Re: encoder head pitch, NO Griping
Posted by
David L Anderson
on 1999-06-08 13:51:52 UTC
Ian
Wrote:
The kind of system I envisage would be easily made on the mechanical
side and would consist of a tape with a number of straight black lines
running the length of the machine - say ten at any even spacing - and a
reading head incorporating another strip of the same kind of stuff and
say ten IR phototransistors looking through the crossed gratings via
small slits in a metal plate. I think that by choosing the number of
lines, the grating angle and the number of phototransistors to give an
accurate repeat rate, it should be possible to achieve good resolution
by simple means. The grating could proboably be made by dragging a strip
of acetate across the cutting adge of a screw tap in a suitable frame.
Here is a "work in progress" that I will describe. I have built a small
proto-type that is partially functional. It has no readout, Just signals
that demonstarte the capability. I shelved the project in 1998 for lack of
funds, access to scales or an acurate means to make them, and lack of
sufficient CPU power to do the job cost effectivley. It is only a matter
of time before silicon speed and prices make this method cheaper than and
HEDS1000 solution. Please note that nothing is free in a DRO solution.
Displays, power supplies, cabinet ,etc can add up to a few hundred dollars
in itself
I have been experimenting with a similar idea, Not using moire effect
but an array of phototransitors and a course pitch grating (10 gratings to
the inch). The photo sensors are in a linear array 2mils per pixel 128
pixels per line 1 line. I consider the sensor spacing etched into the
photolithography of sensor as the defraction grating that Ian mentions. The
sensor is made by TI (TSL401) and costs about $6. It is unique in that it
is analog NOT digital. So instead of just a 1 or 0 in quadtrature, you must
read out each pixel's intensity and interpret the results. Consider it as
a one line television scanning a set of a set of bands on a moveable scale.
As the bands move we get a different pattern of voltages from each of the
128 sensors representing the bar position. We must just scan quickly
enough that the fastest motion will not cause us to jump a band, much like
a stepper motor "cogs" when we step it to fast (only in reverse).
As to potential accuracy, the system might have great potential. The edge
of the bar can be found within +/- .005" with very little signal
processing. I hooked the output of the sensor to a scope and was able to
get this level of accuracy by observing the waveform on the screen. The
scale was made on a laser printer using overhead film printed from a cad
program. This was glued to a piece of clear plexiglass 1" wide by 11" long
to form a "cheap" scale for testing. To get more accuracy I was planning
on using the Analog information that is in the voltage on each sensor.
This requires more complexity . An A/D converter and led bias circuit for
tight control of the light source. The software is much more complex and
approaches image processing or DSP in complexity.
The problems:
1) The system requires high speed processing to achive even modest feed
rates without error. The means (dual) expensive 24 Mhz micro controllers
or an FPGA solution Per axis. This Costs a fair amount of development
dollars that must be recovered. The processor cost could reach $40 per
axis. This is not currently competative with HEDS1000 sensors. With
Silicon getting faster though, this cost could drop in half in 18
months.Anyone have an FPGA development system? Let me know!!!
2) The scales need not be fine pitch but must be accurate and have a low
coeficient of expansion. Plastic is just to non-unifom in it's thermal
expansion characteristics. Glass is fragile. A metal scale is the best
choice, but it must be either precision machined or etched. These are not
easy operations. This requires acurate machinery, but it could be cheap if
done in a reasonable volume. Anyone got a CNC mill that could machine a
1"X36"X3/16" bar of cold rolled with 0.1" wide slots, 1/2" long every 0.1"
along its length? If you did it , would it stay straight? Let me know!!!
TO BE Continued..... Accuracy, Precision, and the home shop ....
[David L Anderson]
Wrote:
The kind of system I envisage would be easily made on the mechanical
side and would consist of a tape with a number of straight black lines
running the length of the machine - say ten at any even spacing - and a
reading head incorporating another strip of the same kind of stuff and
say ten IR phototransistors looking through the crossed gratings via
small slits in a metal plate. I think that by choosing the number of
lines, the grating angle and the number of phototransistors to give an
accurate repeat rate, it should be possible to achieve good resolution
by simple means. The grating could proboably be made by dragging a strip
of acetate across the cutting adge of a screw tap in a suitable frame.
Here is a "work in progress" that I will describe. I have built a small
proto-type that is partially functional. It has no readout, Just signals
that demonstarte the capability. I shelved the project in 1998 for lack of
funds, access to scales or an acurate means to make them, and lack of
sufficient CPU power to do the job cost effectivley. It is only a matter
of time before silicon speed and prices make this method cheaper than and
HEDS1000 solution. Please note that nothing is free in a DRO solution.
Displays, power supplies, cabinet ,etc can add up to a few hundred dollars
in itself
I have been experimenting with a similar idea, Not using moire effect
but an array of phototransitors and a course pitch grating (10 gratings to
the inch). The photo sensors are in a linear array 2mils per pixel 128
pixels per line 1 line. I consider the sensor spacing etched into the
photolithography of sensor as the defraction grating that Ian mentions. The
sensor is made by TI (TSL401) and costs about $6. It is unique in that it
is analog NOT digital. So instead of just a 1 or 0 in quadtrature, you must
read out each pixel's intensity and interpret the results. Consider it as
a one line television scanning a set of a set of bands on a moveable scale.
As the bands move we get a different pattern of voltages from each of the
128 sensors representing the bar position. We must just scan quickly
enough that the fastest motion will not cause us to jump a band, much like
a stepper motor "cogs" when we step it to fast (only in reverse).
As to potential accuracy, the system might have great potential. The edge
of the bar can be found within +/- .005" with very little signal
processing. I hooked the output of the sensor to a scope and was able to
get this level of accuracy by observing the waveform on the screen. The
scale was made on a laser printer using overhead film printed from a cad
program. This was glued to a piece of clear plexiglass 1" wide by 11" long
to form a "cheap" scale for testing. To get more accuracy I was planning
on using the Analog information that is in the voltage on each sensor.
This requires more complexity . An A/D converter and led bias circuit for
tight control of the light source. The software is much more complex and
approaches image processing or DSP in complexity.
The problems:
1) The system requires high speed processing to achive even modest feed
rates without error. The means (dual) expensive 24 Mhz micro controllers
or an FPGA solution Per axis. This Costs a fair amount of development
dollars that must be recovered. The processor cost could reach $40 per
axis. This is not currently competative with HEDS1000 sensors. With
Silicon getting faster though, this cost could drop in half in 18
months.Anyone have an FPGA development system? Let me know!!!
2) The scales need not be fine pitch but must be accurate and have a low
coeficient of expansion. Plastic is just to non-unifom in it's thermal
expansion characteristics. Glass is fragile. A metal scale is the best
choice, but it must be either precision machined or etched. These are not
easy operations. This requires acurate machinery, but it could be cheap if
done in a reasonable volume. Anyone got a CNC mill that could machine a
1"X36"X3/16" bar of cold rolled with 0.1" wide slots, 1/2" long every 0.1"
along its length? If you did it , would it stay straight? Let me know!!!
TO BE Continued..... Accuracy, Precision, and the home shop ....
[David L Anderson]