Optimization of an Axial Pump using CFturbo, PumpLinx & optislang

Transcription

1 13th Annual Weimar Optimization and Stochastic Days 2016 Conference for CAE-based parametric optimization, stochastic analysis and Robust Design Optimization Optimization of an Axial Pump using CFturbo, PumpLinx & optislang Dr. Gero Kreuzfeld, Sebastian Jakisch CFturbo Software & Engineering GmbH

2 Content 1. Motivation 2. Software tools a) CFturbo b) PumpLinx 3. Parameters 4. Optimization workflow 5. Results 6. Conclusion, prospects page 2/ 19

3 1. Motivation Turbomachinery design is Complex Time-consuming Expensive optislang PumpLinx Meshing + CFD/FEA Simulation Conceptual Design Iterative loops required Product CFturbo CAD + Prototyping + Testing page 3/ 19

4 2. Software tools a) CFturbo Modern turbomachinery design software for Pumps Ventilators Compressors Turbines Covers all major types: Radial Mixed-flow Axial Available components: Impeller Volute Vaned/vaneless stators Analytical design theory & empirical correlations Fully parametric geometry model of machines page 4/ 19

5 2. Software tools a) CFturbo Several design steps from design point to 3D geometry All parameters can be accessed via batch mode per XML file optislang integration available page 5/ 19

6 2. Software tools b) PumpLinx CFD system with high solver speed, especially for fluid systems with rotating/ sliding components Geometry model with Δp t Q inflow region impeller outflow region Mesh density after analysis: nodes cells Solver speed accuracy Binary tree mesh page 6/ 19

7 Allowed oscillation Optimization of an Axial Pump using CFturbo, PumpLinx & optislang 2. Software tools b) PumpLinx Additional convergence checks Relative velocity vectors Total pressure Δp t Checked iterations Final value Iterations Pressure colored streamlines page 7/ 19

8 3. Parameters z TE,S # Parameter Reference Minimum Maximum z LE,S Main dimensions 1 d H1 = d H2 176 mm 140 mm 210 mm 2 d S1 = d S2 584 mm 467 mm 700 mm d S1 Shroud d S2 ν = d H1 / d S Δz 204 mm 160 mm 320 mm 4 z LE,H * Meridional contour 5 z LE,S * Hub 6 z TE,H * z TE,S * d H1 z LE,H Δz z TE,H d H2 Simplifications: Hub and Shroud (Tip) axis-parallel r z Meridional view Straight meridional leading and trailing edge page 8/ 19

9 3. Parameters β B2,H # Parameter Reference Minimum Maximum Blade properties 8 n Bl t LE,S m βb2,h 10 t TE,S t TE,H Mean lines 12 m βb1,h * m βb1,h m βb2,s 13 m βb1,s * β B1,H m βb1,s 14 m βb2,h * Leading edge Simplifications: 15 m βb2,s * Free vortex velocity distribution Automatic calculation of blade angles β B1 (shock-less inflow), β B2 (Euler equation) t LE,S m t Mean lines t TE,H t TE,S page 9/ 19

10 4. Optimization workflow Design point Flow rate Q = m³/s Total pressure difference Δp t = bar (H = m) Rotational speed = 780 /min Water, no pre-swirl Objective Max. hydraulic efficiency η Constraints β B2 < 90 Total pressure difference Δp t ± 10% page 10/ 19

11 4. Optimization workflow PumpLinx Meshing, simulation, convergence check, interpretation Check for meshing and simulations problems CFturbo Variation of geometry parameters Geometry update & export (*.spro, *.stl) Check log for warnings and errors β B2 < 90? page 11/ 19

12 5. Results Some geometry examples page 12/ 19

13 5. Results Sensitivity analysis/ Number of samples Advanced Latin Hypercube Sampling Samples Failed CFturbo PumpLinx Valid samples Reduced Q 83.6 % 86.6 % 89.1% CoP Δp t 85.8 % 90.6 % 88.4 % P 91.5 % 92.9 % 94.0 % η 57.4 % 66.5 % 73.6 % page 13/ 19

14 5. Results Sensitivity analysis/ Coefficient of Prognosis CoP η hydraulic efficiency (objective) η = Q Δp t P Q Δp t P Flow rate Total pressure difference Power consumption η: CoP=74% m βb2,s d S1 z TE,S t TE,H Q: CoP=89% Δp t : CoP=88% P: CoP=94% z TE,S t TE,S z LE,S t LE,S n Bl Δz d S1 z TE,S n Bl t TE,S z LE,S t LE,S Δz d S1 n Bl t LE,S t TE,S z TE,S z LE,S Δz d S1 page 14/ 19

15 5. Results Sensitivity analysis/ Metamodel of Optimal Prognosis MOP Response surface for Δp t = f(d S1, Δz) Δp t : CoP=88% Δz d S1 page 15/ 19

16 Optimization of an Axial Pump using CFturbo, PumpLinx & optislang 5. Results Sensitivity analysis/ Parallel coordinate plot parameters objectives constraints page 16/ 19

17 5. Results Optimization Algorithm Samples Simulation time Simulation time/ sample η EA Evolutionary Algorithm h (3.0 d) 13.1 min 69,9 % % ARSM Adaptive Response Surface Method h (5.3 d) 14.0 min 69,3 % + 4.4% Desktop PC 2 x Intel Xeon 3.07 GHz, 6 cores 64 GB RAM Max. 2 parallel simulation jobs page 17/ 19

18 5. Results Optimization Reference Reference ARSM ARSM EA EA page 18/ 19

19 6. Conclusion, prospects Optimization process CFturbo + PumpLinx + optislang successfully tested PumpLinx solver speed is beneficial for optimization and enables optimization on desktop PCs CFturbo initial design can be used as very reasonable starting point ( pre-optimized ) to save optimization time Number of parameters could be reduced to ~ 50% by sensitivity analysis automatic parameter reduction in CFturbo possible in near future Metamodel for efficiency seems to be not possible direct optimization with reduced parameter set seems to be the best solution Removing geometry simplifications is necessary Empirical based performance prediction in CFturbo could be used to further reduce simulation effort CFturbo included metamodel Thanks to Dynardo (optislang) and Simerics (PumpLinx) for licenses and support! page 19/ 19