Determining Weighing / Counting Scale Tolerance - No manufacturer specification

[M Greenaway]

Thanks Ryan

But why calibrate agianst a known weight when we do not count against a known weight ?

We do not say that, for example, 100 pieces weighs 100 grams. so if we weigh out 1 kg we must have 1000 pieces. What we do is a manual count of a sample of say 100 pieces, we put the counted sample onto the scales and then set the scales count to 100. Then we tip in the product up to the required quantity.

To my mind any relationship to true weights is totally irrelevant.

[Ryan Wilde]

etdiyas said:

the scales are calibrated but the uncertainty is not stated.
how about the NIST Handbook 44 and 105-1, some people i know says that these could help, where can i purchase these references thus it cost much?

NIST Handbook 105-1 concerns standard weights, and does you no good for scales. NIST Handbook 44 Section 2 has quite a bit of information, but it is used for legal metrology in the USA (such as scales used to verify the weight of train cars, etc.), and it would not help you all that much. Both Handbooks, as well as many others, are available online at
http://www.nist.gov , and they are free to download (at least in the USA they are, give it a try, see if it works from there).

That said, the tolerance of your scale, if it is made by a reputable manufacturer, will probably be around 2-3 counts, with a repeatability of <1 count (using the RSS of 10 measurements). But, if you are making measurements that are ±200 counts, then you can calibrate it to ensure a 10:1 or 4:1, or whatever your quality system deems acceptable, regardless of the manufacturer's tolerance.

"2 times of scale division" may, or may not, be adequate. If the scale was not designed to meet that criteria, and you are trying to use it at a tighter specification than it was designed, you will have some fairly rigorous proof to accomplish to show that your scale does meet that criteria throughout its calibration cycle.

The 0.5 division that you've read about is "zero accuracy", which is simply that a beam balance must zero within 0.5 division. It does not apply to digital scales.

The best bet is always to find the manufacturer's specification. In the absence of manufacturer's specification, you will have to research what tolerance is required of the unit, and calibrate to that tolerance.

Ryan

[Ryan Wilde]

M Greenaway said:

Thanks Ryan

But why calibrate agianst a known weight when we do not count against a known weight ?

Herein lies a problem. A calibration versus a known weight is not to prove a count, it is to prove the linearity and repeatability of the scale. The NOMINAL weight of the piece must also be known to properly set the scale (for minimum uncertainty).

We do not say that, for example, 100 pieces weighs 100 grams. so if we weigh out 1 kg we must have 1000 pieces. What we do is a manual count of a sample of say 100 pieces, we put the counted sample onto the scales and then set the scales count to 100. Then we tip in the product up to the required quantity.

To my mind any relationship to true weights is totally irrelevant.

Remember, you are counting, but the scale is weighing and using a bit of math to convert to a count. You are, in fact, taking the possible per piece variation and making it the accuracy of the scale. Here is an example:

Number of pieces: 100
Possible variation of weight per piece: 1%

You manually count out 100 pieces from Lot A, and set the scale using the product, which may be 1% below the nominal weight of 100 pieces.

You now count out 100 pieces of Lot B using the scale, and Lot B had a bit less slag in the steel, and weighs 1% more per piece than nominal. You will count 100 pieces, but in actuality, you will have 98 pieces, due to the weight variation. We have not even used any scale linearity error, and we have 2% error. If you now count 5000 pieces of Lot B, you would measure 5000, but actually have 4900 pieces, and that would be with perfect scale linearity. This may be acceptable to your process, it may not.

Example 2:

You calibrate the scale using standard weights. You set the scale using the nominal per piece weight as your reference (the nominal weight of each piece would have to be known). Lot A would measure 100, but actually contain 101, and Lot B would measure 100, but actually contain 99. Your error is halved.

No measurement will ever be perfect, but calibration against known weights reduces the error contribution of the scale itself, leaving the major component of counting error as that of the weight of the piece itself.

Ryan

[M Greenaway]

Ryan

I can appreciate that linearity and repeatability are vitally important for weigh count scales, but I cannot see how accuracy to known weights matters in this application.

Also it would be practically impossible to pick at random 100 pieces all of the same weight. Our sample size would in all probability be representative of the range of weights of the pieces would it not ?

Therefore by taking a count of say 100 pieces and setting our scale to 100 pieces we are in effect taking the actual mean weight of these 100 pieces. What this actual weight is is totally irrelevent, if it were given by the scales as 100 grammes then 1000 pieces would weigh (according to the scales - if linear) 100 kg. Also if the scales weighed the pieces at 1 gramme then 1000 pieces would weigh 1 kg. Whether or not the pieces were actually 100 kg or 1 kg doesnt matter in the slightest. All that is important is the ratio of the weight of the sample to the weight of the total count required.

I talked to my former boss where we used scales to count hundreds and thousands of small brackets made of titanium, stainless and carbon steel and aluminum. Every six months the scales are calibrated by an outside source. Did not dig into the method the lab uses.
1. Weigh the container empty
2. The scale has a function called "Tare" which zeros the machine with the container on it.
3. Put in one piece-"Tare" again.
4. Count 9 more pieces and put them into the container. It must read 10.
5. Add pieces until it reads the amount you want to ship. Great for things like washers. I have counted another 10 and see 20 on the readout. Now I feel better.

For the layman, such as myself, I would only feel comfortable with trusting the count when I periodically calibrate the machine to insure that the conversion mechanism is functioning properly. Whether it be against a "standard" weight suggested by the manufacturer of the scale, or using, for example, 10 or 100 "standard (certified)" weights to duplicate the way we use it.

JMHO Calibrate it!

[David Mullins]

MG
as this thread is under 17025, I think a few of us are assuming a Lab environment.
If you want to weigh against a counted number of parts, go right ahead.
Have you tried holding 100 parts counted in your left hand and then picking up about the same size handful in your right? This would probably prove just as, if not more, accurate. (If you have 2 free hands of course)

[M Greenaway]

Thanks Dave

I think your sarcasm actually hits the nail right on the head. I could perform the same function using a simple balance - calibration is not required !

[Ryan Wilde]

M Greenaway said:

Ryan

I can appreciate that linearity and repeatability are vitally important for weigh count scales, but I cannot see how accuracy to known weights matters in this application.

Here we agree, the actual accuracy is all but moot, BUT calibrated weights are the only method to prove linearity.

Also it would be practically impossible to pick at random 100 pieces all of the same weight. Our sample size would in all probability be representative of the range of weights of the pieces would it not ?

That depends on a few factors, basically. Is your sample taking equal amounts of pieces from separate lots? Does it take from different lots of raw material? I've been doing metrology just long enough to know that the density of metals can vary significantly from batch to batch (Our Class 1 standard weights, which are very controlled throughout their manufacturing process, vary from the nominal density of stainless steel by 0.8%, and I have to compensate for it when calibrating scales due to the difference in bouyancy).

Therefore by taking a count of say 100 pieces and setting our scale to 100 pieces we are in effect taking the actual mean weight of these 100 pieces. What this actual weight is is totally irrelevent, if it were given by the scales as 100 grammes then 1000 pieces would weigh (according to the scales - if linear) 100 kg. Also if the scales weighed the pieces at 1 gramme then 1000 pieces would weigh 1 kg. Whether or not the pieces were actually 100 kg or 1 kg doesnt matter in the slightest. All that is important is the ratio of the weight of the sample to the weight of the total count required.

Again, the actual weight is not an issue, you are correct. The mean weight, as I said, would have to be proved, which I would highly suggest. I also agree that the ratio is what your scale computes. I would also caution that any error in the 100 piece tare is now multiplied by 10 at 1000 pieces. For the ratio to be most accurate, the tare would be on a larger sample than that which is to be used, to minimize variation from piece to piece. A sample of 1000 would provide a much closer mean than a sample of 100.

Another problem is that as the scale approaches 0 parts, it must become increasingly more accurate (as in % of reading, not % of span). Your "calibration" method would actually be most accurate at the higher readings (larger sample), and become less accurate at lower readings, as possible variance from mean would increase with a smaller sample. I do not see how the scale linearity can be proven by any means other than calibrated weights, as it also proves the capability of the ratio of the unit. How would you prove the linearity of the scale?

There is one more point on this that should be made. In the USA, if the counting scale is being used as the final method to count parts that are to be sold to customers, then it is being used for "legal trade", and by law MUST be calibrated with traceable standards.

This is fun!

Ryan