NVH Lecture 7: Operational Modal Analysis

Transcription

1 NVH Lecture 7: Operational Modal Analysis Marie Curie Graduate School on Vehicle Mechatronics & Dynamics, Leuven, 5-8 February 2013 Bart Peeters Operational Modal Analysis Why? Identify models that represent real behaviour Real operating conditions laboratory conditions Effects due to: Non-linearities e.g. car suspension Environment Experimental models that give best linear representation under the relevant operating conditions Practical problems for executing lab-test Access to the actual structure Problems for applying adequate input forces Impossible, difficult Expensive Presence of substantial ambient excitation Health monitoring / damage detection: in-situ Make extended use of available data (ODS) 2 copyright LMS International Copyright LMS International 1

2 Operational Modal Analysis In-operation testing Some applications permit the use of Input - output (FRF) data during normal operation Require special setups for forced excitation Rotating wing-tip vanes Electromagnetic bearings Low-frequency exciters Drop-weights Unbalance shakers Pyrotechnics Control Surface Input Servo-drive inputs (robots) Testing complexity Data quality (undesired ambient sources) Some applications permit simulating in-operation conditions in I/O (FRF) tests (car suspension ) The normal EMA processes can be followed NASA ( ) EMPA KUL 3 copyright LMS International In-operation EMA example: Business jet, wing-vane in-flight excitation In-flight excitation, 2 wing-tip vanes 9 responses 2 min sine sweep Higher order harmonics Very noisy data 0.10 g/n Log Phase 1.00 (m/s2)/n Log 10.0e Phase Hz 1.00 Amplitude / Coherence w ing:vvd:+z/multiple PolyMAX Hz Coherence back:vde:+y/multiple 0.05 Hz 4 copyright LMS International Copyright LMS International 2

3 Operational Modal Analysis The output-only system What identification is Operational problem Modal Analysis? Majority of applications: use of output-only data Objective: Identification of modal parameters from outputonly data measured on a structure during standard operation. Eigenfrequency Damping Mode shape Output-only operational modal analysis = identifying H Based on Y Without knowing U Desired (but unknown) ambient sources: white spectrum Input System Output U H Y 5 copyright LMS International Operational Modal Analysis The output-only system identification problem In the remainder, we will focus on output-only methods for in-operation modal analysis. We will refer to this as OMA Operational Modal Analysis (= operational response modal analysis). In general (papers, software vendors, test procedures), OMA hence implies outputonly modal analysis U H Y White noise Ideal operational excitation White noise + harmonic Less ideal operational excitation 6 copyright LMS International Copyright LMS International 3

4 Operational Modal Analysis in the presence of rotating machinery Sweeping harmonics (run-up) Harmonics excite broad frequency band Useful excitation no filtering! End-of-order effects Fixed (or slowly varying) harmonics Harmonics hamper the identification process Have to be removed from data Helicopter in-flight data Car engine run-up 7 copyright LMS International Animation of Structures Time Animation Operational Deflection Shapes Direct animation of signals in time domain Direct animation of frequency-domain functions Actual deformation as a function of time or frequency Explanation & model through modal analysis 8 copyright LMS International Copyright LMS International 4

5 Operational Modal Analysis More than operational deflection shapes Auto & Cross Powers Peak picking Deformation at a chosen frequency line No damping information Combination of modes and forced responses Combination of closely spaced modes Data reduction through coherence or SVD analysis (Principal Component Analysis) Phenomena ODS OMA Modal model Frequency Damping Mode shape (No modal scaling) Use of system identification methods Structural characteristics Separation of closely spaced modes Root causes Vibration problem root cause discriminator 9 copyright LMS International OMA as part of an engineering workflow Wind turbine operational testing 1 2 Component testing Classical Modal Analysis Validation of component models Assembled structure testing Operational Modal Analysis Validation of assembly models 3 Z Y X MN FE MODEL MX EMA MODEL 4 Y Z X MN M Z UPDATED FE MODEL MX Z Y X MN MX X MN Y Source: Laurent Bonnet, GE Energy LMS Conference Europe, Nürburg, 2-3 March copyright LMS International Copyright LMS International 5

6 0.10 g/n Log ( ) ( ) Traditional (IO) modal parameter estimation Modal model Frequency domain Time domain e Phase Hz Inverse Fourier transform g/n Real s 6.00 FRF H n * n Ai Ai it ( ) h( t) * Ai e Ai i1 j i j i i1 * e * t i IRF T A Q v l T i i i i i i * 2 i, i ii j 1 i i 11 copyright LMS International Stabilization diagram H( ) n v i i1 T li j i * v H i li * j i Try a whole range of model orders Compare modal parameters at current order with previous order n Stability o : new f : freq d : damp + freq v : part. vector + freq s : all 12 copyright LMS International Copyright LMS International 6

7 Stabilization diagram PolyMAX?! Model order problem shifted to problem of separating true from computational poles? 13 copyright LMS International Preprocessing operational data Operational data y k Periodogram ( classical ) s) Y DFT[window y S S ( ( s) k ( s) ( s) ( s) H yy Y Y P 1 ( s) yy ( ) S yy ( ) P s1 ] Correlogram ( half spectra ) 1 N 1 T y k0 ki Ri yk N Syy ( ) DFT[window { R L,..., R0,..., RL}] S ) DFT[window { R / 2,..., R }] yy ( 0 L 10.0e-9 ( ) Log Periodogram with Hanning window 1.00e-9 (m/s 2 ) 2 Log Correlogram with exponential window autopow er_spectr roof:1:+z / roof:1:+z crosspow er_spect roof:1:+x / roof:1:+z AutoPow er roof:1:+z CrossPow er roof:1:+x/roof:1:+z 1.00e Phase copyright LMS International Hz e Phase Hz 8.00 Copyright LMS International 7

8 Why half spectra? 10e-9 (m/s 2 ) 2 Real Time roof:1:+x Unw indow ed Time roof:1:+x Lower order models S ( ) S ( ) S ( ) S yy yy ( j) yy n v i i1 yy gi j i H * v * i gi * j i -10e-9-90 (m/s 2 ) 2 db Phase s AutoPow er roof:1:+x Unw indow ed AutoPow er roof:1:+x 0.00 Hz Exponential window Reduces the effect of leakage Reduces the influence of the higher time lags having a larger variance Compatible with the modal model ( Hanning window with biased damping) 15 copyright LMS International Operational Modal Parameter Estimation Exponential Window No window 10% exponential window 1% exponential window 16 copyright LMS International Copyright LMS International 8

9 ( ) Overview of OMA methods frequency domain Basically, all EMA methods that operate on FRFs can be transferred to OMA methods operating on auto- and cross-spectra FRF (EMA) Auto/cross spectra (OMA) -90 Methods evolution: 0.00 Hz Method Nonparametric Peak picking / ODS SVD-based PCA, CMIF, FDD Can be complimented by simple SDOF curvefitters Parametric PolyMAX (Polyreference) Maximum Likelihood Features Subjective No damping Still picking peaks (subjective / difficult) Better than PP for closely spaced modes SDOF curve-fitting Clear stabilization Highly automatic Stochastic (incl. noise information) Iterative -35 ( ) db 0.21 ( ) Imag : Real 0.13 (m/s2)/n ((m/s2)/n) FRF moto:15:+z moto:15:+z FRF moto:15:+z / moto:15:+z 0.00 Hz Imag copyright LMS International Overview of OMA methods time domain Basically, all EMA methods that operate on IRFs can be transferred to OMA methods operating on auto- and cross-correlations IRF (EMA) Auto/cross correlations (OMA) Methods evolution: Method Output time histories Output correlations Stochastic time series modelling; e.g. ARMA models Subspace identification (BR, CVA, : differences very subtle) Polyreference LSCE Subspace identification Features Not practical for OMA No convincing cases in literature Successful applications Limited number of samples / channels, large computation time Fast Data reduction step (time series correlations) Relatively fast 1.00 ( ) Real e-9 (m/s 2 ) 2 Real -10e s 6.00 Data reduction w/o loss of information Time roof:1:+x Unw indow ed Time roof:1:+x 0.00 s copyright LMS International Copyright LMS International 9

10 Case studies Aerospace engineering Civil Engineering Automotive Engineering 19 copyright LMS International Test FEM correlation Frequency correlations within 5% FEM updating with delta stick approach 20 copyright LMS International Copyright LMS International 10

11 ( ) Next: aero-elastic simulation and in-flight flutter testing Traditional FEM, GVT-updated FEM, or direct GVT M x( t) Cx ( t) Kx( t) F ( x) 0 a Aerodynamic panel method Due to presence of aero-dynamic term, modes of structural system are changing with airspeed and altitude Flutter analysis = assessing evolution of modes (zero-crossing of damping value) FE Model Test Model (GVT) Aerodyn. Panel Model Physical prototype 21 copyright LMS International Flight flutter testing turbulence excitation Background Testing Analysis Damping Flutter m/s2 Real s Airspeed Telemetry link g 2 Amplitude Hz Context: FLITE 2 research project with Airbus France, Dassault Aviation, PZL, LAE, VUB, KUL, AGH, UMAN, INRIA, SOPEMEA, ONERA, LMS 22 copyright LMS International Copyright LMS International 11

12 Application to aircraft flight test data Generally only a few sensors 23 copyright LMS International A380 in-flight tests But sometimes a bit more: 150 accelerometers 24 copyright LMS International Copyright LMS International 12

13 OMA PolyMAX db Sum Crosspow er SUM Synthesized Crosspow er SUM Phase Hz 25 copyright LMS International In-flight OMA mode shape (1/4) 26 copyright LMS International Copyright LMS International 13

14 In-flight OMA mode shape (2/4) 27 copyright LMS International In-flight OMA mode shape (3/4) 28 copyright LMS International Copyright LMS International 14

15 In-flight OMA mode shape (4/4) 29 copyright LMS International Conclusions In-flight testing OMA: some important flutter-critical modes not excited EMA: some modes mainly excited by the turbulences may not be identified Conclusion: beneficial to use artificial excitation, but data analysed with stochastic methods that also take into account the unknown excitation Airbus flight test team evaluated LMS Test.Lab using large-aircraft data We actually achieved better results using operational techniques than with classical EMA. We found more modes. The synthesis was better with higher correlation and fewer errors. And the in-flight mode shapes looked much nicer! We found that the exponential window, which allowed for cross-correlation calculations was a good de-noising tool for our in-flight data. Altitude (feet) 40,000 30,000 20,000 10,000 MACH 0.95 MACH 0.90 MACH True Air Speed (knots) 30 copyright LMS International Copyright LMS International 15

16 OMA has its roots in Civil Engineering 7 November 1940 Tacoma (USA) 10 June 2000 London (UK) 31 copyright LMS International Applications: civil engineering Design verification FE model correlation / updating Identification of damping properties Indirect identification of cable forces Structural health monitoring Verify effectiveness of rehabilitation Øresund Bridge g 2 db Guadiana Bridge Deck:117:+Z Synthesized Crosspow Hz er Deck:117:+Z 0.30 Linear Hz 3.00 Millennium Bridge 32 copyright LMS International Copyright LMS International 16

17 23/11/2002: Bradford City Sheffield United: m/s2 Real ( ) 0.20 Real m/s2 ( ) Goal 1 Goal 2 Goal 3 Goal 4 Goal 5 End of game F B time_record roof:1:+x / Root Mean Square time_record roof:1:+x s Filling Seated Half time Emptying Empty Data acquisition: 4 h Sampled at 80 Hz (down-sampled to 20 Hz) Sliding RMS value ( ) 1000 samples, 50% overlap 33 copyright LMS International Continuous monitoring of the Bradford and Bingley Stadium Data: Paul Reynolds, Sheffield University 34 copyright LMS International Copyright LMS International 17

18 Ambient Vibration Testing of the Millau Viaduct 35 copyright LMS International Millau Viaduct LMS Test.Lab PolyMAX results Source: E. Caetano, F. Magalhaes, A. Cunha (univ. Porto), O. Flamand, G. Grillaud (CSTB) EVACES Conference, Porto, copyright LMS International Copyright LMS International 18

19 Guadiana Bridge Modal parameters Mode Freq. [Hz] Damp. [%] Mode Freq. [Hz] PolyMAX single analysis results Damp. [%] copyright LMS International Ship hull vibration measurements 5 Three-axis acceleration sensors At forward, aft ends and center of upper deck At end of front & aft of compass deck (longitudinal, lateral and torsion vibrations measurement of deckhouse) Compass deck 7 Mono-axis acceleration sensors (vertical direction) On the right side of upper deck Data Measurement Long., trans. and vert. direction Vertical direction 前後 左右 上下方向 Upper deck Master/Slave configuration 38 copyright LMS International Test.Lab & SCADAS-Mobile Copyright LMS International 19

20 Change of Vibration Acceleration at aftbody Measurement point 1, Z direction Change of order components for engine speed Change of frequency component for engine speed Rpm Extr (T1) rpm Hz 50 DeckL:1:+Z (CH3) db g Rpm Extr (T1) rpm Hz 50 DeckL:1:+Z (CH3) db g e-3 g F F F Order 1.00 DeckL:1:+Z Order 2.00 DeckL:1:+Z Order 6.00 DeckL:1:+Z 7.40e-3 g F F F F Frequency 3.40 Hz DeckL:1:+Z Frequency 9.10 Hz DeckL:1:+Z Frequency Hz DeckL:1:+Z Frequency Hz DeckL:1:+Z rpm rpm copyright LMS International OMA results: hull vertical vibration modes Hull Vertical 4 nodes 9.1 Hz Hull Vertical 5 nodes 11.1 Hz Hull Vertical 6 nodes 14.3 Hz 30 Glabal vibration mode Hull vertical mode Hull lateral mode Upper structure mode Hull Vertical 3 nodes 6.1 Hz Frequency [Hz] Mode number Hull Vertical 2 nodes 3.4 Hz 40 copyright LMS International Copyright LMS International 20

21 Ship hull vibration measurements conclusions Classical modal analysis requires excitation devices Costly / cumbersome Not enough excitation energy for large-dimension ships (poor SNR) Unbalance shaker Impact test Operational Modal Analysis is powerful tool for identifying the vibration behavior of the global and local structure of a large ship in operational conditions Ambient excitation = white noise + harmonic forces from propulsion (engine and propeller system) Higher-frequency modes successfully identified from engine run-up data Results are in good agreement with anchor drop test and waveinduced vibrations 41 copyright LMS International Validation of Complete Vehicle Model (CVM) LMS Virtual.Lab correlation LMS Test.Lab PolyMAX NASTRAN FEM Test wireframe Simulated operation data 4-poster measurements Test track measurements Source: Martin Olofsson, Peter Nilsson (Volvo Truck) IOMAC, Copenhagen, copyright LMS International Copyright LMS International 21

22 Mode shape correlation (MAC) FEM ODS / OMA vs. FEM OMA shows better correlation with FEM than ODS Not all FEM modes could be found in operational data (not well excited by 4-poster, too highly damped) Others have high correlation ODS OMA Reduced FEM vs. OMA mode shape 43 copyright LMS International Testing and identification of active systems Intelligent and complex systems engineering Multi-physics system simulation: 1D, 3D, control State estimator: processing data by models Applied to Vehicle Dynamics Control (Flanders Drive project) Testing of active system Model validation and updating Case study Ford S-MAX 4-poster and Proving Ground tests Sensors 15 DOF Steering Suspension force elements Suspension force elements Compliances Tires Road input input / 4 / Post 4 Post Shaker Shaker 44 copyright LMS International Copyright LMS International 22

23 Full-scale modal wind turbine tests: comparing shaker excitation with wind excitation NREL: U.S. Wind Resource (50m) 45 copyright LMS International EMA Mode shapes (avi) Tower fore-aft Tower side-to-side 1st rotor symmetric flap 46 copyright LMS International Copyright LMS International 23

24 Shaker excitation (EMA) vs. wind excitation (OMA) Excellent agreement between EMA & OMA Modes not well excited by wind not identified by OMA 47 copyright LMS International Conclusions Operational Modal Analysis is a mature technology High-quality data acquisition Advanced parameter estimation algorithms Commercial software implementations Industrial applications Last decade, evolution in Technology Usability Applicability: no isolated results but part of engineering workflow Civil engineering Aerospace engineering Automotive engineering 48 copyright LMS International Copyright LMS International 24

25 Thank you Related technical papers can be downloaded from Marie Curie Graduate School on Vehicle Mechatronics & Dynamics, Leuven, 5-8 February 2013 Bart Peeters Copyright LMS International 25