5 Checking the model with artificial excitations

Getting high-quality source characterization results is key for NVH engineers. To achieve good quality, it is essential to detect setup and signal problems and ensure the system is modeled correctly.

We recommend performing artificial excitation measurements at the very beginning of the testing campaign. They should be the first measurements after building up the test setup. You use artificial excitations for two different quality checks:

1. In DIRAC, you verify whether the test setup is instrumented correctly by playing the operational deflection shapes (ODS) of the artificial excitations and observing the movement of the sensors.
2. In SOURCE, you calculate the Blocked Forces produced by the artificials and perform an on-board validation. You check whether all transfer paths are taken into account in your model. Additionally, you can perform a transfer validation (read more here) with these blocked forces to check whether the receiver is independent of the source characterization.

How to: Define artificial excitations

An artificial excitation is a simple and noise-free simulation of an operational excitation. Practically, an artificial excitation is an excitation on the active component in a controlled (and measured) manner, e.g., with an impact hammer.

Artificial excitations can be performed in multiple ways. An efficient method consists of doing an FRF measurement in DIRAC with a hammer when the component is not switched on. You simply take your hammer and perform impacts in predefined locations and measure the data in DIRAC. This is convenient because:

1. There are no secondary excitations (noise sources).
2. A hammer hit produces a 1 DoF excitation.
3. The data processing is easy.
4. The data is stored in the DIRAC project.
5. The data is normalized (to 1N), so the measurement is comparable and transferable to other test setups.

Here are some recommendations for performing artificial excitations in DIRAC using a hammer:

• Perform artificial excitations by hammering on the active component(s).
• If you have multiple vibration sources (e.g., 4 wheels), perform artificial excitations on all of them.
• If possible, perform artificial excitations in all directions as you expect in the operational excitation.
• Place the excitations at locations where they can best simulate the operational excitation.

How to: Perform artificial excitations in DIRAC

To perform artificial excitations in DIRAC, complete the next steps:

1. Add the excitations to the design of experiment in Prepare.
2. (Optional) Type “Internal” in the description field of the artificial excitations.
3. (Optional) Rename the excitation adding the location of the excitation in the description field.
4. Go to Measure and perform the measurements (as you would for any other excitation).

How to: Calculate component TPA predictions for artificial excitations

After measuring in DIRAC, you do the post-processing in SOURCE. For artificial excitation analyses, you do not need to import any operationals because all the necessary information is stored in the FRF dataset. For this reason, you will need to use a Fluent TPA script for the analysis. (Read more about Fluent TPA here.)

Step 1: Create the matrix preset

To analyze the artificial excitations in SOURCE, you need to import FRF data containing the artificial excitations, the characterization FRF and the synthesis FRF. To do so, it is recommended to create in DIRAC a custom preset with the following:

• Response channels:
  • Indicator responses.
  • Validation/Target responses.
• Reference channels:
  • Artificial excitations.
  • VP transformed forces and moments.

Step 2: Set the correct channel types

To perform artificial excitation analysis in SOURCE, you must set the channel types correctly. The channels must be set as follows:

• Indicator sensor channels: Indicator responses.
• Target/Validation sensor channels: Target or validation responses.
• Artificial excitations: Internal forces.
• VP transformed forces and moments: Interface forces.

Step 3: Import/Write the fluent script

The artificial excitation analysis is performed in SOURCE by running a Fluent TPA script. You can find the script on the VIBES Portal. You can create a new Fluent TPA analysis and then import the script, or write it yourself, following the instructions in the picture. You then select the correct inputs (FRF dataset and channels) and compute the analysis.

How to: Check the model with artificial excitations

To check the quality of the model with artificial excitations, you perform an on-board validation, comparing measured data with predicted data using Component TPA.

If you have a good signal-to-noise (read the relative article here) and no setup or signal problems, the validation and measured curves must match very well.

How to: React to bad on-board validation with artificials

The measured and prediction curves for on-board validation of artificial excitations must match very well. If this is not the case, there are three possible causes:

1. The test setup has problems.
2. The signal has problems.
3. The model has problems.

If you have followed all steps and performed all quality checks we suggest in the test-based modeling (link) and source characterization testing workflows (link), you can exclude setup and signal problems. This is because you have already checked the test setup and signal.

If you have not performed all suggested quality checks, this is the time to do so. It is very likely that the issue is caused by problems in the test setup (e.g., test setup does not match the design of experiment, wrong impact locations, wrong VP assignments, etc.) or with the signal (e.g., noise, overloads, broken/damaged sensors or channels, etc.).

If you can exclude signal and setup problems but the on-board validation for artificial excitations is still not good, then the issue are likely to be found in the model. Some possible reasons are:

1. Underfitting. Your Blocked Forces do not have enough DoFs to describe the system completely. There are two cases for this:
   • The interfaces are missing some degrees of freedom. To overcome this, you should include more DoFs per interface (e.g., add rotating if they were excluded).
   • You are entirely missing one or more transfer paths (interfaces). To overcome this, add more Virtual Points to the system (e.g., one VP per interface).
2. Overfitting. Your Blocked Forces have too many DoFs, so you are fitting measurement noise in the model. Use X-DoF or remove/exclude DoFs.
3. Non-linearity. Your system does not meet the basic requirements for applying the technology.

Remarks

It is essential to mention that getting good results with artificial excitation doesn’t guarantee that the same modeling approach can be used to also describe an operational excitation with the same accuracy. The artificial excitation is a very simple load case. And also, the operational measurements needed for characterizing operational excitations are more prone to secondary excitations. For this reason, good artificial excitation results do not directly ensure the same high quality for the operational excitations.

For this reason, a good first step consists of analyzing more than one artificial excitations, which excite different DoFs. However, it might still be the case that all single DoFs are modeled correctly, but a simultaneous excitation of multiple DoFs will not reach the same high quality.

How to: Check the test setup

Artificial excitations can also be used to check if the test setup is instrumented correctly and if there are issues with sensors, channels, etc. This can be done in DIRAC in Analyze, checking the ODS of the artificial excitations at low frequency. When playing the ODS, all sensors should move consistently. If that is not the case, you should investigate the cause.

You can find more information about checking ODS in the relevant article.