deliverable [d13.2] 28 08 2012 v2 upat - series1).pdf · properties and damping from dynamic...

SEVENTH FRAMEWORK PROGRAMME Capacities Specific Programme

Research Infrastructures

Project No.: 227887

SERIES SEISMIC ENGINEERING RESEARCH INFRASTRUCTURES FOR

EUROPEAN SYNERGIES

State-of-the-art report for JRA 2

Workpackage WP13 Deliverable [D13.2] – [On software development for data processing]

Deliverable/Editor: LNEC, UNITN Reviewer: UNITN

Revision: Final

August, 2012

1

ABSTRACT

The main objective of this report that covers the research activities of Task JRA2.3 is the

presentation of the development of numerical tools for processing data from experiments on

structures/infrastructures, suitable for model calibration and specimen simulation. Compatibility

with data formats of the distributed database to be developed in Task NA1.3 is ensured; and

some tools are able to assess the uncertain propagation of random or systematic errors in

computer models owing to experimental measurements. In greater detail, the following topics

and objectives are treated:

- development of software tools for system identification of mockups on shaking tables and

relevant data treatment.

- Software development for Modal Parameter Extraction and for Identification of modal

properties and damping from dynamic response of components and structural systems.

- Improvement of a Performance-based Earthquake Engineering (PBEE) toolbox

developed in a Matlab environment, in combination with the FE-based OpenSees

software. In particular if results of pseudo-dynamic tests are available, the PBEE toolbox

enables post-processing of results and identification of errors in structural models.

- Software development to animate 3D modal shapes from OpenSees outputs and

MATLAB software for 3D animations of real-time tests.

- Software improvement for Structural Health Monitoring (SHM) of structures in situ, in

order to track changes in their dynamic characteristics and detect damage. The critical

need is the availability of simple web tools and techniques for real-time data analysis and

interpretation.

Keywords: Data treatment. System Identification. LabView platform. Matlab environment. 3D animations.

3

ACKNOWLEDGMENTS

The research leading to these results has received funding from the European Community’s

Seventh Framework Programme [FP7/2007-2013] under grant agreement n° 227887.

This work has been developed by the partners of the JRA2 activity.

5

DELIVERABLE CONTRIBUTORS

JRC F.J. Molina

D. Tirelli

F. Taucer

KOERI E. Safak

LNEC A. C. Costa

P. Candeias

UL P. Fajfar

M. Dolšek

UNITN O. S. Bursi

Md. S. Reza

G. Abbiati

Z. Mei

7

CONTENTS

List of Figures ................................................................................................................................11

1 Study Overview .....................................................................................................................13

2 LNEC‐SPA, A shaking table test analysis tool ......................................................................15

2.1 INTRODUCTION ..........................................................................................................15

2.2 MODULE DESCRIPTION ..............................................................................................17

2.2.1 Main Module and Data Storing .......................................................................17

2.2.2 Data Generation Module .................................................................................18

2.2.3 Data Acquisition Module .................................................................................19

2.2.4 Analysis Module ...............................................................................................20

2.2.5 Math Channels Module ....................................................................................20

2.2.6 MDOF Model Module ......................................................................................23

2.2.7 System Identification Module .........................................................................23

2.2.8 Visualization Module and Graph Animation Module ......................................24

2.2.9 Strong Ground Motion Module .......................................................................25

2.2.10 Web Shaker Module ........................................................................................26

2.3 CONCLUSIONS ............................................................................................................28

3 PBEE Toolbox for Identification of Errors in Structural Models ..........................................29

4 MATLAB Tools for Structural Identification and for Identification by means of Pseudo‐

Dynamic Tests .......................................................................................................................35

4.1 INTRODUCTION ..........................................................................................................35

4.2 STRUCTURAL IDENTIFICATION BY FAST IMPACT HAMMER TESTING METHOD .35

4.2.1 Introduction .....................................................................................................35

4.2.2 The FIHT Method .............................................................................................36

4.2.3 Structure Description and Methodology employed .......................................37

4.2.4 Signals Processing ...........................................................................................37

4.2.5 Mode Shapes ...................................................................................................41

8

4.3 SOFTWARE FOR IDENTIFICATION OF MODAL PROPERTIES FROM DYNAMIC

RESPONSE OF STRUCTURES .....................................................................................43

4.3.1 Introduction .....................................................................................................43

4.3.2 Identification of Response Frequency and Damping by a Spatial Modal ......44

4.3.3 Identification of Response Frequency and Damping by a Filter Modal .........45

4.3.4 MATLAB Functions for Identification of the Models ......................................46

4.4 USE OF SIMULINK 3D ANIMATION TOOLBOX FOR THE REPRESENTATION OF

REAL‐TIME TESTING OUTPUT ...................................................................................48

4.4.1 Introduction .....................................................................................................48

4.4.2 Simulink 3D Animation Toolbox......................................................................48

4.4.3 The Representation of Real‐time Testing Outputs ........................................49

4.5 CONCLUSIONS ............................................................................................................51

5 Software Developments for the Implementation of Partitioned Algorithms and their

Interaction with OpenSEES ..................................................................................................53

5.1 INTRODUCTION ..........................................................................................................53

5.2 NUMERICAL AND EXPERIMENTAL VALIDATION OF THE PM METHOD COUPLED

TO OPENSEES .............................................................................................................53

5.2.1 Introduction .....................................................................................................53

5.2.2 Preliminary Numerical Simulation ..................................................................54

5.2.3 The PM Method ...............................................................................................54

5.2.4 Implementation ...............................................................................................55

5.3 MATLAB INTERFACE TO OPENSEES .........................................................................59

5.3.1 Generic Description .........................................................................................59

5.3.2 Input Check Capabilities ..................................................................................59

5.3.3 Post Processing Results ...................................................................................59

5.3.4 Sensitivity Analysis and Optimisation .............................................................60

5.4 STRUCTURAL DYNAMIC IDENTIFICATION TOOLBOX .............................................61

5.5 CONCLUSIONS ............................................................................................................63

6 Structural Health Monitoring................................................................................................65

6.1 INTRODUCTION ..........................................................................................................65

9

6.2 JUSTIFICATION FOR SHM ...........................................................................................65

6.3 DAMAGE DETECTION .................................................................................................67

6.3.1 Damage Detection Based on Change of Natural Frequencies .......................67

6.3.2 Damage Detection Based on Permanent Change in Geometry ....................68

6.3.3 Damage Detection Based on Wave Propagation Characteristics ..................69

6.4 DATA ANALYSIS ..........................................................................................................70

6.4.1 Spectral Analysis ..............................................................................................70

6.4.2 Statistical Signal Processing ............................................................................71

6.4.3 Tracking Time Variations of Signal Properties ...............................................72

6.5 CONCLUSIONS ............................................................................................................73

7 Summary ...............................................................................................................................75

References .....................................................................................................................................77

11

List of Figures

Fig. 2.1: Module organization of the software. ......................................................................... 16 Fig. 2.2: Main Module user interface ........................................................................................ 17 Fig. 2.3: Data organization. ........................................................................................................ 18 Fig. 2.4: GSR Seismic Recorder. ................................................................................................ 18 Fig. 2.5: Data Generation Module user interface..................................................................... 21 Fig. 2.6: Data Acquisition Module user interface. ................................................................... 21 Fig. 2.7: Analysis Module user interface. .................................................................................. 22 Fig. 2.8: Math Channels Module user interface. ...................................................................... 22 Fig. 2.9: MDOF Model Module user interface. ........................................................................ 23 Fig. 2.10: System Identification Module user interface ........................................................... 24 Fig. 2.11: Second floor relative displacements of a precast specimen. ................................... 25 Fig. 2.12: Relative displacements of a precast beam-column joint. ........................................ 25 Fig. 2.13: Strong Ground Motion Module user interface. ....................................................... 26 Fig. 2.14: Web Shaker Module user interface. ......................................................................... 27 Fig. 2.15: General view of the Web Shaker. ............................................................................. 27 Fig. 3.1 The four-storey reinforced concrete frame building. ................................................. 32 Fig. 3.2 Moment-rotation relationship with softening of plastic hinges in columns and beams: a) a four-linear and b) a tri-linear. ............................................................................... 32 Fig. 3.3 The base shear versus first storey drift time histories for models 1 and 2 obtained by nonlinear dynamic analysis and by imposed displacement from pseudo-dynamic test. The computed results are compared with the experimental results for the high-level test. 33 Fig. 4.1 Analytical results to find the minimum amplitude of the first six bending modes expected along the beam ............................................................................................................. 37 Fig. 4.2 Example of the mean FRF of one point position on right, the respective coherence function on left............................................................................................................................. 38 Fig. 4.3 Coherence histogram of the FRF on raw data and on position ................................ 39 Fig. 4.4 Example of distribution of the coherence values for on test and for all the transducers .................................................................................................................................. 39 Fig. 4.5 Example of the map of the filtered coherence for 2 types of transducers: displacement (a), accelerometer (b) ........................................................................................... 40 Fig. 4.6 Example of mode shape value calculation from the FRF complex representation at one position for the mode m ....................................................................................................... 41 Fig. 4.7 Mode shape of a flapper mode of the bridge edges in Y direction ............................ 42 Fig. 4.8 Spatial model frequency and damping ratio identified values for both modes ....... 47 Fig. 4.9 Filter model frequency and damping ratio identified values for both modes ......... 47 Fig. 4.10 a) Results and the representation for the outputs of PLRT2 with a time-step of 4ms b) Results and the representation for the outputs of PLRT2 with a time-step of 2ms 49 Fig. 4.11 Results and the representation for the outputs of improved PLRT2 with a time-step of 4ms ................................................................................................................................... 50 Fig. 4.12 The procedure for simulating the real-time testing output with Simulink 3D Animation .................................................................................................................................... 50

12

Fig. 5.1 FE model and mode shapes of the Rio Torto viaduct: a) period T=1.15s; b) period T=1.01s ......................................................................................................................................... 54 Fig. 5.2 Task sequence of the PM method ................................................................................ 55 Fig. 5.3 Arrangement of the experimental equipment ............................................................. 56 Fig. 5.4 Block diagram ................................................................................................................ 56 Fig. 5. 5 a) Emulated SDoF system b) Test set-up of the PS .................................................. 56 Fig. 5.6 Displacement response at the interface DoF and relevant zoom .............................. 57 Fig. 5.7 Reaction force at the interface DoF and relevant zoom ............................................. 57 Fig. 5.8 Displacement response at the interface DoF and relevant zoom .............................. 58 Fig. 5.9 Reaction force at the interface DoF and relevant zoom ............................................. 58 Fig. 5.10 Arrangement of the experimental equipment........................................................... 58 Fig. 5.11 Block diagram .............................................................................................................. 58 Fig. 5.12 DXF plot of the pier 6 of the Rio Torto viaduct ....................................................... 59 Fig. 5.13 FE model and mode shapes of the Rio Torto viaduct: a) period T=1.15s; b) period T=1.01s ......................................................................................................................................... 60 Fig. 5.14 Displacement response of a) pier 8 and b) pier 9 of the Rio Torto viaduct ........... 60 Fig. 5.15 SDIT Graphic User Interface ..................................................................................... 61 Fig. 5.16 Spectrogram for time-frequency domain .................................................................. 61 Fig. 5.17 Cluster diagram ........................................................................................................... 62 Fig. 5.18 Stabilisation diagram .................................................................................................. 62 Fig. 5.19 Instantaneous estimation of the modal parameters of the 1st eigenmode of the Pescara Bridge ............................................................................................................................. 63

13

1 Study Overview

The main objective of JRA2 is the implementation and application of new types of sensors,

control techniques and modelling tools capable of enhancing the measurement of the response of

test specimens and improving the quality of test control. The activity also aims at developing

numerical simulation tools, integrated with data processing, databases and visualisation, for an

improved design of test campaigns, including the equipment and for enhanced interpretation of

experimental results.

In greater detail, the following objectives in Task JRA2.3 were pursued:

- software development for processing data from tests and for database management.

- Modelling tools, e.g. Finite Element (FE) codes, data processing software and databases

were improved, for the ultimate purpose of a better design of the testing equipment and

interpretation of experimental results.

- Requirements for data generated from physical tests were identified, for calibration and

development of numerical models and damage assessment.

[On software development for data processing]

15

2 LNEC‐SPA, A shaking table test analysis tool

2.1 INTRODUCTION

This section describes a new software tool for generation, acquisition, analysis and visualization

of data collected in dynamic tests made by LNEC.

The software is implemented in the LabView platform (NI 2004) using the native signal analysis

toolkits and new user developed routines. To compute the more intensive calculations some

functions are compiled into dynamic link libraries (dlls) using a Fortran compiler (Compaq

2000). The program is compiled into an executable version that can run independently of

Labview using a setup package that deploys all necessary run-time files.

The code is divided into separate modules with specific tasks. All the modules use a common

interface for data and file management. By the use of this separate module architecture one has

advantages in code development, specialization and it enables the creation of sub-versions of the

program compiled only with the required modules for each user.

The modules currently available are:

Main Module – is a common interface for all modules that allows the management of the data in memory and loading and saving it into files of different formats;

Data Generation – is a signal generation module used to define the input for the tests; Data Acquisition – is used to calibrate and acquire signals from DAQ boards; Analysis – is the main signal processing tool; Math Channels – is used to calculate runtime defined mathematical channels using the

data collected during the tests; MDOF Model – is used to assess the global behaviour of lumped mass systems using

only kinematic quantities allowing the evaluation of forces, moments (e.g. base shear) and energies (e.g. input energy);

System Identification – is frequency domain input-output system identification module; 2D Visualization – allows creating real-time 2D dynamic visualizations; Graph Animation – allows creating animations of XY graphs;


16

Strong Ground Motion – is used to analysis strong ground motion (SGM) records and accessing several SGM parameters;

Web Shaker – is an educational tool that enables a remote user to use a small electrical 1 d.o.f. shaking table.

A graphical representation of the software architecture is presented in Fig. 2.1.

Fig. 2.1: Module organization of the software

The program requires a valid license to operate for two reasons: protect intellectual property and

stimulate continuous update of the software. The license is completely free of charge.

The software has a great concern on practical aspects like exporting easily automatically

formatted images of time-histories and functions to files, to the printer or to MS Word and

Excel 1 . Two other important characteristics are the automatic unit conversion procedures

included on some modules and the automatic option loading form the last time each module was

used.

The next section presents the main characteristics of each module and also of the data storing

scheme.

1 MS Word and Excel are licensed software developed by the Microsoft Corporation.


17

2.2 MODULE DESCRIPTION

2.2.1 Main Module and Data Storing

The Main Module, see Fig. 2.2, is an interface that allows accessing all other modules and a

global settings sub-module used for program configuration. This module uses two virtual

memory zones for selecting, ordering and deletion of channels: the File zone is used to load and

save data into files of different formats; and the Buffer which holds the data used for the

calculations.

The data scheme contains not only information about the samples acquired during the tests but

also the channel’s name, type (e.g. relative displacement), unit, trigger time, and time step. It also

saves information common to all the channels in a header (test name, series, date and

observations, see Fig. 2.3). The main module also allows editing most of this information.

An important characteristic of the software is that is possible to use differently sampled time-

histories and that the name, type and unit are used for automatic graph labelling and for the unit

conversion procedures.

Fig. 2.2: Main Module user interface


18

CHANNELS

HEADER

Name (String)Series (String)Date (String)Obs. (String)

Name (String)Type (String)Unit (String)

Trigger Time (Time stamp)Time Step (Double)

Samples (1D Array of double)

Fig. 2.3: Data organization

Data storage is made in tab delimited ASCII text files and in binary files which are much more

efficient. This efficiency is achieved by decreasing the numerical precision of the measured data

samples. Although for calculations double precision format (64

bits) is used when load and saving to files the data is converted

to 16 bits integers which can have (216 = 65 536 levels of

precision) using a scale factor and an offset. The 65 536 levels

of precision are usually enough to store the data without losing

physical precision because most of the analogue–digital

converters (ADC) in the DAQ boards used at LNEC works 16

bits. An example of this greater efficiency is the 30.7 Mbytes

required to store 170 channels with 16384 samples each using the ASCII format and the

equivalent 5.32 Mb of the binary file.

The program reads also 12, 16 and 18 bits records from GeoSig seismic recorder units (see Fig.

2.4) using external routines (dlls) to convert them to the data format used in the program.

2.2.2 Data Generation Module

The Data Generation Module (see Fig. 2.5) allows creating time-histories for using as input in

the tests. Some basic signal functions are implemented (e.g. sine, triangle, square, sawtooth

waves and artificial noise generators), sine sweep functions (with constant amplitude or constant

acceleration for displacement time-histories) and calibration functions, which are slow motion

displacements used to calibrate transducers (e.g. optical displacement transducers). Some

functions are also available for creating seismic records, like creating time-histories from power

Fig. 2.4: GSR Seismic Recorder


19

spectral density functions and, still in development and testing, the generation of time-histories

compatible with a given response spectra.

The principal tools available in this module are time and frequency domain numerical

differentiation and integration procedures, which are very useful to transform records between

accelerations, velocities and displacements. Other basic edit tools are also available (e.g. offset

removal, function fit and removal, filters and windows, etc.)

A future development to this module is to implement functions to create non-stationary seismic

records.

2.2.3 Data Acquisition Module

The Data Acquisition Module (see Fig. 2.6) is used to send data to actuators (analogue-output)

and to acquire data from sensors (analogue-input). The module can work in two different ways:

the continuous mode, where the acquisition and output processes until a user command; and the

buffered mode, where both input and output are made using the memory buffer in the DAQ

board to process a finite amount of samples. This last mode is more reliable because all the

acquisition is buffered in the hardware, avoiding software caused synchronization problems. This

module also incorporate a basic signal generator but it can access any time-history loaded in

memory.

The software supports manual software triggering (button click event), hardware trigger

(hardware switch on DAQ boards triggered by an external device, which allows the

synchronization of the input and output), time trigger (setting a specific trigger time) and

analogue channel trigger (setting a trigger level on a specific channel).

At this stage the module only supports one simultaneous National Instruments compatible board.

A future development to this module is to be able to use more than one DAQ board installed on

the local computer and to control other acquisition stations by computer network (TCP protocol).


20

2.2.4 Analysis Module

The Analysis Module (see Fig. 2.7) is the main signal processing module. It implements all the

commands for edit and analyse the records. Some of the available commands includes time and

frequency domain numerical differentiation and integration functions, basic signal edit tools (e.g.

offset removal, crop, scale, clip, etc.), trends removal (e.g. linear, polynomial, etc.), signal

resampling functions, windows functions (e.g. cosine tapper window, Hanning window, etc.),

filters (Fourier Filter, and the normal signal processing filters, Butterworth, etc.), peak detector

and the most important spectra functions ( Elastic Response Spectra, Fourier Spectra, Power and

Power Spectral Density Spectrum).

This module runs a continuous cycle where each processing and analysis command can be

added, changed, removed and ordered producing immediate results. Another important

characteristic of the module is being able to create or modify the channels based on the results of

the commands. It is also possible to work with a single channel or with a user defined group of

channels.

2.2.5 Math Channels Module

The Math Channels Module (see Fig. 2.8) allows computing run time defined math channels,

which are mathematically combined channels created from the data acquired. The module uses a

hierarchy of mathematic operations which can be defined at run-time and saved into a binary file.

With this module it is possible to create additional channels just after the raw data is retrieved.

Two types of operations are implemented: operations on a single channel (e.g. arithmetic

operations, trigonometric functions, offset removing, filtering and windows); and operations

combining a group of channels (e.g. adding, subtracting or averaging channels, calculation of the

3 d.o.f. cinematic components for rigid floors, and energy calculations).

This module has proven to be extremely useful to process the data into more elaborated

engineering quantities.


21

Fig. 2.5: Data Generation Module user interface

Fig. 2.6: Data Acquisition Module user interface


22

Fig. 2.7: Analysis Module user interface

Fig. 2.8: Math Channels Module user interface


23

2.2.6 MDOF Model Module

The MDOF Model Module allows computing an estimation of the global forces developed on

specimens that can be considered as lumped mass systems (e.g. RC structures with rigid floors)

using a simplified model that assumes that the inertia forces are equal to the restoring forces

(neglecting damping forces):

0

0

i d k

i k

d

F F FF F

F

(1)

This model gives only approximated values with more accuracy when the velocities are small

(viscous damping hypothesis), but enables to compute forces and moments using the measured

cinematic values.

The inertia forces and torque (IF, IT), the story inertia forces and torque (SF, ST), the base shear

(BS) and the base overturning moment (BOM) can be computed as defined in Fig. 2.9.

ns – Number of storey m – Story mass It – Story torsion inertia IF – Inertia force IT – Inertia torque

SF – Story inertia force ST – Story inertia torque BS – Base shear BOM – Base overturning

moment

Fig. 2.9: MDOF Model Module user interface

2.2.7 System Identification Module

The System Identification Module (see Fig. 2.8) can be used to access the specimen’s dynamic

characteristics and its evolution during the tests. It is a usual practice at LNEC to use the shaking

table to perform low amplitude, broadband test series between each earthquake load. These data

can be used to easily estimate the modal frequencies and damping using regular input-output

modal identification techniques, like the peak picking method of frequency response functions


24

(FRF) estimations. This module allows using single input and multiple output series, filtering the

series and processing frames to improve the FRF estimations (Bendat et al. 1986).

In a near future it is an objective to adopt a curve fitting method to the FRFs to compute

automatically the resonant frequencies and damping. Modal configuration is also an objective

like implementing more advanced identification techniques.

2.2.8 Visualization Module and Graph Animation Module

Visualization is a tool that can enhance significantly the presentation of results, but can also help

the researcher to understand easily and deeply the response of structures. A 2D visualization

module is implemented in the program that allows creating 2D animations using time-history

moving nodes connected by lines (see Fig. 2.11 and Fig. 2.12).

Fig. 2.10: System Identification Module user interface


25

The Graph Animation Module can produce animations of XY graphs which can be used to

represent the response of the structure in polar diagrams, hysteretic cycles, etc. A future

development is to implement a 3D visualization module improving greatly the capacities of

the software.

Fig. 2.11: Second floor relative displacements of a

precast specimen

Fig. 2.12: Relative displacements of a precast

beam-column joint

2.2.9 Strong Ground Motion Module

The Strong Ground Motion (SGM) Module (see Fig. 2.13) was developed for the analysis of data

collected by seismic recorders (Fig. 2.4). This module enables computing the velocity and

displacement time-histories from the numerical integration of the acceleration record. Both time

and frequency domain integration is implemented, Fourier filtering and baseline correction.

Better results are usually obtained using the frequency domain integrator and a band pass Fourier

filter between 0.1-0.5 Hz and 30-50 Hz, followed by a linear baseline removal.

With the acceleration, velocity and displacement records it is possible to compute several SGM

parameters, namely: PGA; PGV; PGD; Arias intensity; A95 Parameter; Predominant Period,

Mean Period and others. It is also possible to compute basic spectra diagrams, (e.g. elastic


26

response spectra; Fourier Spectra, power and PSD spectra) and also the Husid Plot and Energy

Flux Plot. An excellent description of some of these parameters can be found in the

documentation of the SeismoSignal Programme (SeismoSoft 2004).

Fig. 2.13: Strong Ground Motion Module user interface

2.2.10 Web Shaker Module

The goal of the Web Shaker Module (see Fig. 2.14 and Fig. 2.15) is to make available through

the internet a small electric 1D shaking table with an analogue control system used for

educational purposes. Several types of specimens can be tested like simple 1 d.o.f. models (see

Fig. 2.14) to more complex m.d.o.f structures.

The module allows a remote user to control the shaking table, define the input, view remotely the

test throw a web camera and visualize the output records (e.g. displacement of the shaking table

and several displacements and accelerations in the specimens). It is also possible to turn on a

spot light to perform the tests at night.


27

A development for a near future is to create a procedure for the program to automatically send an

email to the user with the data collected in each test.

Fig. 2.14: Web Shaker Module user interface

Fig. 2.15: General view of the Web Shaker


28

2.3 CONCLUSIONS

In this section, a shaking table test analysis tool development by LNEC was summarized. Its

application to the aforementioned tests appeared to be very successful.


29

3 PBEE Toolbox for Identification of Errors in Structural Models

The simulation of the seismic response of a structure is a complex task, which is usually solved

in combination with experimental research. The University of Ljubljana has been involved in

several projects (SAFERR, SPEAR, LESSLOSS) where experimental results were used for the

validation of mathematical models using different software. In addition to some commercially

available software, which supports performance-based methodologies, for example, SAP 2000,

ETABS and PERFORM-3D, open-source software (e.g. OpenSees) is also available. University

of Ljubljana is using these commercial programs mostly for design and seismic assessment of

structures by employing simplified nonlinear models and methods (e.g. N2 method). OpenSees,

which has also been extensively used, has advantages in comparison with the commercially

available software since it provides a comprehensive library of nonlinear elements, material

models, analysis types and solvers. However, usually it does not support the performance-based

assessment prescribed by various structural codes, since it is mostly focused on the research

community. It mainly supports researchers in development of applications for the simulation of

the performance of structural systems subjected to earthquakes. So, OpenSees does not provide

sophisticated graphical input or post-processing capabilities of analysis results. Additionally,

there is also a necessity to develop tools which will be able to further extend the applicability of

software for computational simulation, e.g. for the implementation of methods for the seismic

performance assessment of structures, which will then enable the development of applications

for seismic design of structures. University of Ljubljana has developed a simple performance-

based earthquake engineering (PBEE) toolbox in Matlab, which can be used in conjunction with

OpenSees, and can also serve as a link between the experiments and numerical simulations, as

described in the following.

A PBEE toolbox (Dolšek 2009b) for the seismic performance assessment of reinforced concrete

frames has been developed in Matlab in combination with OpenSees. The aim of the PBEE

toolbox is to enable rapid definition of simple nonlinear structural models of RC frames with

concentrated plasticity. Such nonlinear models are permitted in different structural codes. In such

a case the most time-consuming part of the work involves the determination of the properties of


30

the plastic hinges. Since the PBEE toolbox automatically generates the properties of plastic

hinges, based on data regarding material strength, reinforcement and section properties, the

amount of work which is needed to prepare a structural model is reduced significantly. The

PBEE toolbox enables also the post-processing of the results of analyses and structural

performance assessment with different methods. The user can add new functions to the PBEE

toolbox in order to support additional procedures for the seismic performance assessment of RC

frames, or can just change the rules for determining the moment-rotation relationship of plastic

hinges in columns and beams, which are the main source of uncertainty in simplified nonlinear

models. The PBEE toolbox has already been successfully used for different applications, i.e. for

incremental dynamic analysis with consideration of modelling uncertainties (Dolšek 2009a) or

for the estimation of seismic risk with consideration of capacity degradation over time (Celarec,

Dolsek and Vamvatsikos, 2009).

The PBEE toolbox can be used also for the identification of errors in structural models if the

results of pseudo-dynamic test are available. Usually, only the results of nonlinear dynamic

analysis are compared with the results of pseudo-dynamic tests. In such comparisons the

computed response differs from the measured response in terms of the displacements and shear

forces. It is therefore sometimes not easy to determine why differences occur between the

computed and measured results. For this reason it is convenient to impose the same

displacements on the structural model as those obtained in the experiment. Such an approach

enables a comparison of the computed and measured shear force – storey drift relationships at all

storeys of the structure, as explained in (Dolsek and Fajfar, 2002). In this way it is easy to locate

where the differences appear, e.g. between the measured and computed shear force time

histories. The PBEE toolbox can be used to perform such an analysis using the OpenSees, by

minimizing the amount of data needed. Basically, the user has only to define the files where the

measured storey displacement time histories are stored, and call some functions which

automatically create tcl input files, run OpenSees, and store the results of the analysis in the

Matlab structure for further processing.

For example, a four storey RC frame building (Fig. 3.1), which has been pseudo-dynamically

tested in the ELSA Laboratory, was modelled with the PBEE toolbox by using two types of

plastic hinges (Fig. 3.2). Four files containing the measured displacement time histories were

defined in order to control displacements in the model at each storey. Analysis with imposed


31

displacements was performed by using the PBEE toolbox. The results of the analysis with

imposed displacements are presented in Fig. 3. 3 for the first storey drift versus the base shear.

First, it can be observed that the stiffness of model 1 is higher than that observed in the pseudo-

dynamic test, and also higher than the stiffness of model 2, although the plastic hinges of model

1 are more realistic than those implemented in model 2. Note that cracked sections were assumed

in the case of model 2, whereas the moments of inertia corresponding to the sections of the

beams and columns in model 1 were not reduced. Additionally, the effective slab width, which

contributes to the stiffness and strength of the structure, was assumed to have a value close to the

highest level which was measured in the pseudo-dynamic test. Such an assumption corresponds

to deformations higher than those observed at the start of the high-level test. However, although

the effective slab width was rather large, nevertheless the strength of the structure when

simulated by models 1 and 2 was still slightly underestimated. Similar observations have been

made by other researchers.

The University of Ljubljana recently became a member of the University Consortium of

Instructional Shake Tables (UCIST) which involves more than 100 Universities interested in

improving earthquake engineering education. Within the UCIST small-scale research devices

(Quanser Shake Table II, STII) are used for simulating seismic response and other phenomena of

dynamics of structures. The STII has one degree of freedom with the top movable table of about

50×50 cm. The table is capable of achieving an acceleration of 2.5 g with a mass less than 7 kg.

The maximum displacement of the table is ±7.6 cm from the center position. The instructional

shake table is used for the educational process of students through the preparation of small-scale

structural models and execution of experiments. Students can observe the behavior of structures

during earthquakes as a function of various parameters. In partnership with NEES (National

Network for Earthquake Engineering Simulation) UCIST developed high quality tele-

participation and tele-operation experiments. It is intended to use this system also for

experiments performed at the University of Ljubljana.


32

Fig. 3.1: The four-storey reinforced concrete frame building

Fig. 3.2: Moment-rotation relationship with softening of plastic hinges in columns and beams: a) a four-

linear and b) a tri-linear


33

Fig. 3.3: The base shear versus first storey drift time histories for models 1 and 2 obtained by nonlinear

dynamic analysis and by imposed displacement from pseudo-dynamic test. The computed results are

compared with the experimental results for the high-level test


34


35

4 MATLAB Tools for Structural Identification and for Identification by means of Pseudo‐Dynamic Tests

4.1 INTRODUCTION

This report describes the use of two software tools for the purpose of structural identification and

for the representation of outputs of real-time tests. One of the above-mentioned tools is the

software Matlab (R2008a), and the other is Simulink 3D Animation toolbox (R2012a). In the

first part, Matlab is mainly used in the Fast Impact Hammer Testing (FIHT) for extracting mode

shapes (Daniel Tirelli, 2011). In the second part, Matlab is applied for identification of model

properties from the dynamic response of structures (Francisco Javier Molina, Georges Magonette

and Pierre Pegon). The use of Simulink 3D Animation toolbox is presented in the third part,

which represents the output of a real-time test in a visualized way, emphasizing the drift between

the numerical substructure and the physical substructure on the interface (Giuseppe Abbiati).

Moreover the FIHT method and two identification methods based on spatial model and filter

model are presented.

4.2 STRUCTURAL IDENTIFICATION BY FAST IMPACT HAMMER TESTING METHOD

4.2.1 Introduction

The Impact hammer test is currently used on small structures with little damping in the

mechanical field. Large dimensions and important damping are not recommended to use this

method since it works at low level of energy. But nowadays with the quality of the transducers

and with adequate signal processing, small and medium structures could be easily investigated.

The introduction of the Fast Impact Hammer Testing (FIHT) method takes its place in a field not

yet covered by the different partners. Moreover, the objectives provide partners with an

opportunity to use the method for a final insertion as a part of toolbox built by the team of the

Laboratório Nacional de Engenharia Civil (LNEC). The method will be used for dynamic

characterization of some structures built for the SERIES project.


36

Compared with the classical impact hammer method, the main advantage of the FIHT described

here is that the latter enables to obtain results in a very short time without reducing the accuracy.

The results of some tests on different structures with the FIHT method were compared with those

from commercial tools or algorithms and showed good agreement. The FIHT method was also

used for tests described in the bibliography (Daniel Tirelli & Stefano Primi, 2004; Armelle

Anthoine & Daniel Tirelli, 2008; O.S. Bursi & J. Molina; M. Poljansek & G. Bof, 2009; F.J.

Molina & R. Pascual, 2003). The signal processing, a part of the FIHT method, is implemented

under Matlab and requires the installation of the correspondent FIHT toolbox.

4.2.2 The FIHT Method

The FIHT developed here can give a solution to economically, quickly and accurately find the

main dynamical properties governing the behavior of small and medium structures. Though it

cannot fully substitute the commercial tools, it brings an efficient support to most of the needs in

the dynamic field of civil engineering laboratories due to its high degree of automation of the

procedures.

The main difference between the FIHT method and the classical hammer impact method is the

extensive use of the reciprocity of the FRF (transfer function). The FIHT method only uses one

or two outputs rather than a large number of outputs, as is the case with the classical method. The

signal processing is performed step by step in an automatic way. Several assumptions and

criteria governed the acceptance to the next step. The main limitation was the validity of the

transfer function reciprocity used to fasten the measurement process which in some rare cases

could not be verified. The second important inconvenience could be the incapacity of the modal

extraction procedure to isolate modes very close to each other, leading to operational deflection

shape instead of pure mode shapes.

As the FIHT method is based on the output of only one transducer, a crucial point of the method

is the location of the output transducer. The use of a numerical model provides mode shapes that

are used to define the location of transducers for the modes of interest. In the example of Fig.

4.1, the best output locations are calculated analytically for a bridge specimen (Tirelli, 2011).

They correspond to the highest level of amplitude that can be measured for all the modes


37

requested. The lower the slope around the peak found (as pointed by the black arrow in Fig. 4.1),

the smaller will be the error on the transducer position. In Figure 4.1 the red stars show the

positions of existent displacement transducers used for other purposes. In such case they could

be included in the signal processing to give redundant information of the dynamic behaviour.

Each of these transducers will give a representation of the mode shapes.

Fig. 4.1: Analytical results to find the minimum amplitude of the first six bending modes expected along

the beam

4.2.3 Structure Description and Methodology employed

This example of medium structure dimensions was built with both classical and new type of

materials, namely concrete + composite fiber. As the structure has rather small dimensions and

weight, the faster and the most efficient method to get accurate results on the dynamic behavior

of the frame was the FIHT method.

In order to quickly evaluate if the hammer Impacts will be sufficient to excite the main modes of

the structure, the ratio between the dynamic impact force (measured) and the weight (force) of

the structure are calculated. In this case, a typical ratio is about (6000N/310 000N) ~1.94% (a

ratio greater than 0.1 to 0.5% should be sufficient), therefore the FIHT should be well adapted.

4.2.4 Signals Processing

In frequency domain, means of the same measurements are very efficient to remove the noise

results from signals. A short pre-test on the structure shows that in that case, five means are

sufficient to cancel the noise efficiently. The total number of signals generated by the impacts for

testing all the structures is important and equal in our case to: 5impacts 24positions

8 output input transducers 960signals. ‼!


38

Because all of them make a contribution, in a small part, to define the correct modal parameters

values, an automatic signal processing is mandatory to check the validity of the overall signals

and the transducers operation with very few figures. In this example, as 24 points are tested, we

obtain 24FRFs, means of 5 repetitions each. The displays of 960 time history signals which is

common for a structure of these dimensions, is not really adapted for the time needed to inspect

each of the signals. Furthermore, even when the time history showing that the transducers are

working, it cannot show the correlation between input and output. The solution is to adopt an

automatic check of the signals, illustrated in Fig. 4.2.

Fig. 4.2: Example of the mean FRF of one point position on right, the respective coherence function on

left

In Fig. 4.3, an example of a FRF and its respective coherence function is plotted. The pink

circles correspond to problems of signal/noise ratio which can be removed by FRF filtering. Red

circles represent real problems which cannot be removed by automatic check. It represents on

left, the cut off frequency of the hammer (at very low frequency), and at right the filter of the

acquisition system which was chosen at 200 Hz for these measurements. It is then normal that

the coherence values included in the red circle are taken for the processing as they show a real

problem.

For this bridge example, it is very time costly to measure 24 FRF for each Output transducer(7),

which corresponds to (24×7=168) coherence figures to observe. In the FIHT we proposed a

shortening of these operations, condensing the information in two check levels.


39

Fig. 4.3: Coherence histogram of the FRF on raw data and on position

The first is the histogram representation of the coherence function dividing the coherence values

in bins (default =20) and displaying the histogram of the points distribution for all the Output

transducers (OT=7) on the same graph as it is shown in Fig. 4.3. With this example, we highlight

the fact that the hammer induces very small deformation on the structure but not obviously

inaccurate displacement.

Coherence histograms made on raw data are not always adapted to perform an automatic

processing. Processing raw data could conduct to discard measurements which are not erroneous.

From picture 3 we can draw the conclusion that only the part of the FRF with high signals-noise

ratio has to be taken into account in the coherence histogram processing. In the FIHP method,

points the level of which is lower than three orders of the maximum magnitude of the FRF

processed are removed because they represent measurements highly affected by noise and

therefore with poor coherence. Now the filtered coherence histogram has the appearance in Fig.

4.4. This step achieved to process data of coherence could be submitted to a statistical criterion

which is respected for the acceleration signals.

Fig. 4.4: Example of distribution of the coherence values for on test and for all the transducers


40

The second level of automation could be performed to reduce in one graph the check of all the

points position tested. In Fig. 4.5, the spatial distribution of the impacts of the example of the

PROMETEO beam are shown, where each stick length represents the mean of the filtered

coherence.

In this way the user can understand quickly the overall test validity. The user can then verify for

the position if one of the time signals has to be rejected.

Fig. 4.5: Example of the map of the filtered coherence for 2 types of transducers: displacement (a), accelerometer (b)

The last step of signal processing, before the modal parameter extraction, is the representation of

the FRFs amplitude in the direction of the motion measured. As the FIHT used mainly the

hammer impact for input, the generated displacement is small, so it is rather convenient to

represent the acceleration FRFs to get visible mode shape of high frequency. It is important to

have a display of the curves for the FRFs magnitude for a number of positions which have the

highest level of signal in the total frequency band analyzed, as from each of the single FRFs the

values of the frequency, damping and mode shapes are extracted. Another way to read the curves

is to represent the level of spectral energy distribution in a defined frequency band. The map of

the spectral energy is automatically plotted. The dynamic behavior is then easier to understand

by this kind of mapping. If the material is dispersive or if the structure is old, the FRFs

magnitudes measured must be multiplied by a coefficient:


41

⁄

where i is the input position and j is the output position; d is the propagation distance of wave

between i and j; α is the attenuation coefficient of the material; ω is the angular frequency; c is

the wave velocity in the material. For steel structures, this coefficient could be neglected.

4.2.5 Mode Shapes

The final step of the method is to represent the mode shape. In the FIHT method, each

experimental point P of the structure for each mode m, when subjected to vibrations, takes the

following coordinate at each time step t, with 1 1 chosen for the mode shape animation.

P , t pointcoordinateatrest X , Y , Z A ∙ M ∙ sign I , t

where A is the amplification factor of the dynamic part (2nd member) of P. And the default value

of A is 200. In Fig. 4.6, the red arrow and point, shows which values are taken for the mode

considered.

All the information from the process of modal extraction is stored in matrices of Matlab. Here is

an example of the command of extraction to show how easy the processing is for user which can

change one parameter to get the results expected.

Fig. 4.6: Example of mode shape value calculation from the FRF complex representation at one position for the mode m

[NFR NDA NAM FRFP NCOL FDM]= modal (Vec, Nbm, Crm ,Four,Fi);

INPUTS: (the 3 first are compulsory)

- Vec=list of FRF measured a nd stored in vectors,

- Nbm=nb. of modes searched in each FRF


42

- Crm = criteria on modes =in how many FRFs we expect the presence of the modes. In other

words, peaks at same frequency (or nearly) are considered modes is they are observed in at least

Crm FRFs or positions.

The following parameters are optional:

-Four = dispersion range requested in % if different of the default (=[0.1 to 10%])

-Fi= filter value (if different of the default, to extract, by example closed modes (define in

point number)

OUTPUTS parameters:

Matrix of Peaks frequencies (NFR), damping (NDA), amplitude (NAM), address (FRFP),

NCOL =nb. of the columns which respect the criteria Crm,

FDM=matrix of frequency damping and standard deviation of each mode found.

To fasten the work for the user who has to design the structure, two simple functions in

MATLAB are implemented for the structure at rest (see annex 1). The display of the structure

and of the mesh of all the FRFs positions and surface is then automatically generated when the

coordinates and the direction of motion are previously inserted in a matrix of coordinates. An

example of mode shapes obtained by the FIHT method for this structure is illustrated in Fig. 4.7.

The picture presents the maximum curvature (the software stops its animation at this point and

save the picture).

Fig. 4.7: Mode shape of a flapper mode of the bridge edges in Y direction

The modes are very well identified from the different shapes as it is shown in pictures which are

not presented here. And the pictures also highlight how the FIHT is accurate, as the

measurements show perfectly the exact boundary condition. The accuracy of the method is still

well proved by the regularity of the acceleration mode shape with respect to the displacement

mode shape.


43

Finally, the main achievements of the FIHT method are:

For the tests, it is the use of the reciprocity of the transfer function.

For the processing, it is the use of coherence histogram coupled with the FRF filtering and

mapping, which introduce a novelty in the field. Finally the modal procedure extraction is

drastically reduced and eased by the use of criteria and statistic treatment based on a large

campaign of experimental modal analysis.

The FIHT method modal may improve the experimental methods used in non destructive testing

of civil structures.

4.3 SOFTWARE FOR IDENTIFICATION OF MODAL PROPERTIES FROM DYNAMIC

RESPONSE OF STRUCTURES

4.3.1 Introduction

Two different time-domain identification methods and the correlative software are presented.

The identification of the modal parameters is from results of dynamic response which is typically

done for the pseudo-dynamic tests performed in the European Laboratory for Structural

Assessment (ELSA) (Pegon et al., 2008). Both methods have been successful in the

identification of the natural frequencies, damping ratios and mode shapes of real-size structures.

Two time-domain identification methods, based on two different linear time-invariant models are

described in this chapter (Molina et al., 1999, 2011ab, 2012).The first method is based on the

spatial model in which the stiffness and damping matrices are directly identified from the

measured or the integrated displacements, velocities and restoring forces. Using the combination

of the obtained matrices and the theoretical mass, the eigenfrequencies and modes can be

obtained. The second method is formulated as a filter model of specified order in which the

ground acceleration or external forces acts as input and the experimental displacements act as

output. The eigenfrequencies and modes are obtained from the identified coefficients of the filter.

This method is suitable for dynamic experiments in which the restoring forces cannot be

measured. The filter model identification is also applied to PsD results.


44

4.3.2 Identification of Response Frequency and Damping by a Spatial Modal

This technique is based on the identification of an equivalent linear model in terms of stiffness

and viscous damping matrices of the structure using a short time window of the experimental

response. Within the spatial model, it is assumed that the measured restoring forces, the

corresponding displacements and the velocities are linked, for every discrete time n of the

original accelerogram, in the form

1 (1)

where r(n), d(n) and v(n) are the results of the dynamic or PsD test, K and C are unknown

matrices of stiffness and damping and o is an unknown vector of offset (residual) forces. Once K,

C and o have been estimated by a least squares solution, the complex eigenfrequencies and mode

shapes can be obtained by solving the generalised eigenvalue problem (Ewins, 1984)

00

00 (2)

where M is the theoretical mass matrix, the complex conjugate eigenvalue couples can be

expressed in the form

* 2, ( 1 )i i i i is s j

(3)

where ω is the natural frequency and ζi the damping ratio. The corresponding i th mode shape is

also given by the first DoFn rows of the associated eigenvector iφ .

In order to use the method for the assessment of the consequences of the control errors on the

PsD response, the identification process is repeated for two sets of variables entering in the

model of equation (1), that is,

- First set: measured forces, measured displacements and derived velocities,

- Second set: measured forces, reference displacements and derived velocities.

The first set of variables takes into account the measured forces and displacements on the

specimen so that the identified stiffness and damping matrices, frequencies and damping ratios

from this set are considered the real ones of the specimen not affected by the control errors. The

second set, the considered forces and displacements are the ones entering in the PsD equation

and the eigenfrequencies and damping linking them are the ones that explain the test response.


45

4.3.3 Identification of Response Frequency and Damping by a Filter Modal

Let be the input and the output of a system. A filter model of orders and can be

defined by the constant-coefficient difference equation (Hayes, 1996)

1 ⋯ ⋯ 4

Formulating the matrix equation

1 ⋯ ⋮ ⋯ ⋮ 5

In which, a constant offset term o has been added.

And the equation will be adapted for a system containing inputs, , and outputs,

⋯ ; ⋯

This filter model (5) may be estimated from the input and output data at N time instants if the

number of equations is equal or larger than the number of unknown coefficients. Once the

unknown coefficients are estimated, the free response of the system is defined as

1 ⋯ ⋮

This expression, by successively transposing, recursively substituting and eigenvalue

decomposing, can be abbreviated as

1 0 0

In which, ⋯

Contains the eigenvalues of A and V contains the eigenvectors. The conjugate couples of poles

can be written as

, ∗ ∆

where and are the same as in the spatial model. But the i th mode shape is given by the first

nout rows of the associated eigenvector.

To apply this identification method to the results of a PsD test, the displacements may be taken

as outputs while the external forces in the equation of motion would be taken as inputs. The

number of output variables does not necessarily need to be equal to the number of DoFs. For a

passive linear mechanical system, the theoretical order of the filter should be two. In other

situations, for example if the number of DoFs is larger than the considered number of outputs, a


46

higher order filter model is recommended. To solve the same problem, different orders can be

chosen to make more candidates as physical eigenfrequency, among which the best one will be

picked due to the stability.

4.3.4 MATLAB Functions for Identification of the Models

Several MATLAB functions have been created for the identification of spatial and filter models

and for their related modal parameters from dynamic response of experiments –or even numeric

simulations. These functions can be freely downloaded as a package from the MATLAB server

(Molina et al, 2011b).

A two storey masonry house example of use of the functions on the PsD experiment is presented

in the following.

Within the activities of the ESECMaSE project, masonry house models were tested at the ELSA

laboratory [ESECMaSE project, 2009, Anthoine and Molina, 2008]. PsD seismic tests, taking

into account two DoFs, were conducted on this real-size model of a 2-storey terraced house.

For the spatial model:

The reliability of the obtained PsD response assessed by means of the previously mentioned

checking of the identified frequency and damping ratio is shown in Fig. 4.8 for 0.02g test. The

results are acceptable even though at the beginning and the ending parts some negative values

are shown.


47

Fig. 4.8: Spatial model frequency and damping ratio identified values for both modes

For the filter model:

The identified frequency and damping ratio from the data of the same experiment is shown in Fig.

4.9. The lines are plot for different orders. This is shown to better compare with the spatial

model.

Fig. 4.9: Filter model frequency and damping ratio identified values for both modes

From this test, with two DoFs, in terms of ground accelerogram, measured and reference

displacements, restoring forces and time variable can also be obtained.


48

The commands to perform the identifications and to produce the graphs can be obtained by

asking “help spatial_model_hist” or “help filter_model _hist”.

4.4 USE OF SIMULINK 3D ANIMATION TOOLBOX FOR THE REPRESENTATION OF

REAL‐TIME TESTING OUTPUT

4.4.1 Introduction

A series of results can be obtained from the hybrid test by using the parallel LSRT2 and

improved parallel LSRT2 methods. There are several ways to make the outputs clearer, more

visualized and more understandable. The approach proposed here is to exploit the Simulink 3D

Animation toolbox to represent the outputs of a real-time test and to visualize dynamic system

simulations. A procedure is described in order to give an understanding of the process of

applying 3D animation toolbox.

Simulink 3D Animation provides an interface linking simulink models and MATLAB algorithms

to 3D graphics objects. It enables to visualize and verify dynamic system behavior in a virtual

reality environment. Objects are represented in the Virtual Reality Modeling Language (VRML),

an open 3D modeling standard. You can animate a 3D world by changing object properties such

as position, rotation, and scale during desktop or real-time simulation. We can also access 3D

animation data in Simulink or MATLAB for post-processing.

4.4.2 Simulink 3D Animation Toolbox

Simulink 3D Animation includes a viewer for rendering and recording high-quality animations.

With the 3D World Editor, we can author detailed scenes assembled from 3D models exported

from CAD- or Web-based sources. We can incorporate multiple 3D scene views inside

MATLAB figures and interact with these views via a force-feedback joystick, space mouse, or

other hardware device. The key features of Simulink 3D Animation are:

• Simulink blocks and MATLAB functions for connecting models to virtual reality worlds

• 3D World Editor for authoring 3D worlds

• Video recording and animation playback


49

• Visualization of real-time simulations

• Client/server architecture

• Interaction with 3D views via a joystick, space mouse, or other hardware device

To use virtual reality worlds to visualize dynamic system simulations, these tasks should be

followed:

• Set up Your Working Environment

• Build a Virtual Reality World

• Link to a Virtual Reality World

• View Dynamic System Simulations

• Share Dynamic System Simulation Visualizations

4.4.3 The Representation of Real‐time Testing Outputs

Fig. 4.10a shows the results and the representation for the outputs of real-time test by using the

PLRT2 method with a time step of 4ms; Fig. 4.10b shows the presentation with a time step value

of 2ms. In the simulink 3D model, the pink part acts as the numerical substructure (NS), and the

blue part acts as the physical substructure (PS).

a)

b)

Fig. 4.10: a) Results and the representation for the outputs of PLRT2 with a time-step of 4ms b) Results and the representation for the outputs of PLRT2 with a time-step of 2ms

0 5 1 0 1 5 2 0 2 5-0 . 0 1 5

-0 . 0 1

-0 . 0 0 5

0

0 . 0 0 5

0 . 0 1

0 . 0 1 5

y n

y e

0 5 10 15 2 0 2 5-0 .015

-0 .01

-0 .005

0

0 .005

0 .01

0 .015

y n

y e


50

The results and the representation for the real-time test from the improved PLSRT2 with a time-

step value of 4ms are shown in Fig. 4.11.

Fig. 4.11: Results and the representation for the outputs of improved PLRT2 with a time-step of 4ms

Figures 4.10 and 4.11 indicate that, sometimes, only from the plotting curves of outputs, the

results could not be seen in a clear and visualized way. The representation obtained from the

Simulink 3D Animation could let us vividly understand the different drifts between the NS and

the PS on the interface and it can emphasize the drift in the solution.

The procedure of simulating the dynamic behavior by using Simulink 3D Animation is shown in

Fig. 4.12.

Fig. 4.12: The procedure for simulating the real-time testing output with Simulink 3D Animation

Simulink 3D Animation toolbox can not only be used for dynamic system, it can also be used for

other systems, such as the mechanical system.

0 5 1 0 1 5 2 0 2 5-0 . 0 1 5

-0 .0 1

-0 . 0 0 5

0

0 . 0 0 5

0 .0 1

0 . 0 1 5

y n

y e


51

4.5 CONCLUSIONS

The FIHT method was introduced in order to emphasize several aspects:

1) difference between the FIHT method and the classical impact hammer method;

2) signal processing;

3) extraction of mode shapes;

4) advantages and limitations of the FIHT method.

Two different identification methods based on spatial model and filter model have been

presented. The results of the FIHT method and the two different identification methods were

good enough with respect to their respective physical tests. The MATLAB was very suitable for

the identification of model properties from the dynamic response of structures. Application of the

Simulink 3D Animation toolbox has provided a novel approach to represent the output of

dynamic tests.


52


53

5 Software Developments for the Implementation of Partitioned Algorithms and their Interaction with OpenSEES

5.1 INTRODUCTION

The PM method is an interfield parallel partitioned integration algorithm formulated for

continuous Pseudo Dynamic testing whose favourable stability and consistency properties were

proved by Bonelli et al. (2008). The strong nonlinear NS is simulated by means of the well-

known OpenSEES FE software. Computational burden is not compatible with the fine time step

of the controller if the NS and algorithm are available from the same OpenSEES FE framework

resulting in such a monolithic algorithm failed to complete a complex NS characterized by

several DoFs. In order to solve this problem, MATLAB software and Simulink model are

employed to implement the PM method, whilst the OpenSEES framework is exploited to

simulate the NS. Moreover, the interface between MATLAB and OpenSEES is investigated. Pre

processing of input data and post processing of the displacement results are presented in this

Section.

In order to overcome the basic shortcoming that methods belonging to the subspace family often

produce spurious modes, the use of the Structural Dynamic Identification Toolbox code (SDIT)

originally conceived at the Department of Structural Engineering of Politecnico di Torino to

provide a complete framework for experimental modal analysis (R. Ceravolo, G. Abbiati, 2009)

is presented in the following sections.

5.2 NUMERICAL AND EXPERIMENTAL VALIDATION OF THE PM METHOD COUPLED

TO OPENSEES

5.2.1 Introduction

A full scale testing program is foreseen in the RETRO TA project, a research program of the

European SERIES project (Taucer 2011). The case study consists of an old concrete viaduct


54

where two independent roadways are supported by 12 couples of portal piers. Two isolation

systems, yielding-based and friction-based bearings, were designed and characterized. Initially,

the OpenSEES (2009) FE model of the bridge is described. Then, hybrid simulations with

dynamic substructure are selected to prove the effectiveness of the seismic retrofit. The PM

interfiled-parallel integration algorithm (Pegon and Magonette 1992, Bursi et al. 2008) is

adopted to apply the continuous-time testing method. Two implementations of the PM method

based on OpenFRESCO (2009) are proposed. The connection between the integration algorithm

and both the NS and PS is carefully analyzed.

Moreover, fast hybrid simulations are necessary because the isolation devices might be subjected

to strain rate. To this end, two feasible delay overcompensation strategies are suggested (Wu et

al., 2012). Finally, a way to take into account the non-linear behavior of piers belonging to the

NS during testing is proposed.

5.2.2 Preliminary Numerical Simulation

One of the two roadways of the viaduct was modeled by means of OpenSEES (Paolacci and

Giannini, 2011). Nonlinear fiber beam elements were adopted and only the flexural behavior was

taken into account. Both the non-linear shear behavior of transverse beams and the influence of

fix-end rotation effects owing to strain-penetration of steel bars were neglected. Some mode

shapes corresponding to periods of 1.15s and 1.01s are shown in Fig. 5.1.

Fig. 5.1: FE model and mode shapes of the Rio Torto viaduct: a) period T=1.15s; b) period T=1.01s

5.2.3 The PM Method

The PM method is an interfiled parallel partitioned integration algorithm originally formulated

for continuous Pseudo Dynamic (PsD) testing by Pegon and Magonette (2002); it allows for the


55

coupling of implicit and explicit time integrators dealing with different time-step lengths for each

subdomain. Its favourable stability and consistency properties are numerically and

experimentally proved by Bonelli et al. (2008). Fig. 5.2 shows the task sequence for the case

with substepping, ss=2.

Fig. 5.2: Task sequence of the PM method

The strong nonlinear NS is simulated by means of the well-known OpenSEES FE software.

Computational burden is not compatible with the fine time step of the controller if the NS and

algorithm are available from the same OpenSEES FE framework resulting in such a monolithic

algorithm failed to complete a complex NS characterized by several DoFs. Therefore, two ways

to take advantage of the partitioned PM method without any change of the original source code

of OpenFRESCO are proposed herein:

Implementation #1 – the MATLAB software is employed to implement the PM method, whilst

the OpenSEES framework is exploited to simulate NS; the OpenFRESCO manages the data

transfer between MATLAB and the experimental control system.

Implementation #2 – a Simulink model of the PM algorithm is implemented on a real-time xPC-

Target machine (Mathworks, 2012), whilst the OpenSEES framework is exploited to emulate

NSs on the Host-PC.

OpenFRESCO connects finite element models with control and data acquisition systems to

standardize the deployment of such tests on structural systems.

5.2.4 Implementation

Implementation #1:

By using OpenFRESCO, trial displacements calculated with PM algorithm by MATLAB are

sent both to the PS and NS, respectively, and the corresponding reaction forces are measured


56

back to MATLAB to calculate the next displacement command. To this end, the middle-tie

server architecture is adopted. The corresponding experimental equipment is shown in Fig. 5.3.

MATLAB and OpenFRESCO codes are implemented in the Host-PC, and the Predictor –

Corrector model runs on the xPC-Target real-time machine. The NS is analyzed by the

OpenSEES framework on the Host-PC whilst the PS is handled by MTS controller. The xPC-

Target links the Host-PC and the MTS controller.

Fig. 5.3: Arrangement of the experimental equipment

Fig. 5.4: Block diagram

Fig. 5.4 shows the implementation block diagram of PM method with OpenFRESCO. The xPC-

Target accesses the Host-PC through an Ethernet connection and runs a Simulink predictor-

corrector model. The MST controller was connected with xPC with Shared Common RAM

Network (Scramnet) so that the control signal written locally to the xPC-Target is

instantaneously copied to the controller; the measured signal written locally to the controller is

instantaneously copied to the xPC-Target.

This implementation was tested on a split-mass SDoF system at the University of Trento shown

in Fig. 5.5a, and the corresponding specimen set-up is shown in Fig. 5.5b. The time scale with

values of 10 and 1 were chosen respectively for a PsD and a fast-time test.

Fig. 5.5: a) Emulated SDoF system; b) Test set-up of the PS


57

The properties of the analyzed system are, 5.31 10 , 2.10 10 / , 1.94 10 / ,

3.34 10 /

An equivalent numerical damping of 5 percent was added to the system.

For the PsD test, the displacement response and the reaction force are shown in Fig. 5.6 and Fig.

5.7.

From Figures 5.6 and 5.7, it can be seen that the numerical solution agrees with the physical one

for both the displacement response and the reaction force, though the signals were characterized

by a greater noise than the numerical solution. Also a fast time test was conducted. See Abbiati

et al. (2012) for further details.

For the fast time test, the displacement response and the reaction force are shown in Fig. 5.8 and

Fig. 5.9.

Fig. 5.6: Displacement response at the interface DoF and relevant zoom

Fig. 5.7: Reaction force at the interface DoF and relevant zoom


58

Fig. 5.8: Displacement response at the interface DoF and relevant zoom

Fig. 5.9: Reaction force at the interface DoF and relevant zoom

Implementation #2:

A simulink model of the PM algorithm is implemented on a real-time xPC-Target machine,

whilst the OpenSEES framework is exploited to emulate the NS on the Host-PC. And the

Generic Client Element Simulink block provided by OpenFRESCO manages the data transfer

between the xPC-Target and the Host-PC. The relevant arrangement is shown in Fig. 5.10. In

contrast to the previous implementation, there is no need to employ any predictor-corrector

algorithm.

Fig. 5.10: Arrangement of the experimental equipment

Fig. 5.11: Block diagram

The PS is handled by the MTS controller. Fig. 5.11 describes the implementation of the second

architecture proposed by means of a block diagram. The TCP\IP protocol manages the data

transfer between the xPC-Target, the MTS controller and Host-PC.


59

5.3 MATLAB INTERFACE TO OPENSEES

5.3.1 Generic Description

Since the OpenSEES Finite Element (FE) framework is not provided with a Graphic User

Interface (GUI), a set of methods for the pre and post-processing of OpenSEES model and

relevant results are developed within the MATLAB framework. The whole library is written for

OpenSEES 2.3.2. The MATLAB interface communicates with OpenSEES by means of self-

generated TCL code.

5.3.2 Input Check Capabilities

The present object oriented toolbox can be used to check the data input of a generic FE model.

The input data, such as node coordinates element tables, is converted into a DXF drawing that

can be easily compared with the original CAD picture of structures. Pier 6 plotted by DXF is

shown in Fig. 5.12.

Fig. 5.12: DXF plot of the pier 6 of the Rio Torto viaduct

5.3.3 Post Processing Results

In order to easily animate displacement results from time-history modal analyses, similar

software is adopted which is completely based on the standard MATLAB library. So third-party

software is not needed. Moreover it exploits the capabilities of the graphic objects like lines,

points, patch, splines etc. provided by Mathworks in order to create complex pictures and movies.


60

The geometrical models are described by means of a few Excel tables containing node and

element coordinate matrices. Fig. 5.13 shows the two first deformed modal shapes of the Rio

Torto Viaduct FE model:

Fig. 5.13: FE model and mode shapes of the Rio Torto viaduct: a) period T=1.15s; b) period T=1.01s

A comparison of displacement results between two different FE models of a structure, a pier of

the Pri Torto viaduct, is shown in Fig. 5.14. Both the simulated models are conducted by means

of the above mentioned MATLAB script.

Fig. 5.14: Displacement response of a) pier 8 and b) pier 9 of the Rio Torto viaduct

5.3.4 Sensitivity Analysis and Optimisation

With the proposed interface, it is possible to manage batch analysis directly from the MATLAB

environment which makes it very easy to perform sensitivity analysis and to solve optimization

problems by using OpenSEES as FE solver. It must be emphasized that MATLAB code is fully

compatible with OpenSEES FE fiber models. This PBEE toolbox has been stated in detail in

chapter 3 by Matjaz Dolšek.


61

5.4 STRUCTURAL DYNAMIC IDENTIFICATION TOOLBOX

Structural Dynamic Identification Toolbox code (SDIT) implemented at the Department of

Structural Engineering of Politecnico di Torino, provides a complete framework for experimental

modal analysis (R. Ceravolo, G. Abbiati, 2009). It operates on acceleration records which can be

pre-processed by subsampling, detrending, low-, high-, and band-pass filtering etc. Fast Fourier

Transform and Welch Power Spectral Density plots can be easily produced. Fig. 5.15 presents

the SDIT graphic user interface.

Fig. 5.15: SDIT Graphic User Interface

Spectrograms for time-frequency domain investigation can be calculated. Fig. 5.16 provides an

example of spectrogram that can be computed during the signal pre-processing task. This kind of

plots is very helpful to check the presence of non-persistent spurious frequencies.

Fig. 5.16: Spectrogram for time-frequency domain


62

The SDIT identification code implements output-only identification techniques in the time

domain in order to determine the modal parameters of a system from its structural response.

Among the deterministic methods, the ERA (J.N. Juang, R.S. Pappa, 1984) is based on a Single

Value Decomposition of Hankel’s matrix and has been extensively studied in the literature. ERA

is typically applied to impulse response signals, though in the case of stochastic input it is

customary to replace them with output correlation functions. The time-domain family of SSI

methods stems from Ho and Kalman’s classical realization theory that was extended to stochastic

systems by Akaike and Aoki. Van Overschee and De Moor (P. Van Overschee, B. De Moor,

1996) collected and systematically linked together contributions from the following different

fields: system theory, statistics, optimization theory and linear algebra. The quality essential to

all the algorithms within the subspace family is their ability to work out the matrices that

describe a linear system starting from subspaces that contain the projections of data matrices. In

particular, these algorithms project the space of the matrix rows of future outputs into the space

of the rows of past outputs. One of the basic shortcomings of these methods is that they often

produce spurious modes, whose true nature, however, can usually be identified by means of

simple modal form correlation indicators, or, as an alternative, with the aid of numerical models.

It is possible to discard all the non-recursive mode shapes by means of statistical recurrence

analysis procedure. In order to check the reliability of the identifications, tools for generating

cluster and stabilization diagrams are supplied which are illustrated in Fig. 5.17 and Fig. 5.18.

Fig. 5.17: Cluster diagram

Fig. 5.18: Stabilisation diagram

Computational modes were systematically discarded by using modal assurance criteria (R.J.

Allemang, 2002). In greater detail, all signals coming from different acquisitions were

segmented and a large number of SSI identification sessions were performed. Stabilization

diagrams were used to identify frequencies in each segment (stabilization criterion: maximum

frequency deviation: 2%); then additional tolerance criteria were used for Modal Assurance


63

Criterion (5%) and for damping (0pxp20%). By performing a statistical recurrence of the system

natural frequencies identified by the SSI algorithm and by averaging values, the distinctions was

proved between the authentic modes of the structure and modes that appeared occasionally. Such

outcome was possibly due to the exogenous components.

With reference to time-frequency analysis, instantaneous estimation of modal parameters

capabilities is provided (R. Ceravolo, 2004). In this way, the equivalent viscous damping and

frequency can be estimated along with the full acceleration signals. Fig. 5.19 shows the

instantaneous estimation of frequency, damping and amplitude of the response characterizing the

first eigenmode of the Pescara Bridge subjected to environmental excitation.

Fig. 5.19: Instantaneous estimation of the modal parameters of the 1st eigenmode of the Pescara Bridge

This identification technique shows robustness also when nonlinearities or non-stationary input

occur and thus the identification strategies conceived for linear systems fail.

5.5 CONCLUSIONS

Without any change of the original source code of OpenFRESCO, two strategies that take

advantage of the partitioned PM method are proposed. With these implementations, the

numerical solution agrees with the physical one both for displacement responses and reaction

forces. With MATLAB software, the input data to OpenSEES is animated and easy to be

examined; and the model analysis results from OpenSEES are clear and vivid to be read.


64

By means of the Structural Dynamic Identification Toolbox code (SDIT), acceleration records

can be preprocessed by subsampling, detrending, low-, high-, and band-pass filtering etc. Fast

Fourier Transform and Welch Power Spectral Density plots can be easily produced. This

identification technique shows robustness even when nonlinearities or non-stationary input occur.


65

6 Structural Health Monitoring

6.1 INTRODUCTION

In Civil/Structural Engineering, the majority of Structural Health Monitoring (SHM) applications

are directed towards studying the response and damage from natural hazards, such as

earthquakes and strong winds. The monitoring typically involves measuring continuously the

vibrations of the structure by acceleration sensors. Some recent applications have also included

GPS sensors, which provide superior accuracy for measuring displacements. Although a

significant number of structures are now installed with SHM systems, the utilization of data for

practical applications is still lacking. Some of the new findings resulting from SHM include the

significant influence of environment on structural frequencies and damping, strong dependency

of damping on amplitude and frequency, exponential decay in modal damping values with

increasing building height, and the prevalence of 3D modes and non-proportional damping. A

critical need in SHM is the simple tools and techniques for real-time data analysis and

interpretation. Since data come continuously, the analysis cannot be done in batch mode; it

should be done in real-time. This section summarizes the latest developments in SHM, including

some new techniques for data analysis and damage detection.

6.2 JUSTIFICATION FOR SHM

Recently developed seismic and wind design codes for tall buildings all require installation of

monitoring systems in those buildings. The seismic and wind design codes for tall buildings for

Istanbul and Dubai, which are developed at the request of the Istanbul and Dubai Municipalities

by the Department of Earthquake Engineering of Kandilli Observatory and Earthquake Research

Institute of Bogazici University in Istanbul require minimum eight acceleration sensors in

buildings with heights above 75m (DEE-KOERI, 2008a; 2008b; 2009a; 2009b). The reason for

eight channels is that a tall building primarily vibrates in four dominant modes: horizontal

translations in the two orthogonal directions, rotations with respect to vertical axis, and a rigid

body rocking around the base of the building. It can be shown that eight uni-axial sensors can

capture these modes with a senor layout as schematically shown in Figure 6.1.


66

Fig. 6.1 Dominant vibration modes of a tall building Fig. 6.2 Schematic representation of the changes in and the corresponding eight-sensor monitoring system frequency and damping with vibration amplitude

Considering the relatively low cost of monitoring in comparison to the overall cost of a high-rise

building, a SHM system for a typical high-rise building should have many more sensors than

eight. The alternative tall building seismic design code prepared by the Los Angeles Tall

Buildings Design Council recommends the minimum number of sensors to be 15 channels for

buildings between 10-20 stories, 21 channels for buildings between 20-30 stories, 24 channels

for buildings between 30-50 stories, and 30 channels for buildings above 50 stories (LATBSDC,

2008).

In general, extreme loads do not occur frequently. Therefore, most of the data collected by a

SHM system are the vibrations of the structure caused by ambient forces. Low-amplitude

vibration data generated by ambient forces or small excitations provide a means to predict

behavior under large excitations. This involves the following steps:

1. Develop a linear analytical model of the structure calibrated by the vibration data from the

small earthquake.

2. Estimate ground input for the large earthquake by extrapolating the recorded ground input

from the small earthquake.

3. Estimate the response to the large earthquake by using the analytical model and allowing

nonlinear behaviour.

As the amount of data from instrumented structures are increasing, it is now possible to find

sufficient number of structures that have multiple sets of data under different levels of excitations.

Such data would allow studying the correlations of modal characteristics with vibration

amplitudes for different structural categories. Figure 6.2 shows schematically how these

correlations would look like for natural frequency and damping.


67

6.3 DAMAGE DETECTION

6.3.1 Damage Detection Based on Change of Natural Frequencies

It is natural to use the change in natural frequency as a damage indicator because the natural

frequency and the damping are the only parameters that are needed to describe the response of a

structure. Damage detection typically involves analyses of acceleration response data from a

damaging event in order to see if there are any changes in the structure’s natural frequencies.

However, the dynamic response of a damaged structure is nonlinear and in most cases hysteretic,

as schematically shown in Figure 6.3. The stiffness, and consequently the natural frequencies,

rapidly change during the damaging vibrations and are hard to track for short-duration, transient

loads such as earthquakes.

Fig. 6.3: Hysteretic force-deformation curves for damaged structures

Moreover, natural frequencies of a structure can also change due to soil-structure interaction and

environmental factors. A study from the Millikan Library building at Caltech, Clinton (2004) has

shown that there is a strong correlation between the changes in the natural frequency and the

rainfall. And multiple sets of earthquake records (http://nsmp.wr.usgs.gov/) from a 40-story steel

building in Los Angeles have shown that small nonlinearities, which are always present in

buildings, and the variations in damping can also cause changes in the observed frequencies.

Only the frequencies near the fundamental frequency are shown in Figure 6.4. The figure

confirms that there are significant shifts in the fundamental frequency, although the building did

not suffer any damage (Safak, 1995). But in some buildings, although it is damaged, no changes

in the frequency can be observed from the records (Trifunac, 1999). Analytical studies also

confirm the unreliability of using frequency changes for damage detection.


68

Fig. 6.4: Foundation-to-roof transfer functions of a 40-story steel high-rise building during six earthquakes

6.3.2 Damage Detection Based on Permanent Change in Geometry

An important characteristic of hysteretic behaviour is that the structure shows permanent

deformations, such as permanent displacements and/or rotations, after the earthquake. Unlike the

trigger-based monitoring, the continuous monitoring can detect permanent deformations. This is

accomplished by comparing pre- and post-earthquake ambient records. Analyses of pre- and

post-earthquake records, along with the earthquake records, provide a more reliable approach to

damage detection.

The earthquake-induced damage in a structure can be detected by continuous monitoring, based

on two criteria. In terms of signal properties, these criteria correspond to the following: (1) the

spectral characteristics of the signal change during the earthquake; (2) the mean values of the

signal before and after the earthquake are different. Accelerations are not the best quantity to

measure when trying to detect permanent displacements and rotations, which can best be

measured by special sensors, such as GPS sensors and tiltmeters. Once the presence of

permanent displacements and rotations are confirmed, the question becomes whether they

represent damage or not. Statistical hypothesis tests can be used to make such decisions (e.g.,

Lehmann, 1959).

Applications of the tools (Durbin, 1959; Burg, 1968; Griffiths, 1977; Widrow and Stearns, 1985;

Brammer and Siffling, 1989) to real-time vibration data from structures are outlined in Safak

(2004). Adaptive filters and Kalman filters are more appropriate for real-time data. Instead of

monitoring the changes in the frequencies and damping of the structure, it is easier and faster to

monitor the changes in the parameters of such filters. In an automated system, it is advisable to


69

have two parallel adaptive identifications because the change in signal characteristics would

occur much faster during a damaging earthquake than during ambient vibrations. One with a

longer time window detects slow changes and the other with a shorter time window detects the

sudden changes.

6.3.3 Damage Detection Based on Wave Propagation Characteristics

The vibrations of structures under dynamic loads can be considered as a wave propagation

problem. For multi-story buildings, for example, the vibrations can be characterized in terms of

wave propagation parameters; namely, wave velocities, attenuation of wave amplitudes, and the

wave reflections and transmission coefficients (Safak, 1999). Recorded earthquake motions from

instrumented structures clearly show the propagation of seismic waves. For example, a 7-story

steel-frame building instrumented with four accelerometers at every floor is studied, and we

recorded its accelerations during a small earthquake (Kohler et al, 2005). If we take a closer look

to a one-second long segment which shown in figure 6.5, the propagation of waves becomes very

clear.

Fig. 6.5: Propagation of seismic waves during a one-second interval in the 17-story building

For system identification and damage detection, it has been shown that, when compared to modal

parameters, the wave propagation parameters are more reliable and robust, and also more

sensitive to damage (Safak, 1998). For historical structures, the utilization of wave propagation

approach for system identification and damage detection is particularly convenient because, in

most cases, due to their age, geometry, construction material, and the structural system historical

structures do not meet the requirements of the classical modal analysis, such as elasticity,

linearity, mass and/or stiffness proportional damping (Safak et. al., 2009). For earthquake

induced waves, another wave propagation approach is the Seismic Interferometry (Snieder and


70

Safak, 2006).

6.4 DATA ANALYSIS

6.4.1 Spectral Analysis

One of the critical requirements in SHM is that, since the monitoring is done continuously and in

real time, the data processing and analysis should also be done in real time. Otherwise it would

be difficult to justify the real-time monitoring and continuous data collection. The Fourier

spectral analysis has been the standard method to analyze vibration data from structures. When

used for SHM data, the main source of errors in spectral analysis is the noise in the records and

the time-varying characteristics of the signals under transient loads. Noise alters the amplitudes

and the frequency content of Fourier spectra, and introduces spurious resonant peaks. Fourier

spectral analysis can give misleading results, particularly for records from stiff structures.

There are advanced alternative techniques to the standard Fourier analysis that can improve the

accuracy of the identification. Some of them are briefly summarized below. More detail can be

found in Safak, et.al. (2010).

Segmentation and Averaging

For stationary signals with added noise, it can be shown that the mean values of the Fourier

coefficients of the noisy signal converge to that of the noise-free signal, suggesting that the

averaging of the Fourier spectra of a large number of equal-length data segments can reduce the

noise effects. It can also be shown that the variances of the Fourier coefficients of the noisy

signal are inversely proportional to the record length; that is, the longer the record length the

smaller the more accurate the results. These observations suggest that we should consider long

segments of records when calculating the Fourier spectra of ambient data, provided that the

signal characteristics remain stationary. If the stationarity condition is not met, the alternative

would be to divide the signal into stationary segments, and calculate the Fourier spectrum of

each segment separately and then average them.

Selection of Optimal Smoothing Windows

A widely used technique to reduce the influence of noise in Fourier spectra is to apply

frequency-domain smoothing windows. Too short smoothing windows may not provide


71

sufficient noise reduction, whereas too long smoothing windows may eliminate some of the real

peaks. A simple technique for selecting the optimal smoothing window length is suggested in

(Safak, 1997).

Least-squares Estimation of Fourier Spectra

For a given signal length and sampling interval, the discrete frequencies of the Fourier spectra

are set. The only unknowns are the Fourier coefficients. Instead of the standard Fast Fourier

transforms, we can calculate the Fourier coefficients of the noise-free signal by minimizing the

error V between the noise-free signal, s(t), and the recorded signal, x(t) by using the following

equations:

where,

∙ cos 2

⁄

∙ sin 2

⁄

min , →

∂∂

0and∂∂

0

The minimization results in a linear set of equations for the coefficients, which can easily be

solved by matrix inversion. The calculated values represent the least-squares estimate of the

Fourier coefficients of the noise-free signal.

6.4.2 Statistical Signal Processing

The data from SHM systems are mostly stationary. Moreover, the SHM signals are also infinitely

long with low SNR. These properties make statistical signal processing tools very appropriate for

the analysis of SHM data. Some of the simple statistical signal processing tools are presented

below.

Autocorrelation functions and optimal filter

For stationary signals, the auto-correlation function depends only on the time lag τ. It can be

shown that the expected autocorrelation function of a sinusoid buried in noise has the same

frequency as the sinusoid. That is


72

∙ cos → E ∙ cos

where E[ ] denotes the expected value. It can also be shown that the SNR in the autocorrelation

of a signal is higher than that of the original signal. Therefore, when calculating Fourier spectra

of ambient noise, it is advantageous to use the autocorrelation functions of the records instead of

the original records.

A concept directly related to autocorrelation functions is the optimal filtering. Optimal filtering

aims to remove noise by searching correlated (i.e., periodic) components in the record. There are

numerous variations of the procedure suggested in the literature with their unique names such as

Wiener filtering, Recursive Least Squares, Least Mean Squares, Durbin Algorithm, Burg

Algorithm, and Yule-Walker Algorithm. More detail on these methods can be found in textbooks

on optimal filtering and linear estimation (e.g., Kailath et al., 2000).

Eigenvalues of autocorrelation matrix

Another set of powerful tools to separate signal from the noise can be developed based on the

eigenvalues and eigenvectors of the autocorrelation matrix. The autocorrelation matrix, Q, is

defined by the following equation

0M O M

L 0

where

∑ ∙ and , , 0, ,

Q is a (M+1)x(M+1) dimensional matrix that has (M+1) eigenvalues and eigenvectors. It can

also be shown that the eigenvalues that correspond to the correlated (i.e., periodic) components

of the record are much larger than those that correspond to the uncorrelated (i.e., noise)

components in the record. Therefore, the eigenvalues and eigenvectors of the correlation matrix

can be used to separate the noise from the signal.

6.4.3 Tracking Time Variations of Signal Properties

The simplest and most straightforward approach to analyze continuous data is the block-data

approach. In this method, the records are handled in blocks of specified length. Each block is

processed and analyzed as soon as it is full, and while the data for the next block are being


73

acquired. More efficient ways to analyze continuous data can be developed by utilizing running

time windows. Running windows are in essence weighting functions that emphasize recent data,

while gradually deemphasizing past data. The windows ensure that any property calculated from

data contains measurements that are relevant to the current state of the structure.

More on tracking the time-varying properties of signals can be found in Ljung and Soderstrom

(1983), Safak (1988, 1989a, 1989b) and Ljung (1999).

6.5 CONCLUSIONS

Structural Health Monitoring (SHM) involves continuous monitoring of the dynamic

characteristics of a structure by digital instruments. The main objective in SHM is to track the

changes in the structure’s dynamic characteristics and detect damage. The current data from

SHM systems show that the natural frequencies of the structure are not always a reliable

indicator of damage. Two additional parameters that can be used for damage detection are the

permanent change in the geometry of the structure and the changes in the characteristics of

propagating waves within the structure.

Data from SHM systems are mainly ambient vibration data, which typically have very low

signal-to-noise ratios (SNR), particularly for stiff structures. Standard Fourier-based spectral

analysis approach is not always reliable for noisy data. Statistical signal processing tools, such as

autocorrelation analysis, optimal filters, and eigenvalue-based spectral analysis are more

appropriate.

Also, since SHM data are continuous and recorded in real-time, data processing and analysis

should ideally be done in real time. Adaptive filtering and system identification techniques

provide tools to analyze SHM data in real time.


74


75

7 Summary

This report covered the research activities of Task JRA2.3. Several parts of Deliverable D13.2

edited by LNEC and reviewed by UNITN were issued. With regard to the software developed by

LNEC through the LabView platform, it was implemented by CEA. UL improved the

Performance-based Earthquake Engineering (PBEE) toolbox developed in Matlab, in

combination with the FE-based OpenSees software. JRC developed software for Modal

Parameter Extraction and for Identification of modal properties and damping from dynamic

response of components and structural systems. UNITN developed software capable of

animating 3D modal shapes from OpenSees outputs, improved System Identification software

and implemented MATLAB software for 3D animations of real-time tests. KOERI improved its

software for Structural Health Monitoring (SHM) of structures in situ.


76


77

References

Abbiati G., Bursi O.S., Molinari M., Xu G. (2012) Use of Partitioned Time Integration Schemes for Parallel

Simulations of Hybrid Systems. Proc. of the 5th European Conf. on Struct. Control, Genoa, 18-20 June.

Anthoine, A. and Molina, F. J. (2008) “Pseudo-dynamic testing of full scale masonry structures: preparatory work”

Proc. Of the 14th International Brick and Block Masonry Conference, Sydney, Australia.

Armelle Anthoine and Daniel Tirelli. Project No. Coll – Ct - 2003 – 500291. ESECMaSE. Enhanced Safety and

Efficient Construction of Mas onry Structures in Europe. Work Package N° 8. D 8.2 Preliminary tests and

dynamic identification of the specimens. 04/07/2008. Draft N°1. European Laboratory for Structural

Assessment (ELSA). Joint Research Centre of the European Commission .21020 Ispra (VA). Italy .

Bendat, J. and A. Piersol 1986 - "Random Data-Analysis and Measurement Procedures", John Wiley and Sons.,

New York.

Bonelli A., Bursi O.S., He L., Magonette G., Pegon P., (2008). Convergence analysis of a parallel interfield method

for heterogeneous simulations with substructuring. I. J. for Num. Meth. Eng., 75, 7, 2008, 800-825.

Brammer, K. and Siffling, G. (1989) Kalman-Bucy Filters, Artech House, Norwood, MA, 1989.

Burg, J.P. “A new analysis technique for time-series data”, NATO Advanced Study Institute on Signal Processing,

Enschede, The Netherlands, 1968.

Celarec D., Dolšek M., Vamvatsikos D. 2009. Estimation of the seismic risk with consideration of capacity

degradation over time. Proceedings of COMPDYN 2009, Rhodes Greece, CD217.

Clinton, J. (2004). www.ce.caltech.edu/~jclinton/thesis.html

Compaq 2000 "Compaq Visual Fortran v.6.6c". Compaq Computer Corporation. Houston, USA.

Daniel Tirelli, Stefano Primi, Eugenio Gutierrez. Experimental Modal Analysis of PUMACOM Carbon-Fibre

Composite Highway Bridge at Crossing PK 1+640 of the Autovía del Cantábrico. Dynamic load test. Report

No:1 .2004. European Laboratory for Structural Assessment, IPSC, Joint Research Centre. Commission of the

European Communities, Ispra, Italy.

Daniel Tirelli. Modal Analysis of Small & Medium Structures by Fast Impact Hammer Testing Method. JRC

Scientific and Technical Reports, 2011.

DEE-KOERI (2008a). İstanbul Yüksek Binalar Deprem Yönetmeliği, prepared for the Istanbul Municipality by the

Department of Earthquake Engineering of Kandilli Observatory and Earthquake Research Institute, Bogazici

University, Istanbul, Turkey.

DEE-KOERI (2008b). İstanbul Yüksek Binalar Rüzgar Yönetmeliği, prepared for the Istanbul Municipality by the

Department of Earthquake Engineering of Kandilli Observatory and Earthquake Research Institute, Bogazici

University, Istanbul, Turkey.

DEE-KOERI (2009a). Seismic Analysis and Design Requirements for Buildings in Dubai, The United Arab

Emirates, prepared for the Dubai Municipality by the Department of Earthquake Engineering of Kandilli

Observatory and Earthquake Research Institute, Bogazici University, Istanbul, Turkey.

DEE-KOERI (2009a). Wind Analysis and Design Requirements for Buildings in Dubai, The United Arab Emirates,

prepared for the Dubai Municipality by the Department of Earthquake Engineering of Kandilli Observatory and

Earthquake Research Institute, Bogazici University, Istanbul, Turkey.

Dolšek M 2009a. Incremental dynamic analysis with consideration of modelling uncertainties. Earthquake

Engineering and Structural Dynamics, 38:805-825.


78

Dolšek M. 2009b. Development of computing environment for the seismic performance assessment of reinforced

concrete frames by using simplified nonlinear models. Submitted for Journal publication.

Dolšek M., Fajfar P. 2002. Mathematical modelling of an infilled RC frame structure based on the results of pseudo-

dynamic tests. Earthquake Engineering and Structural Dynamics, 31:1215-1230.

Durbin, J. (1959). Efficient estimators of parameters in moving average models, Biometrica, 46:306-316, 1959.

ESECMaSE project (25 March 2009) http://www.esecmase.org

Etienne Balmes, Orthogonal Maximum Sequence Sensor Placements Algorithms for modal tests, expansion and

visibility. SDTools and Ecole Centrale.Paris/MSSMat, [email protected]

Ewins, D. J. (1984). Modal Testing: Theory and Practice, Taunton Research Studies Press.

Francisco Javier Molina, Georges Magonette and Pierre Pegon. Software for Identification of Modal Properties from

Dynamic Response of Structures. European Commission, JRC, IPSC, ELSA laboratory, I-21027 Ispra, Italy.

F. J. Molina, R. Pascual and J.-C. Golinval, DESCRIPTION OF THE STEELQUAKE BENCHMARK, Mechanical

Systems and Signal Processing (2003) 17(1), 77–82.

Hayes, M. H., (1996). Statistical Digital Signal Processing and Modelling, John Wiley.

J.N. Juang, R.S. Pappa, An Eigensystem Realisation Algorithm (ERA) for modal parameter identification and modal

reduction, NASA/JPL Workshop on Identification and Control of Flexible Space Structures, 1984.

Kailath, T., Sayed, A.H., and Hassibi, B. (2000), Linear Estimation, Prentice Hall, Upper River Saddle, New Jersey.

Kohler, M.D., Davis, P.M. and Safak, E. (2005). Earthquake and Ambient Vibration Monitoring of the Steel Frame

UCLA Factor Building, Earthquake Spectra, Vol. 21, No. 3, August 2005, pp.715-736.

LATBSDC (2008). An Alternative Procedure for Seismic Analysis and Design of Tall Buildings Located in the Los

Angeles Region, by Los Angeles Tall Buildings Structural Design Council.

Lehmann, E.L. (1959). “Testing Statistical Hypotheses”, John Wiley & Sons, New York, NY.

Ljung, L. (1999). System Identification: Theory for the User, Prentice Hall PTR, Upper Saddle River, NJ, 1999.

M. Poljanšek, G. Bof, P. Capéran, J. Esteves, E. Gutiérrez, O. Hubert, R. Kiefer, D. Tirelli, B. Viaccoz, A. Zorzan.

Report on Testing and Analysis of PROMETEO Fibre Reinforced Composite Bridge Beam. Limited Dist

ribution PUBSY JRC54320 – 2009.

Mathworks, MatLAB 2012, “Simulink, Stateflow, xPC Target, Parallel Computing Toolbox”

<http://www.mathworks.com>.

Molina, F. J., Magonette, G, Pegon, P. & Zapico, B. (2011a): Monitoring Damping in Pseudo-Dynamic Tests,

Journal of Earthquake Engineering, 15:6, 877-900

Molina F.J., Marazzi, F., Viaccoz, B. and Bosi, A. (2012). Stability and accuracy in a hybrid test example,

Earthquake Engineering and Structural Dynamics. (EQE-11-0289.R1, accepted).

Molina, F. J., Pegon, P. and Verzeletti, G. (1999) “Time-domain identification from seismic pseudo dynamic test

results on civil engineering specimens” Proc. of the 2nd International Conference on Identification in

Engineering Systems, University of Wales Swansea.

Molina, F. J. (2011b) “Spatial and Filter Models” MATLAB functions freely available at MATLAB CENTRAL

FILE EXCHANGE: http://www.mathworks.com/matlabcentral/fileexchange/32634.

NI 2004 - "Labview v7.1". National Instruments Corporation. Austin, USA.

OpenSEES, (2009), Open System for Earthquake Engineering Simulation <http://opensees.berkeley.edu>.

OpenFRESCO, (2009), “Open Framework for Experimental Setup Control”

<http://OpenFRESCO.neesforge.nees.org>.

O.S.Bursi, J.Molina, W.Salvatore, and F. Taucer. Dynamic Characterisation of a 3D full scale steel concrete

composite building at ELSA. EUR21206 EN.


79

Paolacci F., Giannini R., (2011). An experimental and numerical investigation on the cyclic response of a portal

frame pier belonging to an old reinforced concrete viaduct, Earthquake Engineering & Structural Dynamics,

DOI: 10.1002/eqe.1175

Pegon P. Magonette G., (2002). Continuous PSD testing with nonlinear substructuring: presentation of a stable

parallel inter-field procedure. Technical Report 1.02.167, E.C., JRC, ELSA, Ispra, Italy.

Pegon P., Molina, F. J. and Magonette, G. (2008). “Continuous pseudo-dynamic testing at ELSA” in Hybrid

Simulation; Theory, Implementation and Applications, ed. Saouma V.E. and Sivaselvan M. V. (Taylor &

Francis/Balkema) 79-88.

P. Van Overschee, B. De Moor, Subspace Identification for Linear Systems: Theory and Implementation -

Applications, Kluwer Academic Press, Dordrecht, The Netherlands, 1996.

R. Ceravolo. Use of instantaneous estimators for the evaluation of structural damping, Journal of Sound and

Vibration, 274 (2004) 385-401.

R. Ceravolo, G. Abbiati. SDIT Release 3.1 User's Manual. Structural Identification Toolbox for Matlab (in Italian),

Politecnico di Torino, 2009.

R.J. Allemang. The Modal Assurance Criterion (MAC): Twenty years of use and abuse, Proceedings of the 20th

International Modal Analysis Conference, Los Angeles, CA, USA, 2002.

Safak, E. (1988). Analysis of recordings in structural engineering: Adaptive filtering, prediction, and control, Open-

File Report 88-647, U. S. Geological Survey, Menlo Park, California.

Safak, E. (1989a). Adaptive modeling, identification, and control of dynamic structural systems: Part I- Theory,

Journal of Engineering Mechanics, ASCE, Vol.115, No.11, November 1989, pp.2386-2405.

Safak, E. (1989b). Adaptive modeling, identification, and control of dynamic structural systems: Part II-

Applications, Journal of Engineering Mechanics, ASCE, Vol.115, No.11, November 1989, pp.2406-2426.

Safak, E. (1995). Detection and identification of soil-structure interaction in buildings from vibration recordings,

Journal of Structural Engineering, ASCE, Vol.121, No.5, May 1995, pp.899-906.

Safak, E. (1997). Models and methods to characterize site amplification from a pair of records, Earthquake Spectra,

EERI, Vol.13, No.1, pp.97-129

Safak, E. (1998). Detection of seismic damage in multi-story buildings by using wave-propagation analysis,

Proceedings (in CD-ROM) of the Sixth U.S. National Conference on Earthquake Engineering, Seattle,

Washington, May 31-June 4, 1998, Elsevier Science Inc., New York, NY.

Safak, E. (1999). Wave propagation formulation of seismic response of multi-story buildings, Journal of Structural

Engineering, ASCE, Vol. 125, No. 4, pp.426-437.

Safak, E. (2004). “Analysis of real-time data from instrumented structures, Proceedings (CD ROM), The 13th World

Conference on Earthquake Engineering, Vancouver, B.C., Canada, August 1-6, 2004, Paper No. 2193.

Safak, E., Cakti, E. and Kaya, Y. (2009). Seismic Wave Velocities in Historical Structures: A New Parameter for

Identification and Damage Detection, Proceedings of IOMAC 2009 – 3rd International Modal Analysis

Conference, Ancona, Italy, 4-6 May 2009.

Safak, E., Cakti, E. and Kaya, Y. (2010). Recent developments on Structural Health Monitoring and data analysis, in

Earthquake Engineering in Europe (Ed. By M. Garevski and A. Ansal), Springer Int. Pub.

SeismoSoft 2004 - "SeismoSignal v.3.1.0". www.seismosoft.com.

Snieder, R. and Safak, E. 2006. Extracting the building response using interferometric imaging: Theory and

application to the Millikan Library in Pasadena, California, Bulletin of the Seismological Society of America

Vol. 96, pp.586-598.


80

Taucer F., (2011). Seismic Engineering Research Infrastructures for European Synergies: Activities of the Joint

Research Centre in large scale experimental testing within the framework of the Eurocodes, 7th International

conference on structural defects and rehabilitation, 2-4 June, Fortaleza, Brazil.

Trifunac, M.D., Ivanovic, S.S., and Todorovska, M.I. (1999). Instrumented 7-Storey Reinforced Concrete Building

in Van Nuys, California: Description of Damage from the 1994 Northridge Earthquake and Strong Motion

Data, Report CE 99-02, Dept. of Civil Engrg., Univ. of Southern California.

Widrow, B. and Stearns, S.D. (1985). Adaptive Signal Processing, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1985.

Wu B., Wang Z. and Bursi O.S., (2012), Actuator dynamics compensation based on upper bound delay forreal-time

hybrid simulations, Earthquake Engineering and Structural Dynamics, submitted.

deliverable [d13.2] 28 08 2012 v2 upat - series1).pdf · properties and damping from dynamic...

Documents