Failure Mode SPC

[sanjoh]

I want to graph the failure modes (A, B, C) on a tensile test A is worst, C is best. The test is performed on each production lot. Anyone have any suggestions on what the approprate tool would be?

The goal is to detect shifts in the process that affects failure mode.

Thanks

JS

 

[Geoff Cotton]

Sanjoh,

Why not change "A B C" and make them equal "1 2 3" (or some other score)?

Then its back to good old stats, or am I missing something in your question?

 

[sanjoh]

Geoff,

That works ok for general trends. It leads to lots of false positives when using xbarnr due to the low "measurement" resolution.

John

 

[Bill Ryan - 2007]

Would a simple bar graph, with 3 bars (1 bar for each category) for each lot (each lot being noted on the X-axis) show what you are looking for?

BTW - Welcome to the Cove

Bill

 

[M Greenaway]

The question sounds a little confused.

If you want to monitor the types of failures you get then I suggest an attribute control chart with proportion defective, showing a split of the proportion by failure mode.

If you are creating a linear A to C scale then, as already suggested, you could turn this into figures and plot variables.

I dont understand the 'failure modes A B C, A worst C best' bit of the statement. A failure mode is not quantative in itself.

How are you doing this tensile test ?

If its measured with a force gauge to the point of failure then a variables chart is best.

If you cannot measure the force, just have a pass or fail (break or not) to a prescribed fixed load then an attribute chart is best.

 

[Geoff Cotton]

John,

I'm a bit slow on the uptake, so could you re-phrase your question and help me out please.

I feel like I'm shovelling fog at the moment.

Welcome to the Cove, it's a good place to be.

 

[Darius]

Sanjoh,

First at all, welcome to the Cove

I tink it could help to change the A,B,C to a variable chart, If you are interested on SPC, you can as Geoff said "1 2 3", and use other methodologies like weighted moving average of the last 5-10 values to obtain the value to control, I agree that looks more like a forecasting, but I tink it could work fine.

ie. the sequence 1,2,1,2,1 with the weights 1,2,3,4,5 => (1*1+2*2+1*3+2*4+1*5)/(1+2+3+4+5) =1.4

It's important to weight more the last value, because it's value is more signifficant.

Good luck

 

[Bill Ryan - 2007]

It's important to weight more the last value, because it's value is more signifficant.

Could you please explain. I don't understand why any value would have more "weight"/significance in a tensile testing inspection. I work with die castings where the homogeneity (probably the wrong term to use) of the parts varies from cavity-cavity as well as shot-shot. Each value is just that - a value/reading. We don't use attribute charting of any kind but rather chart the variable values in an X-bar & R chart.

I have not ever had to use a (E?)WMA chart and, thus, have "lost" everything I learned about them for the CQE exam (9 yrs. ago ). I will try to review the methodology myself, but that sounds like it could hurt

Thanks,
Bill

 

[Darius]

Agreed that EWMA could be used, but it has it's complexity.

The different weights of a time series is used in forecasting to give more importance to some values (specially to the last) so the trend of the data is affected more from the last values.

For example if I am forecasting sales and I take the last 10 years to do the job, I can't give the same weight to the values 10 years ago and to the values of the last year/month, because the grow of the market, new tecnologies and many other factors.

WMA is one of the forecasting methods (the simplest ones, and technically is just smothing) but you can use EWMA, Winters or many other.

As I said, it is an uncommon aproach, but if Sanjoh has an SPC/Statictics program with EWMA, it could be a good idea to give a try, but I recomended WMA because it's easy to try in EXCEL and see if could give him what he was looking for.

Other way to do the job, Sanjoh said to Geoff about the direct use of 1,2,3 coding "That works ok for general trends. It leads to lots of false positives when using xbarnr due to the low measurement resolution.", is to change the rules of detection acording to his process. One important thing about rules is that not always apply (for example in chemical process were a measurement has too much to do with the last measurement).

By the way, you said "Would a simple bar graph, with 3 bars (1 bar for each category) for each lot (each lot being noted on the X-axis) show what you are looking for?"

Just a question, how a bar chart can give, as Sanjoh said "detect shifts in the process"?

 

[Bill Ryan - 2007]

Darius

Thank you for the explanation. I'm still a little "fuzzy" with it, but I'll let it try to sink in.

Just a question, how a bar chart can give, as Sanjoh said "detect shifts in the process"?

I understand your confusion because I'm not sure why I posted that . It doesn't make sense to me either. (Must have read Sanjoh's post and "heard" something else - those brain farts get more common as I age )

Have a great weekend!!!

Bill