I have a problem...several years ago, I threw together a balancing machine that ran with great accuracy/no difficulty for over 1/4 million production cycles, never had any issues...THAT workpiece though was basically a 'plane' disk mounted on a small collet...(about 20 pound workpiece on a one pound collet) the machine was designed to flex enough that as the workpiece rotated at fairly high speed, the spindle would find its own center, and a sensor would simply measure the 'orbit' of the spindle bearing mount-and it ran like a clock for almost 2 years...cheap/simple/reliable was kinda proud of that little machine.

just built (3) more machines of similar design but scaled up for much larger workpiece- and THIS workpiece has substantial offset from the collet bore/locating surface to its center of mass AND the collet is huge...nearly 100 pounds, about the same as the workpiece...

early on it was very apparent that 'tooling offset' would be required to minimize effects of inherent imbalance in collet/locator/spindle- the way I attempted to perform 'fixture compensation' is as follows:

1) mount workpiece on spindle(put mark at 60 degree position on collet)

2) put master weight at 60 degree position on locator

run machine and store the deflection amount and high point angle as 'cal#1'.

3) rotate workpiece mark to 180 degree position, and move test weight to 180 also.

run/store that data as 'cal#2'

4) rotate workpiece mark to 300 degrees and move test weight to to 300 also.

run/store that data as 'cal#3'

5) leave test weight at 300 degrees, and rotate workpiece mark to 120 degree

position. run/store that data as 'cal#4'.

subtract deflection amount in cal4 from cal3 and divide by 2: this gives amount of deflection contributed by unknown workpiece imbalance. Subtract this value from cal1~3 to get true counts resulting from test weight only.

using the remaining amounts and angles, create a 3 point circle, and calculate distance to center from machine zero, and angle relative machine zero- store this as 'tooling offset' vector...done with the 'comp' part of the program.

remove test weight, run any part.

create triangle- side 1 = amount/angle detected, side 2= tooling offset vector, create side 3 which connects the endpoints. length of third side= amount of imbalance in workpiece after removing 'tooling offset' amount. (endpoint to center of collet rotation)

if the workpiece is deliberately loaded with heavy side at the 60/180/300 positions that the original circle was calculated from, the numbers are nearly perfect(under 1% error), however if angles in between those above are used, errors of near 20% are occurring...

plotting the data in CAD illustrates quite easily, but I am totally at a loss as to how to correct this issue...

the root of the problem appears like a mechanical issue, but everything checks out OK, and all three machines show similar effects...

example: (ACTUAL NUMBERS FROM A TEST TODAY)

1236 PULSES @ 76 DEG. CAL#1 (TEST WEIGHT/PART MARK AT 60)

1154 PULSES @ 205 DEG. CAL#2 (TEST WEIGHT/PART MARK AT 180)

1398 PULSES @ 325 DEG. CAL#3 (TEST WEIGHT/PART MARK AT 300)

1111 PULSES @ 327 DEG. CAL#4 (ROTATE PART ONLY TO 120)

143 PULSES AT 340 DEG. VECTOR TO CENTER OF INSCRIBED CIRCLE

1117 PULSES = RADIUS OF INSCRIBED CIRCLE

(note the 60/180/300 actual angles are detected a bit off...some

gyroscopic/precession effects going on maybe?)

the above data was verified accurate in cad- so that part of the math is correct...

HOWEVER If I run the same procedure using 3 points at 0/120/240, again CAD

verifies math is correct, yet the vector angle changes by up to 20 degrees...in effect the (6) mastering holes in the fixture appear to be in the wrong locations- but layout says otherwise...3 points (60/180/300) gives 143 @ 340, but other 3 points (0/120/240) give about 140 @ 355 (dont have actual exact numbers here...)

Ive tried running without workpiece just running test weight in the 6 master holes, and consistently get 6 points that never fall on a common circle.

pondering the potential issues:

1) Non-linearity of suspension? seems likely, but output at the angles used to calibrate gives true output at hi/mid/lo test weights, so 'appears' quite linear

2) effects of plane differences from tooling center of mass(high on spindle) relative to workpiece lower on spindle

3) most likely: as output during workpiece cycle is actually resultant of tooling offset+ workpiece imbalance, angle displayed will always be incorrect(but maybe lining up with calibration points makes errors common, giving reason for good output?)

4) gyroscopic precession effects due to very high mass of tooling/workpiece trying to 'wobble' to a natural center point?

the #3 is biggest reason why I came here: if I have a tooling vector offset as above: 143 counts @340 degrees, then I add a workpiece and run it, giving raw data of 1000 counts at 270 degrees, currently I am using the 1000 @ 270 as a line, the 143 @ 340 as a line, and connecting the ends to get the distance from that inscribed circle center point- BUT thinking about it, I know the 270 MUST be incorrect as it is a resultant of both

vectors...how can I get the true value of this angle?

any simple math formula would be greatly appreciated- I'm just a CNC/PLC retrofit guy and the math in this thing has got my head spinning already... took what seemed like forever just to get the 4 pages of rather cryptic looking ladder logic to trig out the center point of that inscribed

circle...can draw with a compass and straightedge (or CAD) in a few seconds, but in PLC logic it was not my idea of fun. Someone suggested tossing this in one of the math guru forums to seek advice...so here I am

the other reason I'm here is simply to ask if anyone has any suggestions as to why the 6 test holes wont orbit a common circle with the same test weight moved around to the various positions? only thing I can think of is that as the test weight plane and collet plane are inches apart, maybe moments are tugging things in different directions...

(BTW I tried storing counts at every 10 degrees during any one run, to see if a circle was formed by the endpoints, and its only elliptical by maybe 1%... not enough to explain the variations seen.)

well, If anyones taken the time to read this far, thanks- and thanks in advance for any suggestions- I'm all ears.