Automatic shape optimisation in hydraulic machinery using OpenFOAM

Transcription

1 Automatic shape optimisation in hydraulic machinery using OpenFOAM 5 th OpenFOAM Workshop, June 2010, Gothenburg Jakob Simader Andreas Ruopp Ralf Eisinger Albert Ruprecht Institut of Fluid Mechanics and Universität Stuttgart

2 Contents 1. Motivation 2. Workflow: 1. Parameter check 2. Grid generation and Grid conversion 3. CFD and simulated boundaries 4. convergence check and evaluation 3. Optimisation results 4. Conclusions and outlook 2 25

3 Motivation Automatic shape optimisation reducing calculation time Robust calculation shemes Low costs Low manpower Multi design criterias (part load, BEP, overload) Multi parameter setup 3 25

4 Perl interface Universität Stuttgart Sequential workflow Setup configuration Optimiser design parameters Parameter check Grid generation Grid conversion CFD simulation optimised design Objective function Convergence check Sequential workflow 4 25

5 Parallel workflow Multilevel parallel run on cluster nodes Advantages: Massive parallel cluster nodes available Fast design studies possible Calculation of many designs at the same time No license costs using OpenFOAM Available node number on cluster is the only limitation Parameter check Grid generation Grid conversion CFD simulation Convergence check Level 1 Level

6 Optimisation: Genetic algorithm (recombination) Simplex algorithm Stochastic algorithm Optimisation and CFD CFD and grid: OpenFOAM Grid size between and nodes SST model Used quality function: pressure recovery factor c p 2 pout p v 2 in in 6 25

7 2 checks: Limited depth No intersection of cross sections: sign Determinate of the four bounding vectors of each tube segment must have same sign d b det (a b) a c sign sign det (b c) det (c d) Parameter check Second check First check Parameter check 7 25

8 Grid generation Grid generation Grid conversion Definition of cross sections Width Radius Height Radius Width Height 8 25

9 Grid generation Grid generation Grid conversion Definition of position of cross sections Position of middle point of cut (x,y) Angle of normal vector 6 parameters for one cut In total 48 parameters for 8 cross sections Grid generator build for in house CFD-code converter needed for OpenFOAM data files 9 25

10 Inlet velocities CFD simulation Measured velocity profiles for francis turbine: part load BEP full load One run for uniform velocity distribution (c = 6 m/s) without any circumferential component Draft tube flow high sensitive to inlet flow field 10 25

11 Convergence check Convergence check Criteria: Number of max. Iterations c p(i = max) 1 c p(i) c p(i+1) ĉ p c p(i = max) check for pressure quantities therefore Runtime output of: - Abs. mass flow ave. of p - Abs. mass flow ave. of ptot Ensuring a good convergence behaviour 11 25

12 48 Parameter setup Optimisation setup Parallel setup with up to 30 individuals CFD-Solver OpenFOAM Calculation time approximately 24 hours Single design criteria: uniform velocity Part load, BEP and over load (8 segments) Detailed elbow discretisation (12 segments, but still 48 Parameter) Multi design criteria: Averaged c p of part load, optimum and over load (1/3 each) Weighted c p of part load, optimum and overload (part load: 30%, optimum: 50%, over load: 20% 12 25

13 48 Parameter setup Optimisation setup Parallel setup with up to 30 individuals CFD-Solver OpenFOAM Calculation time approximately 24 hours Single design criteria: uniform velocity Part load, BEP and over load (8 segments) Detailed elbow discretisation (12 segments, but still 48 Parameter) Multi design criteria: Averaged c p of part load, optimum and over load (1/3 each) Weighted c p of part load, optimum and overload (part load: 30%, optimum: 50%, over load: 20% 13 25

14 Single criteria results: uniform velocity c p = Optimised geometry: Area distribution fits common conventions smooth tube geometry, except bottom shape detailed view of bottom shape 14 25

15 Single criteria results: uniform velocity c p = Hand smoothed geometry: Smaller pressure recovery The contraction after the elbow has positive effect on the secondary flow phenomena 15 25

16 Single criteria results: uniform velocity Cutting plane Secondary flow fills up separated region 16 25

17 Single criteria results: optimum c p = Optimised geometry: Similar shape to the one with uniform velocity Also: smooth tube geometry, except bottom shape 17 25

18 Single criteria results: optimum Evolution of c p along optimisation run 18 25

19 Single criteria results: optimum detailed elbow c p = Optimised geometry: smooth tube geometry, except bottom shape doing the right thing might be a bit wrong, better than the wrong thing right 19 25

20 Single criteria results: optimum detailed elbow 20 25

21 Single criteria results: comparison c p (over load) = c p (optimum) = c p (part load) = Optimised geometries: Very smooth shape for part load conditions Different shape for different inlet flow optimising an averaged pressure recovery 21 25

22 48 Parameter setup Optimisation setup Parallel setup with up to 30 individuals CFD-Solver OpenFOAM Calculation time approximately 24 hours Single design criteria: uniform velocity Part load, BEP and over load (8 segments) Detailed elbow discretisation (12 segments, but still 48 Parameter) Multi design criteria: Averaged c p of part load, optimum and over load (1/3 each) Weighted c p of part load, optimum and overload (part load: 30%, optimum: 50%, over load: 20% 22 25

23 averaged weighted c p (averaged) = c p (over load) = c p (optimum) = c p (part load) = c p (weighted) = c p (over load) = c p (optimum) = c p (part load) =

24 Optimisation effort summary Single Weighted T total (<24 h) (<24h) No.indiv No.of died indiv No.of calc. Indiv No.of best Indiv No.of nodes per Indiv 1 1 No.of cpu s per node 4 4 All runs on 24 nodes on xeon cluster / 2xQuadcore 2.8Ghz Time consumption not tuned yet 24 25

25 Conclusions Introduced optimisation scheme is applicable for high numbers of parameters Fast calculation time due to parallel setup Multi design criteria optimisation Multi-generation optimiser Outlook Faster parallel perl algorithm to reduce calculation time 25 25