The Use of System Modelling Techniques to Filter Test Measurements and Drive a Physical Vehicle NVH Simulator

© Altair Engineering 2011 1

The Use of System Modelling Techniques to Filter Test Measurements and Drive a Physical Vehicle NVH Simulator

David Fothergill Engineering Consultant DJ Fothergill Consulting [email protected] [email protected]

Frank Syred Senior Consulting Engineer Sound & Vibration Technology Ltd Station Lane, Millbrook, Bedfordshire, England, MK45 2YT [email protected]

Paul Kennings Technical Leader: Mounting Systems Chassis & Powertrain NVH Bentley Motors Limited Pyms Lane, Crewe, Cheshire, CW1 3PL, England [email protected]

Abstract

This paper describes the use of a system model to create filters with which to modify test measurements for playback in a physical vehicle Noise, Vibration and Harshness (NVH) simulator, and so to provide directional guidance in the design of a luxury passenger vehicle. The technical approach is described, whereby a whole vehicle model is developed, using the HyperWorks suite, specifically to characterise steady speed and light impact ride responses, and then used to predict the effect of design changes. The updated model results are then manipulated with the baseline results to produce “change filters” in the frequency domain. These filters modify, rather than replace, the results. This obviates the need for explicitly representative road surface simulation, since it is mathematically removed in the process of the (unit-less) filter creation. A physical simulation process is outlined whereby detailed operating measurements were taken on a test vehicle, under specific driving conditions, and accurately reproduced in a driving simulator. This enabled NVH assessments to be completed in a fully immersive environment. Thus, jurors were able to subjectively assess the effects of simulated changes to the chassis design based on a combination of test and mathematical modelling. Keywords: NVH, FVS, HyperWorks, RADIOSS

mailto:[email protected]




© Altair Engineering, SVT & Bentley Motors Limited 2011' The use of System Modelling Techniques to Filter Test Measurements and Drive a Physical Vehicle NVH Simulator 2

1.0 Introduction

A process has been devised whereby a system model is used to create change filters, with which to modify test measurements for playback in a physical vehicle NVH simulator. This was in response to a requirement from Bentley Motors Ltd to understand the NVH implications of installing alternative powertrains in existing and new vehicles, principally those concerning the control of road induced shake. Altair‟s HyperWorks provided a convenient all-in-one modelling package in which to perform the vehicle modelling and response prediction.

1.1 A Brief Introduction from the Manufacturer’s Perspective

Bentley requires world class levels of ride comfort and refinement in their products. These attributes complement Bentley‟s tradition of power, poise and craftsmanship. The technique described in this paper supports the delivery of luxury motor cars which offer their drivers class leading levels of ride comfort. The ride shake element of this attribute was achieved by tuning the power-unit mounting system for optimal shake. Initially, the technique was augmented by vehicle testing to rank alternative specifications and provide correlation with the simulation process. These tests were also used to confirm the attribute balance between powertrain refinement and power-unit shake. Once the technique had been established, and confidence in correlation was accumulated, it was possible to optimise the power-unit mounting system with much less physical testing than before. The system model represented a very attractive investment for Bentley, as it would be for any car company, since it reduced the budget required to develop the car.

1.2 An introduction to the NVH Simulator

Sound & Vibration Technology Ltd (SVT) has been involved in the development of noise and vibration simulation for the automotive industry for many years. The primary function of the noise and vibration simulator is to provide an immersive environment for the occupants to experience the actual sounds and vibrations they would experience if driving a vehicle in "real life", and to present them in the correct perceptual context.

Sound-only simulation utilises Desktop NVH Simulator software from Brüel & Kjær. which provides a portable NVH simulator. This enables existing and modified design iterations to be evaluated interactively as the simulator is „driven‟ in a virtual scenario displayed on a desktop monitor

SVT has also developed a Full Vehicle noise and vibration Simulator (FVS) which provides calibrated vibration stimuli to the driver in addition to the sound stimuli (Figure 1). The Full Vehicle NVH Simulator uses the same Desktop NVH Simulator software, but is a complete static vehicle which is positioned in front of a projection screen. The FVS is a "driver-in-loop" system, meaning that the driver's inputs to the vehicle are monitored and used to calculate the vehicle parameters as it is „driven‟. The sound and vibration stimuli for the occupants of the vehicle are similarly calculated in real time, and presented to the occupant as they „drive‟ in a virtual scenario displayed on the projection screen.


Figure 1: The NVH Simulator Both the Desktop NVH Simulator and the Full Vehicle Simulator allow project managers, engineers, and customers to make sound quality assessments of vehicles in a controlled environment. This enables evaluations of vehicles which do not exist, allowing the loop between design and the customer to be closed early in a vehicle programme, before committing to large development and tooling costs. The brand image of the vehicle can thus be "designed in" rather than problem noise engineered out at the end of the programme. The FVS adds the ability to evaluate vibration quality in addition to the sound quality. The FVS can also be used to “replay” events such as fixed speed runs over a particular section of road, the negotiation of harsh irregularities such as pot holes or speed bumps, acceleration, steady speed under load etc. In this case each member of a jury can experience exactly the same chosen condition, removing the variations due to individual driving styles [1]. Different assessment scenarios can be prepared or directly modified / filtered in real-time from the measured NVH data to create a target NVH performance, e.g. in response to the feedback from a jury-based evaluation session. Typically filters are applied to enhance or attenuate identified features (e.g. noise and / or vibration peaks). A jury assessment of stimuli, with different characteristics determined by these filters, can be used to select the preferred characteristics. The paths contributing to these features are then addressed to identify the scope for improvement. The novel feature introduced in the work discussed in this paper is the use of a CAE model to create the modification filters, specifically aimed at the refinement of road induced shake.


2 The process

2.1 The NVH Database

An early decision was taken to formulate a series of events for fixed replay, comprising surfaces with content most likely to represent challenges to the vehicle ride comfort and for interactive driving for powertrain noise and vibration. The fixed replay conditions deliberately encompassed surfaces that the jury were familiar with and with which they had certain expectations of the level of ride comfort. These included both steady speed random surfaces and discrete irregularities (small impacts). Care was taken to run at more than one speed to avoid potential masking by wheelbase frequency filtering effects. Recordings were made of binaural sound and driver tactile vibration locations. The latter included the driver‟s seat cushion, seat back, the heel point and the steering wheel (centre and rim). Furthermore, a series of auxiliary vibration and sound measurements were taken to allow the calculation of the contributions of various NVH sources, and the tracking of their paths to the driver tactile points / ears. Notably, these included the noise entering the cabin due to forcing from the engine mounts. It was anticipated that any alterations made to the engine mount dynamic stiffness to alter ride shake could also be applied in parallel to the structure borne engine noise paths. Thus, the jurors could experience both the result of the CAE based vibration filter on the ride characteristics and the effect of this change on the powertrain noise contribution to the vehicle sound.

2.2 The application

Figure 2: The Process

The objective of the process (Figure 2) was to use the simulator to both give directional NVH targets and to evaluate physical changes predicted by the CAE model. In the first instance the simulator was used to replay fixed events measured on the base vehicle, and correlation was confirmed between the measured and simulated events.


The base simulation was then used to set a benchmark for the desired levels of NVH refinement. An expert jury used this to define targets to be achieved with the alternative power-unit. A CAE model was constructed to represent the vehicle, and was correlated to ensure correct modal alignment, representative levels of damping, and realistic vibration transfer functions. The CAE model was then modified to change the vehicle for the alternative specification. Unity excitations were applied to both models representing the fixed replay events (continuous running and light impacts at a variety of speeds), and change filters were produced using the CAE response acceleration prediction for the base vehicle as the denominators, and those for the modified vehicle as the numerators. That is, the filter is defined as

(ä (ω) Revised Specification / excitation force) Filter = -----------------------------------------------------------------

(ä (ω) Base Specification / excitation force)

Where ä (ω) is a matrix of acceleration spectra.

It should be noted at this point that as both of the excitation functions are identical, the filter created is both dimensionless and independent of the forcing. Each measured tactile vibration was modified by the appropriate predicted filter, and the results were then replayed in the simulator. Thus, a subjective impression was gained of the effect of changing the power-unit mounts on the ride shake performance. The events selected for the simulator assessments had been chosen such that there was little powertrain noise contribution during the events. Thus, the same binaural sound was used for all the assessments of a specific event to ensure that they were focussed on the change to the ride characteristics as a result of the applied change filter. Concerns were identified and change targets were generated for the filters. The CAE model results were then analysed to identify the causes of vibration peaks and design studies were performed to minimise them. Filters were then re-generated and the results evaluated in the simulator. Subsequent CAE model refinements were introduced to assess the effects of changes to suspension components and body flexibility. In one instance, Bentley engineers were able to assess the effect of completely removing a body modal response and thus experience the “ultimate” trimmed body rigidity.


3.0 The CAE model

3.1 Modelling Philosophy

Altair‟s HyperWorks CAE suite was chosen to construct and run the system model, as it brought together in one package all the necessary technologies. HyperMesh was used to read CAD data, form the geometric databases and construct some of the component models. All of the models were formulated as RADIOSS bulk data decks, which were read into HyperMesh and solved using RADIOSS. Even in its original form the model represented a complex system, combining real scalar springs and masses with both viscous and hysteretic dampers, which would have been hard to meaningfully characterise using simple proportionally damped modal modelling techniques. Therefore the model was run, from the outset, using direct forced response. The model was primarily developed to meet the need for the simulation of changes to the powertrain mounting system, with the minimum of peripheral detail. Effort was focussed in representing those parts which directly attached to, and influenced, the powertrain. The powertrain and body were modelled as rigid lumped masses. The exhaust system was comprised of a reasonably representative beam, mass and shell model and the hydraulic engine mounts were modelled as dynamic sub-systems. Initially, the objective was to separate the model into fixed and tuneable parts. In the first approach, the suspension systems were represented as purely parametric models. Their mass and stiffness properties, taken from Kinematics and Compliance (K&C) test data, were represented as scalar properties, with no attempt at any physical modelling. The spring and damper (viscous and hysteretic) parameters were arranged such that the correct recession characteristics were represented, reproducing the appropriate cross coupling between vertical and fore and aft response. Together with simple split (two spring) tyre contact patch models, this proved sufficient to simulate representative responses to unit road inputs. In particular, in the correlation phase, it facilitated experimentation with the trade-off between the recession angles and the vehicle vertical and horizontal shake. By contrast, the engine mounts were semi-detailed representations of functions of the elasto-hydraulic parameters and could be tuned to represent alterations to the physical design. As the need arose to investigate potential interactions with the powertrain, the model was developed to include fully articulate (albeit small displacement) suspension systems, complete with a hydraulically mounted subframe, isolated differential unit, drive shafts and propshafts (Figure 3).


Figure 3: The CAE Model Ensuing work went on to add a flexible (modal) trimmed body model, but the prime concern was always to create a meaningful simulation with the minimum of data in the shortest possible time.

3.2 Modular approach to model construction

A modular approach was taken to model assembly, whereby a master “deck” was created to pull together and control all of the subassemblies. This contains “include” statements to select all the required component models (suspension systems, exhaust, powertrain, wheels, etc), bush sets, individual hydromounts and the load and phase controls for all of the inputs. A tool was built in Microsoft Excel to create the master deck. This was configured so that the user simply needed to select a load case (e.g. steady speed, front impact, or rear impact) and enter the required speed. The loads were then created automatically, together with the phase delay tables (as functions of frequency) appropriate for the contact patch lengths, wheelbase and speed. The tool was further developed to assist powertrain model configuration, where, for example, the entry of a particular powertrain type acronym would control the selection of an appropriate base set of mounts. In this way it was possible to alter loadcases, speeds, configurations etc extremely quickly. The alternative would be to work through the SPCD, TABLED. RLOAD, DAREA and DLOAD cards and manually edit them, or to find and alter them in HyperMesh, either of which would be much more time consuming. It is, however, acknowledged that the HyperMesh Load Step and Load Collector Browser features (Figure 4a) greatly ease the task of over-checking the applied load logic.


Figure 4a: Typical load step tree and load collector set table, generated using the HyperMesh Browsers

The Excel based master deck also provided a way of automatically documenting models, leaving a configuration record at the top of each model file (Extracts shown in Figure 4b). It was then simply a matter of writing the “deck” out from Excel as an appropriately formatted text file which could then be directly read into HyperMesh as a RADIOSS bulk deck; ready to run.

Figure 4b: Extracts from Excel based master modelling tool

Powertrain type acronym

Mount choice controlled in Excel by engine type acronym:

IF/OR/AND logic chooses “$” or

“Include”


3.3 Detailed Sub-Models of Hydromounts

The hydromounts, used as engine mounts and as subframe or suspension bushes, were modelled as subsystems. These models were created using an Excel based technique whereby a parameterised model is correlated to a rig test result, prior to being written out in RADIOSS bulk format. The model can be manipulated interactively by means of slider bars which control the individual mass, stiffness and damping parameters. A “best fit” correlation can be obtained by using the solver to minimise the least squares difference between the test curve and the prediction. Typically, the sliders are then used to explore the effect of altering the mount design.

Figure 5: Excel based hydromount modelling tool Within the limitation of representing fixed amplitudes, these models very closely replicate the rig test results. It should be stressed that the resultant model is only meaningful if it is used in the context of a fully complex solution. For example, the moving mass component

Excel mount model stiffness compared to rig test data Magnitude Out of phase

Solver parameters Interactive controls

Least squares difference


(representing the fluid) is connected to the top of the system via a scalar spring, but to the base only via a damper. Similarly, the main structural spring path has parallel hysteretic and viscous components, none of which would be meaningfully used in (for example) a real modal model. Most of the model bushes and hydraulic devices were created using complex models based on actual test results. In the simplest instance, a bush would be modelled as a scalar spring with a viscous damper and a hysteretic loss factor. The more complex hydromounts and hydrobushes were all modelled using the technique described above. In the design study and “what-if” investigations, the Excel based model was modified such that the parameters were inter-related in a way that reflected the mount designs under consideration, e.g. equations were added to make the hydraulic “puffing” stiffness a function of the mount base stiffness etc. This constrained the optimization process to produce realistically bounded parameters that would be physically possible to manufacture.

3.4 Model Excitation and Forced Response, use of Unity Excitation and Wheelbase Frequency Filtering

The industry standard approach to vehicle NVH modelling is to develop models of existing vehicles, correlate them to modal and operating data, and then modify them to predict the effect of changes. Judgements are then made regarding the likely effect on passenger comfort. For a current vehicle the technique is then to try to understand how the predicted changes will relate to any known concerns, by comparing graphs of test and model results. A degree of sophistication may be introduced by the use of optimization techniques, where particular response features are targeted for reduction, but the final analysis is still comprised of a visual inspection of the results. As confidence is gained in the reliability of the modelling techniques, changes are made to existing models to represent future vehicles, but in order to make sense of results, predictions need to be absolute rather than relative. The predicted results are then compared to standards or company targets. This requires, in the case of road induced shake, the simulation of a road surface. The analyst is then faced with the choice of enforcing measured motion at the wheels (as for an endurance rig test), or applying boundary enforced motion at the road surface. The former approach inevitably embeds the suspension dynamics of the measured vehicle, and so is not suitable for use in future vehicles, or in those with suspension modifications (i.e. any potential suspension response changes will be prevented by the enforced motion). The latter approach is only reliable for a particular road surface, but is more rigorous. An alternative method is to apply unit excitations to the modelled road surface boundary, and to only predict changes i.e. a set of “filters” (as described in section 2.2, the application). Since the excitation function set is effectively in both the numerator and the denominator of the filter, it is cancelled out, and its absolute level becomes irrelevant. Thus unit load cases were developed to represent steady speed, front wheel impact and rear wheel impact, to be replayed at chosen speeds. Having demonstrated that the excitation amplitude is (in this particular context) unimportant, it is noted that the accurate representation of the phasing between the contact patch points is essential (both within the length of the individual patches and from front to rear axle). Figure 6a shows an example derivation of a phase delay from a front to a rear axle.


Figure 6b shows historical measured data illustrating the “wheelbase frequency filter” effect, whereby the phasing of excitation between the front and rear wheels combines to create peaks and troughs in the frequency domain. Given that the spacing of these features is directly influenced by the vehicle wheelbase and the speed, it is also important to investigate a range of speeds to avoid a potential resonant response being cancelled out by the filter.

Figure 6a: Derivation of phase delay and wheelbase frequency filter (fictitious example only)

Figure 6b: Typical measured wheelbase frequency filter (historical data from SVT database)


As mentioned above, the technique was extended to include the creation of filters for light impacts. This involved the production of independent filters for the front and rear axles. The test results were then manipulated to isolate the response to the individual axles prior to the application of the filters. Whilst this lacked the mathematical rigour of the constant speed work, it has proved sufficiently accurate to provide indicative results; provided that the simulation speeds are slow enough that the response to the front impact has decayed sufficiently before the rear impact is initiated, i.e. they are approximated as essentially independent events.

3.5 Filter Production

The CAE model forced response results were manipulated in HyperGraph to produce the raw filters. This process was controlled by the use of a template file that read in the numerator and denominator files and performed the complex division. The page (see Figure 7 for a typical example) was then written out as a CSV text file, where it was read into Excel to be smoothed and re-formatted for use in the simulator control software.

Figure 7: Typical filter set created in HyperGraph

It should be stressed that the filter results do not represent a forced response, but they are the change that could be expected from the introduction of a model modification. Examples of the typical modifications to vibration time histories and spectra before and after applying the calculated change filter are shown in Figures 8 and 9. Changes such as these can be quite subtle, and difficult to quantify when presented in graphical form. However, they can produce a distinct shift in the character of the vibration which can be perceived in the simulator.


Time response

-1.000

1.000

-0.5000

0.00

0.5000

0.00 1.000 0.2000 0.4000 0.6000 0.8000Time (seconds)

Figure 8: Typical filter applied to vibration time history

Spectrum

0.0

0.4

0.1

0.2

0.3

0 40 10 20 30Frequency (Hz)

Figure 9: Typical filtered modified and original vibration spectra

3.6 Concern Investigations and Development

It can be said that a test result will show what is happening, where the mathematical model will show why it is happening. And so the model has been used for more than simply trying out a series of potential design modifications. It has been used to develop a detailed understanding of some of the interactions between the different powertrain and chassis systems, and the influence of the various elastomeric and hydraulic devices. Further, it has been used to develop the mount system parameters which will minimise road surface induced shake, within design boundaries imposed by practical considerations. This has been done in parallel with guidance from listening studies to estimate the effect of the mount stiffness changes on powertrain sound quality.


4.0 Experience with the Process

During the course of the project, the simulation process has been used to produce in excess of 150 individual simulations, all of which have been “driven” by engineers and used to gain directional guidance in the choice of powertrain and chassis mounts in the context of their effect on both road induced shake and powertrain sound quality. A series of “correlation” exercises have been undertaken, whereby the guidance from the simulation process has been compared to that gained from road and track testing. The conclusions to date have been that the process works. It is not perfectly mathematically rigorous (ref the application to light impacts etc), but rather it is a deliberately pragmatic approach which seeks to make the best use of limited time, resource and budget that is available in a fast moving development programme.

5.0 References

[1] 'Interaction of vibration and sound stimuli in transient impact events', F. Syred, J Hargreaves, J Richards, EIS seminar “Right Sound your product” December 2009.

[2] Altair HyperWorks 2011

The Use of System Modelling Techniques to Filter Test Measurements and Drive a Physical Vehicle NVH Simulator

Technology

virtual scenario

system modelling

filter test

unit mounting

road induced

modify test

powertrain

powertrain