How often do you use of CFD (Computational Fluid Dynamics) ?

I don't know exactly the issue you had with the "coke-bottle SWATHS", but I can garantee you the same effects in every optimization-based loop (gradient, Hess, conjugated gradient, genetic, Laplance multipliers, neuronal networl -AI-... you named it) IF
- a wrong objective fonction has been entered, in the first place
- the parasitic drag of the junctions of the keels with the submarines is overestimated by the code.
In that case, it is easy to understand what's happened during the optimization. Basically, since the XVIIIth century, it is well know that the drag is function of the underwater area curves. Each "bump" in these area curves creates a additionnal parasitic drag, that can be minimized by removing some areas in the section where these "bumps" appear. Without any balancing paramater in the objective function, an optimizer will automatically decrease the section area of the submarines at the junctions, and this decrease will be as strong as the calculated parasitic drag.
Please, observe in the following picture, the result of a correct optimization process, in way of the empennage.

Of course, the testing procedure should be able to validate the calculations. As you said, it is very hard, perhaps almost impossible with limited budgets, to be able to simulate with accuracy real conditions. In consequence, it is also very hard to get test results that don't bring more question on the table than answers to the questions initially posed.

We have worked on the problem and have already figured out the issue. Having too many chefs is indeed a problem, but the main difficulty is that both CFD and tank tests were inline. There are reasons for that, and I will surely develop these reasons, on which I base quite a hard septicism about the seriousness of, not only incompetent actors, who are incompetent by definition, but, overall, of very competent actors, whose advices are not generally put in question. Being critical with tools is important, and being critical with people, whatever degree of incompetence they have, is equally important.

I personnaly mark this excellent advice.

Thank you for the link. Very interesting and educating. Here is a good man, indeed. What is said here is exactly the inverse of what is proposed in Simple Sabotage Field Manual https://www.gutenberg.org/cache/epub/26184/pg26184-images.html , The Simple Sabotage Field Manual, created by the Office of Strategic Services, OSS, published in 1944, to help resistants practice subversion during the WWII. An edifying document, where you find basic management and teamwork methods, that can be found, very astonishingly, in modern pseudo-management methods that aim, supposedly, to increase productivity...

 

Yes, Very true, as most CFD codes are also based on discretized schemes and discretized models. That's why, not only do the boundary conditions affect the result, but also the mesh of the domain of computation itself. In addition to the sensitivity analysis that JehardiMan is talking about, a mesh sensitivity should also be conducted, but is very often forgotten or just skipped. Since the cost&time of CFD is also multiplied.

Yes, I agree with you, this is what bothers me, about CFD. Tools are tools, theories are theories, both bound to a specific domain of applications. Ok. Charmed by the educative and nice outputs made by the companies that sell them, it is hard not to have the feeling of great power in our hands, resulting in very high expectations sometimes. But what makes some CFD customers think that they can spare the strong and fundamental knowledge that is required for CFD to be rightfully used ? Their lesser knowledge ? It bothers me that I have a quite a significant amount of examples, in my working area, where CFD is used just because it goes super easy, as time went by. Auto-meshing, pre-determined boundary conditions (virtual water tank, wind tunnels) , code that converges everytime, delivers R&D like pictures.

Everybody can understand, with the precious comments of you all, that the reality of CFD is something else. Not being cautious with this analysis could potentially lead to serious issues. I think that what is said, in this thread, is very important and should be read by anyone who is looking after CFD.

In addition, I wish everyone can have from you more details about the winning strategies that you used in implementing CFD in your design loops.

 

If I (a hobbyist designer) may be allowed a ramble, from the woolly world of piano tuning, up until the end of the 19th C, temperament (how to tune the notes in a scale) was a moveable feast, down to the level of different tunings for specific pieces. The idea that a twelve tone scale with equally spaced tones involved, mathematically, the 12th root of 2 caught on, and methods were developed to actually tune pianos to it, either using 12 tuning forks, or an ear method. Then in the 20th C, electric machines were developed to tune pianos to perfect equal temperament. Then a funny thing happened- people weren’t happy with the results, but tuners would show that the resulting tunings were accurate. Truth be told there was some cultural bullying going on. Some of us started messing with the temperaments by ear, and discovered that some pianos (and pieces) were happier ( that is people liked the results better) with slightly different tunings. After a few decades of tension between the machine camp and the ear camp, machine manufacturers started offering machines that had storable memory of a ‘good tuning’, so it could be accurately repeated. Which kind of works, as long as temperature and humidity haven’t changed too much, or a recital involves radically different pieces.

So what does his have to do with CFD? CFD is a mathematical expression of reality, as is the mathematical expression of equal tone temperament.

So when I tune, I tune my initial temperament octave to a small digital tuner (12th root of 2 stuff), and see how it sounds on the piano, and play some typical chords inside the octave, and see if it sounds natural, and then start fiddling with it. If I get lost in all that, I go back to the mathematical model and start over- this usually takes 5-10 minutes out of a 45 - 80 minute process. Once I know look for as far as deviation, I can check back occasionally to the machine as I go up and down the keyboard, mainly not to waste time turning pins.

So, when I started messing with the Vacanti, I started looking at a plethora of Master’s theses comparing wind tunnel data to mathematical modelling, and found the phrase “the results show a reasonable fit” of whatever mathematical model was used to replicate wind tunnel testing. That was back in the 90’s, and things nibble away at more fidelity.

Which is a way of repeating what others are expressing here.

But! CFD does afford a really fun way to mess with wild shapes (and incrementally morphing known shapes) without anything except 1s and zeros being harmed, which is good for me, since building stuff takes time and expense. It’s kind of like using a tunable programmable synthesizer to test by ear alternate temperaments- and this is useful, because, for example, I have a PhD thesis on my shelf with 2,000 different historical temperaments inside it, and different temperaments are showing up on the web, so it’s good to know this stuff, as the focus of temperament seems to be getting fuzzy with time. But! Hearing a temperament on a synth isn’t the same as hearing it on a piano. I once had a piano teacher insist on tuning her 7’ Steinway to her Kurzweil- told her the piano would sound like crap, but she insisted, so I tuned her piano note for note to the synth, she almost had a nervous breakdown with the results, even when we sat down and played note for note, and they were the same.

So when I mess with the Vacanti, I run a lot of different hull and water possibilities, heel, trim, etc etc until I get a general idea of what the software is showing me, and then pull back, and try to look at everything from an intuitive viewpoint, which pretty pictures help a lot! (I try to imagine flow in my brain too, and some fluid flow professionals have said that I have a pretty good idea of that, which is nice- staring at enough wind tunnel images, massive numbers of telltales and water itself does work, a bit at least….)

These parallels between two incredibly complex events- tuning and boat design are close enough, to me, that I think they are different expressions of the same cultural phenomena- experience, and taste interacting with mathematical modeling. A cerebral dance, really. With clients involved!

Hope that made sense.

 

Lol. I'd like to stay close the whole time when my piano is tuned. Can be quite a stress, I reckon...


Please, Paul, correct me if I'm wrong. In this analogy, the digitalization of the physical principles and of the physical models, resulting in a "numerical picture of flows" in CFD is replaced by the 12th root of 2 based temperament in music, which could represent the mesh size of the flow. As for the sound of a piano, a the flow around a boat cannot be reduced to its numerically model, but universal trends can be observed from them, from which guidelines can be drawn. The knowledge, or the intuition of the the difference between the real piano, or the real boat, and its idealized version could then lead to the appropriate tuning, that takes into account the real nature of things.

If correctly understood, I must say that this analogy could be pushed further. Regarding the piano, its material, its harmony table, the way its elements are put together, as well as its imperfections give each piano an unique signature, by the way the harmonics develop inside the sound. Although being not much talented, I've got a perfect ear, and if a piano has not been recently tuned, I'm able to tell what kind of music has been played on. Success 100%, if the piano is played by only one person, on a regular basis. My feeling is that the structure of the whole piano interact strongly with the player, and, as time goes on, small defects or modifications appear in the structure, in the tension in the instrument. As a result, the strings and the sympathy they convey in their harmonics tend to fit the player and its music play, somehow.

For instance, it is very easy to know if a classical music has been played, or, instead, if the player was more a jazzman. Just play any classical piece on the piano that has only known jazz, and my ears are assaulted with unpleasant sound. The inverse is also true. It goes the same for a car, and the way an engine is used, the way the car is driven. And for a boat, also, in a similar manner. I think. Each geometrical or structural detail or arrangement make each boat very unique, regarding its response to external excitation. And no numerical approximation will ever be able to capture the influence of these details, that can have great effects on the boat, its driven flow, and its performance. Like harmonics, some details can be the source of effects, which, when combined, can even override every other classical and awaited phenomena.

I hope this will make a little sense, as my musical lexicon in english is very poor.

 

There was an extensive survey conducted. The results were not just significant, they were unanimous! QUACK!
10,00 ducks can't be wrong!
The point is the opinion depends to a large extent on who you ask.
Post #2 is what I would expect from a good large professional outfit with NAs secure in their positions.
-CFD is useful where there is no sufficient experience, or it is simply easier than measuring anything relevant.
-trust is TBD
My only caveat to this reply would be the date -the cost of CFD has dropped and the quality improved dramatically.

"ORCA3D Marine CFD allows the naval architect to :
- Avoid/minimze expense and time associated with model and full scale test programs
- Evaluate the effects of specific unique design features
- Perform high fidelity design evaluations in earlier stages of design when decisions have the largest impact
- Discover and avoid potentially costly design flaws before the design is built (bolt and underline by the quoted engineer himself)

What do you think of these arguments ?"

I think they are a management wish list. "Pay Orcad a large sum of money and get the payback from your NA". From the perspective of the NA, this is money and time charged to the project to get a second opinion. If the design is within the experience of the NA, the second opinion is counterproductive.
Even if the software does the work and the NA doesn't duplicate the effort, who will be responsible for the outcome? Orcad?? NO! The manager that directed the program? NEVER!

CFD is not a sham. It is a tool to gain insights. Those insights are not always valuable or worth the time and expense.

If y0u are looking for positive responses, look at how CFD could be key to an NA winning higher paying work developing vessels he/she DOSEN'T have past experience.

If you are afraid of good NAs losing work to bad CFD, wait until you see what AI can do. CFD will be your best defense.

 

Thank you Skyak. Lot of informations in few words. That's great.

Is it the quality or the accessibility of CFD that has improved with the time ? The closing equations of the approximations of the NS equations have improved a bit, but back in the 70's, CFD was already considered very accurate. Isn't it that, instead, every average people is given, right in his hand, with a simple smartphone, the memory and the power of computations that only top researchers had access in the past ?

In my opinion, here is a very wise comment.

Yes, indeed, but how can the CFD results could be trusted in the NA has no past experience on what is being simulated ? Let's take a very simple case of a keel bulb. The amount of drag is related to the transition & stagnation zones of the flow around its shape. Every automated mesher is able to numerize the shape, but the accuracy of the drag computationt is given by a correct capture of these zones. An adaptative remesh scheme will produce a very good looking definiton of these zones, bu how can you tell, just looking at your CFD result, how close the result is from reality ?

What is your experience with neural networks applied for flow simulations ? Do you think that, like CFD, the simplicity to use AI will also conquer the heart of NA ?

 

This analogy really makes a perfect sense to me. Still pushing it to its bound, may I try this :

Buy a Bösendorfer and a Kawai Shigeru of comparable sizes. Put your fingers at random notes simultaneously. Do the same on each piano. What do you note ? The Bösendorfer will produce an unpleasant sound, while, although unpleasant, the discordance of sounds will be less agressive. I guess this is because pianos are systematically given a color by their manufacturers. Some are made to produce a pleasant sound only if they are used a certain way, within a certain repertoire. Hardly no jazz man will play a Bösendorfer, because a very high talent is required to make pleasant the jazz harmonies on this piano. They will usually prefer Kawai or Yamaha in concert. The same goes also for the speakers. In the studio recording, the monitoring speakers produce a very flat, unbalanced rendering of the sound. Making the sound engineers able to finely distinguish unpleasant harmonics of sound, to be re-worked. Lounge speakers, on the contrary, and even if you buy some of the most expensive, will filter and highlight the most pleasant combinaison of harmonics, for you to have the rewarding feeling of the quality of the sound that you hear. But it's artificial, and has nothing to do with the intriseque quality of the recorded music that you hear.

The same goes with CFD. It produces pleasant and good looking results, but that could have nothing to do with flows in reality. If, now, you put your fingers on a synth at random notes simultaneously, it willalmost always produce an acceptable sound. This is because the synths have a limited bandwidth for harmonics to develop. A natural sound produces by a natural instrument has a infinite number of harmonics, and the interaction of them, even those at very very very high frequencies, not in the range of human ears, can produce effects that do happen in this range. Real flow are shaped by the infinite precision of real shapes, whose precision cannot be obtained by the act of numerization. That's why, in my opinion, the precision of numerical results will asymptotically converge to an impassable boundary, at distance from the reality. Skyak has mention the AI as a new way to do CFD. There is also the Quantum Computing. In my opinion, changing the bits by Q-bits will certainly increase the speed and open the gates to wonderfull real-time RANS CFD, but as long as a quantization of the model and of the physic's principle will be made, this kind of simulation won't overpass this boundary.

But I guess also that, in the same time, we will be overflown by optimistic communications and papers made by very serious scientists extolling the quantum realm, proudly chanting how wise, intelligent and remarkable are their approaches.

 

what you may be hearing is what pianistic touch (hard or soft), does over time to tuning (besides changes in temperature & humidity) and the shape and hardness of the striking surface of the hammers- when felt hits metal, felt packs down, and grooves form in the felt (take a look at your hammers), which if you think about Fourier transforms, flat grooves with corners change the shape of the induced waves on the strings. Hammers, which can thought of as 3D springs can have different hardness, as well as fillers, such as melted plastic being added to them to make a very hard surface. The shape of the tip of the hammers, such as round, elliptical (Fourier analysis again) also influence the waveforms (overtones) of a string. Add to this how musicians process all this, and what they want, the sophistication of their technique, the resulting feedback loop back to the fingers, or for listeners, room acoustics, listening experience, and it starts getting interesting. And I haven’t addressed touch weight (how much weight it takes, statically or dynamically to push to hammer escapement) which has its own can of worms.

But is this much different from the interaction of sailor and boat? For example, some 5.5 meters class boats have come with a technical handbook explaining how to get the most out of that boat. So much for intuitive helming! Some boats respond to a more brutal hand on the helm, some a more gentle hand. Some boats have a brutal helm, some lighter. Some boats need to be worked on more after each sail. Boats have grooves too, but more dynamic, like upwind performance grooves. Some don’t. Some are balky, some not. I’d like to think CFD at some point will be able to explain these differences, and possibly be connected with the neurophysiology of the individual sailor to create boats tailored to a particular sailing style. Of course, gifted designers have been doing this all along, while others practice more around a specific design approach that forces the sailor to sail to the boat in a particular style. Pianos are no different, and each manufacturer has either appealed to different tastes, artistic neurophysiologies or forced the artist into its particular aesthetic. In the piano world, this sometimes referred to as ‘playing to’ a piano, or ‘playing through’ a piano. Piano makes have also evolved or morphed to different aesthetics as time, criticism, economics, ownership, music and market forces March onward. Different technologies have disrupted everything, creating different tastes, which evolve, etc etc

I think this sort of dynamic might help explain, for example, the Finn Dinghy vs the Laser, for example. Laser to Windsurfing to Kitesailing. Archimedean to planing to foils. I’ve never really seen a CFD presentation of the differences.

The delicious aesthetic and scientific complexities (some above^?) might explain why so many boat designers are also musicians & artists?

 

Ah! I was a concert station technician for both Bosendorfer and Kawaii (among others), so a few thoughts-

- Bosendorfers are profoundly sensitive to hammer hardness, voicing, shape, strike point and the amount of power (touch) they’ll tolerate before the tone distorts (or splatters). Softer hammers were more the norm until an Artist who shall remain nameless who was used to playing Steinways (which revel in massive amounts of power) got his hands on a Bosendorfer and applied the same amounts of Steinway power hammering to it, and the poor beastie distorted like crazy. BUT! This happened during a period of Rock and Roll (and this is my personal apocryphal opinion) when distortion was coming on, so Bosendorfer distortion was proclaimed cool among some part of the cognoscenti, and Yamaha, which is usually a bright piano, owns Bosendorfer now, cultural things creep in, and the soft/harsh battle rages on. (FWIW, when I prepped a Bosendorfer for Oscar Peterson, he liked voicing under the distortion zone). Bosendorfer, to my mind, is a piano whose tone is more like a warm candle flame than an LED, transparent in tone color and subtly but infinitely expressive.

Kawai as I understand it, was influenced by Bechstein, which at the time (1930’s) was a very warm aesthetic, but has brightened over the decades. Their soundboard technology and design have become fantastic, which supports a very delineated opaque bright sound that also is expressive, and can be voiced to meet a broad range of taste. However, fewer changes in tonal color are available.

and this is where taste come in-do you like a transparent tone, where you can hear overtones clearly, almost as a chord for every note? (It helps to remember that Bosendorfer started making pianos long before modern pitch and equal temperament, so that culture is still in the mix. Kawai was founded after modern pitch and temperament were established as the norm). Would a synth programmed to sound like a 7’4” Bosendorfer sound good to you?

Like most things with increasing complexity, it starts boiling down to taste. And things are likely to get more interesting (diverse?) with quantum computing, since what to you do when 1 might not equal 1? What does a piano sound like? What does a sailboat do? What does CFD replicate? Can CFD replicate the feel of the sailor/boat interface? Who buys a boat because of how it feels?

Which begs a question about sailing- does the expected feeling from a boat culturally influence or even define design? Can CFD define the feel of a boat? And then replicate it?

 

Ask the same NAs "Do you trust Navier Stokes equations?" If the answer is no there is no point in further discussion but I doubt any will say no.
CFD estimates a solution to those equations to a confidence level. Major sources of error include;
-inaccurate or insufficient definition of inputs
-insufficient resolution
-running outside model comprehension -ex. elements defined as liquid become gas
-too high a tolerance on convergence

All of these were compromised greatly to fit the economics of computation. Those computation economics have changed exponentially over time from a hardware perspective and code efficiency and capacity has gone through several breakthroughs.
I would replace your "average user with a smartphone..." statement with -gone are the days of the PHD spending days meshing a model and saying "EH, time to run this for a week and see if it comes back". The average (untrained) person still can't achieve accurate CFD and can't tell right from wrong.

A good keel bulb is well within the capabilities of CFD, but then it is likely also well within the capabilities of an experienced NA without CFD. The most likely source of error would be the condition of the water passing over it. To be of value in this case CFD could do a series of cases for different angles of attack and turbulence.

AI can easily learn to make images that look like CFD. AI can make those images or even video in real time today. AI is NOT producing a valid solution to NS equations.
AI is a "black box" solution to everyone. CFD is only a black box solution to people who don't understand CFD.

"Do you think that, like CFD, the simplicity to use AI will also conquer the heart of NA ?"
Ducks will still say quack. By that I mean that replies reflect distrust of bosses and customers as much as distrust of what CFD shows. What I would tell NAs is that your posturing and distrust of CFD is misplaced. AI's ability to influence is beyond your comprehension and if you don't learn from CFD, AI will.

After taking a walk (I think better walking) I felt the need to add this edit to the post.
If anyone feels any disrespect or insult in my two posts here, please disregard whatever offends. I have great respect for both experience and mathematical simulation and am grateful for what is shared here.

 

What does AI learn? “You are thing and I love you?” Granted old school, as the last couple weeks have shown. If you can persuade it, I understand that AI is really good at repetitious stuff, which is one mode of complexity, but as Bing has shown us, AI is pretty much the product of hir maker, as those of us who have lived near the Microsoft light cone instantly grasped.

(What was it that Steven Wright said? “I have a map of the United States at home. It’s full size.”)

And Alan, you have a point about synths- it has been coming clearer to me since retiring that synthesized sounds are the real competition to analogue expression. But playing better than a machine requires substantial risk.

 

That's a delicate attention. You've said what you had to say, and it's valuable, and very clear, Skyak.

 

It makes a perfect sense. When trying to "capture" the color of a given piano, I usually put my attention on the feed back I have on my fingers, while focusing on what's happening between the keys and the strings. Bowed instruments players use to play with this, but I didn't know how much a hammer can be thought as "3D springs". I'm very happy to have made a small progress in my comprehension of sound, and, how it is produced.

I'm not a NA, so, I don't know if, like in music, they got to learn also the neurophysiology associated with the sailing practice. From the NA I know, It may be that, as you said, some are natural of this approach, some are not.

I too ! Would be a great insight into the work of a NA, that is not only technical, but has also a lot to do with understanding un-expressed wishes, advising and guiding their clients. For that, appropriate words are used in this exchange, those words that qualifies and explains their desigh. Like having Debussy on your sofa explaining to you the meaning, the inspiration and the goal of its music.

Good point. Not to forget that, in the Classical Antiquity, the Artist was, in his true definition, a man who uses a precise technical knowledge in his realization. This meaning just faded away during the Middle Age, to be definitly buried by the Industrial Revolution. In the extreme, artists were no more than just a bunch of crazies under the influence of drugs and alcohol. Still today, it is very common to name an engineer "an artist", meaning that he is not rigourous and do things by inspiration only. But every body who is practicing any Art knows that, on the contrary, being an artist means working on his technique, skills and knowledge to the maximum, and follow strict rules to guarantee certain desired results. Although there is, always, some rare exception to this rule, artists who do not get to know, but just know, by the Gods themselves or by a strong natural instinct. In relation with CFD, the word "Guru" appears many times in this forum. I worked with HDT (Hughes de Turkeim, windsurf designer for Tiga and hydrodynamic Guru) who was consulted, later after its odyssey with Tiga and Audi, about foils of very advanced racing projects. Like T.Speer, he had a real "magic touch", and all the technical explanations he gave of his realization were mere afterwards attempts to convert his mere intuitions into shareable knowledge.

To me, AI is just an old gradient descent optimizer that has been refined and adapted to the actual computing power. I've programmed myself neuronal networks (NN), in order to estimate what can be achieved with this code, for a few different tasks. In itself, its automatised process of information is nothing new under the sun, but, indirectly, the classification on which NN operates, and the abilities given by syntax analysers, coupled with these NN, could perhaps, in a near future, gives us very important clues on the links between subconscious, neurophysiology and technical concepts.

But at any time, I think that, should be kept in mind, that, these results will only be "pictures" on our own cultural and sociological perspective on things and concepts, pictures that do not contain no real truth, but that evolves with the entiere society, like the Organisms Classification or the Periodic Table, have evolved to reflect the progress of the science, acquired by other means of investigation.

 

Here may be a very important take-away point about the why? of CFD.

In my career, in all the Naval Architects and engineers I have worked with, the number of people who I have met that I would call a "hydrodynamicist" I could count on the fingers of one hand. In my opinion, a "hydrodynamicist" can mentally visualize complex flow and its effects. A "hydrodynamicist" just is, and while education and experience can hone the ability, they cannot create it. A "hydrodynamicist" does not need CFD, that is for us lesser mortals who need a crutch. It is like comparing Langley to the Wrights to Curtiss; Langley never got it, the Wrights plodded through it, but Curtiss visualized it.

 

Since we have had the example from the world of music,I hope I can be excused for introducing another field that uses CFD heavily.The world of top end motor racing is dependent of the entire package working at peak efficiency and wind tunnel testing is the final hurdle to clear before a new development gets put on a car.Even with state of the art 3D printers,it takes time to create the parts for the wind tunnel.To avoid chasing down too many blind alleys a range of concepts are run through CFD evaluation and only the most promising reach the 3D printer.The people running the CFD simulations are specialists,the software is top end and the cars (at least in Formula 1 these days) are meshed with 30 million cells.That level of detail has been found to be required and since there is now a restriction on the amount of CFD done, we may reasonably assume that it isn't squandered.

To deal with this level of processing takes serious hardware and it won't be an otherwise idle workstation,it will be a large cluster of dedicated machines.In a different branch of competition,I read an account of the development of Bloodhound SSC ,which for those who are unfamiliar with the name,is a vehicle intended to surpass the 1000 MPH barrier.The aerodynamics team is led by the distinguished Ron Ayers and he had no illusions about the scale of the challenge.The CFD work was done at a Welsh university that had the computing capacity and it took nine weeks of processing to achieve a steady state result.Other than specialist military programmes,I doubt that many nautical projects would justify this level of resources. Which does make me wonder what could be achieved if only it were possible to do so.