I’ve just returned to Sheffield from a two day meeting discussing latest developments in Computational Fluid Dynamics (CFD) and Multi-Disciplinary Optimisation (MDO). I tried to cover the event live on my twitter account but as the talks were too captivating I missed out on loads of ideas that I wanted to share with you all. Therefore, in this entry I will try to summarize the main points that were addressed.
Background to RAeS
The Royal Aeronautical Society (RAeS) was founded all the way back in 1866 under the original name of “The Aeronautical Society of Great Britain”. Ever since then, it has grown in size substantially, reaching more than 22,000 members this year. As a global organization, RAeS aims to bring together the aerospace community through support of latest developments in the industry. A selection of publications, events and news can be easily accessed from www.aerosociety.com
Aims of the conference
Based on the programme (which can be accessed from: CFD & MDO programme), the conference really aimed to bring together leading experts in CFD and MDO to present an overview of current capabilities and challenges. As such, the talks were given by figures from a broad range of research institutions in Europe and North America.
Summarizing each individual talk would take too long so I decided to bring together ideas from multiple speakers to give an overview of current state-of-art, challenges and future prospects.
Numerical simulations, in particular CFD and FEA (Finite Element Analysis), are widely used in the design cycle of air vehicles, independently of each-other. Various talks showed the ability to predict flow physics of air vehicles. I was already familiar with the typical commercial aircraft flight scenarios but was also happy to find out about the following:
- vortex aerodynamics: discovery and analysis
- how CFD can be used to improve flight simulators of helicopters landing on aircraft carriers.
- various fighter aircraft scenarios: weapon bay aerodynamics, in-flight fueling, vortex lift, etc.
- how quantum computing could make simulations cheaper
Examples of MDO usage were also present in multiple talks. Some results, particularly from the Michigan MDO Lab showed some great examples of how aircraft performance could be optimised by considering both aerodynamics and structures. Yet, there was little evidence of MDO being extensively used in the industrial environment.
The figure below, taken from NASA’s CFD Vision for 2030 report shows the TRL (Technology Readiness Level) for various technologies.
1. Physics: It was a common opinion of many speakers that we lack the ability to predict flow physics outside or close to the edges of flight envelope. Complex flow phenomena governed by boundary layer separation, vortex interaction or unsteadiness can easily be “dissipated” when simulated. Below I adapted a sketch presented by Murray Cross (Airbus). We can observe a typical aircraft flight envelope on top of which the confidence in numerical results is given. It’s shown that as we approach complex flows, the confidence decreases.
Up to recently, U-/RANS (Unsteady-/Reynolds-Averaged Navier-Stokes) simulations represented the backbone of HiFi (High Fidelity) analysis within the industry. RANS-based simulations have limitations imposed by various turbulence modelling techniques. Yet, they can still be quite expensive and require many hours of CPU power to run. On top of that, CFD-only simulations are unable to predict MD behaviour.
As such, there’s a great deal of work trying to develop ROM (Reduced Order Modelling) tools which require fewer resources and can be developed to predict both aerodynamic and structural behaviours. As presented in the lead paper, ROM could eliminate the issue with “finding out info too late” and also help discover unfavourable behaviour since early in the design process.
At the same time, many desire the ability to run higher order simulations such as DES (Detached Eddy Simulations) or LES (Large Eddy Simulations). Making such simulations cheaper can require advancements in many fields, including, computer technology and algorithms.
2. Algorithms: Efficiency and robustness of algorithms is also of great importance. How well does an algorithm scale on thousands of CPUs is as crucial as how well are the numerical schemes implemented. A very nice sketch was included in Dr. Richard Jefferson-Loveday’s talk showing that as the fidelity of the simulations increases, so does the dependency on numerics.
3. Pre-processing: This is something I experienced myself a couple of times already. Pre-processing can take a significant amount of time. And sometimes this is not enough. We’re now looking at automation of some pre-processing steps. In particular, automatic grid-adaptation has a very good potential to improve accuracy of simulations whilst reducing cost. An example of grid adaptation can be observed in the video below.
4. Data extraction: One talk mentioned about 4TB of data being generated for every point that was simulated. By comparison, a 2TB external hard-drive was enough to back-up three versions of the data I computed in the first year of my PhD. Visualisation of all the data is not necessary, but different parts could help various design engineers. The way this data is stored can accelerate or slow down the design cycles. There’s been loads of mentioning of standardised data formats, centralised frameworks and tweaking of the outputs generated by simulation tools.
5. MDO: There are enough capabilities to independently analyse different disciplines. Similar to CFD, FEA (Finite Element Analysis) solvers can be used to predict the structural response of an aircraft wing. The challenges arise when trying to couple them together.
A first challenge is the need to standardise the I/O (Input/Output) processes such that, for example, the output from the CFD solver can be easily read by the Structural solver.
Secondly, paralelisation was also a big topic discussed by many speakers. According to them, CFD software can be easily run on hundreds if not thousands of CPUs whilst FEA solvers still struggle with this. Thus, the need to either develop a new type of structural solver, or a framework that facilitates the current mode of operation.
Finally, possibly the most discussed topic when it comes to MDAO was the need of developing efficient frameworks which facilitate the communication between different disciplines. This includes of course, communication between software, databases and designers. A few examples were given such as those developed as part of the AGILE project, or NASA’s openMDAO.
The previous sections discussed the current state of art and challenges faced. The future of CFD & MDO was also addressed throughout the two-day conference. There are many roadmaps (such as the NASA CFD Vision 2030 given above) which promise extensive improvement of our numerical capabilities in the future. I’m expecting to see a lot more of MDO at both ROM and HiFi levels. One thing is clearer now, that in the end, all these developments can only be achieved through a collaboration among the scientific community.
And as always, write down your opinions in the comment box below and contact me for any further questions you may have.