Areva10 Technical Day

Preview:

Citation preview

1

Using reduced system models for vibration design and validation

Etienne BalmèsSDToolsArts et Métiers ParisTechAREVA Technical day, December 10, 2010

A system = I/O representation

Prototype Virtual prototype

☺ all physics (no risk on validity) � limited physics (unknown & long CPU)☺ in operation response � design loads� limited test inputs ☺ user chosen loads� measurements only ☺ all states known� few designs ☺ multiple (but 1 hour, 1 night,

several days, … thresholds)

� Cost : build and operate � Cost : setup, run, manipulate

In Out

EnvironmentDesign point

System

Model complexitySimulation• Geometry (nominal, variability, …)

• Material behavior (viscoelastic, contact/friction, …)

• Input : dynamic environment

• Objectives : static deflection, frequencies, dynamic amplitudes, stresses, cycle counts

• …

Test (modal analysis)• Bandwidth, how many modes

• Number of in/out, reciprocity, residual terms

• Non-linear characterization

• …

3

Welded plates

Clamped end

Spot weld 1 gun

Spot weld 2 gun

Outline• Micro/macro behavior : equivalent behavior

– Homogeneization, updating, …

– Modal damping

• Model reduction : subspace representationCMS, variable separation, POD, PGD, …

– Classical modal synthesis : spatially simple models– Coupling test & FEM models

– Energy coupling & revised CMS

– Design phases / uncertainty

4

Equivalent models : honeycomb example

• Micro : cell walls, glue, face-sheet, viscoelastic material

• Macro : shell/ orthotropicvolume/ shell

• Equivalence: waves/modes

• End result orthotropic law

• Loss of detail

5

Detailed 3Dhoneycomb

Shell/volume/shell

Numerical homogeneization

Updating from test

PhD ECP Jan. 2010 : Corine Florens

6

Equivalent time domain modal damping• Modal damping = assume viscous damping matrix diagonal in modal basis

• Rayleigh damping:

– Physical domain

– Modal domain

• Modal + piece-wise Rayleigh

Reality

Mass Stiffness

Bianchi ISMA Sep. 2010

Equivalent model building• Homogeneization (equivalent material, equivalent model)

• Updating : identical static,frequency, dissipation (weld spot, screw, beam, …)

• Modal damping

with loss of detail

• Model reduction

with restitution7PhD 2005 Abbadi (PSA)

Outline• Micro/macro behavior : equivalent behavior

– Homogeneization, updating, …

– Modal damping

• Model reduction : subspace representation– Classical modal synthesis : spatially simple models– Coupling test & FEM models

– Energy coupling & revised CMS

– Design phases / uncertainty

8

9

System models of structural dynamics

Simple linear time invariant system

Extensions• Coupling (structure, fluid, control, multi-body, …)

• Optimization, variability, damping, non linearity, …

When

Where

Sensors

Large/complex FEM

Modal analysisSuperelementsCMS, …

10

Component mode synthesisReduction (Ritz analysis) based on restrictions :

• Excitation (space & freq)• Responses• Coupling …

σ(x,t)f(x,t)

u(x,t)σ(x,t)

Coupling : state dependent loads

+

+

{q}N=qR

Nx NR

T

11

Moving complexity in the coupling part

Reduced model

• Coupling : test/FEM, fluid/structureactive control, …

• Local non-linearities : machining, bearings, contact/friction, …

• Optimization / uncertainty

In Sensors

12

CMS current practice• Craig-Bampton (unit displacements + fixed interface modes)

– Very robust, guaranteed independence • McNeal (free modes + static response to loads)

– Tends to have poor conditioning (residual flexibility)

• Well established applications– structural vibrations– multi flexible-bodies– vibroacoustics

• Limits– Very large models– Large interfaces– Parametric design of component– Non local or strong coupling (reduction not independent)

– Hybrid test/analysis– …– Ease of use

Example : structural dynamics modification

In

response

Feedback : modification

System : identified

Motivation: • System model very costly (no blue-print, internal complexity)• Need to predict impact before implementing solution

PhD ECP. Corus 2002, Groult 2008

Test model limitations

• Very limited if non-linear

• Typically inconsistent– Channel dependent noise

– Not exactly reciprocal

– Residual terms, not well excited modes

• Spatially incomplete– Few inputs

– Limited outputs

14

System : identified

PhD ECP. Corus 2002, Groult 2008

Hybrid test/FEM using expansion

Instrumented area

Local model}FEM of modification

Structure under test

Problem : know outputs but states (DOF) needed for coupling

Solution• Local model

•Covers instrumented area•Includes the modification

• Expansion•model based estimation•gives knowledge of states

PhD Corus 2002, Groult 2008

Extended SDM handles• Spatial inconsistence• Mass/stiffness/damping modificationsBut requires consistent, linear model of tested system

Outline• Micro/macro behavior : equivalent behavior

– Homogeneization, updating, …

– Modal damping

• Model reduction : subspace representation– Classical modal synthesis : spatially simple models– Coupling test & FEM models

– Energy coupling & revised CMS

– Design phases / uncertainty

16

17

Interfaces for couplingClassical CMS : continuity coupling

• Reduced independently• All interface motion (or interface modes)• Assembly by continuityDifficulties• Mesh incompatibility• Large interfaces• Strong coupling (reduction requires knowledge of coupling)

Disjoint components : energy coupling

• Assembly by computation of interface energy (example Arlequin)

Difficulties• Use better bases than independent reduction

Energy coupling• Disjoint components with interface energy

• Subspace for each component can be arbitrary:valid Rayleigh-Ritz

• Component Mode Tuning method– free/free real modes (explicit DOFs)– trace of the assembled modes on the component

+

Component mode tuning method• Reduced model is sparse• Free mode amplitudes are DOFs

• Reduced model has exact nominal modes(interest 1980 : large linear solution, 2010 : enhanced coupling)

• Change component mode frequency ⇔ change the diagonal terms of Kel

DiscDiscDiscDisc

OuterPadOuterPadOuterPadOuterPad

Inner PadInner PadInner PadInner Pad

AnchorAnchorAnchorAnchor

CaliperCaliperCaliperCaliper

PistonPistonPistonPiston

KnuckleKnuckleKnuckleKnuckle

HubHubHubHub

ωj21

[M] [Kel] [KintS] [KintU]

CMT & design studies

• One reduced model /multiple designs

Examples

• impact of modulus change

• damping real system or component mode

20

Component redesign

Sensitivityenergy analysis

Nom.

+10%

+20%

-20%

21

Classical CMS (Craig-Bampton)• System is brake without contact area

• Reduction : modes of system and interface loads

• Many interface DOFs needed heavily populated matrix

Revised notion of interface

Disjoint component with exact modes

• No reduction of DOFs internal to contact area

• Reduction : trace of full brake modes on reduced area (no need for static response at interface)

PhD ECP. Vermot Jan 2011

22

Full system transient simulation• 800e3 DOF FEMmodes can’t be used because of contact area200e3 time steps = 1.2 To⇒ Need piece-wise reduction

Local detail accessible• Contact pressure/stiffness• Modal damping for accurate instability study

• Post-processing modal amplitudes, component energy

Exact system modes + local NL

PhD ECP. Vermot Jan 2011

23

Disjoint component bases • Reduction by component : minimize basis

storage

• Use system predictions for correct coupling with minimal number of interface modes

Example full shaft model

• Use cyclic symmetry to build

• CMT for mistuning

PhD ECP. A. Sternchüss 2009

Outline• Micro/macro behavior : equivalent behavior

– Homogeneization, updating, …

– Modal damping

• Model reduction : subspace representation– Classical modal synthesis : spatially simple models– Coupling test & FEM models

– Energy coupling & revised CMS

– Design phases / uncertainty

24

25

Parametric families & reanalysis

Reduction basis T can be fixedfor range of parameters

In Out

Design space (p)

System

• Evolutions of frequencies with uncertain parameters

• Effective stiffness of a damping strut

• Campbell diagram• …

26

• Multi-model

• Other + residue iteration

• Example : strong couplingWith heavy fluids : modes of structure & fluid give poor coupled prediction

Bases for parametric studies

Example water filled tank

With residualWithout residual

[T(p1) T(p2) … ]

Orthogonalization

[T]

[Tk] Rdk=K-1 R(q(Tk))

Orthog [Tk Rdk]

27

Conclusion 1Reduced / equivalent models• Reduction gives access to states : typically superior if local detail needed

Reduction methods : • Rely on a approximation of subspaces using bases that can be piece-wise in space and/or time

• Basic tools to build subspaces•Krylov iterations, static response•Conjugate gradient/Lanczos•Eigenvalue/SVD/POD/PGD

• In vibration validity & model complexitydepends on assumptions on loads and frequency range : not FEM model size

In Out

EnvironmentDesign point

System

qR

Nx NR

T

28

Conclusion 2Linear time invariant reduced model still allows• Coupling (test/FEM, structure/component, fluid/structure)• Variability/design studies

Top issues• SDTools, as software editor, aware that first cost is model setup ⇒ ease of use

• Equivalent/reduced models rely on assumptions ⇒ how can these be clear and controlled by the user ? (control accuracy)

• Understanding comes from result analysis at system and component level ⇒ handling restitution ?

• Handling design studies ?• Design methods for non-linear vibration

www.sdtools.com/publicationsProducts : SDT, OpenFEM, Visco, Rotor, Runtime for use within MATLAB

29

30

Post-processing with reduced models• Restitution

– Many DOFs a few DEF (energy, strain, …)

– A few DOFs many DEF (animation, test/analysis correlation)

– Time simulation sub-sampling

• Understanding the response– Component energies

– Time/freq SVD

• That’s the real frontier

31

Multi-frontal solvers / AMLS• Graph partionning methods ⇒group DOFs in an elimination tree with separate branches

• Block structure of reduction basis

• Block diagonal stiffness

• Very populated mass coupling

• Multi-frontal eigensolvers introduce some form of interface modes to limit size of mass coupling

KM

Recommended