Optimization algorithms for aeronautical engine components: CFD design applications 1
Outline CFD Optimization Research projects Combustor applications Injection system design à swirl number Cowl design à air management Turbine application Tip cooling design à HTC Conclusion
CFD - Optimization EnginSoft has been active in CFD since the early 90s and in optimization since the mid 90s Skills have been developed: Linking through different scripting CFD codes to optimizers Parametrization of geometry and parametrization of meshing CPU management: networks, clusters (linux), large data handling Applications and models have been developed for three market sectors: Automotive Aerospace/defense Home appliance
AeroSpace - Research Funded Projects EC FP5-FP6-FP7 TATEF CLEAN ANTLE MAG-PI à Subcontracting VERDI NEWAC FIRST ERICKA à Partner Clean-Sky SGO Vocal-FAN à Partner SGA Wing-Tech à Partner ESA-ESTEC Alisse Melissa FC1 à Partner
Aeronautical Engine Components Turbines CFD optimization Target Design external aerodynamics à efficiency Design internal impingement cooling à HTC mass flow Design tip cooling à HTC mass flow Design impingement cavity à HTC - mass flow Combustor CFD optimization Target Design Vaporizer à uniform mixing Design Dilution holes à uniform RTDF - NOx Design Injector à swirl number Liner effusion holes à porosity characterization Design cowl à air management 5
6 Injection system
Engine components - Combustor Liners Swirler - Injection Cowls Diffuser Dilution holes & bleed
Objective 8 1999 Component validator for Environmentally friendly Aero engines (CLEAN) Injector Develop CFD parametric model of injection system Link CFD code with optimization software Develop optimized design via CFD-Optimization Problems Limited CPU capability Limited CFD physics: atomization, Nox modeling... Solution Simplified geometry with cylindrical coordinates DOE & MOGA optimisation Achievement Methodology developed allowed design of swirl vanes to obtain: swirl number, uniform premix, no flashback
Injector system Simplified with cylindrical coordinates from primary inlet to combustion can 1999 - CLEAN Injector system CFD Paramertric model Parameters - Constraint 2 inlet swirl angles Axial displacement: mass flow ratio Air_inlet 1 / Air_inlet 2 Throat position Objectives - Constraints Uniform Vaporization Swirl number Nox Flashback Swirl Air inlet 1 Swirl Air inlet 2 Inlet axial Δ Liquid fuel Throat axial Δ Combustor 9
modefrontier: 1999 - CLEAN Injector system Logic setup linking CFD to optimization integration of the CFD model into an optimisation software execution of several simulations in batch mode driven by an optimisation algorithm. INPUT VARIABLES DOE INPUT FILES MOGA ANSYS-ICEM: geometry and mesh MESH OBJECTIVES and CONSTRAINTS CFX: CFD solution and postprocessing OUPUT FILES 10 OUTPUT VARIABLES: fuel vaporisation,
DOE: 40 design SOBOL to explore the design space 1999 - CLEAN Injector system DOE: Did we choose the right parameters? to acquire information about the behaviour of the system to create the starting point (1 st generation) of the optimisation algorithm. Statistical analysis of the DOE results: Tstudent test, correlation matrix The swirl angles are the most relevant parameters with respect to the specified objectives definition interval for Alfa1 from [0 70 deg] narrowed to [0 45 deg] definition interval for Alfa2 from [0 70 deg] narrowed to [30 60 deg]. T h e T- s t u d e n t t e s t highlights the significance of the input variables with respect to the objectives 11
1999 - CLEAN Injector system MOGA: Do we converge? Multi Objective Genetic Algorithm: a set of designs evolve from a parent generation to the next using three operators: selection, cross-over, mutation Graph displays MOGA convergence on a constraint variable: Flashback history chart In red the unfeasible designs: breaking 2 constraints burning before the throat Swirl number dev. DOE MOGA In green the feasible designs 12
Reduce flashback Reduce swirl number deviation Maximise vaporization 1999 - CLEAN Injector system Filter: Understanding the results Optimizer view Input Variables Objectives Constraints 13
DESIGN_124 is chosen as the best design Alfa 1 and Alfa 2 are the most relevant parameters. The flashback is avoided with a correct positioning of the throat, and with the tuning of the two swirl angles and their mass flow. A method to obtain a desired swirl number tuning 2 mass flows and 2 swirl angles has been developed BASELINE DESIGN: Temperature 1999 - CLEAN Injector system CFD Results: Designer view DESIGN_124: Objective Temperature Improvement Min NOx -7% Min dev. mixing -8% Min dev. Swirl # 5% 14
15 Air management
MESH 2008 NEWAC NEW Aeroengine Core concepts Injector system - PERM RANS URANS P R O T O T Y P E 16
2008 - NEWAC PERM Combustor ACARE identified the research needs for the aeronautics industry for 2020, amongst others, the following targets regarding the engine are set, which will be looked for by NEWAC: 20% reduction in CO 2 emissions per passenger-kilometre whilst keeping specific weight of the engine constant significant reduction of the NO X emissions during the landing and take-off cycle (-80%) and in cruise (-60%) respect to CAEP/2 limit The role of Enginsoft in NEWAC, within Sub-Project SP6, is to bring a contribution to the design of Ultra Low NOx AVIO Single Annular Combustor, by means of: The optimisation of an innovative injection system technology called PERM (Partially Evaporating Rapid Mixing), applied for medium overall pressure ratios (20 < OPR < 35) The design of a Ultra Low NO X combustor chamber, focusing on the optimisation of the architecture.
Objective 2008 - NEWAC PERM Combustor Develop CFD parametric model of liners, injection system, cowls Develop optimized design of air distribution/management Problems Very complex geometry with huge mesh with more then 10M mixed elements, hence CPU and parameterization capability is still an issue for the optimization of the whole system Solution Different local models DOE & ANN optimisation Achievement Design method to obtain Cowl geometry and distribute air between PERM injector and inner-outer annulus with the assigned air split and constrain on pressure drop. 18
2008 NEWAC Air management Cowls
Cowl shape evolution and flowfield Evolution 2008 NEWAC Air management
21 Cooling system
Engine components Turbine Aerodynamics Cavities - Sealings Cooling
modefrontier Process Integration Design Of Experiment Decision Making Graphical Analysis MOGA Gradient Optimisation Algorithms Robust Design Expert System RSM, ANN Statistical Analysis
Expert Systems Expert systems (RSM, ANN...) used to support the MOGA for expensive CAE calculations (CFD, explicit FEM, ) and to speed up the search for the optimal solutions. INPUT Mathematical transfer functions are estimated using the available input-output database. RSM gaussian ANN sigmoidal These functions act then as an Expert System to substitute the CFD model EXPERT CFD SYSTEM MODEL OUTPUT OPTIMISATION ALGORITHM Expert systems: Advantage: no CPU time compared to CFD Drawback: accuracy of predictions depends on size of database 24
Objective Design the tip cooling holes via CFD & optimization Improve heat transfer and reduce mass flow Problems 2005 Tip cooling Reynolds number à Yplus à Micron à Mesh à Aspect ratio Average CPU capability Solution Refined mesh in the area of interest Mixed DOE MOGA ANN optimization strategy Achievement Design method to obtain tip cooling holes layout with improved performance compared to standard best practice methods 25
We have 6 geometrical input parameters # of holes, diameter, normalized holes camber line position linear distribution law nozzle angle squealer height Multi-Objective Minimize cooling massflow Maximise heat-flux Constrain efficiency 2005 Tip cooling Parameters choice Concur on the nature of parameters and objectives is non trivial Process integration Logic is assigned once
Tip Cooling: Optimization strategy Initial guess via SOBOL DOE First expert system ANN Further Evolution of 100 designs via a MOGA MultiObjective Genetic Algorithm on the ANN: with non CFD CPU cost Error estimation of virtual CFD DESIGN OF EXPERIMENT: 24 CFD analyses" ANN 1" Optimisation 1: MOGA" ANN 2" Optimisation 2: MOGA" ANN 3" MCDM:" Single Objective" Selection a n d verification: 6 C F D analyses" Selection a n d verification: 9 C F D analyses" Optimisation 3: MOGA" Updating with CFD training cycles the ANN (improve the expert system with real simulations) Final MCDM Single Objective optimization (via normalization) Selection and verification: 9 CFD analyses" OPTIMAL SOLUTION"
Tip Cooling: Results The evolutionary MOGA algorithm and the ANN expert system have allowed an improvement of the pareto frontier, thus devolping a series of designs which are better then the best practice design which was regarded optimal. Tutti Pareto DOE Pareto Ottimizzazione 2 Pareto Ottimizzazione 3 8.E-03 Tutti Objective 7.E-03 9.E-03 Min mass flow 6.E-03 Pareto Completo Finale 5.E-03 Max efficiency 4.E-03 DES_00 DES_AVIO DES_42 Improvements 8.E-03-21% 7.E-03 Max heat transfer minportata (kg/s) minportata (kg/s) Pareto Ottimizzazione 1 6.E-03 1.50% 5.E-03 Unchanged 4.E-03 3.E-03 3.E-03 2.E-03 1.E-03-650 2.E-03-600 -550-500 maxscambio (W) -450-400 -350 1.E-03-650 -600-550 -500 maxscambio (W) -450-400 -350
Conclusion: history - critics CFD Optimization CPU Human Mesh limitations Hexa Few algorithms Single WS Personnel Physics limitations Only DOE Only GA High Price to train P A S T All elements Single SW environment Linux clusters More Skill Improving physics DOE MOGA DFSS Price drop personnel Robust Design Expert System P R E S E N T Robust vs. accurate Parameters choice? Network Expertise & parametric Dynamic parametric connections Meshing Many models space restart CFD Optimization Design IT C R IT I C S