This presentation represents work performed at QinetiQ and its predecessor organisations over a period of years, to meet the requirements of the Ministry of Defence Applied Research customer. I would like to acknowledge the contributions made to this work by my colleagues within the Aerodynamics and Systems Performance and Control groups at QinetiQ, particularly my co-author, John Doherty.
This presentation has an introduction, three main topics and a conclusion. The first, and most important, of the main topics is the assessment process. This ranges from the initial definition of a concept to meet a set of requirements, through fleshing out the details of the design, analysis of the resulting configuration, followed by comparison of the derived performance with the estimates used to define the original concept.
The second main topic is a description of the technologies which underpin the assessment capability, how these have contributed to reducing the time and cost of the process while improving its accuracy and flexibility. My third main topic is a brief description of where future customer requirements are heading, and why it is increasingly important to understand the impact of aerodynamics on the performance, cost and flexibility of a system as a whole
So, without further ado:
The capability described here has been developed with funding by the Ministry of Defence Applied Research Programme to meet the needs of stakeholders within the MOD procurement system. MOD procures complex systems, and indeed, systems of systems. Some of these are developmental but there is increasing emphasis on procurement of Commercial-Off-The-Shelf (COTS) items. In all cases, equipment is procured against Operational Requirements. These are defined in a number of ways, but increasingly, the top level drivers are identified, leaving the technical detail to be addressed by the contractor, along with the MOD representatives, in Integrated Project Teams. What MOD requires is an assessment capability that informs them of the impact of technologies, and how these relate to top level drivers, at system level. This is essential to providing MOD with an intelligent customer capability. A key element in the following processes is the extent to which they are driven by the end user requirement. Ideally, the development of any product starts with the definition of the requirements.
The most important part of the presentation.
It is a legitimate question to ask why MOD is spending money on the development of new assessment processes for air vehicles. Historically, the process has been based on a combination of simple ‘project level’ assessment tools, similar to those in data sheets, followed by extensive (and expensive) wind tunnel testing. Initial phases of this process were potentially inaccurate, particularly for novel concepts. This has acted as a deterrent to the adoption of novel or radical solutions. The time and cost associated with generating extensive wind tunnel data sets for a range of configurations became increasingly prohibitive in the post-Cold War environment. Over the last twelve or so years, we have implemented a number of new technologies to improve the assessment capability. Most of these were radical in the early days of their implementation, but are now mature. The most important individual factor has been the emergence of CFD as an accurate means of calculating lift and drag. Combined with numerical optimization, and now with Computer-Aided Design capability, this has enabled a much more accurate and flexible assessment process to be implemented.
The main tool used within QinetiQ and MOD for assessing air vehicle performance at system level is the Multivariate Optimization design synthesis method, or MVO, as it is generally known. Although the origins of MVO, and earlier MVA, methods arose from civil transport studies in the late 1960s and early 1970s, the current system has evolved from a manned combat aircraft design synthesis dating from the early 1980s. At it’s simplest, the design synthesis consists of a parametric representation of the outline of a combat aircraft. The parameters are variables under the control of a numerical optimization process. In recent years, the MVO method has evolved to reflect the changing nature of MOD requirements. There have been multiple versions of the synthesis, but the current version has been developed largely to represent recent Unmanned Combat Air Vehicle (UCAV) concepts. However, to illustrate the process, we have included illustrations drawn from an earlier exercise involving a manned concept. In recent years, the range of concept classes in which MOD has expressed interest has expanded significantly. It is thus in the best interests of both QinetiQ and MOD to make the MVO tools as flexible and generic as possible. We will describe how we are achieving this later on in the presentation.
The starting point of any MVO synthesis is the capture of requirements and their representation in quantitative form. MVO works on the basis of both mission and point performance requirements. The mssion requirements consist of the weapons load, and a description of the mission to be flown, as a series of legs, which may include cruise, climb, low-level ingress, dash etc. The point performance requirements are expressed as targets for turn, acceleration or specific excess power at various combinations of mass, altitude and Mach number. The number of performance requirements can go into double figures. This is a key factor in the design of combat aircraft. The requirements are frequently conflicting. Determining which requirements are actually driving the design, and determining the optimal balance between them, is a major part of the rationale for using numerical optimization.
This diagram represents the process by which MVO generates a solution concept for a given requirement. Note that the performance requirements are an input to the process, rather than a fall-out. The synthesis generates a parametric representation of the air vehicle concept. The mass, aerodynamic characteristics and performance are calculated. These outputs are then compared with the performance requirements, which are set as constraints. If these are not met, or the concept is not feasible, the optimizer changes the parameters in the synthesis. The optimization loop continues until the performance requirements are met. The objective function, which the optimization algorithm seeks to minimize, is usually Basic Mass Empty, although life-cycle cost models are available, and these are undergoing further development. At the end of the process, the final versions of the parameters are output. For the concept illustrated, the output consisted of simple 2D sections, but we have recently changed the form of the output, in a way which we will describe later.
The geometry output from MVO is essentially a series of 2D sections and wireframe representations of the internal and external lines. To determine the actual performance of the vehicle requires representative lines to be generated. Initially, these are constructed using 3D CAD from the initial MVO representations. The performance of the vehicle is sensitive to the detail of the external lines, thus it is necessary to perform a detailed aerodynamic design of the configuration. For this we use the QinetiQ CODAS design method, to produce a multi-point optimised aerodynamic design of the configuration. A single CODAS run involves approximately 250 individual CFD analyses, whereas the 2000 cycles of an MVO run typically take 5-10 minutes.
The MVO description, as we can see, is fairly basic, but it does include detailing of the concept interior and some volume accounting.
The CAD representation puts flesh on the bare wireframe description. At this point, more details, such as the section profiles, the radome, canopy and wing-body blending emerge. This part of the process is reasonably manpower-intensive and time-consuming, and we will show how we have improved this later in the presentation. At this stage, the concept is ready for the start of detailed aerodynamic design. In the past, also, at this stage, the lines would be suitable for use on a low-speed wind tunnel model.
CODAS has been in use for over twelve years now, and a good deal of experience has been gained in its use for a wide range of requirements. This diagram describes the CODAS optimization design method. Essentially, the process starts, as with MVO, with a parametric representation of the geometry. The performance of the shape is analysed at a range of conditions, using CFD. We have used multiple design points, variable geometry, and even different analysis methods for different design points, depending on the spread of required performance points. For each iteration of the geometry, a whole range of constraints are evaluated, to establish if particular requirements have been met. As experience in the use of CODAS has developed, as with MVO, the choice of constraints has become a more important aspect of the process than the choice of objective function. The optimisation algorithm cycles through the process until all of the constraints have been met, or the conflicts between requirements have been defined. Determining the balance between design requirements is a fundamental part of the process, and it is usual for the process to be repeated, yielding useful information on the detailed design requirements
This is a the final version of the external lines generated by CODAS. The sharper-eyed among you (and I’ll forgive the rest of you at this late stage of the afternoon) will notice a number of subtle changes from the earlier image of the original shape. The wing is now cambered and twisted, to meet a range of performance requirements. Much of the fuel volume from the inner portion of the wing has migrated into the fuselage. The wing has moved slightly forward, while the tailplane has increased in area. To meet trim requirements, CODAS has added camber to the tailplanes and fins, while the boat-tail angle has been changed. It should be noted that the changes made to the external lines are subtle, yet they result in an enormous improvement to the aerodynamic performance of the configuration. This is then representative of a real vehicle, whereas the earlier shape would not be.
Having generated a set of external lines, the time taken to manufacture and test a wind tunnel model can be significant. In many cases, it is desirable to know whether it is worth proceeding to the stage of experimental testing. As a consequence, it is now usual to perform series of CFD analyses between definition of external lines and the availability of a wind tunnel model. These analyses are also used to make first estimates of which elements of a model to build for high speed test, including choice of flap settings and tailplane angles to trim. This part of the process is entirely predicated on the assumption that CFD can be used to give reliable predictions of overall aerodynamic loads, particularly drag. Thus far, this has effectively limited the applicability of this approach to attached flow conditions. I will make some further comments about the scope of CFD versus experiment at a later stage. It is likely that in future, a greater proportion of air vehicle concepts will reach this level of maturity than before. It is also likely that a smaller proportion will go on to experimental testing.
These next two slides give an indication of the value of the post-design CFD analysis. One of the performance requirements for the concept illustrated was for a time taken to accelerate from transonic to supersonic conditions. MVO generated a figure based on its best estimates of available thrust and drag, with acceleration being determined by specific excess power available through the manoeuvre.
The green line on this plot indicates the level of drag predicted by CFD analyses, in comparison.with the original red line predicted by MVO. This indicates the relative success of the CODAS design, but also indicates the limitations of the MVO model for prediction of transonic performance. Determining that the performance of the vehicle is likely to be significantly better at this critical point might also allow an improved trade-off against other performance requirements.
Despite the time and cost taken to manufacture and test wind tunnel models, where large numbers of data points are required, wind-tunnel testing remains the most cost-effective means of generating these. The repeatability of wind tunnel balance measurements is still an order of magnitude less than the best accuracy in loads achievable with current CFD. Wind tunnel data can be gathered at conditions beyond the capability of CFD methods. Indeed, wind tunnel models can also gather data at or beyond the limits of the flight envelope for most aircraft. Low speed wind tunnel testing is significantly cheaper than transonic or supersonic testing. The range of configuration variables which can be tested at low speed is significantly greater than can be afforded at high speed. The role of high speed testing is increasingly to validate the predictions made by CFD, and to provide data for conditions beyond the capability of CFD methods. Surface oil flow visualisation is an important tool for understanding how fluid mechanics impact on the performance, stability and control of air vehicles. It does appear to be an endangered skill at the moment, particularly for transonic flows. We use CFD results to help put together our test programmes. In this particular case, the choice of trailing edge flap settings was fundamentally driven by RANS analyses.
This model was tested on a sting rig in a transonic tunnel. This provides the bulk of the aerodynamic data set for this configuration, although another test entry took place to investigate supersonic performance. This model is representative of those used for the assessment process. The model is fitted with a strain gauge balance, The wings are pressure tapped,. The intakes are faired over, and separate processes are used to determine the effect of propulsion installation. In this particular case, motorised tailplanes were installed.
Wind tunnel testing is not the end of the process of generating a performance data set. Some of the CFD results will be trimmed, but these are less accurate than experimental measurements. They may also be at model-scale , full-scale or infinite Reynolds number. The tunnel data will, in general, be at sub-scale, and untrimmed. We have developed a technique for deriving trimmed data from an untrimmed wind tunnel data set, using response surfaces and, yet again, numerical optimization. The process also generates optimal device schedules as a by-product. There still remains the issue of correcting tunnel measurements from model to full-scale. We tend to estimate a correction to zero-lift drag by using flat-plate skin friction estimates applied to Euler solutions, or by comparing RANS results at model and full scale.
This plot shows a comparison of the original MVO performance predictions with trimmed data derived from wind tunnel tests using the response surface method. In this case, each curve corresponds to a particular setting for both leading and trailing edge flaps. The optimum trimmed drag polar will follow a curve which cuts each of these as angle of incidence, and hence lift coefficient, varies. It should be noted that the experimental data has not been corrected for Reynolds number, while the MVO points are for a range of altitude and mass conditions at the same Mach number.
Because MVO is the only tool that can illustrate the impact of technologies at system level, it is important that it is accurate. To ensure confidence in MVO results, it is desirable that MVO performance predictions match those from CFD and experiment. Where consistent or systematic differences are found, these can be rectified. It is also important that CFD methods continue to be validated against experimental data. An interesting example follows later.
What are we using to improve the process ?
We started to analyse and design blended configurations in the early 1990s. It became immediately apparent that CAD techniques were needed to describe the geometry, let alone modify it. After initial experience of working through a drawing office, it was decided to train aerodynamicists to use 3D CAD systems for surface generation and modification. Roughly simultaneously, it became apparent that the most time-consuming part of the CFD analysis process was not mesh generation, or flow solution, but geometry pre-processing. Over the last twelve or so years, these capabilities have evolved and expanded. Our two main CAD-based tools are now the GEMS pre-processor and the CATIA V.5 parametric knowledge-based 3D CAD system. The latter is now being more closely integrated with MVO. GEMS and CATIA V.5 are increasingly forming the basis for a truly multidisciplinary assessment capability.
GEMS has evolved from a CAD application which allowed us to pre-process geometries for block-structured Euler and RANS analyses. GEMS is a stand-alone CAD application, built on a set of Open Source libraries. It allows us to pick up CAD geometry from CATIA or third parties, and prepare it for analysis. This will frequently involve repair,resampling and reorganisation of the CAD surfaces. GEMS also has the ability to build surfaces from scratch, where necessary. GEMS supports structured and unstructured surface and field mesh generation, for analysis with vortex lattice, panel, Euler and RANS methods. It is easily extensible to other analysis tools. All functions in GEMS can be recorded as replayable scripts, allowing whole processes to be automated.
As we have said earlier, the conversion of MVO output into 3D CAD geometry has been a time-consuming and manpower-intensive process. We have used the parametric and knowledge-based features within CATIA V.5 to produce a parametric copy of the MVO model. The parameters for this model are supplied by MVO in the form of a Microsoft Excel spreadsheet. Revising the model takes a matter of minutes. The five configurations shown here were created from the same parametric model, using different sets of parameters. It is possible to build more detail into the CATIA model, to automatically provide fidelity beyond what is possible with MVO. Increasingly, this model can be used as a vehicle for more detailed multidisciplinary assessment. Initial studies include automation of internal structural layouts.
The majority of methods used in MVO for prediction of aerodynamic characteristics are reasonably similar in capability and vintage to data sheet methods. These are limited in applicability to classical configurations, and are largely based on ARC and NACA sources from the 1940s and 50s. We need to be able to produce methods which are more accurate and flexible, without adversely impacting on the runtime or complexity of MVO. We have identified a set of technologies to replace the existing MVO methods with response surfaces. This involves the generation of parametric data sets using CFD analysis, and their conversion into algebraic expressions. Commercial software is available to determine optimal sample points for accuracy, and automatically generate fits to the data. Initial experience with simple swept-tapered wings has been very encouraging.
We now have a wide-ranging CFD capability: Euler methods are a mature tool, with both mesh generation and flow solution being routine. They are also a fast and accurate means of predicting aerodynamic loads in attached flows. We are increasingly looking at panel methods as a means of mass production of parametric databases, for populating response surfaces and generating stability and control models. We make extensive use of RANS, but aspects of this capability are not yet fully mature and require improvement. We have some remaining issues with the robustness and accuracy of the numerical schemes. We see the primary rationale for RANS being the ability to predict the onset of flow separation and to extend analysis capability to unsteady, separated flows.
This comparison shows an interesting case. It should be noted that both the Euler and RANS analyses were performed several months before the wind tunnel test. These results are for a transonic case, with a deployed wing leading edge flap. The Euler analysis predicts a high suction peak on the leading edge flap knuckle, with a strong shock developing outboard. The upstream Mach number normal to the shock at the outboard pressure station would normally be considered more than adequate to produce a shock-induced flow separation. The RANS analysis indicates that the flow behind the flap knuckle suction peak separates, forming into a vortical flow over the outboard part of the wing. Based on both CFD analyses, we expected the flow over the wing to be separated at this condition. The experimental pressures, confirmed by surface oil flows, showed that the flow behind the knuckle suction peak remained attached, even with a strong shock. Flow separation eventually developed at higher alpha, with a myriad of small shock-induced bubble separations developing along the line of the flap knuckle. This indicates some of the issues arising from use of RANS to predict flow separations from designed wings, rather than wings for which the location of flow separation is known a priori. It also underlines the reason why experiment is, and will remain, necessary.
So, where do we go from here ?
MOD is showing increasing interest in top-level driving requirements, with less emphasis on the underlying details of how these translate into equipment roles and capabilities. There is more emphasis on outcomes and effects of using ‘systems of systems’, rather than on pursuing individual technologies for their own sake. Given that the impact of aerodynamics, and other technologies, can only really be assessed at system level, the importance of MVO remains fundamental, particularly when looking at balance of investments to achieve a given outcome. There is a need for tools like MVO to cover a full spread of air vehicle options, from surveillance and communications platforms to high speed weapons.
Many of the technologies that we have demonstrated have their origins outside aerodynamics. Similarly they have applications beyond aerodynamics. The combination of parametric CAD, automated analysis processes and response surface generation is completely generic, and the data structures we have produced can be enriched to take into account other issues related to system performance. These technologies make much more accurate and flexible synthesis methods feasible. They also expedite the processes for detailed analysis of specific concepts. Building design rules and engineering knowledge into the CAD model offers a more detailed level of understanding than is possible with the current MVO tool, this understanding can then be used to improve the fidelity of the MVO process.
So, to conclude:
The requirements of MOD and other customers for future systems have been evolving rapidly over the last few years, and are likely to continue to do so. To provide intelligent customer capability for a range of systems requires more flexible assessment tools, particularly for novel concepts. We have described the process used to assess the aerodynamic performance of new concepts, and its impact at system level. New technologies, from a wide range of sources, have been adopted to reduce the time and cost associated with the assessment of concepts. New aerodynamic technologies, particularly those related to CFD, have been matured, validated and harnessed into the assessment process. Although there are significant areas where aerodynamics technology can be improved, particularly with respect to increasing the range of applicability of CFD, the customer focus on system-level issues means that multidisciplinary analysis and trade-offs will be a main area of interest.
Thank you, and good evening, any questions ?
The synthesis, design and assessment of air vehicle concepts S C McParlin and J J Doherty Royal Aeronautical Society Aerodynamics Research Conference 10-12 June 2003