[suspension]

Hi,
remember that the design I analyzed was plated top and bottom, much increasing the column torsional rigidity which in turn increases the rigidity in the X direction
Remember also that I have two very stiff cast iron axis beds bolted to the frame which go a long way to simulate the plating effect, namely increasing the torsional rigidity.
The two side plates are locked so that they cannot shear relative to each other and vastly increases the stiffness.

Craig

Hi Craig

I updated the design as below and now X stiffness is 250! (Y - 91, Z - 500). But I doubt this is economical, so need to reduce thicknesses, but I guess now I am in the correct path.


Thanks
Sus

[peteeng]

Hi Sus - I have looked at your model:
1) You have all the X,Y and Z loads in the same study. They need to be in separate studies. Is this how you ran the model? and calculated the compliance?
2) I changed the mesh to 20mm as I find the default mesh settings a bit coarse. This should not change the deflection results much
3) Using the 1000N load at the small hole and constraining the table I got the following

X 36N/'um Y 400N/um and z 454N/um We can address the X result, The others are spectacular.. But then you have 3.6 tonne of steel in the design.

So if you look at the X image you can see the column is lozenging (and the base as well) as the section is open at the front and top. If you closed the front-top I expect you will be way up in the stiffness numbers. Thats because the front and back provide shear stiffness in this direction. Also the connection between the outreach and the column is via a small vertical part of the structure. It needs to be better connected to provide better global load transfer, again the face plate will do that. I expect adding the face plate will put the X way up there in the 100's.

What stiffness/compliance are you aiming at? Peter

edit - internal webs - unless bolted these are difficult to include in practice. If the walls are thick they will not be needed. So suppress them in a model and compare the unribbed and ribbed results. They are secondary structures, they keep the outside plates in shape so they keep their efficiency. You mentioned noise so if you do use webs do not place them at equal intervals as this creates a natural amplifier at certain frequencies. They need to be spaced unevenly so the harmonics are suppressed.

[suspension]

Hi Sus - I have looked at your model:
1) You have all the X,Y and Z loads in the same study. They need to be in separate studies. Is this how you ran the model? and calculated the compliance?
2) I changed the mesh to 20mm as I find the default mesh settings a bit coarse. This should not change the deflection results much
3) Using the 1000N load at the small hole and constraining the table I got the following

X 36N/'um Y 400N/um and z 454N/um We can address the X result, The others are spectacular.. But then you have 3.6 tonne of steel in the design.

So if you look at the X image you can see the column is lozenging (and the base as well) as the section is open at the front and top. If you closed the front-top I expect you will be way up in the stiffness numbers. Thats because the front and back provide shear stiffness in this direction. Also the connection between the outreach and the column is via a small vertical part of the structure. It needs to be better connected to provide better global load transfer, again the face plate will do that. I expect adding the face plate will put the X way up there in the 100's.

What stiffness/compliance are you aiming at? Peter

Hi Pete

Yes I always kept all three load cases inside a single study. I hoped they will be evaluated in isolation in a single simulation run. Otherwise for each axis, I need to run the simulation 3 times.
I think what you propose to increase X stiffness is already included in my latest model and the jump was massive and it is 200+ now. Since the amount of steel in this model is too much as you highlighted, I am in the process of reducing it now.

Thanks
Sus

[suspension]

Did few more simulations, but it looks like getting the required stiffness (ideally 130-150 in each direction) with steel will cost two times the UHPC path (assuming no stress reliving) unless I can efficiently put mass at right places (still keeping the structure simple). I wonder if topology optimizations can help. For example, I can set the position of table and spindle at a particular position and then optimize the topology of the frame to understand where the load path is. This assumes that this path does not drastically change when table and spindle are moved however.

[peteeng]

Hi Sus - Aiming at 130N/um is in commercial stiffness territory and will take tonnes of anything to achieve. The X dirn will always be the tough one if you only want to run once. This is because its a combined load on the column. It is bending it and torqueing it at the same time. The Y&Z are only bending which can be optimised for easily. The C or G frame geometry is the limiting factor. If you made a gantry type machine with high rails you will immediately improve stiffness yet reduce weight, Since your happy to cast this is the next step I think. Peter

[joeavaerage]

Hi,
when I did this about three-four years ago it was the first time I'd really attempted to use Fusion FEA. What I did was break the design down into smaller sections, analyse each section and then
add all the deflections together to get a result.

Since then my experience with Fusion FEA has improved.

Couple of notes on the model. The sides are 32mm steel. The top and bottom are 10mm. The webs (arranged as I previously pictured) are 20mm. The model is one single piece, that is to say there
is no allowance for the webs being bolted nor the plating being welded. You might ask what are the vertical upstand on the base, and what is the horizontal section protruding from the column.
They are a solid 240mm x 240mm steel section to represent the Y and X axis stack, and the Z axis, saddle and headstock respectively. They would ideally be modelled as a separate perfectly stiff component,
so that the deflections measured are the frame itself, not any contribution from the axes....but I'm not sure how to model that. As an approximation I assumed a solid steel extension to the model.

As Peter has said the X axis load represents bending (side to side) AND torsion of the column. The peak deflection under 100N load is 13um, but actually at the intersection of the Z axis with the column it is about 7um.

The deflection under 100N Y axis load is straight bending (for and aft) and at the Z axis intersection of the column it is about 9um.

The deflection under 100N Z axis load is straight bending (fore and aft) and at the Z axis intersection of the column is about 5um.

I have yet to model the effect of the Y and Z axes being bolted (at four corners) to the frame. I think I would have to model a bolted connection as a separate part.

Craig

[peteeng]

Hi Sus et al - The approach I now use is the following:
1) Ignore material initially as the stiffness ratio is only 200/50=4x ie steel vs concrete. The concrete flexural modulus value could be <E30 and needs to be proved by test
2) Model the machine from the spindle outwards. Include rails and cars in the model. Make all parts "blocks" of solid material. Maximise the size of every block as geometry is your friend. Big geometry is the key to big stiffness. If you use closed loops you can see that it has an 8x advantage vs open loop structures
3) Once you have a block model that has the right working envelope and maxed out sized parts, use steel or any material value to run your first stiffness run. Steel will be your top stiffness value, concrete will be 1/4 of that etc. See where you are relative to your target value
4) You should be well above your target value if not, check thru the model. Use bonded connections everywhere
5) Once you think the model is good and meets the working envelope, start shelling the parts and see how the stiffness goes. Initially with thick parts it won't change much but at some point the model will get soft. This is called a "trade study" or trade off study. Its to get a feel for the result.

That's when you can start thinking about materials. If the parts are solid then you have to cast. If they can be thin yet need to be stiff then they can be steel etc By exaggerating the deflection plots you can find where internal secondary structures need to be. For those with GD apply100N and tell it to deflect max 0.001mm (if your target is 100N/um) and see what happens. In the X loadcase you need to specify a symmetric Y plane otherwise the optimum will be a skewed structure. The X load can be both directions that's why you state it needs to be symmetric. Good Luck Peter

[suspension]

Hi,
when I did this about three-four years ago it was the first time I'd really attempted to use Fusion FEA. What I did was break the design down into smaller sections, analyse each section and then
add all the deflections together to get a result.

Since then my experience with Fusion FEA has improved.

Couple of notes on the model. The sides are 32mm steel. The top and bottom are 10mm. The webs (arranged as I previously pictured) are 20mm. The model is one single piece, that is to say there
is no allowance for the webs being bolted nor the plating being welded. You might ask what are the vertical upstand on the base, and what is the horizontal section protruding from the column.
They are a solid 240mm x 240mm steel section to represent the Y and X axis stack, and the Z axis, saddle and headstock respectively. They would ideally be modelled as a separate perfectly stiff component,
so that the deflections measured are the frame itself, not any contribution from the axes....but I'm not sure how to model that. As an approximation I assumed a solid steel extension to the model.

As Peter has said the X axis load represents bending (side to side) AND torsion of the column. The peak deflection under 100N load is 13um, but actually at the intersection of the Z axis with the column it is about 7um.

The deflection under 100N Y axis load is straight bending (for and aft) and at the Z axis intersection of the column it is about 9um.

The deflection under 100N Z axis load is straight bending (fore and aft) and at the Z axis intersection of the column is about 5um.

I have yet to model the effect of the Y and Z axes being bolted (at four corners) to the frame. I think I would have to model a bolted connection as a separate part.

Craig

Hi Craig,

With these readings, isn't the stiffness you get very low? Like 7.6, 11.1, 20 in X, Y, Z?

Thanks
Sus

[joeavaerage]

Hi,

With these readings, isn't the stiffness you get very low? Like 7.6, 11.1, 20 in X, Y, Z?

Well 6, 9 and 5 actually, but notwithstanding the numbers, yes, that is correct. When I measure it I end up with higher, about double or treble. Why there is such a disparity I don't know.
The point being this is what you might expect of a fabricated steel column that uses a little over 200kg of steel. If you want stiffer then use bigger sections. For instance the overall width of
the 'L' is 240mm and the height is 150mm. Your model is 300mm x 300mm and just the geometric increase
in stiffness is probably eight fold....but then you used three tons of steel (by peteengs estimate) whereas I've used 206kg.

If your target is 130N/um to 150N/um get used to the idea of a column and base of several tons. If you look at any industrial machines of that type, the VF1 for example is 4tons. You can be sure if Hass thought
they could make it lighter and still be stiff enough they would have done so.

The simple fact is that stiff machine REQUIRES large geometry to get the required stiffness and that in turn requires mass and therefore cost. Fact of life.

Craig

[peteeng]

Hi Sus - "Low" is relative statement. A small benchtop mill maybe 0.5 to 1.5N/um. On the English cnc site there are a couple of people who have measured theirs and they don't get past 2N/um. There's a middle size Chinese machine on the forum that's been measured at 8N/um from memory. I have articles that measure commercial machines at 150N/um... As you add detail to the model, rails cars, ballscrews etc your stiffness will drop. So keep at it and see where the figures get to. Geometry is the answer and closed structures, they overtake material stiffness pretty quick. Material stiffness is sort of the bonus at the end of the game... Peter

when you actually measure a machine you need to apply a preload to the spindle to take out the mechanical play. There will be some. Thats where your zero point should be. But don't use the data directly, plot it and calculate its slope using 3 or 4 loads, that's the accurate way to remove inconsistencies, hysteresis and other stuff...

[suspension]

Hi Craig and Pete

Yes, I understand that in order to get 100+ stiffness I need lots of mass. When I started simulating steel plate designs very long time back, that is what I also figured. The reason I re-evaluated steel plate path is due to Craig's comment earlier here where he mentioned that stiffnesses of 125 in Y/Z and 90 in X was achieved with much less mass.

Thanks
Sus

[peteeng]

Hi Sus - Its not about "mass" - its about geometry. Machine parts are generally in bending so its bending stiffness you need. This is a function of its geometric inertia. So maximising every parts second moment of inertia is paramount. It is possible to make parts stiffer yet lighter then its benchmark solid part. That's because the central "mass" of the part does not contribute much inertia. That's how I beams and hollow beams work.... Peter