Position calibration & accuracy

With the discussion of making electronics, be it SMD or printing circuits, and other forms of fabbing where we introduce parts that are not fabbed all in one run (e.g. building captive nuts) it seems that we need to spend a little time thinking about how to detect positions and how to calibrate the location of a reprap's extrusion.

This, naturally, brings up the question of machine vision. People talk about this problem like it's going to be very difficult to solve. I think we can make it pretty easy.

For example, the host computer can use a webcam attached to the head carriage of the reprap capture images, then process them.

For extrusion calibration, this would work as follows:

The toolhead controller stores an approximate offset from the nozzle to the camera, which it sends to the host PC. The reprap controller prints a simple "+". The PC adjusts the toolhead by the vector given by the toolhead controller, then captures an image. It looks for the "+" and if it finds it, it calibrates the toolhead offset vector by the distance from the center of the image to the "+". If more accuracy is required, it can do a calibration loop where multiple +'s are printed until the + winds up in the middle of the captured image.

For non-printing toolheads, this is a lot more difficult. In addition, we will need a method for detecting the location of a placed part.

Neat idea. I see a lot of the home machinists using a webcam plugged into their milling machine to find centers and edges on their workpiece. They use a human in the loop to do the image recognition right now though.

On my Darwin, X and Y absolute calibrations don't matter much, as I don't care exactly where the part gets built, I just have to not run into the limits of the machine. The steps/mm doesn't change much either.

The Z axis, however, is very sensitive to height differences between the nozzle and the bed. 0.05 mm makes a difference in part adhesion. And it changes as the build progresses, as the PFTE slowly heats up and warps.

I do a sort of human image recognition right now - I look at the spacing between the extra wide filaments on the 1st raft layer; if they're too wide or too close, I stop the build, readjust the Z axis, and start over. A test patch could work for that too, but the programming to get that info from a webcam is well beyond my skills.

Along a similar line, when doing PCB's for GPS receivers, we would include an RF test patch on an un-used area of the panel, and then test the RF properties of each PCB to makes sure we were getting what we asked for in terms of PCB interlayer properties.

Wade

Not that I want to stand in the way of progress, but I am not sure this would be it. Loss of position over the course of a print is not a problem I am aware of anyone having, due to the lack of backlash in Darwin. Even if this does present a problem ( as Wade mentions with his Z adjustments) regular returns to a test point are called for, unless you you plan for some very high end AI which can identify and diagnose print flaws. The most realistic concern is that when Mendal starts using multiple heads they will need to be "registered" so that they print to identical co-ordinants.

Currently Darwin has opto endstops on the X and Y axis which make calibration of those axises very simple. If we assume that a datum called toolhead offset can be stored for each toolhead giving the relationship of the print point to the endstop flags, we are well on our way. Someone, I believe it was Forrest, has a simple micro-force touch down sensor which can be built into the print bed to calibrate the Z axis. Problem solved electro-mechanically. Computing (not to mention parts) cost as near zero as possible. Remember, one goal of this project is to bring the power of manufacturing to EVERYONE, even those without webcams, which is still most of the world.

I was thinking: Instead of webcams and "force touch sensors", think cheap and just use a simple pressure switch?

I dont know who invented the opto endstop (but I do know its foolproof) but if you have a simple on/off button it will work the same way.

Firstly, I cannot fathom a place where it's feasible to get a PC, all the electronics to run a reprap, stepper motors, and not a webcam. Honestly, they cost $23 CAD for a 1.3MP (interpolated) camera ( [www.ncix.com] ). The reprap steppers sells for $30 CAD.

Secondly, the point always was for calibrating the x/y-axis position of a swappable toolhead. The problem is that two extruder nozzles may not hold the same positional relationship to the toolhead carriage. Consider the difference between a filament extruder and a syringe extruder. This makes endstops useless for calibration.

The electromechanical Z-axis calibration is totally viable, but the x/y-axis calibration holds a tad bit more complexity, hence my attempts to solve it.

Edited 2 time(s). Last edit at 01/28/2009 04:38PM by Annirak.

>> Firstly, I cannot fathom a place where it's feasible to get a PC, all the electronics to run a reprap, stepper motors, and not a webcam. Honestly, they cost $23 CAD for a 1.3MP (interpolated) camera ( [www.ncix.com] ). The reprap steppers sells for $30 CAD.

Then by all means, get a web cam with your reprap, and start working on your enhancements.

That is the nice thing about open source; you can work on your own pet peeves rather than expecting others to work on them for you.

Hi Annirak,

i solved a similar problem with two toolheads and exchangeable millheads or syringe-needles with a webcam in a corner looking upright with a crosshair in the monitor.

Adjusting the Zero-start-position with an accuracy of nearly 0,01mm in XY and 0,1mm in Z was made by moving the actual tool in the range of the camera and fine-tuning in X,Y,Z until the tooltip was in the middle of the crosshairs and the tip sharp visible in the focus-plane.

For accuracies of 0,01mm in Z too with only one camera you can use a mirror inclined 45 degrees and the camera oriented horizontally - so you can see two orthogonal images of your tooltip at once ...

Viktor

Viktor, That sounds like a great solution! Have you posted details anywhere?

Hi Annirak,

... yes - look here: [forums.reprap.org]

I sketched some hints with reversing optics too and linked to some images made with the second type of my microscope-camera, but the old link isn't valid anymore - here the actual position: [mikroskopkamera.de]

Viktor

Firstly, while I am sure a webcam can be gotten anywhere those of us who are now building RepRap's live, the vast majority of the world lives without computers or cameras of any kind. It's known as poverty. And while attempts to alleviate it might well involve sending collections of parts to remote places, the fewer such parts ("vitamins" in RepRap terms) it takes to make a machine the faster and cheaper they will spread. If a problem can be solved with extrudable parts and software it makes no sense to solve it with a webcam, at least not in RepRap terms.

As I said if we imagine that one of the specifications of a toolhead is its' print point offset, which is the measured distance in three dimensions of the print tip from the opto-end flags and z dimension touchdown point then returning the toolhead to those endpoints would serve to locate the print point via elementary math. Under normal handling the offset numbers should not change over the life of the tool, except in a few cases (such as an EDM tool).

The Z sensor I had in mind is simply a flat circle of plastic with two separate conductive strips (each going 120 degrees or so around the edge on opposite ends of a diameter); it is pressed up by a spring set in a hole in the print table (probably directly under the x/y calibration endstops) and held in place by 4 sharpened wires which are connected in series via the conductive strips on the plastic circle. Thus even the tiniest depression of the circle breaks the circuit. The problem with buttons is that every button has travel which needs to be compensated for. Breaking a low voltage circuit requires as close to zero travel as one is likely to see. Obviously the whole set of endstops and z switch could be fabbed on a Mendal except for a spring, some diodes, and a few bits of wire, which are much more likely to be found in a scrap pile than is a working webcam.

As for Viktor's work and Annirak's ideas, they are both great and I do wish Viktor would detail his work on the wiki; but .01mm seems like way more accuracy than Darwin can benefit from given the thickness (.25mm) of its build elements. It's rather like trying to tell time with a stopwatch. RepRap is an open source project and people should feel free to improve it anyway they want, but I think that the core machine needs to stay focused on self reproduction and thus simplicity. I hope we are building a machine that can change the world, not a shiny toy for well off techies.

Hi Brian,

no question - one of my most favorite mottos is "keep it simple, stupid"

My interest in higher accuracies is more private as i'm going a sidepath of repstrapping in direction of microns-resoultion with fabbing and reworking with different technologies.

I think that for the mainstream-user the best solution will be a triplet of switches for finding the XYZ-zero-position in the range of 0,05 to 0,1 microns (actual/wished printing accuracy).

But sometimes there are questions for different solutions or even higher accuracies too ...

I think reprap could better evolve like a growing tree with a main branch and some smaller twigs

Viktor

Edited 1 time(s). Last edit at 01/30/2009 06:55AM by Viktor.