Community Research and Development Information Service - CORDIS

FP7

RBF4AERO Report Summary

Project ID: 605396
Funded under: FP7-TRANSPORT
Country: Italy

Final Report Summary - RBF4AERO (Innovative benchmark technology for aircraft engineering design and efficient design phase optimisation)

Executive Summary:
The RBF4AERO was funded by the EU Seventh Framework Programme (FP7/2007-2013) under the Grant agreement n° 605396 and it ran from September 2013 until August 2016. The RBF4AERO project aims at creating software, referred to as Benchmark Technology, conceived to handle the high-demanding requirements of aircrafts design and optimizations. The key components of such a numerical platform are tools existing before the inception of the project, which have been improved and suitably integrated in a unique working environment.
The impetus for the project implementation was the constantly increasing need of aircraft manufacturers for new solutions to improve performance and reliability of aircraft constituting components. As a consequence, they are constantly urged to invest in innovative design technologies so as to reduce aircraft development costs and delivery time. The Benchmark Technology reached this latter target by parameterizing the numerical models through a mesh morphing technique based on Radial Basis Functions (RBF). To accomplish relevant computational studies for aircraft design such as multi-objective optimization in a multi-physics context, fluid-structure interaction, icing accretion simulation as well as analyses exploiting the adjoint based data, the RBF4AERO consortium put in place original and effective computational procedures.
The consortium partnership is composed by trans-national (2 Small-Medium-Enterprises and for 4 Large-Enterprises) companies, 1 research institute and (2) academic partners with an international reputation in the areas of software design and development, integration of complex system, numerical modelling, computer-driven engineering design and experimental testing. They are D’Appolonia S.p.A. (Project coordinator), Hellenic Aerospace Industry SA, University of Rome “Tor Vergata”, Pipistrel d.o.o. Ajdovscina, National Technical University of Athens, Piaggio Aero Industries S.p.A., Tusas Engine Industries, Institut von Karman de Dynamique des Fluides and Italian National Research Council.
An accurate verification and testing activity was performed to evaluate the platform capabilities and its actual applicability to the design optimization phase. In particular, the RBF4AERO optimization approach was verified against well-documented computational models as well as real case applications of industrial use. On the other hand, the optimized outputs gained for a set of applications planned to be manufactured were also tested by comparing them with the corresponding results experimentally obtained for both the baseline and calculated optimal (morphed) configuration.
The optimization process based on evolutionary algorithm supported by off-line trained metamodels was built with the capability to also include, during computing, the elasticity of the deformable components. The icing simulations demonstrated the desired accuracy and effectiveness, whilst the adjoint-morphing coupling procedures showed all their relevant capabilities. When necessary, a physical validation of the numerical outputs was carried out. Both commercial and open-source tools were used and both computational fluid dynamic and structural solvers were accounted.
The numerical procedures of the Benchmark Technology allowed to improve the sought performance of the analysed models as well as to reduce the time needed to accomplish the computational studies. Such enhancements have a great significance for the aviation sector because they effectively lead to money and time saving for aircraft manufacturers. Therefore, the optimal computer-aided design models gained at the end of the optimization studies proved to be effectively and reliably used for manufacturing purposes.

Project Context and Objectives:
Due to the strong competitiveness and increasing of the requests for new solutions to improve performance and reliability of constituting components, aircraft manufacturers are constantly urged to invest in innovative design technologies so as to reduce aircraft development costs and delivery time. Generally speaking, the whole duration of both the Research and Technology Acquisition phase and the Product Development phase can be estimated to last about two decades. The 80% of the development costs and the 70% of Life Cycle Costs of an aircraft are determined in the conceptual design phase and as the program advances to the engineering and manufacturing development, the possibility to make changes is reduced exactly when the knowledge about the system increases. Due to this, the need to reduce the Research and Technology Acquisition Phase time is really high and, as such, more powerful design tools are needed. CAE-based techniques, such as CFD and CSM, have become indispensable means in conceptual aircraft design. Moreover, to deal with requirements of aeronautical design, the optimization of geometrical shapes is requested to fulfil the contrasting design targets of speed (time required to the overall optimization process), accuracy (achieved using large CFD meshes) and extent (related to the number of different configurations fully calculated during the optimization process).
RBF4AERO project aims at tackling all these aspects providing the user with the RBF4AERO Benchmark Technology, an integrated set of numerical tools built up to handle top-level shape optimization design studies. The core idea is to make the shape optimization possible by an innovative optimization environment based on a high performance meshless morphing technique.
This technique is founded on RBF theoretical approach which offers a number of distinct advantages over traditional optimization approaches, in terms of increase of computational speed and extent of numerical domain maintaining high level of accuracy and without make any trade-off between accuracy, time of computing and number of analyzed configurations.
The final objectives that RBF4AERO Project pursues are:
• capability to reduce the time to prepare an optimization scenario, that is the whole set of baseline configurations needed for an optimization study, to one day against the average of one month according to current practise;
• capability to extend the optimization domain to a wider extent in the respect of the above requirements by means of shape parameters interaction by DoE tables including enough configurations to produce a meaningful response surface;
• capability to run fast and detailed CFD optimization analyses involving large meshes without losing accuracy with respect to the baseline configuration;
• capability to perform numerical optimization analyses in a multi-physics context and with multi-objective target.
The features of the final product of the project research are:
• a very accurate control of the geometrical parameters with an extremely fast mesh deformation capability;
• high level of integration within the CFD computing process by the direct parametrisation of the CFD computational model;
• generality of use and high level of versatility due to intrinsic mesh independency of the RBF technique;
• high level of computing efficiency since the deformed field is defied regardless of the parallel partitioning of the CFD model;
• consistency in mesh treatment because of maintaining mesh topology and characteristics;
• high precision, since the shape modifications are performed through a nodal movement, which is intrinsically exact.
All these characteristics will be integrated in the final goal of the project, which is the development of the RBF4AERO Benchmark Technology: a dedicated numerical platform capable to allow aeronautical design engineers to build up the novel optimisation environment using their own numerical models and computing platforms.

Project Results:
RBF4AERO Platform
The main result of the RBF4AERO project is the implementation of the RBF4AERO Benchmark Technology: an advanced Computer Aided Engineering (CAE) platform fully integrated with High Performance Computing (HPC) hardware and tailored for generic CAE simulations (Computational Fluid-Dynamics (CFD), Finite Elements Analysis (FEA), Fluid-Structure Interaction (FSI), Ice accretion simulations, Adjoint-based optimization). The core of the platform is a Radial Basis Functions (RBF) Morpher Tool (MT) based on the commercial software RBF-Morph, which allows to achieve the results of the optimization studies in shorter times with respect to the current best practices, with no need to face typical limiting trade-off constraints and without changing the tools used for the numerical analyses. The chosen implementation allowed building a modular software architecture for an easy maintenance and growth. All the CAE functionalities can be accessed through the Graphical User Interface (GUI) of the platform: to perform optimization studies it is not required to use coding skills and extra features can be added using platform configuration files.

The RBF4AERO platform allows the user to perform in an automatic way Evolutionary Algorithm (EA) based optimizations: the implemented workflows allow the users to perform Single-Objective Optimizations (SOO) and Multi-Objective Optimizations (MOO); both SOO and MOO can be constrained or not and, moreover, the optimizations studies can involve FSI. The platform can also perform local optimizations through a gradient-based optimization procedure and, finally, can automate ice accretion studies and can exploit adjoint-morphing coupling techniques to perform shape optimizations. All these capabilities are grouped into the following operative scenarios:
1. EA-based optimization, which includes constraint and SOO and MOO which can be coupled with the FSI option;
2. Icing studies;
3. Adjoint-morphing coupling.
The results obtained with the EA-based operative scenario can be reviewed and post-processed using an advanced post-processing module. Furthermore, the platform can schedule and monitor simulation jobs and has the support to multi-user and multi-hardware management.

Morpher Tool
The Morpher Tool, which, thanks to its capabilities allows the parametrization of the computational grid of a CAE model, is the core of the RBF4AERO platform, and is the evolution of the commercial software RBF-Morph. It shares with its predecessor the capability to import stereo lithography (STL) or CFD General Notation System (CGNS) meshes; it allows the user to set-up mesh morphing using RBF and to save the setup and solution; it allows the user to interactively preview the morphing on surfaces and in volumes (using cutting planes); it allows to apply the shape modifications to NURBS geometric representation allowing thus the back2CAD procedure; it allows the batch usage of RBF for the following mesh formats: OpenFOAM, ANSYS Fluent, CGNS (structured and unstructured), ANSYS APDL, STL. To be successfully implemented and used in the RBF4AERO platform workflows, the MT has been enhanced including: new batch features needed by the advanced workflows and needed to support all the RB4AERO solvers; interfaces for data mapping and filtering needed by the FSI and adjoint-morphing coupling procedures; batch usage with silent GUI for support the new RBF fit procedure. In particular, the back2CAD feature has been object of research and development, being judged an important feature to be added to the platform. The back2CAD option will allow the End Users to revert back to the CAD model, which is required to convert the optimised model to a physical product, the shape modifications identified by the optimisation procedures. Rather than a CAD regeneration, it can be seen as a synchronization of the CAS surfaces with the optimized shapes obtained on the mesh; the method takes advantage of the meshless character of the RBF approach, so, similarly to mesh morphing, the CAD module need to be fed with the original baseline and the RBF solution, with the latter one that can be the result of a combination of whatever number of RBF solutions.
The MT developed during the RBF4AERO project successfully managed FSI problems, successfully implemented ice accretion models and has been successfully coupled with adjoint sensitivities to perform adjoint self-sculpting and adjoint preview.

Optimization Environment
The RBF4AERO Platform optimization environment is based on two different optimization strategies:
1. stochastic population based optimization exploiting Evolutionary Algorithms with off-line trained metamodels (EA-based optimization) available for all implemented CFD solvers;
2. gradient-based optimization exploiting continuous adjoint methods available only with the OpenFOAM solver.
The EA-based optimization process operates according to the following workflow, which is managed by the Workflow Manager (WM) module:
1. generate the Design of Experiment (DoE) table;
2. evaluate the DoE entries and fill the platform Data Base (DB);
3. train the metamodel;
4. while the stopping criteria are not satisfied:
a. generate new population
b. use the metamodel to approximate the new population
5. Evaluate the current optimal solution with the CFD tool and update the DB.

The OpenFOAM based adjoint solver implemented in the platform has the following features and capabilities:
• continuous adjoint solution to the incompressible Navier-Stokes (NS) equations;
• sensitivities with respect to some aerodynamic objective functions (lift and drag);
• frozen turbulence assumption (no differentiation of turbulence models).
The optimization environment has been integrated into a GUI designed and developed from scratch to be cross platform thanks to the use of Qt libraries. The user through the GUI can:
• set up a benchmark for the different operative scenarios and applications;
• run the submitted benchmarks on the user hardware platform;
• monitor the benchmark and hardware status;
• post-process the computed benchmark data;
• download the generated data, the analysed numerical models and the benchmark logs after the jobs completion.
The operative modes that can be accessed through the GUI are:
• EA-based optimization (including both SOO and MOO, constrained and unconstrained, with the possibility to include the FSI effects);
• Icing simulations both on 2D and 3D model, exploiting the constrained icing approach and the evolutionary icing approach;
• FSI;
• adjoint-morphing coupling studies.
The numerical solvers, both FEA and CFD, which models and results can be managed by the platform are:
• OpenFOAM (CFD)
• SU2 (CFD)
• NUMECA FINE/Turbo (CFD - turbomachinery)
• ANSYS Fluent (CFD)
• ANSYS APDL (FEA)
• rbf4aeroFSI (FSI solver – based on previous FEA and CFD ones).
With this set of implemented numerical solvers, which, as stated before, can be easily widened thanks to the modular structure adopted in the development, the RBF4AERO platform is very attracting because it can cover a large amount of applications: from internal aerodynamics to external aerodynamics, from subsonic flows to transonic and supersonic flows, from turbomachinery to cooling optimization, from whole aircraft optimization to single engine component optimization. All these applications have been tested in the Project execution through specific test cases which involved both commercial and open source codes. The latter ones, indeed, can also increase the interest towards the RBF4AERO platform: OpenFOAM and SU2 codes can cover a large amount of application and a wide range of flow conditions with no costs in licensing.
Special attention has to be given to rbf4aeroFSI solver: it is basically a set of procedures developed by the Project members to enable the execution of FSI studies exploiting two approaches: the modal superposition and the two-way. These procedures have been implemented according to two main purposes:
1. use the models and solvers already implemented in the platform;
2. be seen by the platform as a new solver, so that it can be used in the EA-based operative scenario and enabling thus the possibility to account for structures stiffness in the normal optimization procedures.
In the final implementation the rbf4aeroFSI solver in the two-way approach can analyse OpenFOAM, ANSYS Fluent and SU2 CFD models coupled with ANSYS APDL FEA models. In the mode superposition approach, the structure modes can be assigned (i.e. precomputed or obtained from experimental tests) or not assigned and so computed with ANSYS APDL, whilst the CFD model can be analysed with OpenFOAM, ANSYS Fluent and SU2.

Technical and Scientific Results
During the RBF4AERO project development many results were achieved both from the technical and the scientific point of view.
EA-based optimizations. The main operative mode of the RBF4AERO platform is the EA-based optimization benchmark. This operative mode allows the user to perform a stochastic optimization exploiting Evolutionary Algorithms with off-line trained metamodels. The procedures implemented allow the user to perform numerical simulations with both commercial and open-source codes; in particular the already implemented ones are: OpenFOAM, SU2, CFD++, ANSYS Fluent, Numeca FINE/Turbo. This initial, but easily expandable, set of CFD solvers, allowed the Consortium members to perform during the Project’s development, a wide range of applications, with both medium-size and large-size CFD meshes. All the End Users that performed stochastic optimization with the EA-based procedure of RBF4AERO platform, experienced benefits both in terms of time savings to complete the overall optimization process and in terms of performance of the final optimized models. From the point of view of the time savings the benefits came principally from two aspects of the RBF4AERO procedures:
1. thanks to the mesh parametrization via the RBF solutions, there is no need to prepare a parametric CAD model to regenerate the mesh or to regenerate the mesh on the modified geometry; this allowed the End Users to obtain a time savings higher than 80% in most of the RBF4AERO Project applications. Obviously, as every novel procedure adopted ex-novo, some time is needed in order to become acquainted with the new workflow, so the End Users advantages can increase with time if the RBF4AERO platform is permanently adopted;
2. thanks to the high level of automation, the End Users do not need to follow every step of the optimization workflow and human resources, during the benchmark execution, can be assigned to other tasks.
From the point of performance increase, the End Users reported that, even if working on shapes and geometries that were already the result of an optimization process, the RBF4AERO procedures allowed to reach a further performances increase.

Icing studies. The icing procedures were developed for both constrained approach (in which ice growth profiles are precomputed and applied to the CFD model at specific simulation times) and the evolutionary approach (in which the CFD solver is coupled with an ice growth model that drive the mesh nodes displacements according to the CFD results). The constrained procedure was tested with both 2D and 3D wing profiles, being the latter ones the most challenging, difficult and time consuming if tackled with standard techniques, whilst the evolutionary approach was tested on a 2D profile. In both approaches the numerical procedures implemented are characterized by a high level of automation and are completely manageable by the RBF4AERO platform. The test performed to verify the approaches demonstrated the correct working of the platform as well as confirmed the local control and the high accuracy (a fundamental requirement in ice accretion studies) of the RBF approach adopted for mesh morphing. Furthermore, the procedures had proven to be very flexible and have the characteristics to be easily adapted to other multi-physics phenomena that involve a modification of the structures shapes.
The adoption of the RBF4AERO icing procedure had also positive effects for the End Users that adopted it. In fact, compared with the standard adopted approach, in the pre-processing phase of the most challenging case (3D icing constrained) the RBF4AERO procedure allowed a time saving of 90%, since there is no need to prepare a CFD model for each configuration, only the baseline one is needed and, of course, the RBF solutions to morph the mesh.

FSI studies. FSI approaches developed and implemented in the RBF4AERO platform are two: the modal superposition approach and the two-way approach. In the modal superposition a modal representation of the structural model is used to make flexible the CFD model: thanks to the mode decomposition, it is possible through the MT to modify the CFD mesh nodes according to the structural stiffness and the aerodynamic loading. The structural modes and frequency can be precomputed and passed to the procedure or the user can supply a FEA model to the procedure which will evaluate the normal modes and frequencies to be used during the FSI study.
In the two-way approach, the user supplies both the FEA and the CFD model, the automated procedure will analyse the CFD model and transfer the loads to the FEA one; once solved the FEA model, the displacements are imposed to the CFD model through the MT and the procedure iterates until the convergence of displacement is reached. The FSI procedures have been effectively implemented and successfully validated using data registered for the HIREASD configuration of the 1st Aeroelastic Prediction Workshop (AeWP) as reference data (test #132 in steady state condition) and adopting the high fidelity and extensively tested numerical models made available by the AeWP committee. Results obtained with the RBF4AERO FSI procedures compared very well with experimental and numerical data from the AeWP.
The adoption of the FSI procedures led the End Users of the project to achieve time savings with respect to their standard approach to the FSI studies: a first End User adopted a two-way approach on hexa-block structured meshes which required a lot of efforts in the pre-processing phase. Adopting the novel approach the End User registered a 90% time saving in the pre-processing phase, thanks to the fact that the mesh re-generation is no longer needed. A second End User adopted a two-way approach with hybrid mesh: in this case, the End User registered an advantage in the solution phase, thanks to the fact that the information exchange between CFD and FEA models is completely automated; the time saving registered is 66% per FSI cycle.
Once extensively validated, the RBF FSI approaches were implemented in the platform in the form of an additional solver: the rbf4aeroFSI solver, so that the FSI can be enabled in the standard EA-based optimization. The rbf4aeroFSI was successfully adopted during the project in solving industrial applications. In the case of the optimization of an aircraft propeller, for example, the FSI procedure coupled with the EA-based optimization workflow allowed to reach the optimized configuration with a time saving of 85% for each analysed Design Point (DP). In the optimized configuration, an increase of 3.5% of the objective function was reached with respect to a starting baseline geometry which, being an industrial test case, was already the result of optimization process by the End User.

Adjoint-morphing coupling studies. The adjoint solver available for the RBF4AERO platform user is based on OpenFOAM toolbox. In the platform are available three operative modes exploiting the adjoint-morphing coupling and a gradient-based optimization method based on the adjoint results. The adjoint morphing modes are:
1. Adjoint-preview: the adjoint results are used to estimate the sensitivities of a set of RBF solutions, so that the ones with an higher impact on the objective function optimization can be identified;
2. Adjoint self-sculpting: the adjoint results (sensitivity maps) are filtered and used to generate an RBF solution which will modify the shape in order to achieve the objective function optimization; the procedure can be iterative to maximize the effects of the optimization;
3. Adjoint sculpting: the adjoint results are used to manually define an RBF solution to be manually amplified in order to obtain an optimized configuration.
In the gradient-based optimization method derivatives are used in the optimization path, in which the RBF solutions amplifications are updated at each cycle using the steepest-descent approach.
All the adjoint-morphing coupling operative modes have been tested and applied successfully by the End Users. In particular the adjoint-morphing procedures were not part of the standard optimization methods of some of the End Users and their adoption was seen as a good improvement with respect to manual optimization procedures that were in use. Furthermore, the high level of automation introduced by such novel approaches, allowed the End Users to obtain great advantage also from the human resource management point of view.

In all the operative scenarios in which the RBF4AERO platform was exploited, the saved time to complete the workflow and the increased performance reached, allowed the End Users to gain a competitive advantage in the final product industrialization. Furthermore, the increased performances in most industrial cases have turn in a more energy efficient product.

Foregrounds
The RBF4AERO project developed a wide range of foreground expertise for all partners of the consortium. In particular, the strict collaboration put in place between the partners enriched each of them of the knowledge of the other ones. The phases of the project which registered a high level of collaboration between the consortium members were:
• the platform development;
• the benchmark set-up;
• the benchmark results analysis;
• the experiments set-up;
• the experimental results analysis and comparison with numerical ones;
• the reports writing.
In all the above listed phases the partners had to share their best practices and know how, in order to successfully accomplish the project goals. In some cases the foreground exploitation was clearer, as, for example, in the case of gradient-based optimization procedures, which were not part of the technical capabilities of some project End Users and that were successfully implemented and exploited in real industrial optimization cases. In other cases the End Users already had in their optimization best practice the methodologies implemented in the RBF4AERO platform, but the adoption of the platform allowed them to, first of all, gain a significant time saving, and, moreover, to obtain further optimization for the tested geometries.
Regarding the experimental tests performed during the Project development, also the partners in charge of them obtained improved expertise from the RBF4AERO foregrounds. One of the test facilities was upgraded for the Project investigation, allowing the research in U-bends under rotation. This possibility results to be interesting for the gas turbine community, since very few investigations are present in the open literature with the same level of detail, reliability and accuracy reached during the Project investigations. The upgraded configuration brings the opportunity to easily replace the test section and apply the experimental methodology to study the aerothermal performance of different geometries.

Potential Impact:
Apart from the significant industrial value of the new optimization platform, the figures dealing with performance represent another success, the benefit for the environment due to the reduction of fuel consumption experienced adopting the optimal models configuration. This fuel consumption reduction has been quantified by the End Users that adopted the RBF4AERO platform to the value of 3%. As already underlined, this improvement, which at a first glance can seem small, is in actually obtained on configurations that were the result of a previous optimization process. So, the RBF4AERO platform, in industrial applications, allowed the End Users to improving cost efficiency and also to contribute to the greening of the air transport activity, since reducing the fuel consumption, the amount of pollutants introduced by aircraft engines is reduced.

An extensive activity of dissemination has been carried out. As hereinafter detailed in section 4.2, the efforts in this context led to 18 publications (one peer-review paper concerning glider optimization and entitled “Glider Fuselage-Wing Junction Optimization using CFD and RBF Mesh Morphing” has been recently accepted for publication on Aircraft Engineering & Aerospace Technology - Journal Emerald Group Publishing DOI: 10.1108/AEAT-12-2014-0211.R1) and 28 dissemination actions such as oral presentations to scientific events, articles published in popular press, flyers, posters, exhibitions as well as MSc (1) and PhD theses (2).

The platform reached a good level of maturity and lots of processes were successfully tested and many bugs fixed. Consortium partners agreed on debating an exploitation agreement to plan joint initiatives aiming, on the one hand, to promote the use of the platform and, on the other hand, to offer cloud-based computational aided engineering service. The possible diffusion of the platform among the End Users in the next future could be facilitated by the quite straightforward usage of the graphical user interface, and due to the fact the open-source computational fluid dynamic codes are included. The numerical capabilities of these codes are such to cover all the flow regime of external and internal aerodynamic cases.

The foreground generated during the project activities will be exploited in two directions: in the academic/research field and in the industrial/business field. In the academic research perspective all the partners had benefits and will continue in benefit of the increased know-how related to methods, procedures and facility usage gained during the collaboration. These academic benefits, then, in many cases End Users adopted procedures and tools that were not included in their standard approach before the project, and experienced real advantages from them. These procedure, tools and know-how will be re-employed in future applications by all the consortium partners. For example, the acquired expertise in approaching FSI problems, and in general multi-physics ones, can be exploited in future projects, applications, consultancy agreements. Also the improvements introduced in the Morpher Tool (MT) allow the possibility to offer a more powerful tool for mesh morphing to be used in both academic and industrial field. From the experimental point of view, due to the specific tests to be performed for the Project’s goals, let the partners in charge of them to both improve their experimental equipment and their expertise in such specific fields. Both these improvements will be exploited by the partners in future academic and industrial application in which this kind of expertise is required.
During the Project activities execution, the RBF4AERO consortium get in contact with other project, such as the FP7 project RIBES and FORTISSIMO, which, apart the already put in place collaborations, the interactions with them can trigger further actions both for the academic and industrial field. In particular, the Consortium is in touch with the FORTISSIMO project board to discuss joint-business opportunities in joining their applications marketplace with the RBF4AERO platform.

List of Websites:
Project Website:
www.rbf4aero.eu

Contacts:
Technical issues: emiliano.costa@dappolonia.it
Financial issues: sara.parodi@dappolonia.it

Related information

Documents and Publications

Contact

Emiliano Costa, (Project Engineer)
Tel.: +39 0659450300
Fax: +0039 0659450399
E-mail

Subjects

Transport
Record Number: 192696 / Last updated on: 2016-12-14