Opinions on FEA for home/hobby user

[Bowber]

I use Alibre to design small parts and assemblies for motorcycles and other home projects and was wondering about using FEA to optimise designs for things like frame structures to get an idea of stress areas before actually making them, trying different shapes of gusset, tube and hole placement etc.

I've been looking at Lisa FEA but not got far with it yet, so before I commit too much time to it I wanted to know peoples opinions on FEA for small projects.

I'm quite happy doing simple hand calculations, but it's been a while since I did any, I have an OND in engineering and used to design industrial equipment for dust control (extraction systems on quarries etc).
I've also made bike frames and other brackets and fixings for years so I've got a good practical feel for thinking problems through.

I'm also well aware that FEA isn't a magic pill that'll tell me everything, just with the simple bits I've looked at it takes a lot of work and thinking about to decide on the correct load, fixings etc.

So is it worth learning to use Lisa or is it hard to use, are there other better solutions that are cheap or free.
I've also looked into FEM Designer and while it's cheap in the industrial sense it's still a lot of money, also I object to a product that's just a download and is $299 ($189) in the US being over £210 from the UK reseller! Just like they made Alibre Std £99 from $99

Or do I just stick to the old "if it breaks make it bigger" method?

Thanks
Steve

 

[bigseb]

... I object to a product that's just a download and is $299 ($189) in the US being over £210 from the UK reseller! Just like they made Alibre Std £99 from $99...

+1

Back on topic: I really need to learn how to do this as well but am struggling to get started. Been playing around with Lisa for a couple of months with getting anywhere and recently downloaded Z88 Aurora. Someone coming round tomorrow to give me some pointers.

 

[DavidJ]

Steve,

One reason for the pricing in the UK is that based on past experience and having used FEMDesigner, quite a lot of user support is expected to be required - that support time costs money, on a product which hasn't got much margin in it. ..

I've been through this all myself previously - there is good free FEA software, but it is of no earthly use to a newcomer to FEA.

 

[Bowber]

I've already found it's a steep learning curve David
As I said £210 isn't much in industrial terms, I pay that much for updates to my signwriting program, but it's earning money so I can justify it.
Is there FEA software that's usable by a newcomer or is it still a very dark art that takes a lot of training to get the best from it?

Steve

 

[DavidJ]

I failed to find anything useful for a relative newcomer (I looked on and off for 2 years or more). It seems to me that most of the free stuff has very poor GUIs and the instructions/tutorials assume a great deal of familiarity with FEA and even FEA software.

I use FEA occasionally in my work, so in the end spent around £3k for a package which was relatively easy to get started with (then the developers got bought ought and the product was killed). There are many paid for FEA packages which also have pretty poor GUIs if my search is anything to go by!

FEMDesigner has the potential to be awesome value, the non-linear materials capability is a fraction of what you'd pay anywhere else. The interface is reasonable and with a bit more development could be good.

To do FEA cheaply or for free, the best chance is to find a friend who has alreday mastered one of the free packages and get that friend to tutor you...

I gave up on free FEA, it was just too frustrating.

 

[H-L-Smith]

Steve,

Like you, I'm a total beginner with FEA although I understand the principles involved including the simplification process behind it, the mathematics and mechanics. That said, getting started is non-trivial and finding some material for self-education can be nearly as expensive as elementary software with many books costing more than $100 U.S. But, you found a good free product in LISA that is also very reasonable if you buy the commercial version later. Several compatriots in an automotive forum (2 are engineers) have been using it with good success for several months. One fellow also has access to ANSYS (the industry-leading and expensive commercial product) at work and has had a chance to compare LISA output to the results from ANSYS for the same models. He has found LISA to be within 1-2% in all cases thus far.

Below is the rendering of an example piece I modeled in Alibre Design. It is a simple pedal box structure for a small sports car. Perhaps it is the same order of complexity as the various motorcycle items you're modeling? I then created a STEP file from it with Alibre, which I input to LISA (actually the FEA modeling program provided with LISA) and the second model was created automatically for analysis within LISA itself. As I said, I'm a rank beginner. The model should be much, much simpler and would have been if I'd been smart enough to specify the correct simplifying geometries when creating the FEA model automatically from the STEP file. That knowledge will come from the education process. You do have a good degree of control over the complexity of the FEA model with LISA. However, the main point is that you can get an FEA model very handily from your Alibre models with almost no effort, which is what you really want.

If you search on YouTube you'll find multiple lecture series on FEA in English that have been video taped. I've gone through a couple of class lectures. They're a little dry, but could provide enough information to get you started with FEA. In the end, it looks like one just has to start doing it and carefully check your output for reasonableness, and above all, learn how to simplify the geometry of the FEA models, but still preserve enough complexity to get real-world accuracy.

Cheers,

Lonnie

 

[Bowber]

Thanks for the replies.

I've also just looked at the Z88 Aurora (thanks Seb) and it does seem a little simpler to start with, but I've not yet managed to import anything into both Lisa and Z88.
It doesn't help that my wife insisted I do a bit of DIY on the house today :roll: (she's not that bad honest)

Anyway I've run out of T&G wood now so I've been released into the workshop, going to get some welding done and then hopefully this evening I'll look at these programs a bit deeper.

I'd like to know the process you use to get a mesh from an STL in Lisa Lonnie.

Regards
Steve

 

[bigseb]

Steve, I'm actually sitting with Z88 right now. I recommend you download the video from their downloads page. Its about 250Mb. In it they do a complete analysis of a part from start to finish. explaining as they go along. Watch it, it really helps let you understand things better.

 

[jhardy1]

Steve,

Like you, I'm a total beginner with FEA ...

A word of warning to all "FEA Newbies" - in the wrong (untrained) hands, CAD-embedded FEA can be not only misleading, but down-right dangerous!

Pretty well all CAD-embedded FEA systems that I have come across will mesh all of your models using "4-node TET" (tetrahedral) elements, and thin components like plate-work, sheet metal, etc (as shown in H-L-Smith's embedded Lisa plots, for example) are almost always meshed as a single layer of very squashed-out tetrahedral elements.

TET4 auto-meshes are often fine for "thick" components (think of machine parts like shafts, gears, wheels, etc), but are generally next to useless (or worse!) for "thin" components.

If a plate item is meshed in a single layer of 4-node tets, the resulting mesh is completely unsuitable to determine stresses arising from out-of-plane behaviour (plate bending etc). For many plate items, plate bending is precisely the mode you are interested in, and a single layer of TET4s just can't do it! (If you want to know why, I'm afraid you’re going to have to learn some of the theory, and yes, that means getting hold of a $100+ text book, attending university, ...)

The software vendors very rarely mention this fundamental limitation of the auto-meshing algorithms they use, relying instead on vague disclaimers along the lines of "this software should only be used by trained professionals", but they should know full-well that by embedding the software in CAD, it will be placed in the hands of many, many untrained individuals.

I have been tinkering with the demo of FEMdesigner AD. Like most low-cost CAD-enabled FEA systems, it does not have any option to create "plate / shell element meshes" from Alibre models (which is what you REALLY want for analysing the behaviour of thin plate components - read the FEA literature to understand why), but it does at least offer the option of second-order (10-node) tets (yes, more expert training and reading is required to understand what that means!). A single layer of 10-node tets may be able to give you a “glimpse" of out-of-plane behaviour, but PLEASE people, don't ever design a critical component while relying upon FEA stresses which have been derived from an auto-mesh generated using a single layer of tetrahedral elements!

 

[Ralf]

Julian, you're absolutely right :!:

To understand how the software works is one thing, but how do I interpret the results is decidedly (much) more difficult,
and without a thorough knowledge of FEM/FEA/Structural analysis AND Materials Science (Engineering), very dangerous.

 

[H-L-Smith]

. . . <snip> . . .
Pretty well all CAD-embedded FEA systems that I have come across will mesh all of your models using "4-node TET" (tetrahedral) elements, and thin components like plate-work, sheet metal, etc (as shown in H-L-Smith's embedded Lisa plots, for example) are almost always meshed as a single layer of very squashed-out tetrahedral elements.

TET4 auto-meshes are often fine for "thick" components (think of machine parts like shafts, gears, wheels, etc), but are generally next to useless (or worse!) for "thin" components.

If a plate item is meshed in a single layer of 4-node tets, the resulting mesh is completely unsuitable to determine stresses arising from out-of-plane behaviour (plate bending etc). For many plate items, plate bending is precisely the mode you are interested in, and a single layer of TET4s just can't do it! (If you want to know why, I'm afraid you’re going to have to learn some of the theory, and yes, that means getting hold of a $100+ text book, attending university, ...)
. . . <snip> . . .

Thanks for taking the time to point out these issues. For me personally, I just wanted to establish an avenue for FEA analysis of my Alibre models with LISA and I did realize the model you see in the graphic was not a good one. It wasn't actually used for any serious analysis.

Some of those $100+ books on FEA also reference a specific software system such as ANSYS (or others) and have you working through FEA problems hand-in-glove with that software, learning both as you go. While that approach can be helpful, It does represent a serious commitment of time to master either topic. And, as been pointed out, you have to have some understanding of the underlying math, science and engineering to keep your FEA model relevant and understand the limitations of the results.

That said, even if the FEA output can't be taken as the gospel truth, it can be very helpful to see where the stresses or heat are greatest in your design as a qualitative indicator and a number of the current generation of software products do that visually very well, assuming your FEA model is representative.

Cheers,

 

[Bowber]

Thanks for the info Julian, As I said in my original post I realise that FEA can give incorrect results if used wrong, hence my question.
I've got a book coming about practical FEA that has good reviews from students etc so hopefully that'll give me some pointers.

Like H-L-Smith my intrest is in addition to me practical knowledge, I have no intention of designing critical parts and relying on some software telling me the part is fine, real problems have a lot of unknowns and my practical knowledge will always be used.
What I'm hoping to achive is looking at problems like a frame, apply a series of load conditions and see the high stress and displacement areas and make a design change to move or eliminate the point.

Regards
Steve

 

[jhardy1]

For anyone who is looking for a book - I can highly recommend "Building Better Products with Finite Element Analysis" by Adams & Askenazi. (My copy is 1st Edition, OnWord Press, 1999 - I don't know if it has been updated since.) This book is fairly light on theory (but enough to help you understand the basic mechanics of what is going on "under the hood"), and provides some really practical guidance on the use of CAD-embedded FEA. It is largely written around PTC ProMechanica - of course, the software technology has moved on a long way in the last 12 years or so, but all of the fundamentals are still absolutely valid. In particular, it highlights the issue of using auto-mesh generated linear-TET (TET4) meshes on thin aspect ratio parts.

The stress and deflection plots that are generated by your CAD-embedded FEA system may look very convincing (anything in Technicolour that is generated by a computer MUST be real!), and they may show "hot spots" in the right general areas, but don't think for an instant that the magnitude of stresses or deflections that are reported from a TET4 mesh are necessarily "accurate".

For example, I have been doing some very basic benchmarking and testing of FEMdesigner AD. I have made a simple model of a piece of cantilever plate 100 mm wide x 10 mm thick x 200 mm projection. (See attached Alibre file; I'm not sure whether the FEMdesigner loads and reactions etc will stay attached to this file, or if they exist in an external database.) The plate is assumed to be fixed at one end and is loaded by a 1 MPa distributed load at the tip (giving a total load of 1kN acting on the free end of the plate), acting perpendicular to the top face, so the plate goes into bending.

Using steel with an Elastic Modulus of 193,100 MPa, the classic solution for this plate would be a tip deflection of 1.66 mm, and an average bending stress of 120 MPa across the anchored end of the plate. Allowing for 3D stress distribution effects, the bending stress will actually vary across the width, peaking at about 134 MPa in the middle of the plate.

FEMdesigner actually did a reasonably good job using default auto-mesh settings - even though it meshed with a single layer of TETs, the default is to use 2nd Order elements, and it produced a result of 136 MPa maximum stress and 1.6 mm deflection.

However, if you mesh using linear TETs (TET4s - which is what many CAD-enabled FEA systems will do), you get a peak stress of 36 MPa (yes! 36, not 136!), and a tip deflection of 0.3 mm (yes, 0.3 mm, not 1.3 mm). Using linear TETs, you MUST force the mesher to give you a mesh several elements thick (across the whole model where ever plate bending effects are significant - not just the "hot spots"), if you want to have any hope of making reasonable predictions about stress or deflection. In the right hands, linear (4-node) TETs can be fine; when used by inexperienced operators, the results may not mean anything!

 

[Bowber]

I'd like to even get that far Julian, I'm not using an in CAD FEA, I've been trying Lisa and Z88 Aurora and I've not been very successful in getting my Alibre generated stl or step files to mesh.
Would the single layer mesh problem be relevant to tubes, I suppose not as you have 2 walls, but it would be relevant to angle or structural sections as well as flat plate.
If I had a frame made from tube and gusseted with flat plate then is the single layer mesh still a problem?

My book has just arrived so I'll be looking through it tonight.

Steve

 

[Sauzer]

I completely agree with julian.
I'm a beginner in FEA, but I'm learning a lot of things that convinced me that the initial results that I had on my first jobs were completely wrong!
(Fortunately they were only exercises )
Unfortunately FemDesigner AD can't handle shell and thin elements (for this moment I hope).
Surely you will need more money for books and training than with low cost software.

 

[jhardy1]

I'd like to even get that far Julian, I'm not using an in CAD FEA, I've been trying Lisa and Z88 Aurora and I've not been very successful in getting my Alibre generated stl or step files to mesh.
Would the single layer mesh problem be relevant to tubes, I suppose not as you have 2 walls, but it would be relevant to angle or structural sections as well as flat plate.
If I had a frame made from tube and gusseted with flat plate then is the single layer mesh still a problem?

My book has just arrived so I'll be looking through it tonight.

Steve

The short answer is - it depends ... but yes, it matters. You simply can't get a good detailed analysis of thin walled structures using only a single layer of TET4 elements, and that is what most auto-meshers will generate (whether they are CAD-embedded or stand-alone). At least when you are using tubular members, there is a reasonable approximate load-path which does not rely on local "plate / shell bending", but instead assumes that the stress at any point in the tube is constant through the wall thickness, but that is a gross over-simplification of the real situation, because there WILL inevitably be some local bending of the tube wall, which a single layer of TET4s will not be able to simulate.

How much does it matter? Aah, well - you're the responsible designer, so I'll let you be the judge of that ...

(But if you are designing bicycle frames, I wouldn't want to trust the design on a steep downhill descent without some physical laboratory tests first!)

 

[H-L-Smith]

Julian,

It looks like your recommended book "Building Better Products with Finite Element Analysis" by Adams & Askenazi must be a popular one. There is one new copy for sale through Amazon.com (USA), but it isn't $100, it's $200. :cry:

This conversation has proven particularly valuable to me since the things I like to design most all involve members made of relatively thin materials like tubing and RHS of 14 and 16 gauge, thin plates (3-6mm) and sheet metal of various types. Truthfully, I wasn't aware that FEA models for those materials required special consideration, so I thank you for that information. It did motivate me to poke around some on Amazon and see if I could find a introductory textbook more suitable for those materials, but I haven't found one yet.

I did uncover one very interesting book by Dr. Alan Morris which is way above the introductory level, but relevant in a general, cautionary way to our conversation here. It's all about controlling errors in modeling FEA analysis and just reading the introduction and preface using the "Look Inside" feature will drive the point home pretty well. Here's the Amazon reference:

http://www.amazon.com/Practical-Reliabl ... 482&sr=1-3

Thanks again,

Lonnie

 

[jhardy1]

This conversation has proven particularly valuable to me since the things I like to design most all involve members made of relatively thin materials like tubing and RHS of 14 and 16 gauge, thin plates (3-6mm) and sheet metal of various types. Truthfully, I wasn't aware that FEA models for those materials required special consideration, so I thank you for that information.

To analyse thin-walled structures which are modelled in solid modelling CAD systems such as Alibre, you are really looking for the capability to auto-mesh with plate / shell elements. I don't know of any low-cost FEA systems with that functionality. In particular, the tool you are looking for is commonly called "mid-plane extraction". This allows the software to recognise two closely-separated but related surfaces, and automatically generate a single average surface which can be meshed using plate / shell elements, which will mesh compatibly where two components meet (tee joints, elbows, etc).

I don't know of any low-cost FEA software that can do this; for example, FEMdesigner AD apparently does not have any ability to mesh with plate / shell elements at all – it is strictly a TET-mesher. Other low-cost software systems may be able to mesh with plate / shell elements, but you will typically end up with two surfaces of elements for each component ("inside" and “outside"), as well as numerous "sliver" elements around plate edges, cut ends of tubes etc. This requires an intensive amount of manual “mesh cleaning" to make the mesh suitable for use.

If anyone can point me at a low cost FEA system which can successfully mesh solid models of thin walled structures using plate / shell elements, I would love to hear of it!

 

[Bowber]

I'd forgotten how much math was involved in these calculations and to make matters worse I've forgotten all the math as well

Going to have to get all my old study books back out and go through them all again to remember how to do the basic stuff!
I wish my brain wasn't so full of useless information, I don't need to remember my first cars reg number :lol:

Steve

 

[bigseb]

Bowber, how are you getting on with the software? Getting nice results now with Z88 *happy dance*