Capability analysis - What is going on this chart?

[reunalla]

Hi All,

so I tried to find out more about this process but no luck. All I know is that it is from the same machine (same process stream), same production batch samples have been taken after each other in sequence.

 

[Miner]

Okay, so it is not a mixture, but you still have random process shifts that apparently self corrects. What kind of machine is this from?

What I am visualizing is something mechanically loose on the machine that randomly allows a physical setting to shift, but normal forces tend to keep it in the desired position. The abnormal points tend toward the same value as if there is a physical stop on either end that it bounces back and forth between, but favoring one extreme.

 

[Proud Liberal]

I still question the measurements themselves. Parallelism is easily corrupted by something as simple as dirt on a surface plate or burrs on the part. Since the "fliers" are so far from the tolerance, it would seem unlikely that anyone would take notice (unless you were actually using a control chart when performing a capability study).

 

[Bev D]

Yogi Berra famously said: "It's amazing the things you can see when you look". I use this a lot to remind people that you cannot solve problems - or understand capability - by processing a pile of data through software no matter how sophisticated the math or programming are.

 

[Steve Prevette]

I would recommend a better analysis of the data on hand. Aside from the points outside the UCL, there are a number of shifts occurring within the data that are ignored by just throwing all the data into one baseline average and UCL. At the very least, we should throw the obvious outliers out (the ones above the UCL) and recalc the average and control limits to see if the remaining data are in control. Eyeballing the data indicate several likely shifts where there are seven in a row above average / below average. Another piece is the termination of data at zero and it appears to be a highly skewed distribution. This is still okay for SPC (SPC does not require normality) but we may better understand the data by understanding its distribution, perhaps by taking the logarithm of the data.

True, I agree on lets find out about the process. But since someone paid to gather this data, let's milk everything we can from it.

 

[toniriazor]

What is your customer requirement for the capability study and what are you manufacturing? I always check myself first if data collection is OK and measured in a proper way. If data collection is poor, not logical and not measured ok, what you will get at the end will not reflect reality.