/sci/ - Science & Math » Thread #12615332

154KiB, 494x422, fea.png

View Same Google iqdb SauceNAO

Anonymous Fri 22 Jan 21:17:42 2021 No.12615332 View Reply Original Report

Quoted By: >>12615374 >>12615413 >>12615426 >>12615478 >>12615545 >>12615622 >>12615933 >>12616784 >>12616796 >>12617577 >>12617719 >>12617740 >>12618198 >>12619329 >>12623276

>working on capstone project for my Computer Science/Systems degree
>asked to use "finite element analysis" to model the effects of ocean waves on an autonomous ocean glider
>Furthest I got in math/physics was vector calculus 1 and intro to physics with calculus
How fucked am I? They made it sound easy, but looking at the wikipedia for the finite element method is making me think I'm way out of my depth. I don't even know how to solve differential equations.

Anonymous

Anonymous Fri 22 Jan 2021 21:22:22 No.12615357 Report

Quoted By: >>12615373

it would be impossible for you to do it by hand but maybe you could do it on wolfram alpha or matlab or something like that.

Anonymous

Anonymous Fri 22 Jan 2021 21:25:29 No.12615373 Report

Quoted By:

>>12615357
>impossible to do by hand
Well I figured as much. Otherwise they'd have a physics student working on it.

Anonymous

Anonymous Fri 22 Jan 2021 21:25:43 No.12615374 Report

Quoted By:

>>12615332
there will be a learning curve for sure
you don't need the tools of diffeq to solve FEM but the experience of working with diffeq would be helpful.
good luck and don't rely on wikipedia for a great intro to math topics.

Anonymous

View Same Google iqdb SauceNAO cow.jpg, 55KiB, 520x403

Anonymous Fri 22 Jan 2021 21:34:51 No.12615413 Report

Quoted By: >>12615453 >>12622955

>>12615332
you need a firm grasp of integrals and linear algebra. familiarity with the physics involved is also a huge plus, i couldn't imagine trying to do numerical simulations without some idea of what i'm modeling.

look up Galerkin Method. essentially you take the partial differential equation (PDE) you want to solve, multiply it by a family of spatial "test" functions and integrate to get a weak form, then relate all the "test" functions on the input and output side of the equation in a big fat matrix. you also have to choose your test functions such that they satisfy the boundary conditions (BCs) of the domain you are solving on.

none of this even scratches meshing, which is extremely non-trivial.

i'm a PhD candidate that's implemented several FEM solvers for various electromagnetism problems. it's not easy. a full derivation from scratch took like 20 pages of LaTeX on one homework. you need to know your shit, and all things considered EM is easy. acoustics and navier-stokes, hoooooly shit you're fucked.

i might copy/paste my derivation so you can see what steps go into it

Anonymous

Anonymous Fri 22 Jan 2021 21:37:33 No.12615426 Report

Quoted By: >>12615433 >>12615439 >>12615453

>>12615332
Use MIT and Berkeley resources on ODE’s and FEM methods. You’ll find a lot of resources for FEM in particular in stuff related to mechE.
Don’t be intimidated anon - you should be able to pick this up quickly. You should have probably done more math in CS (proofs, applications of generating functions, etc.) but engineering diffEq isn’t hard to get into if you have up to vector calculus.

Anonymous

Anonymous Fri 22 Jan 2021 21:38:54 No.12615433 Report

Quoted By: >>12615447

>>12615426
this is a numerics problem solving a diff eq. numerics is a minefield

Anonymous

Anonymous Fri 22 Jan 2021 21:39:52 No.12615439 Report

Quoted By: >>12615453

>>12615426
Oh, I assumed you also have linear algebra down your belt. That’s arguably the most important experience you can have for scientific numerical methods next to calculus. Proof based is better, but elementary linear algebra up to eigenvectors, diagonalization, and *maybe* Jordan form will do in a pinch

Anonymous

Anonymous Fri 22 Jan 2021 21:41:40 No.12615447 Report

Quoted By:

>>12615433
Yeah you’re right. he could be using anything from Fourier split step methods to something more sophisticated

Anonymous

View Same Google iqdb SauceNAO fig3rir.png, 55KiB, 382x264

Anonymous Fri 22 Jan 2021 21:43:37 No.12615453 Report

Quoted By:

>>12615413
>acoustics and navier-stokes, hoooooly shit you're fucked
Well, at least if I fail I have an excuse.

>>12615426
>>12615439
I've done all the discrete maths/proof based stuff. I never took a "linear algebra" class, but I did learn up through diagonalization, so I think my Uni just has a retarded naming scheme for some classes.

Anonymous

Anonymous Fri 22 Jan 2021 21:48:28 No.12615478 Report

Quoted By: >>12617719

>>12615332
Try looking into ANSYS. You will have to learn how to draw out the CAD for your project and how to implement the tools for FEA but at least with ANSYS you won't have to code it in yourself, especially the fluid mechanics which is not trivial.

Anonymous

Anonymous Fri 22 Jan 2021 21:59:45 No.12615511 Report

Quoted By:

Somewhat fucked. The solutions is that the domain is divided to simple geometrics, for example n-simplexes or n-intervals. The partial diff eq. should be mathematically formed to be pushed thru this. You turn basically it to a matrix.

Anonymous

Anonymous Fri 22 Jan 2021 22:10:13 No.12615545 Report

Quoted By:

>>12615332
dude tiny triangles lmao

Anonymous

Anonymous Fri 22 Jan 2021 22:20:47 No.12615582 Report

Quoted By: >>12615608 >>12615676

Ok. You're on calc1 and hs physics. Probably no, or very little linalg and no diff eq.
How much time do you have? What is your workload and how much time you can spend on working through the theory behind it?

Anonymous

Anonymous Fri 22 Jan 2021 22:28:47 No.12615608 Report

Quoted By: >>12615628 >>12618412 >>12619766

>>12615582
3-6 weeks. I can essentially just grind for 8 hrs a day (my current plan).

Anonymous

Anonymous Fri 22 Jan 2021 22:33:13 No.12615622 Report

Quoted By:

>>12615332
use FEniCS and you will be saved. FEniCS makes it easy

Anonymous

Anonymous Fri 22 Jan 2021 22:35:23 No.12615628 Report

Quoted By: >>12615673

>>12615608
>3-6 weeks
Well, fuck.
Find someone you can pay to help you. You will get stuck by yourself. Maybe if you study it hard enough you will be at least able to understand it enough to bullshit your way through it.

Anonymous

View Same Google iqdb SauceNAO fig2rir.png, 5KiB, 285x74

Anonymous Fri 22 Jan 2021 22:49:50 No.12615673 Report

Quoted By: >>12615720 >>12622737

>>12615628
One of the capstone project proposers is an ex-physics/math professor, so in theory I could just ask him to hold my hand throughout the entire project.

He suggested I start by trying to model a simple cylinder, and he gave me formulas for cylindrical surface coordinates, unit vectors normal at each coordinate, and an equation for the area of each finite element, area = dz * dtheta * rho. None of that seems particularly useful though, at least based on what I'm reading.

Anonymous

Anonymous Fri 22 Jan 2021 22:50:31 No.12615676 Report

Quoted By: >>12615720

>>12615582
Are you telling me comp sci tards doesn't know linear algebra???!!?!

Anonymous

Anonymous Fri 22 Jan 2021 22:59:50 No.12615720 Report

Quoted By: >>12615731 >>12615760

>>12615673
>One of the capstone project proposers is an ex-physics/math professor
That's even worse. He's probably assuming you have the groundwork of a physics student, especially since you've picked such a project. Well, you can try talking with him and see how it goes, but I'd rather try to change the topic of the project.
>>12615676
I'm telling you this one doesn't. It's not just "cram one thing and it will somehow work out", it's "learn two things, one depending on another in order to understand the third one". It might not go so well.

Anonymous

Anonymous Fri 22 Jan 2021 23:04:32 No.12615731 Report

Quoted By:

>>12615720
Does he even need to do the FEM solver by himself? He could just use existent solver and make a program to call that. I think that was the task.

Anonymous

Anonymous Fri 22 Jan 2021 23:11:23 No.12615760 Report

Quoted By:

>>12615720
>thinks you have the background of a physics student
Probably not, since it wasn't a requirement for the project that they proposed. I'm actually the only person in our group that has taken ANY physics at all, so if I can't crack it we're really fucked.

Anonymous

Anonymous Sat 23 Jan 2021 00:10:30 No.12615933 Report

Quoted By:

>>12615332
>finite element analysis
I'm assuming software will be able to do most of the grunt work for you, if not then RIP.

Anonymous

Anonymous Sat 23 Jan 2021 05:45:37 No.12616784 Report

Quoted By:

>>12615332
When solving differential equations, just remember that you need to shrink the step size by an order of magnitude per call in the recursive differentiation stack

Anonymous

Anonymous Sat 23 Jan 2021 05:50:02 No.12616796 Report

Quoted By: >>12617545

>>12615332
cutting edge on fluid dynamics is training a neural network to do it. I recommend seeing if a NN is available open source somewhere and train it for desired fluid behavior and then see how your glider does. Use datapoints NN uses to generate effect for your analysis.

Anonymous

Anonymous Sat 23 Jan 2021 11:24:05 No.12617527 Report

Quoted By: >>12617534

Why did everyone just immediately assume he needs to derive the whole model and solve it all by himself from scratch? What does "effects of icean waves on autonomous ocean glider" even mean in the first place? The structural loads from water pressure? The dynamic stability? Something else? Is there really a need to involve FEM at all, maybe a lumped model will do?
Won't it be enough to simply throw a 3D glider model into a fluid dynamics FEM program and slam waves into it? Or even just look up the peak wave pressures and distributions from the design guidelines shipbuilders use (I believe thise thould be fairly standardized and available to public) and apply those and see what happens?

Anonymous

Anonymous Sat 23 Jan 2021 11:29:08 No.12617534 Report

Quoted By:

>>12617527
This.

So many tards here talking about doing derivations... take an existing model and tweak it to suit the purpose.

Anonymous

Anonymous Sat 23 Jan 2021 11:36:38 No.12617545 Report

Quoted By:

>>12616796
is cutting edge on everything is training a neural network to do it now?

Anonymous

Anonymous Sat 23 Jan 2021 11:51:09 No.12617577 Report

Quoted By:

>>12615332
This is a multi disciplinary project and you should be working with specialists. Just because you know computers doesnt mean you know physics and definitely dont have a reason to know the specialized physics of hydrodynamics/aerodynamics.
This is why everyone thinks all college professors should be executed.

Anonymous

Anonymous Sat 23 Jan 2021 12:51:55 No.12617719 Report

Quoted By:

>>12615332
I dunno how in depth you have to go with the theory, but you can use shit like ANSYS like >>12615478 said. There’s a student version for free but it’s limited in model size. Look up simulation tutorials on Youtube.

FEM has loads of complicated theory that filters people. In reality you only need to know how to set up and run the simulation, as opposed to how it actually works on the inside.

>t. studied mech

Anonymous

Anonymous Sat 23 Jan 2021 13:01:39 No.12617740 Report

Quoted By: >>12618051

>>12615332
>asked to use "finite element analysis" to model the effects of ocean waves on an autonomous ocean glider
Find the asshole that asked you to do this and slap him. This isn't something you ask a CS guy with no understanding of computational mechanics to do.

>t. mech engineer specializing in CFD

Anonymous

Anonymous Sat 23 Jan 2021 14:59:42 No.12618051 Report

Quoted By: >>12618193

>>12617740
I’ve seen CS guys in computational mechanics and comp. physics. We had like 2 PI’s doing comp. physics research in the high performance computing labs.
It’s definitely not off the table, but OP should have spent a lot more time looking into it while his PI should have given him way more resources and direction to get it done

Anonymous

Anonymous Sat 23 Jan 2021 15:40:17 No.12618193 Report

Quoted By: >>12618206

>>12618051
Well it's obviously not off the table. CS guys are actually pretty valuable for this kind of stuff because very few Mech/Aero engineers understand how to write efficient code. But at the very least, they need a year or two to make their way through the mathematics of it all.

Anonymous

Anonymous Sat 23 Jan 2021 15:42:20 No.12618198 Report

Quoted By:

>>12615332
Computers do all the hard work today, but if you can't express the problem in simple terms (a simplified version of it using equations) then yeah, it's ogre for you.

Anonymous

Anonymous Sat 23 Jan 2021 15:44:28 No.12618206 Report

Quoted By:

>>12618193
I mean it’s more than just efficient code - I’ve seen core methods research from CS people
>it takes a couple of years for the mathematics
I agree, though at the level of research, that’s what’s it takes for most people with most backgrounds. I should mention that the CS people in FEM research tend to be the people with a good math background anyway.
The analysis and algebra aren’t too bad, but it’s tricky to be clever and find new research

Anonymous

Anonymous Sat 23 Jan 2021 16:38:37 No.12618412 Report

Quoted By:

>>12615608
>>3-6 weeks. I can essentially just grind for 8 hrs a day (my current plan).
Time to give up and switch to Abaqus or some other commercial FEA program. Copy and paste their documentation into your report and call it a day.

Anonymous

Anonymous Sat 23 Jan 2021 20:28:40 No.12619329 Report

Quoted By:

>>12615332
>How fucked am i?
That depends. If you can assume the "ocean waves" can ve simplified down to a boundary condition you can apply to the FEM of the glider itself, its not that bad. If your glider is a simple geometry, that will make meshing easier. If you can use a pre existing mesher, even better. If all you have to do is actually implement a matrix solver for the FEM simulation it shouldnt be too bad. The more things in my comment you said "no" to, the more fucked you are.

Anonymous

Anonymous Sat 23 Jan 2021 22:05:51 No.12619766 Report

Quoted By: >>12621468

>>12615608
not to be negative, but that's like saying i am going to solo play a recital to an audience in 3-6 weeks on an instrument that i've never used.

Anonymous

Anonymous Sun 24 Jan 2021 07:29:23 No.12621468 Report

Quoted By:

>>12619766
pretty much this

Anonymous

Anonymous Sun 24 Jan 2021 16:55:53 No.12622737 Report

Quoted By: >>12623071

>>12615673
>One of the capstone project proposers is an ex-physics/math professor, so in theory I could just ask him to hold my hand throughout the entire project.
>He suggested I start by trying to model a simple cylinder, and he gave me formulas for cylindrical surface coordinates, unit vectors normal at each coordinate, and an equation for the area of each finite element, area = dz * dtheta * rho. None of that seems particularly useful though, at least based on what I'm reading.

t.engineering professor here. Please speak to your prof some more and ask for guidance. If he had to explain cylindrical coords to you (which is babby tier), he has to know FEM is way over your ability.

Also, have you spoken with your fellow students? They are probably as overwhelmed as you. I tell my students never to feel ashamed to admit they don't know something. I can't even imagine why he is trying to get a CS class to do something even physics people have a hard time with.

Anonymous

Anonymous Sun 24 Jan 2021 17:52:09 No.12622955 Report

Quoted By:

>>12615413
acoustics ain't that hard ... depending on what exactly you try to simulate. NSE on the other will always rape you, true.

Anonymous

Anonymous Sun 24 Jan 2021 18:24:08 No.12623071 Report

Quoted By: >>12623233

>>12622737
>t.engineering professor here.
> (which is babby tier)
This amuses me for some reason

Anonymous

Anonymous Sun 24 Jan 2021 18:54:49 No.12623233 Report

Quoted By:

>>12623071
I mean, it is babby shit. I learned it in my first or second uni math class. Im not sure why that poster (and the project partner) think I havent learned about it, but whatever.

I agree that it doesn't make sense that they think I dont know pretty simple calculus, but also think I'll be capable of doing some super technical physics computing. Maybe he means something else by finite element analysis. Its totally possible that my project partner is retarded too I guess.

Anonymous

Anonymous Sun 24 Jan 2021 19:05:04 No.12623276 Report

Quoted By:

>>12615332
This seems like a lot to handle if you specifically don't have any kind of applied mechanics or physics background, are you expected to use any commercial software or some pre-written code?
Building an entire finite element solver from the ground up is no joke even if you had the theoretical background. The problem you've mentioned is fluid-structure interaction problem which would require a decent amount of leg-work to get to speed...

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches