CFD METHODS: Overview of CFD Techniques

okay let's get this one straight the water effects from your video game is not CFD [Music] you know I have seen some people claim to have CFD capabilities when it was just a free extremely limited program that they downloaded from the internet in contrast one of the most powerful actual CFD programs currently in use today is actually an open source program and it's widely used in academic and commercial circles so when you hear the word CFD what are the different methods available and how do they compare CFD is a whole ecosystem of different mathematics methodologies all combining together the easiest way to classify this is with five different categories so first off you've got your mathematics then you've got your dimensions your time domain your turbulence and then your motions and how you stack it all up together depend decides on how simple or how complicated your simulation can be you can go all the way from a simple little desktop application that solves in 30 seconds to an extremely complicated application that requires a massive computer cluster first off let's talk about mathematics you have two major options here the first one is the boundary element method also called panel codes these are extremely fast you can solve these in about 30 seconds so the cost to run these programs is extremely cheap these require potential flow theory so you're completely disregarding viscosity that means there are sections of your problem that you have to completely exclude from your calculations that means you have to start applying some interesting meshing characteristics and some unique adaptations these were around in the 60s 70s and 80s back before computers got powerful the mathematicians came up with some very interesting workarounds to solve the problems at the time but that did mean a lot of approximations and assumptions so very fast and still very useful if you're doing preliminary studies or optimizations but I wouldn't really recommend them for final design and options if you're looking for something a little more robust the next option up is the finite volume method what also called rands codes these do require quite a bit longer to calculate you're talking more powerful computers taking 30 minutes to eight hours to run a scene simulation now but it does give you a lot more control over the solution accuracy rather than just meshing the edges of your bot your domain you are now meshing the entire volume of it and as well this makes the finite volume method an excellent general purpose tool when most people talk about CFD this is actually the normal baseline that they're talking about when they say CFD so this is your normal entry point for CFD prices is this finite volume method the next question you have to consider is dimensions are we talking 2d or 3d obviously if you want to do simulations in the real world we need 3d most CFD simulations are going to be 3d three dimensions however if you're trying to optimize a specific section shape like a wing section or a trust shape you can actually go to a 2d section and that allows you to do a lot of optimization and really work very quickly and very cheaply to do a lot of iterations so 2d is useful when you're trying to get the detailed performance characteristics for a single component or other element or cross section so again we're talking trusses wing sections rudders control fins that kind of thing what's really useful is when you combine 2d simulations with 3d simulations next up we have the time domain this isn't really something you get to pick yourself it's going to probably be dictated by the physics of whatever you're trying to simulate steady simulations our steady-state they're sort of the normal simulation they're shorter where we don't really have anything that changes with time you're just looking at steady flow characteristics so that's constant velocity flow in a pipe or ship traveling at a steady velocity a car traveling at a constant speed those types of things and that's your normal case for normal cost then have unsteady cases that's going to be much longer typically we actually have to first run a steady case to get our initialization condition and then we have to run the unsteady case to find out what's actually changing with time takes extra labour to do all of that there's now also an extra dimension to check CFD operators have to check the quality of resolution in their mesh in all three dimensions while that dimensional resolution also applies to the fourth dimension time so they have to check that as well so it's about 50% extra cost compared to a steady case how much extra and that really depends how long of a time domain you're trying to simulate and your CFD operator can tell you about that the physics really are what drives the cost for time domain there's also a third option for time domain a lot of people don't think about which is cyclic time domain this happens for things that are on a repetitive cycle things like propellers engines turbines in these cases we can actually use cyclic motion where the simulation itself it doesn't actually move doesn't actually vary with time and what we actually do is we take the the net effect of that rotational motion and we apply it as an additional velocity term into the simulation and that's really nice because you don't actually have to deal with the unsteady effects it just gets added in as its own extra velocity saves you a lot of effort and so that's not much extra cost in a steady simulation be sure to ask about that if you're dealing with something that's rotating turbulence this is the big question in CFD if you weren't really sure what's going to differentiate the different types of CFD methods out there the different types of CFD software's it's turbulence turbulence makes all the difference in terms of price one option that a lot of people overlook is laminar this is not something you're going to use when you're trying to get an accurate solution this is something you'll use for preliminary cases at high Reynolds number situations where you're trying to just run a lot of fast iterations so you'll separate out the turbulent viscosity and calculate that it's own separate component you'll be doing that as a hand calculation and then the CFD simulation is just going to do laminar viscosity and so again I would only do that high Reynolds numbers and only for preliminary simulations let's get into the cases where turbulence actually matters in that case you're going to Ran's le s de s or DNS those are the big boys and this is what we're really starting to talk about when we really talk about CFD the first option up is Rance Reynolds average navier-stokes equations you see turbulence happens just too quickly you would need time steps on the orders of Pico seconds to model it accurately and the computer powerful enough to accurately model that just flat-out doesn't exist it's not that we can't buy it it just doesn't exist it hasn't been made so instead Rand's came up with idea of trying to find some mathematics to come up with the average effects of turbulence we're not trying to track every single picosecond we're just trying to find out what the average effect is of all those picoseconds think of it like the lights in your house you know that the electricity in your house is cycling really really fast the electricity and the wires is actually cycling at 60 Hertz if you're in the u.s. 50 Hertz if you're over in Europe well you don't see your lights cycling that fast all you see is the average brightness in your house same thing for rands to get that average effect there are various turbulence models available you don't really have to worry about that that's what the engineer deals with brands is currently the mainstream method of CFD modeling and it works pretty well when you're talking streamlined objects things like cars trains boats ships planes pretty much anything that's meant to move brands has solutions for that they've been working on this for quite a few decades now and they're pretty good at it I would say this is again right about on the average of what you're expecting when you say a typical cost for CFD now let's take a step up let's go to the le s des methods a lot of the engineers started to find out that the rands methods were not working for all scenarios a lot of applications especially like offshore structures industrial buildings especially cylinders circular objects with flow running across them those were having troubles the tribulus models were not working correctly and the reason for that was because we were just ignoring all turbulence and forgetting it and really you need to look at turbulence and realize that there are different scales of turbulence there are small little Eddie's little tiny circles that we can ignore but then there are actually larger Eddie's that do influence the flow patterns and we need to capture those enter large Eddy simulations capture the large Eddie's and so those are where we are going now these are quite a bit more complicated their goal is to not ignore all of the turbulence they want to actually capture and model those larger scale Eddie's but still replace the micro scale Eddie's with a turbulence model unfortunately you don't get this added math for free this is quite a bit higher cost than Ran's methods I would say about a hundred percent higher cost and it gets worse the higher your Reynolds numbers go Reynolds number is a non-dimensional number to characterize the speed of whatever your object is the faster you move the more expensive your large Eddy simulation is going to be and remember this is a non-dimensional number so for those of you in the shipping world I actually have to say sorry I non-dimensional you're considered to be very fast so ships large Eddy simulation is incredibly expensive actually this is actually why they tried to do detached Eddy simulation to try and cut back on some of that cost it tried to use a regular rands approach near the object where most of the mesh is concentrated and then use the large Eddy approach away from the object where you've got less of your mesh it works somewhat but you still pay a fairly high cost so it helps but you're not getting it for free these methods I would say don't go to them just because they sound great only go to them if you have a definite reason to go there and again that's you're looking at applications like offshore structures industrial buildings cylinders specifically cylinders if you have things like waves going around a cylinder pilings that type of thing le s and des are definitely something you want to consider but be ready to pay an extra price for it the last turbulence modeling method is DNFs Direction numerical simulation there are no simplifications to the Navy or Stokes equations this should throw up a lot an alarm bell for you remember how I said that there are no real practical applications for modeling turbulence directly you would need time steps on the order of Pico seconds and the computer for the computer powerful enough for this just flat-out doesn't exist right now that's what direct numerical simulation is is they're actually trying to do that and yeah there really is no commercial application for that yet there have been a few academic cases of people trying to do that where they're really picking an extremely tiny problem to dry it to try and do this just to prove that it can be done so this is mostly in a warning in an alarm bell if vendors offer to do this for you to say I am so good I can do a direct numerical simulation for you run just run they're lying as you can tell we're now into the high end of the CFD spectrum the last thing you can do is add motion to your body your object whatever it is this actually requires you to morph the mesh of the simulation remember that the mesh is the thing that we put around our object to allow the to define the fluid around our body and there are several different ways that you can have motion the very first option which is the typical option is you can have no motion this is the cheapest of the options the next one is a prescribed motion where you're describing a mechanical motion you actually know what the motion is you tell the simulation and that's not too bad it gives the simulation a lot to work with and it's not too bad in terms of stability but it still does add a fair amount to the call so you're looking at about 20% extra and then the last option is D FBI D FBI stands for dynamic fluid body interaction this is pedigree this is top of the line for CFD where we are actually getting interactions between the emotion of the body and the simulation and solution of the flow around the body those two actually feedback on each other I have done this personally for solid body motion of a ship actually moving as a single solid body what happens is the computer first solves for flow and pressure around the object what we call the body that's your star generic term for it that determines the body motion it adds up all the pressure to get the force that gives your motion then that body motion feeds back into the flow pattern because the flow physics is aware that okay my body is now moving that has an effect on the flow patterns that can be highly unstable not to mention the fact I'm also changing the shape of my mesh as I go so definitely expect that to add cost time to wrap things up so if you look at this table this is a generic comparison of how all of these different CFD methods stack up together the blue line across the middle is basically your baseline when somebody says CFD this is what I think of as your baseline for CFD finite volume 3d dimensions steady-state rands simulation with no mesh motion and then you can see how you can go up in each one of those categories and the rough increase in cost that would happen for each one of those now the catch is if you start going up in multiple categories it's not necessarily purely additive if I were to say add on unsteady motion and dynamic fluid body interaction DFB I that's not exactly fifty percent plus fifty percent there might be an additional compounding factor in there so that might add up to an additional hundred and twenty percent that's one of those things you have to talk with your CFD engineer about and a lot of it really turns into what is your particular problem but this table gives you a general idea of what you should expect and most importantly now this gives you the tools so that the next time somebody says I can do CFD you are properly armed to your eyebrow a thumb and come back and say oh really which CFD thanks very much I'm Nick the naval architect thanks for watching don't forget to click that like button and subscribe for more videos and did you know that we produce more than just videos at DMS check out our website to find more articles free downloads and other help with ship design we offer a host of engineering services for budgets large and small so check us out to see if we can make your next project easier

Channel: DMS | Marine Consultant

Views: 21,366

Rating: 4.9519725 out of 5

Keywords: naval architect, ship design, engineer, marine, ship science, marine engineer, marine engineering, CFD, Computational Fluid Dynamics, Computational, Fluid Dynamics, Fluid Mechanics, Navier Stokes, Mesh Generation, CFD engineer, ANSYS, ANSYS CFX, ANSYS Fluent, OpenFoam, StarCCM+, CD-Adapco, CD-Adapco StarCCM+, Unsteady simulation, RANS, LES, DES, DNS, DFBI, prescribed motion, Finite volume, time domain, turbulence

Id: x-pbe3S6YSo

Length: 16min 47sec (1007 seconds)

Published: Mon May 06 2019