What does the FEA process look like at your company? 9

[FEAChamp]

Can anyone chime in with what procedures they go through for an FEA study? I'm supposed to be picking up more responsibility as the "Lead Analyst", and I feel like my view of how an FEA project should go is different than my bosses.

For example lets say we're analyzing a shelf that mounts to the frame rails of a truck with clamps, one assembly, 15 parts. It has quite a few contacts and bolted connections, so when all setup it might take 16 hours to solve.

Do the analysts work independently for the most part, and report back when they have results? Or does the whole team (project lead, FEA lead, management) review every setup before and after it is run? Like sit down and discuss if its the correct way to load, review all the contact sets in detail, review all the bolts, etc? Last week one of my coworkers got the static study ran for a shelf and the setup was reviewed by the owner who has experience in FEA. We reviewed the results on Friday, and determined that it was time to do the fatigue loading studies. They should have the same setup except for loads, so the study could just be copied out and modified. They were setup to run over the weekend. Well something went wrong and the results weren't available until Tuesday. It was then we discovered two sets of contact sets were incorrect and the whole study needs to be re-run, including the static load case. Now my coworker and I are getting flak for not reviewing the setup with the whole team, despite the fact that he reviewed that setup with the owner, and the three people needed to review setup (myself included) were out of the office Friday afternoon. If I'd had him wait until Monday, we would have gotten flak about not taking advantage of the weekend.

They keep wanting the FEA dept to get better at scheduling, and hitting deadlines, but how does that work when studies take 16 hours to solve. It always sounds like they expect us to stay overnight, or whatever to make sure it solves and/or the results are OK. They always like to say that the FEA dept has no urgency and often has a "well it didn't work today, lets try this change, cross our fingers and see if it solves overnight". But to me, that's a big part of the FEA process. It takes time. Staying late doesn't help, outside of instances where its going to finish solving soon, or finishing the setup to run overnight. If we stayed, we'd have to find busy work because technically we're not supposed to charge machine time.

I just feel like FEA is naturally a longer process, and they want to compact it as much as humanly possible. "That study only takes 4 hours to run, well then we should have results for the customer tomorrow". Sometimes thats true, we can get results pretty quick, maybe the next day. But often something happens and pushes the results back and its "just another instance of FEA not having urgency or sticking to a schedule."

So, FEA analysts of Eng-Tips, if you were handed an assembly similar to the one above, what would your general schedule and process look like? Additionally, how do you get your load cases and setups? Usually for us the project lead (young engineer) just knows they need FEA run, they hand it over to us, and then its up to FEA dept to develop all the load cases, design limits, fatigue parameters, etc.

 

[GregLocock]

I don't work in FEA, gave that up 35 years ago, but I do work with MBD sims. Much of my time is occupied with correlating with real world tests, and the unfortunate fact is that a given correlation that normally takes only 5 days sometimes stretches out to one or even, in one case, 3 months. In the latter case it was a safety related signoff, so they just had to wait.

It seems to me your schedules don't allow for multiple runs and screwups. Both are inevitable. If they want faster results get a Linux cluster. You don't mention how you validate your models.


Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[IRstuff]

I would check to see if you're meshing, etc., correctly/efficiently, since a 16-hr turn around does seem awfully slow.

So, if the turnaround time cannot be changed, then each simulation is precious time, so a thorough review needs to be done to ensure not wasting time. But, it would seem to me that some sort of procedural thing is awry, since the notion of needing 3,4, or 5 people looking at a model in detail likewise seems absurd, on the face of it. I assume that the review is going to wind up taking several hours?

If the turnaround time were 2 hrs, would you still want to do a thorough model review? I think not.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[GregLocock]

My background is make it test it break it fix it. As such quick turnaround is essential.

Oh sorry, to answer the OP, specifically for FEA, ie durability, crash, noise, vibration, and probably three other things i've forgotten about.

The process is - CAD comes up with some pretty shapes meeting Design's requirements. Core FEA then turn that into 3 resolutions of mesh. Various departments run those for their tests. Problems are found, CAD draw up more pretty shapes. Rinse and repeat twice. These get signed off. Parts are built, cars are made and tested, more problems are found, and failures to perform in line with the model predictions are found. Hopefully one more cycle of design sorts it all out. All of this is done inside the overall V framework for developing systems, cascade requirements down the left hand side, build up tests (verification) on the RHS. Loop as low down as you can.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[SWComposites]

Why do you need to run an FEA of a shelf??

 

[cohoman]

I've been working as an engineer for 33 years now, the majority of the time with FEA. I've worked at 7 different companies, big and small. Here's my thoughts on this subject:

Before you start on a project, think hard if you can do the analysis using hand calcs. Often times younger engineers tend to jump into using FEA tools because it is flashier and more fun than using traditional classical methods. I've also seen engineers wanting to use FEA because they don't want to bother with understanding fundamental load paths through their structure, etc. They assume the FEA solver will handle all of this for them. I've analyzed equipment shelves on aircraft racks and 80% of our analysis is using hand calcs with free body diagrams, section analysis, etc.

If you decide you need to use FEA for a project, here's the process I would recommend:
[ol 1]
[li]Work with your designer on understanding the geometry of the structure. How the various parts are connected together (with rivets, bolts, etc.), how the assembly is connected to your primary structure. What the section properties are of the structure being analyzed.[/li]
[li]Confirm with your lead engineer or engineering manager that using FEA is appropriate for this analysis.[/li]
[li]Begin making the FEM using the correct element types (shells, bar/beams, solids, connectors, springs) while also noting all assumptions you've made for the modeling. Don't forget to include any relevant stiffness with the structure you are mounting your assembly to.[/li]
[li]Once completed, run some basic model checks (linear statics run verifying applied load = reaction load, parts are connected correctly, deflections seem reasonable). You should also try to validate your model by comparing hand calc results with FEA results where possible.[/li]
[li]Have a review with your designer, peers, and/or "expert" FEA person in your group. Review basic modeling strategy, how the joints are modeled (which degrees of freedom are constrained and free), applied loads, initial test results. Make sure everyone is in agreement to how you're simulating the actual structure with your FEM.[/li]
[li]Decide which load cases are critical for your analysis. You probably don't need to run all potential load cases, just the critical cases.[/li]
[li]Once you've had your review and feel confident that your model is running properly, make your runs.[/li]
[li]Have another meeting with your peers to review your results to make sure they make sense. It's good to have a fresh pair of eyes looking over your work and results.[/li]
[/ol]

These are the basic steps I take when working on a FEA project, and I often suggest the same steps to the novice FEA users in my group. Hope this helps.

 

[L_K]

One practical tip, which was already mentioned is to start with fully linear model. Make
sure the model is working, reaction loads etc.

Then move to non-linear analysis but try to save elements. Make sure the model is working,
for example in ANSYS you can visualize all the contacts at same time. 16h turnaround time
might be fine for the final analysis but not for troubleshooting and fixing errors. HPC might help.

 

[JamesCrist]

I work for a software sales company so I don't do much "real" analysis but I do spend a lot of time troubleshooting customer models and helping them with workflow so I'll offer some suggestions (in no particular order):

1) Make sure you understand what actually needs to be modeled. I see people throw everything under the sun into their model when the could get away with a much smaller scope and some proper boundary conditions.
2) Make sure you're using the right tool for the job. There are tons of "good enough" designer-oriented products that could drastically reduce your turnaround time.
3) Set up debug models. This varies with the type of analysis (fea/mbd/cfd) but starting off with a crappy mesh and rough boundary conditions will help you spot a bad boundary condition before you waste 2 days on an analysis. One thing I like to do is overload forces or something similar to force an exaggerated response, makes spotting errors easier.

Senior Application Engineer
enginsoftusa.com

 

[Vitkacy1989]

I once worked in a company which was better in creation of quality processes than in FEA (say, Company A) and once in another company which was better in FEA than quality of the process (name it Company B). This is how the process looked like:

COMPANY A
1. Clarify with Client any technical doubts
2. Create four-page summary explaining in general how FEA will look like (peer to check it)
3. Create FE model (peer to check it)
4. Apply loads, constraints and other important settings (peer to check it)
5. Debug model using simplified features (e.g. all contacts bonded)
6. Run model overnight
7. Interpret results (peer to check it)

COMPANY B
1. Create final model and run overnight
2. Debug problems found next day
3. Interpret results (peer to check it)

I would vote for the first process. The essential was this short summary describing what FE-Engineer intends to do. That was great and clear reference, worth 3h to be created so all people participating in decision could give it a green light or stop it. As FEA gets more popular, it is crucial to use quality system along all the process. But, if your top management is not aware of it, then I think a big success in FEA career should not be expected in this particular company. It's simple as that, either everyone understands role of the FEA or we have got a typical miscommunication.

 

[IceBreakerSours]

Good comments have already been offered. Just one additional thought: I would advise involving an FE expert outside of your immediate group, especially one who uses a different set of tools for their analysis needs. Diversity of perspective helps tremendously.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[Doodler3D]

This was a helpful post. I'm currently stuck on the same boat. I had CFD experience but have been recently asked to do FEA. At this point of time i'm trying to learn hand calcs with design codes along with enrolling in basic NAFEMS courses. FEA is miles apart from CFD - and it does not help that I studied mechanics of materials 7-8 years ago.

 

[FEAChamp]

Thanks everyone for the input, there's some good food for thought. I guess my concern is more about the actual work load and how you handle solving times and if you fill that time in with other work or how you balance multiple projects at once. Or even how many projects would you work on at the same time?
We're a very small company. Most of the FEA we do is quoted as a specific FEA project. My boss (who did FEA but hasn't really done it in 5-6 years and is pretty knowledgeable about setup and all that, but his post processing experience/skill is basically set the scale to yield and remove all red elements by changing the geometry) creates an estimate, the customer approves it, he hands the project off to me and says I have 15/30/60/whatever hours to do the analysis. Its not unusual at this point for the customer to think we've already been working on the project for a week or two before this since the estimate review was a few weeks before and they signed the estimate then. Then its up to me to have the kickoff meeting, determine any hard due dates, talk with the customer to get a feel for what they really want, put the project plan together, and be the project manager/lead/contact point and update the customer on hours used, hold weekly meetings and also (generally) do the analysis or work with someone under me to do the analysis.

For short and to the point projects/studies, its fine. A simple steel structure with a handful of loads, sure I can put a plan together and all that and it doesn't require much babysitting. But then we get projects with tons of moving parts (plan wise) or where we've never really done that specific type of analysis before and things get messy. In my bosses view, I would never just be waiting for something to solve. Doing a new type of study and waiting to see how it solves, or how fast it solves, or what effect this setting has on the convergence? We don't charge for solve time, so cram another project in there to keep generating billable hours. Then I end up with 3-5 projects and feeling like I can't really focus, especially if I'm having to hold weekly update meetings, having to update project plans, dole out work unreleated to FEA on non-FEA projects I've been assigned lead on, record all customer questions, requests, and info in the punch lists, for all of my projects. Not to mention proofreading and redlining FEA reports, or other tasks like that.

It feels like they think you should be working (updating plans, doling out work or reviewing models) every minute of every day, and that burns me out. Maybe its just me, and I'm not a great worker. Maybe its just this company. I feel like my ideal setup would be that I work with an actual project lead who can manage customer expectations and set deadlines with them and who tells me what kind of analysis or information they want, and I'm not working directly with customers, except for maybe when I'm needed to explain results or brought in for brainstorming, etc. I'd work with the lead and get deadlines like "we need this information or study done by Friday", and that's essentially my main project I'm working on. Maybe I have another task or two to fill large gaps of time, but it wouldn't be 100% hours driven. Obviously if i'm in the middle of a solve that would take 18 hours, or 3 days then yeah, I would try and fill that time with other productive work. But waiting 45 minutes to see what effect this boundary condition has on a draft mesh study, or an hour to see how this study converges with X setting changed wouldn't be abnormal or seen as "wasting customer's money". But they're also quick to use hours worked as a metric for how much effort is put in. If a study we're having trouble with getting to pass is taking 10 hours to solve, and I spent 15 hours on it through a week because I spent 3 hours each day looking at the errors/results from the previous day's study, researching solutions and testing solutions on a draft quality mesh then setting up and running the next run, they'll bring up that I worked way less than 40 hours on this and this project is hot, so overtime should have occurred. But that involves staying late and waiting for the computer to finish solving. But we aren't supposed to charge for solve time, so it really means that I need to find another project to work on while I'm staying late to get the other study to solve because the way the company is setup they rely on every single of the 40 billable hours every week. In some ways it feels like I'm a mechanic who has to fill the time he's waiting for the fluids to drain out with another task just to stay busy.

I think my point I'm trying to feel out is are there places out there that do FEA work that operate like I think a job like this should be, or am I just wrong in my thinking and expectations on how things like this should go? I hope this makes sense. I'm trying to make sense of it all in my mind at the same time.

 

[STrctPono]

I spent 3 hours each day looking at the errors/results from the previous day's study, researching solutions and testing solutions on a draft quality mesh then setting up and running the next run

so it really means that I need to find another project to work on while I'm staying late to get the other study to solve

This is not a day in and day out process for me like it sounds like it is for you but I dislike when I have analysis like this. Switching modes while doing technical work and trying to change mental gears is never easy for some individuals. For me, productivity on the project is reduced when I have to find filler tasks to do between FEA run times.

 

[FEAChamp]

Thanks for the input, knowing that others really do have issues like that is reassuring. One of the other things that annoys me is I try to bring up some of these issues and immediately get dismissed, like its totally unreasonable to have trouble switching gears like that. Or if I say I'm not really great at an aspect of my work or project management (read: don't enjoy it and struggle with it) they tell me "well you're a smart guy, you can figure it out eventually."

 

[FEAChamp]

I'm back with more thoughts/questions.

How often do you have to re-run a study because (for instance) you measured a distance on the original model and not the prepped model, so a part being displaced isn't traveling as far as it should? Or you've gotten a new model with a setup that's pretty well understood, and you forgot to turn on that one option in the mesh settings after you imported the new part? Or many other reasons that you might need to re-run a study?

For my original post, I agree that 16hrs was a long study time, but it was in solidworks, and just had a ton of contacts to solve, and in that case a mistake is pretty costly in terms of time. But now I'm working on an explicit analysis in Abaqus that takes 20min to a few hours to solve, and the powers that be are really trying to make the point that they believe practically every run should be set up perfectly, and get all melodramatic about stuff needing to be adjusted based on the results.

They also want to review practically every study before it is run, option by option, setting by setting. Again, one of the guys (owner) used to to linear static FEA until he started doing mainly estimates, meeting with potential customers, etc. The other guy knows little about FEA, but is a mechanical engineer, and is basically trying to get me and the analysts to teach him how dynamic FEA works as we go. Its just super frustrating to be questioned on everything (and not from someone with higher knowledge, especially since I'm basically self taught in dynamic FEA), or told that a study is now pretty much worthless because I used a mass scaling target time increment of 2e-6 instead of 1e-6, or because the perscribed displacement was off and I might as well have thrown away the customers money (they're exaggerating a little bit on that, but not much), or whatever their idea of what *should* have happened is.

I sort of feel like i'm just straight complaining about work in general. I'd love to see if anyone else has any more insight into what happens at their company, or just insight in general.


A more general question, as an engineer how often do you deal with being the main point of contact with a customer. Updating them on how much money is left in the estimate, dealing with determining/negotiating due dates, etc?

 

[GregLocock]

Runs fail, mistakes are made, iterations are sometimes needed, some parameters are a judgement call or don't actually matter, those are all facts of life.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[IceBreakerSours]

You will run into all sorts of people with varied backgrounds. Some immediately get what you are doing, its importance, why you made the decisions you made, etc., some are clueless or couldn't care less, and others tell you on your face - "I do not trust computer models". You have to pick hills you are willing to die on Furthermore, modelers aren't superhuman either; they have their own beliefs. Some marry the tools they picked up in grad school or their first analyst job and actually start believing in the crap sales folks tell them over and over again.

Once you find your hill, then you had better be armed with irrefutable evidence.

Also, I would suggest keeping an open mind. It is difficult to do but it has helped me become better. The actual wording/phrasing from someone who doesn't know much about modeling may not be important but the underlying sentiment might be worth thinking about and, at an appropriate time/setting, addressing.

Finally, I think you might also benefit by talking to some like-minded folks. I would consider joining some organizations like the ASME Verification and Validation committee, NAFEMS, etc. where people get into the weeds of modeling. Getting a certification from NAFEMS wouldn't hurt your career either; a glossy certificate or two will also help give you credibility and hopefully shut down some of the ignorance. Perhaps, you also ought to seek out a mentor in those organizations or, these days, in LinkedIn modeling groups.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[Sparweb]

But...
it's a shelf!

we're analyzing a shelf that mounts to the frame rails of a truck with clamps, one assembly, 15 parts.

Have you heard of Roark?
I read this thread with a growing sense of impatience since I would not usually consider analyzing a structure like the one you describe with FEA. But that's a knee-jerk opinion. I don't actually know what this shelf looks like or what's loading it. Maybe it would be an obvious FEA problem once I saw it. But in all your discussion you haven't mentioned any traditional calculations. I find these essential even when doing FEA to establish boundaries and get the preliminary sizing close enough not to waste a run of the simulation. The tools available for traditional calculations can work very quickly and provide surprisingly exact answers for many common structural problems that LEND THEMSELVES to simplified manufacturing and lower costs of production. When I need that last 10% improvement, I reach for FEA.

There are a lot of analytical tools that are more efficient for many problems than FEA. They also cost much less. And I'm not talking about loose-leaf paper in a 3-ring binder like you had in high school. If you want alternatives, just put your hand up!

Some customers, bless their hearts and pocketbooks, are only going to pay if you give them the pretty graphics in the power-point. You can't do much about them, but the other thing you seem to be looking for are effective tools to manage schedule crunches caused by corrections and re-runs. Well sometimes you can get the right answer from one method fast, and the pretty illustration from the other method without needing it to be right (the other method gave you that).

A more general question, as an engineer how often do you deal with being the main point of contact with a customer. Updating them on how much money is left in the estimate, dealing with determining/negotiating due dates, etc?

Here, too, having direct analytical tools at your disposal can help you. Providing quicker estimates without having to build a model and run a solution (1/4 to 1/2 the time) sets the customer expectations more realistically, earlier.
As often as costs, they are concerned with delivery dates. I try to be as transparent as possible, but I have seen some project managers put in awkward positions where they had to choose between lying to the customer or losing their job. Glad I don't work for folks like that. Explaining delays, forecasting new delivery dates while the fix isn't done yet, expanding or shrinking scope can all be difficult to explain to your customer. If you're doing this without adequate support from your superiors, you can feel like you're the scapegoat. It is possible that making a simple, specific request to your boss of what you need at the time you need it will get you the support you require.

http://www.sparweb.ca

 

[human909]

I've been puzzled by this thread from the very beginning... But I have been operating on the assumption that I do the simple and easy stuff and the problem being discussed isn't simple and easy. (A shelf sounds simple and easy, but I'm sure if I was designing one for a fighter jet then maybe things would be different...)

Normally I try for FEA solutions that are going to take <10mins to solve unless I am doing a 'final run'. Use every tool in your book to get simplify your model into bite size pieces. It is an awfully complex model if you cannot reduce 15 bolted connections down to something closer to 3. Using symmetry is pretty standard so I'd hope you are doing that. But likewise making reasonably assumptions about constraints at certain slices and you can slice and dice you model until you are analyzing just a few pieces...

But like I said, it depends on how complex your design field is.

 

[swimfar]

I usually have between three to five different projects that I'm actively involved in at any one time. I agree, that it would be easier to just have one project. But with the work that I do, that's not a reasonable expectation. The advantage is that I get to work on lots of different projects, which keeps things interesting continually brings new challenges.

If your simulations take a long time to run, then ideally you have other things to do during that time. If that's your only project, I see how it could be difficult. Can you start working on the introduction part of the report or briefing? Can you start setting up the next simulation so you're ready to hit go as soon as the first one is done? What are you doing while the simulation is running? If you really have nothing else to do, I guess you could triple-check the model setup.

We have at least two engineers assigned to each project. The project manager deals with most of the customer relations, contract setup, and budget. They also usually have a technical role, and may be the main active engineer on small projects. For many small projects, the second engineer just helps to review all the work that the project manager is doing. In this case they will often be more senior than the project manager. For larger projects, the hierarchy is flipped around.

Regarding setup mistakes. They happen. So the question isn't, "is it okay to make mistakes?". It's "how much do my mistakes cost?" and "Is that acceptable?" If you have a multi-day simulation, then it's worth it to spend a little extra time reviewing the model before starting. If your simulation only takes 10 minutes, mistakes are more acceptable. But, if I'm making the same mistakes over and over, that's not good, regardless of the cost. I know it sounds cliche, but think about how you can learn from your mistakes, and keep yourself from doing them again.

Most of our simulations take from 1 hour to 1 month on an HPC, with 12-24h runs probably being the average. If we're running a brand new model, two engineers will always review the model setup together, often before we even start to build the model. What are our boundary conditions? How are we going to apply the loads? How are we going to handle contact? What should the mesh look like? What can be simplified as a rigid cube, and what needs more geometric accuracy? How are we going to model/analyze bolted or welded connections? What material models are we going to use? If you feel that you're forgetting little things often, a good check-list can catch a lot of mistakes. Also, if you have a long run, don't wait until it's finished to check and see if it ran correctly. As soon as stresses start building up, look at the results to see if they seem reasonable. Is the model behaving as expected? Do your reaction forces match the load that you (thought you) applied? Check all contact locations for penetration.

Maybe I'm reading it wrong, but it sounds like you're simultaneously complaining that your colleagues want to do a more thorough review of your models than you think is necessary, and also saying they are unreasonable in being upset when a simulation needs to be rerun. What is the reason for not letting them spend an hour to review the model with you? Would that have caught some of these mistakes? It sounds like you're still learning (we all are), so it can be good to have more eyes on the problem, especially if you're self-taught. It's good to have someone question and make you back up the assumptions that are built into your model. Non-linear, dynamic analyses are tricky. Mass-scaling is a good example of something that I see abused all the time, even by experienced people who should know better. It's really easy to make pretty pictures that don't match reality.

I also often wonder how things are done in other companies. You're asking the right questions. You obviously want to improve, and there's a lot of good feedback in this thread.