Periodic Reporting for period 2 - MADELEINE (Multidisciplinary ADjoint-based Enablers for LArge-scale INdustrial dEsign in aeronautics)
Reporting period: 2019-12-01 to 2021-11-30
So far, during aircraft or engine design phases, most design optimisation studies performed using high-fidelity tools had been focused on a single discipline. This industrial practice was partly a legacy of the organisation structure of most companies (different departments focusing on different disciplines). The effect of single discipline optimisation was that the design process proceeded iteratively from one discipline to another, with significant time-delays to the overall process, making it hard to exploit multi-disciplinary trade-offs. A typical design optimisation with state-of-the-art gradient-based algorithms can involve dozens or hundreds of design iterations. If these designs require adaptive methods or unsteady CFD, like in acoustic problems, simulation run time can become prohibitive for industrial use. Simulation time increases further when moving from single discipline to large-scale MDO problems.
In this context, MADELEINE strengthened the capabilities and use of multi-physics adjoint solvers to maximise the benefit obtained from the computationally intensive simulations that are key enablers for future airframe and engine design. This includes the design of the complete systems but also of sub-systems which can be performed either by aircraft and engine manufacturers or by subcontractors of the supply chain for specific components (including SMEs). The multi-physics interactions considered in MADELEINE are:
✓ For airframe: focus on wing/fuselage aero-structure interactions which represents a key driver for aircraft design;
✓ For airframe/engine interactions: focus on aero-acoustics interactions;
✓ For engine: focus on fan and high-pressure turbine blades of Ultra-High Bypass Ratio engines which are very challenging components to design due to very stringent aero-structure and aero-thermal requirements.
In this context, MADELEINE strengthened the capabilities and use of multi-physics adjoint solvers to maximise the benefit obtained from the computationally intensive simulations that are key enablers for future airframe and engine design. This includes the design of the complete systems but also of sub-systems which can be performed either by aircraft and engine manufacturers or by subcontractors of the supply chain for specific components (including SMEs). The multi-physics interactions considered in MADELEINE are:
✓ For airframe: focus on wing/fuselage aero-structure interactions which represents a key driver for aircraft design;
✓ For airframe/engine interactions: focus on aero-acoustics interactions;
✓ For engine: focus on fan and high-pressure turbine blades of Ultra-High Bypass Ratio engines which are very challenging components to design due to very stringent aero-structure and aero-thermal requirements.
The MADELEINE consortium addressed the objectives of the project around three main pillars:
• Capability (robustness and accuracy of multi-physics adjoint solvers; efficient exploration of large design space; manufacturability oriented design)
• Efficiency (fast adjoint-based MDO capability for large industrial test cases on next generation HPC infrastructure; industry-compatible development time)
• Usability (physics-based parameterisations, appropriate end-user MDO formulations)
The achievement of the project objectives was measured, verified and monitored using the specific Success Criteria (SC) metrics:
• Capability:
SC1: Multi-physics adjoint sensitivities verified.
SC2: Impact of manufacturing criterion in the MDO process on performance and costs quantified.
SC3: Removal of constraints on design space exploration.
• Efficiency:
SC4: MDO adjoint solvers as robust and fast as direct disciplinary solvers even in the presence of flow separation on heterogeneous HPC systems.
SC5: Reduction of MDO development time for industrial deployment.
• Usability:
SC6: MDO parameterisations defined with and approved by industrial aircraft and engine designers.
SC7: Comparison between results obtained through coupled adjoint-based MDO and sequential/uncoupled single objective optimisations on various industrial test cases and definition of best practices and guidelines.
The main results achieved in MADELEINE contribute to all the Success Criteria established in the project:
Adjoint solvers:
* Robustness and accuracy of the adjoint solvers
* Comparison of linear solvers dedicated to adjoint solver
MDO formulations: development of scalable models representative of industrial optimisation problem used to benchmark MDO formulations
Uncertainty Quantification and Robust Design: integration of gradients in Non-Intrusive Polynomial Chaos methods and validation on simplified test-cases
Parameterisation:
* Adaptation of topology optimisation methods for High Pressure turbine application
* Benchmark of parameterisation methods for fan blade optimisation
* Integration of geometric constraints in the Vertex Morphing approach
Mesh deformation: benchmark of mesh deformation techniques on generic test-cases
Time-efficiency of MDO process: evaluation of different strategies to increase the time-efficiency of MDO process (by reducing the number of iterations or by improving the data transfer between the different solvers/tools used in the MDO process)
• Capability (robustness and accuracy of multi-physics adjoint solvers; efficient exploration of large design space; manufacturability oriented design)
• Efficiency (fast adjoint-based MDO capability for large industrial test cases on next generation HPC infrastructure; industry-compatible development time)
• Usability (physics-based parameterisations, appropriate end-user MDO formulations)
The achievement of the project objectives was measured, verified and monitored using the specific Success Criteria (SC) metrics:
• Capability:
SC1: Multi-physics adjoint sensitivities verified.
SC2: Impact of manufacturing criterion in the MDO process on performance and costs quantified.
SC3: Removal of constraints on design space exploration.
• Efficiency:
SC4: MDO adjoint solvers as robust and fast as direct disciplinary solvers even in the presence of flow separation on heterogeneous HPC systems.
SC5: Reduction of MDO development time for industrial deployment.
• Usability:
SC6: MDO parameterisations defined with and approved by industrial aircraft and engine designers.
SC7: Comparison between results obtained through coupled adjoint-based MDO and sequential/uncoupled single objective optimisations on various industrial test cases and definition of best practices and guidelines.
The main results achieved in MADELEINE contribute to all the Success Criteria established in the project:
Adjoint solvers:
* Robustness and accuracy of the adjoint solvers
* Comparison of linear solvers dedicated to adjoint solver
MDO formulations: development of scalable models representative of industrial optimisation problem used to benchmark MDO formulations
Uncertainty Quantification and Robust Design: integration of gradients in Non-Intrusive Polynomial Chaos methods and validation on simplified test-cases
Parameterisation:
* Adaptation of topology optimisation methods for High Pressure turbine application
* Benchmark of parameterisation methods for fan blade optimisation
* Integration of geometric constraints in the Vertex Morphing approach
Mesh deformation: benchmark of mesh deformation techniques on generic test-cases
Time-efficiency of MDO process: evaluation of different strategies to increase the time-efficiency of MDO process (by reducing the number of iterations or by improving the data transfer between the different solvers/tools used in the MDO process)
The integration of high-fidelity simulations in Multi-Disciplinary Optimisation (MDO) is a necessary evolution of the design process in order to meet short, medium and long-terms industrial objectives in terms of:
❖ Competitiveness: by reducing development time and cost but also the cost of manufacture;
❖ Environment: by designing more efficient engine and aircraft configurations with better multidisciplinary compromises and fostering the integration of greener technologies earlier in the design phase.
The adjoint method is a key enabling technology for efficient gradient-based optimisation with a large number of design variables. It can also contribute, outside of the optimisation loop, to identify crucial areas in a design which have the largest impact on performance. In this way, the adjoint approach enables the designer to focus on the most critical area for design optimisation, significantly shortening the design process. At the same time, the presence of excessively large sensitivities in candidate designs can be reduced, leading to more robust designs with smaller real-life performance degradations.
Within MADELEINE, rather than employing cheaper low-fidelity methodologies that have very limited potential to deliver enhanced designs, the consortium deployed efficiently the MDO optimisation process on next-generation HPC infrastructure in order to:
❖ design configuration with improved performance benefiting exploiting multidisciplinary trade-offs
❖ drastically reduce simulation times and make them compatible with industrial product development cycles.
In this context, MADELEINE achieved two particular goals:
❖ reduced the barriers to set-up an efficient MDO process by developing re-usable modules and standardising the interfaces between the disciplines;
❖ reduced the simulation run-time by exploiting heterogeneous HPC architectures through optimised libraries and efficient job orchestration.
❖ Competitiveness: by reducing development time and cost but also the cost of manufacture;
❖ Environment: by designing more efficient engine and aircraft configurations with better multidisciplinary compromises and fostering the integration of greener technologies earlier in the design phase.
The adjoint method is a key enabling technology for efficient gradient-based optimisation with a large number of design variables. It can also contribute, outside of the optimisation loop, to identify crucial areas in a design which have the largest impact on performance. In this way, the adjoint approach enables the designer to focus on the most critical area for design optimisation, significantly shortening the design process. At the same time, the presence of excessively large sensitivities in candidate designs can be reduced, leading to more robust designs with smaller real-life performance degradations.
Within MADELEINE, rather than employing cheaper low-fidelity methodologies that have very limited potential to deliver enhanced designs, the consortium deployed efficiently the MDO optimisation process on next-generation HPC infrastructure in order to:
❖ design configuration with improved performance benefiting exploiting multidisciplinary trade-offs
❖ drastically reduce simulation times and make them compatible with industrial product development cycles.
In this context, MADELEINE achieved two particular goals:
❖ reduced the barriers to set-up an efficient MDO process by developing re-usable modules and standardising the interfaces between the disciplines;
❖ reduced the simulation run-time by exploiting heterogeneous HPC architectures through optimised libraries and efficient job orchestration.