[G-Spot]

[QUOTE=peteeng;2581894]Hi G - I can see the attraction of the twin gantry /QUOTE]

OMG! they hacked into my computer and stole my designs.....lol...and they put the track under the beams and their Z axis is rotated 90degs to the x axis.

Now I want to go back to twin beams again, I was tempted to abandon it due to the complexity to build, you need 2 screws instead of one on 2 axis, not just the one.

Also from looking at bearings I think I want to go for the 32mm cars and you lose so much track mounting under the beams, so that makes them criminally expensive. If I think I have a chance of producing a frame that is super stiff then I am going to go 32 otherwise 25.

[G-Spot]

may as well buy a D300 heavy wall tube and weld flats to it. Has been done before....

Yes I was thinking of that, while normally a pipe would be a nightmare to bolt flat things to, You cant just bolt stuff on to the flattish faces of a square beam .Could be simple in that you can laser cut the mounting flanges and it's easy to measure the dimensions they need to be. I would rather fit a new flat face to a pipe than fix an existing crooked face. Start with an existing flat plate, align with a laser and mirrors, weld it on, stress relief then recheck... I see one of those super accurate machines that you sent me in that pdf on machine building was built completely from pipe even the gantry posts

My build plan is for complete DIY on this though I will get my metal supplied precut and machined. Plan to grind the track mounts flat, using laser/camera/mirror setup that I have used before to machine super flat.

I have downsized this machines dimensions already, the research so far showed me that I am going to pay DOUBLE for every meter extra on top of 750mm for each axis. And it would be a lot cheaper and simpler to make a custom machine to weld and mill the plates because it is done on a single axis and stuff like needing the purge gas on the other side of the plate. So I am going to build at a max of 1m machinable space on each axis.

But the decider was availability of bearings and screws, I can order from reliable suppliers and get top grade used parts easily shipped if 1.5m or less. Anything over 1.5m and the shipping is going to cost more than the parts.

Yeah the science was starting to look really bad for a metal milling machine that also welds together really large flat plates. And also moving gantry is back on the cards because with a fixed gantry you have to achieve the same level of stiffness with your moving bed across it's entire area of more than double your machinable area, so that means you will need double the steel that you used for the rest of the machine. You would be looking at $5000 just for that one axis.

[G-Spot]

Talk about serendipity, If I hadn't made that mistake of using a 90-45 triangle instead of an equilateral I would have been stuck on Hexagons.
But from this octagon I can see clearly it is a far better shape to use as a construct.

Also the optimisations are working on my complex shell structures while before I could only use them on solids.
Learning how to align and connect parts properly really paid off.
Working with shapes with so much symmetry pays off because if your connections are off you can see it clearly in the optimisations lacking symmetry.

Early FEA was all about the freaking meshes which you needed a degree in trigonometry to understand. A connections based approach works far better if you want to make something useful.

Attached below is the Original 100 ton Octobeast.Then I remove the centre triangles and the rest of the pics are based on the reduced Octagon with the 5000N test load acting in bridge mode, I am sure when I do the other optimisations in other force directions it will look different. This is how the optimisation believed material could be removed based on that load.

The Octagon with the centre removed only showed a 2 micrometers increase in deflection, What's very interesting is that it weighs 600Kg which is the same as the twin beam machine that came before it. And that twin beam was only capable of holding about 50 tons and had double the twist in it.

[peteeng]

Hi G - When I started doing FE I had to make meshes manually. ie calculate every node point in the structure using simple CAD and a spreadsheet... (some 40+ years ago) you and others now have it easy with automeshing and optimisation and solid models vs surfaces etc. You still have to understand the basics. Garbage in garbage out... My observation of your systems optimisations are that its a stress based subtractive system.eg it creates a relatively fine mesh then removes elements that are under a certain stress. Does it create new geometry? I don't think so. Fusions optimisation system can just start with the parts you want to connect and it will evolve an optimised shape dependent on the rules you use for the algorithm. eg minimum mass, max stiffness, target weight, exclusion zones, even the manufacturing route (milling, 3D printing, etc are built in) if your serious with optimisation I think you need to move to Fusion. Peter

edit - it has created an arch because the ends are fixed and it has taken advantage of the foundations infinite stiffness. In reality your gantry is connected to a compliant structure and will need to be different....

[G-Spot]

That escalated quickly.

I now have OctaSpawn which is an Octagon with its belly cut open and it guts inside out.

Like crazy stiffness figure. Into sub micrometer stiffness levels, theoretical of course.

If you look at whats emerging it looks like 2 router tables joined back to back, feeling like I might just have reinvented the wheel.....lol

Anyways this design is very weldable and boltable, I have long arms and I am cunning and this is a DIY build, we have nights as long as days.

I wont be welding bolting it up in this configuration, I will split the double back to back plates with a single piece and the long diagonals will be a single sheet. Exactly like they did with those super tall buildings to get great stiffness.

Definitely going moving gantry or as Peter would say "gantry"' going to add some stubby legs leading to 750mm apart legs, I don't care that I will lose that much of my axis, the extra in dollars is just for rail length and not the screws the screws can stay at their original length. Also to remain balanced you cant go higher with your beam height than your width of legs, so if I go 750mm spacing on my legs, I only pay for an extra 250mm of rail than I planned but I gain the ability to machine to a depth of 750mm and still remain machining within that triangle around the centre of mass.

I can also make the base that my gantry runs on just Octagons, filled with cement and re-barred into 5 tons of concrete compacted into the ground.

[peteeng]

Hi G - A couple of points:
1) If these triangles are analytically bonded together make a monolithic version and compare results. Your bonding area is large compared to the structure and its a known issue that bonds are overstiff
2) Your elements are one plate thick. Remesh finer say you have 3 elements thick so say a 3mm mesh size and compare. Depending on your element type the thru thickness needs more elements to capture shear and membrane effects again it could be overstiff
3) Note the overall dims and loading etc and I'll replicate something in my system to see what happens....
4) Why is the bottom of the section open?
5) The across wise loading (usually X dirn) is always the issue not the downward load as you have it. A machine never sees a load in that direction. Plunging loads move the gantry up but it should be the same stiffness up and down being linear but its good to check. Sideways loads (X dirn) place the gantry in torsion and bending which is a more difficult case to get stiff.
6) You have the entire end of the beam restrained. This means the foundation stiffness is infinite. Better to just restrain the bottom edges and let the ends rotate as this will be closer to reality...
Peter

[G-Spot]

Fusions optimisation system can just start with the parts you want to connect and it will evolve an optimised shape dependent on the rules you use for the algorithm. eg minimum mass, max stiffness, target weight, exclusion zones, even the manufacturing route (milling, 3D printing, etc are built in) if your serious with optimisation I think you need to move to Fusion. Peter

edit - it has created an arch because the ends are fixed and it has taken advantage of the foundations infinite stiffness. In reality your gantry is connected to a compliant structure and will need to be different....

Thanks for that info, will definitely check it out. The restrictions on this student version make it impossible to do any proper optimisation. I am using it just as a guiding tool and am doing my own optimisations, which I can see are clearly better than anything it is suggesting. I know the legs being fixed does not represent the real world but because of the restrictions I cant work on the base and gantry together. And it wouldn't let me model in free space, so I have to model assuming the rest of the machine is theoretically perfect, its not a problem because when I have a full design I can then strip away all the parts to basic components then test for real world. I also have decades of experience in FEA spending hours editing meshes manually before any graphical interface. I am no expert by any stretch of the imagination but I think I can use it well enough to produce something that works better by design and not pot luck. Solving a problem using exact numbers is not necessary if you know whether a change made something better or worse, you can iterate to the correct design. I am confident that the the deformation data is good enough to do this. But keep the valuable info and critique coming I value your experience in helping me along the way.

[G-Spot]

OK after doing 100s of simulations I am finding that just making a simple box shape form using thick 20mm plate is not even close to how stiff you can get with complex forms using only 10mm.

That inside out Octagon is deadly when it comes to stiffness, it seems if you wrap your structure in a chain of 90-45 triangles it takes on magical properties of stiffness, it's like you wrapped it in a bridge.

[G-Spot]

If you look at the deformations closely, you can see the triangles distribute the load around the skin perfectly no matter where it comes from.

[G-Spot]

Hi G - A couple of points:
1) If these triangles are analytically bonded together make a monolithic version and compare results. Your bonding area is large compared to the structure and its a known issue that bonds are overstiff
2) Your elements are one plate thick. Remesh finer say you have 3 elements thick so say a 3mm mesh size and compare. Depending on your element type the thru thickness needs more elements to capture shear and membrane effects again it could be overstiff
3) Note the overall dims and loading etc and I'll replicate something in my system to see what happens....
4) Why is the bottom of the section open?
5) The across wise loading (usually X dirn) is always the issue not the downward load as you have it. A machine never sees a load in that direction. Plunging loads move the gantry up but it should be the same stiffness up and down being linear but its good to check. Sideways loads (X dirn) place the gantry in torsion and bending which is a more difficult case to get stiff.
6) You have the entire end of the beam restrained. This means the foundation stiffness is infinite. Better to just restrain the bottom edges and let the ends rotate as this will be closer to reality...
Peter

1. OK I created monolithic versions and they showed the same super stiffness.
2. I cant remesh any finer because of the restrictions on the student version.
3. See new drawings for how I am testing.
4. It's open because I am doing a twin beam design and solving the problem of lack of distribution of the forces between the beams by closing the top off, the spindle and z-axis will live inside the beam.
5. Yes very aware of this, all the standard shaped beams are terrible against x axis.
6. Fixed that using friction-less supports that can move in the same plane as the their face.

[G-Spot]

This shows that with 10mm wall thickness you cant really go more than 250mm unsupported.

So if you used a 300mm x 300mm box section that was only 10mm thick you wouldn't really gain anymore than a 250mm x 250mm

This is basically 2 box sections 500 x 200 joined at the top and the deformation figures have dropped off massively instead of increasing as you might think.

I had to take it to 30mm thick for it to perform as good as the stiffened sections I used

My design from the Octagon was 750kg for the same figures.

So it looks like if you got the time to spend fiddling around installing stiffeners within a beam then the rewards will be very much worth it.

Looks like a great strategy to build a machine with similar accuracy as big manufacturers of CNC machines who definitely haven't got the time to be installing intricate systems of stiffeners and still make a profit.

[peteeng]

Hi G - A 300x300x10t section is geometrically 1.76x stiffer then a 250x250x10t section. So it should be much stiffer. The issue you are hitting is local shear stiffness. The lack of local shear stiffness is reducing the efficiency of the bigger beam as it is warping or lozenging or both under load. You can see this in your deflection plots. You need rails and bearings to be on webs, corners to be thicker so they stay at 90degs etc so the global stiffness can be taken advantage of. If you look at commercial machines where the loads come in and out of structures eg supports, bearings, rails they are thick and connect to something. You played with triangles and that's one reason they are good, they stay in their original shape as its a self-supporting geometry. Machines use thick structures to overcome various shear load issues... As you have found out thin structures need lots of secondary structure (webs ribs doublers) or triangles to stay in shape. Keep at it. Peter

as you have said before all the little things get amplified. See eccentric image for more info.

[G-Spot]

Switched to bearings. What an absolute nightmare.

The CAD drawings for the bearings didn't fit the ones for the rail and they had no bearings inside them.
So I had to scoop out the rails and place about 150 bearings inside, just as well I have been spending the last few months honing my alignment skills.
And it seems like CAD bearings are just as likely to go flying all over the place as real bearings, when you fail to select one out of 148 you don't notice it is missing and then you miss another one, next thing you think you going mad because you notice tiny dots all over the screen...and only then do you realise your mistake.

Does anyone have the CAD drawings for HIWIN model HGW25CC-ZA-C. ? You can download them from their web site, if you live in the USA or EU it would seem. I tried and they said they would email them to me but they never arrived. And I see you can only register on their web site if you live in a few dozen countries.

Anyway I think the drawing of the block I have is the right one, I think the rail is for the standard HG and not the HGW. Who freaking knows, I wanted to do FEA on them and ball bearings and the correct rail are kind of important. Never the less I managed to draw my own bearings and fix the rail and get the FEA done.

I discovered that the simulation will take about 24 hours...per bearing carriage...so I am going to put 4 metal pins where the bearings should be and do the rest of my development that way. I have an old water cooled computer here that was overclocked for about 7 years before it got too stressed and occasionally just resets. I will use that to run the bearing simulation. I found some research on these same bearings looking at computing their stiffness and comparing it with the manufacturers specification. They used a lab grade destructive testing instrument to confirm the FEA simulations. I used the info they shared to get my own simulation working. But making sure my simulations produce the same results they got can help me a lot.

[peteeng]

Hi G - For simulation do not model the ball bearings or use the models from Hiwin. They are overly complex for the situation. Make envelope models for your purposes not detailed eg don't model the ball grooves just the outside dims. Modelling contact between curved surfaces is only done if that's what your really interested in. Its a very mesh/solver dependent exercise so don't go there. I cover how to model cars in a couple of spots on the forum. In early development, model the rail and car envelope, make the car aluminium and that's close enough. Even having the car as steel is OK if the modelling is for comparative purposes... apples to apples... Peter

If you make a contact a sliding connection even though the surfaces are flat and in reality would come apart in the FE they do not. The contact is just given mathematical freedom not real freedom...

attached is Milli a development mill I'm working on. It has cars and rails as I would model them as an example.

If you get to the point of a detailed model, Hiwin and others publish the bearings stiffness. Then you adjust the material property to match the bearing stiffness. I cover this elsewhere in the forum... Looking at your models you need to remove small fillets and radiuses. These require small meshes and then a mesh gradient. this increases mesh count and solver time. this process is called defeaturing. You defeature your model to the level that answers your question. If your interested in a stress at the corner then that needs featuring, if not defeature it. The aim is to make the simplest model that answers the question you are asking, In this case you are interested in global deflections so keep the models very simple...

I've spent may years educating clients about the differences between manufacturing level models and FE level models and how to set up configs for both. Most of my time is spent defeaturing models. I now use simsolid which is a meshless FE system that can work with fully detailed models yet solves in minutes vs hrs, but its a very expensive system but my clients need to know what's going on in 100's of bolts in mining machines. and this seems to be a good economic solution for this level of engineering.

[G-Spot]

OK I managed to get the FEA to process quickly just by replacing the bearings with a single same diameter pin.

Abandoned the twin beam design, I want to use a proper steel milling spindle so too much happening on the top of the Z axis to have to thread it through a small gap.
And like Peter inferred twin beam is the best when you take a beam good enough to do the job on its own and then add another. So they are twice as heavy and twice as expensive.

Going with back to basics and upsizing, thick 50mm front plate with 2 supporting 30mm plates in a 90-45 triangle configuration.

This solved a lot of problems especially when it comes to making it. Only a single milling operation required on my front plate. De-stressed after and because its a triangle configuration it will be super easy to check before paying. When I collect they can stack them against each other and if all 3 dont align you know it's not straight. Not really possible to perform a quick check if its 4 plates and you are making a square section.

Now to tune this design I need to know how much weight is too much weight, this weighs 578Kg including the spindle box, is this too much?

What are the penalties for too much weight? Is there anything you cant do if it's too heavy? obviously it's going to be slower.

[G-Spot]

Hi G - For simulation do not model the ball bearings or use the models from Hiwin. They are overly complex for the situation. Make envelope models for your purposes not detailed eg don't model the ball grooves just the outside dims. Modelling contact between curved surfaces is only done if that's what your really interested in. Its a very mesh/solver dependent exercise so don't go there. I cover how to model cars in a couple of spots on the forum. In early development, model the rail and car envelope, make the car aluminium and that's close enough. Even having the car as steel is OK if the modelling is for comparative purposes... apples to apples... Peter

If you make a contact a sliding connection even though the surfaces are flat and in reality would come apart in the FE they do not. The contact is just given mathematical freedom not real freedom...

attached is Milli a development mill I'm working on. It has cars and rails as I would model them as an example.

If you get to the point of a detailed model, Hiwin and others publish the bearings stiffness. Then you adjust the material property to match the bearing stiffness. I cover this elsewhere in the forum... Looking at your models you need to remove small fillets and radiuses. These require small meshes and then a mesh gradient. this increases mesh count and solver time. this process is called defeaturing. You defeature your model to the level that answers your question. If your interested in a stress at the corner then that needs featuring, if not defeature it. The aim is to make the simplest model that answers the question you are asking, In this case you are interested in global deflections so keep the models very simple...

I've spent may years educating clients about the differences between manufacturing level models and FE level models and how to set up configs for both. Most of my time is spent defeaturing models. I now use simsolid which is a meshless FE system that can work with fully detailed models yet solves in minutes vs hrs, but its a very expensive system but my clients need to know what's going on in 100's of bolts in mining machines. and this seems to be a good economic solution for this level of engineering.

Hey I like your Milli machine, I even found your build thread, so much still to read about, all 100 and something pages. This is how many of the modern CNC machines are made? yes?
I think it is working on the principle that if you got to stick something out unsupported , its better to do it sooner than at the very end of compounding axis' but you already know how much I know on the topic...lol

[peteeng]

Hi G - For machining aluminium I think the design is way overcooked and way over weight. But consider you are refining the concept. Its easier to remove weight once you get the concept right. Remember to align the rails with the webs... you have to remove all eccentric and secondary moments or loads... Peter

[peteeng]

Hi G - In another thread I did a bit of work on gantry shapes and triangular came out very well. Thinking ahead you will need to have lands to enable machining jacks or stands to be used for the final machining. Especially if its on a VMC. or you set up on the rough front face and machine lands along the back then turn and final machine face. At some point you have to figure out how to final machine this sort of part.... Peter

if you send me the step file I can run it through the fusion optimiser, get rid of the rod in the cars and simplify a bit, add a spindle and your close...---- Peter

edit re weight and speed - you have to decide what the max cutting speed you need. Its big structures with small stepovers so you require very high speeds unless you want to take a year to get the job done. so research cutting speeds. I expect to cut very fast you will require a 40,000rpm spindle. Then work back from that using a suitable chip per tooth spec to figure out feed speeds. Then once you know target feed speeds you can figure out air times for a typical job. This will tell you what sort of accel you need and then weight is your enemy. If you accel at typical commercial mill accels say 0.2g then your 500kg gantry weighs 500*0.2*9.81=981kg to move... ie big motors; inertia loads can get very high... Peter

consider some commercial mills are at 1g thats scary big stuff...

[G-Spot]

Already got that x-axis deflection down to 13 um by making the spindle post 30mm thick.
That makes the weight total 650Kg


Don't think I want to go anymore than 700kg with everything attached including the spindle motor.


Well if the FEA is over stiff by 20% and 20 um is what I need to aim for in real life then anything better than 16 um in the FEA should work.

One of the reasons I chose 330mm for the size of my beam is because how the steel plate is supplied. I know the cut and supply people will cut me plate of different thicknesses but if their price is bad then I might just get directly from the mill. That means I can get the 3 beam lengths from a single 1m wide sheet.

[peteeng]

Hi G - Start talking about the static stiffness of the machine. 1000/13= 76N/um which is quite stiff. Even half of that will be fine for machining aluminium... but that's real stiffness not model stiffness. Peter