Journal of Sound and Vibration (1995) 184(3), 503–528

MODELLING OF A HYDRAULIC ENGINE MOUNT

FOCUSING ON RESPONSE TO SINUSOIDAL AND
COMPOSITE EXCITATIONS
J. E. C, C.-T. C, Y.-C. C, W. K. L  L. M. K
Department of Mechanical Engineering, Northwestern University, Evanston,
Illinois 60208, U.S.A.

(Received 14 June 1993, and in final form 24 June 1994)

Frequency response characteristics of a hydraulic engine mount are investigated.

The mount studied is highly non-linear due to an amplitude-limited floating piston (the
‘‘decoupler’’) which enables the response to large amplitude (typically road-induced)
excitations to differ markedly from the response to small amplitude (typically engine-
induced) excitations. The effect of the decoupler on frequency response as well as
composite-input (sum of two sinusoids) response is considered.
New experimental data for a production mount and several specially prepared mounts
are presented and discussed. Two linear models, one for large amplitude excitations
and one for small amplitude excitations, are developed and shown to be effective over a
5–200 Hz frequency range. The latter model explains a moderately high frequency
(0120 Hz) resonance which is often observed in the data, but which has not previously
been described in physical terms.
A piecewise linear simulation and an equivalent linearization technique are used to
explain the amplitude-dependence of frequency response, as well as the composite-input
response. The applicability of equivalent linearization is justified by demonstrating that
high order harmonics contribute very little to the transmitted force. Moreover, this
technique is found to be computationally efficient and to provide insight into decoupler
dynamics.
7 1995 Academic Press Limited

1. INTRODUCTION
Engine mounts serve two principal functions: vibration isolation and engine support.
In the past decade, the automotive industry’s shift to small, four cylinder engines and
transversely mounted front-wheel-drive powertrains has made these two functions increas-
ingly incompatible. For instance, the lower firing frequencies of four cylinder engines
coupled with lower engine inertia tend to excite higher amplitude vibrations. To avoid
significant chassis vibration and passenger compartment noise, softer mounts become
necessary—it is not uncommon for the elastic rate (stiffness) of a mount in a front-wheel-
drive automobile to be 20 times less than that of a rear-wheel-drive automobile [1]. Engine
mounts, however, must also limit or ‘‘control’’ engine excursions caused by rough roads,
idle shake, vehicle accelerations and wheel torque reaction (which is especially an issue in
front-wheel-drive). To provide control, it is important that the engine mounts be stiff and
heavily damped. This growing disparity between isolation characteristics and control
characteristics has profoundly changed the way in which the industry approaches mount
design.
503
0022–460X/95/280503 + 26 $12.00/0 7 1995 Academic Press Limited
504 . .    .

Figure 1. A schematic of a hydraulic engine mount.

Change has also been forced by an increased market emphasis on passenger comfort.
Comfort encompasses interior noise and vibration as well as the feel of the vehicle on rough
roads and under extreme acceleration. This serves only to heighten the conflicts that arise
in design.
To meet the conflicting requirements of isolation and control, the automotive industry
has turned increasingly to hydraulic engine mounts. A typical hydraulic mount is
illustrated in Figure 1. To provide a basis for the design and analysis of such a mount,
good models are essential. Toward this end, a variety of articles presenting mathematical
models as well as design procedures has been published in the past decade [2–15].
An excellent review is provided by Singh et al. [11]. It should be noted that, while this
paper focuses on passive engine mounts, a number of recent studies have explored
semi-active or adaptive engine mounts as well [3, 10, 16–18].
Until recently, most attempts to model hydraulic mounts assumed linearity and were
restricted to rather limited sets of operating conditions (usually corresponding to test
conditions). Unfortunately, production hydraulic mounts exhibit a variety of non-linear
characteristics and, in application, are subject to a broad range of excitations. The
investigation of non-linear behavior appears to have begun with Ushijima and Dan [13],
who used numerical simulation to investigate nonlinear flow characteristics. More recently,
Kim and Singh [19, 20] began a systematic study of hydraulic mount non-linearities.
Among the effects they have considered are non-linear compliance, non-linear flow
characteristics, cavitation and ‘‘decoupling’’.
This paper contributes further to the understanding of mount non-linearity associated
with the ‘‘decoupler’’. The decoupler, shown in Figure 1, is essentially an amplitude-
dependent switch which is intended to improve the performance trade-off between
vibration isolation and control of engine excursion.
Other contributions of this paper include the presentation of new experimental data
describing the response of a hydraulic mount to composite, dual frequency excitations, as
well as the presentation of a novel ‘‘small amplitude linear model’’. This linear model,
developed as a preliminary to non-linear models, is the first in the literature to capture an
important resonance associated with decoupler inertia. All models presented in this paper
consider single axis excitation only, though extension to multiple axes is possible.
    505
2. HYDRAULIC ENGINE MOUNT CHARACTERISTICS
2.1.  
In Figure 1, xe (t) and xc (t) represent the displacement of the engine and the
chassis, respectively. The relative displacement, x0 (t) = xe (t) − xc (t), is referred to as
the ‘‘excitation’’ of the moment. Physical components which contribute significantly to the
dynamics of the mount include the following. The primary rubber (1) is a rubber cone that
serves several purposes. It support the static load of the engine, it contributes (significantly)
to the elastic rate and (modestly) to the damping of the mount and it serves as a piston
to pump fluid through the rest of the mount. The bulge rate of the primary rubber (ratio
of pressure change to volume change) is also an important design parameter. The
secondary rubber (2) is a rubber septum that serves principally to contain the fluid. It also
contributes modestly to the elastic rate. The orifice plate (3) is a metal plate (actually a
sandwich of two plates) that separates the ‘‘upper chamber’’ (enclosed by the primary
rubber) and the ‘‘lower chamber’’ (enclosed by the secondary rubber). Cast in the orifice
plate are the ‘‘inertia track’’ and ‘‘decoupler orifice.’’ The inertia track (4) is a lengthy spiral
channel that enables fluid to pass from the upper chamber to the lower chamber. The fluid
inertia in this channel is significant, and is usually selected so that it experiences resonance
at the natural frequency of the engine/mount system. The damping of the track is also
significant. Thus, the inertia track acts as a tuned damper, and is introduced for the
purpose of control. The decoupler (5) is a plastic plate which acts as an amplitude-limited
floating piston that provides a low resistance path between the upper and lower chambers.
Thus, for small amplitude excitation, most of the fluid transport between chambers is via
the decoupler orifice, which effectively short-circuits the inertia track. For larger amplitude
excitations, the decoupler ‘‘bottoms out’’ and most of the fluid flow is forced through the
inertia track. The inertia of the decoupler is also important at high frequencies, a point
which will be highlighted in this paper. The fluid (6), ethylene glycol, completely fills the
interior of the mount.

2.2.   

The behavior of an engine mount is usually reported in terms of its frequency response
for different amplitude excitations. Frequency/amplitude ranges of greatest interest
include [6, 11, 14]: (1) 5–15 Hz, 0.5–5.0 mm—these excitations are in the range of engine
resonance and large enough to require significant damping; (2) 25–250 Hz, 0.05–0.5 mm—
these excitations can cause noise and vibration, and require good isolation. Even higher
frequency excitations, which may result from combustion noise [14], have received
attention recently, but are beyond the scope of this paper. Interest has also arisen in the
extent to which hydraulic mounts can provide control and isolation simultaneously [14].
This may be important, for instance, while driving on rough surfaces, or during extreme
accelerations on smooth surfaces. Thus, the response to composite inputs is of interest, and
will be considered in this paper.
The frequency response is typically evaluated with a conventional servo-controlled
hydraulic test rig. The chassis bracket is fixed to an inertially grounded force sensor, while
the engine bracket is sinusoidally excited at a fixed amplitude. Force and displacement
records are collected at a series of frequencies and each record is transformed to the
frequency domain via a discrete Fourier transform. For this study, time domain records
all consist of 8192 points collected at a sample interval of 0·0005 s. To analyze high
frequency behavior (5–200 Hz increments), each record is broken into four contiguous
sections which are independently windowed (Hanning window) and transformed. The four
transforms are then ensemble-averaged to obtain estimates of the Fourier transform
506 . .   .
and coherence. To analyze low frequency behavior (1–40 Hz in 1 Hz increments), each
record is broken into eight interleaved sections (each with an effective sample rate of
0.004 s), which are independently windowed and transformed, then ensemble-averaged.†
In all cases, only those data corresponding to the excitation frequency, v, are retained.
Coherences obtained in this way are not reported because they are in all cases extremely
close to unity. The Fourier transforms, F(jv) and X0 (jv), correspond to the fundamental
harmonics of force and displacement. The ratio of these transforms, known as the
‘‘dynamic stiffness’’ is the principal quantity of interest:
Kdyn (jv) = F(jv)/X0 (jv) = K(v) + jvB(v). (1)
The real part of the dynamic stiffness, K(v), is termed the ‘‘elastic rate’’, while the
imaginary part divided by frequency, B(v), is termed the ‘‘damping’’. In comparing these
data with others in the literature, it should be noted that dynamic stiffness is often
presented in terms of the magnitude (‘‘dynamic rate’’) and phase (‘‘loss angle’’).
Figure 2 shows representative data corresponding to the two classes of excitation
considered above. The two curves correspond to 0.1 and 1.0 mm excitations. These data,
as all comparable data in this paper, have been normalized by the stiffness and damping
of the primary rubber (which are assumed to be amplitude and frequency-independent).
It is clear that the response of the hydraulic mount is strongly amplitude-dependent.

Figure 2. (a) Normalized elastic rate and (b) damping of a production hydraulic mount for small amplitude
(—w—, 0·1 mm) and large amplitude (—— E , 1·0 mm) excitations.

† Computational details are given in reference [21].

    507

Figure 3. (a) Normalized elastic rate and (b) damping of a production hydraulic mount for various amplitudes:
—q—, 0·05 mm; —— E , 0·1 mm; —+—, 0·2 mm; ——, 0·4 mm; ——, 0·6 mm; ––––, 0·8 mm; –·–·–·–, 1·0 mm.

The following points are also noteworthy: (1) the peak in the damping for large amplitude
excitation (015 Hz) has been ‘‘tuned’’ so that it corresponds to the natural frequency
of engine bounce, thus, for large amplitude inputs, the mount serves as a tuned damper;
(2) for frequencies less than 050 Hz, both the elastic rate and damping are much less for
small amplitude inputs, resulting in superior vibration isolation; (3) the peak in damping
for small amplitude excitation (0120 Hz) and the corresponding increase in elastic rate
are the result of decoupler inertia, and are undesirable effects.
For this study, a more extensive set of data has been collected. These data are plotted
in Figures 3–5. Figure 3 shows the response of the mount to different amplitude excitations.
It is evident that, with increasing amplitude, the behavior shifts from the prototypical
‘‘small amplitude’’ response to the prototypical ‘‘large amplitude’’ response. The develop-
ment of a model which captures this shift is a principal objective of this work.
Figure 4 displays frequency responses of three specially prepared engine mounts. The
first is a mount from which the fluid was drained, leaving the primary rubber solely
responsible for the dynamic behavior. The data show that the assumption of amplitude
independence is very good, while the assumption of frequency independence is reasonable,
but will introduce a certain degree of systematic error. The second specialty mount was
assembled without a decoupler. The behavior is, not surprisingly, quite similar to the small
amplitude response of a production mount. The third specialty mount had the decoupler
fixed in place. The behavior resembles the large amplitude response of a production mount.
These data are particularly useful in identifying model parameters.
508 . .    .

Figure 4. (a) Normalized elastic rate and (b) damping of three specially prepared engine mounts: —q—,
mount with no decoupler, 0·05 mm; —— W , mount with fixed decoupler, 1·0 mm; —— E , mount with no fluid,
0·05 mm; — + —, mount with no fluid, 1·0 mm.

Figure 5 displays frequency response of a production mount excited by composite

waveforms:

x0 (t) = xb (t) + xs (t), (2)

where
xb (t) = 1·0 sin (2pvb t) mm, and xs (t) = 0·1 sin (2pvs t) mm.

The ‘‘base frequency’’, vb of the large amplitude component is fixed for a given data set
at 5, 10 or 15 Hz. The frequency, vs , of the small amplitude component is swept from 20 to
200 Hz. These data have been collected to assess the mount’s capacity to provide isolation
in the presence of a large amplitude disturbance. Toward this end, the relationship between
the small amplitude input and the force is of particular interest.
The two frequencies, vb and vs , are generated by separate analog instruments,
and therefore may be considered independent and uncorrelated [22]. This allows the
computation of transfer functions relating each of these inputs to the output force:
b s
Kdyn (jv) = F(jv)/Xb (jv), Kdyn (jv) = F(jv)/Xs (jv). (3)
F(jv), Xb (jv) and Xs (jv) are Fourier transforms of their respective time domain
records. It can be shown that transfer functions computed in this way are optimum
    509

Figure 5. (a) Normalized elastic rate and (b) damping of a production mount excited by composite input
consisting of a 1·0 mm sinusoid at a base frequency (5, 10, 15 Hz) and a 0·1 mm sinusoid (frequency on horizontal
axis): —q—, 5 Hz base; —+—, 10 Hz base; —w—, 15 Hz base; —— E , 0·1 mm single frequency response; ——Q ,
1·0 mm single frequency response.

linear approximations to the underlying non-linear system [22]. Because we are interested
s
specifically in the response to small amplitude inputs, only the value of Kdyn (jv)
corresponding to the excitation frequency vs is retained. The coherence associated with
this value approaches unity.
s
The results are quite interesting. They indicate that Kdyn (jv) measured in the presence
of a large amplitude disturbance bears a strong resemblance to the mount’s large amplitude
response (Figure 2). This implies that the mount’s capacity to provide isolation from
small amplitude, high frequency inputs is significantly degraded by the simultaneous
presence of a large amplitude, low frequency disturbance. A similar conclusion was reached
by Ushijima et al. [14] who performed a similar experiment.

2.3.  

The data described above give a more complete picture of hydraulic mount behavior
than previously available. Based on them, it seems reasonable to assert that a good
mount model should be able to capture: the low frequency, large amplitude resonance
(which describes the mount’s performance as a tuned damper); the high frequency, small
amplitude resonance (which, though undesirable, figures prominently in the mount’s
performance as an isolator); the amplitude dependence of elastic rate and damping; and
510 . .    .
the mount’s response to composite inputs (which measures the mount’s ability to provide
simultaneous isolation and control).
The next section introduces two linear models. These models embody our basic physical
understanding of the hydraulic mount, and are sufficient to capture the two resonances.
Section 4 presents two approaches to estimating non-linear frequency responses, a
piecewise linear simulation and an equivalent linearization analysis. These techniques are
also extended to predict the mount’s response to composite inputs. Section 5 presents a
summary discussion.

3. LARGE AND SMALL AMPLITUDE LINEAR MODELS

In this section, two linear models are introduced. The first is tailored to large ampli-
tude sinusoidal excitation (q0·5 mm), and makes the assumption that the decoupler
is ‘‘bottomed out’’ at all times. The second is tailored to small amplitude excitation
(Q0·5 mm), and makes the assumption that the decoupler never bottoms out. Both models
assume that all other important physical effects may be represented by lumped, linear
time-invariant elements.

3.1.   

The principal physical effects are taken to be those associated with the primary rubber
(shear stiffness and damping; bulge stiffness; piston area) and with the inertia track (fluid
inertia and damping). The stiffness of the secondary rubber is small enough that it can be
ignored (reducing the requisite state dimension by one).† The fluid is assumed incompress-
ible (the bulge compliance of the primary rubber is much greater), and the fluid inertia
and damping in the upper chamber are ignored (inertia and damping in the track are much
greater). The interconnection of these elements is straightforward—both bond graph
[23, 24] and mechanical equivalent models are shown in Figure 6. Similar models have
been presented in references [2, 5, 7, 11, 14].
State equations and an output equation for the reaction force may be derived from the
bond graph. They are

&' & '& ' & '

ẋr 0 0 0 xr 1
Vb = 0 0 −1 Vb + Ap ẋ0 , (4)
Qt 0 Kb /It −Bt /It Qt 0

&'
xr
F = [Kr Ap Kb 0] Vb + [Br ]ẋ0 .
Qt
The states are shear displacement of the primary rubber (xr ), bulge displacement of the
primary rubber (Vb ) and volumetric flow in the inertia track (Qt ).
Parameter values were obtained as follows. Ap , the ‘‘piston area’’ of the primary rubber,
is equal to the area of the orifice plate top surface, which is easily measured. Initial
estimates of bulge stiffness, Kb , track inertia, It , and track damping, Bt , were made using
the analytical formulas given in Singh et al. [11]. These formulas, however, are necessarily

† This may be understood by comparing the response of the mount without fluid to the responses of the
fluid-filled mounts. At low frequency, both rubber shear stiffness and secondary rubber stiffness should contribute
to the elastic rate of the latter class; yet, these mounts exhibit nearly the same values as the dry mount (Figure
4). Measurements of primary rubber bulge compliance and secondary rubber compliance presented by Kim and
Singh [20] also indicate that the latter is at least an order of magnitude greater.
    511

Figure 6. The large amplitude model.

based on idealized geometries, such as straight inertia track and a hemispherical upper
chamber. Also idealized are the assumptions of laminar and fully developed flow through
the inertia track. Finally, even such basic assumptions as the absence of leakage paths
tend to short-circuit the inertia track may be questionable. Thus, analytically determined
parameters must generally be considered no better than order-of-magnitude estimates.
In this study, more accurate estimates of Kb , It and Bt , as well as primary rubber shear
stiffness, Kr , and damping, Br , were obtained from the frequency responses of the specialty
mounts.
The specialty mount containing no fluid gives a direct measure of Kr and Br . Though
these parameters vary somewhat with frequency and excitation amplitude (Figure 4),
representative values are selected and used in all models. These values are also used to
normalize the elastic rate and damping plots.
The specialty mount with fixed decoupler should behave very much as the large
amplitude model would predict; thus, it is useful for estimating Kb , It and Bt . These
estimates are obtained by relating characteristics of the frequency response measurements
to model predictions. The transfer function associated with the large amplitude model
(equation (4)) is:

F(s) I s 2 + Bt s
Kdyn (s) = = Kr + Br s + Ap2Kb 2 t . (5)
X(s) I t s + Bt s + Kb

It is convenient to define a natural frequency and damping ratio associated with the fluid
part of the system:

vn = zKb /It , z = Bt /2 zKb It . (6)

512 . .   .
The frequency response may then be written in terms of real (elastic rate) and imaginary
(damping) parts:

$ %
v 2(v 2 − vn2 ) + 4z 2vn2v 2
Kdyn (jv) = Kr + Ap2Kb
(v 2 − vn2 )2 + 4z 2vn2v 2

$ %
2zvn3
+jv Br + Ap2Kb , (7)
(v − v ) + 4z 2vn2v 2
2 2 2
n

At frequencies well above vn (q30 Hz), the elastic rate approaches Kr + Ap2Kb . Because
Kr and Ap are known, this measure of elastic rate may be used to estimate bulge stiffness.
In addition to the high frequency elastic rate, salient and reproducible features of the data
are the resonant frequency, vr1 , at which the peak in damping occurs, and the slope of
the elastic rate curve at resonance, mr1 (see also Figure 7). These measures may be used
to estimate z and vn , from which It and Bt can be determined. An expression for vr1 is
found by maximizing the damping term in equation (7):
vr12 = vn2 (1 − 2z 2 ). (8)
An expression for mr1 is found by differentiating the elastic rate term in equation (7), and
evaluating at v = vr1 :
mr1 = (Ap2Kb /2vr1 )((1 − 2z 2 )/z 2(1 − z 2 )). (9)

Figure 7. (a) Normalized elastic rate and (b) damping of large amplitude model versus data: ——, large
amplitude model; –—–, mount with fixed decoupler, 1·0 mm.
    513

Figure 8. Relative errors of large amplitude and small amplitude linear models: ——, large amplitude, ––––,
small amplitude. (a) Elastic rate; (b) damping.

Given measurements of vr1 and mr1 , equation (9) may be used to estimate z: subsequently,
equation (8) may be used to estimate vn . Determination of It and Bt is then straight-
forward, completing the parameter identification process. The frequency response of the
identified large amplitude model is shown in Figure 7.
A quantitative measure of model performance is the frequency-dependent relative error,
which for elastic rate will be defined as
eK (v) = [K
(v) − K(v)]/max (K(v)), (10)
v

where K(v) is the measured elastic rate and K

(v) is the predicted elastic rate. The relative
error in damping is defined analogously. These quantities are plotted in Figure 8 (along
with those for the small amplitude model). Relative errors in elastic rate are generally less
than 10%, while those in damping are less than 20%.

3.2.   

The principal physical effects are taken to be those incorporated in the large amplitude
model, plus those associated with the decoupler orifice (orifice inertia and damping).
Figure 9 illustrates the model of decoupler orifice dynamics which is used. The decoupler
itself and surrounding fluid are treated as a lumped inertia, I0 , while the fluid shear layer
between this inertia and the orifice plate is treated as a lumped damper, B0 . The pressure
drop between upper and lower chambers (Kb Vb , where Vb is the bulge displacement of the

Figure 9. The orifice plate model.

514 . .   .
primary rubber) drives the orifice flow, Q0 . A pressure balance gives
Kb Vb = I0 Q0 + B0 (Q0 − Ad ẋc ). (11)
Ad is the cross-sectional area of the orifice, and xc is the displacement of the chassis, to
which the orifice plate is fixed.
An expression for the force transmitted to the chassis can also be derived with the
assistance of Figure 9. Contributions are the forces transmitted via rubber shear, the
normal forces on the orifice plate due to pressure, and the shear forces due to drag:
F = Kr (xe − xc ) + Br (ẋe − ẋc ) + (Ap − Ad )Kb Vb + (Ad2 B0 )(Q0 /Ad − ẋc ), (12)
where xe is the displacement of the engine.
The effect of orifice dynamics, as represented in equations (11) and (12), can be added
to the large amplitude model by recognizing that the pressure drop across the inertia track
is the same as that across the decoupler orifice. A complete small amplitude model is shown
in Figure 10. A mechanical equivalent to this bond graph can be found, but is quite
non-intuitive.
One interesting aspect of the small amplitude model is that it requires two inputs (xe and
xc ) rather than a single input (x0 = xe − xc ). The relative displacement is an appropriate
input only in instances when either the engine or the chassis serves as an inertial ground
and therefore a proper reference for the orifice inertia.† While no such instance occurs in
an automobile, it does occur in the test fixture described in section 2.2 which fixes the
chassis bracket to ground (xc = 0). In this case, state and output equations are
K
GVr L
ẋ
G K
0
G0
0 0 0 LK xr L K 1L
GGV G GAG
0 −1 −1
G QbG= G0 K /I −B /I 0
GG QbG+ G 0G
p
ẋ0 , (13)
G t
G G b
kQ0l k0 Kb /I0
t t
0
t
GG t
G G
−B0 /I0lkQ0l k 0l
G

K
GVxr L
G
F = [Kr (Ap − Ad )Kb 0 Ad B0 ]G bG+ [Br ]ẋ0 .
Qt
G
kQ0G l
All variables retain previous definitions. The only parameters which did not appear in the
large amplitude model re Ad , I0 and B0 .
The decoupler cross-sectional area is easily measured. Orifice inertia and damping can
be estimated by a procedure similar to that used to find track inertia and damping. To
simplify the analysis, it is first assumed that, at the frequency of orifice resonance
(v2 1 120 Hz), flow through the inertia track can be neglected. This simplification gives
the dynamic stiffness of the small amplitude model a form very similar to equation (5):
F(s) (1 − e)I0 s 2 + B0 s
Kdyn (s) = = Kr + Br s + Ap2Kb , (14)
X(s) I 0 s 2 + B 0 s + Kb
where e = Ad /Ap . Equation (14) may be used to develop expressions for the resonant
frequency (vr2 ) and slope of the elastic rate plot at resonance (mr2 ). These expressions
may be equated to values measured using the speciality mount with no decoupler, and
solved to yield estimates of I0 and B0 . This process is similar to that used to estimate

† This is not an issue with the track inertia because flow in the track is perpendicular to the assumed axis of
motion.
    515

Figure 10. The small amplitude model.

It and Bt , but somewhat more involved due to the presence of e. Details are given in
Appendix A.
The frequency response of the fully identified small amplitude model is shown in
Figure 11. Relative errors are shown in Figure 8, and are comparable to those of the large

Figure 11. (a) Normalized elastic rate and (b) damping of small amplitude model versus data: ——, small
amplitude model: ––––, mount with no decoupler, 0·05 mm.
516 . .   .
magnitude model. Although the linear models successfully capture the two resonances,
they cannot represent in any greater detail the amplitude dependence of the mount’s
response. To meet this objective, two non-linear models are introduced.

4. NON-LINEAR MODELLING
A variety of non-linear effects contributes significantly to the behavior of a production
hydraulic mount [20]. These include, for instance, entrance and exit effects in the inertia
track flow, and amplitude-dependent softening of the primary rubber. In this paper,
however, attention will be focused on the decoupler, which is principally responsible for
the amplitude dependence.

4.1.   

A direct approach to incorporating decoupler behavior is via a piecewise linear model
[19, 25]. In essence, the piecewise linear model reduces to the small amplitude linear model
(augmented with a state for decoupler position, xd ) whenever the decoupler is not
bottomed out, and to the large amplitude model when the decoupler is bottomed out. The
behavior of the model can be investigated using numerical integration. The model
developed in reference [25] uses the following scheme for switching during a time domain
integration (d is half the gap width):
If the previous step was integrated with the small amplitude model: if =xd = Q d,
integrate small amplitude state equations; if =xd = e d and sign (Vb ) = sign (xd ), set Q0 = 0,
set xd = sign (xd )d, and integrate large amplitude state equations.
If the previous step was integrated with the large amplitude model: if sign (Vb ) = sign
(xd ), integrate large amplitude state equations; if sign (Vb ) $ sign (xd ), integrate small
amplitude state equations.
Note that Vb is linearly related to the pressure in the upper chamber, which determines
whether a decoupler at the limits of its travel will remain bottomed out or not.
A piecewise linear model has several attractive features. No new dynamic elements need
be introduced to account for bottoming out, nor do the (possibly very fast) dynamics of
the bottoming out process need to be directly considered. Moreover, the linear continuous
state equations can be directly mapped to difference equations which are guaranteed stable
[26], so that an efficient simulation can be performed. Finally, excitations of arbitrary shape
are easily accommodated.
A significant disadvantage, however, is that a tremendous amount of time domain
simulation data must be generated to produce a modest amount of frequency domain data.
Frequency responses have been computed as follows. A sinusoidal excitation of amplitude
X0 and frequency v is assumed. A simulation time increment of 2p/v/64 seconds is picked,
and time domain records of approximately 1150 points are computed. Records include
position input and force output information. The beginning of each record is cut off to
remove the transient response, leaving a record of 1024 points. This record is broken into
three overlapping sections of 512 points each, which are Hanning windowed and
transformed with an FFT. A complex transfer function is computed at the frequency of
excitation.
Results are shown in Figure 12 (which may be compared with Figure 3), and relative
errors in Figure 13. Generally speaking, the simulation is quite accurate. Relative errors
in elastic rate and damping are typically less than 20%. The greatest errors occur in the
vicinity of the high frequency resonance, due both to noise in the measured values and
to poorness of fit. It is suspected that the poor fit near resonance is due to the lack of an
appropriate model of the bottoming out process. This is addressed further below, but it
    517

Figure 12. (a) Normalized elastic rate and (b) damping of piecewise linear model for various amplitude
E , 0·1 mm; ——
excitations. Compare with Figure 3: —q—, 0·05 mm; —— Q , 0·2 mm; ——, 0·4 mm; ––––, 0·6 mm;
–·–·–·–, 0·8 mm; – – – –, 1·0 mm.

should be noted here that limited insight into bottoming out is a significant weakness of
the piecewise linear model.

4.2.  

An alternative approach to non-linear modelling, equivalent linearization,
provides roughly comparable results (to piecewise linear modelling) at a greatly reduced
computational cost, and with the added benefit of improved physical insight. Equiv-
alent linearization techniques have been applied to similar problems with considerable
success (see, for instance, Dubowsky and Freudenstein [27] and Comparin and Singh
[28]). To introduce the technique, the simplified mechanical system in Figure 14 will first
be considered. Note that the mass representing decoupler inertia is constrained by a
‘‘cage’’.
The idea of equivalent linearization is, given a prescribed excitation (e.g., x0 (t)=
X0 sin vt), to replace the cage with a linear element or elements that would yield the
same motion, at least up to the fundamental of the Fourier series describing that motion.
The equivalent linear element(s) are then, in effect, parameterized by the amplitude and
frequency of excitation.
The specific approach taken here is illustrated in Figure 15. As the figure indicates, only
one element, the viscous damper, is affected by equivalent linearization. There is some
physical justification for this model. First, there is no physical reason to associate
518 . .   .

Figure 13. Relative errors for piecewise linear model: —q—, 0·05 mm; —— E , 0·1 mm; ——
Q , 0·2 mm; ——,
0·4 mm; ----, 0·6 mm; –·–·–·–, 0·8 mm; – – – –, 1·0 mm. (a) Elastic rate; (b) damping.

significant energy storage (potential or kinetic) with bottoming out. Second, the decoupler
is immersed in a fluid which must be displaced as the cage boundaries are approached.
Thus, a squeeze film [29] is developed. It is well-known that the squeeze film between
parallel plates produces a damping force which varies as the inverse cube of the gap
[29, 30]:
Fsq = g(h /h 3 ) (15)
Here, h is the gap thickness and g is a geometric parameter. For simple geometries,
equation (15) is readily derived from the Reynolds equation for viscous flow [30].
The geometry of decoupler/cage interaction is not simple, but it has been assumed that
the inverse cube form holds nonetheless. The implications of this assumption will be
reassessed below. Because the cage in Figure 14 has two sides, the appropriate damping

Figure 14. The simplified model.

    519

Figure 15. The model which treats the decoupler cage as an amplitude-dependent dissipator.

relation is

$ % $ %
1 1 . 1 + 3x̄ 2 .
Fsq (xd , ẋd ) = g + ẋ , Fsq (x̄, x̄) = bd x̄, (16a, b)
(d − xd )3 (d + xd )3 d (1 + x̄ 2 )3
where b is a geometric constant with units of damping and x̄ = xd /d.
The non-linear damping relation of equation (16b) may be converted to an equivalent
linear damping via a procedure outlined by Gibson [31]. To begin, assume that
x̄ = a sin vt. Then the force that results from equation (16b) can be represented by a
Fourier series:
Fsq (a sin vt, av cos vt) = F0 + Keq da sin vt + Beq dav cos vt+higher harmonics. (17)
It is easily shown that the constant term (F0 ) and in-phase component of the fundamental
(Keq da) are zero. Ignoring higher order harmonics, the equivalent behavior is that of a
viscous damper with an amplitude-dependent coefficient,

g
b
2p
cos2 u(1 + 3a 2 sin2 u)
Beq (a) = du, (18)
p 0
(1 − a 2 sin2 u)3
where u = vt. It is interesting to note that this amount of viscous damping would
ensure precisely as much energy dissipation per cycle as the squeeze film damper, assuming
the same sinusoidal form for x̄. This function is plotted in Figure 16. Note that

Figure 16. Equivalent damping as a function of normalized decoupler amplitude (equation (18)): ––––. Effect
of equivalent damping on normalized decoupler amplitude, assuming fixed excitation amplitude (equation (19)):
E , v=0·2; ——
—— Q , v=0·6; —e—, v=1·0; —— W , v=1·4; —q—, v=1·8; where v is normalized frequency.
Simplified form of Beq (a) ———.
520 . .   .
the equivalent damping approaches infinity as the normalized decoupler amplitude, a,
approaches 1.
If B0 f(x) in Figure 15 is replaced with Beq , a transfer function may be found relating
X(jv) to X0 (jv). The magnitude of this transfer function can be considered an expression
for the normalized decoupler amplitude in terms of equivalent damping, excitation
amplitude and frequency:
X Kb X0
a(Beq , X0 , v) = = . (19)
d z(K − I v 2 )2 + (B v)2 d
b 0 eq

This relation has been plotted in Figure 16 for a single value of X0 and several frequencies.
Given specific values for X0 and v, equations (18) and (19) may be considered simultaneous
equations to be solved for a and Beq . This solution corresponds to a point of intersection
in Figure 16. It is evident that, at each frequency, only one intersection will be found; thus,
a uniquely defined frequency response will be computed (the uniqueness of the result can
be proven rigorously [32]). The result is shown in Figure 17.
This same technique has been used to compute the frequency response of the
hydraulic mount at various amplitudes. The linear orifice damping of the small amplitude
model is replaced by the squeeze film damping of equation (16), with b = B0 (thus, for
an infinitesimal excitation, the non-linear model reduces to the small amplitude linear
model).

Figure 17. Frequency response —— of the system in Figure 14 treating the cage as an amplitude-dependent
dissipator (Figure 15). Small amplitude (linear) frequency response ––––. (a) Elastic rate; (b) damping.
    521

Figure 18. Normalized elastic rate and damping of equivalent linear model for various amplitude excitations.
Compare with Figure 3: —q—, 0·05 mm; —— E , 0·1 mm; ——Q , 0·2 mm; ——, 0·4 mm; ----, 0·6 mm; –·–·–·–,
0·8 mm; — — —, 1·0 mm. (a) Elastic rate; (b) damping.

The results are shown in Figure 18 (compare with Figure 3), and the relative errors
in Figure 19. Equivalent linearization proves, in fact, to be superior to the piecewise
linear model for high frequency inputs, but poorer for low frequency, large amplitude
inputs. The former point is expected, because equivalent linearization treats the bottoming
out process in a more physically meaningful way and because, at high frequency, the
assumption of sinusoidal decoupler motion is qiute reasonable. By the same token,
the latter point may be understood: for large amplitude, low frequency inputs, the
decoupler motion is nearly a square wave, and the force generated by the squeeze film is
dominated by pulses which occur as the decoupler approaches and departs the cage limits.
Indeed, most of the energy of the force signal is then found in the fifth and higher
harmonics.

4.3.     

Given the amount of spectral information which is lost, it may be surprising that
the equivalent linearization technique works as well as it does. This is in part because
comparison is being made to experimental data which has been analyzed in a similar way,
ignoring higher harmonics in the force data. A second reason of importance, however, may
be that the measured force is affected principally by the pressure in the upper chamber,
522 . .   .

E , 0·1 mm; ——
Figure 19. Relative errors of equivalent linear model: —q—, 0·05 mm; —— Q , 0·2 mm; ——,
0·4 mm; ––––, 0·6 mm; –·–·–·–, 0·8 mm; — — —, 1·0 mm. (a) Elastic rate; (b) damping.

not the pressure in the squeeze film; moreover, while higher harmonics may contribute
significantly to the latter, they do not to the former.†
To illustrate these points, the various effects contributing to measured force may be
estimated and compared. The greatest effect, in response to low frequency, large amplitude
excitations, is upper chamber bulge caused by the imposed displacement. The magnitude
of this effect is
F1 1 Ap2Kb X0 . (20)
This force is purely sinusoidal, contributing no harmonics. There are two effects associated
with the decoupler. The first is associated with the amount of upper chamber bulge caused
by decoupler motion. At low frequency the decoupler motion will approximate a rectangle
wave; therefore, the associated force will exhibit harmonics. The magnitude of the nth
harmonic will be approximately proportional to 1/n for the first few harmonics. Higher
harmonics will be even smaller because the decoupler motion is smooth rather than truly
rectangular. The force contributed by a lower harmonic will be, approximately,
F2 1 Ap Kb (Ad (d/n)). (21)
The magnitude of this effect is much smaller than F1 . For the fundamental, F2 /F1 1 0·1,
given a 1 mm excitation amplitude.
† Shear force in the rubber contributes heavily to the measured force as well, but is relatively simple to describe
and not particularly interesting in the present discussion.
    523
The second force term is associated with decoupler inertia. This term is simply decoupler
mass times acceleration, which, for the nth harmonic, is of the order
F3 1 (I0 Ad2 )(d/n)(nv)2 1 Ap2Kb e 2nd(v/vr2 )2. (22)
For n = 5 and v = 30 Hz, F3 /F1 1 5 × 10 . Thus, because the inertia of the decoupler is
−4

very small, contributions of squeeze film pressure pulses to the measured force are very
small.

4.4.     

The point was made early in section 4.2 that no particularly compelling reason exists
to assume that the squeeze film relation applies to a hydraulic mount. Results of the
equivalent linearization analysis, however, support the idea that a reasonable physical
picture of bottoming out includes amplitude-dependent damping. With this in mind,
a simplified form of Beq (a), shown in Figure 16, was investigated. Here, the equivalent
damping is set to B0 unless the decoupler amplitude is observed to exceed its physical limits.
In such a case, an equivalent damping is selected which is precisely large enough to ensure
that the decoupler motion remains within limits. The results (not shown) are very similar
to those obtained with squeeze film damping. This similarity suggests that the precise form
of equivalent damping relation is not very important, as long as it conforms to the general
shape seen in Figure 16.

4.5.  

A final objective of this work has been to predict the hydraulic mount’s response to
composite excitations of the form given in equation (2). This is straightforward with the
piecewise linear model, but less so via equivalent linearization.
The use of the piecewise linear model remains exactly the same as with a single
frequency, except that the input changes to that given in equation (2). As with the
experimental data, the transfer function is computed at vs , the frequency of the small
amplitude component, only. This focuses attention on the capacity of the mount to provide
isolation in the presence of low frequency disturbances.
As in the experimental measurements, it is necessary to ensure that the two components
of the excitation may be treated as statistically independent. Because the amplitude and
frequency of each component are fixed, the phase angles must be independent. This will
not be the case if the two sinusoids are harmonically related. Thus, in performing a
simulation it is important to pick vb and vs carefully so that, to within machine precision,
no harmonic relationship exists.
Results are presented in Figure 20, which has the same format as Figure 5. Both
simulation and experiment suggest that the response to small amplitude inputs in the
presence of a low frequency, large amplitude disturbance is roughly comparable to
the response one would expect for large amplitude single frequency inputs (which of
course cannot be readily generated at such high frequencies). It is not surprising that the
piecewise linear simulation would predict this because it employs the large amplitude
model whenever the decoupler is bottomed out; moreover, the large amplitude component
of the excitation ensures that the decoupler is bottomed out a significant percentage of
the time.
524 . .   .

Figure 20. Response to composite excitations predicted by piecewise linear model. Compare with Figure 5.
—q—, 5 Hz base; —+—, 10 Hz base; —w—, 15 Hz base; —— E , 0·1 mm single frequency response; ——
W ,
1·0 mm single frequency response. (a) Elastic rate; (b) damping.

It is also instructive to use the equivalent linearization approach, which begins with the
assumed decoupler motion (normalized)
x̄ = ab sin (vb t + fb ) + as sin (vs t + fs ), (23)
where ab + as Q 1. The equivalent linear behavior can be found using a double Fourier
series [33], which has the form
a a
6Fsq = s s [Pmn sin (mub + nus ) + Qmn cos (mub + nus )], (24)
m = 0 n = −a(m $ 0)
n = 0(m = 0)

where ub = vb t + fb , us = vs t + fs . Note that many frequencies other than har-

monics appear in the Fourier series. For instance, output frequencies of vs 2 vb ,
vs 2 2vb ,..., 2vs 2 vb , 2vs 2 2vb ,... can be expected. These terms are often called
‘‘combination tones’’ [34]. If the ratio of vs to vb is rational, then an infinite number of
combination tones will occur at the frequencies of excitation. Thus, it is once again
necessary to ensure that a harmonic relationship does not exist between the two sinusoidal
components. Even so, combination tones will exist at frequencies close to excitation, and
it is necessary to assume that they have minimal effect on the response. This assumption
is supported in part by the reasoning presented in section 4.3.
    525
It is readily shown that Q00 , P10 and P01 are all zero, while Q10 and Q01 are non-zero and
are related to the equivalent damping at vb and vs , respectively:

gg
p p
Q10 1 a s vs
Bb (ab , as ) = = (cos ub + cos us )
ab vb 2p 2 −p −p
ab vb

1 + 3(ab sin ub + as sin us )2

× cos ub dub dus
(1 − (ab sin ub + as sin us )2 )3

gg
p p
1 (1 + 3(ab sin ub + as sin us )2
= cos2 ub du du , (25)
2p 2 −p −p
(1 − (ab sin ub + as sin us )2 )3 b s

gg0 1
p p
Q01 1 ab vb
Bs (ab , as ) = = cos ub + cos us
as vs 2p 2 −p −p
as vs

1 + 3(ab sin ub + as sin us )2

× cos us dub dus
(1 − (ab sin ub + as sin us )2 )3

gg
p p
1 (1 + 3(ab sin ub + as sin us )2
= cos2 us du du . (26)
2p 2 −p −p
(1 − (ab sin ub + as sin us )2 )3 b s
These equations may be used, together with two transfer function relations similar to
equation (19) (one at each frequency) to yield a set of four non-linear equations in four
unknowns (Bb , Bs , ab , as ).
A very useful result can be obtained, however, without solving any sets of equations.
It is readily shown by direct computation that Bs q Bb whenever ab q as . Clearly, we
can expect that the latter condition holds because the excitation amplitude at the base
frequency is ten times larger than that at the higher frequency. Moreover, the equivalent
damping at the base frequency, Bb , must be large enough to limit the decoupler motion. As
seen in the single frequency analysis, this ensures that the frequency response approaches
that of the large amplitude model. The inequality in equivalent damping values then
implies that the response at the higher frequency, vs , also approaches that of the large
amplitude model. This is in agreement with the behaviors seen in Figures 5 and 20.

5. CONCLUSIONS
Novel experimental data and mathematical models describing a hydraulic engine mount
have been presented. Piecewise linear and equivalent linear models have been shown to
represent hydraulic mount behavior over a broader range of excitations than previously
possible. While these results provide an excellent basis for hydraulic engine mount analysis
and design, they also suggest a number of topics for future research.
For instance, while the large amplitude and small amplitude linear models fit the data
remarkably well, the incorporation of certain additional effects, such as lower chamber
compliance, upper chamber bulge damping and leakage past the decoupler, can lead to
moderately improved fits [25]. More importantly, it is possible that some of these effects
could be enhanced to improve the performance of future mount designs. For instance, a
significant increase in upper chamber damping can be used to eliminate the high frequency
resonant peak (with the cost, however, of higher damping at frequencies above resonance).
Along similar lines, an interesting adaptive hydraulic mount was recently proposed by Kim
and Singh [17]. This concept uses intake manifold vacuum and an electronic controller to
switch the upper chamber compliance between high and low values according to vehicle
operating conditions.
526 . .   .
Non-linearities other than those associated with the decoupler, such as entrance and exit
effects in the inertia track, may also exert a significant influence on the mount’s behavior
[20]. For instance, the over-estimation of damping at frequencies less than 10 Hz (Figure 7)
is probably a consequence of having ignored such effects. Thus, future models should
incorporate other non-linear effects.
The piecewise linear model presented here was particularly adept at capturing low
frequency amplitude dependence while the equivalent linearization technique was adept at
higher frequency. This was because the former did not incorporate a meaningful model
of bottoming out, while the latter was incapable of representing substantially non-sinu-
soidal signals. A valuable contribution would be the development of a computationally
efficient model combining the best characteristics of each. This is a challenge because
higher order equivalent linearization techniques generally require the solution of sets of
non-linear equations, while a simulation which represents bottoming out explicitly will be
stiff.
The two models substantiated experimental results obtained with composite waveforms.
The equivalent linearization technique, however, left open the tantalizing possibility that
the mount’s ability to provide simultaneous control and isolation could be improved with
some other decoupler characteristic. This possibility has also been raised by Ushijima et al.
[14], who argue that a rubber membrane decoupler gives better performance than a rigid
plate decoupler, as considered here.
Finally, it would be inappropriate to close a discussion of hydraulic engine mounts
without recalling the greater context. Engine mounts are but one contribution to the
noise, vibration and harshness characteristics of an automobile. Hydraulic mounts, in
particular, are designed for significant energetic interaction with the engine and chassis
(as an example, the ‘‘apparent inertia’’ of the inertia track, Ap2 It , is comparable to the engine
mass). It is reasonable to expect, therefore, that the performance of a hydraulic mount
will be sensitive to the dynamic characteristics of the vehicle in which it is
placed. Moreover, performance must ultimately be considered a matter of subjective
(passenger) impression. These factors have to date, and will for the foreseeable future,
necessitate a rather lengthy cycle of design, testing and redesign. Thus, it is not simply the
hydraulic mount itself, but its role in this broader context which must be the subject of
future studies.

ACKNOWLEDGMENTS
The authors would like to thank Jim Frye, Tony Villanueva and Bill Resh of Chrysler
Corporation for their support of this research. The authors would also like to thank
Ed Probst and his staff at Delco Products for their assistance in supplying and testing
hydraulic engine mounts.

REFERENCES
1. W. R 1994 Chrysler Corporation. Personal communication.
2. M. B 1984 SAE Paper 840259. A new generation of engine mounts.
3. S. B, R. S, T. K and P. S 1989 SAE Paper 891160. Optimal
tuning of adaptive hydraulic engine mounts.
4. R. M. B and A. G. H 1993 SAE Paper 931321. On the dynamic response of
hydraulic engine mounts.
5. M. C 1985 SAE Paper 851650. Hydraulic engine mount isolation.
6. P. E. C and G.-H. T 1984 SAE Paper 840407. Hydraulic engine mount character-
istics.
    527
7. W. C. F 1985 SAE Paper 850975. Understanding hydraulic mounts for improved vehicle
noise, vibration and ride qualities.
8. K. K 1989 SAE Paper 891138. Hydraulic engine mount for shock isolation at
acceleration on the FWD cars.
9. R. A. M 1984 SAE Paper 840410. Hydraulic mounts—improved engine isolation.
10. R. S, P. L. G and T. L. H 1986 SAE Paper 860549. Adaptive hydraulic
engine mounts.
11. R. S, G. K and P. V. R 1992 Journal of Sound and Vibration 158, 219–243. Linear
analysis of automotive hydro-mechanical mount with emphasis on decoupler characteristics.
12. M. S and E. A 1986 SAE Paper 861412. Optimum application for hydroelastic engine
mount.
13. T. U and T. D 1986 SAE Paper 860550. Nonlinear B.B.A. for predicting vibration
of vehicle with hydraulic engine mount.
14. T. U, K. T and H. K 1988 SAE Paper 880073. High performance hydraulic
mount for improving vehicle noise and vibration.
15. J. P. W 1987 Automotive Engineer 12, 17–19. Hydraulically-damped engine mounts.
16. P. L. G, D. N, R. S, J. S and R. W. S 1988 SAE Paper
880074. Active frame vibration control for automotive vehicles with hydraulic engine mount.
17. G. K and R. S 1993 in Advanced Automotive Technologies (M. Ahmadian, editor).
247–255. New York: ASME. A broadband adaptive hydraulic mount system.
18. R. S, P. L. G, S. B, T. K and R. W. S 1988 SAE Paper
880075. Open-loop versus closed-loop control for hydraulic engine mounts.
19. G. K and R. S 1992 in Transportation Systems—1992, 165–180. New York, ASME.
Resonance, isolation and shock control characteristics of automotive nonlinear hydraulic engine
mounts.
20. G. K and R. S 1993 Journal of Dynamic Systems, Measurement and Control 115, 482–487.
Nonlinear analysis of automotive hydraulic engine mount.
21. C. T. C 1992 Master’s Thesis, Northwestern University. Dynamic analysis of hydraulic
engine mounts.
22. J. S. B and A. G. P 1986 Random Data: Analysis and Measurement Procedures.
New York: John Wiley.
23. D. K and R. R 1975 System Dynamics: A Unified Approach. New York: John
Wiley.
24. H. M. P 1961 Analysis and Design of Engineering Systems. Cambridge, Massachusetts:
MIT Press.
25. Y. C. C 1993 Master’s Thesis, Northwestern University. Computer simulation of hydro-
elastic engine mounts.
26. G. F. F, J. D. P and M. L. W 1990 Digital Control of Dynamic Systems.
Reading, Massachusetts: Addison-Wesley.
27. S. Dubowsky and F. Freudenstein 1971 Transactions of the American Society of Mechanical
Engineers, Journal of Engineering for Industry 93, 310–316. Dynamic analysis of mechanical
systems with clearances, Part 2: dynamic response.
28. R. J. C and R. S 1989 Journal of Sound and Vibration 134, 259–290. Non-linear
frequency response characteristics of an impact pair.
29. D. F. H 1963 Journal of Basic Engineering 85, 243–246. Squeeze films for rectangular plates.
30. W. S. G, H. H. R and S. Y 1966 Journal of Basic Engineering 88,
451–456. A study of fluid squeeze-film damping.
31. J. E. G 1963 Nonlinear Automatic Control. New York: McGraw-Hill.
32. P.-T. D. S and W. D. I 1978 International Journal of Non-linear Mechanics 13, 71–78.
On the existence and uniqueness of solutions generated by equivalent linearization.
33. A. G and W. V-V 1968 Multiple-input Describing Functions and Nonlinear Systems
Design. New York: McGraw-Hill.
34. A. U and L. O. C 1984 IEEE Transactions on Circuits and Systems CAS-31(9),
766–778. Frequency-domain analysis of nonlinear circuits driven by multi-tone signals.

APPENDIX A: ORIFICE PROPERTIES

A method of estimating orifice inertia, I0 , and damping, B0 , based on resonant
frequency, vr2 , and slope of the elastic rate plot at resonance, mr2 , is presented. The starting
528 . .   .
point is equation (14), which provides an approximate expression for dynamic stiffness.
It is convenient to define a natural frequency and damping ratio as
vn = zKb /I0 , z = B0 /2zKb I0 . (A1)
The frequency response may then be written in terms of real (elastic rate) and imaginary
(damping) parts,

$ %
av 4 + (4z 2 − a)vn2v 2
Kdyn (jv) = Kr + Ap2Kb
(v 2 − vn2 )2 + 4z 2vn2v 2

$ %
2zvn2 − (1 − a)v 2
+jv Br + Ap2Kb 2zvn , (A2)
(v 2 − vn2 )2 + 4z 2vn2v 2
where a = 1 − Ad /Ap . An expression for vr2 is found by maximizing the damping term in
equation (A2) with respect to frequency:

vr22 = vn2 [1 − z4(1 − a)z 2 + a 2 ]/(1 − a). (A3)

It is now helpful to define b as

b = [1 − z4(1 − a)z 2 + a 2 ]/(1 − a). (A4)

An expression for mr2 is found by differentiating the elastic rate term in equation (A2) with
respect to frequency, and evaluating at v = vr1 :
2Ap2Kb
mr2 =
vr2

$ %
(2ab 2 + b(4z 2 − a))((1 − b)2 + 4z 2b) − 2(b 2 − b(2z 2 − 1))(ab 2 + b(4z 2 − a))
× .
((1 − b)2 + 4z 2b)2
(A5)
Equation (A5) may be solved for z using a Newton–Raphson technique; subsequently,
equation (A3) may be solved for vn . Determination of I0 and B0 is then straightforward
using equations (A1).