Results from conducting a Capability Studies

Hello
I have been tasked with preparing a work instruction for 'next steps' when it is determined there is an issue with the results from conducting a capability study.

I have investigated this but there seems to be no indicators of what to do next

Can anyone help

Thanks to all

Tora3484 said:

Hello
I have been tasked with preparing a work instruction for 'next steps' when it is determined there is an issue with the results from conducting a capability study.

I have investigated this but there seems to be no indicators of what to do next

Can anyone help

Thanks to all

Click to expand...

You really need to go back to the basics that should be addressed before doing a capability study.

1. Review the data. Is it "in control" on an SPC chart? If it is not "in control", deal with the trends / out of control points. Get the process stable and in control. By the way, a capability number on an "out of control" process is meaningless.

2. If the data are in control, check to see if there is a "gap" in the data just outside the spec limits. A lot of times, if the data are "close enough" operators will round the number off so it meets spec.

3. If the data are stable and predictable, and the histogram of the long term data is reasonable, now you need to start analyzing why the process isn't meeting specification.

3a. Is the specification correct? Is it "unreasonable"? Can the / Should the spec be changed?

3b. Look at lower level (leading) data that leads to the final spec data. For example, annealing temperature may have an effect on strength of the metal. What detailed data are available about the process that may correlate to the overall result.

3c. At an extreme, there is the Shainin Red X process where you compare a "good" part to a "bad" part and try to tell what led to the difference.

3d. Oh, and then there is the step most people forget - go to the "Gemba" (work location). Observe the work be done. TALK TO THE WORKERS!

Adding to Steve's post:

- make sure you are sampling appropriately for a capability study. small samples and non-representative samples form a non-homogenous process are both notoriously variable but not truly indicative of a non-capable process
- make sure you are taking rational samples; a non-homogenous process will look 'out-of-control' in a standard Xbar, R or Xbar, S chart when it is actually rock solid stable. See the attached presentation on Non-Homogenous Processes and SPC.
- understand the underlying distribution. non-normal distributions can have a small Cpk value because of the larger SD.
- Understand the specifications: were they engineered to ensured no stack-ups or other interactions? if so you are actually capable as long as there are no out of tolerance parts. If not you want to understand where the stackups occur to understand how much reduced variation you actually need.

If the process is truly incapable, you will then need to initiate problem solving to understand causal mechanism and develop a solution....

Thank you to Steve Prevette and Bev D

Just thought of this one: make sure you don't have "chunky data". (low resolution gauge that produces data where there are few possible results)
Chunky data will also inflate your SD resulting an artificially lower Cpk