SPC for Eddy Current Testing

#21
If you have two suppliers that vary from 5% (assumed from vendor 1) and 50% (assumed from vendor 2), I suppose you mainly buy from Vendor 1.
If this assumption is correct, you can plot all of Vendor 1 material separate from Vendor 2 material and see if there is something in your process that can take 50% down to 40%, or 5% down to 3%.
If you already have the data...its just a matter of analyzing it...a pretty cheap enterprise. Typically its the data gathering that costs the most.

5% to 3% might not be worth pursuing, but I'd guess 50% to 40% likely is. But again, at that point it's all cost benefit.
The suggestions above on SPC or at least data mining can help you understand if there is a potential benefit...then you'll have to run it against the cost to see if it's worth doing.

HTH
We already know which supplier is better,cause our operators are seeing daily which suppliers raw materials are producing more defective bars.Its just that sometimes our management finds a discounted raw material and buys in bulk.Cause even %50 percent fails crack control test,we can just sell %50 defective to normal customers(Defective bars actually not bad its just not good enough for automotive customers) and %50 OK materials to automotive customers.In the end we are even better of financially to buy low quality material sometimes.
 
#22
At Mike S - I'm not sure I would find good value in running 100% data through a control chart. Let's talk control limits. We would not expect 100% of the data to be within the control limits, so ... what good would they be? I think I would rather have a running mean and plot +/- 3 sigma limits of the last n samples (30, 60, selectable?) which would show me statistically where I was currently at. I could see range increasing statistically, and how close I was to my limits. I could even have a rolling calculation of my estimated defect percentage based on a normality assumption pretty easily.

Heck, I could run trends of kurtosis and skewness over time of the last 100 or 300 readings and see how that was tracking. If it shifted, there better be an assignable cause (adjustment made, tool changed, something) if I was expecting normally distributed data.

At BevD - I'm being pedantic, but once you are plotting 100% of the data, can you still call it an SPC chart? Isn't it just a run chart at that point?
 

Bev D

Heretical Statistician
Staff member
Super Moderator
#23
ncwalker - of course you can apply control limits to 100% data plotted in time sequence. in fact, the IMR chart is specifically intended for this. there is and never was any restriction regarding sample size for control charts. Sampling is simply a convenience for high volume production.

And of course there will be points out of the limits - this happens with sampling too. that is why the limits are there: to detect non-homogeneity.

the goal is NOT to have all of the points inside the limits - especially in the beginning. it is to determine when assignable causes occur and to fix them. (the OP is saying that in his case this isn't very probable or cost effective)

Statistical control limits (not +-3sigma of the individual values) are the perfect analytical tool for time series data. And remember most data are not normally distributed nor should they be...so a bunch of 'sophisticated statistical analyses are simply not needed for analytic studies...

I have found that the collected works of Donald Wheeler are quite helpful in understanding this - unlike what a lot of what the hack teachers push (Deming's words not mine).

some articles:
Wheeler, Donald, “Probability Models do not Generate Your Data”, Quality Digest, March, 2009 http://www.qualitydigest.com/magazine/2009/mar/department/probability-models-don-t-generate-your-data.html

Wheeler, Donald, ”All Outliers are Evidence”, Quality Digest, May, 2009 http://www.qualitydigest.com/magazine/2009/may/department/all-outliers-are-evidence.html

Wheeler, Donald, “Why We Keep Having Hundred Year Floods”, Quality Digest, June 2013, http://www.qualitydigest.com/inside/quality-insider-column/why-we-keep-having-100-year-floods.html

Wheeler, Donald, “The Secret Foundation of Statistical Analysis”, Quality Digest, December 2015 The Secret Foundation of Statistical Inference | Quality Digest

Wheeler, Donald, “Statistics 101 and Data Analysis”, Quality Digest, March 2016 http://www.qualitydigest.com/inside/standards-column/030716-statistics-101-and-data-analysis-example.html
 

Ninja

Looking for Reality
Trusted
#24
Its just that sometimes our management finds a discounted raw material and buys in bulk.Cause even %50 percent fails crack control test,we can just sell %50 defective to normal customers(Defective bars actually not bad its just not good enough for automotive customers) and %50 OK materials to automotive customers.
Seeing this, I would look in three places immediately:

1. Do you meet all of the demands for your auto suppliers, or do you sometimes short ship due to not having enough crack free?
2. IF you meet all auto customers, do you also sell all of the others, or is there a big pile of them in a warehouse?
3. Do you sell both at the same price?

As I said, it's cost benefit. SPC and other approaches help you do that comparison...but its that comparison all is targeted to.

If you buy 50% crack material for a good price, and you sell it all at the same price you would have buying 5% crack material...there is no justification to fix anything.
If you pay $1.00 for 50% crack material, sell 50% of it for $4 and the other 50% for $1.10...you have a return.
If you pay $2.00 for 5% crack material, sell 95% of it for $4 and the other 5% for 1.10...you have a bigger return...do the math.

Only when you see that 5% crack material gives a better profit can you make any significant case to the mgmt who is buying the cheap stuff.
Cheap is often better...even with more waste.
The purpose of a process is not to eliminate waste, it is to maximize profit (Deming just rolled over and glared at me, sorry).
...before you yell at me, I understand that elimination of waste is one of the primary tools to maximize profit...but don't take your eye off the ball...
 

Mike S.

An Early 'Cover'
#25
We already know which supplier is better,cause our operators are seeing daily which suppliers raw materials are producing more defective bars.Its just that sometimes our management finds a discounted raw material and buys in bulk.Cause even %50 percent fails crack control test,we can just sell %50 defective to normal customers(Defective bars actually not bad its just not good enough for automotive customers) and %50 OK materials to automotive customers.In the end we are even better of financially to buy low quality material sometimes.
Okay, okay, I give up! There is no need for you to learn more about your process, and yields are irrelevant to you. :ko:

Look, part of the reason I continued to reply to this thread is someone, somewhere, just might be one of those rare people :rolleyes: who may have a similar problem and they may find value in learning more about their process and increasing their yields. I felt it important to make the point that, in my humble opinion, SPC could, in such a case, provide valuable insight. 100% sampling does not mean there is no value in using control charts. I am not an SPC expert, but there are a few basics I do know.

Bev pretty much nailed it in her most recent post. :applause:
 

Steve Prevette

Deming Disciple
Staff member
Super Moderator
#26
My father once related a story of buying heavy steel plate for fabrication use and it was swiss cheese riddled with holes that had been filled with chalk and painted over black. Let it be said they found a way to deal with the supplier to express their displeasure even though the supplier was owned by the corporation that owned the fabricator. I am always wary of cases where folks feel they are "victims" of circumstance.
 
#27
Well, I have a reading list now. :)

You will all have to let me digest this - and I apologize in advance if I don't jump on the train. I'm skeptical by nature.
 

Mike S.

An Early 'Cover'
#28
Jump or don't jump, no apology needed.

I will just add this -- Dr. Wheeler is, IMO, the world's foremost expert on SPC, statistical analysis, etc. He is not a BS'er or a bandwagon guy. Some of what he posts is over my head, but the man is a freakin' genius. You could do much worse than to study some of his writings.
 
#29
Okay, okay, I give up! There is no need for you to learn more about your process, and yields are irrelevant to you. :ko:

Look, part of the reason I continued to reply to this thread is someone, somewhere, just might be one of those rare people :rolleyes: who may have a similar problem and they may find value in learning more about their process and increasing their yields. I felt it important to make the point that, in my humble opinion, SPC could, in such a case, provide valuable insight. 100% sampling does not mean there is no value in using control charts. I am not an SPC expert, but there are a few basics I do know.

Bev pretty much nailed it in her most recent post. :applause:
Sorry if i sounded ungreatful,every reply is appreciated.And i agree with what you said and it might be valuable for another person .Also since i'm relatively new in automotive industry,sometimes i cant decide if a requirement is not applicable or just that i cant understand it well.That's why i tried to explain my situation explicitly and with detail
.

And Bev D thanks for all the replies.i think im just going to wait for the next years audit and explain what i think to the auditor myself.And thanks for the reading list,i will definetly check those
 

Top Bottom