optical pattern tracers

I ran a flame-cutter several decades ago that used an optical trace - the thing had to have a fairly high contrast between line and background, and it had an "ey" about the size of a coffee cup. The whiz-kids who kept it in operating order (as it was an *old* model) said that it had a set of pickups surrounding the central sensor, and they read "more" or "less" and the movement was controlled by that.

I personally favored the "witchcraft" explanation however, because if a big gob of slag blew out of the burn-table and hit the paper in just the right place, even uttering harsh imprecations and vile epithets wold not save the job. This usually happened after the job was 65% cut, and generally when cutting 8-up in 8" steel. (No, it was just about impossible to save it and restart).

I spent two nights shoveling out the table while the factory-folk installed a tape head and controlller on it and two weeks while they debugged it, and then went on to significant wealth at piece-rate.

Flash

it is probably a photocell - a long time ago (inthe 60s) there was an article in popular electronics about building a robot that would follow a line - it used a single photo cell and a light bulb, no lenses. it would follow one side of a piece of tape - you set a threshold (it had one transistor, a CK722, and one sensitive and expensive relay) - when the voltage out of the photocell was greater than some amount, the relay energized and it drove one motor, making the thing turn left, when it was less than the threshold, the relay dropped out and that would energize the other motor making the thing turn right.

with the addition of a lens and maybe one more transistor, this could be of use, no?

I was talking to a torch operator (late 70s) when his machine took off after a fly walking across the template!!!! Moved about a half an inch before the fly flew away then the unit just moved in a small circle looking for the line. Oscar saw this happening and shut the torches down before too much bad happened.

Darn funny though, he said it happened often. The drafting machine used for the templates was 30' long and 8' tall hung along the wall in IE.

Matt

One method is to use two photocells, squinted a bit, one to one side, the other towards the opposite side. They have overlapping fields of view. Ratio of the two sides can find a line position. In the center the two signals are equal, and if line is one one direction one cell decreases signal while other increases.

The classic approach is two photocells (or one split photocell) feeding a differential amplifier. If the line is halfway between the cells, output is zero. It the line strays one way the output is negative, and the other way positive. To do full X-Y, there will be four cells in a square, and two differential amplifiers.

This was enough for the gantry servo to steer by.

Although it was easily distracted, as other posters have noted.

Joe Gwinn

What's that Lassie? You say that Flash fell down the old rec.crafts.metalworking mine and will die if we don't mount a rescue by Sat, 6 Dec 2008 23:48:13 -0500:

Thanks.

That is kind of what I figured for the steering of the pickup head. But how does it convert the position of the pickup head to the X and Y drive motors. I could make something follow a line easy enough. But when it comes to going around a corner that would require one(or both) of the drive motors to change direction, that's when I get lost.

If I had an X Y table (or gantry), that was controlled by a pointer that you could rotate to change the direction of travel, I could make a line following eye to move said pointer and trace out a shape on paper.

What's that Lassie? You say that Bill Noble fell down the old rec.crafts.metalworking mine and will die if we don't mount a rescue by Sat, 6 Dec 2008 22:18:09 -0800:

Actually I get how you could use a photocell to sense a line. And I get how to use that to steer something. I could build a toy car that would follow a line. No problem.

But I don't know how to get an X Y table to move when it needs to change directions of one or both of the axes.

On the tracer I saw, the 'eye' rotated to follow the line. The X and Y drives were controlled by the direction that the 'eye' was pointing. That's the part I'm not sure of. How to control two axis, forward and backward as needed, to move in the direction that the eye is pointed.

What's that Lassie? You say that Don Stauffer fell down the old rec.crafts.metalworking mine and will die if we don't mount a rescue by Sun, 07 Dec 2008 11:06:06 -0600:

I get the eye part. And how to steer the pickup head. But how to control the X and Y axes to move in the direction that the pickup head is pointing? That is where I get lost. If the axes didn't have to be able to change direction, then I see how they could do it with a non-rotating head that just controls the speed of one axis to keep up with the other.

But in a tracer, the X and Y axes need to be able to reverse as needed to follow the line. And the speed of each needs to be adjusted to keep the torch constant as the lines angle changes.

What's that Lassie? You say that Joseph Gwinn fell down the old rec.crafts.metalworking mine and will die if we don't mount a rescue by Sun, 07 Dec 2008 13:03:03 -0500:

Do you mean that's what rotated the pickup head? And the position of the pickup head controlled the X and Y for the gantry?

Or was the head stationary? With a stationary head, what would determine witch direction to go on the line?

The head would be stationary, and all adjustment would be electronic.

Let's first deal with one axis at a time.

There are two rectangular cells side by side. If the line to be followed is parallel to the boundary between cells, then the differential amplifier output will be signed and proportional to the offset of the line from the border. If the line is instead perpendicular to the border, the output of both cells will be reduced, but their difference will remain zero.

Now, add a second pair of cells, so we have four cells total laid next to one another like tiles in a square pattern. Let us number the tiles by row and column:

S11 S12

S21 S22

We have two differential amplifiers. The X amplifier has plus inputs from S12 and S22, and minus inputs from S11 and S21. The Y amplifier has plus inputs from S11 and S12, and minus inputs from S21 and S22.

In math:

X output = +(S12 + S22) -(S11 + S21)

Y output = +(S11 + S12) -(S21 + S22)

It's clear what will happen if the follow line is vertical or horizontal, so let's consider the case of a diagonal line crossing the centers of S21 and S12. If perfectly centered, X and Y outputs are both zero. If the line drifts towards S22, what happens? S12 and S21 will remain about the same, while S22 will grow and S11 will shrink. This will cause X output to become more positive, while the Y output becomes more negative.

One can go through this exercise for any line orientation, and get the same answer, so the X and Y outputs provide a sufficient steering signal.

Joe Gwinn

What's that Lassie? You say that Joseph Gwinn fell down the old rec.crafts.metalworking mine and will die if we don't mount a rescue by Sun, 07 Dec 2008 19:22:55 -0500:

Joe,... you're my hero. I get it.

I didn't think I would be able to understand a stationary head tracker. Especially from one posting alone.

Just one thing. What makes the axes move at all? Once centered on the line, why does it follow the line. And witch way does it go? On a vertical line would it go up or down?

And do you know anything about the rotating head type. I get how the tracker head sees the line and turns to follow it, but how is the rotation of the tracker head turned into X and Y outputs? Is it like the stationary head sensors, but fed by something attached to the rotating head?

Thanks a million,

One more thing Joe,

Many of the tracers that I've seen on the WWW, can trace a solid silhouette.

How does that work?

That gets displacement error signals for X and Y, but provides no motion in the direction of the line, only correction for offsets perpendicular to it.

I think there must be some means for maintaining set speed in a direction parallel to the line. This could be done with a form of difference engine, or it might use more straighforward computation of Vx^2 + Vy^2 and Vy/Vx. The difference engine might require a start at known speed and known direction. Could have been analog or digital. Microprocessors have been around since the mid-'70's, CNC since the late '50s, and electronic, mechanical and even pneumatic ways of doing such computations have been known since World War II.

Dan,

ISTR that when I would start the burn, I had to manually set it to a preheat position, manually set X and Y vectors toward an engagement of the pattern trace AT A DIAGONAL, so it would pick up the line and run from the oblique angle, rather than the acute. Then, I would mash a button to begin cutting, and when it had completed the pierce, engage the travel.

Flash

The one I was around had CW/CCW switch, CW would move the torch away from the operator (X+/Y+), and CCW would move the torch towards the operator (X-Y-). The torch could be adjusted to "offset" the line a bit for machining later. So CW cut the part larger than the template and CCW cut an inside feature smaller than the template. Unlike the one Flash used the pierce was done from a 90 to the cut line and a pointer on top of the eye was adjusted to "go that way" until it ran across a line, when it hit the line it would go either CW or CCW to the direction of travel.

Once in a while Oscar had to steer the torch by the pointer on the top of the "eye unit" because if the torch got distracted after the "start travel cycle" and lost the line the eye would quickly "spin in a small circle" trying to find it.

It was kind of funny watching the pointer going round and round, (I suppose "not" funny for Oscar though)

Matt

You are right that the above mechanism would not move unless propelled somehow.

As others have mentioned, the operator sets up and starts the cut. The controller is programmed to move at a constant linear speed, and the servo keeps the head on the line as the curve is followed.

Joe Gwinn

I can think of a number of ways. One way has a 9-sensor grid and detects edges, but I suspect that given the expense per sensor channel back then, it wasn't done that way.

They may have servoed to a constant output (not zero) from the head, but this requires knowing something about the orientation of the edge, which is circular.

Ill think about it.

Joe Gwinn

It's propelled at a constant speed as set by the operator, and the servo only follows the line.

A rotating head can use one or more eccentric sensors to do the work of all four sensors in my example (or nine sensors in the larger array for edge following), so long as the sensor head rotates fast enough that the curves to be followed are sampled very densely.

Joe Gwinn

Video cameras are cheap these days, ditto computers to do the image processing. Software is cheaper than hardware these days.

Stan