Arch Computat Methods Eng (2016) 23:585–594

DOI 10.1007/s11831-015-9145-0

ORIGINAL PAPER

The Vibration Monitoring Methods and Signal Processing

Techniques for Structural Health Monitoring: A Review
D. Goyal1 • B. S. Pabla1

Received: 1 March 2015 / Accepted: 5 March 2015 / Published online: 10 March 2015
Ó CIMNE, Barcelona, Spain 2015

Abstract Machines without vibrations in the working reserved for systems that can undergo an oscillatory
environment are something non-existent. During machin- motion freely without applied forces. Vibration is one of
ing operations, these vibrations are directly linked to the characteristics of mechanical machinery that if un-
problems in systems having rotating or reciprocating parts, controlled, can cause minor or serious performance, op-
such as bearings, engines, gear boxes, shafts, turbines and erational or safety problems. Machine vibration being a
motors. Vibration analysis has proved to be a measure for key element for structural health monitoring (SHM) so
any cause of inaccuracy in manufacturing processes and there is massive interest in acquiring, analyzing and
components or any maintenance decisions related to the quantifying this parameter for improving reliability, life,
machine. The non-contact measurement of vibration signal quality control, productivity and safety against catas-
is very important for reliable structural health monitoring trophic failure. Although there are various causes of vi-
for quality assurance, optimizing profitability of products brations, but about 90 % of all problems arise due to
and services, to enhance manufacturing productivity and to unbalancing or misaligning of rotating parts [1]. Vibration
reduce regular periodic inspections. This paper presents a can be characterized in terms of three parameters: am-
state-of-the-art review of recent vibration monitoring plitude, velocity and acceleration. The sensitivity of sen-
methods and signal processing techniques for structural sors used for measuring these parameters varies with the
health monitoring in manufacturing operations. These frequency of the vibration. The general selection guide-
methods and techniques are used as a tool to acquire, vi- line is to use amplitude sensors to pick up low frequency
sualize and analyse the sampled data collected in any signals, velocity sensors in the middle ranges, and ac-
machining operation which can then be used for decision celerometers at higher frequencies [2]. Machine tool
making about maintenance strategies. structural components, such as gateways, bearings and
ball screws are liable to deterioration in their performance
level with respect to time due to gradual wear and tear,
1 Introduction ageing, unbalance, looseness of parts, etc. On a regular
basis, the testing of such components is needed to reduce
Mechanical systems with reciprocating or rotating parts the risk of failures and machinery breakdown. The present
create vibrations due to mechanical disturbances from global market competition has attracted the attention of
various sources such as engine, sound, noise, etc. In manufacturing industries towards the monitoring of
mechanical engineering, the term ‘‘vibration’’ is often manufacturing processes and equipment conditions to
become a leaner. As cited by Meher-Homji [3], a major
re-insurance company reported that blade failures had the
highest percent of the failures in gas turbines (42 % of the
& D. Goyal total failures in gas turbines). Vibration monitoring has
[email protected]
become an important technique in mechanical structural
1
Department of Mechanical Engineering, NITTTR, product’s research, design, produce, apply and mainte-
Chandigarh 160019, India nance [4].

123
586 D. Goyal, B. S. Pabla

2 Vibrations as a Diagnostics Tool Sensor 1

Diagnosis is the art of identifying machine condition from Sensor 2 Signal

its signs or symptoms to determine its ‘‘cause and effect’’. It Acquisition
is usually used for monitoring, detection and analysis of Sensor 3
machine condition during an operation. Vibration signatures
of the machine can offer an early warning to the operator for
time based maintenance or to make a crucial decision before Data Signal
any serious problem or unscheduled downtime. The ampli- Interpretation Processing
tude of the vibration signature gives an indication of the
severity of the problem, whilst the frequency can indicate Fig. 1 Various steps in structural health monitoring
the source of the defect [5]. The extraction of these signals
can be considered as a valuable diagnostic tool to predict Hybrid classifiers. SHM is also categorized into local and
run-up failures of the machine components. It is a very global methods analogous to perform a human health checkup
challenging task to extract the feature from the acquired [7]. This classification is commonly established on the relation
signal as interferences are occurred due to the presence of of the wavelength of the experimented signals with respect to
noise. It arises a need for developing a processing and the defect size as well as to the whole structure dimensions.
analysis system of high level data to provide veracious ex- SHM technology can also contribute to intelligent aircraft
traction of data about the health of the machine. Various structures for reducing its weight and maintenance [8]. They
feature extraction techniques like statistical domain, fre- showed that SHM is the imitation of human nervous system. A
quency domain, time domain and time–frequency domain popular approach to SHM is modal parameters (natural fre-
are used for obtaining diagnostic information. quencies, mode shapes, mode shape curvatures, flexibility,
Vibration analysis is a very powerful and reliable tech- and changes in strain energy distributions etc.) which are
nique for monitoring the operating conditions of the ma- functions of the physical properties of the structure. So, any
chine. It is becoming more famous and familiar in industry alterations in these properties (mass, stiffness, damping) due
due to non-destructive in nature and allows sustainable to boundary conditions, damage, or other internal defects
monitoring without any interfering in the process. cause variations in modal parameters of the structure. Natural
frequencies are most widely used parameters in vibration
based SHM, but alterations in these frequencies are not much
3 Structural Health Monitoring (SHM) valuable to determine the damages if not very accurate mea-
surements are acquired or large levels of damage have ap-
The environment of machining operations changes unpre- peared [9]. The up-gradation in sensor technology for
dictably and implementing a well consulted SHM is in dire vibration measurement is required to minimize the drawbacks
need to heighten structural safety, reduce maintenance cost of mode shape parameters that they require a single excitation
and avoiding human and economic losses. SHM is a con- point and a roving exciter for their measurement [10]. The
tinuous and an autonomous tool for measuring the various changes in the mode shape curvatures are more suitable than
parameters (i.e. vibration, performance, bearing temperature, mode shape variations for locating damage and also, highly
etc.) to diagnose the real time condition of the distinct localized and sensitive to the region of damage [11]. The
components of the structural and mechanical systems which vibration-based SHM has become a significant and fast
is very useful for improving structural models. SHM can be growing research domain in different fields such as me-
classified as online and offline. In the long term, SHM chanical engineering, aeronautics and civil engineering to
provides current information from time to time about the assess the real time dynamic characteristics of the machine. A
machine’s ability to perform the intended operations in light crack may be undetectable at early stages by visual inspection
of unavoidable failures and degradation from the working technique because it doesn’t make changes in properties of the
place. The SHM [6] consists of three steps, i.e. signal adjacent material, but change in geometry with respect to
monitoring, signal processing, and data interpretation used normal state. Although current observations, such as, visual
in damage identification system as shown in Fig. 1. inspection, non-destructive techniques for stress corrosion
Signal processing techniques includes statistical time series and fatigue cracks are performed in the US Navy’s P-3 Orion
models, Fourier transform, short time Fourier transform, Co- aircraft [12], but a fatigue life enhancement program has been
hen’s class, wavelet transform, Hilbert–Huang transform employed to prolong its life for 10 years. It is estimated that
whereas data interpretation includes artificial neural networks, nearly half of all operating costs of machining operations can
fuzzy logic, support vector machine, Bayesian classifiers, be assigned to repair and maintenance. This is also an ample

123
Vibration Monitoring Methods and Signal Processing Techniques for SHM 587

motivation to study the unmanned SHM that makes the Vibration sensors are considered as the heart of modern
manufacturing industries to operate at low cost with greater computerized SHM systems. The selection of appropriate
customer satisfaction. It is anticipated that an accurate and sensors is needed not only to acquire a precise signal, but
reliable SHM systems could result in an increase in cutting also have a capability of interpreting what the signal means
speeds, reduction in downtime and an overall increase in to improve confidence in sensor based monitoring tech-
savings between 10 and 40 % [13]. The basic objective of nology. The cutting forces and vibration are the most widely
SHM is to determine the location, size, existence and type of used measurement parameters [18]. However, there is a
damage, and prognosis. In general, a SHM system is used for great difficulty in detecting tool faults due to the dynamics
fault detection, breakdown anticipation and problem diag- of the cutting forces. The International Standards Organi-
nostics in the machine. A bathtub curve describes the life of a zation (ISO) establishes universally acceptable metric units
machine [14] as vibration levels are used to indicate the failure for machinery vibration. Figure 2 shows the basic features
of a machine. Sometimes, the deterioration in machine takes of the vibration measurement scheme [14]. A vibration
place so rapidly that there is hardly any gap between fault sensor detects the vibration parameter from a vibrating body
detection and total failure [15]. In such cases continuous or through its mechanical structure and converts these vibration
real time monitoring with automatic shutdown is needed. signals to equivalent electrical signal. This output signal is
then analyzed by various signal processing and feature ex-
3.1 Method of Structural Health Monitoring traction techniques to measure the various characteristics of
vibration such as amplitude, frequency, displacement, ve-
The SHM methods can be divided into two types: model locity, acceleration, phase and period.
based and feature based methods [16]. In drilling, the direct measurements aren’t very efficient
Model-based methods also called failure detection in economical and technical terms [19]. He placed the
methods involve finding a model in which monitoring is various methods applied to condition monitoring that
carried out by either the variations in model parameters, indirectly measure sudden failure and tool wear into three
such as damping ratio or predict errors in the model outputs groups. The dynamic forces in a vibratory system depend
to reveal defects. These methods correlate the observa- on the displacement, velocity and acceleration components
tional signatures of abnormal structures with analytical or of a system. The analysis of dynamic system to base ex-
quantitative models to detect the damage parameters, and citation to understand the operating principle of vibration
also need a precise computational model. Various models instrument is described [20].
such as autoregressive and moving average (ARMA) Vibration pickup The integration of the contact type
model, hidden Markov model (HMM), artificial neural transducer and another device to measure vibrations is
network (ANN) etc. are falling into this category. called a vibration pickup. A seismic instrument, commonly
Feature-based methods are performed in two steps, one used vibration pickup, consists of a mass–spring–damper
is feature extraction from the signal and another is the system mounted on the vibration. The various vibration
decision-making based on these features. Various features transducers used are:
can be used, including
1. Displacement transducers (Vibrometer/proximity
(a) Time-domain features (such as mean, standard probes)
deviation, range, root mean squares, skewness, 2. Velocity transducers (Velometer)
kurtosis, crest factor, etc.); 3. Accelerometers
(b) Frequency-domain features (such as frequencies, damp- 4. Laser Doppler Vibrometers
ing ratios, energy in different frequency bands, etc.);
(c) Spatial domain features;
(d) Time–frequency domain features (i.e. time–frequen-
cy distribution); Vibrating Vibration Signal
machine or transducer conversion
The above signal processing methods will be discussed structure or pick up instrument
in Sect. 5 of this paper.

4 Vibration Measuring Methods Data Display

Analysis unit, or
Vibration measurement is an effective, reliable and non- computer
intrusive technique to monitor the condition of the machine
during startups, shutdowns and normal operations [17]. Fig. 2 Basic principle of vibration measurement

123
588 D. Goyal, B. S. Pabla

Displacement transducers measure relative displace- through tip timing with capacitance probe sensor [28].
ment, whereas the most widely used to measure absolute Capacitive sensors are well suited for high precision, high
motion are velocity and accelerometers [21]. resolution, dynamic and non-contact measurements. These
sensors have disadvantages of high output impedance, high
4.1 Displacement Transducers sensitivity to material variations or contaminants within the
clearance, low grade anti-jamming ability, and its measure
These are quite complex systems having a limited value in the is vulnerable to electrical medium and electromagnetic
study of mechanical vibrations. These are useful in low fre- fields [29].
quency vibration (below 10 Hz) measurements or if there is
small annual radial clearance gap between machine elements. 4.2 Velocity Transducers
These are generally preferred for journal bearing equipped
rotating machinery for shaft vibration measurement. These transducers are used for vibration measurements
within the frequency range of 10–1000 Hz and balancing
4.1.1 Eddy Current Proximity Sensor operations on rotating machinery. These transducers work
on the principle of electromagnetic induction. These are self-
It is widely used non-contact sensor works on the elec- generating and needs no external device for its operation.
tromagnetic induction principle that change in the resis- These are sensitive to lateral vibration so the sensor should
tance of the material due to gap changes. It consists of a be mounted in such a way that surface is straight and slightly
coil, an oscillator, a detector and a triggering circuit [22]. larger than the base. It has the advantages of cost-effec-
When there is a metal object in close proximity, the elec- tiveness, quite sensitivity, simple post-processing circuit, a
tromagnetic field in the coil induces eddy currents in non- large output signal which is linearly proportional to velocity,
magnetic but conductive materials [23]. These are used to negligible maintenance and powerful anti-interference abil-
measure blade condition in a gas turbine engine [24]. The ity; also the disadvantage is that they are susceptible to
prognostics and health management for jet engines and shocks, relatively complex and large structure [30].
other turbomachinery are also described by eddy current Laser velocimeters are a modern type of velocity
sensing [25]. Two proximity probes can be used to obtain transducers that operates on the principle of the Doppler
‘‘orbit plot’’ or ‘‘Lissajous figures’’ of shaft motion in Effect. Due to its expensiveness, these are not commonly
journal bearings [15]. used in diagnostics.

4.1.2 Inductive Proximity Switch 4.3 Accelerometer

These sensors can only be used for the detection of metal Accelerometers are electromechanical transducers designed
objects and is best suited with ferrous metals. It consists of for measuring vibratory acceleration. Its principle is based
a coil wound around a core, an oscillator, detection circuit on the piezoelectric effect that the force caused by vibration
and output circuit. The magnetic field is generated by causes the mass to ‘squeeze’ the piezoelectric materials
passing an alternating current through the coil, when a which produces an electrical charge that is proportional to
metal object comes closer to the coil end, the inductance of the force exerted upon it. The different designs of an ac-
the coil changes. This change is continuously monitored by celerometer are ‘compression type’ and ‘shear type’. In
an L–C oscillator circuit. This circuit further triggers a general the shear configuration gives the best all-round re-
switch to change in preset value of inductance. It is con- sults for general purpose accelerometers and the compres-
sidered an optimum measurement method for rotor-to-sta- sion design is used for particular applications [31]. These are
tor position [26]. employed to measure the vertical vibrations for tool wear
monitoring of turning operations [32]. A Kistler mini ac-
4.1.3 Capacitive Sensors celerometer is used to monitor the online condition of metal
cutting tool, and to estimate the flank wear area in inserts for
It is non-contact type sensor works on principle of mea- CNC turning [33, 34]. The main advantages of the piezo-
suring the electrical capacitance of the gap between the electric accelerometer include, light weight and high fre-
plates. These sensors can also be used as proximity sensors, quency range, compactness, ruggedness and high sensitivity
consists of a single capacitor plate probe with the object as [35]; the disadvantages are that these have a high output
the other plate, which has to be metallic and earthed. These impedance that generate small voltages (order of 1 mV),
are allowed for dynamic measurement in a surface force vulnerable to interference from the external environment,
apparatus within the range of 0–100 Hz [27]. The vibration the weak output signal which requires amplification through
amplitude of turbomachinery blade has also been measured the amplifier circuit and detection by detecting circuit [36].

123
Vibration Monitoring Methods and Signal Processing Techniques for SHM 589

4.4 Laser Doppler Vibrometers (LDV) 5 Signal Processing and Feature Extraction
Techniques
The LDV is an interferometric device which measures the
absolute velocity of a vibrating body through the mea- The extraction of features using signal processing is con-
surement of the Doppler shift of scattered laser light sidered as one of the major elements and challenging aspect
coming from the target. It works on laser Doppler princi- of vibration based machine condition monitoring (MCM).
ple, in which a frequency-modulated coherent laser beam The aim is to accentuate features from the sensed signals,
is reflected from a vibrating surface, and Doppler shift, which are usually noisy and complex and have to be further
fshift of the reflected beam is compared with the reference processed to yield the features of our interest. The selection
beam, fref. A Doppler signal processor demodulates the of the signal processing method is paramount to effectively
photo detector signal, which generates a time resolved identify and diagnose defects that are indicative of potential
velocity of the vibrating body [37]. These techniques are machine failure. Due to the complex processes involved in
capable of high accuracy and sensitivity, and considered to the structural response to dynamic loading, a significant
be a unique instrument for repeatable vibration measure- research effort has been applied to health monitoring of a
ments to overcome the various problems, such as fre- machine with the emergence of a broad range of methods,
quency response, intrusivity etc. These instruments are techniques and algorithms. The various methods based on
used for modal analysis measurement are used for modal level of identification are classified to extract the features for
analysis measurement, high temperature surfaces and noisy obtaining the current health of a machine [42, 43].
environments or to analyze the small and weal structures
Level 1: To determine the presence of damage in
[21, 38]. Laser Doppler Vibrometer is used for observing
structure
changes in mode shapes and natural frequencies for var-
Level 2: To determine the geometric location of the
ious types of damages such as cut-out, impact and de-
damage
lamination in graphite–epoxy composite beams [38]. The
Level 3: To quantify the severity of the damage
result is virtually independent of the transmitter/receiver
Level 4: To predict the remaining service life of the
motion due to the occurrence of the frequency shift at
structure
reflection so it measures the absolute rather than relative
Level 5: Self healing structures
velocity.
These Vibrometers requires less data storage and Park et al. [43] suggests ‘level 5’ in the context of smart
processing as compared with a full field measuring in- structures with emergence of shape memory alloys i.e. Cu–
struments like real time and time-average holography. Al–Ni and Ni–Ti (NiTi) alloys. The ultimate aim of SHM is
The disadvantages include ‘‘speckle drop out’’ (the to predict the remaining service life of the damage. Too much
speckle noise can distort LDV signal), line of sight re- noise in the vibration signals makes impossible to extract the
quired from the laser head to the target, the time required useful information through signal processing techniques so
for testing and demodulation of data [39]. The im- adaptive noise cancellation (ANC) is one such technique that
provements in speckle noise interference would be nec- enhances the signal to noise ratio [44]. The various vibration
essary for a successful application to actual field transducers like accelerometer etc. measures raw data ob-
structures [40]. This technique has a great advantage that tained from the structure in time domain. Although, the time
the measurement point can be changed easily and quickly domain analysis is direct and provide useful information, but
by deflecting the laser beam without loading the target has a limitation that it doesn’t provide clear information
object. These have a very limited application in SHM about all the frequencies present in the response of the
due to its expensiveness. At present, due to its bulkiness structure. Fast Fourier transform (FFT), a real time analysis is
and non-portability, these are unsuitable for intermittent in dire need to perform quick transformations of functions
monitoring [21]. from time domain to frequency domain or vice versa. The
In future, Vibrometers could be developed into a hard further analysis is often undertaken to obtain modal domain
hat, and the operator would just have to look in the ma- data from the frequency domain data. Now, modern tech-
chine direction, utter the ID of the machine into a micro- niques are also available to directly convert the time domain
phone, and the laser and imaging system would locate the data to modal domain data [45]. During the conversion be-
machine and take measurements at a prescribed number of tween the domains, some compression of the data takes
monitoring points on it. Currently, however, they are not place, but for linear systems, there is a little loss of infor-
really a viable option for regular condition monitoring, mation between the time and frequency domains [46]. Cur-
even though they are used in production quality control rently, signal processing methods based on modal domain are
measurements [41]. widely discussed in the literature because mode shapes,

123
590 D. Goyal, B. S. Pabla

modal damping factors and natural frequencies are easily probability function is a non-decreasing function, lies be-
interpreted than features extracted from the time domain (e.g. tween 0 and 1. Its variation with variable is described in
residuals of autoregressive model) and frequency domain [20]. The probability density function (PDF) is used to
(e.g. the distortion identification functions) [11]. obtain the different statistical parameters of a signal by
Obviously, there are no universal physical variables and taking various moments for mean value, x:
signal processing techniques that would be appropriate for Z1
all applications and also, it is not practically possible to x ¼ xpðxÞdx ð1Þ
express all signal processing methods for various applica-
1
tions [47]. To keep it general, this section summarizes the
most common ones in chronological order: The First moment defines the ‘center of gravity’, which
occurs when the area under the curve is unity from Eq. (1).
5.1 Statistical Time Series Models Second moment defines the variance about the mean value,
corresponds to the ‘moment of inertia’. The third moment
The earliest used models in SHM are time series (TS) models defines the ‘skewness’ (S), which is zero for symmetrical
which are developed as an approximate mathematical model functions and large for asymmetrical functions. Fourth
based on a set of input–output measurements. They are moment defines ‘kurtosis’ (K) that measures spiky or im-
classified into two categories, first, the linear statistical TS pulsive signals. The statistical parameter ‘kurtosis’ is used
model (e.g. autoregressive model, moving-average model, for predicting the condition of rolling element bearing by
autoregressive moving average etc.) and second, the non- measuring the bearing housing vibration [54].
linear statistical TS models (e.g. non-linear autoregressive
with exogenous inputs, non-linear autoregressive moving 5.3 Fast Fourier Transform (FFT)
average with exogenous inputs) [48]. In 1951, Peter Whittle
introduced the general ARMA model which is considered as During the vibration transmissions, the vibration produced
a time series model based condition monitoring technique by a machine component consists of certain frequencies
carried out by either the variations in model parameters, such don’t change, although their levels may change from one
as damping ratio or predict errors in the model outputs to location to another. Frequency analysis of the vibration
reveal defects. The time series analysis of vibration signals signal is widely used to diagnose the machine faults [16].
was used to diagnose the damages in mechanical structures Fourier transform (FT), a frequency domain representation
[49]. These are among the most utilized methods for para- that estimates the strength of different frequency compo-
metric representations of a signal and condition assessment nents (the power spectrum) of a time-domain signal. The
of structures under dynamic loading. The time series analysis forward FT is used to convert the signal from time to
of acoustic emission is also performed to monitor the wear of frequency domain, and an inverse FT is used to convert
cutting tool [50]. It is employed to extract the wear sensitive from frequency to time domain. FFT is considered as more
features for the estimation of tool major flank wear in a effective and efficient diagnosis technique to obtain the
turning process and establish the direct relation to the natural Fourier Transform of discretized time signals. This signal
frequency of the tool/holder set by developing a metric cal- is considered for a finite time called the ‘‘frame’’ or ‘‘time
culated based eigen value system [51]. It provides a measure window’’, which is then digitized and stored for feature
of the wear accelerating stage and the region using eigen- extraction. The selection of an appropriate sampling rate is
values and dispersion ratios. ARMA is successfully accepted important for signal digitization to avoid false frequency
to analyze and forecast the rotate speed signal of aero-engine components that take place due to aliasing [15]. As per
and also to predict the value transformation trend of signal Nyquist theorem, the sampling frequency should be at least
[52]. ARMA can’t model noise contaminated measured twice the maximum frequency present in the signal. FFT
structural response and can’t model problems like non-linear has been used in various types of structures to detect
behavior of civil structures [6]. Hence, Statistical non-linear damage in pipes, scaled bridge model made of three steel
TS models have been proposed to overcome the statistical girders subjected to impact test, identification of a 3D
linear TS models. truss-type structure subjected to earthquakes and forced
excitation [55–57].
5.2 Probability Distribution and Density Function In a real environment, the majority of the signals is non-
stationary, i.e. spectra varies with time. During the working
It is a way of characterizing random signal to distribute the mode, machines generate the vibrations, which has a dif-
value at any arbitrarily chosen time. It can be expressed in ferent frequency at each instant of time has a distinct fre-
‘probability distribution’ i.e. the probability of its variable quency [58]. The FFT has significant limitations that it
is less than or equal to a specified value [53]. The can’t depict the change in frequency signals (vibration,

123
Vibration Monitoring Methods and Signal Processing Techniques for SHM 591

speech or biomedical signals) content over the time, also of signal processing it is commonly known as Wigner–Ville
can’t be used for monitoring real structures subjected to Distribution. Cohen’s class (CC) distribution is a technique
dynamic excitations [48]. So the joint time–frequency to evaluate the energy of time-varying systems [66]. The
analysis technique is used to characterize a signal simul- WVD is one of the so-called ‘Cohen’s class’ of time fre-
taneously into a two-dimensional representation (time vs. quency distributions [60]. The WVD is also applied based
frequency) by comparing them with elementary functions on empirical mode decompositions (EMDs) for the fault
such as the frequency modulated Gaussian functions [59]. diagnosis of bearing [67]. Although the WVD has advan-
The major difference between each time–frequency tages of simplicity, excellent accuracy, effectiveness, no
method is to handle the problem of uncertainty. The requirement of window function for its analysis, better
uncertainty principle states that ‘‘one can’t simultaneously resolution than STFT, and invariant to window effects but it
have good frequency resolution and good time resolution’’ also has drawbacks of cross-term interference (i.e. gen-
[60]. To overcome the time information loss problem of eration of spurious frequency not contained in the original
FFT, three signal processing techniques is described in the signal). Choi–William [68] developed alternative kernels to
following sections: short-time Fourier transform (STFT); overcome this problem, but at the cost of lower resolution
Wigner–Ville distribution (WVD), and wavelet transform and an increase in computational resources. Various
(WT). smoothing techniques are suggested to retain optimum
resolution by first locating the interference components, and
5.4 Short-Time Fourier Transform then withdraw from the unsmoothed WVD [69]. A smooth
pseudo WVD is used to monitor the condition of the gearbox
This technique is introduced to overcome the problems of by detecting the broken tooth in a spur gear [70]. WVD is
FFT called a STFT, which is an extension of FFT capa- also suggested to spread the energy of the frequency com-
bility of analyzing non-stationary or noisy signals. The ponent by using a Gaussian window [71]. It is also used for
basic idea is to divide the initial signal into small time contrasting the acoustic and vibration signals to detect the
windows and apply the Fourier transform to each time failures in gear, and compare the time–frequency feature of
segment for representing the variation in signal frequency cracked rotor [72, 73].
content over time that existed in that segment. It has been
used in various applications for e.g. to estimate the modal 5.6 Wavelet Transform (WT)
parameters of a 3D truss-type structures, 7-story RC frame,
3-story 3D steel frame and beams [55, 61–63]. It provides In 1982, Jean Morlet introduced the idea of fine grained
constant absolute bandwidth analysis to identify harmonic time–frequency analysis approach to achieve an optimal
components and offers constant resolution in two-dimen- balance between frequency resolution and time resolution
sional representation, irrespective of the actual frequency for seismic wave analysis. In STFT, complex cosine and
[64]. The size of the window determines the accuracy of sine functions are used to map the signal from time to
time, larger the window the poor time resolution, better frequency domain, whereas WT consists of a family of
frequency resolution and vice versa. elementary functions (as a function of time) that can be
The STFT has a limitation of low-resolution problem. independently dilated and shifted, known as wavelets.
This problem is resulted choosing a small window, which Discrete wavelet transform (DWT) and continuous wavelet
doesn’t permit the dynamic transient behavior of the transform (CWT) are two methods to calculate the wavelet
structure adequately. Also, it doesn’t allow observing two transform of a signal. In DWT, the scaling factor is chosen
closed natural frequencies [65]. as a power of two and it is generally implemented through
a pair of lowpass and highpass wavelet filters [74]. In
5.5 Cohen’s Class CWT, the scaling factor be chosen arbitrarily or by means
of convolution.
The WVD is one of the most general time–frequency ana- This method is more sensitive to stiffness variation as
lysis techniques, seems to violate the uncertainty principle to compared to WVD [73], also crack can be easily found if
provide excellent resolution for accurate examination in the unbalance increases significantly. WT gives a better
both domains, which leads to use this method for monitoring time localization at high frequencies and leads to analysis
the nonlinear behavior of civil structure. In 1932, Eugene with constant percentage (or relative) bandwidth, which
Wigner developed the original Wigner distribution to study makes WT more effective than STFT. WT has great po-
the problem of statistical equilibrium in quantum mechanics. tential for detecting abrupt changes in a signal or locate a
After 15 years, Jean Ville modified the Wigner distribution sudden change in response from an acceleration time re-
and first introduced in signal analysis to eliminate interfer- sponse It is robust and insensitive to changing working
ence between the frequency components. Now, in the field conditions. WT has been used extensively in different

123
592 D. Goyal, B. S. Pabla

fields such as mechanical engineering, transportation amplitude and phase angle from the non-linear or non-sta-
engineering, biomedical engineering, power engineering, tionary signals. HHT has been applied for vibration signal
robotics, earthquake, structural vibration control, image analysis in different fields like mechanical engineering, bio-
processing, and due to its advantage of noise elimination, medical and earthquake engineering. It is used for the crack-
data compression and computational efficiency [48]. induced motor to interpret the non-linear response [78]. It is
Complex harmonic wavelets (typically of one-octave also suggested as damage identification approach for ASCE
bandwidth) are described which are particularly useful for structural health monitoring benchmark structure [79].
denoising purposes with compact in the frequency domain, Although, the HHT has advantages of non-involvement
but infinite in the time domain [75]. They are narrower and of any convolution, less computational time and become a
have the appearance of windowed sinusoids (harmonic promising method to extract the features of non-linear and
functions). The various wavelets are used for SHM [48]. non-stationary signal, but it also has some shortcomings
Complex wavelets have the advantage of non-sensitivity to [80]. First, the EMD generates unenviable IMFs at low-
the phasing of the event being transformed because the frequency region, may lead to result misinterpretation. Se-
imaginary part of the wavelet is orthogonal to the real part cond, the monocomponent property of obtained IMF can’t
(sine rather than cosine). be achieved due to wide frequency range. Third, the EMD
can’t separate the signals of low frequency components. So
5.7 Cepstrum Analysis an improved HHT technique is in dire need to solve these
shortcomings. In the improved version [81], wavelet packet
The ‘‘cepstrum’’ was coined by paraphrasing the word transform (WPT) is used as a preprocessor to separate the
‘‘spectrum’’ along with a number of similarly derived terms analyzed signal into a family of narrow band signals.
[76]. The ‘power cepstrum’ is defined as the inverse Fourier
transform of the log power spectrum or amplitude spectrum. 5.9 Power Spectral Density (PSD)
It is widely used to identify one or more periodic structure
present in the spectra, and each appears as a distinct peak Power spectral density is a powerful tool to demonstrate
corresponding to the spacing of the harmonics and sidebands the strength of the variations as a function of frequency. It
in cepstrum, also for the processing of signals (e.g. seismic, is used to recognize and measure the amplitude of oscil-
underwater, or aero engine noise with ground reflections), latory signals in time series data, and tells at which fre-
speech analysis, and to measure time delays. This technique quencies the variations (energy) are strong or weak which
has an important application in machine diagnostics (e.g. may be quite effective for further analysis. It looks like a
gear box vibrations) to identify the modulating frequencies. simple time series graph which is a function of frequency
Gear vibrations contain rotating frequency harmonics, rather than time. This technique consumes a small amount
meshing harmonics, which are quite complex due to over- of processing power, and it is used for fault diagnoses of
lapping families of sidebands [77]. It provides a practical induction machines [22]. PSD is computed by FFT directly
system for routine prognostic monitoring by condensing the or by transforming the autocorrelation function. So the
frequency domain information into interpreting form. power spectral density, is defined as the FT of the auto-
correlation function divided by 2p taking over a large pe-
5.8 Hilbert–Huang Transform riod of time [20]. The integration of PSD function over the
specific frequency range is used to find the mean square
In 1905, David Hilbert first introduced the ±p/2 phase shift value of the random function and energy. The unit of PSD
operator as a basic property of Hilbert transform. The Hilbert is energy per frequency (width).
transform is a convolution between the Hilbert transformer
and a function [47]. Huang et al. [67] introduced the HHT as a
data analysis tool or an adaptive signal processing technique to 6 Conclusions
extract the properties of stationary, non-stationary and tran-
sient signals, especially for time–frequency–energy repre- Due to evolution and recent development in sensors, signal
sentations. HHT is applied in two steps, first, the EMD i.e. a processing methods, and powerful computers, SHM is
key part of HHT, will decompose the any time series data into paramount for the area of research in engineering struc-
a collection of band-limited quasi-stationary functions called tures. This paper reviewed an overview of the various vi-
intrinsic mode functions (IMF). These monocomponent bration measurement methods and signal processing
functions are kind of adaptive, complete and almost ortho- techniques used in SHM.
gonal representations for the signal to be analyzed. Second, the The most used signal processing techniques in SHM are
Hilbert spectral analysis is applied to each IMF to derive the time series models, wavelet transform, and HHT or
local energy of all the instantaneous frequency, or to obtain its sometimes a combination of any two of them.

123
Vibration Monitoring Methods and Signal Processing Techniques for SHM 593

Although the large amount of work has been reported in 13. Altintas Y, Yellowley I, Tlusty J (1988) The detection of tool
SHM, it is restricted to small and academic problems. The breakage in milling operations. J Manuf Sci Eng 110(3):271–277
14. Rao SS (1995) Mechanical vibrations. Addison-Wesley, Boston
monitoring of smart structures is a big challenge to detect 15. Tandon N, Parey A (2006) Condition monitoring of rotary machines.
the faults or damages from the huge amount of noisy data In: Wang L, Gao RX (eds) Condition monitoring and control for
collected from many sensors on a periodic basis. The vi- intelligent manufacturing. Springer, London, pp 109–136
bration data accumulated form big structures are often very 16. Wang L, Gao RX (2006) Condition monitoring and control for
intelligent manufacturing. Springer, Berlin
noisy. The new methodologies for online monitoring are 17. Scheffer C, Girdhar P (2004) Practical machinery vibration
needed to handle noisy data effectively, and be precise, analysis and predictive maintenance. Elsevier, Amsterdam
ascendable, portable, and expeditious computationally. It 18. Snr D, Dimla E (2000) Sensor signals for tool-wear monitoring in
has been found that the contact type instruments are mostly metal cutting operations—a review of methods. Int J Mach Tools
Manuf 40(8):1073–1098
used for vibration measurements in SHM. The use of non 19. Jantunen E (2002) A summary of methods applied to tool condition
contact type laser based instruments will overcome the monitoring in drilling. Int J Mach Tools Manuf 42(9):997–1010
drawbacks of traditional sensors which will eliminate the 20. Rao JS, Gupta K (1999) Introductory course on theory and practice
need for mounting the sensor on the machine and results in of mechanical vibrations. New Age International, Seborga
21. Randall RB (2011) Vibration-based condition monitoring: in-
quick measurement, high accuracy and non intrusiveness. dustrial, aerospace and automotive applications. Wiley, London
From the above review, it is concluded that there is huge 22. Cusido J, Romeral L, Ortega JA, Rosero JA, Garcia Espinosa A
potential for use of non contact instruments for vibration (2008) Fault detection in induction machines using power spec-
measurement in predicting the maintenance and failure tral density in wavelet decomposition. IEEE Trans Ind Electro
55(2):633–643
aspects of dynamic machines. 23. Bolton W (2003) Mechatronics: electronic control systems in
mechanical and electrical engineering. Pearson Education, Upper
Saddle River
24. Terpay GW, Zipfel Jr GG (1999) Measuring blade condition in a
gas turbine engine using eddy-currents. In: Ninth international
References conference on adaptive structures and technologies. October 14–
16, 1998, Boston, CRC, p 71
1. McMillan RB (2004) Rotating machinery: practical solutions to 25. Dowell M, Sylvester G (1999) Turbomachinery prognostics and
unbalance and misalignment. Fairmont Press, Lilburn health management via eddy current sensing: current develop-
2. Tsang AH (1995) Condition-based maintenance: tools and deci- ments. In: Aerospace conference, 1999. Proceedings 1999 IEEE,
sion making. J Qual Maint Eng 1(3):3–17 vol 3, IEEE, pp 1–9
3. Meher-Homji CB (1995b) Blading vibration and failure in gas 26. Adams ML (2000) Rotating machinery vibration: from analysis
turbines part B: compressor and turbine airfoil distress. ASME to troubleshooting, vol 131. CRC Press, Boca Raton
paper no. 95-GT-419 27. Restagno F, Crassous J, Charlaix E, Monchanin M (2001) A new
4. Cao Y, Rong XL, Shao SJ, He KP (2012). Present situation and capacitive sensor for displacement measurement in a surface-
prospects of vibration sensors. In: Computer distributed control force apparatus. Meas Sci Technol 12(1):16
and intelligent environmental monitoring (CDCIEM), 2012 in- 28. Lawson CP, Ivey PC (2005) Tubomachinery blade vibration
ternational conference on. IEEE, pp 515–518 amplitude measurement through tip timing with capacitance tip
5. Peng Z, Kessissoglou N (2003) An integrated approach to fault clearance probes. Sens Actuators, A 118(1):14–24
diagnosis of machinery using wear debris and vibration analysis. 29. Zhao-Hua W (2006) Capacitive sensor based on active bridge
Wear 255(7):1221–1232 circuit and its measurement system. Beijing University of Che-
6. Adeli H, Jiang X (2009) Intelligent infrastructure: neural net- mical Technology Master Degree Thesis, 11
works, wavelets, and chaos theory for intelligent transportation 30. Li MG, Sun RX, Li Y (2008) Research on vibration sensor of
systems and smart structures. CRC Press, Boca Raton ultra low frequency. Colliery Mech Electr Technol 3:012
7. Fritzen CP, Balageas D, Güemes A (eds) (2006) Structural health 31. Bruel and Kjaer (1984) Mechanical vibration and shock mea-
monitoring, vol 493. ISTE, London surements. Bruel and Kjaer, Denmark, pp 97–120
8. Speckmann H, Roesner H (2006) Structural health monitoring: a 32. Silva RG, Reuben RL, Baker KJ, Wilcox SJ (1998) Tool wear
contribution to the intelligent aircraft structure. In: Proceeding of monitoring of turning operations by neural network and expert
ECNDT system classification of a feature set generated from multiple
9. Doebling SW, Farrar CR, Prime MB, Shevitz DW (1996) sensors. Mech Syst Signal Process 12(2):319–332
Damage identification and health monitoring of structural and 33. Dimla DE, Lister PM (2000) On-line metal cutting tool condition
mechanical systems from changes in their vibration characteris- monitoring. I: force and vibration analyses. Int J Mach Tools
tics: a literature review (No. LA–13070-MS). Los Alamos Na- Manuf 40(5):739–768
tional Lab, NM 34. Trejo-Hernandez M, Osornio-Rios RA, Romero-Troncoso RDJ,
10. Castellini P, Martarelli M, Tomasini EP (2006) Laser doppler Rodriguez-Donate C, Dominguez-Gonzalez A, Herrera-Ruiz G
vibrometry: development of advanced solutions answering to (2010) FPGA-based fused smart-sensor for tool-wear area quanti-
technology’s needs. Mech Syst Signal Process 20(6):1265–1285 tative estimation in CNC machine inserts. Sensors 10(4):3373–3388
11. Carden EP, Fanning P (2004) Vibration based condition 35. Bouche RR (1979) Calibration of shock and vibration measuring
monitoring: a review. Struct Health Monit 3(4):355–377 transducers (No. SVM-11). Shock and Vibrations Information
12. Shah BM, Cannington JC, Ferko EJ, Kwon YS, Saunders SA, Centre, Washington
Phan N, Candela JM (2004) P-3C service life management. In: 36. Fu-Cheng X (2008) Common vibration sensor and its typical
Proceedings of USAF ASIP conference application circuit. Pract Electron 10:13–15

123
594 D. Goyal, B. S. Pabla

37. Tatar K (2006) Machine tool vibrations and violin sound fields 62. Nagata Y, Iwasaki S, Hariyama T, Fujioka T, Obara T, Wakatake
studied using laser vibrometry. Doctoral dissertation, Lulea T, Abe M (2009) Binaural localization based on weighted Wiener
University of Technology gain improved by incremental source attenuation. IEEE Trans
38. Righini GC (ed) (2009) An introduction to optoelectronic sensors. Audio Speech Lang Process 17(1):52–65
World Scientific, Singapore 63. Yesilyurt I, Gursoy H (2013) Estimation of elastic and modal
39. Stanbridge AB, Ewins DJ (1999) Modal testing using a scanning parameters in composites using vibration analysis. J Vib Control.
laser doppler vibrometer. Mech Syst Signal Process 13(2):255–270 doi:10.1177/1077546313486275
40. Khan AZ, Stanbridge AB, Ewins DJ (2000) Detecting damage in 64. Gade S, Gram-Hansen K (1996) Non-stationary signal analysis
vibrating structures with a scanning LDV. Opt Lasers Eng using wavelet transform, short-time Fourier transform and
32(6):583–592 Wigner–Ville distribution. Tech Rev 2, Bruel & Kjaer
41. Vass J, Smı́d R, Randall RB, Sovka P, Cristalli C, Torcianti B 65. Amezquita-Sanchez JP, Garcia-Perez A, Romero-Troncoso RJ,
(2008) Avoidance of speckle noise in laser vibrometry by the use Osornio-Rios RA, Herrera-Ruiz G (2012) High-resolution spec-
of kurtosis ratio: application to mechanical fault diagnostics. tral-analysis for identifying the natural modes of a truss-type
Mech Syst Signal Process 22(3):647–671 structure by means of vibrations. J Vib Control. doi:10.1177/
42. Rytter A, Kirkegaard PH (1994) Vibration based inspection of 1077546312456228
civil engineering structures. Aalborg Universitetsforlag, Aalborg 66. Cohen L (1966) Generalized phase-space distribution functions. J
43. Park G, Muntges DE, Inman DJ (2001) Self-monitoring and self- Math Phys 7(5):781–786
healing jointed structures. Key Eng Mater 204:75–84 67. Huang NE, Shen Z, Long SR, Wu MC, Shih HH, Zheng Q et al
44. Khemili I, Chouchane M (2005) Detection of rolling element bearing (1998) The empirical mode decomposition and the Hilbert
defects by adaptive filtering. Eur J Mech-A/Solids 24(2):293–303 spectrum for nonlinear and non-stationary time series analysis.
45. Maeck J, Abdel Wahab M, Peeters B, De Roeck G, De Visscher Proc R Soc Lond Ser A: Math Phys Eng Sci 454(1971):903–995
J, De Wilde WP et al (2000) Damage identification in reinforced 68. Choi HI, Williams WJ (1989) Improved time–frequency repre-
concrete structures by dynamic stiffness determination. Eng sentation of multicomponent signals using exponential kernels.
Struct 22(10):1339–1349 IEEE Trans Acoust Speech Signal Process 37(6):862–871
46. Friswell MI, Penny JET (1997) Is damage location using vibra- 69. Chiollaz M, Favre B (1993) Engine noise characterization with
tion measurements, practical. In: Proceedings of EUROMECH Wigner–Ville time–frequency analysis. Mech Syst Signal Process
365 international workshop: DAMAS, vol 97 7(5):375–400
47. Li CJ (2006) Signal processing in manufacturing monitoring. In: 70. Staszewski WJ, Worden K, Tomlinson GR (1997) Time–frequency
Wang L, Gao RX (eds) Condition monitoring and control for analysis in gearbox fault detection using the Wigner–Ville distribu-
intelligent manufacturing. Springer, London, pp 245–265 tion and pattern recognition. Mech Syst Signal Process 11(5):673–692
48. Amezquita-Sanchez JP, Adeli H (2014) Signal processing tech- 71. Claasen TACM, Mecklenbrauker WFG (1980) The Wigner dis-
niques for vibration-based health monitoring of smart structures. tribution—a tool for time–frequency signal analysis. Part I:
Arch Comput Methods Eng. doi:10.1007/s11831-014-9135-7 continuous-time signals. Philips J Res 35(3):217–250
49. Sohn H, Farrar CR (2001) Damage diagnosis using time series 72. Baydar N, Ball A (2001) A comparative study of acoustic and
analysis of vibration signals. Smart Mater Struct 10(3):446 vibration signals in detection of gear failures using Wigner–Ville
50. Liang SY, Dornfeld DA (1989) Tool wear detection using time series distribution. Mech Syst Signal Process 15(6):1091–1107
analysis of acoustic emission. J Manuf Sci Eng 111(3):199–205 73. Zou J, Chen J (2004) A comparative study on time–frequency
51. Aghdam BH, Vahdati M, Sadeghi MH (2014) Vibration-based feature of cracked rotor by Wigner–Ville distribution and wavelet
estimation of tool major flank wear in a turning process using transform. J Sound Vib 276(1):1–11
ARMA models. Int J Adv Manuf Technol 76(9–12):1631–1642 74. Mallat SG (1989) A theory for multiresolution signal decompo-
52. Liu ZG, Cai ZJ, Tan XM (2011) Forecasting research of aero- sition: the wavelet representation. IEEE Trans Pattern Anal Mach
engine rotate speed signal based on ARMA model. Proced Eng Intell 11(7):674–693
15:115–121 75. Newland DE (2007) Wavelet analysis of vibration signals. Wiley,
53. Papoulis A, Pillai SU (2002) Probability, random variables, and London, pp 585–597
stochastic processes. Tata McGraw-Hill Education, Noida 76. Bogert BP, Healy MJR, Tukey JW (1963) The quefrency alanysis
54. Dyer D, Stewart RM (1978) Detection of rolling element bearing of time series for echoes: cepstrum pseudo-autocovariance, cross-
damage by statistical vibration analysis. J Mech Des 100(2):229–235 cepstrum and saphe cracking. In: Rosenblatt M (ed) Proceedings
55. Amezquita-Sanchez JP, Osornio-Rios RA, Romero-Troncoso RJ, of the symposium on time series analysis. Wiley, New York, pp
Dominguez-Gonzalez A (2012) A hardware–software system for 209–243
simulating and analyzing earthquakes applied to civil structures. 77. Randall RB (1973) Cepstrum analysis and gearbox fault diag-
Nat Hazards Earth Syst Sci 12(1):61–73 nosis. Bruel and Kjaer, Denmark
56. Lee J, Kim S (2007) Structural damage detection in the frequency 78. Yang B, Suh CS (2004) Interpretation of crack-induced rotor non-
domain using neural networks. J Intell Mater Syst Struct 18:785–792 linear response using instantaneous frequency. Mech Syst Signal
57. Cheraghi N, Zou GP, Taheri F (2005) Piezoelectric-based Process 18(3):491–513
degradation assessment of a pipe using fourier and wavelet 79. Yang JN, Lei Y (2000) System identification of linear structures
analyses. Comput-Aided Civil Infrastruct Eng 20(5):369–382 using Hilbert transform and empirical mode decomposition. In:
58. Shin YS, Jeon JJ (1993) Pseudo Wigner–Ville time–frequency SPIE proceedings series. Society of photo-optical instrumentation
distribution and its application to machinery condition monitor- engineers, pp 213–219
ing. Shock Vib 1(1):65–76 80. Peng ZK, Tse PW, Chu FL (2005) An improved Hilbert–Huang
59. Qian S, Chen D (1996) Joint time–frequency analysis: methods transform and its application in the vibration signal analysis. J
and applications. Prentice Hall Inc, Englewood Cliffs Sound Vib 286(1):187–205
60. Cohen L (1995) Time–frequency analysis, vol 778. Prentice Hall 81. Dequeiroz RL, Rao KR (1993) Time-varying lapped transforms
PTR, Englewood Cliffs and wavelet packets. IEEE Trans Signal Process 41(12):3293–
61. Yinfeng D, Yingmin L, Mingkui X, Ming L (2008) Analysis of 3305
earthquake ground motions using an improved Hilbert–Huang
transform. Soil Dyn Earthq Eng 28(1):7–19

123