10 MIN READ
Error – CFD does not compute
I recently joined Team to head up our in-house CFD (computational fluid dynamics) capabilities and develop our expertise in engineering simulation. CFD is a powerful tool for simulating fluid flows and related physics in a virtual environment and can be of great benefit to the development of medical devices. But creating effective CFD simulations can be a challenging business.
One of my roles is to make sure our clients know more about our skills in this area, and so for my first Insight article I’ve decided to write about the pitfalls of using CFD…
All math models, CFD simulations included, are approximations of the real world and it is important to accept that there will always be some level of error associated with CFD results. But that doesn’t mean the results can’t be useful. We just need to verify the predictions with some carefully planned experiments before we can assess and confirm their usefulness.
Verification of CFD is essential for checking it is providing a sufficiently accurate approximation of the real world to inform design decisions. Without this, CFD is just a bunch of complicated maths solved over many hours by high-powered computers that happens to be meaningless as an engineering tool.
Despite this, the temptation to rely on CFD without verification can remain strong. After all, isn’t a key advantage of CFD supposed to be that it can take the place of experimental testing and thereby save valuable time and resources? Well yes, the number of prototypes and tests can be reduced using CFD, but testing shouldn’t be neglected all together. Take the example of the 2010 Virgin Racing F1 Team, who took the ambitious but ultimately flawed decision to design their F1 car using solely CFD. Considerable time and money was inevitably saved by avoiding verification with wind tunnel testing, but unfortunately the performance of the final product spoke for itself. Two seasons completed and zero points scored.
Simulation and physical testing inherently complement each other and so shouldn’t be viewed as competition. The insight gained from CFD can maximise the reward from testing by helping to explain prototype performance and by identifying key parameters for the effective design of experiments. The results from testing can then maximise the reward from CFD by verifying the simulation approach and exposing inaccuracies. By applying the right combination of both techniques, a seriously streamlined process for device development can be achieved.
Desperately seeking complexity
Fluid dynamics are inherently complex, as will be familiar to anyone who has marvelled at the curling and twisting of smoke rising from a campfire, or wondered at the structures created by milk mixing in a morning coffee (it can’t just be me?). But attempting to capture all of these details using CFD can quickly cause a simulation to become overly burdened by complexity. Computational time suddenly extends into weeks, simulation robustness falls off a cliff and the use of CFD becomes an impractical encumbrance.
The tight budgets and times scales in medical device development demand lean and flexible simulations, and it isa skill of the CFD engineer to make pertinent assumptions and approximations that limit simulation complexity.
This approach is especially relevant in the use of CFD for DPI (dry powder inhaler) development. The finer details of the physics of powder aerosolisation and deagglomeration within DPIs is still not fully understood, and CFD that attempts to simulate all of the processes involved remains in the realm of academic research. Employing such a simulation during device development slows the process, takes weeks to churn out heaps of data that is already out of date before it is calculated and is generally not recommended.
A simpler simulation, such as the flow of air through an important section of the concept geometry, is the starting point to quickly gain a good understanding of device characteristics. This enables fast investigation of large-scale design changes and efficient guidance towards a solid baseline design for the device. Simulation complexity can then be added in a step-wise fashion if required for accuracy when fine-tuning the design and optimising performance. This may involve extending the airflow simulation to the complete device, for example, or tracking discrete dose particles as they travel through the device and then simulating the delivery of a complete powder bolus — but only if this extra complexity adds value.
Of course, leveraging the latest in computing power can increase the levels of CFD complexity that are practical during a project. Processor speeds may be reaching a plateau, but we can now turn to parallel computing for increased performance. Most commercial CFD codes include sophisticated algorithms to efficiently divide a simulation up across multiple processors, and a relatively small investment in high-spec hardware can yield high rewards in CFD effectiveness.
Treating CFD as a Black Box
Commercial CFD software has come a long way since it first appeared in the early 1980s, and development continues at a beguiling pace. Much of this work concerns the commendable improvement of CFD accessibility for the engineering community through automated processes and enhanced user interfaces. But with this comes a dangerous tendency to treat CFD as a black box; define the inputs, press the big green “default settings” button and accept the automated results.
This incurs a significant risk of falling foul of the ‘garbage in- garbage out’ scenario common in computer science. This problem is exacerbated with CFD because solving the relevant fluid dynamics equations is often non-trivial, and so there is no guarantee the software will achieve a valid set of solutions. Even with a perfect set of non-garbage inputs to the black box you can still end up with garbage out.
With little experience of CFD or fluid mechanics it is possible to produce results with CFD software that appear convincing but have little resemblance to reality, which can be very misleading for the project team. The ultimate goal of CFD software — to consistently produce meaningful results from fully automated processes — remains something for the future.
CFD is most definitely not an iPhone (it is supplied with an extensive set of user manuals) and it is important to take time to understand the processes under the hood of CFD software and how these can affect results. This, coupled with a sound knowledge of fluid mechanics, is essential for appropriate interpretation of simulation predictions.
Curing rather than preventing
In the world of healthcare we know that prevention is better than cure, but when it comes to the use of CFD for device development this is sometimes forgotten. It is true that CFD can be a very powerful tool for diagnosing and curing problems with device functionality, but it is even more effective when used to prevent problems arising in the first place. A recent study of engineering simulation strategies employed across a range of industries found a very clear conclusion that “using simulation to gain better insight into product behaviour from the very beginning of the design process is a key differentiator of success”4.
This success is partly due to the ever-increasing cost of design changes as a device progresses through development. Using CFD to help predict and prevent problems in the initial concept stages may appear unnecessary if prototypes are performing well, but it is far more cost effective than attempting to fix design problems discovered when at high volume manufacture.
But implementing simulation early isn’t just about reducing costs, it is more about achieving better devices through improved understanding. A device can be thoroughly tested and proved to work consistently, but if there are gaps in our understanding of how the device works then there is always a risk that it is not optimised and that problems may be encountered further down the line. Simulating early helps mitigate this risk, and the resulting improved understanding can inspire innovative new routes for development at a point when taking new routes is still possible. CFD left late in the development process is often seriously compromised in its ability to benefit device performance.
Thinking you can survive without it
Scepticism towards CFD is sometimes present in medical device development, possibly resulting from the pitfalls I’ve been discussing. It can be healthy to treat results with a critical eye, and this helps prevent mistakes, but if it causes simulation to be discarded altogether this can be a bigger mistake still.
Engineers often take inspiration from nature, and inspiration for simulation can be found in the theory of evolution. The ability to simulate is believed to play a significant role in evolution by natural selection, as noted by biologist Richard Dawkins:“[Organisms] that can simulate the future are one jump ahead of [organisms] who can only learn on the basis of overt trial and error. The trouble with overt trial is that it takes time and energy. The trouble with overt error is that it is often fatal.”5
This sums up the benefits of simulation succinctly. It can achieve a competitive advantage by saving time and resources and by providing the knowledge to develop superior devices with lower risks of failure. If we’re only interested in developing the very best medical devices we can, then we need to make full use of simulation tools when appropriate.
Widespread acceptance of the benefits of engineering simulation for device development is reflected in the fact that the FDA recently published guidelines for including computational studies in medical device submissions6. This is a welcomed and comprehensive document that clearly describes the components of a successful study to ensure it contributes valid scientific evidence; a check-list for simulation effectiveness. CFD has firmly established itself in the device developer’s toolbox. However, as I’ve highlighted, merely ‘doing some CFD’ isn’t enough. If this valuable tool isn’t used correctly it can have significant consequences for device success.
This article was taken from issue 8 of Insight magazine. Get your free copy of the latest issue here.
1. Box, G. E. P., Draper, Empirical Model Building and Response Surfaces, Wiley Series in Probability and Statistics (1987)
2. Pidd, M., Just modelling through: A rough guide to modelling. Interfaces 29:2, 118 – 132, (1999)
3. Slotnick, J. et al, CFD Vision 2030 Study: A path to revolutionary computational aerosciences. NASA (2014)4 Houlihan, D., Jordan, The Impact of Strategic Simulation on Product Profitability. Aberdeen Group (2010)
5. Dawkins, R., The Selfish Gene. Oxford Paperbacks, 2nd Revised Edition (1989)
6. Reporting of Computational Modeling Studies in Medical Device Submissions. Draft Guidance for Industry, FDA (2014)