LOG.md 16.4 KB
Newer Older
Jake Read's avatar
Jake Read committed
1 2 3 4

## 2019 11

Jake Read's avatar
Jake Read committed
5 6

Jake Read's avatar
Jake Read committed
7 8 9 10 11 12 13 14 15 16 17
### Vision Controller Hello-World

We're currently working to build a computer vision based displacement sensing method for the DEX. Since our machine (or, many machines manufactured by novices / in the public domain) are liable to flex (indeed, nothing is infinitely stiff!), the thought is to measure local displacements of the sample, at the sample, rather than measuring open-loop through the machine's structure.

To spin this up, I've written a small / barebones subpixel template tracker in the browser, in cuttlefish. This is conveniently lightweight - the whole cycle (image collection -> analysis) can happen at ~ 10Hz, which is not splendid, but not diabolical either.


Here we can see a desktop test - I am reading the X- position from my tracking system on to a chart, and moving the tracker on a linear stage. This system resolves ~ 15um, which is not bad for a proof of concept.

## 2019 10
Jake Read's avatar
Jake Read committed
18 19 20 21 22

Re-did the machine last week, now much simpler:


Jake Read's avatar
Jake Read committed
23 24 25 26 27 28 29 30 31 32 33
## Comparison to Instron 4411 (2019 10)

To see how we do against a real instron, I tested identical samples on the DEX as well as on an Instron '4411' with a 5kN load cell. In the plot below (I'm using cuttlefish to plot the .csv that I saved from Bluehill, the Instron software), the leftmost plot is taken on the 4411, and the lazier slope belongs to the DEX.

While the samples fail around the same load, the difference in elongation is ~ 1.5mm wide: this is almost surely the machine's own deflection, stretch in the belts, etc.


This obviously warrants correction. One way to do this is to build a stiffer machine, however, we will be chasing up the cost and complexity if we do this. Rather, we should throw some more control at it. To start, we can circle back to our attempts at [subpixel tracking](https://gitlab.cba.mit.edu/calischs/subpixel_tracking), or attach a small linear stage directly to our fixturing elements. For this, I am imagining something like the [AMS5311](https://ams.com/as5311), which should do some 12 bits inside of a 2mm throw (for 0.4um resolution). Either can be added to existing systems, given network controllers / modular browser code. Since I want to integrate it elsewhere, it's likely that the camera option comes first.

**update** the new machine uses a ballscrew transmission, not a belt (as used here) which should elminate most of the creep tested here against the 4411. forward progress has been made with a vision based controller, to improve further, but has not been characterized yet.
Jake Read's avatar
Jake Read committed
34 35 36 37 38 39 40 41 42 43

#### Some Force Maths (and a .xlsx file)

 - want (?) from ex. nylon dogbone
 - for this, w/ motor doing 1.5NM, how much reduction ?

 Have this spreadsheet already, that's great. The D683 dogbone is ~ 8mm wide, and it would be great to test parts at 4 layers of 0.2mm - 0.8mm thick. That's skinny, but here we are: these are all under ~ 500N. OK.

 For that oomph, with a 22T drive pinion, I'll want a 6:1 reduction and a NEMA23 motor (just a short stack) pushing 1.26NM through to a max. 660N linear force. OK, glad to have checked, nice to know about the N23 - I had been thinking of N17s - and will have to watch about getting that 6:1.

Jake Read's avatar
Jake Read committed
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
## uSSM #3


Theres a few new criteria that ussm-3 aims to achieve:
- Manufactured only using simple shop tools, PLA 3d printers, and a standard bed size (24'x 24') laser cutter.
- Pulling force of around 600N

In order to be laser cuttable, acryllic and delrin were the main two materials considered. Delrin was the go to option to avoid shattering.

This design uses mainly the following adapted gantry system: [gantry](https://gitlab.cba.mit.edu/jakeread/rctgantries) along with this beam design: [beam](https://gitlab.cba.mit.edu/jakeread/rctgantries/tree/master/beams) for rigidity.

The main idea is to attach a beam to the linear axis and use it as a carriage to hold fixturing for tensile testing on the top half and compression testing in the bottom half. The gantry system shown above does not supply enough torque, so an adapted version was made shown below.


The uSSM #3 design takes advantage of the beam design by attaching 4 different beams to a larger "O-face", with webs in the corners to prevent torsion. The larger face itself is split into a bottom and top in order to make assembly easier, make each piece smaller to fit into laser cutter bed, and to have the possibility of changing top for larger specimens. The beam design also utilizes custom joinery talked about in the [beam](https://gitlab.cba.mit.edu/jakeread/rctgantries/tree/master/beams) repository. This makes the machine take some time to build, but it allows delrin sheets to be used, allowing many fabs with laser cutters to be able to make this machine.


### fixturing
Fixturing is being designed to use mainly 3D printed parts. First up, PLA is being in the jaws to see if it can withstand the 600N load alongside a few other McMaster parts and this [load cell](https://www.amazon.com/Pressure-Force-S-type-Sensor-Cable/dp/B01F6IOWDG/)
![u3_fixturing]( images/USSM3_fixturing.jpg)

## Drawingboard Return

With \#2 feeling somewhat unloved ('both overdesigned and underdesigned'), I'm back at the basics for \#3. There's a few major selections, and decisions to make:

### Material Selection

-> ALU, FR4(G10), Acrylic?

While aluminum is my go-to for machine design, and is ostensibly possible to mill on a shopbot by a motivated user (see [jens](https://github.com/fellesverkstedet/fabricatable-machines/wiki/Fabricatable-axis)), there is some hesitation to use it.

| Material | Young's Modulus (GPA) | Specific Young's | Cost for 6mm x 24x24" | Machinability |
| --- | ---: | ---: | --- | --- |
| ABS | 2 | ? | 52 | Not Dimensionally Stable, but OK to Machine |
| Nylon 6 | 3 | 2.5 | 130 | Painful |
| HDPE | 1 | ? | 23 | Easy |
| Acetal (Delrin) | 2.8 | ? | 89 | Breezy, also lasers, and non-cracking |
| Cast Acrylic | 3.3 | 2.8 | 46 | Breezy, esp. w/ Lasers |
| 6061 ALU | 69 | 22 | 87 | Breezy with WJ, Painful on Shopbot |
| FR1/CE (Canvas / Phenolic) | 6 | ? | 81 | TBD, probably WJ Pain and Ease on SB |
| FR4/G10 (Fiberglass) | 22 | ? | 98 | Painful on a WJ, Slightly Easier on a Shopbot |


That said, ALU lands pretty well 1 order of magnitude above Canvas Phenolic ('FR1' or 'CE') for strength, while costing a similar amount of dollars. Fiberglass is a nice candidate, so machining G10 is likely a worthwhile experiment. However, both composites have anisotropic-ness and are sensitive to the size of local features (and to localized loads), making them less favourable.

!TODO: beam equations for the above, to size req' depth
!TODO: shear forces for the same,
!TODO: cost not-mcmaster, and acrylic, some composite like hydrostone w/ fiber
 - ... composite vendors at fabclass website

### Transmission Design

#### How Many kNs ?

We want lots of force, with very fine control of position. This means a nice linear transmission. To estimate the forces we might want to see, I wrote a quick table of forces required to rip apart ~ 3mm square (0.001mm^2) samples of a few materials.

| Material | Yield (MPA) | F at Break (N) |
| --- | ---: | ---: |
| ABS | 40 | 360 |
| Nylon 6 | 70 | 630 |
| HDPE | 15 | 135 |
| 6061 ALU | 310 | 2790 |
| 4140 Stainless | 655 | 5895 |
| 6AL-4V Ti | 950 | 8550 |

Brinell hardness tests range from 10N through to 30kN (for steel and cast iron) but non-ferrous materials normally see 5kN only.

So, a ballpark of ~ 10kN would be ideal - this is a big number - off the bat I'm going to estimate that 5kN will be a more reasonable target. 1kN is enough for a complete set of plastics, but that's only allowing for a realtively small sample.

-> Ballscrews, Belt Rack and Pinion, Rack and Pinion

Generating kNs of force is no easy feat, especially when we want to do it *very smoothly* while displacing very small amounts.

I will start by mentioning that this is dead easy with ballscrews. With a 1605 ballscrew, (16mm diameter, 5mm per turn) and a NEMA23 with 3Nm of torque, we can generate about 3kN of linear force (per motor) - to land at 5kN total no problem.

However, these are somewhat cumbersome and expensive - and they land in fixed sizes. Towards more parametric machines, we can look at a rack and pinion type axis.

[ballscrew maths](http://www.nookindustries.com/LinearLibraryItem/Ballscrew_Torque_Calculations)

Because tooth geometry very sensitively affects linear-ness of drive, especially where (down below the mm) we will be driving an entire instron test-cycle inside of one tooth-phase, I want to discount a traditional rack and pinion right off.

I am curious about a belt-driven rack, similar to [this design](https://gitlab.cba.mit.edu/jakeread/rctgantries/tree/master/n17_linearPinion).

!TODO: compare by transmission ratios (abstract from motor) and cost of parts.
!TODO: belt spec for hight (huge) load belts: tooth shear, and stiffnesses.

#### Motor Torques

To generate the force required, we're going to need some motor / transmission oomph. Here's a list of typical NEMA size motors, and the torques they can generate. The [atkstepper](https://gitlab.cba.mit.edu/jakeread/atkstepper23) can supply enough current to power any of these.

| Motor | kg | Nm |
| --- | ---: | ---:|
| N23 56mm | 1.2 | 1.3 |
| N23 100mm | 1.8 | 3.0 |
| N34 68mm | 1.8 | 3.4 |
| N34 156mm | 5.4 | 13.0 |
| N42 150mm | 10.9 | 22.0 |

### Racks and Pinions

We'll be using two linear stages (one on either side of the platform), so, from our N34 156mm motor with 13Nm of torque we'll need 26, 5.2 and 2.6mm lever arms respectively for 1, 5 and 10kN total force.

Considering practical limits on pinion diameter (with a shaft of 14mm, and 19 3mm-pitch teeth, we'll have an 18mm diameter pinion - 9mm lever arm) we will only realistically achieve 1.44kN of linear force per motor with direct-drive rack and pinion on a Nema 34 size motor. This makes a 3kN machine, but to add some safety factor we're at at 2kN goal with this approach.

The next step would be to check against tooth shear stress for 3mm pitch.

From reasonable WEDM (time) and Waterjet (washout) limits, I expect the thickness of any fabricated pinion to have a limit around 12mm. To make this all simple, I'll say 10mm. For the waterjet, this is a bit of a stretch (get it?)...

To guess at resolutions, we'll take our sample above in aluminium having a length of 100mm - with elongation at break being 12%, we're interested in a 12mm 'long' stress / strain curve. For 1000 pts on this curve, we're interested in a step size of 12 microns.

From our 18mm diameter pinion having a circumference of 56mm, this means 4667 counts per revolution. In a 200-step motor, we would need 32 microsteps per revolution - most drivers will go to 256 - but microstepping isn't exactly linear. To do this really well, we will want to finish work on closed loop stepper driving, where we can use a 14-bit encoder to control around ~ 4096 counts reliably.

All in, a direct rack and pinion drive can land us at a 2kN machine with some desired resolution, but we're at or near most of the limits here.

### Belts


Since I was spinning this up to test on a belt-driven gantry system, I was able to confirm that this is a bad idea. We defeat the elastic stretch of the belt by measuring at the specimen (not through the machine's structural loop) but a meagre GT2 belt profile skips teeth pretty quickly as we approach any tens of kilos of load.

### Ball or Leadscrews

Racks aside, a ballscrew is the obvious way to do this. Ballscrews can be had for less than hundereds of dollars. Besides transmitting motion smoothly, their ratios are favourable. For example, with a 5mm pitch ballscrew having an efficiency of 85%, we can drive 14kN with our 13Nm motor - so 28kN for the machine.

For the same ballscrew, to achieve 12 micron resolution we'll only need 416 steps in each rotation - this is easy to get.

## Design Spec / Notes

The goal here is to design and build a machine, which can be fabricated in $250k size fab labs, that can generate stress-strain curves for a wide set of materials as well as perform hardness testing.

To ballpark, I'd like to see 100mm diameter plates having a total travel from 0mm separation -> 500mm, this leaves enough room for fixturing etc.

Sam's note: the ballscrews should be 'pulling' in all cases. Against their fixed, driven side. This is a good note, thank you sam.

### Phoning Home

Samples are ~ < 1mm thick, and data from Materiom shows loads all falling well under 100N. IMO, the machine should be spec'd somewhere positive of 500N pulling force, so that we can reasonably pull apart big plastic samples. That said, 500N < 5kN.

Down here, once-reduced belt-and-pinion rack drives feel good. Since I'm looking for a medium high-force actuator to solve a 3D Printing-Bed problem elsewhere, I'm going to try to combine work here again, and develop a 3d-printing + laser-cutter friendly high force linear system. Conservative spec will allow me to even do this with N17 motors,


## Step Two: Force Measurement

For load measurement, Sam has worked through a great [loadcell](https://gitlab.cba.mit.edu/calischs/loadcell) design.


Ok, doing this with a wheatstone-bridge type load cell now.

I believe I want my ADC to have the green line 'on top' and white on the bottom, of a differential channel.


## Fixturing


! important


# uSSM \#2


**status** machine is designed, fabricated, waiting for adaption of fixturing from 'the generalist' and for squid works controllers.

## Dedicated uSSM

Here I'm using some
- Ballscrews, SFU2005 C7 700mm Long [ebay, $150](https://www.ebay.com/itm/SFU2005-Rolled-Ballscrew-C7-Anti-Backlash-Ballnut-L300-350-400-450-500-600-1000/202274175696?hash=item2f187afad0:m:m2SEqt9q_ykYYkrO5ZM9deA:rk:1:pf:0)
- Nema 23 Motors w/ .25" Shaft
- GT2 5mm 6.35mm Bore Pulley 12T for 15mm Belt 'A 6A55-012DF1508'
- GT2 5mm 12mm Bore Pulley 48T for 15mm Belt 'A 6A55M048NF1512'
- GT2 5mm 15mm Belt with 55T 'A 6R55M055150'

- http://shop.sdp-si.com/catalog/product/?id=A_6R55M055150 55T 15mm Closed GT2 Belt
- http://shop.sdp-si.com/catalog/product/?id=A_6A55-012DF1508 12T 0.25" bore Pulley 15mm
- http://shop.sdp-si.com/catalog/product/?id=A_6A55M048NF1512 48T 12mm Bore
- http://shop.sdp-si.com/catalog/product/?id=A_6A55M044NF1512 44T
- http://shop.sdp-si.com/catalog/product/?id=A_6A55M050NF1512 50T

## FEA

I ran a quick simulation to see that this flexure was OK. Here it is loaded in the direction it is meant to be stiff in, with 2.5kN applied vertically. There are two of these members being tested so this would be equivalent to 5kN of force, about where I expect this thing to top out. Displacing about 40uM (I'm looking at the face-to-face distance between the loaded zone and the fixed zone, not that rotation that appears from my non-rigorous fea constraints) with 70MPa stress maximum.


And out of plane, with 250N applied, about 20uM displacement. So 1/10th the load and 1/2 of the displacement, it's at least 5x stiffer in one DOF than the other. Is that grounds for a decent flexure?



# uSSM \#1: Generalist does uSSM

To test requisite parts for a small stress and strain machine, I attached some jaws and a load cell to [this 'generalist' machine](https://gitlab.cba.mit.edu/jakeread/mothermother).


I used [automatakit](https://gitlab.cba.mit.edu/jakeread/automatakit) network controllers and [atkapi](https://gitlab.cba.mit.edu/jakeread/atkapi) to program the beginnings of a materials testing system: here I'm just stepping the axis along 25 steps at a time and capturing a photo on each step. No data yet.



We're going to try to track the endpoints with CV, rather than fancy encoders etc. Sam has done some prior work on this, [here](https://gitlab.cba.mit.edu/calischs/subpixel_tracking).


# Microphone Stiffness Testing

[Alysia did this in FabAcademy](http://archive.fabacademy.org/2017/fablabsantiago/students/356/project.html) and it would probably be a fun piece of kit / example for the NIST Project. Looks pretty simple... and compelling.