An integrated approach to engine radiated acoustics

With Virtual.Lab Motion, it is possible to run a multibody simulation on an engine – including flexible bodies for e.g. crankshaft - for run-up conditions and to obtain accurate loadings. Alternatively it is possible to obtain these excitations using similar software’s. Virtual.Lab provides all the necessary interfaces to import and convert the data into relevant data for NVH purposes, rpm engine-run data, single rpm frequency spectra or range of engine orders. Typically the original excitation data is time-based. Using the FFT functionalities inside Virtual.Lab it is possible to convert these into the frequency domain, automatically for a complete engine run-up.

Once accurate, frequency based loading information is available, the LMS Virtual.Lab solution is split into two stages. The first is a structural vibration part, from these results you can already achieve the sound power density, the total input power and you can also perform some frequency band tracking and order analysis. The latter is purely based on efficient post-processing of surface vibration data. The second stage is the acoustic part where you will use the boundary element technology to accurately predict the acoustic radiated sound. This process utilizes the ATV technology to reduce computation time considerably.

Virtual.Lab then gives you the advantage of being able to record the process you’ve just followed and repeat it for a different set of loads or even a different engine. This is possible right throughout the entire structural and acoustic stages. A typical process below shows the steps that are taken to perform these stages

The total setup time for this is very fast. Generation of the coarsened mesh can be done in hours with our mesh coarsening technique. Once all the mesh and modal information is available, the setup time is about 30 minutes.

The structural mesh in most cases contains all too detailed features such as ribs or small holes. Although

these features are very important for the structural vibration calculations, they can be omitted when doing the acoustic BEM calculation. This will reduced significantly the complexity and size of the BEM models.

The structural mesh is also typically a combined shell solid mesh. The advantage of having a shell mesh is when modification is needed – e.g. oilpan thickness - you only need to modify the appropriate properties rather than the entire mesh.

Computing the structural modes can result in very large files, sometimes many gigabytes of data in binary format. LMS Virtual.Lab has been designed to reduce as much as possible the amount of structural data without compromising accuracy. Instead of calculating the engine vibration on all nodes of the engine, it is possible to just create the output points relevant for NVH, essentially excitation points and nodes on the surface, the so-called wetted surface. Generating that wetted surface and setting up the structural solver to deal with the wetted surface is straightforward.

The powertrain excitations are function of: 

• Location: crank bearing, combustions, etc
• Engine load conditions: whether different rpm ranges, single rpm or an order set
• Frequency: typically from 20Hz to 3000Hz or more

LMS Virtual.Lab allows to import a full set of excitation data, convert it into the frequency range and store it into Load Function Sets. Likewise excitation data can directly come from Virtual.Lab Motion in which case, the data is directly available to be used for forced response analysis. Analyzing the load data can be done using dedicated NVH displays such as Colormap displays, complex function displays and more. In the case of multi-rpm run-up excitation data, Virtual.Lab offers the possibility to extract order data for further analysis. The particular advantage is that for order-based data excitation, less data points need to be calculated compared to a full rpm full frequency data set. LMS Virtual.Lab contains forced response solvers. These can be modal-based or FRF-based. For this application, the modal-based forced response is advised. Setting up the calculation is straightforward: as a user you refer to the loads (Load Function Sets) and to the modes (Mode Set). If needed, you can indicate graphically which points on the engine need to be retained for the vibration analysis. When doing a full acoustic analysis, it is not needed to calculate the vibrations on the surface. In fact, Virtual.Lab Engine Acoustics uses a modal approach where essentially only the Modal Participation Factors are needed. This modal technique results in a significant data reduction without compromising accuracy. Typically you would retain a few points on the engine for vibration processing. The modal participation factors can directly be used in the acoustic forced response analysis (Modal ATV Response Analysis).

If you prefer not to do a full acoustic analysis, but if you prefer to analyze the surface vibration data, then this is likewise straightforward: using the grouping functionalities of Virtual.Lab you create groups of the surface and request output vibration results in these points. Virtual.Lab will calculate these automatically for all loading conditions (rpm’s or orders) and for all frequencies. Once the vibration forced response is done, you have 2 options. Whether you perform detailed acoustic analysis using the BEM technology or whether you analyze the surface vibration data. Virtual.Lab is scalable to support both.

After the forced response calculation, you can start analyzing the resulting data for overall NVH performance of the engine. Essentially, analyzing surface vibration data will allow you to efficiently identify potential problems areas such as critical rpms, frequencies, orders or locations on the structure of high sound radiation, even without performing a detailed acoustic simulation. A wide variety of post-processing tools is available.

In addition to the response analysis of particular points on the engine, it is also possible to create vibration data images on the 3D model and furthermore perform running mode analysis. The latter allows scrolling through the frequency range and actually seeing the vibrations of the engine, giving you more insight into problematic areas. Conversion of vibration data into sound power is possible, including automatic integration in bands if needed. Band selection for frequency integration is done by a dedicated user interface.

Once you finished the structural vibration calculation, you have the option to continue on to the acoustic section. For this you have to generate the coarsened mesh. When generating an acoustic BEM mesh the first thing you need to do is to skin your solid elements, after this you are left with shell elements both inside and outside the boundary of the mesh. You can then manually or automatically close any open holes. From this point you can already perform an acoustic simulation. But removing the ribs from the structure can save you a few hours in computation time. This process in the past has taken some people several weeks to complete on a full powertrain model. Which as you can image is too long to effectively influence the design process. Virtual.Lab has a tool called mesh coarsening that allows you to do all this in hours, and not just clean the mesh up, but also generate a totally new mesh that is specific to the frequency range you are interested in. This is what we call a wrapped mesh. For example, if you set the frequency range to 2000 Hz, Virtual.Lab will create a Quad dominant BEM mesh valid up to 2000 Hz, with elements of averaged size of 29 mm, corresponding to 6 elements / wavelength.

Simulations were performed to check the validity of the acoustic mesh. The results indicated that there was no loss in accuracy as long as the 6 BEM elements / wavelength criterion is kept. The upside is a significant reduction in calculation time by using a mesh optimized for a certain frequency of analysis.

After you have generated the coarsened mesh you need to define the field point mesh, also known as microphone positions. Virtual.Lab has tools for generating the field point mesh for cylinders, planes, spheres, ISO standard meshes for sound power calculations, directivity field point meshes, etc.

After the generation of the field point mesh you need to compute the Acoustic Transfer Vector (ATV) database. The traditional Boundary Element approach uses the structural vibrations directly to define the boundary conditions for the acoustic radiation problem. The drawback of this approach is that these boundary conditions vary with loading conditions and so a different solution must be found for each loading condition.

The LMS Virtual.Lab implementation uses Acoustic Transfer Vector (ATV) technology. ATVs are transfer functions that link the structural vibrations of the radiating surfaces and the Sound Pressure Levels at the desired output field points. They are dependent only on the geometry, mesh density and media characteristics of the acoustic domain, the acoustic surface characteristics (impedances and admittances), the frequency and the location of the field points. They are independent of the loading, which means that they are especially well suited for multi-case analysis such as engine run-up and the optimization of structural design parameters.

This represents an enormous advantage, in that loading and design parameters (not altering the topology of the FEM) can be varied without having to run original solvers again. Another major advantage is that the ATVs can be calculated for wider frequency bands compared to the structural calculations as the ATVs are smooth functions of frequency. Depending on engine geometry, we advise to use a frequency spacing from 5Hz to 25 Hz. Taking larger frequency bands results in corresponding time saving for solving the BEM model.

NOTE: in case you need to perform an acoustic analysis with at only one load condition, for one design, we

advise to use the High Speed BEM solver (using integrated pade expansion techniques) in order to speed up the calculations. Typically for engine calculations, the speed up gained by a High Speed BEM solver is significant, i.e. up to a factor of 10 to 50.

Looking at the formulas we know that;

{Sound Pressure} = [Acoustic Transfer Matrix] . {Surface Velocities}

We also know that the surface velocities are normal component of the structural velocities since only this normal component plays a role in the generation of sound waves. We can then rewrite the formula to look like this

Therefore the Acoustic transfer vector concept is nothing more than an assembly of Acoustic transfer functions relating to the normal vibrations on the surface of a mesh to the sound pressure at a single microphone location, as illustrated below.

So we know that



We also know that

{velocity boundary conditions} = [modes] . {modal participation factors}

Also the structural displacements {u} can be represented by:

If you then combine both equations, we obtain:

If we want to obtain a direct input-output relation between the structural model response vector and the microphone point (field point) sound pressure we use

P = {Modal Acoustic Transfer Vector}T . {MRSP(w)}

Where the Modal Acoustic Transfer Vector (abbreviated to MATV) is defined as:



This is then interpreted as an assembly of acoustic transfer functions relating the contribution of the individual structural modes to the sound pressure at a single microphone location.

In contrast to ATV’s, Modal ATV’s (MATV’s) cannot be computed purely from acoustic parameters but in addition require knowledge of the dynamic behavior of the vibrating structure in the form of structural mode shapes. The entire process is illustrated in the figure below:

The ATV-based process has a huge advantage in that you can store the ATV database and reuse it again as long as you don’t make any changes to the acoustic model. For example, use it for different loading conditions. Even the structural model can be modified with some degree, without the need to re-run the BEM model: just re-use the existing ATVs. After the modification you simple perform the data transfer analysis case again and re-compute your acoustic response, which is just a multiplication; no matrix inversion is required. In case the structural modes are not changing from one run to another, which is the case for different loading condition, then the data transfer of the modes is not needed either: you just re-use the MATVs, which is a very reduced set of data as the dimension of the MATVs correspond to the number of microphone positions (size of field point mesh) times the number of structural modes. Again a significant increase in productivity using MATVs.

The Data transfer analysis is a key process in the setup of your analysis. This allows you to transfer the detailed structural modes over to the coarsened mesh. The software first makes a bridge between the structural mesh and acoustic coarsened mesh using the Mesh Mapping tool. It then allows you to select the structural modes and transfer them over. After the transfer you will have a reduced set of structural modes on the acoustic mesh. These modes will then be used in your Modal ATV response analysis.

The mesh mapping is an automatic process to link the incompatible structural and acoustic mesh together. Being able to work with incompatible meshes is a great advantage as it allows to reduce the acoustic model size significantly compared to a compatible mesh requirement.

This method is also flexible in that you can modify any of the mesh mapping parameters. To do that it is better to understand what is happening. Mesh mapping links every node on the acoustic mesh to ‘X’ number of nodes on the structural mesh. You can define the search criteria that it has to use by giving in a radius ‘Y’ and the number of nodes you want it to link to ‘X’. It will then scan the envelope of structural mesh for the nearest nodes to link to and using RBEs link the nodes together. The nodes are linked with a weighting factor so when you do transfer over the modes using the data transfer analysis the weighting factor is taken into account. Virtual.Lab as default takes 4 nodes and 10mm. This is normally enough to have a good coupling, but it all depends on your coarsened mesh. Within Virtual.Lab you can generate an image of the min or max distance between the coupled nodes of the structural and acoustic mesh.

The standard ATV based engine acoustic analysis is based on a modal approach, because the data reductions that can be obtained are very significant. Yet, a direct vibration approach is also possible in case this would be required.

The modal Based ATV response analysis case combines all the information you have prepared to perform the radiated acoustic analysis. As you have seen above the formula uses the modes from the structural modal and the modal participation factors which are automatically computed from the modal based forced response.

You can define here the frequency range that you are interested in but as default LMS Virtual.Lab takes the RPM range or frequency range that was defined in the modal based forced response. The solution is efficient and takes minutes (order of magnitude). As shown on the left, from this solution, you have directly the possibility to:

Sound pressure levels around the engine can be plotted for particular frequencies of analysis, rpms or orders (see picture below). One also has the possibility to create multiple images and as such compare easily different cases. Additional operation can be performed on these images, such as integrated the pressure field in frequency bands (e.g. 3rd octaves bands), for all or individual orders or rpms.

As you have seen in the images above the visualization of data is very extensive, you can have 2D or 3D plots, perform comparisons between results and define the format and function of your plot (logarithmic, dB, real, imaginary, 3rd octave etc.). Virtual.Lab engine radiated acoustics also offers you tools to pin point weaknesses in the design and help you make engineering decisions. Some of these capabilities and tools are shown below.

The surface can be defined as a number of panels that represent different structural features of the radiating surface, and they can be easily defined using a number of identification criteria, such as feature angle, position, or material characteristics. Panel contribution analysis enables you to identify the contribution of each of these panels indicating which physical parts are radiating the most.

When assessing the effect of different structural modes, to the overall sound pressure level, the modal participation factor is not sufficient. The contribution to the global noise level is a combination of the modal participation factors, the vibration levels and the radiation factors. For a specific field point you can see a dedicated view on the modes in a frequency range where high responses are obtained meaning that the relative contribution of different structural modes can be assessed.

Understanding the critical modes is crucial; the insight gained into the structural behavior of the vehicle leads to the most effective solutions. The use of such a distinct tool is essential due to large number of modes and the higher modal bandwidth. This is displayed in a similar image to the panel contribution displays.

High levels of acoustic radiation are due not only to high loads (vibrations) but to easy transfer paths. So in

addition to the panel contribution one has the ability to identify critical paths whereby the source of the noise is transferred to the acoustic radiation provides crucially important information. Only by understanding both the levels and the paths can a truly effective optimization of the design be accomplished.

Whether there are distinct problems with target levels, or whether the objective is to improve the design, the goal is to use the analysis data to close the design loop. This can be done through scalable solutions, which involve increasing degrees of modification to the analysis data, the mesh or the structural model.

Modification of the analysis data represents the tightest loop in the design process associated with troubleshooting and problem solving, which would be performed when the design of the vehicle is fixed and it is necessary to make small adjustments. Fast modification prediction allows you to adjust the behavior of the structural modes by the placement of physical characteristics at key locations on the structure. Such physical modifications can be the addition of a lumped mass, the effect of increasing stiffness or damping. Due to the integrated approach within LMS Virtual.Lab, these modifications can be made and the analysis

results re-computed to determine the effect. Graphical representation of the modified locations and a graphical comparison of the ‘before and after’ response functions provide the information you need. This is a fast and efficient means to assess the effects that does not require any re-computation of modes by the original Finite Element solvers.

More significant modifications can be applied to the original FE mesh defining the structure. These modifications can include changing the properties of the existing mesh, by adjusting physical dimensions such as panel thicknesses, material properties or beam stiffness. The nature of the mesh can also be changed by the addition of elements, by changing the shape or by morphing the mesh. 
In both cases the FE solvers need to be re-run, but LMS Virtual.Lab provides drivers to run these external processes allowing you to remain within its integrated environment. The capture of the complete process means that mesh modifications are then carried through the entire analysis process.

Design changes can be more efficiently predicted using advanced substructuring techniques. The unique

approach in Virtual.Lab allows to modally reduce components and to physically connect these reduced components with FE models of the components of design interest. The reduced models should incorporate full dynamic information. Therefore, these reduced models contain rigid body behavior, free-free flexibly modes and the residual vectors in the connection points (bolts).

The full system dynamics are then described by components represented by FE – i.e. physical – properties and of components represented by modes. As some components have a modal representation, their size can be reduced significantly and as such, the total calculation time of the full system is significantly reduced. While as a full-system modal solution can take several hours in case no reduction is applied, the calculation time for the systems dynamics in case of reduction, can be an order of magnitude smaller. Even though some components are (modally) reduced, there is no accuracy drop wrt the acoustic signature. The MATV technology allows dealing with substructures. Typical application: optimization of oilpan or valve cover design wrt the overall noise level of the engine.

In today’s highly competitive environments, the optimum design is one that is close to the design limits. The LMS Virtual.Lab Optimization package offers an integrated set of powerful capabilities, tools, techniques that give engineers rapid insight into all possible design options that meet their requirements. All the capabilities are integrated into a module that allows users to specify the design objective, set design parameters and their distribution, automates, controls and monitors the optimization routines. The Design of Experiments (DOE), technique can be used to carry out virtual experiments, the results of which can be viewed using various Response Surface Modeling (RSM) techniques giving critical insight into design parameters and the tradeoffs involved.

Once an optimum is achieved it is important to investigate the robustness of the optimum, due to tolerances on the design parameters, input design parameters must be considered as distributions rather than single deterministic values. Variation of the design around its optimum values can be evaluated to meet robustness reliability and quality criteria.

Virtual.Lab Engine Radiated acoustics has an end to end solution that is also scalable. If you just want to analyze the vibrations or go into a full radiated acoustic simulation then you have a tool available to get you the most out of both. Such as the extensive and application specific post processing capabilities such as the

acoustic power color map displays.

Virtual.Lab also provides a cross attribute possibility where loading can not only come from test but also from CAE through our Virtual.Lab motion package. Loads can directly be transferred and linked to our radiated acoustic simulation such that any modification to the multi body simulation will directly influence the acoustics. This gives you the possible to watch design changes take place directly and with the ATV approach computation time is a matter of minutes rather than hours.

Combine that with the advanced post processing that gives you the possibility to find those weak points in your design, using as panel contribution, modal contribution and path contribution. All these tools allow you to pin point problem areas.

Then using our advanced substructuing approaches and fast modification prediction you can refine single components without having to re-compute the entire engine. This can again save tremendous computation time. Combine with the capability of Virtual.Lab optimization you have a very fast and effective tool to optimize your engine for acoustic radiated problems.