Process controls on attribute data - Chrome plating manual buffing line

[liquified56 - 2009]

Hey everyone,

I have a question on CPK i hope you can help with. We make chrome plated parts for the plumbing and auto industry and are currently using Control charts in our plating line and diecast line. In a recent Customer audit one of our primary customers mentioned they wanted to see some cpk data or process control data from our manual buffing line. I have alot of history with being able to run charts with actual dimensions to check(variable data) but im not sure how to do this on a purely cosmetic operation. I do have some existing data i can work off of though that we record on a daily basis. Mainly Parts produced/and rejected by operator/shift. Fallout is very high in this operation just due to its nature and we run a wide variety of parts so a cluster map or U chart i think its called isnt feasible.

What i am thinking is running data with a 0 LCL and a management set UCL and setting triggers at the UCL to notify the supervisor that something is happening with that particular operator. Over time we can adjust the UCL as the process tightens(people know we are monitoring). One of the main issues is the Cpk data will be off and the chart will be one sided considering we cant fall below 0 on the LCL. Any ideas or other charts/process studies i should be looking at instead where i can get a measureable cpk/cp?

Thanks for your help

 

[bobdoering]

Re: Process controls on attribute data?

In a recent Customer audit one of our primary customers mentioned they wanted to see some cpk data or process control data from our manual buffing line.

Cpk on manual buffing....oh, brother. To what specification on their print did they want this capability compared to? Was a reflectance range called out, or something? Cpk really only relates to bilateral tolerances, and typically with normal distributions.

Although measuring fallout may be statistical, it really offers little for control. That is what I am guessing is the customer's issue: How do you control a manual process? Training and testing? PM on the buffing wheels? Specification of compounds? Frequency of adding compound? Moreover, how do you prove these activities are effective - which is where the measurable would be handy.

Where are you at in the relationship with the customer? Do you have the order? If no, you might want to reconsider the cost of meeting what the customer is asking for - because it seems like they want a lot of overhead for free.

 

[liquified56 - 2009]

Re: Process controls on attribute data?

To what specification on their print did they want this capability compared to?

Not a print requirement a Quality System requirement for all processes related to their product.

How do you control a manual process? Training and testing? PM on the buffing wheels? Specification of compounds? Frequency of adding compound? Moreover, how do you prove these activities are effective - which is where the measurable would be handy.

Buffing wheels/adding compound is controlled in the preventive maintenance WI & operator WI. Training/effectiveness is done through CAR/NCMR history and operator production outputs. Also we have compound and wheel testing results in our PAR history.

Where are you at in the relationship with the customer? Do you have the order?

One of our biggest customers and where we want to grow our business. We are currently one of their best suppliers and this is getting into potential future/new parts and production.

 

[bobdoering]

Re: Process controls on attribute data?

Not a print requirement a Quality System requirement for all processes related to their product.

If there is no specification, there is no cpk. It is a fundamental part of the calculation! Their Quality System should not indicate requirement for all processes related to their product, it should indicate requirement for all critical specifications related to their product. I am surprised there is no measurable callout for the finish - especially for a end-customer visual area.

One of our biggest customers and where we want to grow our business. We are currently one of their best suppliers and this is getting into potential future/new parts and production.

That is understandable. I just feel the approach you are considering is adding a layer of effort to your system with little value added - just to meet their 'system'.

 

[liquified56 - 2009]

There are cosmetic finish requirements, spec analysis, dirt/pit size criteria, Surface grade callouts. Im trying to meet their requirements with data we already are recording so i dont overwelm production with more inspection/paperwork requirements.

 

[Bev D]

let's try an alternative approach. This is something that has worked effectively for me when dealign with manual processes and categorical requirements (aka attributes data)

I firmly believe that we need to monitor the performance of any critical process output and it appears that your customer feels this is a critical output for them. SO, first principles takes us to a simple run chart of the number of rejections per lot (p chart) or the number of discreet defects per part (c or u chart). IF the different parts are similar enough that the defect rate is more dependent on the process than the part (eg simple vs complex geometries) then a single chart could be used for all parts. only the actual data will tell us which way to go.

Now we can add REAL control limits using the formulas for the appropriate chart type. Just because you have categorical doen't mean SPC can't be used as intended - to assess the statistical control or stability of a process. I woulf run away from 0 and some arbitrary upper goal line as fast as possible - that flawed ineeffective approach was debunked decades ago and is exactly what SPC was intended to replace.

Next we can calculate a Cpk value by going backwards thru the Z table - calculate the ppm value, determine the corresponding Z value and divide by 3 Of course this isnt' a Cpk value in teh traditional sense of understanding the process spread vs the tolerances, adn I personally prefer to just deal with yield numbers over silly Cpk/Ppk values, but some customers just HAVE TO HAVE a single number: Cpk. So we can give it to 'em. Why argue the point? in this case the whole Normal distribution problem for continuous data that relates standard deviation to a ppm value is irrelevant. It's a small thing and everyone gets to check their box. The real value add comes from teh control chart itself: concentrate on that. It will be invaluable in understanding your process.

 

[bobdoering]

I am curious - did they ever require a capability study (versus ongoing control)? And if so, how would you perform one? Did they offer any suggestion as to what they would like to see studied? Maybe it's just me, but if a vendor asks me, I would offer up all kinds of ideas. Of course, they would then generally discuss with me the limitations of each of them...but it sparks a good dialog.

 

[bobdoering]

Now we can add REAL control limits using the formulas for the appropriate chart type. Just because you have categorical doesn't mean SPC can't be used as intended - to assess the statistical control or stability of a process. I would run away from 0 and some arbitrary upper goal line as fast as possible - that flawed ineffective approach was debunked decades ago and is exactly what SPC was intended to replace.

This is absolutely true - and I hope they can sell it to their customer. Usually, when the drop the "cpk" bomb, their expectation is that the process is 'capable' - as is in not only not making bad parts - but not near making bad parts. Using attribute charts with upper control limit above zero will preclude that, although you will be able to determine control, such as shifts in the process (as in making more bad parts than the usual amount of bad parts indicating a special cause). You know - it's the old control versus capability thing.

Best of luck any way you slice it...

 

[Stijloor]

I have the impression that the customer's representative has no clue what they're asking for. Yes, we must be very sensitive to Customer's needs and expectations, but if it does not make (satatistical) sense, then it's OK to diplomatically push back. Regretfully, some (automotive) suppliers have no adequate (alternative) approach to offer resulting in a lot of wasted time and non-value added work.

Stijloor.

 

[bobdoering]

I have the impression that the customer's representative has no clue what they're asking for. Yes, we must be very sensitive to Customer's needs and expectations, but if it does not make (statistical) sense, then it's OK to diplomatically push back.

I agree...to whatever degree that makes sense.

Regretfully, some (automotive) suppliers have no adequate (alternative) approach to offer resulting in a lot of wasted time and non-value added work.
.

Not always - it depends on the customer. Most reasonable customers will defer AIAG books. There is enough detail in them to find appropriate exceptions when needed. My experience has shown that it is generally more than the other fields have to work with.