Tolerances +/- 1 standard deviation?

A

Adam S

#1
I am upgrading and automating a measuring process and this has lead me to examine the coding lying behind a computer-based data capture system that has been in place for a while.

ie various parameters of parts are measured at an AQL, the data is manually entered into the PC and passed/failed on the basis of maipulation of the data entered.

I am looking to have the process automated, removing the manual entry of the data, and have discovered a bit of code in the data handling that stipulates that if the average value of parameter X for the sample exceeds the upper and lower tolerances +/- 1 Standard Deviation , the batch is failed.

ie the specification for paramater X is 10-20, but this data criteria widens the criteria to a lower tolerance of 10-1 SD and the upper tolerance to 20+1 SD.

I don't understand why the creator of the code would have widened the tolerances by 1 SD as they did, but I'm uncomfortable removing the code without justification.

Any opinions?
 
Elsmar Forum Sponsor

Jen Kirley

Quality and Auditing Expert
Staff member
Admin
#2
A computer program for recording inspection could be written by someone who does not understand that tolerances are usually set in dimensions, not a statistical derivative like standard deviation.

If I was using the system, I would expect to be able to enter my own tolerances. Was this system made for this process only, or was it made to be used for different kinds of processes? Was it made by someone in the company, or did you buy the program?

In short, the correct tolerances must be followed. If the software cannot be adapted, it should not be used.

If someone inside your company is capable, he/she could change the code. If it was purchased, the seller might be asked to make the changes and even include a feature for entering your own tolerances. In either of these cases, be sure to test the changed software thoroughly before shipping any product it was used on.
 
A

Adam S

#3
I believe the program was developed by us, but the individual responsible and any justifications behind the addition of the +/- SD to the tolerances are lost to time

So I can be confident that should I go ahead and authorise removal of the SD widening, and just stick with the tolerances, as long as I make mention of and justify the reasons for the removal (ie "adds no value and unnecessarily widens tolerance"), I'm acting with due diligence.
 

Bev D

Heretical Statistician
Staff member
Super Moderator
#4
It woudl be helpful if you could attach some actual examples. the real data of a sample that passed and a sample that failed. then we can assess what is actually happening.

I'm also not clear on the criteria.you state that the sample is done on an AQL but the acceptance criteria is based on the average of the sample? this just doesn't make sense to me. is the AQL an Attributes or Variables system? what data is used to calculate the standard deviation? The sample being assessed or past samples?

where did the 10-20 limit come from? Tolerances shoudl come from the physical requirement of the product; what is the requirement of the product?
 
A

Adam S

#5
Sorry if I haven't been clear but it’s difficult for me to give exact data as I'm not involved with the details of the specification or tolerances per se, it's the principle of understanding why the +/- 1 SD has been added and deciding whether or not to remove it from the revised data manipulation.

I'll give an example, the parameters being measured are dimensions of boxes:

The current system:
A measuring equipment measures height, length and depth of boxes, at a specified sampling rate (as per ISO2859-1/BS 6001)
These parameters are read off the machine by the operator and manually inputted into a computer
Computer software manipulates the data and accepts or rejects the batch based on the parameters entered

We are revising the above system; now, the measuring equipment pushes the data straight into the computer for manipulation, cutting out the manual data input stage

While reviewing the manipulation software as part of the change process, it was noticed that there is a line of code in the data handling software that compares the average height of each batch to the height tolerances, with any batches with a height greater than the tolerance being rejected, which seems a reasonable function to have. On closer inspection it was noticed that the average height data was not being compared to the tolerances for each box (eg 10cm-20cm), but was actually being compared to height tolerances widened by one standard deviation, ie the lower tolerance was 10cm - 1 SD and the upper tolerance 20cm + 1 SD

The SD is calculated from the batch height measurements

Hope this helps!
 

BradM

Staff member
Admin
#6
Hello, Adam! Thanks for being patient with us!:)

My first reaction was to remove that line of code. I believe there should be a controlled method for you to enter whatever tolerance you require for your process.

1. Don't let the software dictate your quality decisions.
2. If it is some value that most in your organization has trouble understanding/defending, that sounds like future trouble.
3. Tolerance/requirements are subject to change. Too, when you grow, you may have customers with different requirement.

I say revise the software and establish a password-driven specification input for it to calculate the acceptance.
 

CarolX

Super Moderator
Super Moderator
#7
I, too has a similar problem with my data collection software - the system has defaults that assumes your process is within certain control parameters.

I have to manually set my zones.
 

Tim Folkerts

Super Moderator
#8
I wonder if this was some form of "guardbanding" -- an effort to be SURE the sample is bad rather than rejecting borderline cases.

For example, if the specs say the average height should be 10-20 cm, and you pick out a few items to check and the average is 20.2, it is possible that the the actual average for the whole batch is indeed between 10-20 cm. Perhaps by bad luck, you happened to choose a few unusually large items. To give the "benefit of the doubt" to the batch, you might reasonably widen the tolerances a little. (Or if you wanted to be "extra sure", you might reasonably narrow the tolerances a bit! )

However, simply adding 1 st dev would not be the best way to do this. The standard error of the mean for the sample would be SE = (sd dev) / n^(0.5). You would want to see how many SE's you're beyond the spec limits. if the batch is actually just inside the spec, then drawing a sample that is 1xSE outside the specs would be fairly common; drawing a sample 2xSE outside the specs would be unusual; drawing a sample 3xSE outside the specs would be quite rare.

A better solution would be to set the limits to (upper spec limit) + 2SE and (lower spec limit) - 2SE.

For example, suppose the USL is 20 and you draw a sample with a mean of 20.5 and st dev of 1. According to your current rule, this would be acceptable.
  • If the sample size is 4, the SE = 1 / 4^0.5 = 0.5. you are 1 SE from the spec, so this would be acceptable by my rule. (You are not confident the mean is truly over 20.)
  • If the sample size is 16, then SE = 1 / 16^0.5 = 0.25. You are 2 SE from the specs, which is borderline by my rule.
  • If the sample size is above 16, then the sample would fail my rule. (now you are pretty confident the mean is truly over 20.)
Tim F
 

Bev D

Heretical Statistician
Staff member
Super Moderator
#9
I wonder if this was some form of "guardbanding" -- an effort to be SURE the sample is bad rather than rejecting borderline cases.

For example, if the specs say the average height should be 10-20 cm, and you pick out a few items to check and the average is 20.2, it is possible that the the actual average for the whole batch is indeed between 10-20 cm. Perhaps by bad luck, you happened to choose a few unusually large items.
well YES. that is one good approach to esnuring the lot average is truly out of spec. (although typically we care about the the number of individual values that are out of spec. but if the characteristic beign measured only has to be roughly in spec, lot averages can do the job quite nicely. However, for BEER fill volumes this woudl be a disaster!)

This approach would be akin to ensuring that the lower confidence interval of the sample was ABOVE the spec limit for the averages...

There are better ways of doing this of course and it will be interesting to seeing the OPs data if he can post it.
 
Thread starter Similar threads Forum Replies Date
A Is there a standard on tolerances of substances ? Medical Device and FDA Regulations and Standards News 0
Skooma Deburring and Tolerances - Is there a Standard for deburring or removing material? Manufacturing and Related Processes 3
Q M-282 - Engineering Standard for Tolerances - Is it ASTM, DIN? ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 4
W Thread Micrometer Standard Tolerances General Measurement Device and Calibration Topics 4
BradM Pipettes - Using ISO 8655 standard tolerances instead of manufacturer's tolerance General Measurement Device and Calibration Topics 27
K What the industry standard is for attribute gage tolerances - In Writing General Measurement Device and Calibration Topics 7
R Tolerances for Analytical Balances General Measurement Device and Calibration Topics 3
B FAI and ANSI Y14.5 Trailing Zeros and Print Tolerances Manufacturing and Related Processes 3
H Thread Plug Gauge Tolerances General Measurement Device and Calibration Topics 3
W Key functional characteristic definition - Tolerances and Deviations Definitions, Acronyms, Abbreviations and Interpretations Listed Alphabetically 0
M Best Practice for setting tolerances on a Drawing Inspection, Prints (Drawings), Testing, Sampling and Related Topics 8
DietCokeofEvil What is the general consensus on Caliper tolerances? General Measurement Device and Calibration Topics 1
S MSA (Measurement System Analysis) - Process Tolerances Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 9
J Tighter Tolerances than Control Plan FMEA and Control Plans 1
P Applying Combined Tolerances of Two Scales - Balance Tolerance Question General Measurement Device and Calibration Topics 9
G MSA (Measurement System Analysis) and Dimensional Tolerances Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
B Changing dimensional tolerances to approve rejected product ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 2
K Tolerances on Drawing Title Block Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 3
V GD&T Interpreting Datums in Two Single Segmented Position Tolerances Inspection, Prints (Drawings), Testing, Sampling and Related Topics 5
C ZZ Pin Gage Tolerances and Calibration General Measurement Device and Calibration Topics 2
D ISO 8062-3 Angle Tolerances - The drawing states ISO 8062-3 DCTG 11 Other ISO and International Standards and European Regulations 1
M Thread Gage Tolerances ?Who?s The Authority?? General Measurement Device and Calibration Topics 9
N Injection Molded Nylon Part Tolerances Manufacturing and Related Processes 6
A Ppk Issue - Some Products have Unilateral Tolerances Capability, Accuracy and Stability - Processes, Machines, etc. 24
H GD&T - How to measure these Positional Tolerances on a CMM Inspection, Prints (Drawings), Testing, Sampling and Related Topics 10
S Improper Use of Tolerances & SPC Practices, or not? Misc. Quality Assurance and Business Systems Related Topics 5
M Vinyl and Cloth Cut Parts Tolerances? (Car Seats) Manufacturing and Related Processes 8
G Geometric Tolerance - Swedish Drawing which states that all tolerances are to NS 1430 Manufacturing and Related Processes 2
optomist1 ASME GD&T Certification Test - Converting Coordinate Tolerances to Profile Tolerances Inspection, Prints (Drawings), Testing, Sampling and Related Topics 8
S S-Type Tolerances: How to calculate Cs? Control Limit(s)? Statistical Analysis Tools, Techniques and SPC 8
S Typical Tools for Particular Tolerances Table? Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 15
M Recording True Positioning (Geometric Tolerances) on an FAI Report Records and Data - Quality, Legal and Other Evidence 15
D Statistical Tolerances in GD&T (Geometric Dimensioning & Tolerancing)? Various Other Specifications, Standards, and related Requirements 31
QMMike Quoted Tolerances Trumping Drawing Tolerances? Contract Review Process 13
D DIN 1683 Casting Tolerances - What does "GTB" mean? Other ISO and International Standards and European Regulations 7
P Gauge Design Tolerances - How to decide tolerances for gauge design? General Measurement Device and Calibration Topics 3
smryan Can Cpk be calculated without tolerances or limits? Capability, Accuracy and Stability - Processes, Machines, etc. 11
D Metric Print from Italy - Tolerances - 38.8mm -0.015/-0.045 Inspection, Prints (Drawings), Testing, Sampling and Related Topics 5
O Sum of Cpk - Updating Product Tolerances Capability, Accuracy and Stability - Processes, Machines, etc. 6
K Basis for Establishing Gage Tolerancing - "Gage Maker's Tolerances" General Measurement Device and Calibration Topics 4
E How to Determine Tolerances for Equipment Verification General Measurement Device and Calibration Topics 18
F Capability Calculation on 1 Population Multiple Tolerances Capability, Accuracy and Stability - Processes, Machines, etc. 5
L UN Class 5 Tolerances - Thread Setting Plug Gage 7/8-9 UNC 5 General Measurement Device and Calibration Topics 23
M No Tolerances given by Customer - Shirring (sewing operation) IATF 16949 - Automotive Quality Systems Standard 5
M DIN6930 What table to use - Tolerances of straight line and radii measurements Inspection, Prints (Drawings), Testing, Sampling and Related Topics 4
D Adjusting Tolerances because of available tooling? Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 9
U Calculating Pp for Unilateral Tolerances Statistical Analysis Tools, Techniques and SPC 17
S Opening Up Tolerances - We have over 6000 live drawings - help please Design and Development of Products and Processes 6
T How to Calculate Process Capability for Unilateral Tolerances? Statistical Analysis Tools, Techniques and SPC 21
R Definition Basic Dimensions - What are basic dimensions? Why don't they have tolerances? Definitions, Acronyms, Abbreviations and Interpretations Listed Alphabetically 14

Similar threads

Top Bottom