[CFD] Residuals in CFD (Part 4) - Global Imbalance

[@jk463]

This series about residuals deserves every existing compliment. I've never seen any other lecture(not even in my univ) in this level in terms of not only context but also in visual/aural quality (although I know that I will still keep struggling dealing with my residuals, even after listening to this masterpiece... tough job to do). This lecture should be designated as an official introductory lecture for all CFD-newbies at schools and CFD software manuals.

[@fluidmechanics101]

Thank you so much for your kind words, it means a lot 😊

[@mrfatah4510]

Thanks for bringing these fantastic lectures series to the small world of CFD (especially beginner) users!

[@EclecticVibe]

Always happy to see FM101 video pop up! Thank you for your effort 👍

[@xuxu119]

thank you!!!!!!!!! your lectures are more useful than the ansys manual.

[@zettaker2181]

Greetings from Russia! Thank you so much for such clear lectures. Love watch every of them and advise my friends at university/colleagues at work. It's pleasure that now everyone can get simple explanation on CFD topics just on KZbin. Good luck with your work! 🙂

[@sergniko]

This was an amazing journey to residuals world! Very useful and helpful! Than you very much!

[@rahulchopde2289]

Thank you Dr. Aidan for this wonderful lecture series. Your lucid explaination of these complex topics will help us in perfecting our CFD simulations.

[@dmitriykogan8879]

Many thanks for this excellent series of videos on residues.

[@rafaortwein6646]

Excellent talk, many thanks for the lecture.

[@alekseisorokin4723]

Good job, Aidan! Thanks. It's a common advice when you have a convergence problem during the calculation try to simplify the model - exclude some equations, supersonic flow, polynomials in gas properties etc - and add this features consequently. Of course, it will not help completely if you have bad mesh cells. Good luck!

[@botonb4eg1]

Thanks a lot for these series of lectures! Appreciate a lot!

[@Mercredification]

This lecture series is pure gold. Just the right mix of practical tips and mathematical background to understand the different concepts. I wonder if there exists a "solver specific" database with threshold recommendations for local and global imbalances in general and maybe for certain applications to provide some guidelines towards more credible simulation results or if they are too much simulation type or intent dependent in general. Anyways, really great work, appreciate it a lot!

[@taroccomatheus]

Most didatic lectures on the subject i had find.

[@wagdimahroussaieed]

i am was struggling with the assessing the solution convergence but after i watched your amazing videos i fount it quite easy process. thank you so much

[@denniii09]

This series was insanely interesting. Although I already knew to not only care about residuals (RMS or MAX) but also about Imbalances, the theory behind of it all was making it even more clear for me. Thank you for these detailed and informative videos! Really excited for the next video...any hints, what topic you'll cover next ? :D

[@fluidmechanics101]

... maybe relaxation

[@mdshupto183]

you are the best teacher. Love you Dr. Aiden

[@kimiyakasaei1661]

Many thanks for these set of helpful lectures!

[@virupakshatm3339]

Very informative series. Thank you!

[@hungnguyenthanh8833]

Your lectures were very fine! Although I am not able to understand alls. But I am also capable of improving my analyses in my CFD problems thanks to your lectures. Thank you very much and good luck for your channel!

[@nishantbhatta9081]

Thank you so much for this residual series Dr. Wimhurst. I would be really grateful if you can explain about Multiphase flows especially VOF methods in some lecture series.

[@turalsuleymanov4529]

Thank you very very much THat was amazing) I have watched all of the series and they are so useful.

[@juliocesarortiz1770]

Thank you so much for your effort. Your videos are very useful. I will share your lectures in office :)

[@akshayghorpade8624]

Great Lecture.

[@abdelhak.keddouda]

Great series of talks as always, thank you Dr. Aidan, I just have a question about if there is any plans for videos on how to account for Nanofluids in the Navier Stokes and Energy equations, thanks again for your efforts.

[@fluidmechanics101]

I haven't considered nano fluids before but I imagine they are more difficult because you are at a scale where the continuum assumption starts to break down

[@sarojmandal981]

Great video. Load of information for me doing Phd in mechanical engineering. Also, as a suggestion to learn more, can you also instruct us how to reduce residuals using ansys fluent, and what are the magnitudes when round-off error also propagates in the solution. A demonstration would be very helpful.

[@fluidmechanics101]

Good suggestions. Thanks!

[@ilhantalih9949]

Thanks sir, I look forward to seeing next lectures.

[@markonabiy6632]

Very helpful, thanks

[@pietromaccari1968]

Great! Thanks for the video

[@hebaalaaeldin2596]

Thank you so much for this great effort!

[@Rau379]

Awesome Content!

[@debayanmazumdar7615]

nice and helpful video. quite detailed god bless you !!!

[@omerkorcanates2150]

Thank you very much, it was very usefull

[@mohamedshaaban2920]

thank you

[@ЛевЯрков-е1ж]

Thank you!

[@Asim408]

Hi. I have one request. Kindly can you make a video about initialization and statistical data analysis in transient simulation. Thanks!

[@denniii09]

statistical data analysis in transient sims. sounds great!!!

[@nac9494]

Great series. Thank you so much!

[@8cold8hot]

This series of videos are extremely clear and practical! May I ask for a special case about residuals: If I run a DPM case with the continuous phase starting from the stagnant state, but the injected Lagrangian particles will bring in extra momentum and interact with the continuous phase, then what is the suitable criterion to judge whether the residuals for the continuity equation of continuous phase are small enough that the case is already converged?

[@fluidmechanics101]

I would probably look at other metrics besides just the residuals. Can you think of any monitors that might be appropriate for your case?

[@8cold8hot]

@@fluidmechanics101 Thank you for the answer. I agree with you, I might look for other quantities like wall temperature...

[@riebi34]

Thank you so much for your work. It's highly appreciated. Please don't stop what you are doing 👏

[@metoosti5987]

Hello Aidan, I do have a question about residuals as well which is provided by ANSYS. I was analyzing the multi element front wing of an F1 car in ground effect and I was able to validate my results for a certain case. My residuals were quite okay (about 1e-4) but when I decreased the ground clearance of the wing, the residuals increase to the level of 1e+1 which is possibly the result of having large velocity gradients. I also get higher residuals when I increase inlet air velocity. I want to use a steady state solution to find a result but the residuals are just too high when I use it. When I use unsteady state solution, residuals actually decrease but I'm not sure if it's necessary to use an unsteady solution there since I'm just trying to find the average value of drag and lift coefficients. Do you have any ideas about how I can decrease my residuals when I use a steady state solution? I did try to refine my mesh and I couldn't see any difference at all.

[@fluidmechanics101]

It sounds like you either have 1) a problem with the mesh (when the clearance reduces the cells are getting skewed and the solver is falling over) or 2) there is some real flow unsteadiness which is trying to develop and the solver is trying to go unsteady. My advice would be to have a look at the mesh quality and flow field and see what you can see. That should point you in the right direction. Have you tried plotting the residual contours? Maybe you can find the cells with the high residuals and work out what is going on

[@lizarettflavour]

great!

[@soroushasadian9100]

Dear Dr. Wimshurst, Thank you for the series, now I have a better understanding of residual plots. Could prepare a video and explain UDF? Because many problems require user-defined functions for input Such as cardiovascular models require pulsative inlet velocity

[@fluidmechanics101]

UDFs are tricky. That is a great idea. Maybe a new series?

[@soroushasadian9100]

@@fluidmechanics101 Sure! Why Not?!

[@mohitraje5901]

I am trying to simulate high-velocity flow (using k-e model) through porous media in unsteady state. Initially, the residuals decrease for nearly 15000 iterations. Then they start to increase and create fluctuations in my output parameters. How can I tackle this issue?

[@fluidmechanics101]

Have a look at the flow field and try and work out what is happening. Maybe look at the solution in iteration 14000, 15000 and 16,000 and try and work out what is happening

[@oguzkirez3944]

Many thanks and a question. Please consider a generic case in CFX. My monitor points and global imbalances converged. However, I have rms residuals from 1e-3 to 1e-6 and they are chaotic (but do not exceed 1e-3). Would you think the solution is okay?

[@class8007]

This may or may not involve global imbalance, but if a particular residual (e.g. Velocity/momentum in u direction) does not converge satisfactorily, but others that I am more interested in for a particular simulation do converge well (e.g. heat flux), is it safe to assume the heat flux results values are as accurate as the residuals lead us to expect? Or could the imbalance in u momentum lead the heat flux to still possibly converge, but to a value not representative of the real solution? If it is case specific, is there a good way of telling how much residual convergence is affecting other residuals?

[@fluidmechanics101]

I wouldn't trust your residuals. I would look at global imbalances and surface monitors and have a closer look. Energy residual does always seem to be lower than momentum, it does not necessarily mean that the heat flux balance is better converged

[@class8007]

Thanks! It was just a hypothetical situation, I've yet to encounter such a situation but its good to know when and what to trust.

[@gksrbqls91]

What's the implication of residual stabilizing but not decreasing anymore to something like 1e-3? does it mean it's converged but accuracy is not ideal?

[@fluidmechanics101]

Yes thats the one. It is indicating that you should probably look at the global imbalances and try to work out what your overall conservation error is. Your mesh / boundary conditions / pseudo time step size could then do with some improvement. But be careful, residuals can be deceiving and don't show the whole picture!

[@gksrbqls91]

@@fluidmechanics101 Thank you! Your videos were super helpful!

[@gerhardbekker646]

What do you think about using steady-state solvers for flows that have some unsteadiness? Like the backwards-facing step you looked at? Do you think the results still provide valuable information or are they actually misleading and wrong?

[@fluidmechanics101]

Yes, the steady state solver will still provide valuable understanding (and it is often useful to do one before the transient anyway because it is a great initial condition for the transient). The steady state result will normally be pretty close to the actual result for the broad flow field. A useful comparison is to compare the steady state with the transient and actually see for yourself what the differences are. Maybe 10-20% difference would be expected but you are unlikely to get 100-200% difference

[@gerhardbekker646]

@@fluidmechanics101 Thanks for coming back to me. Your suggestions make sense. Appreciate it!

[@Jialei-dw3li]

When assessing convergence, if the variable of interest is flattened out, but the residuals are still decreasing, do I need to continue iterating until the residuals are also flattened out?

[@fluidmechanics101]

If the variable of interest is the only thing you are reporting, then you can stop. But if you are looking at lots of aspects of the case (flow field, wakes, turbulence generation) then I would keep going until the residuals are flat as well

[@Jialei-dw3li]

@@fluidmechanics101 What about the transient case? Must the residuals also be flattened out as well in each time step?

[@fluidmechanics101]

Ideally yes. But usually what we do is compare the result of interest (i.e. drag coefficient) with different number of iterations per time step. For example, compare the results with 5, 10 and 15 iterations per time step and see if there is a significant difference in drag coefficient

[@Jialei-dw3li]

@@fluidmechanics101 Really thank you for your guidance.

[@KToMmi]

Hello Dr. Aidan! Even though I'm a more experienced user, this series was a pleasure to watch. If you don't mind a suggestion, in my opinion it would be well worth it to expand the topic of convergence issues diagnosis. Maybe discuss some advanced tool to identify where the issue lies in the domain and how to solve it - for example some tricks to improve mesh quality without necessarily simplifying the underlying geometry. I often find myself struggling with inflation layers quality when trying to reach very low (

[@fluidmechanics101]

Yes, thanks for the suggestion! I think this would be very useful for many CFD users

[@pierrefpv]

How do you deal with residuals and global imbalances in a solver for unsteady compressible flow with strong wave action? Do you iterate during a given timestep to reach convergence for that timestep? Due to the compressibility individual cells can accumulate mass, energy and momentum, so how do we know if the solution is correct (converged) for that timestep or not?

[@fluidmechanics101]

Yep converge within each time step. The accumulation of mass, energy etc is accounted for by the d/dt term which is included in the unsteady calculation (and is ignored in steady state)

[@fluidmechanics101]

If you can get the residuals lows at the end of each time step, you can think of that as the correct amount of accumulation within the cell (for the given fluxes across it's boundary, which aren't equal, hence the accumulation)

[@AirChaser97]

Is the shock one of the transient flow?

[@fluidmechanics101]

You could have a steady shock (in that the shock itself stays in the same position) but most often you will need some type of transient solution to allow the shock to develop and move into the correct position

[@pawankhanal8472]

Great. hey , I really enjoyed your turbulence modeling .how hard it is to do thesis on new turbulence modeling like combining k epsilon with something to improve for boundary layer effect (this is already done but like that ) . can we modified on some existing model ? please give me suggestion I really want to do thesis on turbulence modeling but wanna be sure that can I do or not. And of course I will do lot of literature review but I really want to know your suggestions.

[@fluidmechanics101]

It depends on the application. The existing turbulence models are pretty well calibrated for the applications they were designed for and the coefficients are specified for them. You could modify the existing models for a new application (either by adding additional source terms or recalibrating the existing coefficients) but it would depend on your application. So I would choose your application first and then see how good the existing models are and whether you could improve them. This is definitely something worth discussing with an academic advisor or tutor

[@hungnguyenthanh8833]

Can I ask you question? What is the residuals in CFX that display in Solution part? Local imbalance. So how they are calculated?