SBS - The Best Value in QMS software

Use of an outlier in calculating Cpk/Ppk

  • Thread starter duecesevenOS - 2009
  • Start date
D

duecesevenOS - 2009

#1
Our office has argued back and forth about this and i don't think we have ever come up with a solid answer. If you have a process running very well and very stable. Our Cpk's are outrageously high like 4.0 for about a month. And then a punch tip breaks without stopping the machine.

So when the operator does his/her inspection he/she finds a dimension that is way out of the spec limits. The operator throws out all of the parts produced between the last good inspection and the bad one and then replaces the punch. Then the operator starts running again and everything is once again very good.

Except the Cpk and Ppk on SPC is now showing around .2 . We are not worried that we have supplied our customer with bad product but we don't know if the outlier should be used to evaluate the process. Punch tips breaking is a part of the actual process. It will happen on occasion and product will run out of spec when it does.

I just finished reading the long and old sticky post about Cpk vs. Ppk and it almost sounds like the answer might be yes and no. Am I right in saying that for Cpk we should throw out this outlier but for Ppk we should not? If Cpk is the process capability when instabilities are removed and Ppk is the actual process as it has performed with subsample variation then.

Either way this fairly uncommon outlier is going to make the population non-normal so using either Cpk or Ppk with the outlier has major issues. I'm just wondering what you guys think should happen in this situation. Do you throw the data out for a Cpk analysis? Do you throw it out for a Ppk analysis?

We don't currently have acceptance based on Cpk in our plant but will be going to it in the next few years. These are the questions that boggle the forward looking people around.
 
Elsmar Forum Sponsor

Tim Folkerts

Super Moderator
#2
My gut reaction is that you should in this case ignore the bad data points and leave them out of the calculations. :mg:

Let me explain! Your process is not simply to make the parts and blindly ship them. If that were the case, I would say keep the data points in the calculations and live with the reduced Cpk & Ppk.

However, your process is to make the parts, then test the batch. Testing is part of the process of preparing parts. Think of it this way - the people at your shipping docks don't know what went on with the parts before they arrive for shipping. To them - and to your customers - the parts that actually get shipped ARE the only parts that were produced.

To be statistically rigorous, it might be better to draw a different sample for "testing" as part of the production process versus "quality assurance" and calculating capabilty, but that is probably overkill.

I'll be interested to hear if others agree with me...


Tim F
 

Miner

Forum Moderator
Staff member
Admin
#3
Any metric such as Cpk or Ppk is based on the assumption that the process is in a state of statistical control. In your situation, you have a special cause for which you know the reason. You should not include the outlier(s) from this special cause in your calculation.

The difference in calculation between Cpk and Ppk was intended to pick up the slight wandering of the process around the center line. Not a blatant special cause such as a broken punch.
 

Statistical Steven

Statistician
Staff member
Super Moderator
#4
Those observations are NOT outliers. They are assignable causes. Assignable causes would not be used in the calculation of Cpk or Ppk. I would do an analysis of number of parts before a punch break and institute preventative maintenance to replace the punch before they break.
 
C

Clausterphobic

#5
Statistical Steven said:
Those observations are NOT outliers. They are assignable causes. Assignable causes would not be used in the calculation of Cpk or Ppk. I would do an analysis of number of parts before a punch break and institute preventative maintenance to replace the punch before they break.
:agree1: I agree with what Tim, Miner and Steven said (that the "abnormal" reading should be excluded from computation of Cpk and Ppk)... but I would like to discuss on the comment by Steven above.

Cpk and Ppk are intended to show the capability and stability of a process. Sporadic readings caused by assignable events, does not display the true capability of the process. In my previous work, whenever an assignable cause happens, the "outlier" reading is still recorded on the SPC chart. Then, a space provided at the back of the SPC form is filled-up to let the operator/technician/engineer(who conducted the investigation) write down the reason and what actions has been done.

When the SPC form is due for submission to the company statistician, he will be able to trace the reason of the outlier reading therefore, it will not be included on the computation.

:topic: sorry i cannot share with you the spc form since i'm not with company anymore.:( on the other hand, i will try to recreate it(if i can still remember:bonk:) and post it in this thread :bigwave:
 
D

duecesevenOS - 2009

#6
Thanks guys. This sounds like the conclusions we came up with. Leave it on SPC but do not use it to calculate Cpk/Ppk. :)
 
D

Dave Dunn

#7
duecesevenOS said:
Thanks guys. This sounds like the conclusions we came up with. Leave it on SPC but do not use it to calculate Cpk/Ppk. :)
Going a step further, I would also make sure to note on the control chart the assignable cause, and to not use that point in future control limit calculation.
 

Bev D

Heretical Statistician
Staff member
Super Moderator
#8
one more input: if your Cpk/Ppk values are for external reporting to a customer - you could remove the assignable causes.
BUT if you really are using the Cpk/Ppk values internally to judge the true process capability and prioritize & assign improvemetn projects (or even to just understand your cost of poor quality) I would leave the data in! My thought process: Although the broken punch has an easily assignable cause it is still part of the process - it occurs regularly (inferring from post and my previous experience in these types of operations) and results in scrap. It's presence also requires you to have inspection to detect and remove these nonconforming items. At some point this shoudl be improved.

Although punches do break - they often break with stunning predictability. You can plot the number of units between breaks ans see if there is a stable 'life time' before breaks. then you can pre-emptivley (is that a word? I do make up words) change out punches jsut before they are likely to breal: reduced scrap and better process capability...

Use the data to help you - don't just hide it away if it is inconvenient. I have foudn that inconvenient data is often the first palce to look for process improvements. The hidden factory is a wealth of profit jsut waiting for us to find it!
 

Statistical Steven

Statistician
Staff member
Super Moderator
#9
Mr

Bev D said:
one more input: if your Cpk/Ppk values are for external reporting to a customer - you could remove the assignable causes.
BUT if you really are using the Cpk/Ppk values internally to judge the true process capability and prioritize & assign improvemetn projects (or even to just understand your cost of poor quality) I would leave the data in! My thought process: Although the broken punch has an easily assignable cause it is still part of the process - it occurs regularly (inferring from post and my previous experience in these types of operations) and results in scrap. It's presence also requires you to have inspection to detect and remove these nonconforming items. At some point this shoudl be improved.

Although punches do break - they often break with stunning predictability. You can plot the number of units between breaks ans see if there is a stable 'life time' before breaks. then you can pre-emptivley (is that a word? I do make up words) change out punches jsut before they are likely to breal: reduced scrap and better process capability...

Use the data to help you - don't just hide it away if it is inconvenient. I have foudn that inconvenient data is often the first palce to look for process improvements. The hidden factory is a wealth of profit jsut waiting for us to find it!
Exactly the approach....use the data to change the punch BEFORE it breaks. Then the issue of bad data in the Cpk goes away.
 
Thread starter Similar threads Forum Replies Date
P Are we an outlier for so many variations of Change Management? ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 4
M What is the difference between an Outlier versus Out of Specification ? Statistical Analysis Tools, Techniques and SPC 17
B Grubbs test to determine one outlier to normalize data Statistical Analysis Tools, Techniques and SPC 4
M Lean: Conducting Capacity Study, calculating Cycle times on laser cutting machines Lean in Manufacturing and Service Industries 2
A Calculating Carbon Footprint in the Organization Miscellaneous Environmental Standards and EMS Related Discussions 2
C Calculating your KPIs ISO 13485:2016 - Medical Device Quality Management Systems 7
K Calculating Heat Dissipation Manufacturing and Related Processes 1
Q Calculating Defect Rates ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 4
S Formula for Calculating NoGo Major Diameter for UN gages Inspection, Prints (Drawings), Testing, Sampling and Related Topics 4
P Calculating a weight for a machine for the CE label CE Marking (Conformité Européene) / CB Scheme 0
R Remote Support - Calculating Number of Employees IATF 16949 - Automotive Quality Systems Standard 2
G Calculating Tolerance of a Panel Meter with a 2 degrees of an Arc General Measurement Device and Calibration Topics 0
T Calculating LCL, UCL, Cp, and Cpk in an Excel Spreadsheet Statistical Analysis Tools, Techniques and SPC 3
V Product Development - When to start calculating Process Capability Capability, Accuracy and Stability - Processes, Machines, etc. 3
J Calculating Reliability for Subsystems in Series Reliability Analysis - Predictions, Testing and Standards 15
Stijloor Calculating (3rd Party) Audit Days for Company with Seasonal Employees General Auditing Discussions 3
T Formulas for Calculating Coefficients for RTD / PRTD Calibration According to ITS-90 General Measurement Device and Calibration Topics 5
J Calculating part variation from historic data for GRR study Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
P Purpose of calculating Ta, Tb & R Square in Linearity Study Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
P Purpose of calculating Uncertainty value in calibration study Measurement Uncertainty (MU) 5
M Calculating MTBF for plug-in PCI cards Reliability Analysis - Predictions, Testing and Standards 2
J Calculating impact force, expressed in lbs, of a 5 lb weight dropped 48 inches Inspection, Prints (Drawings), Testing, Sampling and Related Topics 1
M Calculating Cpk when sample size equals to 1 Capability, Accuracy and Stability - Processes, Machines, etc. 12
D Calculating Cpk on Non-Normal Data Distribution Capability, Accuracy and Stability - Processes, Machines, etc. 10
K Calculating Capability of a process when data is skewed in the upper range Capability, Accuracy and Stability - Processes, Machines, etc. 4
A Calculating Accuracy for a Multimeter General Measurement Device and Calibration Topics 1
M Calculating Coefficients for an RTD (Resistance Temperature Detector) Probe General Measurement Device and Calibration Topics 3
T Calculating Plating/Coating Weight for IMDS RoHS, REACH, ELV, IMDS and Restricted Substances 3
D Calculating Lower and Upper 3 Sigma Control Limits Statistical Analysis Tools, Techniques and SPC 2
M Calculating Adequate Receiving Inspection Sample Size Statistical Analysis Tools, Techniques and SPC 2
M Calculating Capability of Delivery Performance Capability, Accuracy and Stability - Processes, Machines, etc. 5
K Advice on Calculating Control Chart Control Limits Statistical Analysis Tools, Techniques and SPC 13
B Calculating Combined DPMO and Sigma Level for Two or More Different Work Areas Six Sigma 3
G Calculating the Fraction Defective Confidence Interval for a Lot Statistical Analysis Tools, Techniques and SPC 2
A Calculating Combined Measurement Uncertainty - VDA 5 Measurement Uncertainty (MU) 1
B Calculating Precision to Tolerance Ratio Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 9
D DPPM - Calculating the Defect Rate of a Software Test Statistical Analysis Tools, Techniques and SPC 9
L FMEA - Calculating RPN = S*O*D FMEA and Control Plans 3
I Shewhart Constants vs Central Limit Theorem in calculating Control Limits Statistical Analysis Tools, Techniques and SPC 18
B Please share a template for calculating Cp Cpk Document Control Systems, Procedures, Forms and Templates 3
S Is a Stable Process (within Control Limits) required for Calculating Pp, Ppk? Capability, Accuracy and Stability - Processes, Machines, etc. 6
M Calculating the IATF TS 16949 Recertification Audit Man Days - Clarification ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 6
J IEC60825 - Laser Safety - Calculating Maximum Permissible Exposure Correctly Other ISO and International Standards and European Regulations 1
P Calculating Process Capability from Data of attached spreadsheet Statistical Analysis Tools, Techniques and SPC 4
I AQL Formula in Excel - Calculating Sample Size based on an AQL Chart AQL - Acceptable Quality Level 8
D Formula for Calculating Warp Measurement of Corrugated Boxes General Measurement Device and Calibration Topics 1
S Need Help Calculating and Implementing Cs - Wright's Index Statistical Analysis Tools, Techniques and SPC 11
S Calculating an Uncertainty Budget for an Optical Comparator General Measurement Device and Calibration Topics 4
D One-Two-Three Sigma Control Limits - Calculating Control Limits for X-bar Charts Statistical Analysis Tools, Techniques and SPC 3
K Calculating (Approximate) Parallelism Inspection, Prints (Drawings), Testing, Sampling and Related Topics 10

Similar threads

Top Bottom