[ardenum2]

Hi Peter,

I could not find any threads about how to model these carriages correctly for FE analysis... Any pointers?
In Bamberg's thesis, he suggests modelling the cars with an equivalent Young's modulus between 2 and 8 GPa. A lot lower than aluminium, and IMO way too low. Did I miss something?

Attachment 496588

why do you want to model them? just download the 3d data from the manufacturer

[jackjr-123]

My question is about how to model the bearings to get accurate stiffness values out of FEA. 3D data from manufacturers doesn't help with that...

[ardenum2]

My question is about how to model the bearings to get accurate stiffness values out of FEA. 3D data from manufacturers doesn't help with that...

you have a 3d model from a company and then you apply the same material in cad, there's nothing more to it. you want to model the ball bearings/rollers inside? why in the hell would you want to do that?

[jackjr-123]

Peter will understand.
If you simply model a piece of steel bonded to a rail, you'll get a much more rigid assembly than a real carriage with ball/roller bearings.

The goal is to check if the linear guides are under or over sized relative to the structure.

[ardenum2]

Peter will understand.
If you simply model a piece of steel bonded to a rail, you'll get a much more rigid assembly than a real carriage with ball/roller bearings.

The goal is to check if the linear guides are under or over sized relative to the structure.

are you designing a $250,000 machine? otherwise it's a pointless endeavor, manufactures give stiffness values for static and dynamic stiffness, that's more than what you'll ever need. you don't really need fem to tell you if a rail is too big. look at big floor type machining centers, the rails are tiny in proportion to the frame. the simple answer is if it doesn't fit geometrically it's too big, you're overthinking and overblowing things out of proportion.

[peteeng]

Hi Jack - The way Bambach does it is to create three bits of material, 2 on the sides and one on the top of the car. Then he adjusts the stiffness of those bits until the car total stiffness agrees with the published stiffness. Depending on the thickness of these materials changes the total stiffness. If you are using Fusion you can create an elastic connection and adjust that connection stiffness until it agrees with the published stiffness. I also suggest you don't use the manufacturers models but make a simpler model. I have found that Hiwin for instance have poor geometric details inside the car that create horrible meshes and very long solution times. For instance you do not need to detail the rail grooves, just use the bounding geometry. Before you go down this path in FE in a machine make a single rail and car and place some loads on it to test it solves correctly.

If you are in early days of the design no need to model at this deep level. It can get complex. If its early days just bond the car to the rail and make the car aluminium or leave it steel. Simple. Then your comparing apples to apples. One thing to be aware of is if you are interested in the real loadpath then you have to use sliding connections at the car and include a drive system, because the car has freedom along its path but no freedom transverse to the rail.... So the drive system (ballscrew, R&P or belt) contributes to the long direction compliance of the machine.

I've just looked at fusion fe and can't find the elastic connection. Maybe they have removed it lately, that's annoying. So you will have to use the 3 material approach. I've just run my std rail and car with sliding contacts and a 100N load. The car slide 7e6 metres before stopping 7km. So the drivetrain needs to be in place to stop it from sliding if you want to include this freedom. I just reran the model with the "remove rigid body motion" and the displacement reduced to 81mm but this is still too far for linear modelling. I'll dig around for the elastic connection... Peter


edit - much of this sort of FE work is comparison work not absolute answers work. It's also called a trade study or a sensitivity analysis. As long as your apples to apples across models and looking for a design direction or clear difference the models don't have to be exact. As long as you understand the assumptions and limits of the result and what you are looking for. In my line of work where customers are looking for exact answers it can take days to set up and test models before answers can be trusted... hundreds of connections to check, manual calcs to confirm, it goes on and on... KISS for this sort of modelling.

[peteeng]

Hi Jack - Heres a typical calculation to match the car stiffness. Also cars come in different preload conditions. I generally use medium as BST can supply med but not heavy. But have used heavy from other suppliers. Prefer heavy... Peter

[peteeng]

Hi All & Sundry - I have squared away a couple of things so I'm back to lapping blocks. I've forgotten where I was up to so I have started again with the P80 grit paper and done some AB and BC work. I have decided to screed the blocks with epoxy filler like I would do on a boat hull to fill the hollows vs bringing the entire surface down to the lowest point. I can still do that, but since this is an experimental project may as well have a go. That may save a lot of time especially when I get to the 3mx600mm one... I'll also use the grit I bought to see what that does. As the table would be for vacuum laminating it has to be sealed anyway and the if the surface is flat but patchy I don't mind... I have made the first bench for Lanky, I suppose its time to start a new thread for that build.... Many things to get done. Oh yes, plus warm up the grout samples as well for testing... Peter

[jackjr-123]

Thanks Peter, I was finally able to get better results. What I was missing is a sliding constraint on the top of the car to prevent it from rolling when applying a lateral load.
Attachment 496614
I've been doing these simulations in Solidworks, and let me tell you, it sucks. It's so buggy... I need to find something better.

I've been doing some FE analysis on a small benchtop C-frame structure (about 500 kg). So far it seems reaching an overall 10N/µm target stiffness will be hard. Linear bearings seem to be a large source of compliance.

To simplify the FEA, is it valid to omit the ballscrews, constrain the axis at the ballnut mount, and then calculate the overall stiffness as a series of 2 springs?
For example:
- calculated stiffness of the ballscrew assembly: 100 N/µm
- FEA stiffness result of the axis without ballscrew: 100 N/µm
=> overall stiffness = 1 / (1/100 * 2) = 50 N/µm

[peteeng]

Hi Jack - Yes as long as this is something you understand and it is done to provide a certain answer. My approach is to bond the cars and most things and accept that in reality the machine will be half the stiffness of the FE model. I build side by side different models or a model with variations to see what the differences are. To get a design direction not the exact answer. Then you have bolted connections to deal with as well. Trial Fusion360 as its FE is pretty good. If you download it as a hobbyist I believe its free. I use F360, strand7 and simsolid all have different strengths (and weaknesses). Benchtop machines are what I'm interested in and its difficult to achieve high rigidity. Large machines have the advantage of large geometry. I would look at other configs vs the C frame if your chasing more rigidity. Portals, moving column can be stiffer. The C frame design is a function of its history and casting limitations and using ways vs linear bearings. It had to have its drives at the "friction centre" so they did not jam or stutter. Linear rails have no friction or stick slip so other geometries are possible... Peter

I don't include the drive system until I have dialed in the structure. If you try to do the drive system plus the structure in round 1 then I think its too complex. Round 1 is to get a good structure going then once you firm this up add the drives in Round2. Be aware of how the drive works in R1 so you don't short sheet yourself somehow. Good Luck

[ardenum2]

solidworks has great coupled thermal-flow simulation. straightforward and easy to setup, for heat transfer between solids and liquids eg. watercooling of components. so far I haven't been able to do the same in nx, the setup is so damn complicated.

[jaguar36]

I've been doing these simulations in Solidworks, and let me tell you, it sucks. It's so buggy... I need to find something better.

I've been doing some FE analysis on a small benchtop C-frame structure (about 500 kg). So far it seems reaching an overall 10N/µm target stiffness will be hard. Linear bearings seem to be a large source of compliance.

To simplify the FEA, is it valid to omit the ballscrews, constrain the axis at the ballnut mount, and then calculate the overall stiffness as a series of 2 springs?
For example:
- calculated stiffness of the ballscrew assembly: 100 N/µm
- FEA stiffness result of the axis without ballscrew: 100 N/µm
=> overall stiffness = 1 / (1/100 * 2) = 50 N/µm

There are alot of better FEA packages than Solidworks, but they all cost alot more. Abaqus and Patran are my favorites. Unfortunately Autodesk removed the simulation module from the free version of Fusion.

Trying to accurately model a complete CNC mill with all of the linear motion components, spindle, tools, etc would be a fools errand without some test data to correlate it too. You should just be trying to model the structural frame of the mill. Not only is the stiffness of the linear bearings difficult to quantify, its also non-linear with load. Here's a bit from the Rexroth linear motion catalog, which has some great information in it and is well worth the read:

[jackjr-123]

Yes Rexroth catalog is quite good. They also provide rigidity charts for a given car size and preload. They are not perfectly linear but it's close enough.

I don't think analyzing just the frame stiffness and then throwing some random size bearings at it is a good approach, as they are a major source of compliance.

[peteeng]

Hi Jack - If you use the bearing suppliers design manual they are selected 1) by application 2) accuracy required 3) dynamic load condition - choose the ballscrew to be the same size as the rail ( 4) static loading 5) preload 6) consider stiffness required 7) consider the bearing life 8) figure out lubrication

I have usually not done this in this order! But modern FE systems can model to high accuracy if you put the time into it. I designed a crane once the prototype was called T-Rex. The owner of the company asked me to predict its deflection when it lifted 1.5T before it was complete. I forget the exact answer but he came down to the test day and was surprised I got it nearly to the mm... or I should say Strand7 got it to the mm Peter

Jack - do you have a thread for your machine?

Hi Jag - Good to know what free Fusion does. Thanks for the link. They have been changing the features for some time. Moving things around and into the extensions....

[jackjr-123]

Peter,

I have no doubt that this can be simulated accurately. I'm trying to learn more about FEA but the learning curve is steep. I never know when the result can be trusted. Sometimes I get completely ridiculous results: I make some part fully rigid and the max displacement increases a lot instead of going down!

As far as I read this thread you always used aluminium (70 GPa) to simulate the cars in your Milli designs. Compared to Bamberg's equivalent Young's moduli (2-8 GPa) this is WAY higher.
Did you try to model them better to get more accurate results?

[peteeng]

Hi Jack - I have modified that to steel recently. I do not worry about car stiffness unless I'm dealing with the drivetrain. I'm more concerned with the overall structural configuration. Drivetrains are considered after the structure is nailed down.

The car material stiffness changes depending on the materials thickness that you use for the modelling. Originally instead of modelling the two directions I decided that the vertical and trans stiffnesses where close enough so I changed the total car stiffness (saving me modelling so many small parts) until it was close to the correct stiffness. Instead of creating a new material I just picked aluminium because at the time it was close. Where your heading requires a lot of small models to be made and checked (eg ballscrew and nut stiffness, car and rail interacting, nuts and bolts etc) usually by hand. Then use the same approaches in the big model knowing that the small bits are correct. Until you get confidence in the total model. With contact models mesh size is important have you looked at that? Just because the code says its "connected" does not mean its correctly connected. Each connection has to be visually checked, the automated connections have let me down in very sophisticated software (simsolid) and some time ago ansys. Have you studied mechanics of solids? statics and basic mechanics? If not then FE is a very big subject and its clearly garbage in garbage out territory.

Regarding displacements changing publish the model here and I can look at it. A step file is good. With FE you have to know the answer your looking for as this can change the approach or methods... and a reality check is always recommended if you can do the manual paper calcs. Peter

I analyse a lot of bolted machinery in the mining industry. I have been using a system for a couple of years and always seemed to get good results. Recently I had to do fatigue work on a machine and the bolts did not make sense. So I did a 2 page calculation to prove out the FE and it turned out that if I preloaded the bolt then the service load was never included in the FE bolt load. Shock and horror. I have been sending the output of this for various things for years in good faith! I sent the calcs and concerns to the company. They confirmed the calcs where correct but to get the correct answers in their FE I would have to update to the advanced module at a cost of $50,000AUD!! I have sent a very serious note to their development manager saying they have to get this right!!

[jaguar36]

I have been sending the output of this for various things for years in good faith! I sent the calcs and concerns to the company. They confirmed the calcs where correct but to get the correct answers in their FE I would have to update to the advanced module at a cost of $50,000AUD!! I have sent a very serious note to their development manager saying they have to get this right!!

Which software was this on?

[peteeng]

Hi Jag - I won't say at the moment. They have been generous and solved a few issues along the way. I give my software providers a hard time. All FE systems I've used have niggles and have to have work arounds. The more automated they have become the more untransparent various things become and it takes time to discover these issues. We can now solve very complex models apparently easily, probably too easily and there needs to be checks involved. But checking 1000 connections in a model is an issue or checking 20km of welding in a structure. I spend more time checking things now then setting up the models!! So keep things as simple as possible to achieve what you are trying to achieve... and reality check as much as possible. Peter

[peteeng]

Evening All - I have started sorting the lapping plates again. I have sealed them with thin epoxy infusion resin and will cook them when they get a set. I was going to use the fairing compound immediately on the liquid epoxy but decided to let it seep in for a while then use the putty. On a couple of projects I have found that using a liquid epoxy as a primer prior to using putty has resulted in very much better results via tensile testing. Seems to me the putty does not penetrate as well as the liquid epoxy does. Peter

[joeavaerage]

Hi peteeng,
a pic of the platter for my fifth axis, I finally got the T slots done!

Craig