How to further reduce Manatee e-NVH computing time?

Introduction

Manatee e-NVH CAE collaborative platform is complementary to general purpose multiphysic software suites. Due to its specialized algorithms, Manatee is much faster for pre-processing (multiphysic simulation workflow setup), running (solving of the multiphysic numerical problem at variable operating points) and post-processing (assessment of variable speed e-NVH metrics, troubleshooting and solving of e-NVH issues).

When using Manatee parameter sweep or optimization, one may want to further reduce computing time without degrading magnetic noise & vibration calculation accuracy. This article presents several methods to reduce e-NVH computing time.

Parallelization

Manatee can run parallel computing over several CPUs. Parallelization is used during parameter sweep, optimization, variable speed or electromagnetic calculations (MLUT calculation). The parallelization can be set-up in Manatee preferences. To increase the number of CPUs on your particular licence please contact EOMYS.

Electromagnetic Vibration Synthesis

Manatee EVS algorithm takes longer when including a high number of surfaces or load cases. The more information you get at the end of the results to troubleshoot e-NVH issues, the longer is the calculation. The calculation time can therefore be reduced with a smaller number of radiating surfaces and smaller number of magnetic load cases.

Electromagnetic model

The choice of operating points to build the Magnetic Look Up Table is automatically set in a conservative way. To reduce the number of operating points (typically 5 to 10 may be enough) you can manually set-up the MLUT.

The number of timesteps of the magnetic model can be tuned to focus on a given frequency range. Generally the higher is the excitation frequency, the lower is the magnitude of the excitation. To help you choosing the maximum frequency to capture you may use the Magnetic Force Signature Analsyis.

The mesh of the magnetic circuit can be made coarser provided that it does not affect the level of response variables. A sensitivity study can be carried manually on the force harmonics or noise harmonics that you track. To run such an analysis (not yet available in GUI) please contact Manatee support.


Structural model

When importing a modal basis from a 3D FEA mechanical model, the frequency range of the imported modes can be manually defined. If you have identified e-NVH problems in a specific frequency range [f1, f2] you can for instance restrict the frequency range of the modal basis import to [0.7*f1, f2*1.3] (30% margin) to speed up structural calculations.

Parameter sweeps

Parameter sweeps should only be carried on the most influential variables. Some responses may be not linear, preventing the use of Design of Experiments method to quantify the most influential variables and limit the number of design variables in parameter sweeps / optimization. If you assume a linear behaviour with respect to response variables, you may start quantifying the influence of design variables by running a two-level parameter sweep (+/-5% on each design variable) to identify the most influential design variables.

To reduce the calculation time of parameter sweeps, you can reduce the number of input variables or the number of different design variable values in a given range. Instead of running parameter sweeps at variable speed, you may also restrict your e-NVH simulation to a certain number of operating points where NVH hot spots have been identified.

During parameter sweeps, make sure that you do not re-run mechanical calculations when it is not necessary (for instance, changing the slot opening may not change significantly the modal basis of the electrical machine, so you may want to import the same modal basis during the optimization).

Optimization

Optimization should only be carried on the most influential variables. Some responses may be not linear, preventing the use of Design of Experiments method to quantify the most influential variables and limit the number of design variables in parameter sweeps / optimization. If you assume a linear behaviour with respect to response variables, you may start quantifying the influence of design variables by running a two-level parameter sweep (+/-5% on each design variable) to identify the most influential design variables.

The computing time of optimizations relying on evolutionary algorithms can be minimized by reducing the size of population or number of generations. A large population size is only required when dealing with a high number of objectives or when you need a fine discretization of the Pareto front.

Instead of running optimization at variable speed, you may also restrict your e-NVH simulation to a certain number of operating points where NVH hot spots have been identified.

During the optimization, make sure that you do not re-run mechanical calculations when it is not necessary (for instance, changing the slot opening may not change significantly the modal basis of the electrical machine, so you may want to import the same modal basis during the optimization).