You are on page 1of 14

Journal of Sound and Vibration 442 (2019) 414e427

Contents lists available at ScienceDirect

Journal of Sound and Vibration


journal homepage: www.elsevier.com/locate/jsvi

Identification of unbalance in a rotor system using a joint


input-state estimation technique
Akash Shrivastava, Amiya R. Mohanty*
Department of Mechanical Engineering, Indian Institute of Technology, Kharagpur, 721302, India

a r t i c l e i n f o a b s t r a c t

Article history: A technique for estimation of unbalance parameters of rotor-bearing systems has been
Received 12 March 2018 presented and verified by numerical simulations and experimental measurements. The
Received in revised form 17 September 2018 method uses system model and response measurements for unbalance identification. A
Accepted 12 November 2018
rigid rotor model is considered in the present work. Unbalance force estimation using a
Available online 14 November 2018
Handling Editor: I. Trendafilova
joint-input state estimation technique is followed by unbalance parameter identification
using least-squares fitting. Results are presented for different shaft speeds and measure-
ment noise levels. Unbalance parameters are identified accurately with noisy measure-
Keywords:
Joint input-state estimation
ments, which verifies the robustness of the proposed technique. The proposed technique is
Unbalance identification verified on a rotor-bearing test rig for different unbalance configurations. Sensitivity
Rotor-bearing system analysis is also performed by varying the values of system parameters.
Model-based technique © 2018 Elsevier Ltd. All rights reserved.
Force estimation
Noisy measurements

1. Introduction

Rotating machines are important assets in the industry. In all kind of industries, rotating machines, e.g., generators,
motors, pumps, etc. are used widely. Rotating machine failures affect production and its cost, therefore, fault diagnosis of
rotating machines has been an active area of research to prevent their failures.
The occurrence of faults in machines causes a change in dynamic behavior, which can be detected from the measured
vibration signals. Conventional signal processing based techniques use vibration signatures for fault identification. Mostly,
these techniques provide qualitative information regarding machines’ condition. Recent developments in the field of con-
dition monitoring indicate that the prediction of the dynamic behavior of rotating machinery can be enhanced by combining
measurements and models [1]. Model-based approaches are found to be more effective in fault diagnosis and provide in-
formation about fault type and its extent [2e6]. The model-based approach is found to be more effective over signal-based
approaches for fault diagnosis of rotor systems where active magnetic bearings are used [7]. Furthermore, model-based
approaches can be employed for detection of incipient faults. In model-based techniques, the change in machine’s dy-
namic behavior due to fault is assumed to be caused by some unknown forces and moments which are termed as equivalent
loads. Firstly, these loads are identified/estimated from residual vibrations and system model, and then compared with
mathematical models of the faults. By comparison, the type, location, and extent of the faults are obtained.
The most common rotating machinery fault is mass unbalance, which is caused by uneven distribution of the mass [8].
Markert et al. [3] presented a model-based technique for the identification of unbalance fault parameters. They used modal

* Corresponding author.
E-mail address: amohanty@mech.iitkgp.ac.in (A.R. Mohanty).

https://doi.org/10.1016/j.jsv.2018.11.019
0022-460X/© 2018 Elsevier Ltd. All rights reserved.
A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 415

expansion technique to obtain residual vibrations at all degrees of freedom of the model. Fault parameters are obtained by
time domain least-squares fitting of calculated equivalent forces and forces from the fault model. The effect of limited
measurements, numerical integration of accelerations, and occurrence of multiple faults is also examined. The smearing of
equivalent loads is found due to modal expansion, which may lead to incorrect fault identification. Later, this technique was
used for the identification of transverse crack [5], unbalance and misalignment [6]. Bachschmid et al. [4] used the model-
based technique to identify multiple faults (unbalance, shaft bow and misalignment) of the rotor system where frequency
domain least-squares fitting of multi-dimensional residual between measured and calculated vibration was used.
Kalman filter based techniques have been presented by several researchers in the rotor-dynamics field. A technique for
crack localization is presented in Ref. [9]. Lee and Chen [10] presented Kalman filter and recursive least-square based
technique integrated with a fuzzy logic technique to estimate forces generated by rotating machines; the technique was
verified using numerical simulations. In Ref. [11], non-linear version of Kalman filter (extended Kalman filter) is used for the
estimation of bearing parameters. Khanam et al. [12] have proposed a Kalman filter based technique to improve the signal to
noise ratio for bearing fault diagnosis. The techniques presented in Refs. [9,11], and [12] assume known excitations to the
system, which may not be the case in practical situations. Meanwhile, some force estimation techniques are developed and
successfully implemented on machines and civil structures. Recently, authors of the present paper have proposed a model-
based unbalance parameter estimation technique using Kalman filter based force estimation technique [13].
Joint input-state estimation (JIS) technique is an input identification technique proposed by Gillijns and De Moor [14] and
later applied in structural dynamics [15]. In Ref. [16], the conditions for the invertibility of linear system model and identi-
fiability of the forces are derived for a reduced order model. In a case study, the JIS technique is applied for identifying time-
varying forces on a bridge [17]. The finite element model of the bridge was updated by experimentally obtained modal pa-
rameters. The uncertainty in the estimated forces has been quantified based on the response of the bridge under wind load. A
technique for quantifying uncertainties due to measurement errors and unknown stochastic excitations is presented in
Ref. [18], where JIS is extended for the case of correlation between process and measurement noise. The JIS technique deals
with measurement noise as well as with modeling error. Therefore, it is found to be more suitable for the estimation of
unknown input to the rotor system.
In the previous study [13], we have presented an approach for unbalance identification using Kalman filter and least
square based technique. The technique used Kalman filter and recursive least square method for unbalance force identifi-
cation. The method was verified by numerical simulation where a single disk rotor bearing system was modeled by finite
element method. The results have shown the accuracy of the technique. However, response measurements at different lo-
cations of the shaft were needed, which may not be available always. Further, the measurements were limited to displace-
ment and velocity only.
The present work aims to estimate unbalance parameters in a rotor-bearing system using a model-based approach in-
tegrated with JIS estimation technique. In this approach, force identification technique is used for fault parameter estimation.
Numerical simulations and experimental verification of the proposed technique have been presented. A rigid rotor supported
by rolling element bearings is considered in the present research work. By comparing estimated unbalance force with
mathematical force model, unbalance parameters (amplitude and phase angle) have been estimated using the least-squares
technique. Results are presented for various shaft speeds and different levels of measurement noise. Accurate unbalance
parameters are estimated for different measurement noise levels which verify the robustness of the proposed technique.
Later, experimental results are presented to demonstrate the effectiveness of the proposed technique. Sensitivity analysis is
also performed with errors in model parameters of the rotor system.

2. Formulations

2.1. Rotor system model

A rigid rotor-bearing system with disk unbalance is shown in Fig. 1. It is assumed that the bearings are identical and
isotropic, i.e., the stiffness of each bearing is same in x and y directions. As shown in Fig. 1, a and b are distances of bearings 1

l
b a
funb c
e
t
x
x2 x1
y y2 y1
Fig. 1. Rigid rotor with disk unbalance.
416 A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427

and 2 from the center of gravity of the rotor, c is the distance of disk from the center of gravity of the rotor. The total length of
the rotor, i.e. a þ b is l. The unbalance mass is located at a distance ‘e’ from the center of the disk.
The linear equation of motion at a defined rotational speed is described by:

€ þ Cd q_ þ Kq ¼ f unb
Mq (1)

where M is the matrix containing mass and inertia, Cd is the damping matrix and K is the stiffness matrix. f unb is the un-
balance force vector, q is a vector of nodal displacement and the dot over the vector represents differentiation with respect to
time. Gyroscopic effect is neglected in the present work. Though, the formulation can be easily extended for the case when the
gyroscopic effect is considered.
Usually, the responses are measured at bearing locations, therefore, the equations of motion (Eq. (1)) are obtained for
coordinates shown in Fig. 1, where the mass, stiffness and damping matrices are given as
2 .  . 3
mðb=lÞ2 þ Id l2 0 mab l2  Id l2 0
6 .  . 7
6 .0 . mðb=lÞ2 þ Id l2 0 mab l2  Id l2 7
M¼6 . 7;
4 mab l2  Id l2 0 mða=lÞ2 þ Id l2 0 5
 . .
2
0 mab l2  Id l2 0 mða=lÞ þ Id l2

2 3 2 1 3
k1 0 0 0 cxx 0 0 0
6 xx 7 6 0
6 0 k1yy 0 0 7 c1yy 0 0 7
K¼6 7; Cd ¼ 6
6
7
7;
6 0 k2xx 0 7 4 0 0 c2xx 0 5
4 0 5
0 0 0 2
kyy 0 0 0 c2yy

8 9 8 9
>
> uu2 ðb  cÞcosðut þ fÞ >
> > x >
>
< 2 >
= < 1>
> =
uu ðb  cÞsinðut þ fÞ y1
f unb ¼ ; q ¼
>
>
> uu ða þ cÞcosðut þ fÞ >
2
>
>
> x >
: 2>
> ;
: 2 ;
uu ða þ cÞsinðut þ fÞ y2

where u is the amount of unbalance (i.e., me), f is the phase and u is the rotor speed. Here, m is the mass of the rotor and Id is
the transverse mass moment of inertia of the rotor. The elements in matrix K and Cd are the stiffnesses and damping co-
efficients of the bearings 1 and 2.

2.2. State-space modeling

In state-space modeling, the state variables of a second order system with n degrees-of-freedom are represented by state
vector of size 2n  1.
The continuous time state-space form can be written as,

_
XðtÞ ¼ Ac XðtÞ þ Bc FðtÞ (2)

YðtÞ ¼ Gc XðtÞ þ Jc FðtÞ (3)


 
0nn Inn
where XðtÞ ¼ f x1 y1 x2 y2 x_1 y_ 1 x_2 y_ 2 g is the state vector, Ac ¼ 1 1 is the system matrix, and
M K M Cd
 
0nn
Bc ¼ is input influence matrix.
M1
The output (or measurement) vector YðtÞ in general, is the combination of displacement, velocity and acceleration
€ þ Sv XðtÞ
response. Thus, output vector can be represented as YðtÞ ¼ Sa XðtÞ _ þ Sd XðtÞ, where Sa , Sv and Sd are the selection
 
matrices for acceleration, velocity and displacement response, respectively. The matrix Gc ¼ Sd  Sd M1 K Sv  Sa M1 Cd
is the state output matrix and Jc is the direct transmission matrix.

2.3. Joint input-state estimation technique

The application of JIS estimation technique in structural dynamics has been presented by researchers. A brief literature
review is provided in the introduction section. The potential of this technique for fault diagnosis of rotor systems is exploited
in the present work. For the sake of completeness, JIS algorithm is presented in this section.
A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 417

The output responses are measured at discrete times, therefore, the continuous time state-space equations (Eq. (2) and Eq.
(3)) need to be converted into its discrete time equivalent. By considering measurement noise and modeling error, the
discrete-time linear dynamic system can be represented as,

xðk þ 1Þ ¼ AxðkÞ þ BFðkÞ þ wðkÞ (4)

yðkÞ ¼ GxðkÞ þ JFðkÞ þ vðkÞ (5)

where k is a time index, xðkÞ ¼ XðkDtÞ, FðkÞ ¼ FðkDtÞ and assuming zero-order hold for the inter-sample behavior of the input,
A ¼ eAc Dt , B ¼ ½A  IA1
c Bc , G ¼ Gc and J ¼ Jc . While discretizing the system matrices, we choose a sampling interval (Dt)
such that it is able to capture the dynamics of the systems. In general, the sampling frequency should be higher than twice the
largest eigenfrequency (Nyquist-Shannon sampling theorem), Dt  maxð p .
ui Þ
wðkÞ and vðkÞ are the system and measurement noise vectors which contain uncorrelated white Gaussian random se-
quences with variance Q and R, respectively.
An initial unbiased estimate and its covariance presented in Eq. (6) and Eq. (7), respectively, are assumed to be known.
The JIS technique consists of three-steps namely: estimation of unknown input (Eq (8), Eq (9), Eq. (10) and Eq. (11)),
measurement update (Eq. (12), Eq. (13), Eq. (14) and Eq. (15)), and time update (Eq. (16) and Eq. (17)).
Initialization

^x0 ¼ E½x0  (6)


 i
T
Px0 ¼ E ðx0  ^x0 Þðx0  ^x0 Þ (7)

Estimation of unknown input

e ¼ GPx T
R k k=k1 G þ R (8)

 1 1
e J
Mk ¼ JT R e 1
JT R (9)
k k

 
^f ¼ M y  G^x (10)
k k k k=k1

 1 1
~ J
Ppk ¼ JT R (11)
k

Measurement Update

1
e
Kk ¼ Pxk=k1 GT R (12)
k
 
^ k=k ¼ x
x ^ k=k1  J^f k
^ k=k1 þ Kk yk  Gx (13)
 
Pxk=k ¼ Pxk=k1  Kk R~  JPp JT KT (14)
k k k

 T
Pxp
k
¼ Ppx
k
¼ Kk JPpk (15)

Time Update

x ^ k=k þ B^f k
^ kþ1=k ¼ Ax (16)

" xp
# 
Pxk=k Pk AT þ Q
Pxkþ1=k ¼ ½ A B px p (17)
Pk Pk BT

^ k=k is the state estimate, ^f k is the unbalance force estimate, Pp is the error covariance matrix for input estimate, Pxk=k is
where x k
xp
the error covariance matrix for state estimate and the matrix Pk is the error cross-covariance (between input and state
estimates) matrix. Kk is the gain matrix, R and Q are the measurement and process noise covariance matrices, respectively.
418 A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427

2.4. Disk unbalance

In model-based fault diagnosis, faults are identified by the equivalent loads acting on a healthy system. A static disk
unbalance is considered in the present work. Schematic diagram of an unbalanced disk is shown in Fig. 2.
By considering only the transverse vibrations, the unbalance forces (Eq. (18) and Eq. (19)) at disk location in lateral di-
rections can be expressed by the following equation,

f x ¼ meu2 cosðut þ fÞ (18)

f y ¼ meu2 sinðut þ fÞ (19)

2.5. Least square fitting

A least squares algorithm is used to achieve the best fit between the estimated forces, ^fðtÞ and the model forces fðb;tÞ. The
objective function is expressed as,
X
minb jjfðb; tÞ  ^fðtÞjj22 ¼ minb ðfðb; tðiÞÞ  ^fðiÞÞ
2
(20)
i

Equations (18) and (19) represent unbalance forces at disk locations. However, the estimated forces are equivalent un-
balance forces at bearing locations, which are expressed in Eq. (1). Therefore, ff unb g41 are used as model forces in least-
square fitting.
The fault model forces (corresponding to one bearing location only) are represented by following equations,

f x ðb; tÞ ¼ bx1 ðb  cÞu2 cosðut þ bx2 Þ (21)


 
f y ðb; tÞ ¼ by1 ða þ cÞu2 sin ut þ by2 (22)

The unbalance parameters will be estimated by comparing fault model forces (Eq. (21) and Eq. (22)) with the estimated
forces ^fðtÞ using least-square fitting (Eq. (20)).

3. Numerical simulations

In this section, numerical simulations are presented for the verification of the proposed unbalance parameters estimation
technique. The rigid rotor model as shown in Fig. 1 is used in the present work. Responses are obtained from Eq. (4) and Eq. (5)
for various shaft speeds with known unbalance force. The measurement (or output) vector contains accelerations and

Fig. 2. Disk unbalance.


A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 419

displacements, the details on sensor requirements can be found in Ref. [16]. A white Gaussian random noise is added to the
generated response to simulate real measurements. The noisy measurements can be expressed as, y ¼ ytrue þ ynoise , where
ytrue is obtained from the system model. Random noise sequence can be expressed as, ynoise ¼ gsrk , where g is the noise level,
s is the standard deviation of the exact response and rk is the normal random sequences. All the responses are calculated at
bearings, as mentioned in the previous section that the states are transverse displacements and velocities at bearing locations.
The proposed technique assumes that prior to unbalance force estimation the model parameters, e.g., rotor mass, bearing
stiffness and damping coefficients are known. For unknown system parameters, the parameter identification techniques can
be integrated with the proposed technique. The details of rotor system are presented in Table 1. It is also assumed that the
location of unbalance is known. This assumption holds true for the rotor system because the unbalance force would be located
where the disk is mounted. For the case of multi-disk system, the equivalent unbalance forces can be estimated instead of
actual force. The aim is to calculate the amount and phase of unbalance from the estimated unbalance force.
The state vector as defined in Eq. (4) contains displacements and velocities at bearing locations in transverse directions,
i.e., x and y directions, whereas, the force vector contains equivalent unbalance forces at bearing locations. Therefore, all the
states are being used for unknown force estimation.
^
The initial state values x are assumed zero and the error covariance matrix, P ^ contains a constant value (1010 in
0=1 0=1
the present work) in its diagonal. The value of the initial state is the same as used in the response generation. Therefore, a
small value of error covariance, i.e., 1010 is selected. However, if the initial state value is unknown, a large (say 108 or 106)
can be chosen. With large initial error covariance, filter ignores the initial estimates. The measurement error covariance R and
process noise covariance Q are usually selected based on the small percentage (1% in the present study) of measurements and
states, respectively [15].

3.1. Constant shaft speed

Here, the results are presented for shaft speeds of 1800 RPM, 2400 RPM, and 3000 RPM. The unbalance mass, and its
distance from the center is kept constant which are 5 g and 0.07 m, respectively, thus the amount of unbalance is
3.5  104 kg-m. In all the cases the phase angle is 40 . Simulations are performed for 1%, 5% and 10% of measurement noise
levels.
In Fig. 3, exact and estimated unbalance force is shown for the shaft speed of 1800 RPM. It can be observed that the joint
input-state estimation technique can accurately estimate the unbalance force. Table 2 shows estimates of residual unbalances

Table 1
Details of the rotor-bearing system.

Component Property Value


Rotor Shaft diameter 16 mm
Mass density 7800 kg/m3
Length 362 mm
Mass 1.5 kg
Inner diameter 16 mm

Disk Outer diameter 152 mm


Thickness 16 mm

Bearings Stiffness 2.6  107 N/m


Damping 250 N/m

(a) (b)
Actual unbalance force Actual unbalance force
10 Estimated unbalance force 10 Estimated unbalance force

5 5
Force [N]
Force [N]

0 0

-5 -5

-10 -10
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time [s] Time [s]
Fig. 3. (a) Time history and (b) details of actual and estimated unbalance force obtained for shaft speed 1800 RPM.
420 A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427

Table 2
Estimated residual unbalances for different shaft speeds and measurement noise levels. Actual unbalance is 3.5  104@40 .

Shaft speed (RPM) Noise level (%) Estimated residual unbalance (kg-m@degrees)
1800 5 3.5  104@40
10 3.5  104@40
15 3.5  104@39.97

2400 5 3.5  104@39.99


10 3.5  104@39.96
15 3.49  104@39.93

3000 5 3.49  104@40


10 3.49  104@40.03
15 3.49  104@39.95

for different shaft speeds and measurement noise levels. It can be seen that the estimates of unbalance amplitude and phase
angle are accurate at all levels of measurement noise. Use of least-squares technique for unbalance parameter estimation
further reduces the effect of measurement noise.

3.2. Fluctuation in shaft speed

For balancing, it is usually assumed that the shaft’s angular speed is constant, which may not be true always. Therefore, the
effect of shaft speed variation on estimation results is presented here. The fluctuation in rotor speed may occur due to
torsional vibration or mechanical faults. Consideration of shaft speed fluctuation for balancing is an active area of research
[19].
The time-varying angular speed can be expressed as

uðtÞ ¼ u0 þ A1 u0 cosðu1 tÞ (23)

where u0 is a nominal angular speed, A1 is the relative fluctuating amplitude, and u1 is the fluctuating frequency.
Numerical simulations are performed for this case where in Eq. (23) A1 and u1 are set to 1% and 5% of the nominal rotor
speed 1800 RPM or 188.49 rad/s. Fig. 4 (a) shows variation in shaft’s angular speed. Actual and estimated unbalance forces are
presented in Fig. 4 (b) and 4 (c). It can be observed that the estimated force matches well with the exact one in spite of
variation in rotor speed. However, error in the estimated amount of unbalance is found to be 73% because in least-square
technique it is assumed that the rotor speed is constant (Eq. (6) and Eq. (7)). Error in estimated phase angle is also
observed. In any further analysis, the shaft speed is assumed constant.

4. Experimental verification

The experimental test rig consists of a motor, helical beam coupling, disk and two ball bearings, as shown in Fig. 5. There
are 36 threaded through holes present on the disk (18 holes are located at 50 mm and other 18 at 70 mm from the center of
the disk). The ball bearings (MB, ER-10K) are considered rigid in the present analysis. The rotor shaft is driven by a 0.373 kW
(0.5 hp) AC motor. A variable frequency drive is used to power the motor at different rotational speeds. All the components are
attached to a base plate of size 915  505  12.8 mm and isolated from the table with six rubber blocks.
Vibration signals are measured using accelerometers (two Bruel & Kjaer 4370 and two Bruel & Kjaer 4526) along-with
Bruel & Kjaer PULSE software platform. The vibration signals are sampled at 4096 Hz and recorded for 120 s. Displace-
ments and velocities are obtained by numerical integrations of the measured acceleration.
Table 3 summarizes different unbalance conditions that are considered for the experimental verification of the proposed
technique. In order to create unbalance, bolts of different masses (6.13 g, 9.04 g, 11.91 g, and 14.78 g) are attached at a distance
of 70 mm from disk’s center. Dynamic balancing was performed prior to the experiments. In Test 1 of all the cases, no un-
balance mass is attached.

4.1. Pre-processing of vibration signals

It is noted here that, the presence of unbalance in a rotor causes an increase in synchronous vibration, i.e., 1  level. It is
assumed here that only unbalance fault is present in the system. Therefore, the 1  vibration signal is used for unbalance
parameter estimation.
The vibration signals measured for Test 1 of all the cases are considered as a signal corresponding to healthy states. Re-
sidual vibration is calculated from the healthy and faulty system at same operating conditions. To generate residual signals,
vibration signals are measured at a particular speed for different unbalance conditions. Therefore, the 1  filtered signals may
A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 421

(a)
191

Angular speed [rad/s]


190

189

188

187

186
0 1 2 3 4
Time [s]

(b) (c)
15 15
Actual unbalance force Actual unbalance force
10 Estimated unbalance force 10 Estimated unbalance force

5
Force [N]
Force [N]

5
0
0
-5
-5
-10
-10
0 0.2 0.4 0.6 0.8 1 0.71 0.72 0.73 0.74 0.75 0.76
Time [s] Time [s]
Fig. 4. (a) Variation in shaft speed (b) time history and (c) details of actual and estimated unbalance force. (Measurement noise 5%).

(a)
AC Motor
Accelerometers
VFD

Coupling
Tachometer Disk

(b)

Fig. 5. (a) Rotor-bearing test rig (b) disk.


422 A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427

40 Tacho pulse
signal

Displacement [ m]
20

-20

0.9 1 1.1 1.2


Time [s]
Fig. 6. 1  filtered displacement signal with tachometer signal for case 3-Test 4.

Table 3
Cases for experimental verification of unbalance parameter estimation technique.

Cases Tests Shaft speed (RPM) Added unbalance mass (g)


Case 1 Test 1 649 Residual
Test 2 6.03
Test 3 9.04
Test 4 11.91
Test 5 14.78

Case 2 Test 1 1130 Residual


Test 2 6.03
Test 3 9.04
Test 4 11.91
Test 5 14.78

Case 3 Test 1 1367 Residual


Test 2 6.03
Test 3 9.04
Test 4 11.91
Test 5 14.78

be out of phase. Fig. 6 represents a displacement and tacho pulse signal. In Fig. 7, signals for healthy and faulty cases are shown
for with and without phase difference. Phase shifts can be avoided by recording a trigger signal during measurement. In the
present case, tacho signals are used to remove any phase shift. Alternatively, the phase shift can also be calculated using cross-
correlation [20].
In Fig. 8, time histories of calculated residual displacements are shown for case 1 after phase-shifting. It can be seen that,
with an increase in the amount of unbalance, the amplitude of residual displacements also increases.

4.2. Results and discussions

For successful implementation of the present technique, knowledge of system parameters is required. Usually, stiffness
and damping properties are determined using experimental methods. For the test rig under consideration, damping co-
efficients of bearings are evaluated experimentally, which is 250 N/m [6]. A frequency domain technique presented in Ref. [21]
for parameter identification, is employed here to estimate bearing stiffness. This technique uses unbalance responses for the
estimation of unknown parameters. Here, different radial stiffnesses are assumed in different radial directions. The following
are the identified average stiffnesses, for bearing 1: k1xx ¼ 1:5  105 N/m, k1yy ¼ 2:78  105 N/m and for bearing 2: k2xx ¼ 7:04 
105 N/m, k2yy ¼ 6:05  105 N/m. The unbalance responses from all the configurations given in Table 3 are used for stiffness
identification.
In all the cases, measurement noise covariance is calculated by the following formula
R ¼ ð0:01  maxÞðyÞ2 , where y is the response signal. The value of process noise covariance Q is selected as 1010  I8 . In
the present approach, unbalance forces are estimated prior to unbalance parameter estimation. Fig. 9 shows actual and
A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 423

(a)
40 td Case 3 - Test 1
Case 3 - Test 4

Displacement [ m]
20

-20

1.6 1.8 2
Time [s]
(b)

40 Case 3 - Test 1
Case 3 - Test 4
Displacement [ m]

20

-20

-40
1.6 1.8 2
Time [s]
Fig. 7. Details of the displacement signals of case 3 and Test 1, 4 (a) before and (b) after phase-shift. (td is the time-delay).

estimated unbalance forces for two different cases. It can be observed that the estimated forces match well with the actual
forces.
The actual and estimated unbalance parameters are presented in Table 4 for shaft speeds of 649 RPM, 1130 RPM, and 1367
RPM. The amplitude of unbalance is quite accurate while phase angles are less accurate for some cases. The fluctuation in
speed or the presence of harmonics is the possible reason for the error in estimated phase angle. Furthermore, the frequency
component other than synchronous may be present even after filtering and cause modulation. Time history of the measured
acceleration for case 1-test 5 is shown in Fig. 10 which reveals the presence of frequency components other than synchronous.

4.3. Sensitivity analysis

The model parameters may not be accurate and may provide wrong estimations. In the previous section, we presented
results for the case where estimated stiffnesses are used. However, it is important to observe the effect of inaccurate system
parameters. In this section, estimation results are presented with the perturbation in stiffness value, which is the most
uncertain parameter. The errors were introduced in estimated stiffnesses prior to unbalance parameter estimation.
Table 5 contains estimated unbalances for different values of percentage increase in stiffness viz., 5%, 10%, and 15%. It can be
easily observed that the error in estimated phase angle is very less when compared with the estimates given in Table 4. The
estimated amount of unbalance (amplitude) is increased with the percentage increase in stiffness. The maximum error in the
estimated amount of unbalance is 23% with respect to the initial estimate. In Table 6, errors in estimated unbalance amplitude
are presented where estimates given in Table 4 are considered as reference values.
424 A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427

(a)
60
Case 1 - Test 2
Case 1 - Test 3

Displacement [ m]
40 Case 1 - Test 4
Case 1 - Test 5
20

-20

0 1 2 3 4
Time [s]

(b)

Case 1 - Test 2
40 Case 1 - Test 3
Displacement [ m]

Case 1 - Test 4
Case 1 - Test 5
20

-20

0.8 1 1.2
Time [s]
Fig. 8. (a) Time history and (b) details of residual displacements at bearing 1 in horizontal direction for case 1.

5. Conclusions

In this paper, unbalance identification technique based on a joint input-state estimation algorithm, has been proposed. The
present technique has been verified by numerical simulations and experimental measurements. The robustness of the
proposed technique is tested by incorporating measurement errors in the numerical study. Results are presented for different
shaft speeds and measurement noise levels. Experimental results have been presented for different unbalance configurations.
Sensitivity analysis is also performed to observe the effect of changes in system parameters. It is found that the estimated
amount of unbalance is close to the actual value. However, some deviations are observed in the estimated phase angles for
some cases.
In the present work, rigid rotor model is used, which may not capture the dynamics of some real-world machines, e.g.,
high-speed rotors. However, these rotating systems can be modeled by other methods, e.g., finite element modeling and can
A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 425

(a) 2
Actual unbalance force
Estimated unbalance force
1

Force [N]
0

-1

0 0.5 1 1.5 2
Time [s]

(b)

3
Actual unbalance force
2 Estimated unbalance force
Force [N]

-1

-2
0 0.5 1 1.5 2
Time [s]
Fig. 9. Time history of actual and estimated unbalance force for (a) case 1, Test 2 and (b) case 1, Test 3.

Table 4
Actual and estimated unbalances for different shaft speeds.

Shaft speed (RPM) Actual unbalance (kg-m@degrees) Estimated unbalance (kg-m@degrees)


649 4.29  104@54 4.21  104@58
6.33  104@54 6.26  104@56
8.33  104@54 8.92  104@71
1.00  103@54 1.10  103@78

1130 4.29  104@54 4.40  104@50


6.33  104@54 6.40  104@60
8.33  104@54 8.52  104@65
1.00  103@54 1.05  103@62

1367 4.29  104@54 4.57  104@41


6.33  104@54 6.62  104@62
8.33  104@54 8.02  104@52
1.00  103@54 1.07  103@55
426 A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427

0.1

Acceleration [m/s2]
0

-0.1

0 1 2 3 4
Time [s]
Fig. 10. Time history of the acceleration signal for case 1, Test 5.

Table 5
Estimated parameters of unbalance, with perturbation in stiffness.

Shaft speed (RPM) Actual unbalance (kg-m@degrees) Estimated unbalance (kg-m@degrees)

Percentage change in stiffness

5 10 15
649 4.29  104@54 4.62  104@58.9 4.83  104@59 5.05  104@59.5
6.33  104@54 7.04  104@56.7 7.37  104@57 7.70  104@57.2
8.33  104@54 9.05  104@71.6 9.48  104@71.9 9.91  104@72.1
1.00  103@54 1.10  103@78.68 1.15  103@78.9 1.20  103@79.2

1130 4.29  104@54 4.80  104@51 5.1  104@51.5 5.31  104@52


6.33  104@54 6.90  104@60.9 7.10  104@61.5 7.4  104@62.1
8.33  104@54 9.10  104@65.8 9.35  104@66.1 9.60  104@66.7
1.00  103@54 1.08  103@62.7 1.16  103@63.2 1.18  103@64.1

1367 4.29  104@54 4.68  104@41.74 4.90  104@41.90 5.12  104@42.05


6.33  104@54 6.69  104@62.24 7.01  104@62.4 7.32  104@62.54
8.33  104@54 8.75  104@52.59 9.17  104@52.73 9.58  104@52.86
1.00  103@54 1.09  103@55.59 1.14  103@55.74 1.19  103@55.87

Table 6
Percentage error in estimated amount of unbalance, with perturbation in stiffness.

Shaft speed (RPM) Estimated unbalance (reference) (kg-m@degrees) % Error

Percentage change in stiffness

5 10 15
649 4.21  104@58 9.74 14.73 19.95
6.26  104@56 12.46 17.73 23
8.92  104@71 1.45 6.27 11.09
1.10  103@78 0.9 4.54 9.09

1130 4.40  104@50 9.09 15.9 20.68


6.40  104@60 7.81 10.93 15.62
8.52  104@65 6.8 9.74 12.67
1.05  103@62 2.85 10.47 12.38

1367 4.57  104@41 2.4 7.22 12.03


6.62  104@62 1.05 5.89 10.57
8.02  104@52 9.1 14.33 19.45
1.07  103@55 1.86 6.54 11.21

be easily used in the present technique. Furthermore, only one fault (disk unbalance) is considered in the present work.
Identification of multiple faults using present approach remains as future work.
A. Shrivastava, A.R. Mohanty / Journal of Sound and Vibration 442 (2019) 414e427 427

References

[1] A.W. Lees, Recent advances and prospects in condition monitoring, in: Vibration Engineering and Technology of Machinery, Springer, 2015, pp. 51e63.
[2] H. Bach, R. Markert, Determination of the fault position in rotors for the example of a transverse crack, Struct. Health Monit. (1998) 325e335.
[3] R. Markert, R. Platz, M. Seidler, Model based fault identification in rotor systems by least squares fitting, Int. J. Rotating Mach. 7 (2001) 311e321.
[4] N. Bachschmid, P. Pennacchi, A. Vania, Identification of multiple faults in rotor systems, J. Sound Vib. 254 (2002) 327e366.
[5] J. Jain, T. Kundra, Model based online diagnosis of unbalance and transverse fatigue crack in rotor systems, Mech. Res. Commun. 31 (5) (2004)
557e568.
[6] A.K. Jalan, A.R. Mohanty, Model based fault diagnosis of a rotor-bearing system for misalignment and unbalance under steady-state condition, J. Sound
Vib. 327 (2009) 604e622.
[7] S. Singh, R. Tiwari, Model based identification of crack and bearing dynamic parameters in flexible rotor systems supported with an auxiliary active
magnetic bearing, Mech. Mach. Theor. 122 (2018) 292e307.
[8] A.R. Mohanty, Machinery Condition Monitoring: Principles and Practices, CRC Press, 2014.
[9] S. Seibold, K. Weinert, A time domain method for the localization of cracks in rotors, J. Sound Vib. 195 (1996) 57e73.
[10] M.H. Lee, T.C. Chen, Intelligent fuzzy weighted input estimation method for the forces generated by an operating rotating machine, Measurement 44
(2011) 917e926.
[11] B.A. Miller, A.S. Howard, Identifying bearing rotor-dynamic coefficients using an extended Kalman filter, Tribol. Trans. 52 (2009) 671e679.
[12] S. Khanam, J.K. Dutt, N. Tandon, Extracting rolling element bearing faults from noisy vibration signal using Kalman filter, J. Vib. Acoust. 136 (2014),
031008.
[13] A. Shrivastava, A.R. Mohanty, Estimation of single plane unbalance parameters of a rotor-bearing system using kalman filtering based force estimation
technique, J. Sound Vib. 418 (2018) 184e199.
[14] S. Gillijns, B. De Moor, Unbiased minimum-variance input and state estimation for linear discrete-time systems with direct feedthrough, Automatica
43 (5) (2007) 934e937.
[15] E. Lourens, C. Papadimitriou, S. Gillijns, E. Reynders, G. De Roeck, G. Lom- baert, Joint input-response estimation for structural systems based on
reduced- order models and vibration data from a limited number of sensors, Mech. Syst. Signal Process. 29 (2012) 310e327.
[16] K. Maes, E. Lourens, K. Van Nimmen, E. Reynders, G. De Roeck, G. Lombaert, Design of sensor networks for instantaneous inversion of modally reduced
order models in structural dynamics, Mech. Syst. Signal Process. 52 (2015) 628e644.
[17] K. Maes, K. Van Nimmen, E. Lourens, A. Rezayat, P. Guillaume, G. De Roeck, G. Lombaert, Verification of joint input-state estimation for force iden-
tification by means of in situ measurements on a footbridge, Mech. Syst. Signal Process. 75 (2016) 245e260.
[18] K. Maes, A. Smyth, G. De Roeck, G. Lombaert, Joint input-state estimation in structural dynamics, Mech. Syst. Signal Process. 70 (2016) 445e466.
[19] H. Cao, D. He, S. Xi, X. Chen, Vibration signal correction of unbalanced rotor due to angular speed fluctuation, Mech. Syst. Signal Process. 107 (2018)
202e220.
[20] H. Herlufsen, Dual Channel FFT Analysis (part I), Brüel & Kjær Technical Review (199e251).
[21] R. Tiwari, V. Chakravarthy, Simultaneous estimation of the residual unbalance and bearing dynamic parameters from the experimental data in a rotor-
bearing system, Mech. Mach. Theor. 44 (2009) 792e812.

You might also like