Wang X. Vehicle Noise and Vibration Refinement

Подождите немного. Документ загружается.

Advanced simulation techniques in vehicle noise 183

weeks; this often causes a signiﬁ cant delay of CAE analyses with respect

to design status and the need to kick off the prototype build process before

the ﬁ nal virtual series could be completed.

8.4.2 Noise and vibration reﬁ nement deliverables for

virtual series

Each component CAE model that is required for a full vehicle noise and

vibration simulation model will ﬁ rst be checked and analysed by the respon-

sible component area to ensure that the component will meet its targets.

Typical noise-relevant component attributes are dynamic stiffness, compo-

nent eigenfrequencies and mode shapes, noise and vibration transfer func-

tions as well as source levels, e.g. shaking forces and noise radiation of

powertrains. If any component signiﬁ cantly fails to meet its noise-related

target, full vehicle noise performance will also probably not meet its targets.

Typical virtual series deliverables for full vehicle noise and vibration

reﬁ nement are:

• Vehicle modal alignment: all component modes should properly be

separated to avoid any ‘resonance catastrophe’ under operating

conditions.

• Vehicle sensitivity to ‘unit excitation’: this can be a vibration transfer

function (vehicle response at the customer perception point due to, e.g.,

unit mount force, calculated by FE tools), a noise transfer function

(p/F, also calculated by FE tools) or noise reduction (sound pressure

difference between engine bay and passenger cabin, calculated by

SEA tools).

• Response to typical vehicle load cases, e.g. structure-borne response to

idle excitation (loads of the powertrain applied to a full vehicle FE

model), response to wheel imbalance (nibble, solved by FE or system

dynamics tools depending on vehicle content for electronic tools to

reduce steering wheel rotation – EPAS), response to road excitation or

road noise (road surface irregularities applied to tyre patch points using

an FE tool); vehicle acceleration noise (including all load paths in the

TPA tool); and high-frequency noises like wind noise (derive the shape-

related pressure ﬂ uctuation on the outer vehicle skin from CFD simula-

tion and apply this as load for SEA analyses).

The ‘status assessments’ that need to be prepared for the virtual series

report will be followed by further activities to identify the root causes for

missing attribute targets. Transfer path analyses as well as modal contribu-

tion analyses are typical methods applied for these noise and vibration

reﬁ nement investigations (see Plates VII and VIII between pages 114 and

115).

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:48 AM

IP Address: 129.132.208.2

184 Vehicle noise and vibration reﬁ nement

Based on such ﬁ ndings, high-level corrective actions like increasing local

stiffness by adding reinforcements to the CAE model or shifting component

modes in the respective CAE model will be investigated. If these changes

solve the issues, these concepts will be communicated to the respective

component areas to develop feasible solutions that need to be incorpor-

ated into the design for the next virtual series or for prototypes, as

appropriate.

8.4.3 Simulation of design status versus

x-functional optimization

Virtual series as described above would just deliver vehicle noise and vibra-

tion reﬁ nement status for nominal design. This would not make use of one

main advantage of virtual analyses: CAE enables analysis of several differ-

ent designs within a fraction of the time, efforts and resources that would

be needed for creating and testing real hardware variants. Consequently,

simulations should

• take into account variability of design parameters such as mount stiff-

ness on attribute performance,

• investigate parameter settings which enable more robust designs,

• include x-functional optimization to ﬁ nd a best compromise between

attribute performances. The attributes offering most potential conﬂ icts

for noise and vibration reﬁ nement are vehicle dynamics, durability,

weight and cost.

Each of these tasks requires a series of simulations using different input

parameter sets. There is no longer a need to submit each simulation

run manually; several optimization packages are available on the market

that automatically submit a series of analyses on individual tools as listed

above, tailored to variability of input parameters, and that carry out post-

processing and optimization.

In real-case applications, every simulation can take hours or even days.

In these cases, the time to run a single analysis makes running more

than a few simulations prohibitive. The design of experiment (DOE) tech-

nique is a smart approach to achieving a maximum amount of knowledge

from a reduced number of calculations or tests [11]. Further analyses can

be performed on a response surface that is a multi-dimensional hyper-

surface created by interpolation of the well-distributed results on the full

vehicle CAE model. This response surface represents a meta-model of the

original problem and can be used to perform the optimization. Just the

outcome of this optimization needs to be veriﬁ ed on the original CAE

model.

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:48 AM

IP Address: 129.132.208.2

Advanced simulation techniques in vehicle noise 185

8.4.4 Conﬁ dence in noise and vibration

reﬁ nement simulations

Quality and conﬁ dence in CAE simulation results depend on the quality of

each single involved sub-model. Some vehicle noise and vibration reﬁ ne-

ment phenomena can be predicted with high conﬁ dence by CAE (e.g.

powertrain rigid body modes in a vehicle, bending frequencies of a power-

train), whereas prediction of some other noise and vibration reﬁ nement

characteristics (e.g. trimmed body noise transfer functions) cannot be pre-

dicted with an accuracy of less than say 3 dB. Consequently, road-induced

low-frequency noise – which involves trimmed body noise transfer func-

tions for suspension attachment points – cannot be predicted with sufﬁ cient

accuracy by CAE.

Consequently, a ‘full analytical sign-off’ for vehicle noise and vibration

reﬁ nement is still not achievable. Therefore all attribute assessments should

be done jointly between attribute experts and simulation experts. In the

case of road noise, any analytical upfront assessment might be based on a

surrogate metric by comparing suspension forces of different suspension

designs.

In order to enable early CAE analyses for the next vehicle program,

CAE teams need to spend reasonable resources on correlating the latest

CAE models with current production hardware.

8.5 Application of virtual reality for vehicle noise and

vibration reﬁ nement

Simulation techniques should be applied to enable evaluations of complete

vehicles or subsystems even before hardware is available. Consequently,

results of CAE simulations should not be shown just as ‘pure numbers’ or

‘colour curves’ but should be presented in a form similar to impressions in

real vehicles.

8.5.1 Visualization of simulation results

Modal analysis results have been animated on computer screens for many

years; animation speed and quality have improved signiﬁ cantly. For illustra-

tion purposes, displacement amplitudes are often magniﬁ ed by orders of 10

to allow for, e.g., mode interpretation. Recent developments allow 3D

display without any magniﬁ cation in a so-called ‘virtual reality cave’ to

evaluate the real visible impression of driving a vehicle. In addition, simu-

lating seat and steering wheel vibrations in such a cave creates a real in-

vehicle ‘look and feel’ impression [12].

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:48 AM

IP Address: 129.132.208.2

186 Vehicle noise and vibration reﬁ nement

8.5.2 Auralization of simulation results

Auralization techniques are all the techniques required to generate and

simulate predicted sounds and noises. The main techniques for auralization

approaches are order extraction, sound decomposition and sound synthesis.

Order extraction techniques allow identiﬁ cation of order content of any

recorded vehicle sound, based on quality ‘tacho’ signals that enable identi-

ﬁ cation of trigger signals. Sound decomposition is required to separate

recorded noise into different orders and ‘random noise’ contributions.

Sound synthesis is the last process step in superposing individual sound

components. It is extremely critical for the quality of any synthesized sound

that phase relationships between individual orders are representative,

otherwise any synthesized sound will sound artiﬁ cial and cannot be used

for jury evaluation. This is relevant for both binaural (two-ear) sounds that

are based on artiﬁ cial head (Aachen Head) measurements as well as mono-

aural (mono) sounds.

8.6 Conclusions

Computer performance is improving from year to year, thus enabling more

noise and vibration reﬁ nement CAE simulations as well as analyses on

CAE models with more degrees of freedom. Finer meshing itself is not

necessarily guaranteeing increased accuracy of simulation results; simula-

tion methods also need to be permanently reﬁ ned. These method enhance-

ments are usually performed as joint projects between software vendors,

universities and OEMs. CAE enhancements and new methods are regularly

reported at user software conferences and are included in software releases.

A list of URL links to vendors of commercial noise and vibration reﬁ ne-

ment software packages is given in Section 8.7. A very good overview on

the latest state-of-the-art noise simulation methods and projects is given,

for example, at the annual SAE Noise and Vibration Conference (Society

of Automotive Engineers).

Although many noise and vibration phenomena can be simulated, there

is no real need to do this for each and every noise and vibration effect.

Strictly following the respective design rules can be more than capable of

avoiding recurrence of typical error states such as gear rattle, transmission

whine or squeak. Therefore any simulation should be requested and set up

only if adding real value to the carline program – never do any CAE simu-

lation just to follow any process. Simulations additionally require dedicated

resources for component ‘meshing’, simulation model creation and ﬁ nally

simulation job execution and analysis; simple component rig tests may be

a better alternative than CAE.

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:48 AM

IP Address: 129.132.208.2

Advanced simulation techniques in vehicle noise 187

A ﬁ nal remark on CAE simulation is that any CAE model that generates

error messages deﬁ nitively delivers wrong results; CAE simulation runs

without warnings could be correct – nevertheless it is important to under-

stand both the physics as well as the restrictions of each CAE method and

model. Always do sanity checks on results. It is signiﬁ cantly better not to

publish doubtful CAE results than to share erroneous CAE results that

drive wrong design directions. Nevertheless, developers should make use

of simulation techniques for noise and vibration reﬁ nement, as they are a

viable contributor to quality vehicles.

8.7 Sources of further information and advice

There are hundreds of CAE tools available on the market that can be used

for noise and vibration reﬁ nement simulation. The following is a small

selection of tools used within the Ford Motor Company for noise and vibra-

tion reﬁ nement:

NASTRAN, Finite Element solver, www.mscsoftware.com

ABAQUS, Finite Element solver, www.simulia.com

RADIOSS, Finite Element solver, www.altairhyperworks.com

AKUSMOD, tool for coupled ﬂ uid–structure analysis, www.sfe1.extern.

tu-berlin.de

ACTRAN, tool for coupled ﬂ uid–structure analysis, www.fft.be

SYSNOISE, boundary element tool for acoustics simulation, now included

in Virtual.Lab, www.lmsint.com

ADAMS, multibody system software, www.mscsoftware.com

SIMPACK, multibody system software, www.simpack.com

SIMULINK, simulation software for multidynamic systems, www.

mathworks.com

VA One, statistical energy analysis tool, www.esi-group.com

SEAM, statistical energy analysis tool, www.seam.com

Star-CD, CFD tool, www.cd-adapco.com

PowerFLOW, CFD tools, www.exa.com

Artemis, sound recording, analysis and playback software, www.head-

acoustics.de/eng

modeFRONTIER, optimization and robustness tool, www.esteco.com

I-sigh, optimization and robustness tool, www.simulia.com

HyperStudy, optimization and robustness tool, www.altairhyperworks.com

8.8 References

1. Zienkiewicz, O., The Finite Element Method, 3rd edition, McGraw-Hill, New

York, 1979

2. Bathe, K., Finite Element Procedures, Prentice-Hall, Englewood Cliffs, NJ, 1982

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:48 AM

IP Address: 129.132.208.2

188 Vehicle noise and vibration reﬁ nement

3. MSC/Nastran, Handbook for Superelement Analysis, MacNeal-Schwendler

Corp., Los Angeles

4. Ciskowski, R.D. et al., Boundary Element Methods in Acoustics, Computational

Mechanics Publications, Southampton, UK, 1991

5. Schielen, W., Multibody Systems Handbook, Springer, Berlin, 1990

6. Harper, B.D., Solving Dynamic Problems in MATLAB, Wiley, New York, 2002

7. Lyon, R.H., Statistical Energy Analysis of Dynamical Systems, Theory and

Applications, MIT Press, Cambridge, MA, 1975

8. Lyon, R.H. and DeJong, R.G., Theory and Application of Statistical Energy

Analysis, 2nd edition, Butterworth-Heinemann, Boston, MA, 1995

9. Ferziger, J.H. and Peric, M., Computational Methods in Fluid Dynamics,

Springer, Berlin, 2002

10. Lietz, R., Mallick, S., Kandasamy, S. and Chen, H., Exterior airﬂ ow simulations

using a Lattice Boltzmann approach, 2002 SAE World Congress, Detroit, paper

2002-01-0596

11. Grove, D.M. and Davis, T.P., Engineering Quality and Experimental Design,

Longman, London, 1992

12. Allman-Ward, M. et al., The interactive NVH simulator as a practical engineer-

ing tool, 2003 SAE Noise and Vibration Conference, Traverse City, MI, paper

2003-01-1505

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:48 AM

IP Address: 129.132.208.2

189

Advanced experimental techniques in vehicle

noise and vibration reﬁ nement

T. AHLERSMEYER, Ford-Werke GmbH, Germany

Abstract: This chapter discusses advanced techniques to correctly

measure vehicle noise and to understand its root causes. The chapter

covers structure-borne and air-borne contribution analysis for low and

high frequencies in general and shows how sound intensity probes and

laser vibrometers can help to analyse the test object. In addition,

psychoacoustic metrics and material tests typically used in the ﬁ eld of

vehicle noise and special aspects of tracked analysis are discussed.

Key words: transfer path analysis, sound intensity, source identiﬁ cation,

sound quality, material testing, tracking analysis.

9.1 Transfer path analysis technique

The classical method of noise source identiﬁ cation is to conduct a discon-

nection test. If a source is disconnected (e.g. by decoupling the engine

mount, putting a total mufﬂ er on the air intake system, etc.) and the noise

amplitude at the receiving point is reduced signiﬁ cantly, the main root cause

is found. However, this only works well if there is a single dominant source

(or very few). Even then there is an inherent risk that the disconnection

test changed the other contributions, which is often the case if engine

mounts are decoupled. With multiple sources of similar contributions this

method is not able to give clear results.

In such cases, Transfer Path Analysis (TPA) works better. The TPA

models each connection between the noise source (e.g. vibrating engine)

and the receiver (e.g. interior compartment noise or seat vibration) as an

excitation (e.g. force input, N) and a transfer function (e.g. receiver sensitiv-

ity, Pa/N or (m/s

)/N). Figure 9.1 shows the source–receiver model. The

general thinking is:

resp

= exci

× TF

(9.1)

where

resp

= response at point i caused by excitation at path j

exci

= excitation at path j

= transfer function between points i and j.

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:51 AM

IP Address: 129.132.208.2

190 Vehicle noise and vibration reﬁ nement

If all paths are modelled correctly the complex sum of all responses

should be equal to the response under operation:

resp resp

oper,

iij

∑

(9.2)

where resp

oper,i

= response at point i measured under operation.

The difference between the calculated response and the measured

response can be used as an indicator of the quality of the TPA model.

However, even if the calculated response equals the measured response

under operation, this is not a proof that the TPA model is exact in every

detail.

The main problem of the TPA is to get the force input.

1–3

There are four

principal methods:

1. Force method: measure excitation force by introducing a force sensor.

2. Stiffness method: estimate excitation force through, e.g., rubber mounts

by multiplying measured relative displacement by measured dynamic

stiffness.

3. Matrix method: estimate excitation force by measuring and inverting

the inertance transfer matrix.

4. Operational TPA: estimate excitation force by a least-squares approach

of operational data.

The features of each method are:

1. Introducing a force sensor (e.g. 15 mm high) will modify the vehicle

(mass and stiffness at the connection point and position of force input)

and therefore potentially all path contributions. It can only be applied

in special cases. The advantage is that the problems of other methods

do not arise (see below).

Vibration•

•

Source

Receiver

Noise radiation

Vibration

Noise

9.1 Source–receiver model.

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:51 AM

IP Address: 129.132.208.2

Advanced experimental techniques in vehicle noise 191

2. The measurement of relative displacement can be difﬁ cult for complex

mounts such as engine mounts. The dynamic stiffness of a rubber mount

depends on the frequency and the preloads in all directions. Unfortunately

the preload changes during operation (wide open throttle = WOT, or

overrun = OVR) due to changing engine torque and with temperature,

so it is very difﬁ cult to test the engine mount under real-world condi-

tions. Commercially available 3D mount testing machines are extremely

expensive and work only up to 100 Hz, so they cannot provide the

required data.

3. The matrix method only works if all relevant transfer paths are included

in the model. A typical vehicle model will include at least 25 structural

paths, a model of a rear- or all-wheel drive vehicle may include up to

80 paths (see Table 9.1).

4. The least-squares approach assumes that transfer functions are constant

over, e.g., all engine rpms. Since the engine torque changes over rpm

this is not valid in detail. As the desired response is used as a target for

the least-squares optimisation, the difference between the measured

response and the calculated response can no longer be used as a quality

indicator of the TPA model.

For the ﬁ rst method the vehicle must be modiﬁ ed signiﬁ cantly; for the

second method the vehicle must be modiﬁ ed if the original mounts are to

be used. For the ﬁ rst, second and third methods, the transfer functions

should be measured with a decoupled excitation side, which again is a

modiﬁ cation of the vehicle. These modiﬁ cations may change the vehicle

signiﬁ cantly, therefore making validation of the model impossible.

During the measurement of the transfer functions the vehicle is at room

temperature, has no occupants (missing volume and mass load) and is not

preloaded with reaction torques from the powertrain. All these conditions

differ more or less from operational conditions and may adversely affect

the result.

Table 9.1 Typical paths to be included in a TPA

Count Component Directions Paths

3–4 Engine/transmission mounts XYZ 9–12

1 AC hose XYZ 3

1 Clutch cable XYZ 3

3–8 Exhaust hangers Z or XYZ 3–8

2 Airborne noise sources S (scalar) 2

1 Powertrain radiated noise S (scalar) Typically 6

2 (4*) Side shafts* XYZ 6–12

* If the contribution of the side shafts is to be further analysed, they must

be split down to link arm (XYZ) + top mount (XYZ) per side shaft.

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:51 AM

IP Address: 129.132.208.2

192 Vehicle noise and vibration reﬁ nement

The only way to avoid this is to conduct operational TPA, which unfor-

tunately gives the least insight into the vehicle because the excitation forces

are not calculated; only the path contribution and the transfer functions

(Pa/(m/s

)) are calculated, which are only valid under the operating condi-

tions used.

4,5

This situation can be improved by performing noise transfer function

(NTF) (Pa/N) and point inertance measurements ((m/s

)/N). With these

values one could better understand the body structure and calculate the

exciting forces. However, for physically exact results it is required to discon-

nect the body and excitation sides in order to ensure that the exciting force

is introduced only into the body. Unfortunately this gives the method a

disadvantage.

Airborne noise sources such as intake and exhaust oriﬁ ce noise can be

included in the model, e.g. by measuring the noise under operation (Pa)

and the transfer function with noise excitation at the oriﬁ ce (Pa/Pa). Even

the radiation from surfaces can be included by measuring the surface aver-

aged acceleration (m/s

), multiplying it by the surface (m

) and measuring

the volume acceleration transfer function (Pa/(m³/s

)).

As the TPA calculates a vector sum of contributions it can only deal with

coherent signals with constant phase relationship (e.g. engine excitation).

For incoherent excitation (e.g. road noise) the inputs must ﬁ rst be repre-

sented by a set of orthogonal vectors (principal components, PRC) by

means of a principal component analysis (PRCA).

The number of principal components is identical to the number of refer-

ence signals. Nevertheless, a certain principal component cannot be inter-

preted as being correlated with a certain reference signal. In order to

explain the measured response best a high number of PRCs are required.

However, for each set of PRCs one has to solve a complete TPA. Therefore

it is desired to ﬁ nd the lowest number of reference signals that can explain

the measured response well enough in the problem areas. The quality of a

road noise TPA depends highly on the selection of the reference signals.

The index of the highest contributing PRC is not constant, so problem

range A may be explained mainly by PRC 1 whereas problem range B may

be explained mainly by PRC 2. In general a noise phenomenon (e.g., a

response peak at a certain frequency) is explained not just by a single PRC

but by a combination of PRCs. This requires that the results of multiple

TPAs have to be judged in parallel or one has to work with the sum of PRCs.

9.2 Sound intensity technique for

source identiﬁ cation

The sound pressure is a scalar value, so by measuring the sound pressure

at one point there is no information available concerning where the noise

is coming from. For a point source in a free-ﬁ eld environment the sound

Copyrighted Material downloaded from Woodhead Publishing Online

Delivered by http://woodhead.metapress.com

ETH Zuerich (307-97-768)

Sunday, August 28, 2011 12:05:51 AM

IP Address: 129.132.208.2