Beruflich Dokumente
Kultur Dokumente
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/228422578
CITATIONS
READS
37
384
2 AUTHORS:
Yi Guo
Robert G. Parker
SEE PROFILE
SEE PROFILE
National Wind Technology Center, National Renewable Energy Laboratory, Mail Stop: 3811, 15013 Denver West Parkway, Golden, CO 80401-3305, USA
Department of Mechanical Engineering, Virgina Tech, USA
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 23 August 2012
Accepted 20 February 2014
Available online 12 March 2014
This paper investigates the dynamics of wind turbine planetary gear sets under the effect of gravity using
a modied harmonic balance method that includes simultaneous excitations. This modied method
along with arc-length continuation and Floquet theory is applied to a lumped-parameter planetary gear
model including gravity, uctuating mesh stiffness, bearing clearance, and nonlinear tooth contact to
obtain the dynamic response of the system. The calculated dynamic responses compare well with time
domain-integrated mathematical models and experimental results. Gravity is a fundamental vibration
source in wind turbine planetary gear sets and plays an important role in the system dynamic response
compared to excitations from tooth meshing alone. Gravity causes nonlinear effects induced by tooth
wedging and bearing-raceway contacts. Tooth wedging, also known as a tight mesh, occurs when a gear
tooth comes into contact on the drive-side and back-side simultaneously and it is a source of planetbearing failures. Clearance in carrier bearings decreases bearing stiffness and signicantly reduces the
lowest resonant frequencies of the translational modes. Gear tooth wedging can be prevented if the
carrier-bearing clearance is less than the tooth backlash.
2014 Elsevier Masson SAS. All rights reserved.
Keywords:
Dynamics
Wind turbine
Planetary gear
1. Introduction
The National Renewable Energy Laboratory (NREL) Gearbox
Reliability Collaborative (GRC) was established by the U.S. Department of Energy in 2006. Its key goal is to understand the root causes
of premature gearbox failures (Musial et al., 2007) through a
combined approach of dynamometer testing, eld testing, and
modeling (Link et al., 2011), resulting in improved wind turbine
gearbox reliability and a reduction in the cost of energy. As a part of
the GRC program, this paper investigates gravity-induced dynamic
behaviors of planetary gear sets in wind turbine drivetrains that
could reduce gearbox life. Planetary gear sets have been used in
wind turbines for decades because of their compact design and
high efciency. Despite these advantages, planetary gear sets
generate considerable noise and vibration. Vibration causing high
dynamic loads may result in gear tooth and bearing failures (Musial
et al., 2007). Fatigue failures are a concern in long life-cycle applications. Analyzing the dynamics of wind turbine planetary gear
* Corresponding author. Tel.: 1 303 384 7187; fax: 1 303 384 6901.
E-mail addresses: yi.guo@nrel.gov, guo.83@buckeyemail.osu.edu (Y. Guo).
http://dx.doi.org/10.1016/j.euromechsol.2014.02.013
0997-7538/ 2014 Elsevier Masson SAS. All rights reserved.
46
Ring
Carrier
Planet
181.6
2633
759.9
104
3.2
144.2
59.1
3.2
21
99
e
39
215.6
1016.4
e
400.4
186.0
1047.7
e
372.9
9
9
ksp 16.9 10 , krp 19.2 10
100
102 106 5 109 6.8 109
3.2 109
0.275
45.8 106 57.4 106 0
0
~ 0 ~f d s; z ~f b s; z ~f s; z F
~ s
~ 00 Cz
Mz
B
m
m
x ~
M ~
C ~d
f dm ~b
f bm
; M ; C
; fm
; fm
;
2
L
M
MLu
MLu
MLu2
~f f B ; F
~ F
B
MLu2
MLu2
(1)
47
h
iT
y
h
x h
x
x
x y
x y
Fg t fcg
; fcg ; 0; frg
; frg ; 0; fsg
; fsg ; 0; f1g
; f1g ; 0; .; fNg
; fNg ; 0
(2)
x m g sinU t
fwg
w
c
y
mw g cosUc t;
fwg
w c; r; s
fjgx mj g sin Uc t jj
h
fjg mj g cos Uc t jj ;
j 1; .; N
(3)
(4)
where the variable Uc um/Nr (if the ring is xed) denotes the
carrier rotation frequency. um denotes the mesh frequency. Nr denotes the number of teeth on the ring.
The mesh stiffness uctuates as the number of teeth in
contact changes and is also an important internal excitation
source of geared systems. These excitations are included through
time-varying mesh stiffnesses. The mesh stiffnesses are calculated using the Calyx program (Nayfeh and Balachandran, 1995).
Its mean amplitudes are listed in Tables 1 and 2 (Appendix). This
program analyzes gear tooth contact and rolling element contact
by using a combined analytical/nite element analysis detailed
in Vijayakar (1991). Its results have been compared against
studies of gear dynamics (Singh, 2010, 2011; Guo and Parker,
2012b).
3.2. Finite element model
The two-dimensional nite element model includes 36 quadrilateral elements per tooth, four nodes per element, and each node
has three degrees of freedom. It uses a combined surface integral
and nite element method to capture tooth deformation and contact loads in geared systems (Kahraman and Blankenship, 1994).
The software developed by Vijayakar (1991) intrinsically evaluates
time-varying tooth contact forces that are specied externally with
conventional simulation tools. Fluctuating mesh stiffness over a
mesh cycle, tooth wedging, and tooth separation are included
intrinsically in the nite element model. The nite element model
also includes clearance nonlinearity at the carrier-ring bearing.
Other bearings are modeled as linear stiffnesses without clearance
in the nite element and analytical models. The nite element
approach has been validated by experiments of geared systems
(Parker et al., 2000a,b; Kahraman and Vijayakar, 2001). It is used to
benchmark the established lumped-parameter model when
experimental data is unavailable.
4. Extended harmonic balance method
The extended harmonic balance method is used to obtain the
dynamic responses of the model in Eq. (1). The formulation includes two excitation sources with excitation frequencies U1 and
U2. The coupling effects between these two excitations are
considered by including their side bands. Other excitation sources
can be considered in a similar way. The response z is expanded
into a Fourier series and assumed to include the R1, R2, R3R4R5, and
R6R7R8 harmonics of excitation frequencies U1 and U2 and their
48
Table 2
System parameters for the 550-kW planetary gearbox (PG-B). The ring mass includes the gearbox housing and parallel stages.
Sun
Mass (kg)
Moment of Inertia (kg-m2)
Number of Teeth
Pitch Diameter (mm)
Root Diameter (mm)
Average Mesh Stiffness (N/m)
Bearing Stiffness (N/m)
Gearbox Trunnion Stiffness (N/m)
Carrier-Bearing Stiffness (N/m)
Carrier-Bearing Clearance (mm)
Torsional Support Stiffness (N/m)
Ring
Carrier
Planet
51
4000
1330
114
61.1
2484
314.7
51.9
16
68
e
26
224
952
e
364
202
980
e
329
9
9
ksp 3.95 10 , krp 5.29 10
100
e
4 109 5.3 109
126 106
3.6 109
1
0
3 106 24.4 106 0
side bands mU1 lU2 and pU1 qU2. Each component zh in z then
has a total of 2(R1 R2 R3R4R5 R6R7R8) 1 terms and is
expressed as
2
6
G6
4
G1 G2
R4 P
R3
P
m1 l1
G3 m;l;R5
R7 P
R6
P
p1 q1
G4 p;q;R8 7
5
n
(7)
R7 X
R6
X
R4
R3
P
P
m1 l1
mU1 lU2 2 A3 z
pU1 qU2 2 A4 z
p1 q1
zh zh;1
R1
X
z0 GU1 B1 z GU2 B2 z G
i1
i1 m1 l1
R7 X
R6
X
pU1 qU2 B4 z
(8)
p1 q1
i1
R5 X
R4 X
R3 n
X
zh;2ai;m;l cos imU1 lU2 s
mU1 lU2 B3 z
m1 l1
R2 h
i
X
R4 X
R3
X
(5)
R8 X
R7 X
R6 n
X
zh;2bi;p;q cos ipU1 qU2 s
i1 p1 q1
o
zh;2bi;p;q1 sin ipU1 qU2 s
~f d Gf d ;
m
m
uN U2
uN U1 ;U2
~f Gf ;
B
B
~ GF
F
(9)
By substituting Eqs. (8) and (9) into the equations of motion, Eq.
(1) yields
3T
6
7
6
7
xc U1
xc U2
xc U1 ;U2
7
z 6
6 z3N3;1 ; .; z3N3;2L 1 ; .; z3N3;2L 1 ; .; z3N3;2L 1 ; .; z3N3;L 7
1
2
3
4
4
5
|{z} |{z} |{z}
uN U1
~f b Gf b ;
m
m
(6)
82
R4 X
R3
<
X
2
G 4 U21 MA1 U22 MA2
mU1 lU2 MA3
:
m1 l1
R7 X
R6
X
p1 q1
R4 X
R3
X
(10)
m1 l1
R7 X
R6
X
fB
d
fm
b
fm
F
K U1 MA1 U2 MA2
2
9
=
Lm;b
r1;i
Lm;b
r2;1
R4 X
R3
X
m1 l1
2 P Lb
c1;i
Lc;B
6
6
6
6
6
vFB
G6
H
6
6
vz
6
6
6
4 Symmetric
Lc;B
Lbr Lc;B
Lbs
Lbc2;1 . Lbc2;N 3
7
7
.
7
7
7
.
7
7
7
0 7
Lbp;1 .
7
7
5
1
Lbp;N
R7 X
R6
X
(17)
pU1 qU2 CB4
p1 q1
KLTI
(11)
Eq. (10) then becomes
b
Kz F f B f m f m
(12)
d
b
v Kz f B f m f m F
(13)
vz
Smoothing functions are employed to approximate the piecewise nonlinear forces so that the explicit formulation of J can be
derived. These nonlinear forces include bearing reaction forces and
tooth loads on the drive-side and back-side of the gears. The inverse
Fourier transformation operator H is dened such that z Hz,
which maps from the time domain to the frequency domain.
2
6
H6
4
n
H1 H2
R4 P
R3
P
m1 l1
H3 m; l; R5
R7 P
R6
P
p1 q1
7
7
5
H4 p; q; R8
n
(14)
where H1eH4 are the inverse Fourier transformation operators
associated with G1eG4, respectively.
The explicit formation of J is obtained as
J KH
d
vf m
3
7
Lm;b
r2;n 7
7
m;b 7
Lm;b
Lm;b
s2;1 / Ls2;n 7
s1;i
7
7
7
Lm;b
7
p;1
7
7
5
1
p1 q1
R4 X
R3
X
(16)
m1 l1
6
6
6
6
6
vf m
G6
H
6
6
vz
6
6
6
4 Symmetric
Lm;b
p;n
R7 X
R6
X
p1 q1
49
d
Hvf m =vzG
b
vf m
vf B
GH
GH
G
vz
vz
vz
(15)
Dz0
Dz00
0
Jt K
I
C
Dz
Dz0
(18)
50
Fig. 2. (a) The lumped-parameter model and (b) side view of the planetary gear set.
Fig. 3. A 750-kW wind turbine gearbox of the NREL GRC during a dynamometer test.
Fig. 4. Carrier radial displacement with respect to the ring calculated using the
extended harmonic balance method (red e) and measured by the GRC project (/). (For
interpretation of the references to colour in this gure legend, the reader is referred to
the web version of this article.)
method largely match those of the experimental data. The planetbearing force for each individual planet bearing over a carrier cycle
is asymmetric because of different pin position errors and bearing
clearances for each planet. The lumped-parameter model predicts a
1.4% difference between the peak amplitudes of the planet-bearing
forces over a carrier cycle because of these errors and clearances.
The measured bearing force of planet 2 deviates from the other two
planets, which is caused by a signal offset error.
5.2. Comparison to computational methods
Dynamic responses computed by the extended harmonic balance method are compared against nite element and numerical
integration analyses for PG-B. In lightly damped systems, nite
element analysis and numerical integration require many time
steps for the transient response to diminish so that steady-state
data can be obtained. The extended harmonic balance method
avoids these long duration integration simulations by calculating
the steady-state dynamic response in the frequency domain. Numerical integration and nite element analysis compute only stable
51
Fig. 6. Measured (/) and calculated (red e) planet-bearing forces over a carrier cycle.
(For interpretation of the references to colour in this gure legend, the reader is
referred to the web version of this article.)
responses. Conversely, the harmonic balance method uses arclength continuation to nd unstable responses by tracing solutions as the speed changes quasi-statically.
The large ring gear mass results from a typical arrangement of
wind turbines whereby much of the gearbox mass is rigidly connected to the ring, which is supported on a relatively compliant
foundation. The input torque is applied to the carrier, and the sun
gear is the output. Gravity acts at the center of mass of all of the
component gears. For the ring and sun, it is a periodically varying
external excitation in the carrier reference frame in which the
model is formulated. One carrier period is analyzed using the nite
element model, giving 68 mesh cycles. Each mesh cycle is divided
evenly into 10 intervals. The nite element solutions have single
precision accuracy while the convergence tolerance is
100 1012 mm for analytical solutions. A global Newton iteration
scheme with line search technique (Baker and Overman, 2000) is
used to obtain accurate numerical solutions of the analytical model.
The line search technique makes the predicted solution at each
iteration always closer to the exact solution by adjusting the iteration step size for each step. For the numerical integration, 120
mesh cycles were simulated at each speed, of which 20 cycles are
considered as transient time. Each mesh cycle is discretized into 50
intervals. For the harmonic balance method, R1 R2 3, R3 1,
R4 4. Each carrier period is discretized into 5291 equally spaced
intervals. The solutions of the harmonic balance method have the
convergence tolerance of 1 106 mm.
The natural frequencies of PG-B below 400 Hz without bearing
clearance and nonlinear tooth contact are listed in Table 3. Modal
analysis is performed numerically on the nite element model of
PG-B by applying torque and force impulses at the carrier and sun.
Numerical impulse tests provide natural frequencies and their mode
shapes as described in Table 3. The natural frequencies predicted by
Table 3
Natural frequencies of PG-B.
Fig. 5. Frequency spectrum of the calculated (red e) and measured (/) carrier
displacement. (For interpretation of the references to colour in this gure legend, the
reader is referred to the web version of this article.)
Natural Frequency
Lumped-parameter
(Hz)
Damping
(%)
40.75
69.49
157.53
301.46
40.94
69.1
163.7
312.1
5.48
5.26
7.43
5.19
0.35
0.56
3.77
3.41
Translational
Rotational
Translational
Rotational
52
Fig. 7. Root-mean-square (RMS, mean removed) dynamic response of PG-B for various mesh frequencies calculated by the extended harmonic balance method (red C) and
numerical integration (/). Unstable solutions calculated by the harmonic balance method are denoted by (red B). (For interpretation of the references to colour in this gure
legend, the reader is referred to the web version of this article.)
Fig. 8. (a) Dynamic carrier-bearing force of PG-B calculated using the harmonic balance (red e) and numerical integration methods (e), (b) percentage of bearing contact in a carrier
cycle of PG-B with various speeds, given Dc 1 mm. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
53
Fig. 9. (a) RMS dynamic response of PG-B carrier translational vibration with gravity (e) or mesh stiffness (e e) excitations. Bearing clearance is zero, innite backlash is used,
H harmonic, T translational mode, and R rotational mode; (b) RMS dynamic response of ring translational vibration when mesh stiffness is uctuating (e) and constant (e e).
The dotted lines denote the unstable branches. Dc 0.1, 0.5, 1 mm.
the carrier, ring, and sun gears. The rotational modes considered
have only rotation of the carrier, ring, and sun gears. The planet
modes considered have only planet motions; the carrier, ring, and
sun gears do not move. These vibration modes are derived from the
Eigenvalue problem of the equations of motion in Eq. (1) without
tooth contact and bearing clearance nonlinearities. The natural
frequencies of PG-A under 800 Hz are the translational modes at
165, 311, and 796 Hz and the rotational mode at 292 Hz. The natural
frequencies and mode shapes of PG-B are described in Table 3.
6.1. Gravity-induced excitation
The effects of gravity and uctuating mesh stiffness on the dynamic response of PG-B are shown in Fig. 9(a). Within the speed
range of interest, the carrier response with gravity as the excitation
source is one order of magnitude higher than that with uctuating
mesh stiffness. This agrees with the observation that low-frequency
Fig. 10. (a) Dynamic response of the RMS of carrier translational (upper graph) and rotational (lower graph) responses of PG-B with different backlash; (b) drive-side and back-side
tooth contact at the rst sun-planet mesh with one-third of nominal backlash.
54
Fig. 11. RMS xr and x1 dynamic responses of PG-A with bearing clearance from 0 to 400 mm.
Fig. 12. Back-side tooth load at the rst sun-planet mesh and planetary load sharing over a carrier cycle (a) with various bearing clearances and speeds; (b) with various sun
stiffnesses and bearing clearances for PG-A. The carrier speed is 0.5 Hz in (b).
drive-side and back-side of the rst sun-planet mesh with onethird of the nominal backlash under the same condition as
Fig. 10(a). In this gure, a y-axis value of 1 indicates a full-contact
condition and a 0 represents an out-of-contact condition. Within
the speed range of 70e85 Hz, the gear rotates signicantly because
of the rotational mode at 70 Hz and loses tooth contact. In the
meantime, the translational mode causes the radial motion between the sun and planet 1. This radial motion exceeds the available
tooth radial gap, which leads to tooth wedging. The coexistence of
tooth wedging and tooth contact loss within the same speed range
reects the competition between the translational and rotational
modes, resulting in the closed loop in Fig. 10(a).
Fig. 11(a) and (b) shows the dynamic responses of xr and x1, with
various values of bearing clearance for PG-A. The resonant frequency of the rst translational mode decreases from 165 to 28 Hz
when the bearing clearance increases from 0 to 400 mm as shown in
Fig. 11(a). Bearing clearance affects the resonant frequencies of the
translational mode shapes of the member with the bearing clearance. For large-scale wind turbine gearboxes, the resonant frequencies can be further reduced into the range of external lowspeed excitations because of structural exibilities of the drivetrain, the blades, and the supporting components (Helsen et al.,
2010). The responses without clearance and with clearance of
400 mm, which is essentially innite clearance, correspond to two
limiting systems. The mode shapes of these two limiting systems
are different from each other as depicted in Fig. 11(a). In Fig. 11(a),
the ring gear displacement increases 4.6 times when clearance
increases from 0 to 400 mm. In the meantime, the planet translational displacement x1 decreases 25 mm when the bearing clearance increases. Clearance in the carrier bearing increases the
vibration of the central members: the sun, ring, and carrier, but its
effect on planet responses is much smaller. The bearing forces at the
carrier support are small because they are nearly balanced by corresponding self-imposed tooth loads. Because of the absence of the
0
60
Gl a; b; c 6
4
0
cosaU1 bU2 s1
cosaU1 bU2 s2
cosaU1 bU2 sn
sinaU1 bU2 s1
sinaU1 bU2 s2
sinaU1 bU2 sn
/
/
/
/
55
3
0
07
7
5
0
(20)
56
Acknowledgment
3
Xcbrj sin jbri sin ar Xcbrj sin jbrj cos ar Xcbrj sin jbrj
6
7
Lm;b
4 Xcbrj cos jbrj sin ar Xcbrj cos jbrj cos ar Xcbrj cos jbrj 5
r2;j
Xcbrj sin ar
Xcbrj cos ar
Xcbrj
(27)
2
3
/ sin Ri Ui s1 0
/ sin Ri Ui s2 0 7
7
/
5
/ sin Ri U1 sn 0
(28)
3
(19)
Xcbsj sin jbsj sin as Xcbsj sin jbsj cos as Xcbsj sin jbsj
7
6
Lm;b
4 Xcbsj cos jbsj sin as Xcbsj cos jbsj cos as Xcbsj cos jbsj 5
s2;j
b
b
b
Xcsj cos as
Xcsj sin as
Xcsj
(29)
where i 1,2.
where l 3, 4, plus or minus signs are used in the operator,
respectively. When l 3, 4, c R5, R8, respectively.
Ai 4
3
A0i
Xcbrj sin2 ar
Xcbrj sin ar
(21)
B 4
1
2
3
B0i
0
6 1
6
6
6
0
Bi 6
6
6
6
4
5
n
0
2
2
0
1
0
Xi
7
7
7
7
7
7
7
7
Xi 5
0
1
2
(31)
(24)
where drj ; dsj denote the back-side mesh deection vectors at the
jth ring-planet and sun-planet meshes. Parameter s controls how
close the smoothing functions are to the original piecewise
nonlinear functions.
(32)
b
s
L
b
dbrj G zyr cos jbrj zxr sin jbrj zzi sin ar zhi cos ar zui zur r
L
dbsj G zys cos jbsj zxs sin jbsj zzi sin as zhi cos as zui zus
jbsj jj as ; jbrj jj ar
(33)
(25)
where jj denotes the position angle of the jth planet. ar, as denote
the pressure angles of the ring and sun gear teeth. br, bs denote the
tooth backlashes of the ring and sun gears.
References
KLTI diag0; 0; Pc ; 0; 0; Pr ; 0; 0; Ps ; 0; .; 0
(23)
3
1
0
Xcbsj
Xcbsj cos as
(30)
(22)
Xcbrj
2 Xcb sin2 a
Xcbsj sin as cos as Xcbsj sin as 3
s
sj
Xcbsj cos as 5
4 Xcbsj sin as cos as Xcbsj cos2 as
Xcbsj sin as
A0i diag 0; 12 ; 12 ; 22 ; 22 ; .; Xi2 ; Xi2 ; 0 ; i 1; 2; 3; 4
Xcbrj cos ar
Xcbrj sin2 jbrj Xcbrj sin jbrj cos jbrj Xcbrj sin jbrj
7
6
b
b
b7
b
b
2 b
b
Lm;b
6
4 Xcrj sin jrj cos jrj Xcrj cos jrj Xcrj cos jrj 5
r1;j
Xcbrj sin jbrj
Xcbrj cos jbrj
Xcbrj
(26)
57
Link, H., LaCava, W., Van Dam, J., McNiff, B., Sheng, S., Wallen, R., McDade, M.,
Lambert, S., Buttereld, S., Oyague, F., 2011. reportGearbox Reliability Collaborative Project Report: Findings from Phase 1 and Phase 2 Testing. National
Renewable Energy Laboratory.
Musial, W., Buttereld, S., McNiff, B., 2007. Improving wind turbine gearbox reliability. In: European Wind Energy Conference, Milan, Italy, NREL/CPe500e
41548.
Nayfeh, A., Balachandran, B., 1995. Applied Nonlinear Dynamics. John Wiley and
Sons.
Oyague, F., Gorman, D., Sheng, S., 2010. NREL Gearbox Reliability Collaborative
Experimental Data Overview and Analysis. NREL.
Parker, R.G., Agashe, V., Vijayakar, S.M., Sep. 2000a. Dynamic response of a planetary
gear system using a nite element/contact mechanics model. J. Mech. Des. 122
(3), 304e310.
Parker, R.G., Vijayakar, S.M., Imajo, T., Oct. 2000b. Non-linear dynamic response of a
spur gear pair: modelling and experimental comparisons. J. Sound Vib. 237 (3),
435e455.
Raghothama, A., Narayanan, S., 1999. Bifurcation and chaos in geared rotor bearing
system by incremental harmonic balance method. J. Sound Vib. 226 (3), 469e492.
Rasmussen, F., Thomsen, K., Larsen, T., 2004. The Gearbox Problem Revisited. Ris
National Laboratory, AED-RB-17 (EN).
Seydel, R., 1994. Practical Bifurcation and Stability Analysis: From Equilibrium to
Chaos, second ed., Springer.
Singh, A., 2010. Load sharing behavior in epicyclic gears: physical explanation and
generalized formulation. Mech. Mach. Theor. 45, 511e530.
Singh, A., 2011. Epicyclic load sharing map e development and validation. Mech.
Mach. Theor. 46, 632e646.
Thomsen, J., 2003. Vibration and Stability: Advanced Theory, Analysis, and Tools,
second ed., Springer.
Turner, K., Walker, H.F., 1992. Efcient high accuracy solutions with gmres(m). J. Sci.
Stat. Comput. 13 (3), 815e825.
Velex, P., Flamand, L., Mar. 1996. Dynamic response of planetary trains to mesh
parametric excitations. J. Mech. Des. 118 (1), 7e14.
Vijayakar, S.M., 1991. A combined surface integral and nite element solution for a
three-dimensional contact problem. Int. J. Numer. Methods Eng. 31, 524e546.
Zhu, F., Parker, R.G., Jan. 2005. Non-linear dynamics of a one-way clutch in belt
pulley systems. J. Sound Vib. 279 (1e2), 285e308.