This action might not be possible to undo. Are you sure you want to continue?

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

**MODELLING OF A HYDR AULIC ENGINE MOUNT
**

FOCUSING ON R ESPONSE TO SINUSOIDAL AND

COMPOSITE EXCITATIONS

J. E. Coic:1r, C.-T. Cn:Nc, Y.-C. Cnioi, W. K. Lii :Nb L. M. Krrr

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

¡. r. coic:1r r1 :i. 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.

n.br:iiic rNciNr xoiN1 505

2. HYDRAULIC ENGINE MOUNT CHARACTERISTICS

2.1. in.sic:i cn:r:c1rris1ics

In Figure 1, x

e

(t) and x

c

(t) represent the displacement of the engine and the

chassis, respectively. The relative displacement, x

0

(t) =x

e

(t) −x

c

(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. rrrqirNc. rrsioNsr cn:r:c1rris1ics

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

¡. r. coic:1r r1 :i. 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 X

0

(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:

K

dyn

(jv) =F(jv)/X

0

(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].

n.br:iiic rNciNr xoiN1 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.

¡. r. coic:1r r1 :i. 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:

x

0

(t) =x

b

(t) +x

s

(t), (2)

where

x

b

(t) =1·0 sin (2pv

b

t) mm, and x

s

(t) =0·1 sin (2pv

s

t) mm.

The ‘‘base frequency’’, v

b

of the large amplitude component is ﬁxed for a given data set

at 5, 10 or 15 Hz. The frequency, v

s

, 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, v

b

and v

s

, 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:

K

b

dyn

(jv) =F(jv)/X

b

(jv), K

s

dyn

(jv) =F(jv)/X

s

(jv). (3)

F(jv), X

b

(jv) and X

s

(jv) are Fourier transforms of their respective time domain

records. It can be shown that transfer functions computed in this way are optimum

n.br:iiic rNciNr xoiN1 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

speciﬁcally in the response to small amplitude inputs, only the value of K

s

dyn

(jv)

corresponding to the excitation frequency v

s

is retained. The coherence associated with

this value approaches unity.

The results are quite interesting. They indicate that K

s

dyn

(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. xobri rrqiirrxrN1s

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

¡. r. coic:1r r1 :i. 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. i:rcr :xiii1ibr xobri

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

&

x˙

r

V

b

Q

t'

=

&

0

0

0

0

0

K

b

/I

t

0

−1

−B

t

/I

t'&

x

r

V

b

Q

t'

+

&

1

A

p

0

'

x˙

0

, (4)

F=[K

r

A

p

K

b

0]

&

x

r

V

b

Q

t'

+[B

r

]x˙

0

.

The states are shear displacement of the primary rubber (x

r

), bulge displacement of the

primary rubber (V

b

) and volumetric ﬂow in the inertia track (Q

t

).

Parameter values were obtained as follows. A

p

, 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, K

b

, track inertia, I

t

, and track damping, B

t

, 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.

n.br:iiic rNciNr xoiN1 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 K

b

, I

t

and B

t

, as well as primary rubber shear

stiﬀness, K

r

, and damping, B

r

, were obtained from the frequency responses of the specialty

mounts.

The specialty mount containing no ﬂuid gives a direct measure of K

r

and B

r

. 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 K

b

, I

t

and B

t

. 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:

K

dyn

(s) =

F(s)

X(s)

=K

r

+B

r

s +A

2

p

K

b

I

t

s

2

+B

t

s

I

t

s

2

+B

t

s +K

b

. (5)

It is convenient to deﬁne a natural frequency and damping ratio associated with the ﬂuid

part of the system:

v

n

=zK

b

/I

t

, z =B

t

/2 zK

b

I

t

. (6)

¡. r. coic:1r r1 :i. 512

The frequency response may then be written in terms of real (elastic rate) and imaginary

(damping) parts:

K

dyn

(jv) =

$

K

r

+A

2

p

K

b

v

2

(v

2

−v

2

n

) +4z

2

v

2

n

v

2

(v

2

−v

2

n

)

2

+4z

2

v

2

n

v

2

%

+jv

$

B

r

+A

2

p

K

b

2zv

3

n

(v

2

−v

2

n

)

2

+4z

2

v

2

n

v

2

%

, (7)

At frequencies well above v

n

(q30 Hz), the elastic rate approaches K

r

+A

2

p

K

b

. Because

K

r

and A

p

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, v

r1

, at which the peak in damping occurs, and the slope of

the elastic rate curve at resonance, m

r1

(see also Figure 7). These measures may be used

to estimate z and v

n

, from which I

t

and B

t

can be determined. An expression for v

r1

is

found by maximizing the damping term in equation (7):

v

2

r1

=v

2

n

(1 −2z

2

). (8)

An expression for m

r1

is found by diﬀerentiating the elastic rate term in equation (7), and

evaluating at v=v

r1

:

m

r1

=(A

2

p

K

b

/2v

r1

)((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.

n.br:iiic rNciNr xoiN1 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 v

r1

and m

r1

, equation (9) may be used to estimate z: subsequently,

equation (8) may be used to estimate v

n

. Determination of I

t

and B

t

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

e

K

(v) =[K (v) −K(v)]/max

v

(K(v)), (10)

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. sx:ii :xiii1ibr xobri

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, I

0

, while the ﬂuid shear layer

between this inertia and the oriﬁce plate is treated as a lumped damper, B

0

. The pressure

drop between upper and lower chambers (K

b

V

b

, where V

b

is the bulge displacement of the

Figure 9. The oriﬁce plate model.

¡. r. coic:1r r1 :i. 514

primary rubber) drives the oriﬁce ﬂow, Q

0

. A pressure balance gives

K

b

V

b

=I

0

Q

0

+B

0

(Q

0

−A

d

x˙

c

). (11)

A

d

is the cross-sectional area of the oriﬁce, and x

c

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=K

r

(x

e

−x

c

) +B

r

(x˙

e

−x˙

c

) +(A

p

−A

d

)K

b

V

b

+(A

2

d

B

0

)(Q

0

/A

d

−x˙

c

), (12)

where x

e

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 (x

e

and

x

c

) rather than a single input (x

0

=x

e

−x

c

). 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 (x

c

=0). In this case, state and output equations are

G

G

G

K

k

x˙

r

V

b

Q

t

Q

0

G

G

G

L

l

=G

G

G

K

k

0

0

0

0

0

0

K

b

/I

t

K

b

/I

0

0

−1

−B

t

/I

t

0

0

−1

0

−B

0

/I

0

G

G

G

L

l

G

G

G

K

k

x

r

V

b

Q

t

Q

0

G

G

G

L

l

+G

G

G

K

k

1

A

p

0

0

G

G

G

L

l

x˙

0

, (13)

F=[K

r

(A

p

−A

d

)K

b

0 A

d

B

0

]G

G

G

K

k

x

r

V

b

Q

t

Q

0

G

G

G

L

l

+[B

r

]x˙

0

.

All variables retain previous deﬁnitions. The only parameters which did not appear in the

large amplitude model re A

d

, I

0

and B

0

.

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

(v

2

1120 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):

K

dyn

(s) =

F(s)

X(s)

=K

r

+B

r

s +A

2

p

K

b

(1 −e)I

0

s

2

+B

0

s

I

0

s

2

+B

0

s +K

b

, (14)

where e =A

d

/A

p

. Equation (14) may be used to develop expressions for the resonant

frequency (v

r2

) and slope of the elastic rate plot at resonance (m

r2

). These expressions

may be equated to values measured using the speciality mount with no decoupler, and

solved to yield estimates of I

0

and B

0

. 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.

n.br:iiic rNciNr xoiN1 515

Figure 10. The small amplitude model.

I

t

and B

t

, 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.

¡. r. coic:1r r1 :i. 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. iircrvisr iiNr:r xobri

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, x

d

) 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 =x

d

= Qd,

integrate small amplitude state equations; if =x

d

= ed and sign (V

b

) =sign (x

d

), set Q

0

=0,

set x

d

=sign (x

d

)d, and integrate large amplitude state equations.

If the previous step was integrated with the large amplitude model: if sign (V

b

) =sign

(x

d

), integrate large amplitude state equations; if sign (V

b

) $sign (x

d

), integrate small

amplitude state equations.

Note that V

b

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

X

0

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

n.br:iiic rNciNr xoiN1 517

Figure 12. (a) Normalized elastic rate and (b) damping of piecewise 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.

should be noted here that limited insight into bottoming out is a signiﬁcant weakness of

the piecewise linear model.

4.2. rqiiv:irN1 iiNr:riz:1ioN

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., x

0

(t)=

X

0

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

¡. r. coic:1r r1 :i. 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]:

F

sq

=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.

n.br:iiic rNciNr xoiN1 519

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

relation is

F

sq

(x

d

, x˙

d

) =g

$

1

(d −x

d

)

3

+

1

(d +x

d

)

3

%

x˙

d

, F

sq

(x¯, x¯

.

) =bd

$

1 +3x¯

2

(1 +x¯

2

)

3

%

x¯

.

, (16a, b)

where b is a geometric constant with units of damping and x¯ =x

d

/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

x¯ =a sin vt. Then the force that results from equation (16b) can be represented by a

Fourier series:

F

sq

(a sin vt, av cos vt) =F

0

+K

eq

da sin vt +B

eq

dav cos vt+higher harmonics. (17)

It is easily shown that the constant term (F

0

) and in-phase component of the fundamental

(K

eq

da) are zero. Ignoring higher order harmonics, the equivalent behavior is that of a

viscous damper with an amplitude-dependent coeﬃcient,

B

eq

(a) =

b

p

g

2p

0

cos

2

u(1 +3a

2

sin

2

u)

(1 −a

2

sin

2

u)

3

du, (18)

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 x¯. 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) ———.

¡. r. coic:1r r1 :i. 520

the equivalent damping approaches inﬁnity as the normalized decoupler amplitude, a,

approaches 1.

If B

0

f(x) in Figure 15 is replaced with B

eq

, a transfer function may be found relating

X(jv) to X

0

(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:

a(B

eq

, X

0

, v) =

X

d

=

K

b

z(K

b

−I

0

v

2

)

2

+(B

eq

v)

2

X

0

d

. (19)

This relation has been plotted in Figure 16 for a single value of X

0

and several frequencies.

Given speciﬁc values for X

0

and v, equations (18) and (19) may be considered simultaneous

equations to be solved for a and B

eq

. 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 =B

0

(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.

n.br:iiic rNciNr xoiN1 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. 1nr rrrrc1 or nicnrr n:rxoNics

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,

¡. r. coic:1r r1 :i. 522

Figure 19. Relative errors of equivalent 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.

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

F

1

1A

2

p

K

b

X

0

. (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,

F

2

1A

p

K

b

(A

d

(d/n)). (21)

The magnitude of this eﬀect is much smaller than F

1

. For the fundamental, F

2

/F

1

10·1,

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.

n.br:iiic rNciNr xoiN1 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

F

3

1(I

0

A

2

d

)(d/n)(nv)

2

1A

2

p

K

b

e

2

nd(v/v

r2

)

2

. (22)

For n =5 and v=30 Hz, F

3

/F

1

15 ×10

−4

. Thus, because the inertia of the decoupler is

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

small.

4.4. sixiiirirb rorx or rqiiv:irN1 b:xiiNc

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 B

eq

(a), shown in Figure 16, was investigated. Here, the equivalent

damping is set to B

0

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. coxiosi1r rxci1:1ioNs

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 v

s

, 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 v

b

and v

s

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.

¡. r. coic:1r r1 :i. 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)

x¯ =a

b

sin (v

b

t +f

b

) +a

s

sin (v

s

t +f

s

), (23)

where a

b

+a

s

Q1. The equivalent linear behavior can be found using a double Fourier

series [33], which has the form

6F

sq

= s

a

m=0

s

a

n =−a(m$0)

n =0(m=0)

[P

mn

sin (mu

b

+nu

s

) +Q

mn

cos (mu

b

+nu

s

)], (24)

where u

b

=v

b

t +f

b

, u

s

=v

s

t +f

s

. Note that many frequencies other than har-

monics appear in the Fourier series. For instance, output frequencies of v

s

2v

b

,

v

s

22v

b

,..., 2v

s

2v

b

, 2v

s

22v

b

,... can be expected. These terms are often called

‘‘combination tones’’ [34]. If the ratio of v

s

to v

b

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.

n.br:iiic rNciNr xoiN1 525

It is readily shown that Q

00

, P

10

and P

01

are all zero, while Q

10

and Q

01

are non-zero and

are related to the equivalent damping at v

b

and v

s

, respectively:

B

b

(a

b

, a

s

) =

Q

10

a

b

v

b

=

1

2p

2

g

p

−p

g

p

−p

(cos u

b

+

a

s

v

s

a

b

v

b

cos u

s

)

×

1 +3(a

b

sin u

b

+a

s

sin u

s

)

2

(1 −(a

b

sin u

b

+a

s

sin u

s

)

2

)

3

cos u

b

du

b

du

s

=

1

2p

2

g

p

−p

g

p

−p

cos

2

u

b

(1 +3(a

b

sin u

b

+a

s

sin u

s

)

2

(1 −(a

b

sin u

b

+a

s

sin u

s

)

2

)

3

du

b

du

s

, (25)

B

s

(a

b

, a

s

) =

Q

01

a

s

v

s

=

1

2p

2

g

p

−p

g

p

−p

0

a

b

v

b

a

s

v

s

cos u

b

+cos u

s

1

×

1 +3(a

b

sin u

b

+a

s

sin u

s

)

2

(1 −(a

b

sin u

b

+a

s

sin u

s

)

2

)

3

cos u

s

du

b

du

s

=

1

2p

2

g

p

−p

g

p

−p

cos

2

u

s

(1 +3(a

b

sin u

b

+a

s

sin u

s

)

2

(1 −(a

b

sin u

b

+a

s

sin u

s

)

2

)

3

du

b

du

s

. (26)

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 (B

b

, B

s

, a

b

, a

s

).

A very useful result can be obtained, however, without solving any sets of equations.

It is readily shown by direct computation that B

s

qB

b

whenever a

b

qa

s

. 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, B

b

, 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, v

s

, 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.

¡. r. coic:1r r1 :i. 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, A

2

p

I

t

, 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. Rrsn 1994 Chrysler Corporation. Personal communication.

2. M. BrrNicnoN 1984 SAE Paper 840259. A new generation of engine mounts.

3. S. BoicniiioN, R. Snoirrsni, T. KNirri and P. Scniiir 1989 SAE Paper 891160. Optimal

tuning of adaptive hydraulic engine mounts.

4. R. M. Br:cn and A. G. H:bbrov 1993 SAE Paper 931321. On the dynamic response of

hydraulic engine mounts.

5. M. Ci:ri 1985 SAE Paper 851650. Hydraulic engine mount isolation.

6. P. E. Corcor:N and G.-H. Ticis 1984 SAE Paper 840407. Hydraulic engine mount character-

istics.

n.br:iiic rNciNr xoiN1 527

7. W. C. Fiovrr 1985 SAE Paper 850975. Understanding hydraulic mounts for improved vehicle

noise, vibration and ride qualities.

8. K. K:box:1si 1989 SAE Paper 891138. Hydraulic engine mount for shock isolation at

acceleration on the FWD cars.

9. R. A. Mizrcnii 1984 SAE Paper 840410. Hydraulic mounts—improved engine isolation.

10. R. Snoirrsni, P. L. Gr:r and T. L. Hois1oN 1986 SAE Paper 860549. Adaptive hydraulic

engine mounts.

11. R. SiNcn, G. Kix and P. V. R:viNbr: 1992 Journal of Sound and Vibration 158, 219–243. Linear

analysis of automotive hydro-mechanical mount with emphasis on decoupler characteristics.

12. M. SiciNo and E. Anr 1986 SAE Paper 861412. Optimum application for hydroelastic engine

mount.

13. T. Usni¡ix: and T. D:N 1986 SAE Paper 860550. Nonlinear B.B.A. for predicting vibration

of vehicle with hydraulic engine mount.

14. T. Usni¡ix:, K. T:i:No and H. Ko¡ix: 1988 SAE Paper 880073. High performance hydraulic

mount for improving vehicle noise and vibration.

15. J. P. Wrs1 1987 Automotive Engineer 12, 17–19. Hydraulically-damped engine mounts.

16. P. L. Gr:r, D. Novo1N., R. Snoirrsni, J. S1:rir. and R. W. S1rvrNs 1988 SAE Paper

880074. Active frame vibration control for automotive vehicles with hydraulic engine mount.

17. G. Kix and R. SiNcn 1993 in Advanced Automotive Technologies (M. Ahmadian, editor).

247–255. New York: ASME. A broadband adaptive hydraulic mount system.

18. R. Snoirrsni, P. L. Gr:r, S. BoicniiioN, T. KNirri and R. W. S1rvrNs 1988 SAE Paper

880075. Open-loop versus closed-loop control for hydraulic engine mounts.

19. G. Kix and R. SiNcn 1992 in Transportation Systems—1992, 165–180. New York, ASME.

Resonance, isolation and shock control characteristics of automotive nonlinear hydraulic engine

mounts.

20. G. Kix and R. SiNcn 1993 Journal of Dynamic Systems, Measurement and Control 115, 482–487.

Nonlinear analysis of automotive hydraulic engine mount.

21. C. T. Cn:Nc 1992 Master’s Thesis, Northwestern University. Dynamic analysis of hydraulic

engine mounts.

22. J. S. BrNb:1 and A. G. Pirrsoi 1986 Random Data: Analysis and Measurement Procedures.

New York: John Wiley.

23. D. K:rNoii and R. RosrNnrrc 1975 System Dynamics: A Uniﬁed Approach. New York: John

Wiley.

24. H. M. P:.N1rr 1961 Analysis and Design of Engineering Systems. Cambridge, Massachusetts:

MIT Press.

25. Y. C. Cnioi 1993 Master’s Thesis, Northwestern University. Computer simulation of hydro-

elastic engine mounts.

26. G. F. Fr:NiiiN, J. D. Povrii and M. L. Worix:N 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. Coxi:riN and R. SiNcn 1989 Journal of Sound and Vibration 134, 259–290. Non-linear

frequency response characteristics of an impact pair.

29. D. F. H:.s 1963 Journal of Basic Engineering 85, 243–246. Squeeze ﬁlms for rectangular plates.

30. W. S. GrirriN, H. H. Ricn:rbsoN and S. Y:x:N:xi 1966 Journal of Basic Engineering 88,

451–456. A study of ﬂuid squeeze-ﬁlm damping.

31. J. E. GinsoN 1963 Nonlinear Automatic Control. New York: McGraw-Hill.

32. P.-T. D. Si:Nos and W. D. Iv:N 1978 International Journal of Non-linear Mechanics 13, 71–78.

On the existence and uniqueness of solutions generated by equivalent linearization.

33. A. Grin and W. V:Nbrr-Vribr 1968 Multiple-input Describing Functions and Nonlinear Systems

Design. New York: McGraw-Hill.

34. A. Usnib: and L. O. Cni: 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, I

0

, and damping, B

0

, based on resonant

frequency, v

r2

, and slope of the elastic rate plot at resonance, m

r2

, is presented. The starting

¡. r. coic:1r r1 :i. 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

v

n

=zK

b

/I

0

, z =B

0

/2zK

b

I

0

. (A1)

The frequency response may then be written in terms of real (elastic rate) and imaginary

(damping) parts,

K

dyn

(jv) =

$

K

r

+A

2

p

K

b

av

4

+(4z

2

−a)v

2

n

v

2

(v

2

−v

2

n

)

2

+4z

2

v

2

n

v

2

%

+jv

$

B

r

+A

2

p

K

b

2zv

n

2zv

2

n

−(1 −a)v

2

(v

2

−v

2

n

)

2

+4z

2

v

2

n

v

2

%

, (A2)

where a =1 −A

d

/A

p

. An expression for v

r2

is found by maximizing the damping term in

equation (A2) with respect to frequency:

v

2

r2

=v

2

n

[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 m

r2

is found by diﬀerentiating the elastic rate term in equation (A2) with

respect to frequency, and evaluating at v=v

r1

:

m

r2

=

2A

2

p

K

b

v

r2

×

$

(2ab

2

+b(4z

2

−a))((1 −b)

2

+4z

2

b) −2(b

2

−b(2z

2

−1))(ab

2

+b(4z

2

−a))

((1 −b)

2

+4z

2

b)

2

%

.

(A5)

Equation (A5) may be solved for z using a Newton–Raphson technique; subsequently,

equation (A3) may be solved for v

n

. Determination of I

0

and B

0

is then straightforward

using equations (A1).

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 amplitudedependent 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.

Frequency/amplitude ranges of greatest interest include [6. For larger amplitude excitations. and is introduced for the purpose of control. 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.’’ The inertia track (4) is a lengthy spiral channel that enables ﬂuid to pass from the upper chamber to the lower chamber. 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. The secondary rubber (2) is a rubber septum that serves principally to contain the ﬂuid. or during extreme accelerations on smooth surfaces. respectively. a point which will be highlighted in this paper. 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). The frequency response is typically evaluated with a conventional servo-controlled hydraulic test rig. xe (t) and xc (t) represent the displacement of the engine and the chassis. ethylene glycol. HYDRAULIC ENGINE MOUNT CHARACTERISTICS 505 2. and require good isolation. 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. 2. while driving on rough surfaces. but are beyond the scope of this paper. the inertia track acts as a tuned damper. have received attention recently. while the engine bracket is sinusoidally excited at a ﬁxed amplitude. 0.05–0. Interest has also arisen in the extent to which hydraulic mounts can provide control and isolation simultaneously [14]. x0 (t) = xe (t) − xc (t). for small amplitude excitation. is referred to as the ‘‘excitation’’ of the moment. The ﬂuid inertia in this channel is signiﬁcant. The behavior of an engine mount is usually reported in terms of its frequency response for diﬀerent amplitude excitations. and will be considered in this paper. The relative displacement. (2) 25–250 Hz. 0. Cast in the oriﬁce plate are the ‘‘inertia track’’ and ‘‘decoupler oriﬁce. It also contributes modestly to the elastic rate. Thus. and is usually selected so that it experiences resonance at the natural frequency of the engine/mount system. 2. completely ﬁlls the interior of the mount.5–5. The ﬂuid (6). which eﬀectively short-circuits the inertia track. To analyze high frequency behavior (5–200 Hz increments). time domain records all consist of 8192 points collected at a sample interval of 0·0005 s. For this study.2. This may be important. In Figure 1. which may result from combustion noise [14]. for instance. The four transforms are then ensemble-averaged to obtain estimates of the Fourier transform . 14]: (1) 5–15 Hz.1. The bulge rate of the primary rubber (ratio of pressure change to volume change) is also an important design parameter. Thus. 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. It support the static load of the engine. most of the ﬂuid transport between chambers is via the decoupler oriﬁce. The chassis bracket is ﬁxed to an inertially grounded force sensor.5 mm— these excitations can cause noise and vibration. The primary rubber (1) is a rubber cone that serves several purposes. the response to composite inputs is of interest. Physical components which contribute signiﬁcantly to the dynamics of the mount include the following. The damping of the track is also signiﬁcant. 11. Even higher frequency excitations. each record is broken into four contiguous sections which are independently windowed (Hanning window) and transformed. Thus.0 mm—these excitations are in the range of engine resonance and large enough to require signiﬁcant damping.

. K(v). and coherence. 0·1 mm) and large amplitude (—— . . is termed the ‘‘elastic rate’’. then ensemble-averaged. To analyze low frequency behavior (1–40 Hz in 1 Hz increments). Figure 2. correspond to the fundamental harmonics of force and displacement. E † Computational details are given in reference [21].506 . while the imaginary part divided by frequency.1 and 1. is termed the ‘‘damping’’. known as the ‘‘dynamic stiﬀness’’ is the principal quantity of interest: Kdyn (jv) = F(jv)/X0 (jv) = K(v) + jvB(v). as all comparable data in this paper. it should be noted that dynamic stiﬀness is often presented in terms of the magnitude (‘‘dynamic rate’’) and phase (‘‘loss angle’’). have been normalized by the stiﬀness and damping of the primary rubber (which are assumed to be amplitude and frequency-independent). v. 1·0 mm) excitations. (1) The real part of the dynamic stiﬀness. (a) Normalized elastic rate and (b) damping of a production hydraulic mount for small amplitude (—w—. . only those data corresponding to the excitation frequency.0 mm excitations. These data. Coherences obtained in this way are not reported because they are in all cases extremely close to unity. The ratio of these transforms. Figure 2 shows representative data corresponding to the two classes of excitation considered above. In comparing these data with others in the literature. B(v). The Fourier transforms. It is clear that the response of the hydraulic mount is strongly amplitude-dependent.† In all cases. which are independently windowed and transformed. The two curves correspond to 0. F(jv) and X0 (jv). are retained. each record is broken into eight interleaved sections (each with an eﬀective sample rate of 0.004 s).

E 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. the behavior shifts from the prototypical ‘‘small amplitude’’ response to the prototypical ‘‘large amplitude’’ response. the mount serves as a tuned damper. –·–·–·–. Figure 4 displays frequency responses of three specially prepared engine mounts. with increasing amplitude. 0·8 mm. The development of a model which captures this shift is a principal objective of this work. (2) for frequencies less than 050 Hz. but will introduce a certain degree of systematic error. 507 Figure 3. quite similar to the small amplitude response of a production mount. ——. 0·6 mm. The second specialty mount was assembled without a decoupler. —+—. The behavior resembles the large amplitude response of a production mount. 0·05 mm. For this study. (3) the peak in damping for small amplitude excitation (0120 Hz) and the corresponding increase in elastic rate are the result of decoupler inertia. thus. The data show that the assumption of amplitude independence is very good. —— . a more extensive set of data has been collected. It is evident that. The ﬁrst is a mount from which the ﬂuid was drained. resulting in superior vibration isolation. . while the assumption of frequency independence is reasonable. not surprisingly. These data are particularly useful in identifying model parameters. The behavior is. The third specialty mount had the decoupler ﬁxed in place. ——. 1·0 mm. leaving the primary rubber solely responsible for the dynamic behavior. 0·2 mm. for large amplitude inputs. These data are plotted in Figures 3–5. and are undesirable eﬀects. ––––. both the elastic rate and damping are much less for small amplitude inputs. 0·4 mm. (a) Normalized elastic rate and (b) damping of a production hydraulic mount for various amplitudes: —q—. 0·1 mm. Figure 3 shows the response of the mount to diﬀerent amplitude excitations.

It can be shown that transfer functions computed in this way are optimum . s Kdyn (jv) = F(jv)/Xs (jv). Figure 4. the relationship between the small amplitude input and the force is of particular interest. are generated by separate analog instruments. This allows the computation of transfer functions relating each of these inputs to the output force: b Kdyn (jv) = F(jv)/Xb (jv). mount with no ﬂuid. 10 or 15 Hz. . Xb (jv) and Xs (jv) are Fourier transforms of their respective time domain records. (a) Normalized elastic rate and (b) damping of three specially prepared engine mounts: —q—. —— . mount with ﬁxed decoupler. 1·0 mm. . —— . Toward this end. The two frequencies. These data have been collected to assess the mount’s capacity to provide isolation in the presence of a large amplitude disturbance. 0·05 mm. The frequency. vb of the large amplitude component is ﬁxed for a given data set at 5. where xb (t) = 1·0 sin (2pvb t) mm. — + —.508 . Figure 5 displays frequency response of a production mount excited by composite waveforms: x0 (t) = xb (t) + xs (t). and xs (t) = 0·1 sin (2pvs t) mm. of the small amplitude component is swept from 20 to 200 Hz. (3) F(jv). vb and vs . mount with no decoupler. vs . (2) The ‘‘base frequency’’. mount with no ﬂuid. 1·0 mm. and therefore may be considered independent and uncorrelated [22]. W E 0·05 mm.

low frequency disturbance. high frequency inputs is signiﬁcantly degraded by the simultaneous presence of a large amplitude. 2. E Q 1·0 mm single frequency response. small amplitude resonance (which. 15 Hz) and a 0·1 mm sinusoid (frequency on horizontal axis): —q—. 5 Hz base. ﬁgures prominently in the mount’s performance as an isolator). large amplitude resonance (which describes the mount’s performance as a tuned damper). only the value of Kdyn (jv) corresponding to the excitation frequency vs is retained. The coherence associated with this value approaches unity. 10 Hz base. —— . [14] who performed a similar experiment. 15 Hz base. linear approximations to the underlying non-linear system [22]. The data described above give a more complete picture of hydraulic mount behavior than previously available. Based on them. A similar conclusion was reached by Ushijima et al. the high frequency. 509 Figure 5. 0·1 mm single frequency response. and . (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. 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). —+—. it seems reasonable to assert that a good mount model should be able to capture: the low frequency. 10. This implies that the mount’s capacity to provide isolation from small amplitude. —w—. s The results are quite interesting.3. though undesirable. the amplitude dependence of elastic rate and damping. —— . Because we are interested s speciﬁcally in the response to small amplitude inputs.

and the ﬂuid inertia and damping in the upper chamber are ignored (inertia and damping in the track are much greater). State equations and an output equation for the reaction force may be derived from the bond graph. and makes the assumption that the decoupler is ‘‘bottomed out’’ at all times. The interconnection of these elements is straightforward—both bond graph [23. linear time-invariant elements. LARGE AND SMALL AMPLITUDE LINEAR MODELS In this section. 7. The principal physical eﬀects are taken to be those associated with the primary rubber (shear stiﬀness and damping. The stiﬀness of the secondary rubber is small enough that it can be ignored (reducing the requisite state dimension by one). . 24] and mechanical equivalent models are shown in Figure 6. 5. These models embody our basic physical understanding of the hydraulic mount. Both models assume that all other important physical eﬀects may be represented by lumped. Section 4 presents two approaches to estimating non-linear frequency responses. 3. bulge stiﬀness. and makes the assumption that the decoupler never bottoms out. The ﬁrst is tailored to large amplitude sinusoidal excitation (q0·5 mm). 3. Similar models have been presented in references [2. The second is tailored to small amplitude excitation (Q0·5 mm). a piecewise linear simulation and an equivalent linearization analysis. The next section introduces two linear models. Section 5 presents a summary discussion. 11. two linear models are introduced.† The ﬂuid is assumed incompressible (the bulge compliance of the primary rubber is much greater). These techniques are also extended to predict the mount’s response to composite inputs. the mount’s response to composite inputs (which measures the mount’s ability to provide simultaneous isolation and control).1. and are suﬃcient to capture the two resonances. piston area) and with the inertia track (ﬂuid inertia and damping). They are &' & xr ˙ 0 Vb = 0 . .510 . 14].

Qt .

Parameter values were obtained as follows. Bt . It . These formulas. which is easily measured. both rubber shear stiﬀness and secondary rubber stiﬀness should contribute to the elastic rate of the latter class. 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. Kb . track inertia. however. the ‘‘piston area’’ of the primary rubber. '& ' & ' &' xr 1 Vb + Ap x0 . At low frequency. Initial estimates of bulge stiﬀness. Ap . is equal to the area of the oriﬁce plate top surface. ˙ Qt 0 (4) . and track damping. bulge displacement of the primary rubber (Vb ) and volumetric ﬂow in the inertia track (Qt ). [11]. ˙ Qt The states are shear displacement of the primary rubber (xr ). yet. were made using the analytical formulas given in Singh et al. are necessarily † This may be understood by comparing the response of the mount without ﬂuid to the responses of the ﬂuid-ﬁlled mounts. 0 0 0 Kb /It 0 −1 −Bt /It F = [Kr Ap Kb xr 0] Vb + [Br ]x0 . these mounts exhibit nearly the same values as the dry mount (Figure 4).

I t s + Bt s + Kb X(s) (5) It is convenient to deﬁne a natural frequency and damping ratio associated with the ﬂuid part of the system: vn = zKb /It . (6) . even such basic assumptions as the absence of leakage paths tend to short-circuit the inertia track may be questionable. 511 Figure 6. The specialty mount with ﬁxed decoupler should behave very much as the large amplitude model would predict. z = Bt /2 zKb It . It and Bt . representative values are selected and used in all models. In this study. such as straight inertia track and a hemispherical upper chamber. more accurate estimates of Kb . Finally. analytically determined parameters must generally be considered no better than order-of-magnitude estimates. as well as primary rubber shear stiﬀness. These estimates are obtained by relating characteristics of the frequency response measurements to model predictions. based on idealized geometries. Thus. Br . These values are also used to normalize the elastic rate and damping plots. The large amplitude model. Also idealized are the assumptions of laminar and fully developed ﬂow through the inertia track. Though these parameters vary somewhat with frequency and excitation amplitude (Figure 4). Kr . it is useful for estimating Kb . The transfer function associated with the large amplitude model (equation (4)) is: Kdyn (s) = I s 2 + Bt s F(s) 2 = Kr + Br s + Ap Kb 2 t . were obtained from the frequency responses of the specialty mounts. thus. The specialty mount containing no ﬂuid gives a direct measure of Kr and Br . and damping. It and Bt .

. (9) Figure 7. and the slope of the elastic rate curve at resonance. 1·0 mm. 2 (v − v ) + 4z 2vn v 2 2 2 2 n (8) An expression for mr1 is found by diﬀerentiating the elastic rate term in equation (7). In addition to the high frequency elastic rate. .512 . salient and reproducible features of the data are the resonant frequency. from which It and Bt can be determined. These measures may be used to estimate z and vn . Because Kr and Ap are known. mount with ﬁxed decoupler. $ 3 2zvn . . vr1 . An expression for vr1 is found by maximizing the damping term in equation (7): 2 2 vr1 = vn (1 − 2z 2 ). (a) Normalized elastic rate and (b) damping of large amplitude model versus data: ——. and evaluating at v = vr1 : 2 mr1 = (Ap Kb /2vr1 )((1 − 2z 2 )/z 2(1 − z 2 )). at which the peak in damping occurs. The frequency response may then be written in terms of real (elastic rate) and imaginary (damping) parts: 2 Kdyn (jv) = Kr + Ap Kb $ 2 2 v 2(v 2 − vn ) + 4z 2vn v 2 2 2 2 2 2 2 (v − vn ) + 4z vn v % % (7) 2 +jv Br + Ap Kb 2 At frequencies well above vn (q30 Hz). –—–. large amplitude model. mr1 (see also Figure 7). the elastic rate approaches Kr + Ap Kb . this measure of elastic rate may be used to estimate bulge stiﬀness.

while the ﬂuid shear layer between this inertia and the oriﬁce plate is treated as a lumped damper. v (10) where K(v) is the measured elastic rate and K (v) is the predicted elastic rate. . I0 . equation (9) may be used to estimate z: subsequently. 513 Figure 8. completing the parameter identiﬁcation process. while those in damping are less than 20%.2. The relative error in damping is deﬁned analogously. Relative errors of large amplitude and small amplitude linear models: ——. Determination of It and Bt is then straightforward. The pressure drop between upper and lower chambers (Kb Vb . The decoupler itself and surrounding ﬂuid are treated as a lumped inertia. 3. Figure 9 illustrates the model of decoupler oriﬁce dynamics which is used. 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%. B0 . A quantitative measure of model performance is the frequency-dependent relative error. (a) Elastic rate. The frequency response of the identiﬁed large amplitude model is shown in Figure 7. large amplitude. which for elastic rate will be deﬁned as eK (v) = [K (v) − K(v)]/max (K(v)). 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). ––––. The oriﬁce plate model. small amplitude. (b) damping. equation (8) may be used to estimate vn . Given measurements of vr1 and mr1 . where Vb is the bulge displacement of the Figure 9.

.514 . . primary rubber) drives the oriﬁce ﬂow. . Q0 . A pressure balance gives Kb Vb = I0 Q0 + B0 (Q0 − Ad xc ).

In this case. state and output equations are ˙ 0 0 K xr L K0 0 LK xr L K 1 L GV G G0 GGV G GA G 0 −1 .2 which ﬁxes the chassis bracket to ground (xc = 0). The eﬀect of oriﬁce dynamics. to which the oriﬁce plate is ﬁxed. ˙ (11) Ad is the cross-sectional area of the oriﬁce. and the shear forces due to drag: 2 F = Kr (xe − xc ) + Br (xe − xc ) + (Ap − Ad )Kb Vb + (Ad B0 )(Q0 /Ad − xc ). but is quite non-intuitive. as represented in equations (11) and (12). it does occur in the test ﬁxture described in section 2. 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 ). A mechanical equivalent to this bond graph can be found. 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. Contributions are the forces transmitted via rubber shear. 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. the normal forces on the oriﬁce plate due to pressure.† While no such instance occurs in an automobile. An expression for the force transmitted to the chassis can also be derived with the assistance of Figure 9. ˙ ˙ ˙ (12) where xe is the displacement of the engine. A complete small amplitude model is shown in Figure 10. and xc is the displacement of the chassis.

b −1 G Q G= G0 K /I −B /I GG QbG+ G 0p x0 . G˙ .

0 t t GQ t G G0 K b /I t GGQ t G G 0 G k .

These expressions may be equated to values measured using the speciality mount with no decoupler. 0l k 0 −B0 /I0lk 0l k l b 0 (13) F = [Kr (Ap − Ad )Kb K xr L GV G 0 Ad B0 ]G bG+ [Br ]x0 . X(s) I 0 s 2 + B 0 s + Kb (14) where e = Ad /Ap . it is ﬁrst assumed that. ˙ G Qt G kQ0l All variables retain previous deﬁnitions. The decoupler cross-sectional area is easily measured. This simpliﬁcation gives the dynamic stiﬀness of the small amplitude model a form very similar to equation (5): Kdyn (s) = F(s) (1 − e)I0 s 2 + B0 s 2 = Kr + Br s + Ap Kb . Oriﬁce inertia and damping can be estimated by a procedure similar to that used to ﬁnd track inertia and damping. I0 and B0 . Equation (14) may be used to develop expressions for the resonant frequency (vr2 ) and slope of the elastic rate plot at resonance (mr2 ). ﬂow through the inertia track can be neglected. . The only parameters which did not appear in the large amplitude model re Ad . 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. To simplify the analysis. and solved to yield estimates of I0 and B0 . at the frequency of oriﬁce resonance (v2 1 120 Hz).

Details are given in Appendix A. mount with no decoupler. but somewhat more involved due to the presence of e. The small amplitude model. and are comparable to those of the large Figure 11. 0·05 mm. Relative errors are shown in Figure 8. . The frequency response of the fully identiﬁed small amplitude model is shown in Figure 11. It and Bt . (a) Normalized elastic rate and (b) damping of small amplitude model versus data: ——. small amplitude model: ––––. 515 Figure 10.

nor do the (possibly very fast) dynamics of the bottoming out process need to be directly considered. This is addressed further below. attention will be focused on the decoupler. excitations of arbitrary shape are easily accommodated. due both to noise in the measured values and to poorness of ﬁt. Note that Vb is linearly related to the pressure in the upper chamber. integrate small amplitude state equations. leaving a record of 1024 points.1. two non-linear models are introduced. Frequency responses have been computed as follows. NON-LINEAR MODELLING A variety of non-linear eﬀects contributes signiﬁcantly to the behavior of a production hydraulic mount [20]. however. The beginning of each record is cut oﬀ to remove the transient response.516 . A simulation time increment of 2p/v/64 seconds is picked. Records include position input and force output information. In this paper. A signiﬁcant disadvantage. xd ) whenever the decoupler is not bottomed out. integrate small amplitude state equations. A complex transfer function is computed at the frequency of excitation. if =xd = e d and sign (Vb ) = sign (xd ). which is principally responsible for the amplitude dependence. Results are shown in Figure 12 (which may be compared with Figure 3). 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. they cannot represent in any greater detail the amplitude dependence of the mount’s response. so that an eﬃcient simulation can be performed. set Q0 = 0. if sign (Vb ) $ sign (xd ). which are Hanning windowed and transformed with an FFT. 4. the simulation is quite accurate. magnitude model. which determines whether a decoupler at the limits of its travel will remain bottomed out or not. These include. The behavior of the model can be investigated using numerical integration. Generally speaking. set xd = sign (xd )d. and relative errors in Figure 13. entrance and exit eﬀects in the inertia track ﬂow. Although the linear models successfully capture the two resonances. No new dynamic elements need be introduced to account for bottoming out. 25]. To meet this objective. It is suspected that the poor ﬁt near resonance is due to the lack of an appropriate model of the bottoming out process. A direct approach to incorporating decoupler behavior is via a piecewise linear model [19. If the previous step was integrated with the large amplitude model: if sign (Vb ) = sign (xd ). but it . and integrate large amplitude state equations. A sinusoidal excitation of amplitude X0 and frequency v is assumed. however. and to the large amplitude model when the decoupler is bottomed out. the piecewise linear model reduces to the small amplitude linear model (augmented with a state for decoupler position. . and amplitude-dependent softening of the primary rubber. A piecewise linear model has several attractive features. the linear continuous state equations can be directly mapped to diﬀerence equations which are guaranteed stable [26]. Moreover. In essence. . Finally. 4. Relative errors in elastic rate and damping are typically less than 20%. for instance. integrate large amplitude state equations. The greatest errors occur in the vicinity of the high frequency resonance. This record is broken into three overlapping sections of 512 points each. and time domain records of approximately 1150 points are computed. is that a tremendous amount of time domain simulation data must be generated to produce a modest amount of frequency domain data.

in eﬀect. – – – –. There is some physical justiﬁcation for this model. 0·4 mm. The equivalent linear element(s) are then. The idea of equivalent linearization is. the viscous damper. Equivalent linearization techniques have been applied to similar problems with considerable success (see. —— . 4. Compare with Figure 3: —q—. there is no physical reason to associate . only one element. the simpliﬁed mechanical system in Figure 14 will ﬁrst be considered. The speciﬁc approach taken here is illustrated in Figure 15. 0·8 mm. and with the added beneﬁt of improved physical insight. provides roughly comparable results (to piecewise linear modelling) at a greatly reduced computational cost. ––––. 0·1 mm. ——. x0 (t)= X0 sin vt). As the ﬁgure indicates. 0·05 mm. 0·6 mm. E Q –·–·–·–. 517 Figure 12. equivalent linearization. 0·2 mm. to replace the cage with a linear element or elements that would yield the same motion.2. given a prescribed excitation (e.g.. Note that the mass representing decoupler inertia is constrained by a ‘‘cage’’. —— . (a) Normalized elastic rate and (b) damping of piecewise linear model for various amplitude excitations. is aﬀected by equivalent linearization. for instance. First. should be noted here that limited insight into bottoming out is a signiﬁcant weakness of the piecewise linear model. Dubowsky and Freudenstein [27] and Comparin and Singh [28]). parameterized by the amplitude and frequency of excitation. An alternative approach to non-linear modelling. 1·0 mm. at least up to the fundamental of the Fourier series describing that motion. To introduce the technique.

0·2 mm. Figure 13. ——. – – – –. 0·05 mm. . 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. Relative errors for piecewise linear model: —q—. –·–·–·–. Second. 0·8 mm. (a) Elastic rate. 1·0 mm. —— . E Q 0·4 mm. Thus. 0·6 mm. the decoupler is immersed in a ﬂuid which must be displaced as the cage boundaries are approached. . (b) damping. signiﬁcant energy storage (potential or kinetic) with bottoming out. 0·1 mm.518 . 30]: Fsq = g(h /h 3 ) . a squeeze ﬁlm [29] is developed. ----. —— .

h is the gap thickness and g is a geometric parameter. The implications of this assumption will be reassessed below. The geometry of decoupler/cage interaction is not simple. Because the cage in Figure 14 has two sides. the appropriate damping Figure 14. The simpliﬁed model. For simple geometries. equation (15) is readily derived from the Reynolds equation for viscous ﬂow [30]. (15) Here. . but it has been assumed that the inverse cube form holds nonetheless.

Equivalent damping as a function of normalized decoupler amplitude (equation (18)): ––––. Note that ¯ g 2p 0 cos2 u(1 + 3a 2 sin2 u) du. (1 − a 2 sin2 u)3 (18) Figure 16. E Q W Simpliﬁed form of Beq (a) ———. v=0·2. v=1·0. ¯ (1 + x 2 )3 ¯ % (16a. —— . where v is normalized frequency. $ 1 + 3x 2 . ¯ x. Then the force that results from equation (16b) can be represented by a ¯ Fourier series: Fsq (a sin vt. xd ) = g where b is a geometric constant with units of damping and x = xd /d. v=1·8. relation is ˙ Fsq (xd . assuming the same sinusoidal form for x . ˙ (d − xd )3 (d + xd )3 d % Fsq (x . . To begin. x) = bd ¯ ¯ . b) It is easily shown that the constant term (F0 ) and in-phase component of the fundamental (Keq da) are zero. —e—. assuming ﬁxed excitation amplitude (equation (19)): —— . Ignoring higher order harmonics. av cos vt) = F0 + Keq da sin vt + Beq dav cos vt+higher harmonics. (17) $ 1 1 + x. 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. —q—. 519 Figure 15. assume that x = a sin vt. Eﬀect of equivalent damping on normalized decoupler amplitude. The model which treats the decoupler cage as an amplitude-dependent dissipator. —— . ¯ The non-linear damping relation of equation (16b) may be converted to an equivalent linear damping via a procedure outlined by Gibson [31]. This function is plotted in Figure 16. the equivalent behavior is that of a viscous damper with an amplitude-dependent coeﬃcient. Beq (a) = b p where u = vt. v=1·4. v=0·6.

If B0 f(x) in Figure 15 is replaced with Beq . Figure 17. d z(K − I v 2 )2 + (B v)2 d b 0 eq (19) This relation has been plotted in Figure 16 for a single value of X0 and several frequencies. (b) damping. . v) = X Kb X0 = . only one intersection will be found. The linear oriﬁce damping of the small amplitude model is replaced by the squeeze ﬁlm damping of equation (16). for an inﬁnitesimal excitation. Frequency response —— of the system in Figure 14 treating the cage as an amplitude-dependent dissipator (Figure 15). . excitation amplitude and frequency: a(Beq . This solution corresponds to a point of intersection in Figure 16. the non-linear model reduces to the small amplitude linear model).520 . Given speciﬁc values for X0 and v. at each frequency. The magnitude of this transfer function can be considered an expression for the normalized decoupler amplitude in terms of equivalent damping. It is evident that. X0 . (a) Elastic rate. equations (18) and (19) may be considered simultaneous equations to be solved for a and Beq . a transfer function may be found relating X(jv) to X0 (jv). approaches 1. a uniquely deﬁned frequency response will be computed (the uniqueness of the result can be proven rigorously [32]). thus. . Small amplitude (linear) frequency response ––––. the equivalent damping approaches inﬁnity as the normalized decoupler amplitude. a. with b = B0 (thus. 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.

in fact. because equivalent linearization treats the bottoming out process in a more physically meaningful way and because. low frequency inputs. 521 Figure 18. most of the energy of the force signal is then found in the ﬁfth and higher harmonics. (b) damping. 0·4 mm. 0·2 mm. large amplitude inputs. Normalized elastic rate and damping of equivalent linear model for various amplitude excitations. Compare with Figure 3: —q—. (a) Elastic rate. —— . 4.3. at high frequency. The results are shown in Figure 18 (compare with Figure 3). ----. — — —. to be superior to the piecewise linear model for high frequency inputs. and the force generated by the squeeze ﬁlm is dominated by pulses which occur as the decoupler approaches and departs the cage limits. A second reason of importance. Given the amount of spectral information which is lost. —— . the latter point may be understood: for large amplitude. ignoring higher harmonics in the force data. . the assumption of sinusoidal decoupler motion is qiute reasonable. The former point is expected. 0·05 mm. 0·6 mm. may be that the measured force is aﬀected principally by the pressure in the upper chamber. Equivalent linearization proves. 0·1 mm. By the same token. and the relative errors in Figure 19. but poorer for low frequency. This is in part because comparison is being made to experimental data which has been analyzed in a similar way. the decoupler motion is nearly a square wave. –·–·–·–. 1·0 mm. E Q 0·8 mm. ——. Indeed. it may be surprising that the equivalent linearization technique works as well as it does. however.

given a 1 mm excitation amplitude. moreover. E Q 0·4 mm. † Shear force in the rubber contributes heavily to the measured force as well. Relative errors of equivalent linear model: —q—. For the fundamental. 0·2 mm. Figure 19. while higher harmonics may contribute signiﬁcantly to the latter. At low frequency the decoupler motion will approximate a rectangle wave. (20) This force is purely sinusoidal. in response to low frequency. There are two eﬀects associated with the decoupler. 0·1 mm. . contributing no harmonics. Higher harmonics will be even smaller because the decoupler motion is smooth rather than truly rectangular. The greatest eﬀect. is upper chamber bulge caused by the imposed displacement. ——. not the pressure in the squeeze ﬁlm. the associated force will exhibit harmonics. therefore. approximately. The magnitude of this eﬀect is 2 F1 1 Ap Kb X0 . 1·0 mm. they do not to the former. The magnitude of the nth harmonic will be approximately proportional to 1/n for the ﬁrst few harmonics. F2 /F1 1 0·1. —— . —— . .522 . the various eﬀects contributing to measured force may be estimated and compared. . but is relatively simple to describe and not particularly interesting in the present discussion. (21) The magnitude of this eﬀect is much smaller than F1 . F2 1 Ap Kb (Ad (d/n)). 0·6 mm. (b) damping. 0·05 mm. –·–·–·–. — — —.† To illustrate these points. The ﬁrst is associated with the amount of upper chamber bulge caused by decoupler motion. 0·8 mm. (a) Elastic rate. large amplitude excitations. ––––. The force contributed by a lower harmonic will be.

−4 (22) For n = 5 and v = 30 Hz. 4. as long as it conforms to the general shape seen in Figure 16. With this in mind. This is straightforward with the piecewise linear model. Thus. F3 /F1 1 5 × 10 . The results (not shown) are very similar to those obtained with squeeze ﬁlm damping. but less so via equivalent linearization. This focuses attention on the capacity of the mount to provide isolation in the presence of low frequency disturbances. Both simulation and experiment suggest that the response to small amplitude inputs in the presence of a low frequency. shown in Figure 16. As in the experimental measurements. it is necessary to ensure that the two components of the excitation may be treated as statistically independent. This will not be the case if the two sinusoids are harmonically related. The use of the piecewise linear model remains exactly the same as with a single frequency. Here. 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). was investigated. In such a case. the large amplitude component of the excitation ensures that the decoupler is bottomed out a signiﬁcant percentage of the time. however. the transfer function is computed at vs .5. to within machine precision.2 that no particularly compelling reason exists to assume that the squeeze ﬁlm relation applies to a hydraulic mount. only. is of the order 2 2 F3 1 (I0 Ad )(d/n)(nv)2 1 Ap Kb e 2nd(v/vr2 )2. support the idea that a reasonable physical picture of bottoming out includes amplitude-dependent damping. Results of the equivalent linearization analysis. because the inertia of the decoupler is very small. 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. the equivalent damping is set to B0 unless the decoupler amplitude is observed to exceed its physical limits. Thus. 4. which. . The point was made early in section 4. 523 The second force term is associated with decoupler inertia. a simpliﬁed form of Beq (a). an equivalent damping is selected which is precisely large enough to ensure that the decoupler motion remains within limits. no harmonic relationship exists. except that the input changes to that given in equation (2). moreover. As with the experimental data. the phase angles must be independent.4. Results are presented in Figure 20. in performing a simulation it is important to pick vb and vs carefully so that. contributions of squeeze ﬁlm pressure pulses to the measured force are very small. for the nth harmonic. 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). which has the same format as Figure 5. This term is simply decoupler mass times acceleration. the frequency of the small amplitude component. This similarity suggests that the precise form of equivalent damping relation is not very important. Because the amplitude and frequency of each component are ﬁxed.

Compare with Figure 5.. —w—. The equivalent linear behavior can be found using a double Fourier series [33]. (24) m = 0 n = −a(m $ 0) n = 0(m = 0) where ub = vb t + fb . 2vs 2 2vb . can be expected. it is once again necessary to ensure that a harmonic relationship does not exist between the two sinusoidal components. Response to composite excitations predicted by piecewise linear model. —— . Figure 20. E W 1·0 mm single frequency response. It is also instructive to use the equivalent linearization approach. and it is necessary to assume that they have minimal eﬀect on the response. vs 2 2vb .3.. —+—. 15 Hz base. 10 Hz base.. (b) damping... 0·1 mm single frequency response. . Even so.. Note that many frequencies other than harmonics appear in the Fourier series. If the ratio of vs to vb is rational. then an inﬁnite number of combination tones will occur at the frequencies of excitation. —q—. which begins with the assumed decoupler motion (normalized) x = ab sin (vb t + fb ) + as sin (vs t + fs ). For instance. Thus. These terms are often called ‘‘combination tones’’ [34]. . which has the form 6Fsq = s a s a [Pmn sin (mub + nus ) + Qmn cos (mub + nus )]. —— . . 5 Hz base. This assumption is supported in part by the reasoning presented in section 4. us = vs t + fs . output frequencies of vs 2 vb .. (a) Elastic rate. ¯ (23) where ab + as Q 1. combination tones will exist at frequencies close to excitation. 2vs 2 vb .524 .

Clearly. such as lower chamber compliance. an interesting adaptive hydraulic mount was recently proposed by Kim and Singh [17]. Bs . this ensures that the frequency response approaches that of the large amplitude model. the incorporation of certain additional eﬀects. Moreover. can lead to moderately improved ﬁts [25]. P10 and P01 are all zero. must be large enough to limit the decoupler motion. also approaches that of the large amplitude model. This is in agreement with the behaviors seen in Figures 5 and 20. . 525 It is readily shown that Q00 . 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. A very useful result can be obtained. The inequality in equivalent damping values then implies that the response at the higher frequency. Piecewise linear and equivalent linear models have been shown to represent hydraulic mount behavior over a broader range of excitations than previously possible. Bb . More importantly. however. however. (1 − (ab sin ub + as sin us )2 )3 b s These equations may be used. (1 − (ab sin ub + as sin us )2 )3 b s ab vb cos ub + cos us as vs (25) Q01 1 = as vs 2p 2 gg0 −p −p 1 (26) × = 1 + 3(ab sin ub + as sin us )2 cos us dub dus (1 − (ab sin ub + as sin us )2 )3 1 2p 2 gg p −p p cos2 us −p (1 + 3(ab sin ub + as sin us )2 du du . while the large amplitude and small amplitude linear models ﬁt the data remarkably well. as ) = Q10 1 = ab vb 2p 2 gg p −p p (cos ub + −p a s vs cos us ) ab vb 1 + 3(ab sin ub + as sin us )2 cos ub dub dus × (1 − (ab sin ub + as sin us )2 )3 = Bs (ab . respectively: Bb (ab . While these results provide an excellent basis for hydraulic engine mount analysis and design. upper chamber bulge damping and leakage past the decoupler. 5. ab . It is readily shown by direct computation that Bs q Bb whenever ab q as . as ). vs . while Q10 and Q01 are non-zero and are related to the equivalent damping at vb and vs . As seen in the single frequency analysis. without solving any sets of equations. 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 . CONCLUSIONS Novel experimental data and mathematical models describing a hydraulic engine mount have been presented. they also suggest a number of topics for future research. For instance. For instance. a signiﬁcant increase in upper chamber damping can be used to eliminate the high frequency resonant peak (with the cost. of higher damping at frequencies above resonance). it is possible that some of these eﬀects could be enhanced to improve the performance of future mount designs. as ) = 1 2p 2 gg p −p p cos2 ub p p −p (1 + 3(ab sin ub + as sin us )2 du du . 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. the equivalent damping at the base frequency. Along similar lines.

This possibility has also been raised by Ushijima et al. ACKNOWLEDGMENTS The authors would like to thank Jim Frye. Personal communication. B and A. S. The equivalent linearization technique. necessitate a rather lengthy cycle of design. R. Hydraulic engine mount isolation. while a simulation which represents bottoming out explicitly will be stiﬀ. the ‘‘apparent inertia’’ of the inertia track. G. REFERENCES 1. 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. E. the over-estimation of damping at frequencies less than 10 Hz (Figure 7) is probably a consequence of having ignored such eﬀects.526 . testing and redesign. such as entrance and exit eﬀects in the inertia track. This was because the former did not incorporate a meaningful model of bottoming out. These factors have to date. H 1993 SAE Paper 931321. 3. K and P. B. it is not simply the hydraulic mount itself. left open the tantalizing possibility that the mount’s ability to provide simultaneous control and isolation could be improved with some other decoupler characteristic. 6. but its role in this broader context which must be the subject of future studies. P. R. Tony Villanueva and Bill Resh of Chrysler Corporation for their support of this research.-H. Engine mounts are but one contribution to the noise. that the performance of a hydraulic mount will be sensitive to the dynamic characteristics of the vehicle in which it is placed. are designed for signiﬁcant energetic interaction with the engine and chassis 2 (as an example. . R 1994 Chrysler Corporation. W. as considered here. who argue that a rubber membrane decoupler gives better performance than a rigid plate decoupler. For instance. 4. Finally. This is a challenge because higher order equivalent linearization techniques generally require the solution of sets of non-linear equations. B 1984 SAE Paper 840259. and will for the foreseeable future. S. C and G. A valuable contribution would be the development of a computationally eﬃcient model combining the best characteristics of each. 5. T 1984 SAE Paper 840407. Ap It . . The two models substantiated experimental results obtained with composite waveforms. M. performance must ultimately be considered a matter of subjective (passenger) impression. Moreover. Thus. M. 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. Thus. It is reasonable to expect. On the dynamic response of hydraulic engine mounts. Non-linearities other than those associated with the decoupler. therefore. M. T. may also exert a signiﬁcant inﬂuence on the mount’s behavior [20]. Optimal tuning of adaptive hydraulic engine mounts. while the latter was incapable of representing substantially non-sinusoidal signals. is comparable to the engine mass). it would be inappropriate to close a discussion of hydraulic engine mounts without recalling the greater context. . [14]. C 1985 SAE Paper 851650. vibration and harshness characteristics of an automobile. however. Hydraulic engine mount characteristics. Hydraulic mounts. in particular. future models should incorporate other non-linear eﬀects. 2. A new generation of engine mounts. S 1989 SAE Paper 891160.

S. A broadband adaptive hydraulic mount system. 23. R. editor). G. G. C 1992 Master’s Thesis. 17. A. 30. P. 310–316. K and R. G 1963 Nonlinear Automatic Control. 9. S. M. Dynamic analysis of mechanical systems with clearances. U and T. J. J. 34. U. 527 7. D. B0 . 20. Massachusetts: MIT Press. Journal of Engineering for Industry 93. B. P 1986 Random Data: Analysis and Measurement Procedures. F. Dubowsky and F. M 1984 SAE Paper 840410. Open-loop versus closed-loop control for hydraulic engine mounts. G. M. G and W. H. 11. Hydraulic mounts—improved engine isolation. New York: McGraw-Hill. Massachusetts: Addison-Wesley. 165–180. S. 14. mr2 . J. K 1989 SAE Paper 891138. G. W. Frequency-domain analysis of nonlinear circuits driven by multi-tone signals. 16. Linear analysis of automotive hydro-mechanical mount with emphasis on decoupler characteristics. 12. S and W. V-V 1968 Multiple-input Describing Functions and Nonlinear Systems Design. R 1975 System Dynamics: A Uniﬁed Approach. 247–255. H 1986 SAE Paper 860549. R. is presented. and slope of the elastic rate plot at resonance. K and R. S 1989 Journal of Sound and Vibration 134. Nonlinear B. for predicting vibration of vehicle with hydraulic engine mount. L. 10. P. 243–246. O. Northwestern University. F. K and R.B. APPENDIX A: ORIFICE PROPERTIES A method of estimating oriﬁce inertia. Squeeze ﬁlms for rectangular plates. Part 2: dynamic response. H. Understanding hydraulic mounts for improved vehicle noise. 766–778. 13. S and E. L. K and R. I 1978 International Journal of Non-linear Mechanics 13. A study of ﬂuid squeeze-ﬁlm damping. L.-T. R and S. S 1988 SAE Paper 880075. F. 71–78. isolation and shock control characteristics of automotive nonlinear hydraulic engine mounts.A. S 1993 in Advanced Automotive Technologies (M. G. 25. 8. J. R. ASME. and damping. 451–456. New York: John Wiley. G. A. R 1992 Journal of Sound and Vibration 158. C. C 1993 Master’s Thesis. Ahmadian. T and H. U and L. Y 1966 Journal of Basic Engineering 88. G. 18. Resonance. vr2 . K. C and R. K and P. 259–290. S and R. Optimum application for hydroelastic engine mount. D. 22. D. W 1987 Automotive Engineer 12. Reading. P. Measurement and Control 115. S. New York. K and R. Computer simulation of hydroelastic engine mounts. 482–487. T. P. New York: John Wiley. High performance hydraulic mount for improving vehicle noise and vibration. S. Adaptive hydraulic engine mounts. K 1988 SAE Paper 880073. Cambridge. 24. 21. vibration and ride qualities. S 1992 in Transportation Systems—1992. G. The starting . W. D 1986 SAE Paper 860550. 17–19. D. New York: McGraw-Hill. J. C. P and M. 27. R. H 1963 Journal of Basic Engineering 85. Freudenstein 1971 Transactions of the American Society of Mechanical Engineers. J. 29. S. Nonlinear analysis of automotive hydraulic engine mount. P 1961 Analysis and Design of Engineering Systems. S 1993 Journal of Dynamic Systems. D. 26. N. I0 . D. V. 15. New York: ASME. A 1986 SAE Paper 861412. C. Northwestern University. Dynamic analysis of hydraulic engine mounts. 28. Non-linear frequency response characteristics of an impact pair. F 1985 SAE Paper 850975. Y. K. G. S. W 1990 Digital Control of Dynamic Systems. A. W. R. H. based on resonant frequency. P. R. 31. 32. L. T. G and T. C 1984 IEEE Transactions on Circuits and Systems CAS-31(9). 19. 219–243. T. W. L. B and A. T. Hydraulically-damped engine mounts. S. On the existence and uniqueness of solutions generated by equivalent linearization. Active frame vibration control for automotive vehicles with hydraulic engine mount. S 1988 SAE Paper 880074. 33. Hydraulic engine mount for shock isolation at acceleration on the FWD cars. E.

. . subsequently. and evaluating at v = vr1 : mr2 = 2 2Ap Kb 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)) . $ 2 2zvn − (1 − a)v 2 . point is equation (14).528 . which provides an approximate expression for dynamic stiﬀness. (A1) The frequency response may then be written in terms of real (elastic rate) and imaginary (damping) parts. 2 Kdyn (jv) = Kr + Ap Kb $ 2 av 4 + (4z 2 − a)vn v 2 2 2 2 2 2 2 (v − vn ) + 4z vn v % % (A2) 2 +jv Br + Ap Kb 2zvn where a = 1 − Ad /Ap . equation (A3) may be solved for vn . z = B0 /2zKb I0 . (A4) An expression for mr2 is found by diﬀerentiating the elastic rate term in equation (A2) with respect to frequency. ((1 − b)2 + 4z 2b)2 (A5) % Equation (A5) may be solved for z using a Newton–Raphson technique. 2 2 2 (v − vn )2 + 4z 2vn v 2 (A3) It is now helpful to deﬁne b as b = [1 − z4(1 − a)z 2 + a 2 ]/(1 − a). . Determination of I0 and B0 is then straightforward using equations (A1). It is convenient to deﬁne a natural frequency and damping ratio as vn = zKb /I0 . An expression for vr2 is found by maximizing the damping term in equation (A2) with respect to frequency: 2 2 vr2 = vn [1 − z4(1 − a)z 2 + a 2 ]/(1 − a).