Applied System Identification

for Constructed Civil Structures

Dionysius Siringoringo, Ph.D

Bridge & Structure Lab, Univ. of Tokyo
July 22, 2010

Series of lecture on Asia-Pacific Summer School on Smart Structures and Control

Outline
• Introduction
- Definition
- Objectives
- Scope

• Experimental Methods
- Classification
- Type of Excitation
- Type of Response

• Analysis Methods
- Classification : Parametric vs Non-parametric
- Type of Model : Structural vs Modal vs Non-Physical Numerical Model
- Domain: Time vs Frequency vs Cross Time-Frequency

• Uncertainties

• Examples of Application

• Discussions and Closure

2
1. Introduction : Definition

• A process to develop or improve mathematical representation of

a structural system using experimentally obtained structural
response(s).

• Mathematical representation of a structural system: Mass,

Stiffness, Damping, Flexibility, Connectivity

• Experimentally obtained structural response : vibrations, static

response/deflection, strain response etc.

• System identification is a broad term, when ‘system’ refers to

structural system, the term structural identification is commonly
used.
1.Introduction : Objectives

Why System Identification for constructed structures ?

1. Model Validation of newly constructed structures

- verify assumptions in design model (e.g. boundary condition, nonlinear
behavior, energy dissipation mechanism/ damping)
- verify performance of control system (e.g. base-isolation, Tuned Mass
Damper, etc)

2. Model Updating
- obtain FEM calibrated structural model
- adjust structural parameters after retrofit or modification

3. Structural/Condition Assessment and Health Monitoring

- detect structural changes possibly due to defect or damage
- recognize environment/loading influence or pattern on the structure
1.Introduction : Objectives

Why System Identification for constructed structures ?

4. Earthquake Engineering
- performance of structure during earthquake
- post-earthquake structural assessment

5. Wind Engineering
- verification/comparison with wind tunnel results
- aerodynamic performance (e.g. aerodynamic damping of long-span bridges)

6. Soil-Structure Interaction
- characterize and quantify parameter of surrounding soil medium

7. Traffic-structure Interaction
- characterize structural response due to certain type of vehicle/train
- detect changes in structure-vehicle interaction medium (e.g. pavement effect on bridge
response, railway track effect on train comfort measure)
1.Introduction : Scopes – Global and Local
Example of Global Structural Identification : Modal identification
of instrumented bridges for global assessment of the
structure

Yokohama Bay Bridge

1.Introduction : Scopes – Global and Local
Example of Local Structural Identification : Evaluation of damping
on stay cable of cable-stayed bridge to asses the effectiveness of
cable damper system.

Stonecutters Bridge Cable Hydraulic damper

Single mode decay response

of the cable : f = 0.49 Hz

Log of Envelopes of Filtered Acceleration-Cable119SDATA15-Channel1-Mode1]

-1.5
peak
valley
average
=0.055487
-2

-2.5

Log (Peak Acc) log(m/s2/s)

-3

-3.5

Free-vibration test of stay cable by pull-and- Cable damping

-4

release test. (logarithmic decrement :

-4.5

d= 0.055)
-5
80 100 120 140 160 180 200 220
Time (s)
2.Exp Methods : Classification of Required Data

The required data is data collected during experiment and can be

classified into:

• Excitation : measurements made of disturbance forces,

pressure, impact, stress applied to the structure.

• Response: measurements made of the reactions of the

structures to the applied disturbance, such as, deflection,
displacement, velocity, acceleration, strain etc.
2.Exp Methods : Type of Excitation

The excitation can be classified as:

1. Dynamic or static (i.e. according to whether or not they engage
inertial effects)
2. According to controllability, and
3. According to measurability

4. Controllable (measurable and un-measurable) static loads

5. Uncontrollable (measurable and un-measurable) static loads
6. Controllable (measurable and un-measurable) dynamic loads
7. Uncontrollable measurable dynamic loads
8. Uncontrollable un-measurable dynamic input (ambient
dynamic excitation)
2.Exp Methods : Static Loads

Controllable (measurable and un-measurable) static loads

Relatively rare for full-scale experiments on real structures because of the
scale of the load required to generate a measurable effect.

Common example is proof testing of bridges often involving use of heavy

vehicles, either stationary or moving

Uncontrollable (measurable and un-measurable) static loads

Generally include elements of dynamic load and response monitoring,

particularly in the case of traffic and wind which generate quasi-static and
dynamic response.
2.Exp Methods : Dynamic Loads – Controllable Measurable

• Forced vibration test (FVT)

Transfer functions or frequency response functions (FRFs) scale
input (forcing) to output (response) via either mass or stiffness
so can both be identified, along with high quality information
about dissipative effects (mathematically realised as viscous
damping)

Examples: mass exciters, Electro-dynamic shakers ,

instrumented hammer
2.Exp Methods : Dynamic Loads – Controllable Un-Measurable

• Manual excitation
Impulse response functions (IRF) or free-vibration response.
Neither mass nor stiffness can be identified. Modal frequency
and damping can be estimated quite accurately.

Examples: Impact hammer, people jump, drop weight test,

Snap-back or or step relaxation test
2.Exp Methods : Dynamic Loads – Controllable Un-Measurable Excitation

Controllable but unmeasurable dynamic loads

• Manual excitation : Impact Hammer Test

Giving excitation to a short

span bridge by impact hammer

Free vibration response of the bridge subjected

to impact hammer
2.Exp Methods : Dynamic Loads – Controllable Un-Measurable Excitation

Manual excitation : Drop Weight Test

Giving excitation to a short span

bridge by dropping sand bag weight

Note: while drop weight test is effective

in exciting the free-vibration response
of the structure, additional damping is Example of Free vibration response of the bridge
expected as the dropped weight tends excited by dropped weight
to increase the damping.
2.Exp Methods : Dynamic Loads – Controllable Un-Measurable Excitation

Manual excitation : Pull-and-released test of stay cable

1.5

1
Example of free vibration
0.5
response of a stay cable

A c c e le ra tio n (m
2
) /s
0

-0.5

-1

-1.5
80 100 120 140 160 180 200 220
Time (s)

Flowchart to obtain damping value of a stay cable

Free-Vibration Response
Frequency Response
Raw Data

Filtering mode of interest

Single-mode free-vibration
response

Logarithmic Decrement
using envelope of decay response
Giving excitation to a stay cable by pull-and
released test Single-mode damping value
2.Exp Methods : Dynamic Loads – Controllable Un-Measurable Excitation

Vehicle excitation/ Controlled Traffic

Example of strain response
when a truck passing a bridge

Vehicle excitation:
1. Response larger than ambient vibration
response.
2. Stress and acceleration responses can
be conducted simultaneously
3. Effect of vehicle-bridge interaction
should be considered in analysis.
Example of acceleration response when a truck
passing a bridge
2.Exp Methods : Dynamic Loads – Uncontrollable Measurable Excitation
Seismic excitation
Transfer functions or frequency response functions (FRFs) between seismic
input (base excitation) to output (structure response) . Structural properties,
modal properties and modal participation factor can be estimated.

Example: instrumented bridges and buildings in Japan and California US.

Example : Yokohama-Bay Bridge, instrumented cable-stayed bridge near Tokyo

2.Exp Methods : Dynamic Loads – Uncontrollable un-measurable Excitation

Ambient excitation : wind, traffic, and unmeasured

micro-tremor.
Correlations between response are used to estimate modal
properties. Mode-shapes unscaled. Treated as stochastic
system identification.

Example: periodic ambient vibration measurement and

instrumented bridges and buildings.
2.Exp Methods : Dynamic Loads – Uncontrollable un-measurable Excitation

Example of ambient excitation : wind-induced vibration of

suspension bridge.
Tower acceleration response

Treated as stationary random process.

2.Exp Methods : Dynamic Loads – Uncontrollable un-measurable Excitation

Example of ambient excitation : traffic-induced vibration of bridge.

Example of vertical acc and the spectrum of a medium span highway bridge to traffic.

Bridge response subjected to open traffic usually treated as stationary random process,
since the input is unknown. Effect of vehicle mass is usually neglected. However, in case of
short span bridge, the effect of vehicle mass may not be negligible and influence the
identified bridge frequency.
2.Exp Methods : Type of Response Excitation

Static : Strain, Deflection

Dynamic:
• Acceleration
• Relative of absolute displacement
• Velocity
• Inclination
• Strain
• Stress
• Water Pressure
• Structural and environmental temperature
• Wind Velocity
• Wind Direction
3. Analysis Methods: Classification

Parametric and Non-parametric Models

Parametric Model
Structural model and initial estimate of model parameters are
known a priori. Measured responses are fitted to obtain the
best estimate of model parameters.

Non-Parametric Model
Model structure is not specified a priori. Structural responses
are used to obtain model parameters by means of dynamical
system and quantities such as cross/auto-correlations, transfer
function/frequency response function.
3. Analysis Methods: Classification
Example of Parametric Model
Output-Error Minimization for system identification using seismic
response.
Theoretical model or structural
model is required

F = objective function

Minimize the difference between the

Example : comparison between recorded
measured modal parameters and model
response and computed response of a bridge
generated modal parameters by updating
deck subjected to seismic excitation
parameters of the model iteratively
3. Analysis Methods: Classification
Example of Non-Parametric Model
State-Space System identification using seismic response.

By modeling the input-output relationship of seismic-induced vibration using state-space

model and realization of observability matrix, system matrices A, B, R and D can be
obtained and modal parameters are realized.
3. Analysis Methods: Types of Model
1. Structural Model
• System is modeled in terms of mass, stiffness, or flexibility, and
damping matrices.
• Geometric distribution of mass, stiffness and damping are known
• Structural connectivity between degree of freedom is preserved

Equation of motion : Mu(t )  Cu (t )  Ku (t )   Bz(t )

System matrix A in state-space form for  0 I 

A  exp  1 1 
t
discrete data:  M K  M C

Equation of motion in discrete x ( k  1)   A x ( k )   B  z ( k )

dynamic system, where system
matrix A is to be indentified y ( k )   R  x (k )  [ D ]z (k )
Goal : To Indentify system matrix A in its original form, from which the mass, stiffness
and damping matrices can be retrieved.
3. Analysis Methods: Types of Model
2. Modal Model
• System is defined in modal coordinates describing the vibratory motion of
structures in terms of modal frequency, modal damping and mode shapes
(also mode phase angle for complex modes)
• Geometric distribution of mass, stiffness and damping and information on
structural connectivity are not preserved.
• Describes the resonant spatial (mode shapes) and temporal of the structure.
• Modal parameters are analogous to eigensolution of stuctural mass and
stiffness.
Equation of motion in discrete dynamic system,
where system matrix A is to be indentified

x ( k  1)   A x ( k )   B  z ( k )

y ( k )   R  x (k )  [ D ]z (k )

Goal : To Indentify system matrix A by solving the

Eigenvalue problem and determine the modes.
3. Analysis Methods: Types of Model
3. Non-Physical Numerical Model

• Does not have physical relationship with the structure (i.e. no spatial information, no
geometry distribution of mass, stiffness, and damping)
• Simply a parameter curve-fit of the given mathematical model to the measured data.
• Examples : Auto Regressive Moving Average (ARMA) and its variants, Rational
Polynomial Model etc.
• Some can be converted to modal model form.

Example : Auto Regressive Moving Average (ARMA) Model where the auto regressive
coefficients can be related to modal parameters

d n y (t ) d n 1 y (t ) dy (t )
n
 a n 1 n 1
  a1  a 0 y (t )
dt dt dt
d m u (t ) d m 1u (t ) du (t )
 bm m
 bm 1 m 1
  b1  b0 u (t )
du du dt
3. Analysis Methods : Domain

1. Frequency Domain

• Peak Picking Method

• Transfer Function / Frequency Response Function/ Impulse
Response Function.
• Average Normalized Power Spectrum Density (ANPSD)
• Complex Exponential Frequency Domain Method (Schmerr
1982)
• Eigensystem Realization Algorithm in Frequency Domain (ERA-
FD) (Juang & Suzuki 1988)
• Frequency Domain Decomposition (Brincker et al. 2001)

28
3. Analysis Methods : Domain

2. Time Domain

• Ibrahim Time Domain (ITD) (Ibrahim & Mikulcik 1973)

• Least Squared Complex Exponential Method (LSCE) (Brown
1979)
• Polyreference Complex Exponential Method (PRCE) (Vold et al.
1982)
• Eigensystem Realization Algorithm (ERA) (Juang & Pappa 1985)
• Stochastic Subspace Identification (Overschee & De Moor
1991)

29
3. Analysis Methods : Domain
3. Cross Time-Frequency Domain
Represents frequency evolution as time progresses.
Can detect non-linearity and non-stationary signals

• Short Time Fourier Transform (STFT)

• Wavelet-based system identification
• Empirical Mode Decomposition – Hilbert Huang Transform
(should be carefully applied since EMD lacks physical meaning of
signals)

30
3. Analysis Methods : Direct and Indirect Method Time Domain
Direct Method
When the IRF/FRF is available, they can be use as input directly to
system identification method.

Indirect Method
When the IRF/FRF is unavailable such as in case of ambient
vibration measurement, an additional method is needed to
construct synthetic IRF , ex. through cross-correlation (Natural
Excitation Technique (NEXT) or through Random Decrement.

Raw Data NEXT ERA

Randec ITD
31
3. Analysis Methods : Checklist
Types of Inputs and Outputs
Controllable dynamic loads
Measured Input(s)
Mass exciter and Shaker
Instrumented Impact Hammer
Unmeasured Input (s)
Manual excitation (people jumping)
Snap-back, or step relaxation
Swinging bell to excite a cathedral tower
Uncontrollable dynamic loads
Measured Input(s)
seismic excitation (Single Input)
seismic multiple excitation (Multiple Input)
Uncontrollable dynamic loads
Unmeasured Input(s)
ambient vibration test (operational modal
analysis)
wind excitation
traffic excitation
microtremor with unmeasured input
System Identification Method
 
 
Transfer Functions (Input-Output SysID)
Single Input Single Output (SISO) or Single Input Multi
Output (SIMO) system
 
Impulse Response Function/ Free-vibration Response
Impulse Response Function/ Free-vibration Response
Impulse Response Function/ Free-vibration Response
 
 
Transfer function, SISO or SIMO
Transfer function matrix, MIMO
 
 
Stationary broadband assumption
Output Cross-correlation
Covariance-Driven System Identification
Data Driven Stochastic Subspace Identification
32
4. Uncertainties

Uncertainty is unavoidable in understanding the results of system

identification. Modal properties are susceptible to variation
even when structural condition remains the same.

How to quantify the confidence of the identified modal properties?

1. Error propagation analysis using perturbation method

2. Monte Carlo Simulation
3. Bootstrap Method

33
4. Uncertainties : Error propagation analysis using perturbation
Example of error propagation in System Realization using Information Matrix

Input [Up] Correlation Matrix :

Correlation Matrix Information Matrix
Up(e) = Up(0) + eDUp Ryy(e) = Ryy(0) + eDRyy
Of Input-Output [Rhh]
Rhh(e) = Rhh(0) + eDRhh
[Ryy],
Ryu(e) [Ryu],
= Ryu(0) + eDRyu
[Ruu]
Yp(e) = Yp(0) + eDYp Singular Value
Output [Yp] Singular Value
Ruu(e) = Ruu(0) + eDRuu Decomposition
Decomposition
Realization of System
Realization of System
Matrix
Matrix [A]
A(e) = A(0) + eD A

Objectives:
To define and quantify the error on Realization of Modal
Realization of Modal
the modal parameters as the effect Parameters
Parameters
of input and output noise w(e) = w(0) + eD w
w, z, j
z(e) = z(0) + eD z
34
f(e) = f(0) + eDf
4. Uncertainties : Bootstrap Analysis
Example of investigation of the effect of variability and to estimate the confidence
bounds of identified modal parameters by NEXT-ERA

Randomly selected Ensemble 1 ERA

CCF1, CCF2, CCF3,…CCFN (M component) w1, x1,f1
CCF1, CCF5, CCF3,…CCFM
Compute CCF average

Randomly selected
Ensemble 2 ERA
(M component) w2, x2,f2
CCF7, CCF2, CCF1,…CCFM
Compute CCF average

. .
. .
. .
Randomly selected Ensemble P
(M component) ERA
CCF4, CCF6, CCF2,…CCFM wp, xp,fp
Compute CCF average

Estimate mean value and

CCF : Cross-correlation function 95% Confidence bound
35
4. Uncertainties : Bootstrap Analysis
Distribution of modal parameters and their mean values and
confidence level can be obtained

Examples of bridge 1st frequency statistical distribution on different structural conditions

using Bootstrap Method

Modal parameters are considered as stochastic variable that have distribution

with certain statistical characteristics . Therefore decision made on structural
condition involved statistical confidence.
36
5. Examples: Ambient Vibration Measurement of Suspension Bridge

3-Span Suspension Bridge Tower Height : 130 m

Bridge Type: Simply-supported at the 21m at base
tower Tower width : 18 m on top
Length: 1,380m Girder material : Streamlined steel-box
Span: 330-720-330 m Tower material : Steel box (welded)
Total Deck Width: 20m Completed : 1998

37
5. Examples: Ambient Vibration Measurement of Suspension Bridge
Using NEXT -ERA Modal Parameters

38
5. Examples: Ambient Vibration Measurement of Suspension Bridge

Structural identification : effect of friction force and aerodynamic forces on identified

frequency and damping (Nagayama et. al 2005)

39
5. Examples: Seismic-Induced System Identification of Cable-Stayed Bridge

40
5. Examples: Seismic-Induced System Identification of Cable-Stayed Bridge
A data driven identification method was applied considering multiple input excitation
and multiple responses (MIMO System)

41
5. Examples: Seismic-Induced System Identification of Cable-Stayed Bridge
With dense instrumentation and good quality of seismic records we identify bridge
modal parameters until high order

42
5. Examples: Seismic-Induced System Identification of Cable-Stayed Bridge
Observation of the performance of seismic isolation devices using 1st longitudinal
mode (Siringoringo & Fujino 2008 )

(a) Typical slip-slip Mode (Earthquake 1990-02-20) (b) Typical Mixed Slip-Stick Mode (Earthquake 1995-07-03 )

From the first longitudinal mode we can observe behavior of Link-Bearing Connection
during earthquake.

Different behaviour of Link-Bearing Connection at the end-piers was observed during

different level of earthquake excitation.

It was found that the expected slip-slip mode only occurred during large earthquake. 43
Suggested Readings Materials:

Theoretical and Experimental Modal Analysis by Maia, Silva, He, Lieven et al.
Applied System Identification by Jer Nan Juang
Monitoring and Assessment of Structures by GST Armer
The State of the Art in Structural Identification of Constructed Facilities (ASCE Report 1999)

Q&S
Questions and Sharing?

• Dionysius Siringoringo
dion@bridge.t.u-tokyo.ac.jp

44