I've got everything mentally modelled and I've just got to slowly input the data into Fusion360. This is one of 4 parts, where the other 3 is the top cover for this part, the PCB + motor holder and the drive pulley.

It's a good thing I decided to see if the model I have right now is printable because I'm already having printing issues. 

Something useful that I didn't realise when I bought the small scale is that I can reuse the screws for Tetrinsic. So far I think only the LCD and diametric magnet needs adhesive.

As mentioned in a previous log (I think), it's cheaper to buy an entire mini scale than to buy the 300g or 750g load cell by itself. First I took the AAAs out of my 2Kg scale to test it and it indeed works:

Both scales agree on the same weight. Suprisingly, the cheaper scale looks to have a higher contrast LCD.

Well time to open it.

8 phillips head screws later and I got to the load cell, which seems to actually be rated for 1kg and has 2 4.15mm holes instead of M3 threads. Looking at the inside geometry, it seems that they used to be M3 threads and drilled out. The scale plate has 4.10mm stems that friction fit though and held in place by self tapping screws.

Another 5 small phillips screws later, and I can see the board:

The S+ and S- have a small C1 capacitor inbetween them, so I probably should also add that into my PCB design since a £1.50 scale is probably cutting as many corners as it can (thus, if the cap wasn't needed, it would have been ommited). As expected, the LCD contacts the board via pads. While transparent, the LCD is quite a bit darker than just clear acrylic or something, likely to do with polarisation. Here's it infront of the glossy white box:

The LED doesn't magically look really bright though. I've put this next to the glossy white box so you can see the exposure level. The diffuser film (that's over the clear plastic) is quite the dust/lint magnet, but it easily wipes off.

The LCD actually looks rather cool without a backlight:

Not sure when I got a fingerprint in there but I didn't want to break the LCD trying to rub it off. The LCD now looks like it's got some cool, techy 3D background. It's like a hacker edition precision scale.

As you can see, there's a tiny trace of threads in the 4mm holes. It's kind of like Fusion360's textured threads IRL. 

Anyway, the dimensions are still the same 6 x 9 x 45mm I hoped for. The thickest height is 7.3mm but I'd use 7.5mm in CAD. The wire relief on the side stops 4mm from the edge. Wires for V+, S+, S-, V- are red, blue, white and black respectively. Wire relief blob is an extra 1.6mm and the wires are 0.64mm diameter and around 10cm long.

It seems I forgot to add the project log where I mention that I had an idea to have magnets under the LCD so that they could be mounted on the #Tetent TimerSpy [gd0136] screen-down. There'd be another set of magnets such that the distance between both of them is only the screen thickness appart. The magnet calculator gave 2.3kg of pulling force from 2 sets of 30 x 3 x 5mm rectangle N35 magnets, which is likely more than enough considering I could probably fit those magnets in a single Tetrinsic. 

The result is that the design becomes much more streamlined, LCD protected and a pocket doesn't have the potential to catch on the ball chain or the edge of the cutout wall. I also might get a few more mm's for the PCB.

The LCD and steel tubes arrived yesterday. I forgot that I ordered those tubes, and this was before I decided to cut the load cell (which more or less eliminated the need for the tube).

These tubes are advertised for use with jewelry, but they're unreasonably precise for the price and quantity I paid for:

Maybe this specific tube was just a lucky pick. I didn't try any of the other 29.

The LCD looks small yet advanced. 

This changes my perspective a bit on how technologically advanced this input device probably is. I probably shouldn't expect anything less considering that the button-pressing keyboard has been mostly undisputed for over 40 years, giving engineers plenty of time to have obtained a faster solution.

When I realised that I only had 15 days remaining last log when I needed to have had a PCB ready to go a week prior, I knew the Feb 1st deadline wasn't going to be feasible and started focusing on other projects. I do feel better rested to tackle Tetrinsic now, and that might be because I've had a 2 week break after 4 weeks of straight R+D for this project, and I've moved other projects (namely #T^2 TyMist [gd0138]) forward after 4 weeks of stagnation. It really doesn't help that I'm doing the equivalent of 4 - 5 people's main hobby project as a singular individual (and #SecSavr Suspense [gd0105] would probably be a 3 team effort minimum:  a programmer, an engineer and a chemists/material researcher).

I'm still not quite sure what the project timeline is going to be. It's like those rare times when you're copying a large directory and midway though it finds more stuff to copy, so that the progress bar looks like 30% -> 40% -> 52% -> 39% -> 45% -> 22% -> ...

Chinese New Year is also coming up soon so that has the potential to push things back.

Solution mining... ends, in 0 days.
My itenerary is being recalculated.

Cheaper Load Cells

So I was watching this teardown video of scales because Detail is considering them for one of his projects and I was wondering if I should chance the 300g load cells for more precision or stick with the planned 750g. I thought "If I could get the same precision as those 500g scales, I'd be fine with it."

I see this and think "Hmm this looks awfully familiar."

Of course it is. It's the 750g load cell I'm planning to use right now. It seems they're typically installed into 500g pocket precision scales, which makes sense considering they'd usually also have to hold a plate.

Well I go onto AliExpress to see their prices and they're low -- low enough that there was a possibility I could find an entire pocket scale cheaper than buying the load cells alone.

This is the price to beat:

And this is the cheapest scale + shipping I could find after an hour of searching manually (aka no AliTools):

Now, this is a different scale. The 2 options of 200g and 500g make it sound like they use the 300g and 750g load cells (respectively) but I couldn't find a teardown.

I got AliTools into this to deepen my search, and I found this image:

"Oh cool. They took a picture of the battery compartement. Haven't seen that one yet. WAIT. ZOOM ZOOM ENHANCE!!"

That cutout... that cutout in the metal is TOTALLY the load cell, right? It seems to be the same length as the AAA batteries on each side.

Yeahhhh! This has totally got to be the load cell I'm after. This is great news. The specs of the load cell recommend 10V, but the guy in the video measured only 2.48V going to the load cell. There are loads of pocket scales claiming 0.1 and 0.01g of precision with only 3V from the batteries! I can only imagine that the 0.01g scales just have better electronics or a different LCD. Reviews seem positive for the precision of these bargain-bin scales.

I did see a video on the listing and the (similar looking, but ultimately different) scale took like 1000ms to go from 0->100, which implies that whatever cheap circuity is used for pocket scales (likely the very popular HX711) does not obtain enough samples per second to be usable for this application. Remember that I need to be able to get a reading fast enough to generate the haptic feedback of a finger going through multiple actuation-force levels.

Another good thing to note is that the cell is able to get such precision even while mounted to cheap -- and somewhat thin looking -- plastic. 

I have noticed that it's possible to get the 200g scales slightly cheaper than the 500g ones, which is probably perfectly fine for data-entry use cases. The max overload should be approx 900g anyway. I'll just play it safe with the 500g scales.

Cheaper chips

I put the new load cell source in the BOM spreadsheet and now the ADC sticks out as the most expensive single component. I found the same ADC on AliExpress from a new store (they only have 3 folowers) that sells a variety of chips:

After VAT, it's around £32 which is decently cheaper than the £47 from DigiKey. £15 savings for 10 mins of work? Now that's more like it. Can I get more AliExpress savings?

This saves £5.

And that's about it. So total savings are just under £28. That's an entire Tetrinsic with change right there.

Motor shortlist

I'm going to hold off buying all the motors as I still don't yet know if the one I've got is strong enough for the job. In the case that it's not strong enough (or I just want to save some BOM money), these are the motors I've found:

Wait...

Pardon me? Free savings! That's another £12 down!

Anyway, as I was saying. The £1.70 one is the smallest and has an approx 1500KV. Then there's the 2010 and 2015 motors. These are the only one's I've found so far with torque data:

It's quite suprising that the motor that's 5mm shorter has 1/9th the torque. The larger one has a 940KV. The 50p listing is kind of suspicious because the title and listing all describe the 2015 model, yet all the product images are the 2010 model. 

Those readings are concerning.

This is the torque converted to grams. It means that, if the radius of the chain pulley was 10mm, the force exerted would be 174.3g. The actual pulley radius is likely 5mm, so the exerted force is about 350g. That's plenty to stop all fingers in their literal tracks and give the thumbs a serious workout. 1/9th of that, or 39g is going to be a "y'all feel smthn?" from literally all fingers. The question is where all the other motors fall within those minmax values.

I think stator size should be a good indication of how strong a motor is. The 2015 motor has a 16.6mm diameter, 7mm thick stator. The 580KV I've just opened now is 13.1mm by 7mm. This site says "stator volume is a factor of motor torque assuming all other variables being equal" and "narrower stator width, [resulting] in a smaller rotational inertia that allows for faster RPM changes". That leads me to believe:

350 * (13.1/16.6)^2 = 218.0g

 218g should be decent enough for science, right?

gets scale out and mounts it vertically

ehh... my fingers are actually hovering around that 220g-and-up ballpark. 240 seems to be the average. 

Maybe I should go back to my OG strat of dual axis strain guages. Having to create some 2-axis resultant vector just doesn't sound like the greatest idea though, even if the current Tetent layout is already designed to always go back to the E column. Nah, cancel that idea; traditional gaming wouldn't be possible. I'm trying a fixed chain and it doesn't feel that great either.

I'm looking at the datasheets of "high dedent force encoders" and they're 210 +/- 70gf, so the 580KV motor might be good enough for science.

Conclusion

The idea of being able to tweak actuation forces down to a tenth of a gram is just not something your standard input device is able to do. Exciting! Someone could have a <10g or >100g "switch" and be able to effortlessly switch between them. The only real questions I've got is "will the BLDC motor be able to create the haptic responses I want?" and "Is the BLDC even strong enough for decent finger resistance?"

Lastly, I've noticed that the proposed chain path and PCB is awfully similar to what I was planning for Tetrinsic 2.0, so I'm renaming Tetrinsic Concept4 to Tetrinsic Gen 2X2. 2 x 2 = 4 anyway.

Solution mining... ends, in 15 days.

It's possible that the thumbs would use joysticks instead of Tetrinsics. 

In the latest TestCut concept, the current thumb Tetrinsics are in the best ergonomic position and yet they still don't comfortably want to move whilst Finger2-4 are in motion, due to mental overhead and feeling like a muscle is about to cramp. With TimerSpy, I don't think a Tetrinsic for the thumb would even fit in the space available. 

Joysticks would be useful for non-typing tasks such as mouse movement or gaming, they're cheaper and takes up less space.

Solution mining... ends, in 18 days.

[17:30] Sorry, but the motor sticking out of TestCut is an eyesore. I started looking into drawing up the new PCB, viewing the AS2040 for reference since I found it through EasyEDA and without the header pins, seems close to the target Tetrinsic PCB size. This then lead into considering if I want to solder the LCD directly or use a ribbon cable extension. Then I was wondering if I should include MemoryLCD support too (for TimerSpy) or have something off-the-shelf.

Eventually, this turned into trying to remine for a solution that didn't have the motor sticking out from the top. It doesn't look good in TestCut and I imagine it wouldn't look that great on the back of my hand either for TimerSpy.

I've finally found my calipers, and the bolt holding the stator and rotor together has a thickness of 1.33mm, so I'd imagine this means that it's an M1.4 bolt. There's no standoffs for that, so I'd likely have to use a long bolt and a steel tube.

Seems that M1.4x16mm is the longest I can get on ebay, which should be good enough for... this run:

It's suprisingly an improvement on all metrics, with the catch that the front slope is 10mm high instead of 6.5 or so. Not sure how Me In The Future is going to be doing the sliding for TimerSpy.

I've printed out the shape envelope (I should've done this for the previous design too but I didn't) and it's kind of tight in pockets, so the solution could still fail. The main issue is the initial slope edges for this 60mm (approx 3 Tetrinsics) print, and that the back pushes on my hand the most. This makes me believe that the concept with the motor sticking out would have fared even worse.

I've tried this too and it's still kind of problematic. Certainly looks cooler on my hand though. That's also a problem, because I've now rejected the solution that has a motor that sticks out. If I can't successfully mine this solution, I'd once again be solutionless.

Increasing the length to decrease the 60degree slope height to 6mm is likely to do more harm than good... And then there's also the question of the ribbon cable for the screen.

I also got this bolt location dimension wrong.

[19:30] I'll try this, and continue mining for a solution if the print turns out to be successful:

I've also filleted-chamfered the edges:

[19:40]

An ergonomic solution...
may exist.

 Me:

Okay... now I need a solution that physically works. There's no encoder in the sketch.

Doesn't seem like that would work. I don't get that much more space for the encoder and the fact that there's a free gap in the latest print (see below) seems to help with ergonomics when the wrist is bent.

It certainly helps with the ergonomic aesthetic at least.

Let's consult the datasheet for advice.

Nothing in the realms of off-axis technology?

Seems like 60mT is ideal.

So 600 gauss.

This is a square 6x6x2.5mm diametric square magnet (because a circle isn't an option) and it seems very overpowered for the job. It looks to be >6mm distance from the magnet before reaching a workable location. It seems that something even smaller than the 3mm OD x 2mm thickness radial magnets I got for Tetwin would be closer to the ideal:

[21:30] Ok, so now that I know that I can use a much smaller magnet, I've got an idea to actually have a cogwheel 90 degrees to the motor-driven cogwheel. The ball-chain would act as the teeth.

Additionally, I think a shallower initial slope angle of 50 degrees would help instead of 60 degrees. 

The LCD extension cable will have to do quite a bit of origami gymnastics. Actually, I should check to make sure I can even get a cable of the same pitch.

2.54? That's awfully generous. Too generous...

Something like this should be fine. 

[21:55] Ok print finished and it doesn't seem like 50 degrees did all that much other than poke into Finger5. It seems that 75mm total length is a bit iffy. Yeah, I really should've printed out the second Gen3X1 concept before going into TestCut because it probably would've failed.

If I get a file, I probably could get another few mm's in...

[20 minutes later...]

That should be manufacturable.

[22:50] I printed it and 

An ergonomic solution...
may exist.

It visibly looks small.

Not a fan of the fillets though... 

Filleted chamfers or nothing is going to have to do.

It's a new PCB method so it's essentially a whole new concept now.

It's taken 6.5 hours to get to this point. If I average 1h/day after that Feb 1st deadline, it means that the progress I'd usually manage in a day will then take a week. Ouch.

Solution mining... ends, in 19 days.

[13 Jan, 05:38] Render:

It actually looks like something's missing when there's no motor sticking out.