As its name suggests, computational fluid dynamics (CFD) uses numerical analysis and data structures to solve fluid flow problems. Life would be easy if you could simply convert 3D CAD models into a mesh of simple shapes and nodes, allowing the behaviour of fluids over the surface to be predicted and designs under development optimised.
But very few real-world problems can be solved easily, as most situations involve complex interactions. So modelling fluid behaviour using CFD requires a lot of time, skill and computational power.
If you are familiar with the process, you may be reminded of one of the final scenes of the film Apollo 13, when one officer is briefing another onboard the USS Iwo Jima:
“Let me put it this way. The trajectory may be off, their thrusters may be frozen, their guidance system might be malfunctioning, their heat shield could be cracked, and their parachutes might be three blocks of ice. Clearly we have got some obstacles to overcome.”
CFD is a complex and delicate discipline with many potential obstacles. It requires a lot of resources, both in terms of experienced labour and computational power. To allow for better, quicker analysis and also to reduce the number of potential issues encountered in each step, other unrelated disciplines can be brought into the mix.
So, where do we start? There are multiple disciplines and scenarios which can have desirable outcomes to solve our CFD complexities. Automation and the application of Artificial Intelligence (AI) methods are now areas being investigated to overcome these challenges. Examples of these in CFMS touch all three elements of the CFD cycle.
When receiving a CAD file from a design department or a customer the first problems occur because this data does not necessarily meet the needs of CFD engineers. CAD geometries must be literally ‘watertight’, so no gaps can exist between surfaces and there are many superfluous features that must be removed, literally by hand.
CFMS have developed tools to automatically find these issues, allowing the designer to rapidly ‘defeature’ the model, by removing unnecessary elements (bolts, holes, etc) at the click of a button. This, for complicated geometries can mean weeks of preparatory work is saved before CFD can be conducted.
Automated shape refinement in response to a requirements change can also accelerate the creation of CFD meshes and solutions. This is especially valid for fluid structure interaction cycles, but also when trying different feature designs in a component. CFMS is using an approach driven by key product parameters to automatically generate a shape and mesh for a given component. Since a single mesh can take weeks, it is easy to understand how automating the creation can shave off months of work from a single project if a range of options need to be evaluated.
CFMS has a high performance computing (HPC) facility available for CFD which comes into its own when used for complex unsteady analyses with high cell number meshes. While this seems an easy solution to the computational resource problem, even with thousands of cores an analysis may take days or weeks, which comes at a cost.
CFMS are now adopting AI methods to reduce the level of computational power needed to solve complex CFD problems. Using a combination of simulated and real world data an AI surrogate can be trained to create a digital twin of the problem in hand simplifying the computational need and potentially realising analysis in real-time.
Overall, the approach to combine CFD with automation and AI is one of many examples of integrating different disciplines for faster, better workflows. Much like CFD is a branch of fluid mechanics, CFD and Data Science is already an integrated discipline at CFMS. By supplying each part of the CFD process with new solutions and methods these are helping to reduce uncertainties, saving time and costs, obtaining better solutions every day.
The beauty of this approach is that, while traditionally CFD is associated with aviation, any application involving a fluid interacting with an object is an ideal field of application. From automotive to trains, to buildings, to microchips… if you have something flowing onto something, CFD and AI are there to make life easier and better.
If the dialogue from Apollo 13 sounded like “let me put it this way. All potential issues have been resolved automatically, no more obstacles to overcome”, it would have probably been a boring movie but a desirable engineering outcome. In reality, this is exactly what the CFMS simulation team is now doing.