Obscure bug found in G212 stepper driver

[Richards]

Recently, while developing a new controller interface, labeled the G4, for the Shopbot CNC Router, Ted Hall, President of Shopbot Tools, Inc. found an obscure bug in the Gecko G212 stepper driver that can cause serious problems when the stepper driver is used on a machine that makes very small moves, even when the step multiplier is set to 1:1. According to Brady Watson, who is intimately involved in testing various devices for Shopbot, the problem occurs when very small moves are made, particularly moves consisting of fewer than ten steps followed by a direction change. Some of the steps can be truncated, leading to errors that can ruin the parts being cut.

When Mariss Freimanis, President of Gecko, learned of the thread on the Shopbot Forum, he responded with an explaination of what was happening, why it was happening and how to avoid the condition that caused the error. His entire response can be found on the Shopbot Forum, but part of it is quoted here: "The G901s synthesize 10, 5, 2 or 1 internal step pulses for every pulse sent to it. This is done using a phase-locked loop circuit . . . This works OK in CNC applications provided enough time (20 milli-seconds) is allowed to elapse before the next direction change and acceleration occurs."

Since I have been extensively testing Gecko G212s and G340s for more than a year, I was interested in the possibilty of having a bug in the G212 cause problems in projects that may materialize sometime in the future. As a precaution, I removed the step multiplier cards in the G212 drivers and in the G340 driver and replaced those parts with optocouplers, part no HCPL2531-ND from Digi-Key. Those parts, which cost only $1.30 each, turn the G212 into a G202 and the G340 into a G320.

Today, after extensive testing and with several suggestions from Brady Watson, I was finally able to duplicate the error. Here is a listing of Mach 3 G-code that produced the error for me:

G01 Y0.005 F360
G01 Y-0.0075
G01 Y0.0025

Here is a listing of code that fixed the error:

G01 Y0.005 F360
G4 P20
G01 Y-0.0075
G4 P20
G01 Y0.0025
G4 P20

Mariss suggested inserting a 20mS delay before changing direction to give the step multiplier time to clear. His suggestion worked perfectly.

(For those who would like to duplicate the test for themself, I set 2000 steps per inch, 740 velocity and 40 acceleration in the Mach 3 setup. I called the section of code, listed above, as a subroutine and repeated the cycle 100 times.)

It should be noted that when these extremely small steps were doubled that NO error occurred. My suggestion is to use the G202 instead of the G212 whenever there is a possibility that you will be making extremely small moves.

I have to applaud Ted and Brady for finding this bug. Although it is obscure it is exactly the kind of bug that can drive us CNC operators insane when parts are ruined for no apparent reason. I also have to applaud Mariss for knowing exactly what could cause the bug and know exactly how to avoid it.

-Mike Richards

[braidmeister]

Mike,
Thanks for the kudos. It should be noted that Dirk Hazeleger from www.botrods.com was also instrumental in observing & tracking this condition down, with his AGek ShopBot upgrade electronics package.

-B

[Richards]

Thanks for reminding me. Dirk is a great guy. He and I have exchanged several emails about this problem and about several other things that are of interest to him and me. My apologies to you, Dirk. You were probably the first one to alert me to the bug.

I also just received an email from Scott Worden, at Timber Lake Creations, who may have been the very first to find the bug when he was cutting a file of an Indian Headress. He reported that his Z-axis had moved out of position by about 0.125" while cutting the file.

-Mike

[routerman]

The same problem happens when using the step multiplier board for the G320's. The PLL looses steps when you accel too fast.

Marris thinks that resonance in a stepper is caused by the software feeding it, but this is not true. Stepper drives with out that fancy 'anti-resonace' circuit work just fine.