Operational Modal Analysis Course

INCLUDING INTRODUCTION TO STRUCTURAL HEALTH MONITORING

Niels-Jørgen Jacobsen
October 5+6, 2023

www.hbkworld.com | © HBK – Hottinger, Brüel & Kjær | All rights reserved

CONFIDENTIAL - EXTERNAL
The Presenter
Education
 Joined Brüel & Kjær in 1986 after graduating from the Technical University of Denmark
 Degree in International Business from Copenhagen Business School in 1992

Job Positions
 Digital Signal Processing (DSP) algorithm Software Developer on Analyzer Systems
2035 and 3550
 Project Manager on PULSE software applications including Order Analysis, Balancing,
Sound Power, Acoustic Material Testing etc.
 Application Specialist in Structural Dynamics, Machine Diagnostics and Industrial Acoustics
 Manager of Vibration Specialist Group
 Product Manager for Structural Dynamics Solutions and Array Acoustics Solutions

 Member of the Permanent Committee and the Scientific Committee of IOMAC

Niels-Jørgen Jacobsen
Product Manager
Technical Papers and Articles Structural Dynamics Solutions
 Author of many technical papers and articles and Array Acoustics Solutions
• Conference papers, Magazine articles, Technical Reviews etc.

“It’s all about dedication”

2
CONFIDENTIAL - EXTERNAL
 Attendee introductions
 Name, company, position/duties
 Your experience with Operational Modal Analysis

1 2 3 4 5

• If 3 or higher:
˗ Which techniques you have used
˗ Which software/tools you have used
 Your experience with Experimental Modal Analysis
 What you hope to get out of this course

3
CONFIDENTIAL - EXTERNAL
 Objectives
To give all attendees the necessary theoretical and practical knowledge to
successfully conduct proper Operational Modal Measurement and Analysis

By explaining
 The terminology, possibilities, advantages and limitations of Operational Modal Analysis (OMA)
 How OMA can be used as a complement to experimental modal analysis (EMA) based on
typically hammer and shaker testing
 How to perform proper measurement and validation of the vibration signals used for OMA
 The different time and frequency techniques used in OMA and how to validate results
 Specific requirements to instrumentation used for OMA

In addition
 Introduce the idea behind and benefits of vibration-based Structural Health Monitoring (SHM)

4
CONFIDENTIAL - EXTERNAL
 Agenda Day 1 – October 5th, 2023 – Morning

Welcome, objectives and agenda

Introduction to Operational Modal Analysis (OMA)
- Basic idea
- Applications of OMA
- Advantages, assumptions and limitations
- Data acquisition (measurement)
- Overview of modal parameter identification techniques for OMA
- Comparison of OMA to Experimental Modal Analysis (EMA)
Coffee break
Measurement example using BK Connect
Measurement validation tools
- Singular Value Decomposition (SVD) plots
- Contour plots
Exercises - Measurement and measurement validation
Lunch break

5
CONFIDENTIAL - EXTERNAL
 Agenda Day 1 – October 5th, 2023 – Afternoon

Projection channels
Frequency domain techniques
- Frequency Domain Decomposition (FDD)
- Enhanced FDD (EFDD)
- Curve-fit FDD (CFDD)
- Consequences of harmonic components
- Detection and reduction of harmonic components in the frequency domain
- Automatic mode estimation
Coffee break
Analysis Validation tools
- Complexity plots and mode normalization
- MAC plots and tables
- Comparison of different estimation techniques
Exercises - Frequency Domain Decomposition Techniques
Q&A

6
CONFIDENTIAL - EXTERNAL
 Agenda Day 2 – October 6th, 2023 – Morning

Recap and questions to yesterday's training

Frequency Response models and State Space models
Time domain techniques
- Stochastic Subspace Identification (SSI) overview
- SSI - Main elements
- SSI techniques (UPC, PC, CVA)
- Analysis validation tools (Response validation, Prediction error)
Coffee break
Time domain techniques cont.
- Uncertainty estimation using UPCX
- Hints and tips using SSI
- Reduction of harmonic components in the time domain
Exercises - Time domain techniques
Lunch break

7
CONFIDENTIAL - EXTERNAL
 Agenda Day 2 – October 6th, 2023 – Afternoon

Exercises - Time domain techniques cont.

Batch processing using SSI
Structural Health Monitoring (SHM)
- Introduction to SHM
- Vibration-based SHM using SSI
Coffee break
Structural Health Monitoring (SHM) cont.
- Damage detection example
Further information
Q&A, Course evaluation

8
CONFIDENTIAL - EXTERNAL
 Introduction to Operational Modal Analysis (OMA)

9
CONFIDENTIAL - EXTERNAL
 Contents
 Structural Testing Overview
 OMA - One-page Overview
 Basic Idea and Applications
 Advantages, Assumptions and Limitations
 Excitation and Data Acquisition
 Identification Techniques
 Examples:
• Z24 Highway Bridge
• Wind Turbine Blade
 When to use EMA and when to use OMA

10
CONFIDENTIAL - EXTERNAL
 Structural Testing Overview
Problem
Non-optimal Dynamic Behaviour of Structures

Market Drivers
Safety, Reliability, Performance, Consumption, Comfort

Controlled Measurements In-operation Meas. Test-FEA Integration

Applications Applications Applications

Mobility Measurements Time ODS Pre-test Analysis
Experimental Modal Analysis 1) Spectral ODS (FFT or Order) Model Correlation & Validation
Single/Poly Ref. Hammer Testing Run-Up/Down ODS Sensitivity Analysis
1) Often referred to as SIMO/MIMO Shaker Testing Operational Modal Analysis Model Updating
Classical Modal Analysis or
Normal Mode Tuning Flutter Analysis (Pole Trending) Model Refinement (damping)
Mobility-based Modal Analysis
Forced Response Simulation Structural Health Monitoring … Structural Modifications
as OMA is also Experimental
Structural Modifications Hybrid Modelling …
Inertial Property Calculations ...

Answers
What is the structure’s actual behaviour and/or inherent dynamic properties ?
How can we improve the structure’s behaviour and/or dynamic properties ?
11
CONFIDENTIAL - EXTERNAL
 Operational Modal Analysis (OMA) – One page overview
To obtain a mathematical model of the dynamic properties of a structure by
experimental means using unmeasured natural excitation (modal testing)

In-operation measurements (in situ)

– True boundary conditions
– Actual excitation signals
– Actual environmental conditions

Broadbanded ambient or self-generated

natural excitation signals
– Machinery running, vehicles driving,
aircraft flying, ships sailing ...
– Civil engineering structures excited
by wind, waves, traffic loads …

Results – Modal Parameters

– Natural Frequency Operational Modal Analysis answers the questions:
– Modal Damping What is the structure’s inherent dynamic properties?
– Mode Shape How can they be improved?
12
CONFIDENTIAL - EXTERNAL
 Operational Modal Analysis – Basic Idea
 Determination of a mathematical Modal Model under operating conditions by response testing only
(output-only modal analysis)
• In situ testing or testing in simulated natural environment
• Natural input is used (ambient or self-generated vibrations)
• No measurement of input forces required
• Measurement procedure quite similar to ODS analysis

 Originated from Civil Engineering applications

(ambient modal analysis, natural input modal analysis)
• Bridges, buildings, towers, wind turbines, stadiums, dams, off-shore platforms, power plants etc.

13
CONFIDENTIAL - EXTERNAL
 Operational Modal Analysis – Applications
 Today widely used for Mechanical Engineering applications
(operational modal analysis)
• In-flight testing of aircraft for flutter prediction
• Helicopters, space structures etc.
• Maritime: Ships, submarines, diesel engines …
• On-road full-body testing of vehicles, trains, trams etc.
• Engine, transmission and exhaust system testing
• Operating rotating machinery (engines, generators, turbines etc.)
• Consumer Products: White goods, power tools, disk drives etc.
• Universities and Research Institutes for industrial research and education
• And many others …

14
CONFIDENTIAL - EXTERNAL
 Operational Modal Analysis – Advantages
 Modal model represents real operating conditions
• True boundary conditions
• Actual force & vibration levels and frequency range
• Actual environmental conditions

 Structures impossible or difficult to excite can be tested

• Large or fragile structures
• Structures with internal or ambient excitation not possible to control or accurately measure
• Structures with boundary conditions difficult to simulate
• Non-linear structures e.g., composite materials

 Simple, straightforward and fast setup, measurement and analysis

using only accelerometers
• No elaborate fixturing of structure, shakers and force transducers
• Simple measurement procedure quite similar to ODS analysis

 Reduced instrumentation cost

• No test rigs, shaker stands, shakers, power amplifiers, force transducers,
15
signal generators, impact hammers etc.
CONFIDENTIAL - EXTERNAL
 Operational Modal Analysis – Assumptions and Limitations
 Same as for Classical Modal Analysis
• Linear and time-invariant structure
• Broad-banded excitation needed
˗ Sufficient energy for all modes of interest. Run-up/down of engines often sufficient

 Mathematical
• Stationary input signals can be approximated by filtered zero mean Gaussian noise
˗ Signals are completely described by their correlation functions or auto- & cross-spectra
˗ Synthesized and measured correlation functions or auto- & cross-spectra are similar

 Deterministic signals (harmonic components) might destroy the results

• Must either be removed from the time signals or identified in the frequency domain
and their influence reduced/eliminated
˗ Can, in general, not be removed by simple filtering
• Time domain methods based on orthogonal projection and non-linear optimization are available
• Frequency domain methods based on Kurtosis available

 Unscaled (uncalibrated) Modal Model

• No absolute Forced Response Simulations or Structural Modification simulations possible
• Scaling can be found by driving points measurements from Classical Modal Analysis
• Scaling can be done by moving mass method
16
CONFIDENTIAL - EXTERNAL
 Excitation and Data Acquisition – Stochastical Approach
 Excitation should be shaped Gaussian white noise

 Excitation should be (approximately) random in time and space (uncorrelated)

acting over the entire structure

 OMA is based on true Multiple-Input Multiple-Output (MIMO) techniques.

In general, the more excitation points, the better

 The measurement time depends on the period of the lowest mode of interest, the
complexity of the structure and the type of excitation.
As a rule of thumb:
1
T  1000  [s]
f min
 Artificial unmeasured uncorrelated broadband excitation can be added to the
natural input excitation for weakly excited structures, for example, to extend the
excitation frequency range or increase the excitation level

17
CONFIDENTIAL - EXTERNAL
 Data Acquisiton – Measurement Procedure
Operational Modal Analysis and Operating Deflection Shapes (ODS) Analysis
 Measurements can be done in a single data set (fast, data consistency) or by using multiple data sets
(limited number of transducers)
Operational Modal Analysis
 Single data set: No reference transducers needed
 Multiple data sets: One or more fixed transducers are needed as references for all data sets
ODS Analysis
 One reference transducer is always required for each frequency/order of interest
 Advantageous to use multiple reference transducers to ensure sufficient energy in
the reference signal at all frequencies/orders of interest
Accelerometers are fixed or roved for each data set

Fixed Reference
Accelerometers

18
CONFIDENTIAL - EXTERNAL
 Data Acquisition – High Dynamic Range required
Dyn-X
As the input is neither measured nor controlled,
the output levels are unknown and unpredictable

Consequences:
 Overloads destroy the measurements
 Under-range situations give poor signal-to-noise ratio

Solution:
 The dynamic range of the DAQ system should match the best transducers available
• No overload or under-range situations will occur due to the data acquisition hardware
˗ If overloads occur, then it is the transducers being overloaded
• There will be no need for time-consuming trial runs and autoranging
• You will not end in a situation, where the measurements must be re-done (often not possible)

19
CONFIDENTIAL - EXTERNAL
 Identification Techniques

 Non-parametric
• Modal parameters are estimated directly from curves, functional relationships or tables

Experimental Data Signal Processing Modal Parameters

 Parametric
• Modal parameters are estimated from a parametric model fitted to the signal processed data

Experimental data Signal Processing

Parameter Estimation Modal Parameters
Parametric Model

20
CONFIDENTIAL - EXTERNAL
 Identification Techniques

Measured Data fn, , 

Signal Processing Parametric Fit
Time Histories
Natural Frequency, fn
Damping, 
Mode Shape, 

EFDD (SDOF, Basic fit)

FDD
(Pure Signal Processing) CFDD (SDOF, Advanced fit)
SSI (MDOF, Advanced fit)

 Non-parametric technique
˗ Frequency Domain Decomposition, FDD

 Parametric techniques
˗ Enhanced Frequency Domain Decomposition, EFDD
˗ Curve-fit Frequency Domain Decomposition, CFDD
˗ Stochastic Subspace Identification, SSI
21
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
Singular Value Decomposition of matrices
n
R = V  S  V  H
=  vi si viH = s1v1v1H + s2 v2 v2H +  + sn vn vnH
i =1

The Singular Value Decomposition

of the response matrices is
performed for each frequency

Singular values Singular vectors

 s1 0 0 0 . . 0.
0 s2 0 . . . 

0 0 . . .
[S] = 
0 . . s .

[V] = v1 v2  v3  . . . vn 
. . 0 .
 
. 0 .
. 0 .
 
 0 . . . . . . 0

• A real diagonal matrix • Orthogonal columns

• Number of non-zero elements in • Unity length columns
the diagonal equals the rank • Approximates the Mode shapes
• Identify Natural frequencies
from Singular Value spectra
22
CONFIDENTIAL - EXTERNAL
 Identification Techniques – Frequency Domain
 Dedicated peak-picking techniques
Modes are picked from SVD (Singular Value Decomposition) curves 1)
• Frequency Domain Decomposition (FDD)
• Enhanced FDD (EFDD)
• Curve-fit FDD (CFDD)

 Standard FRF-based techniques

• Based on standard curve-fitters known from
Experimental Modal Analysis, for example, PTD and RFP-Z

1) The Complex Mode Indicator Function (CMIF) used in EMA is based on SVD, but in
OMA we are not just using the SVD for mode indication, but also for mode extraction

23
CONFIDENTIAL - EXTERNAL
 Identification Techniques – Time Domain

 Stochastic Subspace Identification (SSI) techniques

Modes are found from stabilization diagrams based on Discrete-time
Stochastic State Space Models
• Unweighted Principal Components (UPC)
• Extended Unweighted Principal Components (UPCX)
˗ Includes uncertainty estimates of all modal parameters

• Principal Components (PC)

• Canonical Variate Analysis (CVA)

 UPC, PC and CVA differ in how the data is weighted

24
CONFIDENTIAL - EXTERNAL
 Stochastic Subspace Identification (SSI) – Overview
Combined System A

Stationary
zero mean Measured
Gaussian Responses
White Noise
Excitation Filter Structural System
(linear, time-invariant) C xt →  → yt
wt (linear, time-invariant)

Unknown excitation forces vt

Discrete-time Stochastic State Space Model

State Equation xt +1 = Axt + wt Model of the dynamics of the system
Observation (Output) equation yt = Cxt + vt Model of the output of the system

wt: Process noise – vt: Measurement noise – Model order: Dimension of A

Modes extracted using Crystal Clear Stabilisation Diagrams

25
CONFIDENTIAL - EXTERNAL
 Example – Z24 Highway Bridge – Automation

• 9 data sets with 33 DOFs each except data set 5 with 27 DOFs.
• 5 reference channels used. Total of 251 DOFs. fs = 100 Hz; T = 655 s
• Ambient vibration: Traffic under bridge; Wind

26
CONFIDENTIAL - EXTERNAL
 Example – Z24 Highway Bridge – First 8 elastic modes

3.86 Hz 4.88 Hz
9.74 Hz 10.2 Hz

12.2 Hz 13.3 Hz 20.3 Hz 29.2 Hz

27
CONFIDENTIAL - EXTERNAL
 Example – OMA on a 34m Wind Turbine Blade
Rationale: OMA is ideal for wind turbine blade testing Measurements performed by
 Simple and fast setup measuring only accelerometer outputs Brüel & Kjær together with
colleagues from DTU Wind
 Blade easily excited at multiple DOFs using impact hammers
Energy facilities in Roskilde,
• No measurement of input forces Denmark as part of the
• No requirement to impacting the blade in exact input DOFs EUDP project
 Shaker testing would require special long-stroke shakers
 Roving impact hammer testing impractical due to blade size and location
 Fixed impact hammer testing impractical due to insufficient energy distribution

Test Setup
 Blade length: 34 m – composite materials
 Excitation using impact hammers at multiple DOFs - No measurement of input forces
 Response measurements in 78 DOFs - 60 DOFs on blade; 18 DOFs on support system
 Data acquisition and analysis using PULSE
• Distributed LAN-XI DAQ system on blade and on ground; PULSE sw for measurement and analysis

28
CONFIDENTIAL - EXTERNAL
 Pictures from Test

Case Study Pictures courtesy of SSP

29 www.bksv.com/media/doc/bn1630.pdf CONFIDENTIAL - EXTERNAL
 SSI stability diagram and first 5 blade modes
1.8Hz; 0.15% - Edge 4.1Hz; 0.32% - Flap

6.9Hz; 0.35% - Edge 9.0Hz; 1.1% - Flap 12.4Hz; 0.53% - Torsion

30
CONFIDENTIAL - EXTERNAL
 When to use EMA and when to use OMA

Situation EMA OMA

Non-operating conditions X (X)
Operating conditions X
Structures with internal or ambient excitation not X
possible to control or accurately measure
Boundary conditions difficult to simulate in a lab X
Large or fragile structures X X
Scaled modal model X
Linearity studies X
Fast results using impact hammer testing X

If in-operation measurements must be done, but OMA is not applicable, ODS analysis is the preferred tool:
• Many dominant deterministic signals (harmonic components)
• Weak stochastic information
• Highly non-linear structures
• …
31
CONFIDENTIAL - EXTERNAL
32
CONFIDENTIAL - EXTERNAL
 Q&A – Introduction to Operational Modal Analysis (OMA)
Q1: What best describes OMA?
a) A technique to find the modal parameters of both mechanical and civil engineering structures
b) A technique to construct a mathematical model of the dynamic properties of a structure by
experimental means using unmeasured natural excitation
c) A technique that produces very good modal parameter estimation using single shaker excitation

Q2: Which of the below statements are benefits using OMA?

a) The modal model represents real operating conditions
b) Structures impossible or difficult to excite can be tested
c) Simple, straightforward and fast setup, measurement and analysis
using only accelerometers
d) Reduced instrumentation cost

33
CONFIDENTIAL - EXTERNAL
 Measurement using BK Connect
Geometry-guided time data recording

34
CONFIDENTIAL - EXTERNAL
 B&K Demo Plate
Measurement 1
Ref. + 3 roving accelerometers

Measurement 2
Ref. + 3 roving accelerometers

9 nodes measured in total

Freq. span: 1.6kHz Measurement 3

Recording length: 30s Ref. + 2 roving accelerometers

35
CONFIDENTIAL - EXTERNAL
36
CONFIDENTIAL - EXTERNAL
 Measurement Validation in OMA

37
CONFIDENTIAL - EXTERNAL
 Contents
 Singular Value Decomposition (SVD) plots
• Good data quality
˗ Proper excitation and measurement time
˗ Proper selection of projection channels
˗ Repeated roots and closely-coupled modes clearly identified
˗ Enough independent inputs
˗ Sufficient dynamic range
• Poor data quality
˗ Cable breaks, disconnected or defect transducer
• Separating structural modes from harmonic components

 Short Time Fourier Transform (STSF) Contour plots

• Separating structural modes from harmonic components

 Various spectra
• Auto-spectra and Cross-spectra
• Coherence functions
• Average all elements or all the diagonal elements in the spectral density output matrix
38
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Proper excitation
• Modes are clearly observable

39
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Measurement time is sufficient long
• SVD curves are smooth

Smooth curves Noisy curves

40
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD) SVD is the same as the Complex
Mode Indicator Function (CMIF)
known from Classical Modal Analysis
A SVD plot of the Response Spectral Matrix will reveal
 Repeated roots and closely-coupled modes
• Wide peaks at the same or close frequencies in multiple SVD curves

41
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Proper selection of projection channels
• Include at least as many projection channels until the first SVD curve not included is flat (no additional
mode information)

42
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Enough independent inputs
• Number of significant singular values indicate the number of independent inputs

1 input 2 inputs Multiple inputs

43
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Effective dynamic range
• First flat singular value indicate “noise floor”
• Effective dynamic range can be found as difference between highest peak and the noise floor
• Everything below noise floor will be poorly estimated

44
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Poor signal quality - Disconnected or defect transducers, cable breaks etc.
• Shown as flat SVD curves with large offset

45
CONFIDENTIAL - EXTERNAL
 Singular Value Decomposition (SVD)
A SVD plot of the Response Spectral Matrix will reveal
 Harmonic components
• Narrow peaks at the same frequencies in multiple SVD curves

46
CONFIDENTIAL - EXTERNAL
 Contour Plots of Short Time Fourier Transforms (STFT)
 For separating structural modes from harmonic components
• Harmonics are shown as narrow lines

Constant speed:
2100 RPM (35Hz), 60s

Resonances:
~ 340Hz, 480Hz, 690Hz, 830Hz and 930Hz

47
CONFIDENTIAL - EXTERNAL
 Contour Plots of Short Time Fourier Transforms (STFT)
 For separating structural modes from harmonic components
• Harmonic are shown as narrow lines

Run-up/down:
0-6000 RPM (0-100Hz), 60s

Resonances:
~ 360Hz, 500Hz, 725Hz, 875Hz and 975Hz

48
CONFIDENTIAL - EXTERNAL
 Various spectra
 Auto- and Cross-spectrum
• Between all channel DOFs and the projection channel DOFs
 Coherence
• Between projection channel DOFs only
 Average all elements in the diagonal in the spectral density matrix
 Average all elements in the spectra density matrix

49
CONFIDENTIAL - EXTERNAL
50
CONFIDENTIAL - EXTERNAL
 Q&A – Validation of OMA Measurements
Q1: What can the SVD plot be used for?
a) Observe the input level from different excitation points/DOFs
b) Observe the data quality of the measurements
c) Discriminate structural modes from harmonic components

51
CONFIDENTIAL - EXTERNAL
 Exercise
OMA measurement using BK Connect Data Recorder
B&K Demo Plate

52
CONFIDENTIAL - EXTERNAL
 Exercise – OMA measurement (1..4)
Measurement 1
Ref. + 3 roving accelerometers

Measurement 2
Ref. + 3 roving accelerometers

9 nodes measured in total

Freq. span: 1.6kHz Measurement 3

Recording length: 30s Ref. + 2 roving accelerometers

53
CONFIDENTIAL - EXTERNAL
 Exercise – OMA measurement (2..4)
Setup
• 1 fixed accelerometer (ref) in node 7 (upper left corner)
• 3 roving accelerometers in 3 measurements
• Attach the 3 roving accelerometers to node 1, 2 and 3 in the z+ direction on the plate
Attach the reference accelerometer to node 7 in the z+ direction
Connect the roving accelerometers to channel 1, 2 and 3 and the reference accelerometer to channel 4 of LAN-XI

BK Connect
1. Open the Time Data Recorder Type 8402 application and import the ‘Plate’ geometry

2. HW table: Rename the signal names to something more descriptive e.g., Ref, Rov 1, Rov 2 and Rov 3

3. Set Up\DOF Setup task: Define the measurement sequencing (remember to set Fixed for the reference accelerometer)

4. Set Up\Trigger Setup task: Add a Signal Trigger and set the Level and Hysteresis

5. Measure\Time Data Recorder task

• Set Freq. Range = 1.6 kHz
• Monitor: Select Time and FFT
• Time Data Recorder properties: Start Recording = Trigger 1; Stop Recording = Duration; Duration = 30s

54
CONFIDENTIAL - EXTERNAL
 Exercise – OMA measurement (3..4)
6. Perform Measurement 1 for 30s by tapping on the plate with your fingers in the z+ direction randomly in
time and space over the entire structure
• When completed notice that Measurement 1 is saved to the Project Browser

7. Move the roving accelerometers to nodes 4, 5 and 6

8. Perform Measurement 2 for 30s …
• When completed notice that Measurement 2 is saved to the Project Browser

9. Move the first two roving accelerometers to nodes 8 and 9

10. Perform Measurement 3 for 30s …
• When completed notice that Measurement 3 is saved to the Project Browser

11. Optionally trim the recordings in the Process\Time Data Review task
• E.g., if there were overloads

55
CONFIDENTIAL - EXTERNAL
 Exercise – OMA measurement (4..4)
12. Export from the Project Browser the geometry and recordings as individual recordings in UFF file format
• If all recordings are exported together, you must divide them into Test Setups (Measurements) in OMA
• Binary UFF instead of ASCII UFF can be used to reduce file size

13. Open OMA

• Create ‘New Project’ using ‘Internal Storage’ and ‘Universal File Format’
• Import the geometry and the 3 recordings (the recordings can be imported simultaneously)
• Perform signal processing using default parameters
• Do FDD by manual peak picking
• Animate results

56
CONFIDENTIAL - EXTERNAL
 Projection Channels

57
CONFIDENTIAL - EXTERNAL
 Basic Idea
 Reducing redundant information
[Gyy]
• Only a few independent rows/columns exist  G11 G12 G13 G14 . . . G1N 
 
• Many rows/columns are linear combinations of the others G 21 G 22
G31 G32
G 23
.
.
.
. . 
. 
 
• A lot of unnecessary (redundant) information exists G 41 . . . 
 . . . 
 
 . . 
 . . 
 Reduction of linear dependent columns by proper choice of 
GN1 . . . . . .

GNN 
projection channels
• [G(: , [p1 p2])]
 G1p1 G1p2 
• [p1 p2] : projection channels  
G G
 2p1 2p2 
   
 Correlation techniques are used to find the measurement channels  
 Np 1 Np 2 
G G
with m information
N: Number measurement channels

58
CONFIDENTIAL - EXTERNAL
 Determining Proper Projection Channels
 The correlation coefficients of the measured data are used to determine the projection channels

 The correlation coefficient elements and matrix:

 1 C12 C13 C14 . . . C1N 
 
C 21 1 C 23 . . . 
C 31 C 32 . . . 
 

 
C 41 . . . 
E ( y i ( t + )  y j ( t ))2  . . . 
C ij ( ) =
2  

Ey i ( t + )  y i ( t )  E y j ( t + )  y j ( t )   .
 .
. 
. 
 
CN1 . . . . . . 1 

59
CONFIDENTIAL - EXTERNAL
 Projection Channels – Single Data Set
 Find the channel that correlates most with other channels using Correlation Coefficient, Cij
calculations. This channel most likely contains maximum physical information:
N
Wi = C
j=1
ij ji, i=1,…,N

Find max ([W1, W2……..WN])

 Find as many channels needed that correlates the least with all previous found projection
channels. These channels will most likely bring most new information

 If a channel has insignificant correlation with any other channels, it will be disregarded since it
might be ”dead”

60
CONFIDENTIAL - EXTERNAL
 Projection Channels – Multiple Data Set
 All reference channels are chosen as projection channels, because the reference channels
are present in all data sets

 Find as many channels needed that correlates the least with all previous found projection
channels. These channels will most likely bring most new information (as for single data set)

 If a channel has insignificant correlation with any other channels, it will be disregarded since it
might be ”dead” (as for single data set)

61
CONFIDENTIAL - EXTERNAL
 Use of Projection Channels in FDD, EFDD and CFDD
 N: No of measurement channels
 M: No of projection channels

[G yy(: ,[p1 p2…])] = [V]NxM[S]MxM [V]HMxM

• A SVD of the [G yy(: ,[p1 p2…])] matrix is performed

• The mode shapes are found among the M columns of [V]NxM Projection Channels are default shown in purple.
Only 3 automatically selected out 95 DOFs in this case
• [V]HMxM is not used for anything

 The use of projection channels reduces the calculation time

• The optimum number of projection channels can be found automatically by the software or manually
(explicit) selected
˗ It is based on information in the SVD curves

62
CONFIDENTIAL - EXTERNAL
 Explicit Selection of Projection Channels
 Selected in the
• Properties for a Channel in the Data Organizer
• Select Explicitly column in the Channels Information table
 Can be selected either before or after the data has been
processed, but after requires you run the Signal Processing again
 Help in manual selection of Projection Channels
• For each Data Set investigate the Coherence between the channels
 References will always be selected as Projection Channels
 Select ‘No. of Channels’ in Signal Processing configuration

63
CONFIDENTIAL - EXTERNAL
 Summary – Projection Channels
 Advantages of using projection channels
• Increased computational speed and less memory usage
• Reduced redundant information
˗ Removal of unimportant data (general)
˗ Better stabilization of physical modes (SSI)

 Disadvantages of using projection channels

• Missing modes, if number of projection channels is too small
• Missing modes, if they do not appear in projection channels

 Use SVD curves to find optimum number of projection channels, M

• Or let the OMA software do it automatically

 Harmonic detection using Fast Kurtosis Checking requires use of projection channels

 Projection Channels are sometimes referred to as Reference Channels, but there is often
more Projection Channels than the number of Reference Channels
64
CONFIDENTIAL - EXTERNAL
65
CONFIDENTIAL - EXTERNAL
 Q&A – Projection Channels
Q1: What is the main purpose of using projection channels?
a) Use of projection channels reduces redundant information thereby increasing computational speed,
and lowering memory usage
b) Use of projection channels generally improves the OMA results
c) Projection channels can be used instead of bad reference channels for the OMA algorithms

66
CONFIDENTIAL - EXTERNAL
 Frequency Domain Techniques

67
CONFIDENTIAL - EXTERNAL
 Contents
 Proof of the FDD techniques
 Frequency Domain Decomposition (FDD)
 Enhanced Frequency Domain Decomposition (EFDD)
 Harmonic Detection and Reduction
 Curve-fit Frequency Domain Decomposition (CFDD)
 Automatic Mode Estimation

68
CONFIDENTIAL - EXTERNAL
 Frequency Domain Decomposition (FDD)
 Determination of Modal Model from Responses only

FDD procedure:
 Cross Spectral Matrix estimation for each data set, [Gyy()]
 Singular Value Decomposition (SVD) of Matrices
 Identification of Single Degree of Freedom (SDOF) models from SVD curves
 Modal Parameter identification from SDOF models

SVD

69
CONFIDENTIAL - EXTERNAL
 Proof of FDD Techniques
Let’s consider a linear, time invariant dynamic system.
The response spectrum can be expressed as a product:
Y() = H() F()
post-multiplying by its Hermitian (Complex Conjugate and Transposed)
Y()H = F()H H()H
yields:
Y()Y() = H()F()F() H()
 H H H
    
[ G YY ( )] [ GFF ( )]

performing averaging:
GYY () = H() GFF () H()H
where [GYY] and [GFF] are output and input cross-spectra matrices, respectively.
70
CONFIDENTIAL - EXTERNAL
 Proof of FDD Techniques
Let’s make the following assumptions about the input forces {F}
• The forces are un- or weakly correlated
• Their frequency spectra are broadband and smooth
• The forces act all over the entire structure {F}

Then
 G11 G12 ... G1n 
 G G22 ... G2n 
[GFF ] = {F} {F} =
H  21
 [I]
 ... ... ... ... 
 
Gn1 Gn2 ... Gnn 

71
CONFIDENTIAL - EXTERNAL
 Proof of FDD Techniques
Assuming

GFF ()   I 
then

GYY () = H() GFF () H()H GYY ()  H() I  H()H

Conclusion
Under the aforementioned assumptions about the forces, the output cross-spectra matrix [GYY]
holds all information about the dynamic properties of the system

The goal of OMA is to extract the modal parameters

from measured dynamic responses of the structure
72
CONFIDENTIAL - EXTERNAL
 Modal Behaviour
 A mode is described by its Modal Parameters
• Natural Frequency and Damping (Eigenvalue)
• Mode Shape (Eigenvector)

 The dynamic response y of a structure can be decomposed into a set of linear independent
modes  and modal coordinates qx(t)

= + + ++

y(t) = q1(t)1 + q2(t)2 + q3(t)3+    + qn(t) n

m
y =  q =  r qr
r =1
73
CONFIDENTIAL - EXTERNAL
 Frequency Domain Decomposition (FDD)
Singular Value Decomposition of Response Cross-spectrum Matrix
n
R = V  S  V  H
=  vi si viH = s1v1v1H + s2 v2 v2H +  + sn vn vnH
i =1
 SVD performed for each frequency, giving response spectra of modes
 Frequency at peak found from decoupled modes
• Frequency resolution no better than line spacing
 The singular vectors approximates the mode shapes
 No damping estimated [dB] Singular Values of Pow er Spectral Density
S1: At least one
-27 mode exists S2: At least two modes exist

-36 Mode 1

-45
Mode 2

-54

-63

-72

Gyy Mag. Example -81

SVD of [Gyy] Matrix Decoupled Modes
0 5 10 15 20 25
Frequency [Hz]
74
CONFIDENTIAL - EXTERNAL
 Demo – FDD and EFDD of a 34m Wind Turbine Blade

75
CONFIDENTIAL - EXTERNAL
76
CONFIDENTIAL - EXTERNAL
 Q&A – Frequency Domain Decomposition (FDD)
Q1: What is FDD?
a) A non-parametric technique
b) A parametric technique
c) A peak-picking technique, where the modes are picked from cross-spectra
d) A peak-picking technique, where the modes are picked from SVD curves

Q2: Which modal parameter is not estimated using FDD?

a) Natural frequency
b) Damping ratio
c) Mode shape

77
CONFIDENTIAL - EXTERNAL
 Enhanced Frequency Domain Decomposition (EFDD)

78
CONFIDENTIAL - EXTERNAL
 Enhanced FDD (EFDD)
 SDOF function created using a
H
( 0  i )2
MAC = H H
 0,8
MAC Rejection Level (default 0.8): ( 0  0 )  ( i  i )

 Improved natural frequency and damping included from correlation functions

 Mode shape estimation from weighted sum of singular vectors:  weight =   i  si
i
SVD of Spectral Density Matrices Normalized Correlation Function

Validation of Natural Frequency Estimate Validation of Damping Ratio Estimate

79
CONFIDENTIAL - EXTERNAL
 Demo – FDD and EFDD of a 34m Wind Turbine Blade

80
CONFIDENTIAL - EXTERNAL
 FDD vs. EFDD – Advantages and Disadvantages
 Advantages of FDD
• Easy and fast to use – snap feature to peaks
˗ Often used for initial investigation of resonances
• Picking at any Frequency and any Singular Value is possible
˗ Better understanding of structural behaviour

 Disadvantage of FDD
• Frequency resolution no better than linespacing
• No damping estimated

 Advantages of EFDD
• Damping estimated
• Improved estimation of natural frequencies and mode shapes
 Disadvantage of EFDD
• Distinguished peaks required in all datasets for all modes
• Fine tuning of MAC rejction level, min/max correlation may be required for each resonance in each dataset

81
CONFIDENTIAL - EXTERNAL
82
CONFIDENTIAL - EXTERNAL
 Q&A – Enhanced Frequency Domain Decomposition (EFDD)
Q1: What does EFDD add to the FDD technique?
a) Faster estimation of modes
b) Estimation of damping
c) Improved estimation of natural frequencies and mode shapes

83
CONFIDENTIAL - EXTERNAL
 Harmonic Detection and Reduction

84
CONFIDENTIAL - EXTERNAL
 Consequences of Harmonic Components
Techniques Consequences

All techniques ⚫ Potentially mistaken for being structural modes

⚫ Potentially bias the estimation of structural modes (freq, damp, shape)
⚫ Potentially higher dynamic range required to extract “weak” modes

FDD ⚫ The picked FFT line may be biased by the deterministic signal(s)
⚫ Deterministic signals must be away from the resonances
(no damping estimated)
EFDD ⚫ The identified SDOF used for modal parameter estimation may be
biased by deterministic signal(s)
⚫ Deterministic signals must be outside the SDOF function thereby potentially
narrowing the SDOF and resulting in poorer identification (leakage)
SSI ⚫ The SSI methods will (theoretically) estimate both deterministic signals and
(PC, UPC, CVA) modes – even for deterministic signals very close to the modes
⚫ Information in the time signal is used to extract both deterministic signals
and modes, therefore the recording time must generally be longer
⚫ Higher model orders are generally required

Note:
Harmonic components cannot, in general, be removed by simple filtering
85
CONFIDENTIAL - EXTERNAL
 EFDD – Harmonic Components
Aluminium plate. Mode: 354 Hz; Harmonic: 374 Hz

SVD of Spectral Density Matrices SVD of Spectral Density Matrices

Normalized Correlation Function Normalized Correlation Function

86
CONFIDENTIAL - EXTERNAL
 EFDD – Harmonic Components
Aluminium plate. Mode: 354 Hz; Harmonic: 374 Hz

Validation of Natural Frequency Estimate Validation of Natural Frequency Estimate

Validation of Damping Ratio Estimate Validation of Damping Ratio Estimate

87
CONFIDENTIAL - EXTERNAL
 Harmonic Detection and Reduction – Frequency Domain
Reducing harmonic components (deterministic signals) in EFDD and CFDD techniques
 Harmonic components indicated in SVD plots using kurtosis calculations
 Harmonic components removed in SDOF model before extracting the modal parameters

=> Makes OMA much more applicable for operating mechanical structures,
where the excitation are subject to deterministic forces due to the rotating parts

Harmonic Detection Harmonic Reduction

88
CONFIDENTIAL - EXTERNAL
 Probability Density Function - Background
 Probability Density Function (PDF)
– The statistical properties of a narrowband stochastic response of a structural mode (random signal)
and a harmonic component are significantly different

Gaussian PDF Sinusoidal PDF

− ( x −  )2

y = f (x  ,  ) =
1 y = f (x a ) = ( cos(arcsin (x a )))
−1
2 2
e
 2
However, not a robust technique for cases with harmonics and closely-coupled modes
89
CONFIDENTIAL - EXTERNAL
 Kurtosis - Definition
The kurtosis γ* of a stochastic variable x provides a measure of
how peaked or how flat the probability density function of x is

 (x  ,  ) =
 E (x −  )
4

 4
*i.e. 4th normalized moment

 > 3: Leptokurtic (”Peaked” and ”fat tails”)

 = 3: Mesokurtic (Normal Distribution)
 < 3: Platykurtic (Less ”peaked” and ”thinner tails”)

 = 1.5 for a sinusoidal component

Note: Often kurtosis is calculated as  - 3

90
CONFIDENTIAL - EXTERNAL
 Harmonic Detection - Extended Kurtosis Checking
The kurtosis γ of a stochastic variable x provides a measure of
how peaked or how flat the probability density function of x is

 (x  ,  ) =
 E (x −  )
4

 4

Automatic Detection of Harmonic Components

1. Each measurement channel yi is normalized to unit variance and zero mean (scaling)
2. For all frequencies fj between DC and the Nyquist frequency a narrow bandpass filtering of yi
around fj is performed
3. The Kurtosis γj,i for the filtered signal yi around fj is calculated
4. For each frequency, the mean of the Kurtosis γj is calculated across the measurement channels
5. The median m of the Kurtosis of all frequencies is calculated. If the signal is purely Gaussian
distributed this robust measure for the mean will be 3.
6. For each frequency fj the deviation of the Kurtosis γj from the median m is calculated. If γj deviates
significantly from m, then the distribution around fj is different than for most frequencies. Hence γj
can be denoted as an outlier (harmonic) and should not be included in the estimation of the SDOF
function.

91
CONFIDENTIAL - EXTERNAL
 Fast Kurtosis Checking - Overview
A fast supplement to the extended kurtosis check method by only
checking potential deterministic signals in the projection channels
and only at frequencies, where peaks (abrupt changes) are
repeated in multiple SVD curves
Assumptions:

⚫ At least 4 - 6 projection channels should be used

➢ Include channels until lowest singular value curve is approx. horizontal

⚫ The frequency resolution should be sufficient high to reveal abrupt changes

Which method to use?

⚫ The Fast Kurtosis Check method is the Default method

⚫ The Extended Kurtosis Check method should be used,
when deterministic signals are not easily visible in the SVD curves

92
CONFIDENTIAL - EXTERNAL
 EFDD - Removal of Harmonic Components

Harmonic components are indicated by vertical lines

Harmonic components are removed by linear interpolation
Frequencies removed are not included in weighted mode shape estimation

Mode at 354 Hz; Harmonic Component at 374 Hz Mode and Harmonic Component at 354 Hz

Damping estimated too high.

Harmonic Component eliminated! Use higher frequency resolution
and/or polynomial fit!

93
CONFIDENTIAL - EXTERNAL
94
CONFIDENTIAL - EXTERNAL
 Q&A – Harmonic Detection and Reduction
Q1: Why is kurtosis used instead of probability density functions for detection of harmonics?
a) Simpler and faster calculation
b) More robust to cases with harmonics and closely-coupled modes

95
CONFIDENTIAL - EXTERNAL
 Curve-fit Frequency Domain Decomposition (CFDD)

96
CONFIDENTIAL - EXTERNAL
 Curve-fit Frequency Domain Decomposition (CFDD)
 Curve-fitting the SDOF function to smoothen out SDOF errors caused by e.g. harmonics removal
EFDD

Harmonics found
from Kurtosis
calculations

E (x −  )
 (x  ,  ) =
4

 4

CFDD

Significant improvement
but more powerful time
domain techniques exist

97
CONFIDENTIAL - EXTERNAL
 Curve-fit Frequency Domain Decomposition (CFDD)
G yy ( f ) = H ( f ) Gxx ( f ) H H ( f ) B( f ) B0 + B1e 2fT + B2 e 4fT
Response Matrix: FRF of SDOF system: H( f )= =
A( f ) 1 + A1e 2fT + A2 e 4fT
 H( f ) H H ( f )
Algorithm
1. Estimate SDOF spectrum G(f) - Remove harmonic components by linear interpolation
2. Calculate positive half-power spectrum P(f) of G(f)
 A1 
A 
 2
(1+ A1e 2fT + A2 e 4fT ) P( f ) = B0 + B1e 2fT + B2 e 4fT => − P( f )e 2fT
− P( f )e 4fT 1 e 2fT e 4fT   B0  = P( f ) => Ac  = Bc
 
3. Construct the following matrices:  B1 
 B2 

 − P( f 0 )e 2f 0T − P ( f 0 )e 4f 0T 1 e 2f 0T e 4f 0T   P( f 0 ) 

 2f1T 4f1T   P( f ) 
 − P ( f1 )e − P( f1 )e 4f1T 1 e 2f1T e   1 

Ac =  . . . . .  , Bc =  . 
   
 . . . . .   . 
− P( f )e 2f T − P ( f )e 4f T 1 e 2f T e 4f T   P( f )
 

−1
Re( Ac ) Re(Bc )
4. Solve the regression problem: ˆ =    
giving the estimates:  = A1 A2 B0 B1 B2 
T

Im( Ac )   Im(Bc ) 

5. Frequency and damping estimates are obtained from the roots of A(f)
6. Mode shape estimate is obtained from weighted sum of singular vectors:  weight =   i  si
i
98
CONFIDENTIAL - EXTERNAL
 Demo – Removal of Harmonic Components – Ship Structure

Harmonic Components with 2.05 Hz interval Internal Excitation

• Fundamental frequency of main engine at
2.05 Hz
• 4-bladed propeller fundamental frequency
at 8.2 Hz

Test
• 90 min; 16 channels; fs = 128 Hz
• 3 projection channels
• Extended Kurtosis Checking

2nd Mode: 2.03 Hz; Harmonic Components with 2.05 Hz intervals • Curve-fitted FDD, CFDD
Upper display: Before reduction; Lower display: After reduction
Acknowledgements:
Sven-Erik Rosenow, Santiago Uhlenbrock and Günther Schlottmann,
99 University of Rostock, Faculty of Mechanical Engineering, Germany
for providing the raw measurement data
CONFIDENTIAL - EXTERNAL
100
CONFIDENTIAL - EXTERNAL
 Q&A – Curve-fit FDD
Q1: When is it advantageous to use the Curve-fit FDD technique?
a) To use half-power spectra (parametric technique) instead of SVD curves (non-parametric technique)
to extract the modal parameters
b) When harmonic components are close to resonance frequencies to recreate the SDOF
functions which are affected by removing the harmonics before extracting the modal parameters

101
CONFIDENTIAL - EXTERNAL
 Automatic Mode Estimation

102
CONFIDENTIAL - EXTERNAL
 Advantages of doing Automatic Mode Estimation
 Simplifies and speeds up the modal analysis task

 Reduces human errors

 Demands less modal analysis skills

 Allows for unattended testing – Structural Health Monitoring

• Using the SSI techniques

103
CONFIDENTIAL - EXTERNAL
 (C/E)FDD Automatic Mode Estimation - Overview
 Modal Coherence (light blue graph with inverse fill)
• Defines whether you have a Modal Domain (High) or a Noise Domain (Low)

 Modal Domain (light green graph)

• A property defined for all modes
• Defines a frequency region dominated by a mode

 Excludes deterministic signals (harmonics) from identification if detected (vertical black lines)

104
CONFIDENTIAL - EXTERNAL
 Indicator Function – Modal Coherence
N
1
Discriminator function: d1 ( f 0 ) =
N
u ( f
n =1
1 n ) H u1 ( f 0 ) , N: No. of Freq. Lines Avg.

Number of frequency lines, N includes a moving average of singular vectors, u for the line in question.
Provides a more noise free estimate. For closely coupled modes, N should be small to avoid smearing.
If the frequency resolution is high (Δf small), N can be higher.

0.8

Light blue graph with inverse fill

0

105
CONFIDENTIAL - EXTERNAL
 Indicator Function – Modal Domain
For areas, where the Modal Coherence (d1(f0)) is higher than a certain user-
defined threshold, a Modal Domain can be established around peaks.
Peaks must fulfill user-defined criteria to be qualified as a mode.
Deterministic signals are automatically excluded (option).

106
CONFIDENTIAL - EXTERNAL
 Automatic Mode Estimation - Properties
Automatic Mode Estimation Settings
 Immediate Update
• After each parameter change or when box checked
 Dynamic Headroom [dB]
• The algorithm stops when modes are attenuated more than the specified value compared to the highest one
 Maximum Modes to Estimate
• The algorithm stops when the number of modes specified is found
 Test Mode Damping
• Enable/disable estimation of mode damping using the spectral half power method
 Maximum Damping [%]
• Only modes with a damping value equal to or less than the specified value are extracted
 MAC Rejection Level
• Used to define the SDOF function for EFDD and CFDD analysis
 Number of Modal Correlation Lags
• Specifies how many neighbouring frequencies to average when calculating the Modal Coherence
 Enable Post Cross MAC Analysis
• Post-analysis removing ”duplicate” modes with a high correlation
• Compares a mode to another one. If the MAC between them is lower than the specified threshold,
they are two separate modes, and both should be kept.
• The lower the threshold, the fewer modes are kept

Diagram Options
 Show Modal Coherence and Modal Domain, if checked
107
CONFIDENTIAL - EXTERNAL
 Summary – Frequency Domain Decomposition Techniques
 Frequency Domain Decomposition (FDD)
• Each mode based on just the picked frequency
˗ Natural frequency as good as FFT resolution
˗ Mode shape estimation dependent on FFT resolution
˗ No damping estimates
 Enhanced FDD (EFDD)
• Each mode based on an SDOF function around the picked frequency
˗ Improved natural frequency and mode shape estimation
˗ Accurate damping estimate calculated
 Curve-fit FDD (CFDD)
• A curve-fit is made around the SDOF function found in the EFDD technique
˗ Improved removal of deterministic signals (harmonic components)
• The (E/C)FDD techniques are intuitive, easy-to-use and can deal with harmonics
• Automatic Mode Estimation is possible, but not as powerful as for the SSI techniques
• The FDD technique is for first investigations. The EFDD and CFDD techniques provide very accurate results!
• However, they are all limited to lightly and medium damped structures
• FRF-based techniques exist but might have issue with harmonics and leakage!
108
CONFIDENTIAL - EXTERNAL
 Demo – Automated Mode Estimation
 Example data – Plate with Harmonics

109
CONFIDENTIAL - EXTERNAL
110
CONFIDENTIAL - EXTERNAL
 Analysis Validation

111
CONFIDENTIAL - EXTERNAL
 Mode Table
 Natural frequency, damping ratio, mode shape complexity, normalization, man/auto approach etc.
 Shape Details

Harmonics can be
detected as ‘modes’
with low damping ratio

112
CONFIDENTIAL - EXTERNAL
 Animation
 Does it look like a mode shape? Global or local mode?

Mode at 354Hz (left); Harmonic at 374Hz (right)

113
CONFIDENTIAL - EXTERNAL
 Complexity Plots – Modal Complexity Factor (MCF)

(𝑆𝑥𝑥 − 𝑆𝑦𝑦 )2 + 4𝑆 2 𝑥𝑦
𝑀𝐶𝐹𝑟 = 1 − r: mode number
(𝑆𝑥𝑥 + 𝑆𝑦𝑦 )2

Sxx = Rer Rer Syy = Imr Imr Sxy = Rer Imr

T T T

0  MCF  1 0: Real mode (normal mode)

Modal Phase Collinearity: MPCr = 1 – MCFr

Reasons for complex modes include

• Short and/or noisy measurements
• Data inconsistency e.g., due to changing environmental conditions or varying mass loading
• Insufficient dynamic range
• Poor modal parameter estimation
• Non-proportional damping and non-linearities
114
CONFIDENTIAL - EXTERNAL
 Complexity Plots

 To check the mode shape complexity

 To normalize the modes
• For correlation with FEA modes which are real
• For easier identification of mode

 Normalization approaches Original Phase Rotation

• Original
• Phase Rotation (to compensate for a potential offset)
• Normalization w.r.t. Real or Imag. Axis using Normalization Angle

Real Axis Imaginary Axis

115
CONFIDENTIAL - EXTERNAL
 Modal Assurance Criterion (MAC)

A measure of the consistency (degree of linearity) between

estimates of mode shape vectors

cr  dr T 2
*

MAC cdr =
crT  crdrT  dr
* *

cr dr : Estimates of mode shape r from model c and d, respectively

c = d: AutoMAC c ≠ d: CrossMAC r: mode number

Consistency between
• Different experimentally determined mode shapes
– Different curve-fitting algorithms (e.g., FDD versus SSI); Different projection channels (columns);
Experimentally and analytically determined mode shapes

High MAC between modal vectors can be caused by

• Insufficient spatial sampling of the modes. Not sufficient number of DOFs in the measurements
• An estimated modal vector is caused by a forced vibration (that is, an ODS) and not a mode
116
CONFIDENTIAL - EXTERNAL
 MAC plots and tables
 Check of how linear related shapes are

117
CONFIDENTIAL - EXTERNAL
118
CONFIDENTIAL - EXTERNAL
 Walk-through the OMA sw
Example data: Plate with Harmonics

119
CONFIDENTIAL - EXTERNAL
 Exercises
Frequency Domain Decomposition Techniques

120
CONFIDENTIAL - EXTERNAL
 Where to find the Examples Files
All exercises are based on example files included in the PULSE OMA installation.
 Select ”Example Files” in the ”Welcome to PULSE Operational Modal Analysis” dialog that
appears, when you open PULSE OMA:

121
CONFIDENTIAL - EXTERNAL
 Where to find the Examples Files
 Or in the Application menu:

122
CONFIDENTIAL - EXTERNAL
 Exercise 1 – Plate with Harmonics (Single Data Set) (1..2)
Analysis\Prepare Data
 Signal Processing
• Harmonic Detection should be disabled (Default)
• Disable Stochastic Subspace Identification

 Data Processing (Measurement Validation)

• SVD plots, Auto and Cross Spectra, Coherence, Contour Plots (Short Time Fourier Transform) etc.

 Analysis\Estimation
• FDD, EFDD, CFDD
˗ Manual and Automatic Mode Estimation
˗ EFDD SDOF validation (MAC Rejection: Correlation, Damping, Natural Frequency)
˗ MAC plots and tables; Complexity plots; Frequency vs Damping Diagram

 Analysis\Validation
• Compare modes from a single estimation method (e.g. EFDD) and between estimation methods (e.g. EFDD and CFDD)
˗ Animation (Overlaid, Difference, Top-Bottom, Side-by-Side); Complexity Plots; MAC plots and tables

123
CONFIDENTIAL - EXTERNAL
 Exercise 1 – Plate with Harmonics (Single Data Set) (2..2)
Analysis\Prepare Data
 Signal Processing
• Perform Fast Kurtosis Checking
˗ Investigate the Kurtosis Spectrum and the effect of changing the Threshold in the Harmonic Settings properties
• Optional: Perform Extended Kurtosis Checking
˗ Investigate the Kurtosis Spectrum and the effect of changing the Threshold in the Harmonic Settings properties

 Analysis\Estimation
• Perform FDD, EFDD and CFDD

124
CONFIDENTIAL - EXTERNAL
 Exercise 2 – Steel Plate (Two Data Sets; Closely-coupled Modes)
Analysis\Prepare Data
 Signal Processing
• Harmonic Detection should be disabled (Default)
• Disable Stochastic Subspace Identification

 Data Processing (Measurement Validation) for the Two Data Sets

• SVD plots, Auto and Cross Spectra, Coherence and Contour Plots (Short Time Fourier Transform)

 Analysis\Estimation
• Perform FDD, EFDD and CFDD

125
CONFIDENTIAL - EXTERNAL
 Additional Exercises

 Time and Frequency ODS in Analysis\Estimation task

 Geometry Creation in Setup\Prepare Geometry task

 Edit Times Series in the Setup\Manage Measurement task

• Time range; Single and Double Integration

 Report Generation

126
CONFIDENTIAL - EXTERNAL
127
CONFIDENTIAL - EXTERNAL
 End of Day 1

128
CONFIDENTIAL - EXTERNAL
Operational Modal Analysis Course
INCLUDING INTRODUCTION TO STRUCTURAL HEALTH MONITORING

Day 2
Niels-Jørgen Jacobsen
October 6, 2023

www.hbkworld.com | © HBK – Hottinger, Brüel & Kjær | All rights reserved

CONFIDENTIAL - EXTERNAL
 Agenda Day 1 – October 5th, 2023 – Morning – Recap

Welcome, objectives and agenda

Introduction to Operational Modal Analysis (OMA)
- Basic idea
- Applications of OMA
- Advantages, assumptions and limitations
- Data acquisition (measurement)
- Overview of modal parameter identification techniques for OMA
- Comparison of OMA to Experimental Modal Analysis (EMA)
Coffee break
Measurement example using BK Connect
Measurement validation tools
- Singular Value Decomposition (SVD) plots
- Contour plots
Exercises - Measurement and measurement validation
Lunch break

130
CONFIDENTIAL - EXTERNAL
 Agenda Day 1 – October 5th, 2023 – Afternoon – Recap

Projection channels
Frequency domain techniques
- Frequency Domain Decomposition (FDD)
- Enhanced FDD (EFDD)
- Curve-fit FDD (CFDD)
- Consequences of harmonic components
- Detection and reduction of harmonic components in the frequency domain
- Automatic mode estimation
Coffee break
Analysis Validation tools
- Complexity plots and mode normalization
- MAC plots and tables
- Comparison of different estimation techniques
Exercises - Frequency Domain Decomposition Techniques
Q&A

131
CONFIDENTIAL - EXTERNAL
 Agenda Day 2 – October 6th, 2023 – Morning

Recap and questions to yesterday's training

Frequency Response models and State Space models
Time domain techniques
- Stochastic Subspace Identification (SSI) overview
- SSI - Main elements
- SSI techniques (UPC, PC, CVA)
- Analysis validation tools (Response validation, Prediction error)
Coffee break
Time domain techniques cont.
- Uncertainty estimation using UPCX
- Hints and tips using SSI
- Reduction of harmonic components in the time domain
Exercises - Time domain techniques
Lunch break

132
CONFIDENTIAL - EXTERNAL
 Agenda Day 2 – October 6th, 2023 – Afternoon

Exercises - Time domain techniques cont.

Batch processing using SSI
Structural Health Monitoring (SHM)
- Introduction to SHM
- Vibration-based SHM using SSI
Coffee break
Structural Health Monitoring (SHM) cont.
- Damage detection example
Further information
Q&A, Course evaluation

133
CONFIDENTIAL - EXTERNAL
 State Space Models
Modelling of mechanical systems

134
CONFIDENTIAL - EXTERNAL
 Linear Mechanical Systems
Two modelling and analysis approaches:
f(t) x(t)
 Frequency Response Functions

Rijr Rijr* m
Hij () =  Hijr =  +
r r j  − pr j − pr*

k c

 State Space Formulations

x ( t ) = Ax ( t ) + Bu( t )
y( t ) = Cx( t ) + Du( t )

135
CONFIDENTIAL - EXTERNAL
 Equation of Motion – Time Domain
Single Degree of Freedom Model
f(t) x(t)

Newton 2nd law:

m
mx( t ) + cx ( t ) + kx ( t ) = f ( t )

k c

Force Balance: Inertial + Dissipative + Restoring = External

136
CONFIDENTIAL - EXTERNAL
 SDOF Models – Frequency Domain
Single Degree of Freedom Model

f(t)
 Time domain f ( t ) = mx( t ) + cx ( t ) + kx ( t )
x(t)
m

 Frequency domain F() = -2mX() + j cX() + kX()

c k

X() 1
 FRF H() = =
F() − 2m + jc + k

137
CONFIDENTIAL - EXTERNAL
 Frequency Response Function (FRF) for SDOF Model
FRF:
X() 1 f(t)
H() = =
F() −  2 m + jc + k
x(t)
m

On partial fraction form:

c k
1/ m R R*
H( j) = = +
( j − p)( j − p*) j − ( − + jd ) j − ( − − jd )

1 k c2 c
where R= , d = − , =
j2md m 4m2 2m

Rijr Rijr*
MDOF model: Hij () =  Hijr =  +
r r j  − pr j − pr*

138
CONFIDENTIAL - EXTERNAL
 State Space Model of a SDOF System (1..3)
Time domain: f ( t ) = mx( t ) + cx ( t ) + kx ( t ) (1)

f(t)
X() 1
FRF: H() = = (2) m
x(t)
F() − 2m + jc + k

c k

For a second order system we need two states to describe it:

1) Position: x 1 ( t ) = x( t ) (3)
2) Velocity: x 2 ( t ) = x ( t ) (4)

139
CONFIDENTIAL - EXTERNAL
 State Space Model of a SDOF System (2..3)
 The first state equation is (1) written in terms of state variables, their first derivatives
and the input f(t):
mx 2 ( t ) = −cx 2 ( t ) − kx 1( t ) + f ( t ) (5)

 The first derivative is now placed on the left-hand side:

k c 1
x 2 ( t ) = − x1( t ) − x 2 ( t ) + f ( t ) (6)
m m m

x 1( t ) = x 2 ( t ) (second state equation) (7)

 In addition, we need an algebraic output equation describing

the output as a linear combination of states and inputs

y( t ) = x 1 ( t ) (8)

140
CONFIDENTIAL - EXTERNAL
 State Space Model of a SDOF System (3..3)
 State and output equations written in Matrix format:

 x 1   0 1   x1  0 
   = − k c   +  1 f (t) (9)
x 2   m −
m  2 
x m 

 x1 
y = 1 0   (10)
x 2 

 Generic form of state equations:

x ( t ) = Ax ( t ) + Bu( t ) (11)
y( t ) = Cx( t ) + Du( t ) (12)

141
CONFIDENTIAL - EXTERNAL
 State Space Model – Generic Form
 Generic form of state equations:

x ( t ) = Ax ( t ) + Bu( t )

y( t ) = Cx( t ) + Du( t )

 A is the system matrix (natural frequency and damping)

 B is the input matrix (force distribution)

 C is the output matrix (“mode shapes”)

 D matrix represents any direct connection between input and output

(normally zero for mechanical systems)
142
CONFIDENTIAL - EXTERNAL
 x ( t ) = Ax ( t ) + Bu( t )
State Space Model – Generic Form y( t ) = Cx( t ) + Du( t )

[D]

Feed Forward Matrix

+
x ( t ) x( t )
[B] + 1/s [C] +
Input, u + Output, y
Integrator Sum
Control or Sum Output Matrix
input Matrix

[A]

System Matrix

The Feed Forward Matrix, [D] is normally zero for a mechanical (passive) system

143
CONFIDENTIAL - EXTERNAL
 State Space Model – Feed Forward Matrix [D] set to zero

x ( t ) x( t )
[B] + 1/s [C]
Input, u + Output, y
Integrator
Control or Sum Output Matrix
input Matrix

[A]

System Matrix

x ( t ) = Ax ( t ) + Bu( t )

y( t ) = Cx( t ) + Du( t )

144
CONFIDENTIAL - EXTERNAL
 Summary
 Frequency Response model
• Second order differential equation reduced to Two term algebraic equation
Rijr Rijr*
Hij () =  Hijr =  +
r r j  − pr j − pr*
 State Space model
• Second order differential equation reduced to Two first order algebraic equations

k c 1
x 1( t ) = x 2 ( t ) x 2 ( t ) = − x1( t ) − x 2 ( t ) + f ( t )
m m m
• Matrix formulation (efficient computer calculations)

x ( t ) = Ax ( t ) + Bu( t )
y( t ) = Cx( t ) + Du( t )

 Both models describes

• The internal dynamics of systems and the overall input/output relationship
145
CONFIDENTIAL - EXTERNAL
146
CONFIDENTIAL - EXTERNAL
 Q&A – State Space Models
Q1: Why is it beneficial to use State Space models in OMA to model the inherent dynamics of systems?
a) They can be expressed as matrix formulations which significantly improves the calculation
speed when extracting the modal parameters
b) We can avoid using the Frequency Response model which is optimal for EMA but not OMA

147
CONFIDENTIAL - EXTERNAL
 Time Domain Techniques

148
CONFIDENTIAL - EXTERNAL
 Contents
 Stochastic Subspace Identification (SSI) in Brief
• Overview, Classes and User Procedure, Linking, Model Order Selection

 SSI – Main Elements

• Common Input Matrix, SVD Subspace, Model Validation

 Standard SSI Techniques

• UPC, PC and CVA
• Response Validation and Prediction Error Validation
• General advantages of SSI techniques

 Advanced SSI Techniques

• Crystal-Clear SSI (CC-SSI) incl. Automated Mode Estimation
• SSI-UPCX

 Hints and Tips using SSI

 Harmonic Reduction using Orthogonal Projection
149
CONFIDENTIAL - EXTERNAL
 x ( t ) = Ax ( t ) + Bu( t )
Stochastic Subspace Identification (SSI) – Overview y( t ) = Cx( t ) + Du( t )
Combined System A

Stationary
zero mean Measured
Gaussian Responses
White Noise
Excitation Filter Structural System
(linear, time-invariant) C xt →  → yt
wt (linear, time-invariant)

Unknown excitation forces vt

Discrete-time Stochastic State Space Model

State Equation xt +1 = Axt + wt Model of the dynamics of the system
Observation (Output) equation yt = Cxt + vt Model of the output of the system

wt: Process noise – vt: Measurement noise – Model order: Dimension of A

Modes extracted using Crystal Clear Stabilisation Diagrams

150
CONFIDENTIAL - EXTERNAL
 SSI – Classes and User Procedure

Classes of Identification
 Data Driven: Use of raw time data (No data reduction – Most Accurate!) – Used by PULSE OMA
 Covariance Driven: Use of covariance functions (Data reduction – Slightly Faster!)

SSI User Procedure (Manual or Automated)

 Select Identification Class
• Unweighted Principal Components (UPC); Principal Components (PC); Canonical Variate Analysis (CVA)

 Select total max. model order being twice the number of potential modes (structural, harmonics, noise)

 Estimate Modal Parameters from Stabilization diagram

• Select interval of model order candidates (use SVD diagram)
• Estimate models (adjust tolerance criteria)
• Manual: Select the optimal model (use model validation)
• Automated: The software uses all stable modes across models instead of a single model

 Select and link modes across data sets

• Done automatically, when Automated Mode Estimation is used
151
CONFIDENTIAL - EXTERNAL
 SSI – Estimate Modes from Stabilization Diagram – Manual Approach

For each data set a model order is selected.

Results are linked across all data sets, if multiple data sets are used.

152
CONFIDENTIAL - EXTERNAL
 SSI – Linking Results across Data Sets – Manual Approach

Model order selected individually in 4 data sets

4 data sets linked

Linking

153
CONFIDENTIAL - EXTERNAL
 Selecting Proper Model Order for SSI
Final Prediction Error
 Fitting error decreases with increasing model order
 Parameter uncertainty increases with increasing model order

Parameter Uncertainty
Final Prediction Error

Fitting Error

Optimum Choice Model Order

154
CONFIDENTIAL - EXTERNAL
 Demo – Building Model – Manual Mode Estimation

155
CONFIDENTIAL - EXTERNAL
156
CONFIDENTIAL - EXTERNAL
 Outline
 Stochastic Subspace Identification (SSI) in Brief
• Overview, Classes and User Procedure, Linking, Model Order Selection

 SSI – Main Elements

• Common Input Matrix, SVD Subspace, Model Validation

 Standard SSI Techniques

• UPC, PC and CVA
• Response Validation and Prediction Error Validation
• General advantages of SSI techniques

 Advanced SSI Techniques

• Crystal-Clear SSI (CC-SSI) incl. Automated Mode Estimation
• SSI-UPCX

 Hints and Tips using SSI

 Harmonic Reduction using Orthogonal Projection
157
CONFIDENTIAL - EXTERNAL
 Hankel Matrices with Output Data
Matrix with constant (positive sloping) skew-diagonals

 y( 0 ) y(1) y( 2 ) y( 3 ) ... y( j − 1) 
 y(1) y( 2 ) y( 3 ) y( 4 ) ... y( j ) 
 ... ... ... ... ... ... 
 Y0|i −1  Y p   y( i − 1) y( i ) ... ... ... y( i + j − 2) 
Y = = =
 Y f   y( i )
Y     y( i + 1) ... ... ... y( i + j − 1) 
i | 2 i − 1

 y( i + 1) y( i + 2) ... ... ... y( i + j ) 
 ... ... ... ... ... ... 
 y( 2i − 1) y( 2i ) ... ... ... y( 2i + j − 2)

i << j large
i = No of modes Hermann Hankel
2i = State Space Dimension 1839 - 1873
2i+j-2 = No of samples
j = No of samples – State Space Dimension + 2
158
CONFIDENTIAL - EXTERNAL
 SSI – Main Elements
Create Apply Weights Select SVD Estimate A, C
Common Input Matrix W1 Oi W2 Subspace xt +1 = Axt + wt
Oi (UPC, PC, CVA) W1 Oi W2 = U1 S1 V1 yt = Cxt + vt

 s1 0 0 0 . . . 0
0 s2 0 . . . 

0 0 . . .
V1T 
W1OW2 = U1U 2 
 
0
 T  = U 1 S1V1T
. . s .
. .
V2 
. 0
 
. 0 .
. .

 0 . . . . .
0
. 0
 Validate Model
• Stabilization Diagram
• Response Synthesis
Selected Subspace • Prediction Error

State Space Dimension

sp

s5
s3
s4 Validate Analysis
s2 • Auto and CrossMAC
s1
• Complexity Plot …
Singular Value
Subspace shown in stabilization diagram
159
CONFIDENTIAL - EXTERNAL
160
CONFIDENTIAL - EXTERNAL
 Outline
 Stochastic Subspace Identification (SSI) in Brief
• Overview, Classes and User Procedure, Linking, Model Order Selection

 SSI – Main Elements

• Common Input Matrix, SVD Subspace, Model Validation

 Standard SSI Techniques

• UPC, PC and CVA
• Response Validation and Prediction Error Validation
• General advantages of SSI techniques

 Advanced SSI Techniques

• Crystal-Clear SSI (CC-SSI) incl. Automated Mode Estimation
• SSI-UPCX

 Hints and Tips using SSI

 Harmonic Reduction using Orthogonal Projection
161
CONFIDENTIAL - EXTERNAL
 Standard SSI techniques – UPC, PC and CVA
The SSI techniques differ in the weighting of the Common SSI Input Matrix

 Unweighted Principal Components (UPC) Oi

• No weighting applied
• For extracting modes with comparable energy level
• Produces good results using a reasonable small state space dimension
Select SVD
Subspace
 Principal Components (PC) Oi W2 W1 Oi W2 = U1 S1 V1
• Post-multiplication by the W2 matrix
• Application similar to UPC

 Canonical Variate Analysis (CVA) W1 Oi

• Pre-multiplication by the W1 matrix
• For extracting modes with large difference in energy level
• Requires larger state space dimension (model order) to extract weakly-excited modes

The best validation is to compare the results of the different SSI techniques!

162
CONFIDENTIAL - EXTERNAL
 Response Validation and Prediction Error Validation
Shows how good the SSI model represents the measured data

163
CONFIDENTIAL - EXTERNAL
 General Advantages and Disadvantages of SSI
Advantages
 Use stabilization diagrams as known from experimental modal analysis (hammer and shaker testing)
 Suitable for lightly, medium and heavily damped structures
 The most accurate estimates of natural frequency, damping and mode shape
 Unbiased estimation
• No leakage as SSI is a true time domain and data-driven method
˗ No unpredictable overestimation of damping

 No problems with deterministic signals or repeated-roots

• Deterministic signals are extracted as “modes” with very low damping and can be excluded
(Deterministic signals can also be reduced/eliminated beforehand in the time domain)
 Includes uncertainty calculations for the modal parameters and even more accurate parameter
estimation (UPCX)

Disadvantages
• Difficult theory, but easy to use
The SSI techniques are the natural choice for OMA!

164
CONFIDENTIAL - EXTERNAL
165
CONFIDENTIAL - EXTERNAL
 Outline
 Stochastic Subspace Identification (SSI) in Brief
• Overview, Classes and User Procedure, Linking, Model Order Selection

 SSI – Main Elements

• Common Input Matrix, SVD Subspace, Model Validation

 Standard SSI Techniques

• UPC, PC and CVA
• Response Validation and Prediction Error Validation
• General advantages of SSI techniques

 Advanced SSI Techniques

• Crystal-Clear SSI (CC-SSI) incl. Automated Mode Estimation
• SSI-UPCX

 Hints and Tips using SSI

 Harmonic Reduction using Orthogonal Projection
166
CONFIDENTIAL - EXTERNAL
 Crystal Clear SSI – CC-SSI
Crystal Clear Stabilization Diagrams
A maximum number of poles to be estimated is specified
‒ Either automatically by software or manually by user
Remaining poles are considered noise modes and excluded
‒ Their frequencies are forced above the Nyquist and their Damping ratios set larger than 100%

Advantages
 Better overview of the physics
 Consistent damping estimates across models
• Selection of model order becomes less critical
 Very robust in demanding situations
• Heavily damped modes
• Weak modes mixed with dominant deterministic signals
• High mode density, i.e., high number of modes in a narrow frequency range
 Allows for accurate Automatic Mode Estimation by using all stable modes across all model orders
in all data sets
• Ease-of-use; Less skills required; User-independent results; Consistent results
• Easy to automate e.g., for Repetitive Testing or Structural Health Monitoring

167
CONFIDENTIAL - EXTERNAL
 Example – Standard SSI Techniques
Stabilization Diagram using Standard SSI with Highly Damped Data
Extracting modes requires cleaning up of the stabilization diagram by adjusting
stabilization criteria and using decimation, filtering and/or projection channels
- but some modes would still be difficult to extract!

168
CONFIDENTIAL - EXTERNAL
 Crystal Clear SSI Techniques

Stabilization Diagram using CC-SSI with Highly Damped Data

Very clean stabilization diagrams. Modes are automatically identified using default settings!

169
CONFIDENTIAL - EXTERNAL
 The UPCX technique
 UPCX: Unweighted Principal Components eXtended
 Extends the UPC technique with uncertainty calculations for all
modal parameters: Natural frequency, Damping ratio and Mode
shape complexity
• The uncertainties help judging the quality of the modes.
 UPCX is selected in the Stochastic Subspace Identification section in
the Signal Processing form in the Analysis/Prepare Data task
 Calculation of mode shape complexities must be selected in
the UPCX task selecting ”Uncertainty Estimation\Mode Shapes”
• Avoids large project sizes, when not required
• Reduces calculation time if only interest in the uncertainties of the natural
frequencies and damping estimates

170
CONFIDENTIAL - EXTERNAL
 Uncertainty – Stabilization Diagram
 Frequency uncertainty is shown as
confidence intervals

 Frequency, damping and mode shape

complexity uncertainty is shown in
tooltips for each frequency and model

 Coefficient of Variation (CV)

• CV = Standard Deviation / Mean Value
• Modes are shown as stable, if the
Maximum Uncertainties for Frequency CV
and Damping CV are fulfilled
˗ Set in the properties for the UPC-X The UPCX technique
stabilization diagram

171
CONFIDENTIAL - EXTERNAL
 Uncertainty – Natural frequency and damping ratio estimation
 More accurate natural frequency and damping
ratio estimation than from conventional
"mean value" based clustering techniques
• Disregards model orders where the mode
estimates exceed the max. confidence intervals
˗ Modes go from Stable (red) to Noise (brown)
and are not included in the estimation
• Weighs stable modes inversely proportional to
the confidence intervals
˗ Automatic mode estimation becomes more reliable
for e.g., Structural Health Monitoring

172
CONFIDENTIAL - EXTERNAL
 Uncertainty – Modal Alignment View
 The Modal Alignment View is available
for all the SSI modal parameter
identification techniques

 It shows for a mode selected in the

Modes table the natural frequency,
damping ratio and mode shape
complexity as a function of model
order (range is user-defined)

 Includes display of uncertainty

confidence intervals with tooltips,
if the UPCX technique is used

173
CONFIDENTIAL - EXTERNAL
 Uncertainty – Frequency vs. Damping Diagram
 The Frequency vs. Damping Diagram
is available for all the SSI modal
parameter identification techniques

 It shows for a mode selected in the

Modes table the mean value of the
natural frequency, damping ratio and
mode shape complexity

 Includes display of uncertainty

confidence ellipsoids with tooltips,
if the UPCX technique is used

 Confidence Level [%] is set in the

properties
• The smaller the Confidence Level,
the smaller the ellipsoids

174
CONFIDENTIAL - EXTERNAL
 Uncertainty – Modes Tables

 Standard Deviation (Std.) columns added for

• Frequency
• Damping
• Complexity

 Columns hidden when using UPC and the other SSI methods

175
CONFIDENTIAL - EXTERNAL
 Uncertainty – Complexity Plot and Geometry

 For a mode selected in the Modes table

• The Complexity Plot shows ellipsoids for each vector based on the Confidence Level [%] set in the
properties for the Complexity Plot
• An ellipsoid selected in the Complexity Plot is also shown on the animated geometry
˗ Requires nodes are shown on the geometry

176
CONFIDENTIAL - EXTERNAL
 Uncertainty – MAC Plots and Tables

 MAC Plot: The uncertainty is indicated by the width of the 3D bars or cylinders
• Mean value and standard deviation shown as cursor read-out
 MAC Table: Mean value ± Standard deviation

177
CONFIDENTIAL - EXTERNAL
 Example – Fighter Wing – UPC/UPCX and working with SSI
properties

178
CONFIDENTIAL - EXTERNAL
179
CONFIDENTIAL - EXTERNAL
 Outline
 Stochastic Subspace Identification (SSI) in Brief
• Overview, Classes and User Procedure, Linking, Model Order Selection

 SSI – Main Elements

• Common Input Matrix, SVD Subspace, Model Validation

 Standard SSI Techniques

• UPC, PC and CVA
• Response Validation and Prediction Error Validation
• General advantages of SSI techniques

 Advanced SSI Techniques

• Crystal-Clear SSI (CC-SSI) incl. Automated Mode Estimation
• SSI-UPCX

 Hints and Tips using SSI

 Harmonic Reduction using Orthogonal Projection
180
CONFIDENTIAL - EXTERNAL
 Hints & Tips using SSI
 Decimation is often beneficial for extracting lower frequency modes
• Can be beneficial to use FDD to get an ‘overview’ of low frequency modes

 Use of filtering (high-pass, low-pass, band-pass and band-stop) often results in cleaner
stabilization diagrams and at lower State Space Dimensions

 Use of Projection Channels

• Reduces calculation time and memory consumption dramatically
• Reduces the scatter of noise modes in stabilization diagrams significantly giving better stabilization
of the physical modes
• Reduces the required maximum state space dimension

181
CONFIDENTIAL - EXTERNAL
 Use of Projection Channels in SSI – Fighter Wing example
 Reduces the scatter of noise modes in Stabilization Diagram
• Better Stabilization of the physical modes

With 4 projection channels Without projection channels (All 36 response ch. used)

182
CONFIDENTIAL - EXTERNAL
183
CONFIDENTIAL - EXTERNAL
 Outline
 Stochastic Subspace Identification (SSI) in Brief
• Overview, Classes and User Procedure, Linking, Model Order Selection

 SSI – Main Elements

• Common Input Matrix, SVD Subspace, Model Validation

 Standard SSI Techniques

• UPC, PC and CVA
• Response Validation and Prediction Error Validation
• General advantages of SSI techniques

 Advanced SSI Techniques

• Crystal-Clear SSI (CC-SSI) incl. Automated Mode Estimation
• SSI-UPCX

 Hints and Tips using SSI

 Harmonic Reduction using Orthogonal Projection
184
CONFIDENTIAL - EXTERNAL
 Harmonic Reduction using Orthogonal Projection (1..2)
Main advantages:
 Robust to non-stationary harmonic excitation with slightly varying amplitude and frequency
• Machinery not running with stable speed and load producing harmonics varying in amplitude and frequency

 Prior knowledge not required

• Pure post-measurement technique not requiring any prior knowledge of the number of harmonics, their frequencies
and their stability, or that any measurement of the rotational speed of the engines have been made

 Simplified post-processing
• The removal/reduction of harmonics is done in the time domain before performing one or more of the
(E/C)FDD and SSI modal identification techniques. This simplifies the subsequent post-processing
˗ For the (E/C)FDD techniques, the harmonics are already removed from the spectra and simple manual or automatic
peak-picking can be performed to find the modes
˗ For the SSI techniques, the harmonics no longer show up in the stabilization diagrams. Hence, there is no longer a
need to manually discriminate “modes” as true physical modes or harmonics based on damping values

 Removal of modes
• Better estimation of weak modes in presence of strong modes
˗ The strong modes can be removed from the response data before conducting the modal analysis

• Studying the effect on the response spectra by removing different modes

˗ Assumes well-separated modes
185
CONFIDENTIAL - EXTERNAL
 Harmonic Reduction using Orthogonal Projection (2..2)
Basic idea:
1. First predict the response only of the peaks defined by the user
2. Once estimated, the original measurements are orthogonally projected onto the predicted
response of the selected peaks. The projected measurements become “blind” to the peaks,
and as such the peaks are reduced significantly

Steps of the reduction technique:

1. Use Stochastic Subspace Identification (SSI) to estimate a range of state space models as
candidates for calculating the predicted response of the selected peaks
2. Selects the optimal state space model containing the eigenvalues corresponding to the
selected peaks
3. Predicts the modal response of the selected peaks from the state space model
4. Orthogonally projects the original measurements onto the predicted response of the selected
peaks

186
CONFIDENTIAL - EXTERNAL
 Example – Removal of Harmonic Components – Ship Structure

Internal Excitation
• Fundamental frequency of main engine at 2.05 Hz
• 4-bladed propeller fundamental frequency at 8.2 Hz

Test
• 90 min; 16 channels; fs = 128 Hz
• 6 projection channels
• Harmonic components removal using Orthogonal
Projection time domain technique

2nd Mode: 2.03 Hz; Harmonic Components with 2.05 Hz intervals

Upper display: Before reduction; Lower display: After reduction
Acknowledgements:
Sven-Erik Rosenow, Santiago Uhlenbrock and Günther Schlottmann,
University of Rostock, Faculty of Mechanical Engineering, Germany
for providing the raw measurement data
187
CONFIDENTIAL - EXTERNAL
188
CONFIDENTIAL - EXTERNAL
 Q&A – Time Domain Techniques
Q1: What are the unique features that makes data-driven SSI techniques generally recommended as being
the most powerful for OMA?
a) They are MDOF techniques allowing multiple modes to be estimated simultaneously
b) By solely working in the time domain, they provide unbiased estimation thereby eliminating unpredictable
overestimation of damping values
c) Very clean stabilization diagrams can be obtained making automated mode selection very robust

Q2: What is the main difference between the different standard SSI techniques?
a) How the state space model framework is defined
b) How the Common SSI Input Matrix is pre- and post-multiplied by weighting matrices
c) How the validation of the modal model is performed

Q3: What does UPCX add to the UPC technique?

a) Uncertainty calculation on all 3 modal parameters (natural frequency, damping and mode shape complexity)
b) Improved modal parameter results when using automated mode selection
c) Estimation using a lower state space dimension

189
CONFIDENTIAL - EXTERNAL
 Batch Processing for OMA Pro

190
CONFIDENTIAL - EXTERNAL
 Overview
SSI analysis of a large number of time data files and/or multiple analysis
of same or different time data files without user interaction
Benefits
 Increased productivity (unattended operation)
 Ease-of-use (less trained operators required)
 User-independent results (configurations pre-defined)

Applications
 Repetitive testing
• Batch testing of various items
• Testing of same structure under varying operating conditions
• Testing of modified structures
• …
 Structural Health Monitoring
• Trending the same structure over time

191
CONFIDENTIAL - EXTERNAL
 Available Versions and Execution Methods
Versions
 BZ-8527-A-F/N PULSE Batch Processing Option for OMA Pro
 BZ-8527-B-F/N PULSE Batch Processing Option for OMA Pro Academic

The option (ARTXBAT.exe) can be executed from a

 Command prompt
 .bat file
 VBscript file

 The option can be executed whether OMA is closed, open or being used

192
CONFIDENTIAL - EXTERNAL
 Syntax (1..2)
 ”Directory for ARTBAT.exe\ARTXBAT.exe” “/CFG: Directory for cfg file\SomeConfigFile.cfg”
“/UFF:Directory for uff file\SomeResultsFile.uff”

Example:
 "C:\Program Files\Bruel and Kjaer\PULSE Operational Modal Analysis\Bin\ARTXBAT.exe"
"/CFG:C:\NJJ\WebEx Batch Processing\Multiple Measurements\Configuration Files\Plate -
354Hz.cfg" "/UFF:C:\NJJ\WebEx Batch Processing\Multiple Measurements\Results Files\Modes -
354Hz.uff"

 If the directories are not explicitly specified, the CFG configuration file must be located in the
same directory as the executable and the UFF result files will be written in the same directory as
well

193
CONFIDENTIAL - EXTERNAL
 Syntax (2..2)
 ”Directory for ARTBAT.exe\ARTXBAT.exe” “/CFG: Directory for cfg file\SomeConfigFile.cfg”
“/UFF:Directory for uff file\SomeResultsFile.uff”

 The argument /CFG is mandatory and specifies the Configuration File

• Sample interval (in seconds)
• Geometry definition (nodes, lines and surfaces)
• Test Setup (measurement file, DOFs)
• Equations (slave nodes)
• Signal processing configuration
• Modal indicator settings
 The /UFF option specifies where to save a Universal File Format results file

194
CONFIDENTIAL - EXTERNAL
 Configuration File
The signal processing configuration as well as the modal indicator settings used for the SSI analysis in the
program, are all specified in the Configuration File in a new section starting with the keyword Processing
Processing
 Detrend
• Detrending (Enable: 1[On]/0[Off])
 Decimate
• Decimation (Degree: 1,2,3,4,5,10,20,30,40,50,100)
 Filter
• Filtering (Enable: 1[On]/0[Off]), (Type:1[Lowpas]/2[HighPass]/3[BandPass]/4[BandStop], (Order: 1,2,..), (LowCut: [0 to 1]), (HighCut: [0 to 1])
 Projection
• Projection Channels (Enable: 1[On]/0[Off]), (Number of Channels: 1,2,3,4......)
 Harmonics
• Harmonic Detection (Enable: 1[On]/0[Off]), (Fast Checking: 1[On]/0[Off])
 Spectrum
• Spectral Density: (Number of Frequency Lines: 2,4,8,16,32,64,128,256,512,1024,2048,4096,......)
 SSI
• SSI Configuration: (Enable: 1[On]/0[Off]), (Max. Dimension: [1 to 200], (Max. Poles: -1[Disable CC]/[1 to Max. Dimension]), (Estimate Start: [1 to Estimate Stop]), (Estimate
Stop: [Estimate Start to Max. Dimension]), (Estimate Inc.: [1 to Max. Dimansion])
 Estimators
• (Enable SSI-UPC: 1[On]/0[Off]), (Enable SSI-PC: 1[On]/0[Off]), (Enable SSI-CVA: 1[On]/0[Off])
 Stability
• (Enable: 1[On]/0[Off]),(Max.Freq.Dev.),(Max.Damp.Dev.),(Max.ModeShape.Dev.),(Max.InitModalAmp.Dev.),(Min.PhysicalDamp.),(Max.PhysicalDamp.)

195
CONFIDENTIAL - EXTERNAL
 Demo – Batch Processing
Examples using measurements from the B&K demo
plate with and without harmonic components

Example 1: Single Measurement using three different SSI techniques and different directories
 Using SSI-UPC, SSI-PC and SSI-CVA on binary data
• Plate with harmonic at 374Hz
 Different directories for
• Configuration files, Measurement files and Results files (UFF)

Example 2: Multiple Measurements using single SSI method

 Using SSI-PC for three different ASCII files
• Plate without harmonics; Plate with harmonic at 354Hz; Plate with Harmonic at 374Hz
 Different directories for
• Configuration files, Measurement files and Results files (UFF)

196
CONFIDENTIAL - EXTERNAL
 Exercises
Stochastic Subspace Identification (SSI) Techniques

197
CONFIDENTIAL - EXTERNAL
 Exercises using Example Data
 Manual mode estimation using SSI-UPC – Building Model
• Manual linking of two measurements

 Automated mode estimation using SSI-UPC – Building Model

• Investigate effect of using Crystal Clear SSI with Estimation Approach: Disabled, Manual and Automatic
• Investigate effect of changing Modal Indicators

 SSI-UPCX – Fighter Wing

• Investigate uncertainties in Stabilization Diagram, Modal Alignment View, Mode Table, Frequency vs
Damping Diagram, Complexity Plot, Animation, MAC plot and table
• Investigate effect of changing Modal Indicators

 Harmonic Reduction using Orthogonal Projection – Plate with Harmonics

• Followed by EFDD and SSI-UPC(X)

 Batch Processing – Plate with Harmonics

• Different SSI techniques used on the same data
198
CONFIDENTIAL - EXTERNAL
 Structural Health Monitoring (SHM)

Vibration-based SHM using SSI

199
CONFIDENTIAL - EXTERNAL
 Experience
 Your experience with Structural Health Monitoring

1 2 3 4 5

• If 3 or higher:
˗ Which techniques have you used?
˗ Which software/tools have you used?

200
CONFIDENTIAL - EXTERNAL
 Content
 Introduction to SHM
• Definition
• Benefits and solutions
• Applications
 Vibration-based SHM using SSI
• Introduction
• Basic approach
• Damage indicators
 Damage detection example – Dogna Bridge, Italy

201
CONFIDENTIAL - EXTERNAL
 Definition of Structural Health Monitoring (SHM)
Structural Health Monitoring (SHM)
The process of implementing a damage detection and characterization strategy for engineering structures

 Damage is defined as changes to the material and/or geometric properties of a structural system, including changes
to the boundary conditions and system connectivity, which adversely affect the system's performance
 The SHM process involves the observation of a system over time using periodically sampled dynamic response
measurements from an array of sensors, the extraction of damage-sensitive features from these measurements,
and the statistical analysis of these features to determine the current state of system health
 For long term SHM, the output of this process is periodically updated information regarding the ability of the structure to
perform its intended function in light of the inevitable aging and degradation resulting from operational environments
 After extreme events, such as earthquakes or blast loading, SHM is used for rapid condition screening and aims to
provide, in near real time, reliable information regarding the integrity of the structure
 Damage classification in order of complexity: Detection, localization, assessment, and type

In short, monitor the structural state and determine the required health management of the structure

202
CONFIDENTIAL - EXTERNAL
 SHM – Benefits and solutions
Benefits of Solutions
condition-based damage detection include  Structures with permanently installed transducers and
 Avoidance of catastrophic failures distributed data acquisition systems with wireless transmission
 Customized structural inspections  Automatic uploading and processing of measurement
files at remote locations for fast damage notifications
 Optimized maintenance services
 Optimal use of structures
 Minimized downtime
 Extended lifetime of structures
 Predicted life-time of structures
 Improved design of future structures

Consequently, today SHM is a very

important objective for end-users,
manufacturers and maintenance teams

203
CONFIDENTIAL - EXTERNAL
 Applications
 Civil engineering structures exposed to wind,
waves, and traffic loads, operating forces,
seismic activities etc.
• Bridges, buildings, power plants, dams, stadiums …
• Wind turbines, off-shore structures …

 Aircraft and helicopters exposed to engine

and blade vibrations, and wind loads

 Ships and submarines exposed to engine

and propeller vibrations and waves loads

 Freight transportation on lorries and trains exposed

to loading, road/rail disturbances and unloading

 and much more

OMA Teaser SHM Teaser

204 C O N F I D E N T I A L
CONFIDENTIAL - EXTERNAL
 Introduction – Vibration-based SHM using SSI
 Vibration-based SHM refers to detecting damage through changes in a set of features extracted from
the vibration signals collected from structural health monitoring (SHM) systems

 A myriad of different vibration-based damage detection methods exist but their popularity is often
hindered by their complex mathematical formulation

 Commercially available methods often use the modal approach, which presumes that the damage is
manifested through a change in the modal parameters, but these methods have severe limitations
• Modal parameters as damage-sensitive features are often not sensitive enough to identify local damage,
especially if the structure is only excited by low-frequency inputs
• The features are sensitive to changes in environmental conditions such as temperature and humidity

 Vibration-based SHM using SSI

• Based on damage detection in the physical domain using state space models instead of in the reduced modal
domain for highest possible reliability
• No modal parameter estimation necessary
• The full dynamic system is compared implicitly instead of comparing a few modes
• Robust estimation in case of differences in vibration level from measurement to measurement
• Robust to changes in ambient conditions
205
CONFIDENTIAL - EXTERNAL
 Basics of vibration-based SHM using SSI
 An Observability matrix O can be defined containing
the State matrix A and the Observation (Output) matrix C
• It contains the dynamics of the system

 Establish the reference state Observability matrix ORef from multiple measurements.
This matrix contains the dynamic description of the system in the reference state
 Establish a similar Observability matrix ODam from measurements of a potentially damaged state

 Perform the damage test using residual vectors :

• Damage vec( ORef – ODam ) ≠ 0
• Safe vec( ORef – ODam ) ≈ 0

 PULSE OMA contains two damage indicators based on the above framework:
• Classic and Robust
• In addition, the Modal damage indicator based on modal parameters calculated using SSI
and a Control Chart combining one or more of the damage indicators
206
CONFIDENTIAL - EXTERNAL
 Damage Detection Example – Dogna Bridge, Italy
Bridge
 Four-span single-lane concrete bridge
 Length of each span is 16 m
 Lane approx. 4 m wide
 Bridge made of reinforced concrete (RC)
• Slab 0.18 m thick, supported by three
longitudinal beams with cross-section
0.35 m x 1.20 m
 Each pier is an RC wall
• About 4.5m deep and 3.6m high

Event
 Bridge flooded and damaged on
August 31, 2003
Measurements and photos provided by:
 For safety reasons, the bridge was - Professor Morassi, Dipartimento di Ingegneria Civile e Architettura,
demolished May 2008 and replaced Università degli Studi di Udine, Italy
by a new one about 200 m downstream - Professor Benedettini, Dipartimento di Ingegneria Civile, Edile-Architettura
e Ambientale, Università degli studi dell'Aquila, Italy
207
CONFIDENTIAL - EXTERNAL
 Damage Detection Example – Dogna Bridge, Italy
Measurements
Performed April, 2008
 8 measurements with bridge in ”undamaged”
condition
• State after flooding

 14 measurements performed after damage has

been added
• Asphalt overlay was removed, and span
was made independent of adjacent span before
damage was artificially added
• Up to 3 cuts of one of the supporting beams as well
as introduction of a damage in the centerline of the
side beam

 Measurements using 10 uniaxial accelerometers

• Sampling frequency: 400 Hz Simplified geometry
• 59k samples per channel; Measurement time: ~147 s

Data included in installation: C:\Program Files\Bruel and Kjaer\PULSE OMA\Examples\Damage Detection

208
CONFIDENTIAL - EXTERNAL
 PULSE Structural Health Monitoring – Typical Screen Layout

Damage
Indicators Reference
State
Validation

Analysis
Sessions
table
Modal
Parameter
History
Geometry

209
Modes table CONFIDENTIAL - EXTERNAL
210
CONFIDENTIAL - EXTERNAL
 Further Information

211
CONFIDENTIAL - EXTERNAL
 HBK Academy

https://www.hbkworld.com/en/knowledge/academy

212
CONFIDENTIAL - EXTERNAL
 Global scale. Local presence.

www.hbkworld.com
213
CONFIDENTIAL - EXTERNAL
 IOMAC – International Operational Modal Analysis Conference
 Biennial event held in May in uneven years – except for recent changes due to the pandemic
• Conference, exhibition and pre-conference courses (e.g., 1 day OMA Course)
• Dedicated to OMA and related topics such as SHM
• HBK (B&K) was among the initiators of IOMAC
• HBK represented in the Permanent Committee, the Scientific Committee and often in the Organizing Committee
• Technical papers from previous conferences can be downloaded for free

 Next IOMACs
• 2024: Naples, Italy May 21-24
• 2025: Rennes, France

 Previous conferences
• 2022: Vancouver, Canada 2013: Guimaraes, Portugal
• 2019, 2007, 2005: Copenhagen, Denmark 2011: Istanbul, Turkey
IOMAC
• 2017: Ingolstadt, Germany 2009: Portonovo, Italy
• 2015: Gijon, Spain
214
CONFIDENTIAL - EXTERNAL
 Other international conferences for OMA and SHM
 IMAC (International Modal Analysis Conference)
• A conference and exposition focusing on structural dynamics including topics in simulation and modeling,
nonlinear dynamics, sensors, signal processing and control spanning the full range of engineering disciplines
• Held every year in USA (typically Florida and California) end January, beginning February
• Next IMAC: Orlando, FL, USA Jan 29 – Feb 1, 2024
• www.sem.org/imac

 ISMA (International Conference on Noise and Vibration Engineering)

• Main focus is structural dynamics
• Biennial event (even years) held in September in Leuven, Belgium
• Next ISMA: Sep 9-11, 2024
• www.isma-isaac.be

 EVACES
• International Conference on Experimental Vibration Analysis for Civil Engineering Structures
• Biennial event (uneven years) held end August, beginning September
• Different worldwide locations
• Next EVACES: In 2025. Location and dates not fixed yet
• https://www.evaces2023.polimi.it/
215
CONFIDENTIAL - EXTERNAL
 Q&A

216
CONFIDENTIAL - EXTERNAL
 Objectives
To give all attendees the necessary theoretical and practical knowledge to
successfully conduct proper Operational Modal Measurement and Analysis

By explaining
 The terminology, possibilities, advantages and limitations of Operational Modal Analysis (OMA)
 How OMA can be used as a complement to experimental modal analysis (EMA) based on
typically hammer and shaker testing
 How to perform proper measurement and validation of the vibration signals used for OMA
 The different time and frequency techniques used in OMA and how to validate results
 Specific requirements to instrumentation used for OMA

In addition
 Introduce the idea behind and benefits of vibration-based Structural Health Monitoring (SHM)

217
CONFIDENTIAL - EXTERNAL
 Course Evaluation
Please go to: https://www.menti.com/alyifrxk3wd2

and answer the few questions to evaluate the course.

Your feedback is very important to us, and we really appreciate you taking the time to do so.

218
CONFIDENTIAL - EXTERNAL
 Thank You
[email protected]

www.hbkworld.com | © HBK – Hottinger, Brüel & Kjær | All rights reserved

219 CONFIDENTIAL - EXTERNAL