of sound radiation from electric motors

Transcription

1 Titelmasterformat Mid-frequency challenge durch Klicken efficient bearbeiten simulation of sound radiation from electric motors M. Moosrainer, M. Jegham, CADFEM GmbH , DAGA 2018 Munich

2 ANSYS simulation platform Complete Systems Common data models Common IT infrastructure Systems Platform Multiphysics Acoustics Structures Fatigue Fluids Magnetics Electronics Power Cooling Magnetics electronics Software Controller CAD/ PLM CAD-Interfaces Geometry preprocessing Multiphysics Workflows High Performance Computing Parametric Optimization 2

3 DAGA 2017 revisited: electric drives as a source of noise Wikipedia Electro Mobility Wikipedia Marine Propulsion Power Engineering Industrial Drives Schottel Vacuum Cleaner, Fan Universal Motors, 3

4 Concept of FEM-based noise computation: workflow ANSYS Mechanical with Electric Drive Acoustics inside ANSYS Electromagnetic Analysis EM excitation loads DFT Excitation Loads Harmonic Vibration Analysis Oscillation, ERP, Waterfall Plot 4

5 Concept of FEM-based noise computation: acoustic results n Identify critical OPs ERP [db] Vibration shapes for critical OPs Radiated acoustic sound field f DAGA 2017: vibration ERP workflow DAGA 2018: acoustics workflow 5

6 Torque [Nm] Torque Concept of FEM-based structure-borne sound computation Equivalent Radiated Power (ERP) as a rough estimate mean-square structural normal velocity v n on radiating surface A advantage: returns a fast figure of generated noise indicates critical operating points efficient comparison of designs performance factors electromagnetic excitation: rpm interpolation for complex load spectra structural vibration: mean square v efficiently computed in modal subspace acoustics: no acoustic field calculation, no meshing of fluid space 1 PERP c vn 2 A da ERP: radiation efficiency σ =? air-borne sound radiation in the following! n 1 n 2 n 3 n 4 n rpm Speed [rpm] Wibbeler, J; Moosrainer, M.; Hanke M.: FEMbasierte Körperschallanalyse für elektrische Antriebe. In: Fortschritte der Akustik (DAGA 2017), Dt. Gesell. für Akustik e.v., Kiel (2017). 6

7 Helmholtz number: non-dimensional scale for sound radiation tasks Helmholtz number Electric Hermann motor von Helmholtz a=0.2m He k a 2 a f=100 Hz, λ=3.4m, He=0.4 a λ Rule of thumb He < 1: low frequency problem He 10: high frequency problem f=1 khz, λ=0.34m, He=4 f=10 khz, λ=0.034m, He=40 7

8 Mid-frequency challenge for mesh-based acoustics simulation f max controls element size min c f max 10 min, ESIZE f min controls acoustic domain size due to absorbing boundary conditions ABC max c f min 4 max, R Acoustic Cavity E-motor R(f) R wide frequency range f min < f < f max large acoustic domain R & small elements ESIZE huge number of DOFs exponential increase in computational resources idea: f parameterization 8

9 Acoustic HPC performance optimization: frequency sub-ranges Table of parameterized frequency bands full model for total freq. range 3 Mio. DOF's smaller range f min to f max for each band smaller air domain R(f) larger elements ESIZE(f) smaller number of DOFs more efficient solution in each band freq. band DOFs freq. band DOFs freq. band n DOFs 9

10 ANSYS HPC Mesh Domain Decomposition (MDD) decomposes the problem into n subset of elements each mesh group is computed by one core Mesh domain 1 Core 1 n subsets 1 freq point Mesh domain 2 Mesh domain 3 Mesh domain n MDD smaller sub-domains distributed on all cores less RAM is required per core solves one frequency at a time huge MPI data communication Core n 10

11 ANSYS HPC Frequency Domain Decomposition (FDD) decomposes the problem into n frequency bands each band is calculated by NProcPerSol cores NSUBST,50 with 100 CPU cores (-dis -np 100) & DDOPTION,FREQ, NProcPerSol =2 50 parallel sets of calculations, each working on a frequency point using 2 cores for MDD (2 groups of elements per frequency). Freq1 1 Freq 2 Core 1 Core 2 n freq. bands 1 freq band Freq 3 Freq 4 Freq 5 Freq 6 large amount of RAM is required solves one frequency band at a time some CPU's may remain idle in FDD very little MPI data communication Freq 7 Freq n Core 1 Core 2 11

12 Acoustic HPC performance optimization: balancing of MDD & FDD 14 CPU, 180 GB RAM khz, rpm 75 solutions frequency bands: no MDD: no FDD: no frequency bands: no MDD: yes FDD: no frequency bands: yes MDD: yes, balanced FDD: yes, balanced # DOFs 3 Mio. 3 Mio. 0.6 Mio used CPU Elapsed time [hours] RAM [GB] Elapsed time [h] RAM 12

13 Acoustic HPC performance optimization: high-frequency low 10 Hz to high-frequency 10 khz: 30 steps from 500rpm to 5000rpm: 10 load cases elapsed time 7 hours for 300 solutions on a machine with 14 CPU s and 180 GB RAM 13

14 SPL at 1 khz 14

15 Structure-borne (ERP) vs. air-borne sound power level: 500 rpm 15

16 Structure-borne (ERP) vs. air-borne sound power level: 5000 rpm 16

17 Air-borne sound analysis: resulting radiation efficiency σ 1 L 10log [db], complicated behavior no good experience with look-up tables for σ 17

18 Waterfall Diagram: 0-10kHz and n [ rpm] 18

19 Conclusions and outlook for electric drive acoustics electromagnetic structural vibration acoustics workflow from structure-borne (ERP) to air-borne sound radiation ANSYS efficiently computes multi-rpm air-borne sound up to 10 khz SPL waterfall diagram from 500 rpm to 5000 rpm and from 10Hz to 10kHz within 7 hours elapsed time on a computer with 14 cores and 180 GB RAM ANSYS Mechanical HPC performance optimization based on optimal selection of absorbing boundary condition parameterized frequency f, element size ESIZE(f), acoustic domain size R(f) optimized balancing of MDD & FDD adapted to model size, freq. range & hardware outlook to DAGA 2019 acoustic model order reduction for further speed-up 19