural frequency analysis. It can also be used to solve
the following problems:
229
Chapter 6 Running An Analysis
Subspace Iteration
The first step in the subspace iteration procedure is to establish the number of
starting iteration vectors. This should be greater than the number of required
eigenvalues to increase the rate of convergence. It is important to remember that the
number of starting iteration vectors cannot exceed the number of degrees of freedom
of the system. Experience suggests that the number of starting vectors should be
determined from the expression:
nivc = min (( 2*nroot ), ( nroot+8 ), ( nvbz ))
where
q nivc is the number of starting iteration vectors,
q nroot is the required number of eigenvalues and
q nvbz the number of degrees of freedom in the structure.
Occasionally insufficient eigensolutions are computed in the initial eigenvalue
analysis. The number of eigensolutions can be increased by using a RESTART and
re-specifying the Eigenvalue Control. This second eigenvalue analysis utilises the
230
Eigenvalue Analysis
231
Chapter 6 Running An Analysis
232
Eigenvalue Analysis
i≠j
for all eigenvectors Fi and global mass matrix M.
The inverse iteration procedure should always converge quickly since the shift point
is automatically updated to improve convergence if this is not the case. The
maximum number of iterations is therefore set, by default, at 30 since this should
never be required by the algorithm. It is provided merely as a safety measure and the
default value may be changed in the CONVERGENCE data section.
233
Chapter 6 Running An Analysis
Lanczos
The Lanczos eigenvalue analysis facility is based on the inverse iteration procedure
but is applied to only one starting vector. It is significantly faster to converge and can
use significantly less physical memory and hard disk than subspace methods. For
these reasons it is ideal for large numbers of requested eigenvalues and for large
problems.
Note. The load correction terms, due to Coriolis and angular acceleration, are
ignored because they result in non-symmetric damping and stiffness matrices
respectively.
Notes
q The relationship between the eigenvalue, λ, and the angular frequency, ω, is:
234
Eigenvalue Buckling Analysis
235
Chapter 6 Running An Analysis
Although a linear static analysis will ensure that equilibrium is fully achieved and
may also predict stress levels within an acceptable range, the current structural
design may still be unsuitable for the intended use. The additional consideration is
related to the type of equilibrium achieved.
Below the critical buckling load of a structure "stable" equilibrium will usually be
achieved, whilst above this load "unstable" equilibrium may result from geometric
and/or material effects.
This note concentrates on the methods available in LUSAS to determine the
buckling behaviour of a structure as a result of geometric effects (commonly denoted
as "Euler strut buckling").
There are two methods in LUSAS to obtained information regarding buckling loads
and their respective deformation modes,
q Linear Eigenvalue Buckling Analysis
q Full Geometrically Nonlinear Analysis
The following comments describe the principal characteristics of these methods to
assist in the selection of the most appropriate for the structure to be modelled.
236
Eigenvalue Buckling Analysis
237
Chapter 6 Running An Analysis
238
Eigenvalue Buckling Analysis
Linear elements may be used for the remainder. If there is uncertainty, all
elements used should support geometric nonlinearity.
This procedure depends on geometrically nonlinear functionality to determine
buckling effects and cannot be omitted. The LUSAS element library should be
inspected to determine which is available for the proposed element(s).
To establish whether a proposed element supports nonlinearity, the Geometric
Nonlinearity section in the LUSAS element library should be referred to. Geometric
nonlinearity is, as stated above, not used for linear buckling analyses but an
affirmative indication in this section is tantamount to saying that nonlinear
functionality is available. Nonlinear capability may, of course, also be determined
from the availability of nonlinear material models.
If this reference states 'Not Applicable' the element does not support any geometric
nonlinearity. Any other response indicates that nonlinearity is permissible and
defines the options available.
In summary, however; the following elements support geometric nonlinearity and
may be used within one or other of the buckling techniques described
q All BAR elements
q All CONTINUUM elements (except FOURIER and EXPLICIT DYNAMICS)
q All two dimensional MEMBRANE elements
Additionally, the BEAM elements supported are as follows
q 2 Dimensional BM3, BMX3
q 3 Dimensional BS3, BS4, BSL3, BSL4, BTS3, BSX4, BXL4
And the SHELL elements supported are
q 2 Dimensional BXS3
q 3 Dimensional TSL6, QSL8TTS3, QTS4, TTS6, QTS8
Note that PLATE elements do not support geometrically nonlinear applications.
239
Chapter 6 Running An Analysis
Final Notes
There may be genuine numerical difficulties in a linear eigenvalue buckling analysis
which will be rectified by using the alternative eigenvalue buckling solution. Using
this alternative algorithm will always give positive eigenvalues except when the
buckling load factor is less than unity. Adjust the load level to ensure that all the
load factors are greater than 1 if this occurs. In other words the load applied should
be below the lowest expected buckling mode of the structure.
240
Spectral Response Analysis
Starting Procedure
Before specifying the spectral response data the eigenvalues and eigenvectors of the
system are computed using an eigenvalue extraction analysis (note that the computed
eigenvectors must have been normalised to the global mass).
241
Chapter 6 Running An Analysis
Starting Procedure
Since the harmonic response analysis requires a knowledge of the natural vibration
characteristics of the structure (i.e. the natural frequencies) it is necessary to perform
a previous eigenvalue analysis to obtain the natural eigensolutions which are
considered to be of significance. Note that the computed eigenvectors (or
eigenmodes) must be normalised with respect to the mass. This eigenvalue analysis
may be performed as part of the same analysis or may be recovered from a previously
created restart file.
242
Dynamic (Transient) Analysis
override the system defaults by specifying the overriding damping defaults in the
CONSTANTS section.
The loading definition is made up of two parts: firstly numerous loads may be
defined using the LOAD CASE commands as usual. Then, within the
HARMONIC_RESPONSE CONTROL data chapter these loads are defined as the
real or the imaginary components by using one or more HARMONIC LOADING
commands. It must be noted that all loads are considered to have the same frequency
of vibration as given in the FREQUENCIES section but the relative phases may be
altered by correctly defining their complex components.
243
Chapter 6 Running An Analysis
Starting procedure
To start a dynamic analysis a knowledge of the initial conditions is required. The
initial conditions for the Hilber-Hughes-Taylor integration scheme are:
where:
where:
• , ,
This means that in the output for any time step, the displacements will relate to the
current response time while the accelerations effectively lag one time step behind the
displacements
244
Impact Dynamics
Impact Dynamics
In addition to using nonlinear joint models to represent contact and impact, a
specialised procedure is available for modelling impact in dynamic analysis. This
procedure uses a technique, and permits the surfaces of 2D, axisymmetric, and 3D
structures to register and react to contact with one another.
Creep Analysis
Where the relationship between stress and strain is time dependent, a creep analysis
will be required. The creep response is usually a function of the material properties,
stress, strain and temperature history. Unlike time independent plasticity where a
limited set of yield criteria may be applied to many materials, the creep response
differs greatly for different materials.
A creep analysis may be carried out using a linear or nonlinear material model
within a nonlinear, dynamic or thermo-mechanically coupled analysis.
Creep material properties are defined from the Attributes menu > Material >
Isotropic/Orthotropic.
Fourier Analysis
Fourier elements offer an efficient method to solve problems in which axisymmetric
structures are subjected to non axisymmetric loading, provided that the
displacements are small and linear theory applies. The circumferential displacements
and variations of load are expressed as the sum of the components of a Fourier
series, whilst the axial and radial variations are described by the standard finite
element formulation. Each term of the Fourier series is analysed individually and the
results are then combined to provide the overall solution.
Fourier elements can be used to model both solid and thin walled structures; in
particular they offer an ideal method to obtain an initial estimate of the eigenvalues
245
Chapter 6 Running An Analysis
of thin walled structures without the expense of performing a full shell analysis on
the complete structure. The choice between a full structural discretisation using solid
or shell elements and the use of the Fourier element depends upon the number of
Fourier terms that are required to accurately describe the load; if only a few terms
are required then the Fourier element should be considered.
A Fourier analysis can be considered as a generalisation of the standard
axisymmetric analysis. The finite element mesh is defined in the XY-plane and may
be axisymmetric about either the X or the Y axis. Loading is applied to the mesh in
the standard manner using the load case properties, with its circumferential variation
defined using the curve definition. Finally the Fourier components to be computed
are input using the Fourier control as part of the load case properties.
Supports are defined in the usual manner , with the declaration free, restrained or
spring supports. For the n=0 harmonic the spring stiffness per unit radian must
include a factor of 2p for the implicit integration around the surface. For harmonics
other than n=0 the factor should be p. Certain restrictions are applied to the
freedoms of nodes lying on the axis of symmetry. These conditions, in the table
shown, are automatically imposed on the centreline nodes.
Inertial Loading
The operation of the inertial loading, input using the CBF load type (body force), is
slightly different to the other standard loads. Inertial loads are calculated from
246
Fourier Analysis
247
Chapter 6 Running An Analysis
Thermal Problems
If temperature dependent material properties are used the temperature field must be
axisymmetric. For non-temperature dependent materials, a general temperature field
can be input in the same manner as the other element loads. Temperature loads
cannot be used in dynamic or harmonic response analyses. Data from an
axisymmetric thermal analysis can be written to a data transfer file using the
COUPLE WRITE command. This data can then be imported into the Fourier
analysis using the COUPLE READ command; other thermal load data should not be
input.
Post-Processing
Fourier results may be expanded using the Graph Wizard.
Thermal Analysis
Where a solution is required to the quasi-
Application Field Variable
harmonic equation a field analysis may
be performed. The quasi-harmonic Thermal conduction Temperature
equation defines the behaviour of a Seepage flow Hydraulic head
variety of field problems. Some of the Incompressible flow Stream function
more common quasi-harmonic Soap film Deflection
applications, and the associated field
Elastic torsion Warping function
variable, are listed in the table shown
right. Elastic torsion Stress function
Electric conduction Electric potential
Two types of field analysis may be Electrostatics Electric potential
performed using LUSAS:
Magnetostatics Magnetic potential
q Steady State field analysis
q Transient field analysis
Facilities for thermo-mechanically coupled analysis are also available.
The solution of this class of problem follows an identical process to that of the
structural problem. The domain is discretised using a series of field elements,
thermal material properties are specified, thermal loads are applied, and the
equations solved for the values of the field variable at each nodal point. Thermal link
elements or the specification of thermal surfaces determine how heat is conducted,
248
Thermal Analysis
convected or radiated across gaps and spaces between different domains. Since the
most common application is that of thermal conductivity, subsequent discussion will
be directed towards this type of analysis.
In field analysis, the standard loading facilities can still be applied, but with slightly
different interpretation. For example, temperatures may be directly prescribed as
boundary conditions; the standard face loading corresponds to the application of a
heat flux to the boundary; concentrated loads are equivalent to a localised inflow or
outflow of heat located at the point of application of the concentrated load; body
force potentials correspond to internal heat generation per unit volume; prescribed
displacements allow temperatures to be specified at nodes.
Crank-Nicholson 1/2
Euler 0
Galerkin 2/3 (default)
249
Chapter 6 Running An Analysis
Backward difference 1
The transient problem is integrated through time using a 2-point integration scheme.
The type of integration scheme is controlled in the CONSTANTS section of the
TRANSIENT CONTROL data chapter (by specifying the parameter beta). Some of
the more common 2-point integration schemes and their associated beta values are
shown in the table right.
Note. In the limit the final solution should be the same as the steady state analysis
subject to the new loading and boundary conditions; the transient analysis merely
models the thermal inertia in moving from the initial to the final conditions. The
body property which is used to describe this inertia is the effective heat capacity.
When choosing an increment of time, the stability of the incrementation scheme
must be examined. When beta is greater than or equal to 0.5 the solution is
unconditionally stable (the Crank-Nicholson, Galerkin and Backward difference
schemes are of this form).
When beta is between the limits 0 and 0.5 the solution is stable provided that:
where β is the input parameter beta and λmax is the maximum eigenvalue of the
system.
The time step used for implicit algorithms is dependent upon the number of modes
that influence the response of the system. Generally, the major part of the response is
governed by the lower modes so that:
250
Thermo-Mechanical Coupled Analysis
where C is the effective heat capacity including the effects of the latent heat of
evolution due to phase changes and ϕ is the temperature. In the material data input
both H and C may be specified. For analyses where phase changes are not
represented, the effective heat capacity value C is used in the calculations.
For analyses where phase changes are represented, tabular input should be used to
define the variation of H with temperature, together with an initial value of C.
Providing a variation in temperature exists at a point, the effective specific heat is
then interpolated from the enthalpy values. If no variation exists, for example, in an
area of the problem that has experienced no change in temperature from the initial
temperature, then the initial value of C is used.
For nonlinear analysis the nonlinear control parameters are used to define the
iterative strategy. The convergence section is utilised to provide tolerances for
defining steady state, and either a field norm (temperature equivalent to the
displacement norm), or a residual flow norm (equivalent to the residual force norm)
may be used.
q Semi-Coupled analysis
q Fully Coupled analysis
251
Chapter 6 Running An Analysis
In a semi coupled analysis, for instance, the structural response is influenced by the
temperature field, but the thermal response is independent of the structural response,
or vice-versa. In such a case, the thermal analysis is performed prior to the structural
analysis, and either a single or series of nodal temperature tables are created. These
are read during the structural analysis at the required load case or time step.
In a fully coupled analysis, the thermal and structural analyses must be performed
simultaneously with a continuous transfer of information between the two analyses.
For instance, in addition to modelling the influence of the thermal field on the
structural response, the effect of the structural response on the thermal field is
represented. Temperatures are transferred from the thermal to the structural analysis,
and the updated geometry is transferred from the structural to the thermal analysis.
The analyses may be coupled on the incremental or iterative levels (iterative
coupling is machine dependent).
For true full coupling of two nonlinear fields, information transfer has to occur on an
iteration level within each increment, so that in addition to preserving equilibrium of
the local thermal and structural fields, equilibrium of the combined system is
maintained. Iterative coupling is essential for strongly coupled systems, e.g. structure
to structure contact. For weaker thermo-mechanical coupling, information transfer at
an increment level should provide an adequate solution.
252
Thermo-Mechanical Coupled Analysis
Implementation of coupling
To transfer data between the thermal and structural analyses an external data
transfer file, common to both analyses, is set up and data is transferred as specified
in the COUPLE commands. Only one of the two analyses must open the data transfer
file; for a semi coupled analysis, this must be performed by the first analysis to be
run.
As each analysis proceeds, data may be both written to and read from datasets stored
in the external data transfer file. Each dataset is given an unique referencing
number. In addition to this reference dataset number, if the analysis writing the data
is either transient or dynamic, the current solution time of the analysis is appended
to the dataset. When data is read by either a dynamic or transient analysis, the times
of the datasets are checked to match the current solution time of the analysis. If it is
not possible to exactly match the time of the solution with any one of the datasets,
the data is linearly interpolated from the two datasets which bracket the solution
time. If the datasets do not have time labels, they are read sequentially starting from
the dataset number specified with the COUPLE READ command.
The COUPLE WRITE and COUPLE READ commands initialise the writing and
reading of data to and from the data transfer file. Data may be transferred by
specifying either the increment/step number or the solution time. If the specified
time for the data read or write cannot be satisfied exactly, the data is transferred at a
time as close to the specified time as possible.
Once one of the initial transfer conditions, initialised by the COUPLE READ or
COUPLE WRITE commands, has been fulfilled, further corresponding data transfers
may be initiated by using the INCREMENTAL COUPLE READ or
INCREMENTAL COUPLE WRITE commands.
The INCREMENTAL COUPLE READ and INCREMENTAL COUPLE WRITE
commands operate in a similar fashion to the standard LUSAS data output controls
specified by the OUTPUT command. Data may be input/output on every (n)th step or
at intervals of time. For instance, the specification of nfrrd=5 with the
INCREMENTAL COUPLE READ command will cause data to be read on steps 5
10 15 ... etc. Note, these commands only become active once the corresponding
conditions specified on the COUPLE READ and COUPLE WRITE commands have
been satisfied. Further re-specification of these commands will suspend the operation
of the INCREMENTAL COUPLE READ and INCREMENTAL COUPLE WRITE
commands until the new conditions are fulfilled.
The use of time to label the datasets allows the interpolation between datasets. To
extend this facility to incremental steady state analysis, the concept of pseudo time is
introduced. Using the INCREMENTAL COUPLE READ and INCREMENTAL
COUPLE WRITE commands, a particular time may be associated with each steady
253
Chapter 6 Running An Analysis
state solution for both writing to and reading from the data transfer file. The pseudo
time label is calculated as:
( increment number-1 ) * dTime
254
Thermo-Mechanical Coupled Analysis
Simultaneous processing
To run a fully coupled analyses, the command PARALLEL must be inserted into the
coupling command of each analysis. The two jobs are then run. The starting order is
unimportant. The inclusion of the command word PARALLEL in the COUPLE or
COUPLE OPEN command prevents an error occurring if data is not found. Instead
the program will wait until the data is updated by the other analysis.
Iterative coupling
Iterative coupling of the analyses is initiated by setting both the niterr and
niterw parameters in the INCREMENTAL COUPLE READ/WRITE commands.
The analyses continue to iterate until both have converged. The number of iterations
between couplings can be adjusted to offset differences in the rate of convergence of
the two analyses, but this may ultimately lead to excess iterations being performed;
usually the values of both parameters should be set to 1.
The analyses are considered to have converged once one analysis receives
information that the second analysis has converged, and in the next iteration
converges itself. This analysis then proceeds to the next increment. The second
analysis, which converged first, now performs a further iteration. If it converges, it
proceeds to the next increment. However, if the new information causes failure in
convergence it will try to continue to iterate. This will cause termination of the
analysis, since the other analysis has proceeded to the next increment. To overcome
this problem, Option 183 may be set to make both analyses perform an additional
iteration. This is computationally expensive, and in most instances will not be
needed.
Iterative coupling takes place at data transfer points. If both analyses involve time,
both the step size and solution time must coincide for a legal iterative solution.
255
Chapter 6 Running An Analysis
Pre-Analysis Checks
All Analysis Types
1. Check the consistency of your co-ordinate systems between the finite element
model and any engineering drawing that you have worked from.
2. Check key drawing dimensions against co-ordinates of respective points in the
model.
3. Check the mesh for cracks and voids.
4. Check for correct material properties and assignments.
5. Check for consistent units. e.g. comply with one system of units - N, m, kg, sec is
a consistent set of units but N, mm, kg, s is not! Be careful when factoring forces
and densities to comply with a non-standard set of units.
6. Check for correct orientation of beam properties.
7. Check for correct boundary conditions (loads/supports).
8. Check element thickness against original model data (plates/shells).
9. Check reversed normals for plates/shells/2-D.
10. Check element shapes for aspect ratio, skew, warp, taper, curvature and centrality
of mid-side nodes. Warning messages will be present in the output file for all of
the above.
11. Check for duplicate nodes and elements. Warning messages are given in the
output file.
12. From the LUSAS datafile dialog, (when creating the LUSAS datafile, File >
LUSAS datafile), click on the Output button to check that the LUSAS output
256
Post-Analysis Checks
Large Runs
1. Estimate disk and CPU resource requirements.
2. Keep a log of analysis runs for future reference. Note information such as
element types, numbers of load cases, frontwidth, file sizes, run-times, etc.
Post-Analysis Checks
All Analysis Types
1. Check reactions for equilibrium.
2. Check magnitudes of displacements and stresses. Compare to hand calculation.
3. Check for matrix conditioning messages (Pivot & diagonal decay). Small pivot
and diagonal decay warning messages are invoked when the stiffness matrix is
poorly conditioned. Diagonal decay means that round-off error during the
257
Chapter 6 Running An Analysis
solution has become significant which could lead to inaccurate results. A poorly
conditioned stiffness matrix is the result of a large variation in magnitude of the
diagonal terms. This could be caused by large stiff elements being connected to
small less stiff elements or elements with highly disparate values of stiffness (e.g.
a beam may have a bending stiffness that is orders of magnitude less than it's
axial stiffness).
A negative pivot in a non-linear analysis usually means that a limit or bifurcation
point has been encountered. However, negative pivots sometimes occur during
the iterative solution (which sometimes means that the load step is too big) but
disappear when the solution has converged. If negative pivots occur and the
solution will not converge then first try reducing the load step.
If the solution still does not converge a limit or bifurcation point may have been
encountered in which case the solution procedure may need to be changed.
Running the problem under arc length control gives the best chance of
negotiating a limit or bifurcation point. A load limit point can also be overcome
by using prescribed displacement loading.
4. Check the LUSAS output file for other warning or error messages.
5. Check adequate mesh density.
6. Check correct average groups for nodal average stress calculation.
7. Check the model summary information available in the LUSAS output file. This
gives the total length, area, volume and mass for the structure together with the
centre of gravity, moments of inertia and resultant applied load at the origin.
258
Before Calling the Help Desk
FEA implement a ‘named user’ system whereby all queries at a particular site are
channelled through one person. This leads to the named user reaching 'expert’ or
‘power user’ status so that he/she is able to answer some queries directly without
phoning the support desk. This potentially could save both the user and support desk
a great deal of time leading to a more efficient support service. Before calling the
customer support desk the following checks should be carried out by users to validate
their models. It is good practice to systematically carry out these checks as a matter
of course whether or not there appears to be a problem with the solution obtained.
q If you encounter a problem do not suppress the data input printout with
Options 44 and 45.
q Read the data input printout! Is the output file echoing the input data that you
think you've specified? You may have missed a number in a line of input data
- in particular check your Material Properties data echo.
q Read any warning messages.
q Check to see if the current bugs/limitations list contains details of any known
bugs which could be causing your problem.
q Carry out your own QA check on your model (see below)
q Check the model output summary.
The checklists outlined in previous sections form part of the QA procedure adopted
by FEA for carrying out analyses for external clients. Users should create their own
set of QA checks which will be particular to the type of analyses for which they are
responsible. The basic items covered in the procedure used at FEA have been listed
to give a general guide as to the sort of checks that should be included in the QA of a
finite element analysis.
With an ever increasing user base the support desk at FEA is becoming increasingly
busy. With the aim of producing a more efficient service, some general pointers are
given as to what information the user should have at hand when calling the support
desk. This information will help the support engineers get to the bottom of your
problem much more quickly:
q The exact text of any warning or error message(s).
q Any numbers associated with system errors.
q Full trace back information, i.e. routine names.
q A copy of the data file causing problems.
q A list of the last commands typed in LUSAS Modeller or a copy of the session
log.
q The contents of LUSAS Modeller error log LUSAS.ERR
q Try to be logged onto the machine when calling or be close enough to the
machine to type in commands suggested by the support engineer.
259
Chapter 6 Running An Analysis
Superelements
The superelement facility allows a large finite element model to be divided into
smaller, more manageable components or substructures. The superelements are then
added together to form the complete structure. Both linear static and natural
frequency analyses may be performed using superelements. An option also exists to
evaluate the natural frequencies of a large structure from the eigen-solutions of the
individual superelements using modal synthesis.
The superelement facility has the following advantages:
q Cost reduction for re-analysis.
q Better scheduling of computer time by executing several medium size
analyses rather than one large analysis.
q Simpler model generation with individual components which are easier to
check. In the case of user modal data no model generation is required - only
input of the boundary nodes and freedoms.
q Efficient coupling of components which may be analysed by several
organisations.
q Reduced peripheral storage and high speed memory requirements for each
analysis.
A superelement may be defined as an assembly of individual elements together with
a list of exterior freedoms that will represent the superelement once it has been
reduced. This facility is provided in LUSAS under the SUPERELEMENT data
chapter.
260
Superelements
Using Superelements
Analyses involving superelements may be divided into three basic stages:
q Opening databases and generating superelement data.
q Using superelements or user modal data in an analysis (forming the complete
structure).
q Recovering displacements and stresses within each of the superelements.
The superelements may be generated (SUPERELEMENT ASSIGN/DEFINE) in one
or several analyses and stored in several databases. Once generated, the elements
may be used (SUPERELEMENT USE) in either a static or natural frequency
analysis. At this stage user modal data may be included (SUPERELEMENT
MODAL_DATA) in a natural frequency analysis. The recovery procedure
(SUPERELEMENT RECOVER) permits the displacements, stresses and strains of a
superelement to be recovered. Additional concentrated loads and prescribed
displacements are permitted during the USE pass of an analysis.
261
Chapter 6 Running An Analysis
Superelements
Large finite element models may be divided into smaller, more manageable
components which are then added together to form the complete (or residual)
structure. These components are referred to as ‘superelements’. A superelement may
be defined as an assembly of individual elements together with a list of master
freedoms that will represent the superelement once it has been reduced. The
SUPERELEMENT family of commands provides this facility in LUSAS for linear
static and eigenvalue analyses. An option also exists to evaluate the natural
frequencies of a large structure from the eigen solutions of the individual
superelements using modal synthesis.
In addition a component for which the user modal data (frequencies and
eigenvectors) is available can be utilised to obtain the natural frequencies of a large
structure. This component will be referred to as a user modal superelement. A
detailed explanation of these facilities can be found in the LUSAS Theory Manual.
Analyses involving superelements may be divided into three basic stages:
q Opening databases and generating superelement data.
q Using superelements and/or user modal data in an analysis (i.e. solving the
residual structure). This stage could involve user modal data only if
superelements are not used.
q Recovering displacements and stresses within the superelements.
The superelement data may be generated in one or several analyses and stored in
several databases. Once generated, the superelements may be used in either static or
natural frequency analysis. The recovery procedure permits the displacements,
stresses and strains of a superelement to be recovered. Additional concentrated loads
and prescribed displacements are permitted during the combination pass of an
analysis. In the generation phase, both static and natural frequency analyses of an
individual superelement may be performed to validate the data.
The following restrictions apply to the use of superelements:
1. Only linear analyses may be undertaken.
262
Superelements
q If nonlinear elements are used (e.g. nodal joints, radiation) the initial status is
used.
q Plasticity, creep and damage capabilities are ignored.
q Superelements cannot be generated from Fourier elements.
q User modal data may only be used for natural frequency analyses
2. Material properties must be constant. If temperature dependent properties are
input then the values corresponding to the initial temperature will be used.
3. Superelement generation/recovery is limited to one superelement per analysis
run.
4. When a superelement connects to other non-superelements, the connecting node
locations associated with the non-superelements must have the same relative
locations as originally used in generating the superelement. In addition, the
Cartesian sets for the nodes must coincide.
5. If superelements are used without any standard elements the FREEDOM
TEMPLATE data chapter must be included.
6. If superelements are used that have more freedoms than the standard elements,
and any values are to be prescribed for these freedoms, then the FREEDOM
TEMPLATE data chapter must be included. These values may be defined in
SUPPORT CONDITIONS, CONSTRAINT EQUATIONS, RETAINED
FREEDOMS or LOADCASE data chapters.
7. User modal data eigenvectors must be in the structure global axes.
Creating Superelements
A typical data file for creating superelements would contain SUPERELEMENT
ASSIGN and SUPERELEMENT DEFINE commands. The master freedoms for the
superelement would be specified using the RETAINED FREEDOMS data chapter.
This analysis would then create:
q A superelement database
q An output file
q A restart file (optional)
263
Chapter 6 Running An Analysis
The contents of the files created would depend upon the control parameters specified
under the sub-commands of SUPERELEMENT DEFINE. Only one superelement
can be generated in a single analysis.
264
Superelements
nodes and freedoms, frequencies and references to result files containing the
eigenmodes. Additional user modal data components, elements, superelements or
boundary conditions may be specified at this stage. The analysis would then create:
q An output file
q A restart file (optional)
q A LUSAS Modeller plot file (optional)
If a LUSAS Modeller plot file is created it will NOT contain the user modal data
results.
Generation Phase
Analysis 1 Input Output
define1.dat define1.out
define1.mys (optional)
define1.rst (optional)
define1.sfr (rename to recov1.sfr
or delete)
dbname1.sda
265
Chapter 6 Running An Analysis
Recovery Phase
Analysis 4 Input Output
recov1.dat recov1.out
define1.rst recov1.mys (optional)
(optional)
recov1.sfr recov1.rst (optional)
(optional)
dbname1.sda dbname1.srs
The file extensions .sda and .srs relate to superelement data and results databases
respectively. The extension .sfr indicates a file containing the reduced frontal
matrices. If the appropriate .sfr file is accessible during the recovery phase, the
solution is recovered from the back substitution of the equations leading to a more
efficient solution. To take advantage of this facility the .sfr file must have the same
prefix as the data file used in the recovery, i.e. copy define1.sfr to recov1.sfr in above
example. If a superelement restart file (.rst) is not available, recovery may be
performed by redefining the superelement in the datafile (i.e. as in the superelement
creation phase).
Note. The .mys, .rst, .sfr, .sda and .srs files are all binary files. The .dat and the
.out files are ASCII files which can be read and edited directly by the user. The user
modal data files (.def and .adp) are ASCII files.
266
Superelements
def/
dat
rst sfr sda adp sda sfr rst
use1
mys
dat
recov1 recov2 dat
267
Chapter 6 Running An Analysis
268
Superelements
Notes
Superelement Creation
The superelement to be created will contain information on element topology and
usually requires specification of boundary freedoms, loading and reduction options.
The SUPERELEMENT DEFINE command defines the name and title associated
with the superelement.
Superelement Definition
SUPERELEMENT DEFINE defines the superelement name and internal database
that is used to store the superelement information.
SUPERELEMENT DEFINE sname [dbname] [TITLE title]
sname Superelement name.
dbname Internal name of database.
Notes
269
Chapter 6 Running An Analysis
Constants
The CONSTANTS data section defines the control parameters for the superelement
creation phase. A list of valid names defines the output required.
CONSTANTS [LOAD] [STIF] [MASS] [STAT] [EIGN]
LOAD Save the reduced load vector.
STIF Save the reduced stiffness matrix.
MASS Save the reduced mass matrix.
STAT Perform a static analysis using superelement.
EIGN Perform an eigen-analysis using superelement.
Notes
1. If the CONSTANTS command is not specified LOAD and STIF will be invoked
by default. However, if Option 90 is set then LOAD, STIF and MASS will be the
default control parameters.
2. If the command GENERALISED COORDINATES is specified the following
default operations will be carried out:
• The reduced mass matrix will be saved.
• An eigen-analysis will be performed to compute the eigenvectors required.
• The enhanced reduced stiffness and mass will be computed and saved.
Output
The OUTPUT command defines the information to be presented in the results output
file.
OUTPUT [MAST] [LOAD] [STIF] [MASS] [DRCO] [CORD]
MAST Master freedom description.
LOAD Reduced load vector.
STIF Reduced stiffness matrix.
MASS Reduced mass matrix.
DRCO Nodal direction cosine matrices.
CORD Nodal coordinates.
Notes
Generalised Coordinates
The GENERALISED COORDINATES command defines the number of generalised
coordinates to be utilised in the exterior freedom set for the superelement. These
270
Superelements
Modal Supports
The MODAL SUPPORT command defines the support conditions for the master
freedoms for local superelement eigen-analysis. This information is also used for
computation of the eigen-modes associated with generalised coordinates. Further
additional master freedoms may be specified for a local eigenvalue analysis using
Guyan reduction.
MODAL SUPPORT [STIF] [MASS]
N Nlast Ndiff < typei >i=1,n < stifi >i=1,n < massi >i=1,n
STIF Stiffness flag.
MASS Mass flag.
N Nlast Ndiff The first node, last node and difference between nodes of
the series of nodes with identical modal supports.
typei Modal support code for each freedom at the nodes. Valid types are:
R - Master freedom restrained.
F - Master freedom free.
S - Spring support for master freedom.
G - Denotes interior freedoms as masters for local eigenvalue analysis
(see Notes).
stifi Stiffness for spring support.
massi Mass for spring support.
n Total number of degrees-of-freedom at the node.
Notes
271
Chapter 6 Running An Analysis
Using Superelements
Superelements are utilised for two reasons:
q to solve the residual structure
q to redistribute the masses and stiffnesses for eigen-analyses.
The SUPERELEMENT USE command is used in both of these cases.
When solving the residual structure additional elements, loading, boundary
conditions and user modal data may also be added directly with the residual
structure. The residual structure is defined as the combination of the lowest level
superelements. When the residual structure is solved, the displacements relating to
all master freedoms are written to the appropriate superelement database(s).
The masses and stiffnesses of the complete (or residual) structure can be
redistributed to the boundaries of each of the superelements using the
SUPERELEMENT DISTRIBUTE COMMAND (see Distribution of Mass and
Stiffness in Modal Synthesis). In order to do this the residual structure must be
formed by utilising the SUPERELEMENT USE command.
SUPERELEMENT USE sname dbname
sname Name of superelement to be solved.
dbname Database where superelement is stored.
Notes
272
Superelements
FORMAT
FORMAT = File format type (FREE or FIXED)
The header line for each data packet contains the following information:
IT ID IV KC N1 N2 N3 N4 N5
273
Chapter 6 Running An Analysis
25 0 0 1
ID = 0 (n/a)
IV = 0 (n/a)
KC =1
TITLE
TITLE = identifying title, may be up to 80 characters
26 0 0 1
ID = 0 (n/a)
IV = 0 (n/a)
KC =1
DATA
DATA = character string containing Date:Time:Version
274
Superelements
27 0 0 KC N1 N2 N3
ID = 0 (n/a)
IV = 0 (n/a)
N1 = maximum number of freedoms to a node
N2 = total number of boundary nodes i.e. connectivity to residual structure
N3 = number of eigenmodes
KC = 1 + N2 + (N3 + 4) / 5 + N3
FRETYP
FRETYP = freedom types at a node (N1 values specified)
For valid freedom types
FREQ
FREQ = natural frequencies for each mode
275
Chapter 6 Running An Analysis
EIGFLE
EIGFLE = name of results file for this mode number
- this file contains the mass normalised eigenvectors and can be in
free or fixed format as
specified on the first line in the neutral file, this line is repeated for each
mode
TITLE
TITLE
TITLE = identifying title, may be up to 80 characters
SUBTITLE
276
Superelements
SUBTITLE
SUBTITLE = identifying subtitle, may be up to 80 characters
NODID (DATA(J),J=1,NWIDTH)
NODID = node number
DATA = nodal freedom displacements
277
Chapter 6 Running An Analysis
3. If the CONSTANTS command is not specified the results from the most recent
analysis will be recovered.
4. A more efficient solution can be obtained if the appropriate file containing the
reduced frontal matrices is utilised in the recovery stage. These files can be
recognised by the extension .sfr and are produced during the superelement
creation phase. To take advantage of this facility, the previously created .sfr file
must be renamed to fname.sfr, where fname is the name of the data file used in
the recovery. Using this approach the re-assembly of the reduced frontal matrices
is avoided and the solution is achieved directly from the back substitution.
However, it should be noted that the .sfr files are generally quite large and may
require a considerable amount of storage.
278