Documente Academic
Documente Profesional
Documente Cultură
Vibration monitoring
Assoc Prof Assoc. Prof. Dr Dr. Pelin Gundes Bakir
Istanbul Tecnical University
g gundesbakir@yahoo.com @y
References
Pete Avitable Avitables s web page http://faculty.uml.edu/pavitabile/22.515/ME22515_PDF_downloads.htm Safak E., Structural monitoring, what is it, why is it done, how is it done, and what is it worth?,Sixth worth? Sixth National Conference on Earthquake Engineering Engineering, 16 16-20 20 October 2007, Istanbul, Turkey Celebi M. Seismic instrumentation of buildings, USGS Open-File Report 00-157, 2000. 2000 Heylen W., Lammens S. And Sas P., Modal Analysis Theory and Testing, Katholieke Universiteit Leuven, 1997. Ewins D.J., Modal Testing, Theory, Practice, and Application (Mechanical Engineering Research Studies Engineering Design Series), Research Studies Pre; 2 edition ( (August g 2001) ) ISBN-13: 978-0863802188 Maia, N. M. M. and Silva, J. M. M.Theoretical and Experimental Modal Analysis Research Studies Press Ltd,, Hertfordshire, 1997, 488 pp.,ISBN 0863802087
P. Gundes Bakir,
Vibration monitoring
The aim Th i of f vibration ib ti monitoring it i is i to t describe d ib a structure t t in i terms t of f its it modal parameters which are the frequency, damping and mode shapes. If we explain modal analysis in terms of the modes of vibration of a simple plate:
Suppose we apply a sinusoidal force. We will change the rate of oscillation of the frequency but the peak force will always be the same. We will also measure the th response of f the th plate l t due d t to the th excitation it ti with ith an accelerometer attached to one corner of the plate.
P. Gundes Bakir,
Vibration monitoring
If we measure the th response of f the th plate, we will notice that the amplitude changes as we change the rate of oscillation of the input force. There will be increases as well as decreases in amplitude at different points as we sweep in time. ti The response amplifies as we apply l af force with ith a rate t of f oscillation that gets closer and closer to the natural frequency (or resonant frequency) of the system and reaches a maximum when the rate of oscillation is at the resonant frequency of the system.
P. Gundes Bakir, Vibration based structural health monitoring 4
Modal analysis
The time Th i data d provides id very useful f li information. f i B But if we take k the h time data and transform it to the frequency domain using the Fast Fourier Transform then we can compute something called the frequency response function. Now, there are some very interesting items to note. We see that there are peaks in this function which occur at the resonant frequencies of the system. Now, we notice that these peaks occur at frequencies where the time response was observed to have maximum response corresponding to the rate of oscillation of the input excitation.
P. Gundes Bakir,
Modal analysis
Now, if we overlay N l the h time i trace with i h the h f frequency trace what h we will notice is that the frequency of oscillation at the time at which the time trace reaches its maximum value corresponds to the frequency where peaks in the frequency response function reach a maximum. So we can see that we can either use the time trace to determine the frequency at which the maximum amplitude increases occur or the frequency response function to determine where these natural f frequencies i occur. Clearly Cl l th the f frequency response f function ti i is easier i to evaluate.
P. Gundes Bakir,
Vibration monitoring
The figure shows the deformation patterns that will result when the excitation coincides with one of the natural frequencies of the system.
P. Gundes Bakir,
Vibration monitoring
We see that when we dwell at the first natural frequency, q y there is a first bending deformation pattern in the plate shown in blue. When we dwell at the second natural frequency, there is a first twisting deformation pattern in plate shown in red. the p
P. Gundes Bakir,
Vibration monitoring
When we dwell at the third and fourth natural frequencies, q , the second bending g and second twisting deformation patterns are seen in green and magenta, respectively. These deformation patterns are referred to as the mode shapes of the structure.
P. Gundes Bakir,
Vibration monitoring
Now these natural frequencies and mode shapes occur in all structures that we design. Basically, there are characteristics that depend on the weight and stiffness of my structure which determine where these natural frequencies and mode shapes will exist exist. As a design engineer, I need to identify these frequencies and know how they y might g affect the response p of my y structure when a force excites the structure. Understanding the mode shape and how the structure will vibrate when excited helps the design engineer to design better structures structures. Now we can better understand what modal analysis is all about-it is the study of the natural characteristics of structures. Both the natural frequency and mode shape (which depends on the mass and stiffness distributions in my structure) are used to help design my structural applications.
P. Gundes Bakir,
10
P. Gundes Bakir,
11
P. Gundes Bakir,
12
P. Gundes Bakir,
13
From all these simple examples above, it becomes obvious that we need a distribution of points located appropriately such that each mode shape can be uniquely distinguished.
P. Gundes Bakir, Vibration based structural health monitoring 14
P. Gundes Bakir,
15
P. Gundes Bakir,
16
What is the difference between time domain and the frequency domain and the modal space?
First lets consider a simple cantilever beam and imagine that the beam is excited by a pulse at the tip of the beam. The response at the tip of the beam will contain the response of f all ll th the modes d of f th the system t (shown in the black time response plot); notice that there appears to response at se several eral different frequencies.
P. Gundes Bakir,
17
What is the difference between time domain and the frequency domain and the modal space?
The time response at the tip of the beam can be converted to the frequency domain by performing a Fourier Transform of the signal. Th frequency The f domain d i representation of this converted time signal is often referred to as the frequency response function function, or FRF for short (shown in the black frequency plot); notice that there are peaks in the plot which correspond to the natural frequencies of the system.
P. Gundes Bakir,
18
What is the difference between time domain and the frequency domain and the modal space?
We know that the cantilever beam will have many natural frequencies of vibration. At each of these natural frequencies, the structural deformation will take on a very definite pattern, called a mode shape. For this beam, we see that there is a first bending mode shown in blue, a second bending mode shown in red, and a third bending mode shown in green. Of course there are also other higher modes not shown but only three modes will be discussed here.
P. Gundes Bakir, Vibration based structural health monitoring 19
What is the difference between time domain and the frequency domain and the modal space?
Now, the physical beam could also be evaluated using an analytical lumped mass model or finite element model (shown in black in the upper right part of the figure). figure) This model will generally be evaluated using some set of equations where there is an interrelationship, or coupling, between the different points, or degrees of freedom used to model the structure. This means you pull on one of the dofs in the model, the other dofs are also affected and also move.
P. Gundes Bakir, Vibration based structural health monitoring 20
What is the difference between time domain and the frequency domain and the modal space?
This coupling means that the equations are more complicated in order to determine how the system behaves. As the number of equations used to describe the system get larger and larger, larger the complication in the equations become more involved. We often use matrices to help p organize g all of the equations of motion describing how the system behaves which looks like:
P. Gundes Bakir,
21
What is the difference between time domain and the frequency domain and the modal space?
Usually, the mass is a diagonal matrix and the damping and the stiffness matrices are symmetric with off-diagonal terms indicating the degree of coupling between the different equations or dofs describing the system. The size of the matrices depend on the number of equations that we e use se to describe o our rs system. stem
P. Gundes Bakir,
22
What is the difference between time domain and the frequency domain and the modal space?
Mathematically, we perform something called an eigensolution and use the modal transformation equation to convert these coupled equations into a set of uncoupled single degree of freedom systems described by diagonal matrices of modal mass mass, modal damping and modal stiffness in a new coordinate system called modal space described as:
P. Gundes Bakir,
23
What is the difference between time domain and the frequency domain and the modal space?
We can see that the transformation from physical space to modal space using the modal transformation equation is a process whereby we convert a complicated set of coupled physical equations into a set of simple uncoupled single dof systems.
P. Gundes Bakir,
24
What is the difference between time domain and the frequency domain and the modal space?
And we see in the figure that the analytical model can be broken down into a set of single dof systems where the single dof describing mode 1 is shown in blue, mode 2 is shown in red and mode 3 is shown in green. Modal space allows us to describe the system easily using simple single dof s systems. stems
P. Gundes Bakir,
25
What is the difference between time domain and the frequency domain and the modal space?
Now lets go back to the time and frequency responses shown in black We know that the total black. response can be obtained from the contribution of each of the modes The total response shown modes. in black comes from the summation of the effects of the response p of the model shown in blue for mode 1, red for mode 2, and green for mode 3. This applies pp whether I describe the system in the time domain or the frequency domain. Each domain is equivalent and just presents the data from a different viewpoint.
P. Gundes Bakir, Vibration based structural health monitoring 26
What is the difference between time domain and the frequency domain and the modal space?
So we can see that the total time response is made up of the time response due to the contribution of the time response of mode 1 shown in blue, mode 2 in red and mode 3 in green green. We can also see that the total FRF is made up of the part of the FRF due to the contribution of the FRF of mode 1 shown in blue, mode 2 in red and mode 3 in green.
P. Gundes Bakir,
27
What is the difference between time domain and the frequency domain and the modal space?
Please note that we have only shown the magnitude part of the FRF here, here this function is actually complex which is correctly displayed using both magnitude and phase or real and imaginary parts of the FRF. Since we can break the analytical model up into a set of single dof systems, we could determine the FRF for each of the single dof systems as shown with mode 1 in blue, mode 2 in red, and mode 3 in green.
P. Gundes Bakir,
28
What is the difference between time domain and the frequency domain and the modal space?
We could also determine the time response for each of these single dof systems due to the pulse input Or we could simply inverse Fourier transform the FRF for each of the single dof systems systems. Or we could also measure the response of the beam at the tip due to the pulse and filter the response of each modes of the system, and we we would see the response of each of the modes of the system with mode 1 in blue, mode 2 in red, and mode 3 in green green.
P. Gundes Bakir, Vibration based structural health monitoring 29
What is the difference between time domain and the frequency domain and the modal space?
As a result, we see that there is no difference between the time domain frequency domain , modal domain, space and physical space. Each domain is just a convenient way for presenting or viewing the data data. However, sometimes one domain is much easier to see things than another domain. domain For instance instance, the total time response does not clearly identify how many modes there are contributing to the response of the beam.
P. Gundes Bakir,
30
What is the difference between time domain and the frequency domain and the modal space?
But the total FRF in the frequency domain is much clearer in showing how many modes are activated and the frequency of each of the modes. modes So often, we transform one d domain i t to another th d domain i simply i l because the data is much easier to interpret.
P. Gundes Bakir,
31
( Mp 2 + Cp + K ) X ( p ) = F ( p ) or Z ( p) X ( p) = F ( p)
where Z is the dynamic stiffness. Inverting Z gives the transfer function:
H ( p) =
1/ M X ( p) = 2 F ( p ) p + (C / M ) p + ( K / M )
33
P. Gundes Bakir,
H ( p) =
X ( p) 1/ M = 2 F ( p ) p + (C / M ) p + ( K / M )
is referred to as the system characteristic equation. Its roots are called the system poles which are given by:
1, 2 = (C / 2M ) (C / 2M ) 2 ( K / M )
If there h is i no damping, d i the h system under d consideration id i i is a conservative i system (C=0).The undamped natural frequency (rad/s) is then defined as:
1 = K / M
P. Gundes Bakir,
34
1, 2 = (C / 2M ) (C / 2 M ) 2 ( K / M )
equal to zero:
Cc = 2 M K / M
1 = C / Cc
The first equation yields in the time domain a solution to the homogeneous system equation: 1t 2 t
x(t ) = x1e + x2 e
P. Gundes Bakir,
35
1, 2 = (C / 2M ) (C / 2M ) 2 ( K / M )
Yields two complex conjugate roots
1 = 1 + j1
1* = 1 j1
1 = ( 1 + j 1 12 )1
P. Gundes Bakir,
37
Residues
With the knowledge g of the equation q
1 = 1 + j1
the equation for the transfer function
H ( p) =
1* = 1 j1
becomes:
X ( p) 1/ M = 2 F ( p ) p + (C / M ) p + ( K / M )
H ( p) =
1/ M ( p 1 )( p * ) 1
A1* A1 H ( p) = + * ( p 1 ) ( p 1 )
In this formula A1 and A1* are the residues.
P. Gundes Bakir,
with A1 =
1/ M j 21
38
H ( p)
p = j
A1* A1 = H ( ) = + * ( j 1 ) ( j 1 )
The FRF is a subset of the transfer function. The contribution of the complex conjugate part (or negative frequency part) is negligible around resonance Therefore resonance. Therefore, the FRF for SDOF system is often approximated by:
A1 H ( ) = ( j 1 )
P. Gundes Bakir, Vibration based structural health monitoring 39
H ( p)
p = j
A1* A1 = H ( j ) = + * ( j 1 ) ( j 1 )
* t * 1
yields the expression in the time domain: the impulse response function.
h(t ) = A1e + A1 e
1t
The residue A1 is the real part of the pole which defines the initial amplitude 1 is the imaginary part of the pole which defines the decay rate amplitude, and 1 is the frequency of oscillation. The impulse Th i l response of f a system t is i th the system t response t to a Dirac Di impulse at time t=0.
P. Gundes Bakir,
40
System equation and the transfer function The Th equation ti of f motion ti is: i
&}+ [C ]{x &}+ [ K ]{x} = { f } [ M ]{& x
If we transform this time domain equation into the Laplace domain (variable p), assuming the initial displacements and velocities are zero yields:
( p 2 [M ] + p[C ] + [K ]){X ( p )} = {F ( p )}
where [Z(p)] is the dynamic stiffness matrix. The inverse of [Z(p)] is [H(p)]
[H ( p)] = [Z ( p)]
adj[Z ( p )] = Z ( p)
Where adj([Z(p)]) is the adjoint matrix of [Z(p)] which can be expressed as.
adj ([ Z ( p )]) = [ ij Z ij ]t Z ij : the determinant of [ Z ( p)], without row i and column j
ij = 1, if i + j is even; = -1 if i + j is odd
Z ( p ) : the determinant of [ Z ( p)]
P. Gundes Bakir,
43
[ H ( p )] p = j = [ H ( j )] =
k =1
* [ Ak ] [ Ak ] + ( j k ) ( j * k)
hij ( j ) =
k =1
aijk ( j k )
aijk
( j * k)
hij() means a particular output response at point i due to an input force at point j. Since [M], [C], [K] are symmetric, [H(j )] is also symmetric. This implies that hij=h hji which is called reciprocity. This means that you can measure the FRF by impacting point i and measuring the response at point j and get exactly the same FRF as impacting point j and measuring the response at point i. This is what is meant by reciprocity.
P. Gundes Bakir, Vibration based structural health monitoring 44
Residues
The residues are directly related to mode shapes and a scaling factor as:
This shows that the frequency response function can be written in terms of residues. id When written as a mode shape, p , then it becomes very y clear that if the value of the mode shape at the reference point is zero (or almost zero) then that mode will not be seen in the frequency response function. Always select a reference point where all the modes can be seen all the time from that reference point.
P. Gundes Bakir, Vibration based structural health monitoring 45
Residues
Never select the reference point at the node of a mode!
P. Gundes Bakir,
46
P. Gundes Bakir,
47
P. Gundes Bakir,
48
P. Gundes Bakir,
49
P. Gundes Bakir,
50
P. Gundes Bakir,
52
P. Gundes Bakir,
53
P. Gundes Bakir,
54
P. Gundes Bakir,
55
P. Gundes Bakir,
59
Structural response
The methods used in studying structural response records are quite diverse: (a) ( ) Mathematical modeling g( (finite element models varying y g from crude to very y detailed, subjected to timehistory, response spectrum or modal analyses). The procedure requires the blueprints of the structures which may not be readily accessible; (b) System identification techniques: single input/single output or multi input/multi output. In these procedures, the parameters of a model are adjusted for consistency with input and output data (Ljung (Ljung, 1987); (c) Spectral analyses: response spectra, Fourier amplitude spectra, autospectra Sx or Sy, cross-spectral autospectra, cross spectral amplitudes Sxy, and coherence functions ( ) [using the equation : 2xy (f) = S2xy (f)/ Sx (f)Sy (f)] and associated phase angles
P. Gundes Bakir,
60
P. Gundes Bakir,
62
P. Gundes Bakir,
63
P. Gundes Bakir,
65
P. Gundes Bakir,
66
P. Gundes Bakir,
67
P. Gundes Bakir,
69
Performance of base-isolated systems and effectiveness of the isolators are best captured by measuring tri-axial motions at top and bottom of the isolators as well as the rest of the superstructure.
P. Gundes Bakir,
70
P. Gundes Bakir,
71
P. Gundes Bakir,
74
P. Gundes Bakir,
75
P. Gundes Bakir,
77
P. Gundes Bakir,
82
Sensor locations
The number of required sensors and sensor locations depend on the condition that whether 2 dimensional or 3 dimensional motions of the structure are going to be monitored. In 2 dimensions, the degrees of freedom are 2 translations and one rotation. A typical example to such a structure is a multistorey building with shear walls ll and d a rigid i id di diaphragm. h In order to determine these two translations and one rotation, three measurements are needed needed. These three measurements have to satisfy the following conditions:
The measurements have to be taken from two separate locations The three measurement directions should not be parallel. The three measurement directions should not intersect each other.
P. Gundes Bakir, Vibration based structural health monitoring 86
Sensor locations
In 3 D space space, the vibrations of the structure can be fully described by 3 translations and 3 rotations. In such a structure structure, we need at least 6 measurements which have to satisfy the following conditions in order to solve for the 3 rotations and 3 translations from the dynamic equilibrium equations:
The measurements have to be taken at least from 3 separate locations. The measurement locations should not be on a straight line. The 6 measurement directions should not be parallel. The 6 measurement directions should not intersect each other.
P. Gundes Bakir,
87
Sensor locations
Sensor locations should not be determined based on the locations of maximum displacements, because that displacement can be dominated by a single mode. That means, if you put a sensor there you would only identify that mode. The location you should search for is the one whose displacement has contrib contributions tions from a maximum ma im m n number mber of modes modes. That way you can identify more modes (afak,2009). For typical multi-story multi story buildings, buildings there is no reason to put a vertical sensor in the middle of floor slab, unless it has a huge span with no beams and columns to support it. Even then, it would record only the local behavior of the floor slab, not the global behavior of structural system, which is more important when matching analytical models (afak,2009). afak 2009) For torsion, the larger the distance between the two parallel horizontals, the more accurate the calculated torsion (i (i.e., e the better the signal to noise ratio in the torsional signal) (afak,2009).
P. Gundes Bakir,
88
Sensor locations
In general I l the h order d f for placing l i sensors: Roof Ground floor Basement Any location where stiffness and/or mass changes significantly Any location where the curvature of the deformed shape is expected to change.
P. Gundes Bakir,
89
P. Gundes Bakir,
90
P. Gundes Bakir,
91
P. Gundes Bakir,
92
P. Gundes Bakir,
94
P. Gundes Bakir,
95
P. Gundes Bakir,
96
P. Gundes Bakir,
97
P. Gundes Bakir,
98
Maintenance
It is essential to have periodic and consistent maintenance of instruments in order to have a successful program. Unless maintenance arrangements are made, successful recording of data cannot be accomplished. Therefore, routine maintenance is conducted every 3-12 months if circumstances and e perience so allow. experience allo This maintenance includes the following:
1. Remote calibration of period and damping. 2. Inspection p of battery y terminals, load voltage, g and charge g rate ( (batteries are replaced every 3 years). 3. Measurement of threshold of triggering system and length of recording cycle.
As a final maintenance procedure, a calibration record is obtained and then examined for the desired characteristics. All inspection procedures are recorded in the permanent station file at the laboratory.
P. Gundes Bakir, Vibration based structural health monitoring 99
GPS units
Until recently recently, in general, general only accelerometers (single (single, biaxial or triaxial) were used to instrument structures. However, observations of damages during the 1994 Northridge and 1995 Kobe earthquakes, have forced engineers and scientists to focus on performance based seismic design methods and to find new techniques to control drift and displacements. To verify these developments, sensors directly measuring g displacements p or relative displacements p (transducers, laser devices and GPS units) are now being considered.
P. Gundes Bakir, Vibration based structural health monitoring 100