Finding a flat or not modeling distribution: how to manage it?

[Marco Bernardi]

Hi all,

if during a feature analysis the distribution that comes out is a flat model, how can be managed it?
How can be considered and then analyzed respect the specifications?
And if the distribution is not flat but however not modeling, how can be managed?

 

[bobdoering]

First step is to understand whether it makes sense to be "flat" (uniform distribution?) Also need to know if the distribution is continuous or discrete. That comes from looking at the time-ordered sequence data of your capability study, as well as understanding what the process actually is. Can you supply sample data, and a few more clues?

 

[Miner]

In addition to Bob's comments, is the process stable or do you have a mixture of multiple process streams?

 

[optomist1]

for those of us....what is a flat model? Is Bob D's statement spot on?

 

[Miner]

for those of us....what is a flat model? Is Bob D's statement spot on?

It is most likely a uniform distribution, possibly a platykurtic distribution. The most likely causes are process drift from tool wear, unstable processes and mixtures of multiple process streams.

 

[Marco Bernardi]

First step is to understand whether it makes sense to be "flat" (uniform distribution?) Also need to know if the distribution is continuous or discrete. That comes from looking at the time-ordered sequence data of your capability study, as well as understanding what the process actually is. Can you supply sample data, and a few more clues?

The issue was made with a generic topic, not related to a specific case, however, the distribution considered is a continuous one (which case could be with a discrete distribution in an industrial production process?).
Flat distribution is exactly referred to a uniform type distribution, then, is it have to be intended that with a such result, or it's due to a wearing tool condition nor by the way, a clear symptom of out of control process? Then verify the data with a control chart and understand if it's out of control, then try to understand the special causes and remove them? Could it possible a uniform distribution with an in-control process?

 

[Miner]

Bob is definitely the expert on this topic and has many posts and resources on this forum that you can search. He has developed a special type of control chart for processes such as these were tool wear is an expected trait of the process.

 

[Bev D]

I will add that it's not the distributions that matters (although it can provide some insight) it is the run sequence of the data and the knowledge of the physics and actual events of the process that provide the appropriate insight. Remember as Donald Wheeler has said: "Distributional models don't create your data, your process does."

and yes a uniform distribution can be an 'in-control' process. tool wear is a stable and predictable event. if the wear is unexpectedly fast - or slow - that is an 'out of control' process and can be detected by the appropriate type of chart.

 

[Marco Bernardi]

I will add that it's not the distributions that matters (although it can provide some insight) it is the run sequence of the data and the knowledge of the physics and actual events of the process that provide the appropriate insight. Remember as Donald Wheeler has said: "Distributional models don't create your data, your process does."

and yes a uniform distribution can be an 'in-control' process. tool wear is a stable and predictable event. if the wear is unexpectedly fast - or slow - that is an 'out of control' process and can be detected by the appropriate type of chart.

Thank you Bev D for the explanation, but it can be an "in control" process then, only for tool wearing cause? If it's in control, then, how can be considered with process capacity analisys? Is it assimilable to a specific distribution model and capability index calculable?

 

[Miner]

I recommend reviewing Bob's blog series for more insight.