[jackjr-123]

Hi all,

I've been planning on building a small benchtop mill-drill, about the same size of a RF45.
Yes I know I could spend only $3000 and get a chinese RF45 clone, but I prefer starting from scratch and make it exactly the way I want rather than spending a lot of time fixing and improving a chinese kit.

The main constraint of the build is weight (sadly). The floor the machine will rest on is less than ideal and has poor load capacity. So I want to stay below 500kg for the machine. Current initial design is already at 450kg.
This is a machine intended for prototyping, so speed and production features are not a concern. "Slow is smooth and smooth is fast".

Here's the current early design, a lot of details are still missing.


General
- Travels XYZ: 450 x 250 x 400 mm
- Servo motors on all axes. 750W with brake on Z, 400W on XY.
- Absolute linear scales on all axes (Mitutoyo AT715, easily interfaced with custom software).
- BT30 spindle with pneumatic tool changing.
- NC system based on LinuxCNC and EtherCAT. Custom software with emphasis on semi-manual machining and conversational programming.

Headstock
The headstock is built from bolted steel plates, between 15 and 20 mm thick, and weighs about 45 kg.
The spindle is a BT30 model from china, going up to 10k rpm. It is belt driven by a 3.7kW Kollmorgen AKM-54L servo motor. My plan is to start with a 1:1 pulley ratio for simplicity. This would give a max speed of 6000rpm and 9 Nm of continuous torque. If this turns out to be too low, I would switch to a 2 speed PolyV system, with some kind of quick change mechanism.

Motion system of the Z axis: 25 roller guides and 2505 ballscrew.

One of the major feature is to be able to use the machine as a drill press. The traditional way to do it is a quill and a spindle with a long splined shaft. This is complicated to make for a DIY build and an extra axis also hinders rigidity. The simplest way on a CNC machine to manually control an axis is a MPG, but you loose the feel of the machining forces.
My first idea was to design some kind of mechanism to disengage the ballscrew nut from the headstock and make the Z axis move on a rack/pinion drive with the feed lever. So far I haven't figured out any good way to do it.

My latest idea is to connect the feed lever to a small, independent servo motor mounted inside the headstock with a high 1:15 gear ratio. The servo encoder would act as a MPG to control the Z axis movement and the motor would be controlled in torque mode to provide force-feedback and mimic the forces seen by the Z servo-motor. Let me know how crazy this idea is

Column
The column is also built from bolted steel plates, 15mm thick. Current dimensions are 800x300x230 mm and about 150 kg.
It is hollow to allow mounting an air cylinder to counter-balance the headstock.

The best way to build the column is still to be defined (position/spacing/size of bolts).

The rails mounting surfaces are 2 flats of precision ground steel bolted to the column. They are called "precision ground" but are far from precise enough to mount linear guides as is. No shoulders are planned to allow spotting the whole column on a surface plate for scraping. An alternative to the steel flats for easier scraping would be using continuous cast iron but it's even more expensive and needs to be machined. Not sure it's worth it.

Base
Currently the base is a single piece made of E80 UHPC (Durcrete), assuming the 80 GPa modulus is realistic. I will have to make a few samples and check their stiffness.
Dimensions are 750x400x200 mm and weight about 120 kg.

What is still unclear to me is how the base should interact with the stand. Should the stand be part of the base, increasing its stiffness, or should it be independent from the base?
I assume in a C-frame structure the base is weakest in bending, so the thicker the better. If the stand could stiffen the base, maybe some weight could be gained...

There are still many things to discuss, but that's all for now.

Let me know what you think.

[peteeng]

Hi Jack - If your happy with a steel bolted column why cast a base? A steel base will be stiffer? I'm interested in your test for the E80 grout. The column is usually independent so the column can be adjusted vertically and or machined. So the base can be placed into a mill and the lands machined, similar with column.
See rough calc attached. Steel is potentially lighter and stiffer in same envelope... Peter

[jackjr-123]

I also thought that steel had the best stiffness/weight performance for the base, and I tried making a steel base (simple box shape). I ended up with a heavier and less rigid result according to the sims. To match the stiffness of the single piece UHPC, internal ribbing/stiffening was needed and then it was way heavier and complex. Maybe due to the minimum required plate thickness for bolting adding a lot of bulk and the suboptimal load paths?

[jackjr-123]

About machining: the only machining allowed is the one I can do myself. I can use my previously built CNC with 640x500x250mm travels (a fixed gantry design, mainly designed for aluminium milling). It will be suitable for semi-finishing, but scraping will still be required for final accuracy.

[jackjr-123]

This is the steel base (15mm plates, hollow) vs the UHPC one. Quick sim with 3000N on the column bearing surfaces.
Steel base is 132kg vs 120kg UHPC.
Attachment 502110Attachment 502112

[peteeng]

Hi Jack - Can you use the go advanced at bottom of thread box to embed images? This is a better method. The attachment does not work. My conclusion has been that aluminium is a plus for a bolted build to save weight. But may cost more. There is a cost paradox. The lighter it is the more it can cost... I'll have to see your image to comment. For the base I've found deep webs under the rails is the starting point. Peter

[jackjr-123]

Weird the pictures show up fine here. Let me try again.
Attachment 502114 Attachment 502116


I don't mind aluminium, but bolted designs are already a lot of work. If you need to add ribs everywhere it becomes unmanageable.

[jackjr-123]

The above sims are obviously not a pure bending load since the base is fixed on 3 points with 2 of them directly below the column bearing surfaces. With a pure bending load the steel base is indeed stiffer, but not by much. It is 10% stiffer, but also 10% heavier.
The UHPC base is slightly optimised in weight thanks to 2 foam cores inside.

[peteeng]

Hi Jack - I think the way you are loading it is probably not useful. Looking at your original design and putting the base aside for now you need to work on your saddle, I see it's a bit compact and the bearing arrangement and bolt access will need some work. I think the 200mm deep base is a bit thick for this size of machine. Create a simplified model of the machine to start with and load the spindle nose and bed to establish its potential static stiffness. Thats the real stiffness you should be interested in...Peter

I usually start from some philosophical position to investigate. In this case I'd look at the attached image as a starting point to see how that could be developed. I could have these bits laser cut then the lands machined at a machinist. They would then be orthogonal. Then the secondary bits need to be sorted.

[jackjr-123]

Yes the saddle needs work. However as is there is no issue with the carriage mounting bolts. First the saddle plate alone is bolted to the Y axis carriages, then the whole assembled table is laid on top. The inner carriage bolts are accessible with holes through the table.
I did some preliminary whole-assembly fea and it's scary: a whopping 7.5 N/µm stiffness at the tool. And the freaking thing is all bonded! I need to investigate...

Edit: I forgot to define the fixed workpiece on the table as rigid. With that now fixed I get 14 N/µm for XY and 65 N/µm for Z.

[peteeng]

Hi Jack - The 65 for Z is respectable - You will need it for drilling. The torsional stiffness is always a hurdle. But 14 is getting there. Small machines have limited geometry so if you want max stiffness you need to make the column wider and take advantage of all the space you have available within the footprint. Since you have a global model you can restrain the cars at the arm and measure the deflection. Then restrain it at the column base and run it etc so you can determine the "structural loop" that is most compliant. Then deal with that loop... Keep looking for that Grail... Peter

[jackjr-123]

Peter, thanks for all your advices.
Not sure what you mean by "restrain the cars at the arm" ?

I worked on sims all day long and overall stiffness seems to be "death by thousand cuts".
One weird thing I noticed and cannot explain is how the model is restrained. In the above sim the machine was held by a rigid 50x50x50mm cube on the middle of the table. If I bump its size to 100x100 mm the stiffness in XY jumps to 20 N/µm.

So far the base seems to be where most improvement can be obtained. The slightest bending in the base results in a large angular error at the top of the column.

[peteeng]

Hi Jack - An FE model can have several Loadcases (LC 1 LC2 and LC3 etc) eg in your model you have a X load a Y load and a z load. Loads can be forces, pressures and displacements. Have you tried placing a 1um displacement at the spindle and see if its the same stiffness as the "forced" condition? Saves calculating the rigidity.

The other aspect is Freedom cases (FC). You can restrain the model differently to get an understanding of what's going on. So FC1 would be restraining the cars at the outreach. This calculates how far the outreach deflects. Then FC2 you remove FC1 restraint and restrain it at the base of the column, calculate its deflection. Now (Delta2 - Delta1) is how far the column moved. Then restrain at the saddle cars etc. Now you know what each structural part (also called a structural loop) is deflecting so then can concentrate on the one that is the max deflection.

These can be done in copied studies. Some systems have loadcases and freedom case definitions.

In a VMC the loadpath is obvious. In more complex structures the loadpath may not be obvious so you have to do a little work to figure out the primary bits and secondary bits.

You can also just exaggerate the deflection really big and this usually gives you a visual clue on what's going on as the big deflections get exaggerated more than the small ones.
Are we having fun yet?

Keep at it Peter

One thing I have not tested for but have contemplated is the weight of the machine. If you have separating contacts then you can create the machine feet and place them on a surface. Use a real friction co-eff and include gravity. Commercial large machines weigh tonnes and this must create some sort of restoration force. The feet pushing "down" would give the machine more support I expect. But I have not tested this theory so maybe you could...Even without gravity the feet pushing down (due to tool load) may contribute to the machine stiffness. I was going to try this the next time I was analysing a machine. Compare no feet to feet result to find out if this is significant.

see asme B5.54 for clues.

In B5-54 they use the machine to estimate the static stiffness of itself. So the loadpath is nearly exactly as you have used in your modelling. By having a vertical gap between the restraint and the load there is an extra moment created. Perhaps make the rigid block taller so the load line of the load and restraint agree? to remove this moment... I'll re read B5 for clues....

[jackjr-123]

Hi Peter, that makes sense, thanks for explaining.

Today I tried making a small batch (2kg) of UHPC with the standard E45 recipe. It was a complete failure.
I mixed by hand and with the help of a cordless drill, but after 15min not only I was very tired but the result looked like a bunch of sand. Not sure what I did wrong, everything was carefully weighted.
The concrete has been sitting for a year or two, maybe it was too old?

[jackjr-123]

I was finally able to pour a sample of Durcrete E80. Mixing this stuff by hand is a nightmare, even my electric handheld mixer was inadequate.
The sample is currently undergoing heat-treatment (48h @ 80°C). I'll try to make the stiffness tests this week.

[peteeng]

Hi Jack - I burnt out a 240V drill doing a similar mixture. Looking fwd to the test results. Peter