CMM (Coordinate-measuring machine)

I've been going through the page listing ideas for future tool heads, and am surprised to only see output related ideas. so, what about an input device?

what i mean is, what about a tool head that is capable of gathering exact measurements of a object on the reprap bed; or that is capable of creating a solid model of said object.

the tool head consists of a five way switch(north, south, east, west, and down), that when triggered, tells the machine when it has made contact with an object. the machine then records the location. at my work, we have three of these machines. two are programmable and follow a general vector until it finds an object; at which point it follows the next vector programed. the other machine is lead by the tool head, by hand, to measure parts.

more information on CMM's can be found here: [en.wikipedia.org]

so i guess my question is, does the Darwin system allow for input information? i mean, when it comes down to it, the i, j, and k positions must be known for the system to be able to lay down material. all that is left is for the machine to know when to record these position (and a way to control the tool head).



if you need a reason for creating such a toolhead, think about quality controol. very few production processes produce entirely acurate parts each time a part is produced. or how about reverse engineering? wouldn't it be nice to be able to generate a solid model from some object you already have? no need to use height gages, calipers or radus gages to gather data about a part, to make a solid model.

this should also allow a repraper to throw a part into the machine, measure it, and then take the data gathered and make a copy of it.


sorry about the length of this post.

Hi fifo2711,

... we had some talk about measuring the tooltip-position with tactile sensors or fixed cameras and such - maybe i'll find some of the threads later ...

Viktor

I can think of one reason... it's tricky! You need very sensitive switches and a precise cartesian robot to get accurate results, I'd have thought. It doesn't *obviously* have any advantages over the optical 3d scanner either, which is mechanically much simpler and more tolerant of an imprecise setup.

Both mechanisms are going to generate loads of surface point data which is going to require the same sort of software to assemble into a solid object.

Probes seem more useful (to me) as a way of locating objects on the fabricator bed and obtaining their absolute position prior to attacking them with a toolhead.

Greetings fifo, et al,

I think the CMM idea has merit. Whether an optical approach is better than a contact/mechanical one is a lesser issue. Interestingly, the previous post hints at one key use of a CMM head for reprap -- improving on the accuracy of the current mechanism. Using a CMM head on a reprap to measure the shapes of reference objects (whose true shape is known ahead of time) could be used to map the warpage/non-orthogonality of the mechanism. The repeatable part of that warpage could be compensated for by software. This would let us get higher accuracy out of the current design.

Compared to even hobby-grade machine tools, the reprap mechanism (IMHO) leaves something to be desired. It's workable for the low-force additive process reprap uses, and the simplicity of the mechanism is key to self replication. However, the accuracy and repeatability seem to be constraints on performance, as is the de-facto limit to a low-force fabrication process. (Aside: it appears to me that one reason for Nophead's success, in addition to his industry and cleverness, is his use of a sturdy, accurate, nicely repeatable/perpendicular mechansim.)

We could design a stiffer/stronger/more precise reprap mechanism, but that would be far harder for most of us to build, and harder for reprap to reproduce. (It does tempt me though, at least as a thought experiment.) Measurement and feedback often give more improvement than building a stiffer machine. (Confession time: my graduate studies focused on feedback and control systems.) I believe this is one reason that most production machine tools use servo motors, and linear scales, instead of open-loop stepper motors.

If I build a reprap, or something similar, I think a CMM head would be worth the effort to build for it.

-- Larry

There are still ongoing discussions about the use of servomechanisms rather than plain old steppers, and not just on reprap The general consensus seems to be that cheap or home-brew servos are inadequate, and expensive servos are *much* more expensive than steppers, and very often not better *enough* to justify the expense. It doesn't appear that we have reached the limits of the current cartesian robot design.

It looks like having feedback from the extruder is potentially very beneficial however, and I expect you will see that coming soon to a reprap near you.

I don't see any reason why you (or anyone) shouldn't design a more complex and/or expensive cartesian robot, assuming you stick to the same principles as the Darwin frame so as to allow anyone to use a small, cheap, simple system to assemble a bigger and better one. Or were you thinking that the design behind the Darwin bot are sufficiently flawed that it isn't worth continuing down that route at all?

There's no write-up about it on the wiki, but certainly a lot of thought about it.

If you wish to avoid having to add a DoF (degree of freedom) that tilts a needle-shaped probe (mechanically more complex I think), it might be a nicer alternative to have a needle at a 45 degree angle and rotate it towards the surface you wish to scan (pretty simple 4th DoF). We need to be able to measure around the accuracy of what we could still print (preferably a bit better, of course). This means that 45' overhangs can be both scanned and printed. Of course a small servo with a needle on it could make the angle adjustable (5 DoF bot).

Contact can be measured in a couple of ways:
- A pressure sensor (piezo stuff is not really fab-able I guess)
- Use an electromagnetic system like in an electric guitar or that of a pick-up (record player), I guess such a system can only measure vibrations, right? Otherwise no magnetic flux will arise. The needle could be made to vibrate and upon contact become dampened.
- Electrical. If the object is metalic anyway, this is easy. Other options are aluminum foil or perhaps some other (at least low-conductivity) spray on stuff. Probably, very low conductivity is needed if you charge the object and measure that. This might work with any object...
- Any others?

Of course there are optical/laser systems that could be made separate of the RepRap, but making use of the Cartesian bot we have anyway would really simplify things. It will also solve merging the point clouds from different angles, which might not at all be easy. The scanning volume would be similar to our build volume, which is handy for copying physical objects. Besides, the entire stack of solutions we have for the RepRap applies (except for the extruder). Increases in precision (microstepping!), durability, speed, etc. of the printer would make the scanner also improve.

In future versions the electrical tester could also be used to fully automatically test fabricated PCBs. Or actually PCB might be too 2D... I suggest we name it "printed circuit volume"

---

Regards,

Erik de Bruijn
[Ultimaker.com] - [blog.erikdebruijn.nl]

We can also use a CMM as an input system for 3D modeling (as a sort of "digital chisel" or airbrush), and we can mount a laser scanning head on it.

It sounds like a good project.

I like the idea of a measuring probe - a simple mechanical probe sounds a lit easier to build and extrude than a laser scanning head. A needle with a simple electrical contact would enable the measuring of the outline of components, which would be immediately useful feedback for calibration and also for determining the position of supporting material/previously extruded parts.

> Contact can be measured in a couple of ways:
> - Any others?

Yes. Probe support has three pairs of cylindrical contact bars radiating at 120

... here i made a sketch of a high precision 3D-prober (sub-micron-accuracy!) i developed some years ago:


The self-centering of the spheres is made with the ferrofluid-cushions, fixed in the magnetic poles.

The seismic body (the probe-tip with the two atached speheres) should be made of a lightweight material but with a conducting surface - e.g. aluminium or coated plastic.

As it's a completelly stick-n-slip-free process, smallest forces provoke a diplacement senseable with the capacitors embedded in the middle of the magnets, measuring against the seismic body - so with some calculating you can measure every displacement of the probe-tip ...

I think this device is easy reprappable, you need only the magnets, some ferrofluid and small conducting plates for the capacitors.

Viktor

Such an analog measurement is probably most precise. What are the dimensions of this thing? I find this hard to imagine, I hope it's couple of cm high, not microns as well . You'd need a probe to see if you're building the probe right, let alone tiny fingers.

Nice to see that we're approaching ~ 100 x nano scale already I never thought things like this could be made kitchen-lab style... The scanning tunneling microscope might also be possible then... If we can do this, we can perhaps also manipulate matter at atom scales anyway... (okay, now I'm really getting ahead of myself)...

If it's big enough and can measure at a large enough scale we do not need to make the cartesian frame more precise, just the stepper drivers.

With this head (you only need one even if you have multiple repraps) I imagine you could use an (evolutionary) algorithm to automatically tune the print parameters to get the best print quality. You could exactly measure the dimensions of an extrusion at different speeds, etc. Of course there's a software challenge here. I've worked with some evolutionary algorithms though.

By the way: For this device, because equal charges repel, can't we manipulate the probe with carefully controlled charges?

The ferro fluid would be fun to have too: [video.google.com]

---

Regards,

Erik de Bruijn
[Ultimaker.com] - [blog.erikdebruijn.nl]

Hi Erik,

... imagine with magnets and spheres of 10mm diameter - so it wouldn't be so fiddling small

The measuring accuracy can be down to nanometers, limited only by the measuring accuracy of the capacitors.

The box-size can be something between some millimetres to maybe 10 centimetres (then it's problematic with the pure weights of the components)

I built the ferrofluidic actuators and sensing devices in outer dimensions of 3cm to 10cm with strokes (free moving range) of some millimetres.

I have some commercial ferrofluid from Ferrosound, but it tends to separate in the field of a NdFeB-magnet, so a friend mixed some special mixtures, which are more stable in strong magnetic fields - the prototypes are still working after 5 years

Viktor

This is really becoming interesting.

Just out of curiosity:

are the set of sensors closest to the probe necessary?
(or rather, do you know how much precision will be lost if they are omitted?)

Edited 1 time(s). Last edit at 06/18/2008 07:48AM by bjathr.

Hi bjathr,

... i don't know exactly what you mean - this setup is a set of two half-cages with orthogonally oriented capacitive sensors, which measures the distance between the conducting pads and the sphere(s).

From the displacement of the two spheres you can calculate the weight of the probe-object itself, the 3D-orientation in space (tilt-n-nip), dynamic displacement through vibration and every change when the tip touches an object or surface.

You can use every high-resolution sensor-type - e.g. inductive, acoustic (hypersonic) or optical (interferometer)

The basic principle is the 'soft' fixation of an embedded object in the 3D-gap between surrounding ferrofluid-cushions.

This '3D-arrest' is fluid-dynamic without any stick-n-slip, so it has an infinite resolution and the measuring accuracy is only limited by the resolution of your sensors ...

Viktor

bjathr Wrote:



> are the set of sensors closest to the probe
> necessary?
> (or rather, do you know how much precision will be
> lost if they are omitted?)

Greeting all,

It seems to me that:
to the degree that you can neglect any contact torques (about the center of the probe tip), then you can reduce the 6 DOF (three contact forces, and three contact torques about the tip's center) to just 3 DOF (just the contact forces, presumably acting strictly along vectors that *exactly* intersect the probe tip.) For the three unknowns, you'd need only three independent measurements, thus three sensors.

In practice, I think you gain somewhat from the redundancy of more sensors. Assuming negligable contact torques may well be a mistake. (Though if the tip diameter is small compared to the probe's length, then these will be less of an issue.) Cloning more sensor channels of is relatively easy, and differential measurements overcome a multitude of sins/sensor non-linearities.

Some additional details:

We need to support the probe against both forces and moments, so we need the lower magnets/ferrofluid anyhow.

The geometry shown doesn't resist contact induced torques about the probe's long axis. So, stictly speaking, there are only five observable DOF. rotation about the long axis would spin the probe. Provided that the probe is a good body of revolution, this wouldn't be observable by the sensors (unless it dragged the ferrofluid about via viscose friction.)

The geometry shown uses more magnets (and capacitors) than strictly neccessary, ten, so far as I can see. (A diagram would be good here, but I'm a poor artist.) Let me try to describe an alternative geometry, using only six:
Imagine a hollow cube. The probe's axis will be along one of the cube's body diagonal. The two balls nestle inside the cube, skewered on a rod along this body diagonal, and are each tangent to the three faces that meet at the opposite corners. Clip off one of the body-diagonal corners, and extend the rod outside the cube. Where the balls touch the square faces, we place the magnets and capacitor plates. Now (to compare it to the original geometry) rotate the whole thing so that the probe axis is vertical; namely stand the cube up on that body diagonal. Voila, similar characteristics, but fewer magnets and capactive channels to measure. Iff we can neglect the contact moments, then we'd only need to instrument three of the capactor plates, but that'll probably end up costing us accuracy.

Again, this alternative geometry is insensitive to contact torques that spin the probe about its axis. Probably not a big deal.

-- Larry

... you can simplify the system by reducing the amount of sensors ... and you can build a probe-stick with only two magnets, when you atach the magnets to the stick, give the ferrofluid on the magnets and fix the ferrofluid in two sensor-cages too:

But it's much easier to lathe a perfect symmetric shape then find an absolute perfect magnet ...


There are many possible designs, but my first sketch shows a type with an optimal accuracy and with orthogonal differential sensors, what's easier to imagine and no problem with common electronics ...

Viktor

VDX Wrote:

> I have some commercial ferrofluid from Ferrosound,
> but it tends to separate in the field of a
> NdFeB-magnet, so a friend mixed some special
> mixtures, which are more stable in strong magnetic
> fields - the prototypes are still working after 5
> years
>
> Viktor


Viktor,

Do you have any information on what that special mixture is that resists separation around strong magnetic fields? (Or have some idea of how to track that info down.) I've only read about ferrofluids, but haven't used them for anything, yet. Very interesting idea.

In your sketch above, how many capacitive plates would you use?

Best Regards,

-- Larry Pfeffer

Hi Larry,

> Do you have any information on what that special
> mixture is that resists separation around strong
> magnetic fields? (Or have some idea of how to
> track that info down.) I've only read about
> ferrofluids, but haven't used them for anything,
> yet. Very interesting idea.

... my friend made some ferrofluid-research at Charite in Berlin for medical applications, so he had some very special ressources - but some years ago he's retired in old-age-pension and we lost contact ...

For my problem he mixed AFAIK nanodisgerged magnetite (spheres with 30 to 50 nanometers diameter) with aliphatic acid and different types of synthetic or mineral oils.

What made his mixtures special was the process of disperging the magnetite and coating with the aliphatic acid - so he could mix much higher procents of magnetite without solidifying or separating in strong magnetic fields.

Normal mixtures seems to have between 3% and 5% of magnetite, we managed to embed until 15% and i have some probes with different concentration and some residues which have until 20% but stays fluid in the field ...

Actual ferrofluids are comonly used as sealing, damping and cooling media in worthy loudspeakers, in HD-drives and sometimes in stepper-motors.

Here: [www.teachersource.com] - you have a source with the ferrofluid from ferrosound - maybe it's a bit better now, then my charge from 2001?



> In your sketch above, how many capacitive plates
> would you use?

... there should be 5 corresponding pairs, so it's a set of 10 capacitors, where every two opposite are combined for linearisation.

Viktor

... another idea for a low-cost but high-accurate probe-head:

Atach an ultrasonic resonator to a tool-tip and a microphone to the bed.

When the tool-tip touches the surface of the object, the microphone senses the contact ...

Viktor

VDX Wrote:
-------------------------------------------------------
> ... another idea for a low-cost but high-accurate
> probe-head:
>
> Atach an ultrasonic resonator to a tool-tip and a
> microphone to the bed.
>
> When the tool-tip touches the surface of the
> object, the microphone senses the contact ...
>
> Viktor


Wouldn't the behavior/accuracy of such a system be highly dependent on the material that the object being sensed is made out of?

A ball bearing switch with a needle touch probe would be easy to make,
and mostly printable I guess

A small pcb, six steel balls, three small rods, two wires and an old milling bit.
The rest is printable (almost)

I've seen one made this way some days ago, I can't find it right now....

'sid

current song: "the search" - cherry poppin' daddies

[edit]
Here we are:
[www.indoor.flyer.co.uk]

Edited 1 time(s). Last edit at 08/19/2008 07:02PM by sid.

... all mechanical/switching probe-heads have backlash or hysteresis what you have to recalculate and can be inaccurate to some 10 microns.

I need an accuracy down to submicrons and with probe-tip-diameters down to 0,1 mm - i already have some hardalloy- and ruby-spheres down to 0,25 mm and some NiTi-rods with 0,1mm diameter where i 'rounded' the ends with a welding-laser ...

For the object-material: - yes, with 'soft' materials and bigger objects it's not so easy, but you can apply the microphone to the object itself, then you can sense increasing of the sound too (against the sound waves traveling through air or the robot-frame).

My primary 'targets' are mostly very rigid/stiff objects made from metal, ceramics or compounds, so this isn't the problem

There are some other methods - e.g. conductivity sensors and a perfect sphere caught in a 6-sided cage centered by ferrofluid (as showed above) or simply a foam-filling.

But the ultrasonic-contact-probe seems to be the simplest way to achieve hiegh accuracy without expensive or complicated electronics ...

Viktor

Thought it's about a RepRap tool head...
And to be honest I don't think that the need of submicron accuracy is widely spread in here

Most parts that are "scanned" are printed out on the RepRap again, I guess,
so everythings thats a little more accurate than the printing toolhead will do a pretty decent job.


'sid

Hi Sid,

... it's about my special interest in 'micro-fabbing' with self-mixed pastes at room-temperatures.

I've already made some experiments with microdispensing of highly viscous fluids with droplets of 30 microns diameter and 5 to 10 microns height.

On the other side i developed and used a micro-welding system with a 8-watt laserdiode (used at max. 5Watts) for sintering/welding gold-paste droplets of 60 microns size.

With some mixed ceramic-waterglass-pastes and metall-pastes i hope to be able to fab some multilayered 3D-objects as bases and assembly components for microsensors ...

And why not go the way for custom/home-brew electronics for the reprap?

Viktor

I didn't say that it's wrong to have high resolution.
Don't get me wrong, I love the idea of having a micro- or even a nanofabber at home.
All I said is, that for this development stage of the RepRap a mechanical probe head would be sufficient and easier to fabricate from scratch.

As long as the cartesian (darvin in this case) isn't capable of such a high resolution, an ultrahigh resolution probe head doesn't make any sense.

So just one step at a time...

'sid

Hi Sid,

... OK for the conventional reprap, but with my CNC i have a resolution of 12,5 microns in halfstep and up to submicrons with microstepping (i have three 1/256-microstepping drivers which i can temporary switch for the halfstep-drivers)

My next idea is assembling a direct driven 'tilting-arm'-tripod, where 3 strong steppers turns synchronous their arms and the parallel kinematic is mainly emphasized by the delta-robot ( [en.wikipedia.org] ) or the ABB Flexpicker (see appended image).

*** My first concept with the linear drives ( [builders.reprap.org] ) is good for micro-fabbing, but when designing a system for 'normal' use, you need a much bigger working-area ***

With the highest possible microstepping the motors have an angular resolution of 51200 positions (or 0,007

Edited 2 time(s). Last edit at 08/20/2008 08:58AM by Viktor.

... maybe this type of CMM is better DIY-able with high resolution encoders:

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]