How to compare (correctly) e-NVH simulations and tests?
EOMYS | Joins the SIMULIA Brand of Dassault Systèmes
Manatee software | noise & vibration control of electrical systems | Now part of the SIMULIA brand of Dassault Systèmes
Next time right!
The "zero prototype" or "first time right" mottos were probably not invented by real-world engineers... At Eomys , our experience shows that building a realistic and comprehensive virtual prototyping workflow of acoustic noise and vibrations due to electromagnetic excitations requires several iterations between tests and simulation - and that's a great opportunity to leave our cold, silent screens for a warm, noisy hands-on experience at the workshop! Electrical machines do produce noise and heat, and e-motor designers are aware of the multiphysic nature of their discipline. Magnetic noise and vibrations (called "e-NVH" in automotive applications), is indeed sensitive to electrical (e.g. current harmonics), magnetic (e.g. B(H) curve), mechanical (e.g. damping), acoustic (e.g. reverberation) and heat transfer (e.g. magnet temperature) variables.
Multiphysics is one thing, but what makes it even more difficult to correlate simulation models with reality are strong interactions between the physics. This article reviews the few reasons why your e-NVH simulation results, even run with Manatee software , will NOT match with the absolute values of your test results at FIRST simulation run.
When apples meet oranges
Comparisons must be carried on comparable figures, but for acoustics that's easier said than done... Manatee V2.2 e-NVH levels can be obtained as sound power level (SWL), or as sound pressure level (SPL) of the simulated motor.
Electrical machines are usually tested on a dyno testrig designed to assess electromagnetic and cooling performances. The sound power level of the tested machine should not include the load machine noise. This can be achieved with different techniques, such as placing the load machine outside the acoustic chamber or using acoustic intensimetry measurements. Moreover sound pressure is naturally measured with microphones, but it highly depends on the position of the microphone and acoustic environment (e.g reverberation effects).
The best is therefore to run sound power level measurements according to acoustic standards (e.g. ISO 3745:2012, ISO 3744:2010, ISO 3746:2010). This means more complex testbench set-up, longer measurement and post processing times, and that's why it is so tempting to compare pressure levels in different conditions, or to compare power level with pressure levels... In that case, you may obtain 15 dB of difference between apples and oranges!
Real world doesn't go round
Many assumptions are unconsciously made when running multiphysic simulations, especially on the symmetries of the system. However, a certain level of eccentricities and geometrical / magnetic asymmetries is always present in an actual machine due to the manufacturing tolerances. It can introduce new e-NVH resonances due to new magnetic force harmonics. As an example, eccentricities modulate all pulsating forces with additional Unbalanced Magnetic Pull harmonics. If you have simulated a fully symmetrical machine in Manatee, some resonances may be missing compared to experiments, resulting in up to 15 dB differences. The following measurements are therefore recommended:
Note that some asymmetries do not necessarily change the frequency signature of magnetic forces, so they are not visible on a simple vibration spectrum!
The sin of sine
We all start the design process of e-machines running sinusoidal, current-driven electromagnetic simulations... Transient, strongly coupled electrical to magnetic simulations are difficult to be run during early iterations loops and are generally kept for the detailed design phase. However, current harmonics are the source of flux harmonics, which in turn produce force harmonics. Therefore, you may find differences between your ideal sine-driven simulation and experimental results. These differences can be due to:
In that case, it is recommended to measure the three-phase current waveforms and apply them inside a current-driven simulation to quantify their impact on e-NVH.
Damping is not cheating
When magnetic forces resonate with a structural mode, the vibration velocity level (and therefore sound power level) strongly depends on the modal damping value - which no FEA software, no matter how sophisticated it is, can correctly estimate. Modal damping depends on structural mode, temperature, resin type, winding type, you name it... Based on Eomys experience, it varies from 0.5% to 4% in electrical machines, so Manatee e-NVH software uses a default average damping value of 2%. This means that calculations may lead to a -12 dB underestimate (if "real" damping is 0.5%) up to a +6 dB overestimate (if "real" damping is 4%) at some frequencies.
That's why a step-by-step Experimental Modal Analysis is highly recommended to quantify modal damping on your own application. So do not hesitate to input modal damping in your Manatee simulation, the error on sound power level can then be brought down to roughly +/-3 dB.
领英推荐
"Garbage in, garbage out"
A 3D FEA mechanical model can be imported as a modal basis inside Manatee virtual prototyping environment. Large discrepancies between simulation and tests can therefore be obtained if the simulated modal basis is not representative of the reality. This can be due to the following issues:
Advanced effects such as non-linearities, gyroscopic effects, magnetic pre-stress can also affect the dynamic behaviour of some particular designs.
Simulation, more real than experiments?
What you see is not what you hear! The nice pictures you obtain during a run-up with your Dynamic Acquisition System are obtained after complex signal processing including filters, Short Time Fourier Transform (STFT), RPM extraction and order tracking analysis. All these algorithms have some specific parameters which can have a significant impact on sound level. Besides test set-up accuracy, test results should also account for post processing accuracy.
In particular, STFT used in spectrograms requires a trade-off between time accuracy and frequency accuracy. Order extraction must be done integrating energy on a given bandwidth. It is advised to do a sensitivity study on these parameters before comparing dB levels from tests and simulation. Therefore, NVH measurements accuracy can be assumed to be within +/-1 dB even when following standards.
Manatee e-NVH software uses anti aliasing filters, non-uniform Discrete Fourier Transforms and convolution properties to reduce as much as possible numerical artefacts. This means that the acoustic energy is spread over a discrete set of frequencies, in line with e-NVH physics. This gives a more realistic spectrogram compared to what you would obtain from measurements!
A last piece of challenges
As you can see, comparing tests and simulations is a long journey... If you think you're done at this point of the article, here are additional points that can further challenge you:
Twisting the twin
As you can see, an e-NVH digital “twin” must face reality several times before looking alike. But that's a great opportunity for all engineers involved in the development of electrical systems to collaborate, share, and raise their engineering know-how, iterating between simulation model updates and analysis of new experiments. The physics of acoustic noise due to electromagnetic excitations is too complex to get it "first time right", and the digital twin must be partly test-driven.
Manatee software environment has been designed to make these iterations easier. It combines several modelling levels for quicker iterations, and predefined simulation workflows adapted to both design and validation engineering profiles. Contrary to general purpose FEA software, it also includes dedicated tools to troubleshoot and mitigate noise problems.
Besides, Eomys consulting team can help you to design and perform relevant tests to understand the e-NVH behaviour of your electrical system, and build a realistic virtual prototype. Technical trainings are also regularly organized online besides customized e-NVH workshops, we will be happy to meet you during one of these events!
?