bulletin of the jsme vol.3, no.1, 2016 mechanical
TRANSCRIPT
J-STAGE Advance Publication date: 21 January, 2016Paper No.15-00589
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
Vol.3, No.1, 2016Bulletin of the JSME
Mechanical Engineering Journal
0123456789
Abstract
Operational transfer path analysis (OTPA) calculates contributions of reference points to response point
vibration by using only operational data. Through OTPA, effective interior noise and vibration reduction are
achieved by applying intensive countermeasure to the high contributing part. However, it becomes difficult
occasionally when many reference points have similar contributions by a vibration mode. In this case,
obtaining high contributing vibration mode and considering how to reduce the mode become important
information. In this study, we attempted to calculate the vibration mode contribution by modifying OTPA.
Principal component calculated in OTPA procedure is composed of correlated vibration factors among
reference points. We then considered the relationship between the principal component and the vibration mode,
and associated the principal components with the vibration modes of a test structure. As a result, high
contributing vibration modes to the response point could be found. In addition, information about which side of
the structure (response or reference side) had better to be measured intensively was also obtained by evaluating
the influence of each principal component to the response point (principal component transfer function).
Finally, Several countermeasures were applied to the structure considering the principal component and
vibration mode contributions. The result shows effective vibration reduction at the response point could be
carried out. Through these procedures, the modified OTPA became more useful tool for applying effective
countermeasure.
Key words : Transfer path analysis, Principal component, Vibration mode, Transfer function, Contribution
1. Introduction
Obtaining contribution of each vibration source to the response point quantitatively is essential to achieve effective
countermeasure. Transfer path analysis (TPA) has been developed to obtain the contribution and several TPA methods
were proposed until now (Van der Auweraer, 1979; Starkey, 1989; Plunt, 1998; Noumura, 2006; Brandl, 2008;
(OTPA) is one of the TPA methods recently developed (Noumura and Yoshida, 2006). This method informs us the
contributions with smaller man-hour using only operational data for calculating the contributions (Noumura, 2006;
calculated using the operational data and the contribution is obtained by multiplying measured reference signal at the
operational condition with the transfer function (Noumura and Yoshida, 2006). The countermeasure for the reduction of
the response point vibration has been applied typically to the high contributing reference point such as engine mount
attachment point. However, if many reference point vibrations correlate each other strongly by putting the reference
points to close points on a flame, the contributions are occasionally very similar and finding out high contributing part
and applying intensive countermeasure becomes difficult. In this situation, the contribution separation of the reference
point by OTPA becomes ineffective. However, if high contributing vibration mode of the structure could be clarified
instead of high contributing reference point, this information has a possibility to become important information for the
Contribution analysis of vibration mode utilizing operational TPA
Junji YOSHIDA* and Koki TANAKA* *Osaka Institute of Technology
5-16-1 Ohmiya, Asahi-ku, Osaka 535-8585, Japan
E-mail: [email protected]
1
Received 26 October 2015
Lohrmann, 2008; Sonehara, 2009; Tcherniak, 2009; Klerk, 2010; Yoshida, 2013; Yoshida, 2014). Operational TPA
Lohrmann, 2008; Sonehara, 2009; Klerk, 2010; Yoshida, 2013; Yoshida, 2014). In this method, the transfer function is
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]2
effective countermeasure.
Meanwhile, principal component mode has been discussed recently as a CAE technique to extract important
vibration modes excited at the operational condition from many vibration modes (Mochizuki, 2015; Kawai and
Yanagase, 2015). This enables us to focus less vibration modes as the reduction target. If we can find out which the
excited vibration mode at the operational condition affects the response signal significantly, the countermeasure
becomes more effective.
In this study, we propose an experimental analysis method of contribution separation using the principal
component through the modification of the present OTPA method instead of obtaining the reference point
contributions. Then, we tried to extract high contributing vibration mode to the response point by associating the
principal component to the vibration mode after consideration of the relationship between principal component mode
and vibration mode. In addition, we considered an evaluation procedure to determine whether the vibration mode at
around the reference points or the resonance at the response point is the main factor of the large vibration peak at the
response point by comparing the characteristics of principal component and the transfer function to make OTPA
method more effective tool.
2. Calculation procedure of OTPA
In OTPA, the individual contributions are calculated through two-step processes. In the first step, the transfer
function from the reference point to the response point is calculated using various operational data measured at the
reference and response points. In the second step, each contribution is obtained by multiplying the reference point
signal by the calculated transfer function. The transfer function in this method is calculated using a principal
component regression method (Noumura and Yoshida, 2006). The calculation procedure is shown in Fig. 1.
Fig. 1 Calculation image of transfer function of OTPA using principal component regression method.
1) At first, principal component analysis (PCA) is applied to the reference signal matrix [Ain] by the singular
value decomposition (SVD) to remove correlation among reference signals as shown in Eq. (1) and (2). The calculated
uncorrelated signals were named as principal component [T].
Tin VSUA (1)
SUVAT in (2)
2) Noise component in the uncorrelated signal was eliminated using the size of each singular value (Noumura
and Yoshida, 2006).
3) Multiple regression analysis is applied between the remained (signal) principal component [T] and the
response signal [Aout] to obtain the influence [B] of each principal component to the response signal as shown in Eq. (3)
and (4).
BTAout (3)
out
TTATTTB
1
(4)
Fig. 1 Operational TPA using principal component regression.
Principal component analysis
T1Ain1
Ain2
Ain3
Ain4
Ain5
T2
T3
T4
T5
Aout
Referencesignal
Principalcomponent
Siz
e o
f p
rin
cip
al
com
po
nen
t
Responsesignal
Regression analysis
Noisereduction
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]3
4) Transfer function from the reference signal to the response signal [H] is calculated by multiplying the
coefficient matrix [V], that connects the reference signal to the principal component, and the regression coefficient [B],
that connects the principal component and the response signal as shown in Eq.(5).
out
TTATTTVH
1
(5)
In the second step, the contribution from the reference point is obtained by multiplying the transfer function with
the reference signal.
This is the outline to obtain contributions of reference points. Countermeasure to reduce the response vibration is
typically applied to the high contributing reference point. In this study, we calculated the principal component
contribution as shown in Eq. (3) by multiplying the principal component [T] and the principal component transfer
function [B] instead of the reference point contributions to obtain additional useful information by OTPA.
3. OTPA application to a simple structure
As the experimental structure on which reference point signals are supposed to correlate each other, two flat plates
connected by four rubber bushes were used as shown in Fig. 2. Material of these plates was Aluminum and thickness of
them was the same at 3 mm. The weight of the lower plate (supporting plate) and that of the upper plate (response
plate) were 1240 g and 320 g, respectively. Each edge of the supporting plate was hung by a soft rubber cable. A swept
sinusoid force from 10 to 1000 Hz was given at the input point on the supporting plate by an electrodynamic shaker
(Modalshop K2007E01).
Fig. 2 Two plates structure for application of OTPA and force input and vibration measurement points.
Four reference points (from Ref. 1 to Ref. 4 in Fig. 2) for applying OTPA were set on the rubber bush attachment
points on the supporting plate, and the response point (Res. in Fig. 2) was set on the center of the response plate as
shown in Fig. 2. These vibrations at the reference and response points were measured simultaneously using
accelerometers (PCB C35265) weighting 2 g each, when the shaker was activated in the operational condition. The
contribution from each reference point was calculated according to the OTPA procedure. Figure 3 shows the average
spectrum of each reference point contribution to the response point.
Fig. 3 Contribution of each reference point to the response point.
300 mm
200 mm
Ref.1 Ref.2
Ref.3 Ref.4
P.1 P.2 P.3
P.4 P.5 P.6
P.9
P.10 P.11 P.12
Input
P.7 P.8
Res.
30
50
70
90
110
130
0 100 200 300 400 500
Ref1Ref2Ref3Ref4
Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
Horizontal and vertical axes indicate frequency and the vibration acceleration level, respectively. The result shows
contributions of most reference points had vibration peaks at the same frequency and the contribution levels were also
very similar. The vibrations at the reference point seemed to have high correlation each other and depended on the same
vibration mode in this situation because the reference points were set at very close points on the same supporting plate.
This result indicates that there was no dominant reference point affecting the response point. The obtained contribution
of reference point by OTPA is not significant information for conducting effective countermeasure. We then tried to
separate contributions of principal component to the response point instead of the reference point contribution by
modifying the OTPA procedure.
4. Principal component obtained in OTPA
4.1 Principal component mode at the operational condition
Principal component at the operational condition [T] is obtained by Eq. (1) and (2) in the OTPA procedure. The
principal component is composed of correlated factors in the measured reference signals and the component (column
element in principal component matrix [T]) has no correlation with the other component. In addition, the reference
point signal could be obtained by multiplying the principal component matrix [T] with the inverse of the matrix [V] as
shown in Eq. (6). Here, the inverse matrix of [V] could be expressed as the transpose of matrix [V] because this matrix
is unitary matrix obtained by SVD. In case the number of the reference point is only two as an instance for the
explanation, the matrix equation could be developed as Eq. (7).
Tin VTVTA 1
(6)
222211122111
2222212112221121
2212211112121111
21
2221
1211
vtvtvtvt
vtvtvtvt
vtvtvtvt
aa
aa
aa
nnnnnn
(7)
The left side box by solid line denotes principal component 1 (PC 1) included in reference signal 1 (Ref. 1), and the
right side box by solid line denotes PC 1 in the reference signal 2 (Ref. 2). Each PC 1 element has complex value
(having amplitude and phase information) and composes the reference signal. Accordingly, we can separate the
vibration behavior at the reference signals to each principal component behavior using the information. The separated
component is named as the principal component mode in this study.
4.2 Relationship between principal component mode and vibration mode
The principal component is obtained by extracting correlated factors in the reference signals and each component is
no correlated to the other components by PCA. In addition, measured reference signal at each point is composed of the
superposition of all principal component modes. On the other hand, behavior of a vibration mode in a reference point
synchronizes (having perfect correlation) at the other reference points. Vibration mode also has orthogonality to the
other mode. Hence, each principal component mode has been regarded as the vibration mode in some previous studies
(Yoshida, 2013; Yoshida, 2014). However, there is almost no discussion about the relationship theoretically in detail
between them. In this section, we discuss the relationship.
As described in Eq. (2), the principal component [T] is obtained by multiplying the reference signal matrix [Ain]
with the matrix [V]. The principal component mode could be expressed in the element of the multiplied matrix of [T]
and [V]T as shown in Eq. (8).
222211122111
2222212112221121
2212211112121111
21
2221
1211
vtvtvtvt
vtvtvtvt
vtvtvtvt
aa
aa
aa
nnnnnn
(8)
4
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]5
In the principal component mode, the PC 1 vector included in Ref. 1 {t11v11, t21v11 ,,, tn1v11} (left solid box) has a
perfect correlation with the PC 1 vector in Ref. 2 {t11v21, t21v21 ,,, tn1v21} (right solid box) because the relationship
between PC 1 in Ref. 1 and PC 1 in Ref. 2 is always v11:v12 in each row. The PC 1 vector in Ref. 1{t11v11, t21v11 ,,, tn1v11}
(left solid box) and the PC 2 vector in Ref. 1 {t12v12, t22v12 ,,, tn2v12} (left dotted box) has no correlation each other
vector {t12, t22 ,,, tn2}, respectively, which are no correlated by PCA. Therefore, a PC mode has a perfect correlation
with the same PC mode and does not have any correlation with the other PC mode. This is the characteristic of PC
modes composes the reference signals.
On the other hand, the acceleration at the two reference point could be expressed by using eigenmode vector [φ]
and modal coordinate [ξ] as follows.
2221
1211
2121
aa (9)
When the input forces are continuously applied to the system, the above relationship is developed as Eq. (10)
according to the FFT repetition.
2221
1211
21
2221
1211
21
2221
1211
nnnn aa
aa
aa
(10)
The above equation could be developed as follows.
222121212111
2222122121221121
2212121121121111
21
2212
2111
nnnnnn aa
aa
aa
(11)
The solid boxes and the dotted boxes are the first and second vibration modes excited in the operational condition,
respectively. Through the comparison of Eq. (8) and (11), we can see both PC modes and vibration modes consist of
reference signal. The reference point vibration is composed of the superposition of PC modes or vibration modes.
In the upper Eq. (11), if the 1st vibration mode vector in Ref. 1 {ξ11φ11, ξ21φ11,,, ξn1φ11} (left solid box) and the same
mode in Ref. 2 {ξ11φ12, ξ21φ12,,, ξn1φ12} (right solid box) have a perfect correlation, and the 1st vibration mode in Ref. 1
{ξ11φ11, ξ21φ11,,, ξn1φ11} (left solid box) does not have any correlation with the 2nd mode in Ref. 1 {ξ12φ21, ξ22φ21,,,
ξn1φ21} (left dotted box), these relationships are exactly same as the PC modes. And then, the PC modes calculated in
the procedure of OTPA are regarded as the vibration modes.
About the relationship among the vibration modes at the operational condition, the 1st vibration mode in Ref. 1
(left solid box) apparently has a perfect correlation with the 1st mode in Ref. 2 (right solid box) because the
relationship among them at each row is always φ11 : φ12. The relationship between the 1st mode in Ref. 1 (left solid
box) and the 2nd mode in Ref. 1 (left dotted box) depends on the excitation condition to the two vibration modes. In
other words, the relationship depends on the relationship between the 1st vibration mode vector {ξ11, ξ21,,, ξn1} and the
2nd vibration mode vector {ξ12, ξ22,,, ξn2} under the operational condition. Because the 1st mode in Ref. 1 is the result
of the multiplication of the 1st vibration mode vector with φ11, and the 2nd mode in Ref. 1 is the result of the
multiplication of the 2nd mode with φ21.
However, there is a special case at around the natural frequency of the vibration mode regardless the relationship
among the vibration modes. If the frequency of the input force closes at the natural frequency of the 1st vibration mode
in the operational condition, the amplitude of the 1st mode becomes much larger than the 2nd mode. In this situation,
the 1st vibration mode becomes dominant at both reference points as shown in Eq. (12).
because these vectors are just the result of the multiplication of v11 and v21 with PC 1 vector {t11, t21 ,,, tn1} and PC 2
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]6
121111
12211121
12111111
222121212111
2222122121221121
2212121121121111
21
2212
2111
nnnnnnnn aa
aa
aa
(12)
In the above situation, vibration behavior of the reference signals are determined mostly by the 1st vibration mode
(solid boxes) and each reference point vibrates with high correlation each other by the dominant vibration mode.
On the other hand, PC mode in a reference point also vibrates having perfect correlation with the same PC mode in
the other reference points. And the PC mode does not have any correlation with the other PC modes as described
before. In addition, the PC number is determined according to the size of each PC in PCA procedure, therefore, the
reference point vibration dominated by the 1st vibration mode (may include very small 2nd mode influence) is
calculated as the PC 1 mode in the PCA procedure theoretically in this situation. In this case, PC 2 includes very small
non-correlated component to PC 1 like measurement errors or 2nd vibration mode.
This is the basic theory background in which the PC 1 mode indicates the dominant vibration mode at around the
natural frequency of the vibration mode. The above described special cases are expected to occur at around peak
frequencies of averaged acceleration level in the operational condition as shown in the following image.
Fig. 9 Image of the average spectrum of a reference point vibration at an operational condition.
5. Obtaining principal component and vibration modes, and the relationship of them
5.1 Obtaining principal component mode
In this section, we actually obtained the principal component mode from the measured operational data of the
employed simple two plate structure. We verified the relationship between the calculated principal component mode
and the vibration mode of the structure. To obtain the principal component mode and compare the principal component
mode shape with the vibration mode shape, we increased the number of reference point from four to twelve (P. 1 to P.
12 in Fig. 2). Figure 10 shows the average spectrum of the principal component calculated from the 12 reference point
signals. Here, 12 principal components were calculated theoretically by using 12 reference point signals, and principal
component (PC) 1, 2, 3 and 4, those were relatively larger components, are shown in Fig. 10.
40
60
80
100
120
0 100 200 300 400 500Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]7
Fig. 10 Calculated principal component of PC 1, 2, 3 and 4 at the operational condition.
Seven vibration peaks of the principal component 1 (PC 1) indicated by arrows were observed. This indicates
seven vibration mode are supposed to be excited by the actual input force in this operational condition if the principal
component expresses the vibration mode correctly.
5.2 Obtaining vibration mode
Subsequently, we analyzed vibration modes of the supporting plate experimentally by conducting an impact
measurement test to evaluate the relationship between the principal component mode and the vibration mode. Seventy
seven points on the supporting plate were employed in the test for obtaining vibration modes experimentally. Table 1
(a) shows the extracted vibration modes (from 1st vibration mode to 15th vibration mode) under 550 Hz and (b) shows
the frequencies where PC 1 has the peak (named as 1st PC 1 to 7th PC 1) for the comparison.
Table 1 Obtained vibration modes and principal component 1.
As the result, 15 vibration modes were found on the supporting plate, and some natural frequencies of the vibration
modes were similar with the vibration peak frequencies of the principal component. However, the number of the
vibration peaks of the principal component was less than the number of the vibration modes. This difference is
considered to be generated by the difference of the input force condition. In the test for obtaining the vibration modes,
more vibration modes of the supporting plate were excited by conducting the impact measurement at all 77 points on
the plate. Oppositely, the input force was given at a fixed point and the frequency range of the force was increased
constantly by the shaker in the operational condition. Therefore, only some vibration modes of the supporting plate are
considered to be excited in the operational condition. In other words, excited vibration modes at the operational
condition could be extracted in many vibration modes of the structure utilizing the principal component. This leads to
conduct effective countermeasure by focusing only the excited vibration modes at the operational condition if we can
associate the vibration mode with the principal component 1 that has the largest size among all principal components.
5.3 Relationship between principal component mode and vibration mode
Principal component 1 (PC 1) at the operational condition was found to have vibration peaks at the similar
frequencies with the natural frequency of some vibration modes. Then, we investigated the relationship between the
PC 1 mode and the obtained vibration mode by comparing the mode shapes. Figure 11 (a) and (b) shows the mode
shapes of the PC 1 at 59 Hz and 205 Hz, respectively. The vibration mode shapes at 60 Hz (1st mode) and 225 Hz (6th
mode) are also shown in Fig. 12 (a) and (b). Noting that the difference of the measurement point number between them;
the number of principal component was 12 according to the number of the reference point and the number of the
Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
50
70
90
110
130
150
0 100 200 300 400 500
PC1
PC2
PC3
PC4
(a) Obtained vibration modes (b) Obtained PC1peaks
Vib. Mode 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 15th PC. Mode 1st 2nd 3rd 4th 5th 6th 7th
Freq. (Hz) 60 71 153 171 181 225 268 271 323 374 459 484 502 535 543 Freq. (Hz) 59 162 205 252 348 428 479
Damp. (%) 2.7 4.8 1.5 1.4 2.7 1.9 1.3 1.6 1.1 1.3 1.4 0.6 1.2 1.2 1.5
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
vibration mode was 77.
(a) PC mode at 59 Hz (b) PC mode at 205 Hz
Fig. 11 Principal component mode shapes at 59 and 205 Hz.
(a) Vibration mode at 60 Hz (b) Vibration mode at 225 Hz
Fig. 12 Vibration mode shapes at 60 and 225 Hz.
As shown in these figures, the principal mode shapes were observed to be very similar with the vibration modes
each other. Then, we tried to calculate the relationship quantitatively between the PC 1 mode and the vibration mode by
using information of the amplitude and the phase in each mode shape. The relative amplitude and phase ratio of each
principal component could be obtained by using the row elements in Eq. (8) and the ratio of the vibration mode at the
same reference points with the principal component were also extracted from the row elements in Eq. (11). After then,
the correlation coefficient of the relative amplitude and phase among the reference points between each principal
component mode shape and the vibration mode shape was calculated. Table 2 shows the obtained absolute of the
correlation coefficient.
Table 2 Correlation coefficient between PC 1 mode shape and the vibration mode shape.
As shown in the bold underlined number in the above table, each PC 1 mode excited at the operational condition
was found to have very high correlation with one of the vibration modes. On the other hand, some vibration modes
(2nd, 3rd, 5th, 8th, 9th, 11th, 13th and 15th) did not have high correlation. This reveals that the PC 1 expresses
vibration mode excited at the operational condition, and some vibration modes were not excited at the condition. In
addition, the frequency of the PC 1 modes was generally lower than the natural frequency of the associated vibration
mode. This tendency is considered to be caused by the additional weight by the attached 12 accelerometers and the
shaker on the supporting plate in the operational condition. However, even though the frequency of the peak PC mode
at the operational condition is slightly different from the natural frequency of the vibration mode, we will be able to
find the associated vibration mode using the above mentioned method by searching the mode at wide frequency range
in case the mode shapes between them have consistency.
In addition, we discussed the PC 1 mode could be expected to be almost the same as the dominant vibration mode
No 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 15th
No Freq. 60 71 153 171 181 225 268 271 323 374 459 484 502 535 543
1st 59 0.88 0.13 0.18 0.17 0.02 0.20 0.24 0.16 0.17 0.79 0.19 0.03 0.11 0.17 0.17
2nd 162 0.20 0.25 0.26 0.85 0.29 0.23 0.05 0.07 0.21 0.15 0.19 0.28 0.08 0.12 0.10
3rd 205 0.16 0.09 0.14 0.10 0.15 0.94 0.39 0.50 0.08 0.18 0.08 0.47 0.14 0.12 0.13
4th 252 0.13 0.32 0.08 0.05 0.09 0.37 0.76 0.53 0.06 0.13 0.25 0.28 0.17 0.28 0.24
5th 348 0.81 0.20 0.12 0.10 0.07 0.26 0.16 0.20 0.31 0.84 0.25 0.34 0.06 0.12 0.13
6th 428 0.24 0.09 0.27 0.23 0.16 0.42 0.16 0.21 0.23 0.28 0.50 0.89 0.19 0.20 0.17
7th 479 0.09 0.44 0.42 0.15 0.17 0.19 0.11 0.08 0.21 0.23 0.26 0.53 0.71 0.74 0.26
Vibration mode
PC
1 m
od
e
8
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
at the resonance frequency. Nevertheless, some correlation coefficients between the associated PC 1 and the vibration
modes were not very high (<0.8). One of the reasons is considered to relate how degree the single vibration mode
dominates to the other modes in the resonance frequency band in addition to the influence of the measurement error.
The correlation coefficient between the PC 1 mode at 252 Hz (4th PC 1 mode) and the associated vibration mode at 268
Hz (7th vibration mode) was 0.76. The averaged vibration peak level at 252 Hz at the operational condition (Fig. 10)
was not so high comparing with the other peaks. Hence, the PC 1 mode is considered to be affected (or deteriorated) by
the other vibration modes slightly in addition to the dominant vibration mode (7th vibration mode).
About the 7th PC 1 mode at 479 Hz, the coefficient with the associated vibration mode (14th vibration mode) was
also not so very high at 0.74. The 14th vibration mode was 2nd bending mode along the lateral direction as shown in
Fig. 13 (a) and the associated 7th PC 1 mode did not express the 2nd bending mode well as shown in Fig. 13 (b).
(a) 14th vibration mode at 535 Hz (b) 7th PC 1 mode at 479 Hz
Fig. 13 Associated 14th vibration and 7th PC 1 modes.
Because the number of the reference signals along the lateral direction for the PC mode was only three and
describing the 2nd bending mode accurately was difficult using only the three points. The insufficient number of nodes
for representing the 14th vibration mode is one of the main reasons of the correlation decrease. Similar phenomena
(fault high or low correlation) caused by the difference of the node number were observed in the other combination
such as the combination of the 10th vibration mode at 374 Hz and the 1st PC 1 mode at 59 Hz. Although we will be
able to find out the fault combination by checking the wide frequency difference and the comparing the coefficient, this
phenomenon had better to be taken into account for the application of the proposed method.
About the relationship between the 14th vibration mode and the 7th PC 1 mode, even though the number of the
node in the PC mode was not sufficient for describing the vibration mode shape accurately, the correlation coefficient
of them was over 0.7, which is a criterion value regarded as “high correlation” in general, and the PC 1 mode have
already clarified to express the dominant vibration mode at around the natural frequency. From them, the vibration
modes relating the PC 1 mode were considered to be successfully found out at the operational condition in many modes
using the association method.
6. Contribution of vibration mode and transfer function of principal component
If we can extract not only the excited vibration modes at the operational condition but also obtain the contribution
of the each excited modes to the response point, this information becomes more useful for applying effective
countermeasure to reduce the response point vibration. The contribution of principal component to the response point is
obtained by multiplying the principal component [T] with the principal component transfer function [B] as described in
Eq. (3). Figure 14 shows the calculated contributions of PC 1 to 4 having relatively higher contribution in total 12
principal component contributions.
9
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]10
Fig. 14 Contribution of PC 1, 2, 3 and 4 to the response point vibration.
The contributions of principal component were observed to be separated more clearly than the contribution of
reference point (Fig. 3). In addition, PC 1 contribution was observed to have the largest in all contributions. In the PC 1
contribution, four vibration peaks at 162, 205, 348 and 479 Hz indicated by bold arrows were especially high level. In
contrast, the contribution at 59, 252 and 428 Hz indicated by dotted arrows were low level. The vibration modes
excited at the three frequencies seem to be unimportant for the reduction of response point vibration although the
vibration was excited largely before multiplying the transfer function as shown in Fig. 10. The vibration at these
frequencies is considered to be transferred inefficiently between the supporting and response plates comparing with the
other vibration having large peaks at the response point. The PC 1 modes at 162, 205, 348 and 479 Hz, those were
found to be important components, have already known to associate with 4th, 6th, 10th, and 14th vibration modes.
Hence, these analytical results inform us which vibration modes should be measured intensively among many modes of
the structure for the reduction of the response vibration by utilizing principal component contribution in OTPA.
On the other hand, the contribution of principal component at 505 Hz indicating by triangle in Fig. 14 was found to
have high contribution to the response point although the principal component did not have high vibration peak itself at
the frequency as shown in Fig. 10. This phenomenon is considered to be generated by the transfer characteristic from
supporting plate to the response point. Then, we investigated the calculated transfer function of PC 1 (principal
component transfer function) to the response point as shown in Fig. 15.
Fig. 15 Transfer function of PC 1 to the response point.
As shown in this figure, the transfer function had large peak at the frequency (indicated by triangle). This implies
that resonance seems to have occurred at the response plate at the frequency and this is the factor to increase the
contribution of PC 1. Subsequently, we obtained a point inertance of the response point and the vibration mode of the
response plate by an impact measurement test to obtain the vibration characteristic of the response plate. Figure 16 (a)
and (b) show the point inertance at the response point and the vibration mode of the response plate at 501 Hz,
respectively.
Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
30
50
70
90
110
130
0 100 200 300 400 500
PC1
PC2
PC3
PC4
-100
-80
-60
-40
-20
0
0 100 200 300 400 500
Freq. (Hz)
PC
tra
nsf
er f
unct
ion
(dB
)
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]11
-20
0
20
40
60
80
0 100 200 300 400 500
Freq. (Hz)
Tra
nsf
er f
unct
ion
(dB
)
(a) Point inertance at response point (b) Vibration mode of response plate at 505 Hz
Fig. 16 Point inertance and vibration mode of supportng plate.
From the result, a vibration peak was found at the response plate at around the frequency and the response point
had the highest amplitude at the vibration mode. This result indicates the countermeasure had better to be applied on
the response plate not to the supporting plate because the resonance occurred at the response side. From these analyses,
we could obtain which part should be measured intensively at the input side or the response side in addition to the
important vibration modes to the response point by utilizing principal component in OTPA.
7. Countermeasure utilizing the high contributing principal component
Through the OTPA procedure and evaluation method to associate the PC 1 mode with the dominant vibration
mode, we found out high contributing vibration modes to the response point vibration. In addition, we could determine
which response side (response plate) or reference side (supporting plate) affects largely to the high vibration peak at the
response point by assessing the principal component transfer function. We then tried to reduce the response point
vibration effectively using the analytical results. At first, we conducted a countermeasure to reduce the high vibration
peak of the response point at 162 Hz in two ways. In the frequency band, the PC 1 (2nd PC 1 mode) was calculated to
have highest contribution and 4th vibration mode (Fig. 17) was found to associate with the PC 1 mode. Hence, we
applied the countermeasure considering the vibration mode shape.
Fig. 17 Vibration mode shape of the 4th mode associated with 2nd PC 1 mode.
In this countermeasure, we put six weights (total 150 g) to the separated points on the supporting plate. To verify
the efficiency of the countermeasure considering the vibration mode, the weights were put at the antinode points to
inhibit the mode as shown by open circles in Fig. 18 (a) (Case A). In Case B, the same weights were put at the node
point as shown by open circles in Fig. 18 (b) to compare the vibration reduction effect at the response point. We also
measured the original vibration level at the response point as the basis of the level (Case C).
Response point
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
The acceleration vibration level at the response point in the identical operational condition with the previous test
was obtained in these three conditions. Figure 19 shows the average vibration levels at the response point.
Fig. 19 Average response point vibration level at around 162 Hz and comparison in three conditions. Case A: six weights
were put at the antinodes of the high contributing vibration mode, Case B: the weights were put at the nodes, Case
C: no weights (original condition).
Horizontal and vertical axes indicate the frequency and the average acceleration level at the response point,
respectively. Gray solid, dotted and black solid lines indicate the level in Case A, B and C, respectively. As shown in
the above figure, the vibration levels at around 162 Hz (indicating by the dotted circle) were observed to be reduced by
adding the weights in both cases (Case A and B). However, the decreased level was different. Although the vibration
level was reduced only about 3 dB by adding the weights at node of the mode in Case B, the level was reduced about
8 dB by adding them at antinode points in Case A. This indicates that the vibration level at the response level could be
reduced more efficiency by considering the high contributing vibration mode shape.
In the second example, we tried to reduce the response vibration peak at 206 Hz. In this frequency band, 6th
vibration mode (4th PC mode) as shown in Fig. 20 was found to be the important mode that should be measured for the
reduction of the peak vibration at the response point.
Fig. 20 Vibration mode shape of the 6th mode associated with 4th PC 1 mode.
Then, we added the same six small weights (total 150 g) on the supporting plate in two ways as shown in Fig. 21
and compared the reduction level. In Case A, the six weights were put on the antinode points as indicated by opened
circles in Fig. 21 (a), and the same six weights were put at the node points of the vibration mode as shown in the
50
60
70
80
90
100
110
100 150 200 250
Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
CaseA: Antinode
CaseB: Node
CaseC: No weight (Original)
12
(a) Case A: weights at antinode points (b) Case B: weights at node points
Fig. 18 Weight attachment points for the vibration reduction of the response point at 162 Hz.
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
opened circles in Fig. 21 (b) in Case B. In Case C, the weights were not put to measure the original vibration level at
the response point.
Figure 22 shows the average response point vibration levels at the operational condition in these three conditions.
Fig. 22 Average response point vibration level at around 206 Hz and comparison in three conditions. Case A: six weights
were put at the antinodes of the high contributing vibration mode, Case B: the weights were put at the nodes, Case C:
no weights (original condition).
As shown in the results, the vibration level at around 206 Hz (indicated by dotted circle) was reduced largely
especially by adding the weight at the antinode points (Case A) about 8 dB although the vibration was not reduced
largely by adding them at the node points (Case B). From these verifications of the countermeasure, the
countermeasure considering the high contributing vibration mode obtained by the modified OTPA was found to realize
effective countermeasure.
As the final countermeasure, we tried to reduce the response point vibration at 505 Hz, where the principal
component transfer function had a peak. From the analytical result, the resonance at the response point was considered
to be a main factor. Then, we put a concentrate mass of 150 g to the response point to reduce the resonance vibration
peak at the frequency in Case A. In Case B, the six weights (total 150 g) were put on the supporting plate as same as
Case A in the countermeasure example at 206 Hz as shown in Fig. 21 (a). In Case C, these weights were not put to
measure the original vibration level at the response point. Figure 23 shows the average response point vibration levels
at the operational condition in these three conditions.
50
60
70
80
90
100
110
150 200 250 300Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
CaseA: Antinode
CaseB: Node
CaseC: No weight (Original)
13
(a) Case A: weights at antinode points (b) Case B: weights at node points
Fig. 21 Weight attachment points for the vibration reduction of the response point at 206 Hz.
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
Fig. 23 Average response point vibration level at around 505 Hz and comparison in three conditions. Case A: concentrate
weight was put at the response plate, Case B: the weights were put on the supporting plate at the antinode points of
6th vibration mode, Case C: no weights (original condition).
As shown in the above figure, the vibration level at around 505 Hz (indicated by dotted circle) could be observed to
be reduced very largely at about 20 dB by adding the concentrate weight at the response point. The reduction level was
much larger than Case B where the six weights were added on the supporting plate even though the total additional
weight was the same as 150 g. Consequently, more effective countermeasure was achieved by putting the weight at the
response side when the principal component transfer function had a peak at the target frequency.
From these countermeasure examples, we could verify the effectiveness of the application of the modified OTPA
utilizing principal component and the association method of the high contribution PC 1 mode with the dominating
vibration mode.
5. Summary
In this study, we carried out vibration tests using two plates structure and analyzed the measured signal by OTPA
utilizing the principal component to obtain useful information for effective countermeasure at the response point. In
addition, we discussed the relationship between the principal component mode and the vibration mode theoretically and
derived the PC 1 mode expresses the dominant vibration mode at around the resonance frequency. Subsequently, we
calculated the contribution of the principal component by multiplying the principal component and the transfer function
through the modified OTPA procedure.
As the result, we obtained which part (supporting plate or response plate) affects the vibration peak largely at the
response point by comparing the principal component level and the transfer function at the frequency bands where the
response point vibration was large. In addition, we developed a method to associate the high contribution PC 1 mode
with the dominant vibration mode by evaluating the mode shape correlation. Consequently, high contributing vibration
mode to the response point could be found out. At last, we actually attempted the vibration reduction at the response
point to verify the effectiveness of the proposed analytical procedures in several cases. The results showed that the
vibration at the response point could be reduced effectively by applying intensive countermeasure according to the high
contribution vibration mode shape in case the principal component level was a main factor. In addition, in case the
transfer function characteristic of the principal component made a vibration peak at the response point, the vibration
was successfully reduced more by applying the countermeasure at the response side comparing with the
countermeasure at the reference side on the supporting plate. These countermeasures were very normal method itself
for the vibration reduction. Accordingly, the association method of the PC 1 mode with the vibration mode does not
relate the vibration reduction method directly. However, the effectiveness of the normal countermeasure is increased
largely by combining the countermeasure method with the association method proposed in this study obtained by the
modified OTPA by focusing the reduction target vibration modes in many modes.
As described above, more informative result could be obtained by the modified OTPA in case where reference
signals have high correlation each other. Furthermore, if the vibration mode is obtained through a simulation technique,
we can also expect to extract high contributing vibration modes in the simulated many vibration modes to the response
point vibration by applying the proposed method. In addition, the countermeasure could be considered utilizing the
50
60
70
80
90
100
110
450 500 550 600Freq. (Hz)
Acc
eler
atio
nle
vel
(d
B)
CaseA: Response plate
CaseB: Supporting plate
CaseC: No weight (Original)
14
2
Yoshida and Tanaka, Mechanical Engineering Journal, Vol.3, No.1 (2016)
© 2016 The Japan Society of Mechanical Engineers[DOI: 10.1299/mej.15-00589]
simulation technique effectively by focusing only the important (high contributing) vibration mode. This realizes the
combination analysis between the experimental OTPA and the simulation technique for finding out important vibration
mode and the countermeasure to reduce the response vibration more instantly and effectively.
Acknowledgments
The authors thank Mueller-BBM in Germany and TOYO cooperation in Japan for supporting this work.
References
Brandl, S., Priebsch, H. –H., Brandl, F., Biermayer, W., Houldrich, R. and Sontacchi, A., Challenges and New Solutions
for Transfer Path Analysis, Proceedings of International Conference on Noise and Vibration Engineering 2008
(2008), pp.1-16.
Kawai, H., and Yanase, J, Noise Reduction by Use of the Evaluation Index for Vehicle Structure Eigen Modes, JSAE
Paper 20155047 (2015), pp.268–272 (In Japanese).
Klerk, D. de. and Ossipov, A., Operational Transfer Path Analysis: Theory, Guidelines and Tire Noise Application,
Journal of Mechanical Systems and Signal Processing, Vol.24, No.7(2010), pp.1950–1962.
Lohrmann, M., Operational Transfer Path Analysis: Comparison with Conventional Methods, Proceedings of
International Congress on Sound and Vibration 15 (2008), pp.2671-2676.
Mochizuki , T, Understanding of Major Dynamic Mechanism by the Principal Modal Analysis, JSAE Paper 20155046
(2015), pp.262-267 (In Japanese).
Noumura, K., and Yoshida, J., Method of Transfer Path Analysis for Vehicle Interior Sound with No Excitation
Experiment, FISITA2006 Proceedings F2006D183 (2006), pp.1-10.
Plunt, J., Strategy for Transfer Path Analysis (TPA) applied to Vibro-Acoustic Systems at Medium and High
Frequencies, In Proceedings of the 23rd International Conference on Noise & Vibration Engineering (ISMA)
(1998), pp.139–158.
Sonehara, T., A Study on Meaning of Acceleration-based Transfer Path Analysis, JSAE Paper 100/20095547 (2009),
pp.19–24 (In Japanese).
Starkey, J., and Merril, G., On the Ill-conditioned Nature of Indirect Force Measurement Techniques, International
Journal of Analytical and Experimental Modal Analysis, Vol.4, No.3, (1989), pp.103–108.
Tcherniak, D., and Schuhmacher, A.P., Application of Transmissibility Matrix Method to NVH Source Contribution
Analysis, IMAC 2009 (2009).
Van der Auweraer, H., Wyckaert, K., Hendricx, W., and Van Der Linden, P., Noise and Vibration Transfer Path
Analysis, Lecture series - van Kareman Institute for fluid dynamics, van Kareman Institute, Rhode-Saint-Genèse,
Belgium (1979)
Yoshida, J. and Yamashita, D, Target Level Setting Method for the Reference Signal of Operational TPA, Journal of
System Design and Dynamics Vol.7, No.4 (2013), pp.317–327.
Yoshida, J. and Yamashita, D, Meaning of Principal Component in Operational TPA, JSAE Paper 20145009 (2014),
pp.1–6 (In Japanese).
15