Single Cavity Blow Molded Bottle - process capability

I may be able to help you. What variable are you trying to control?

the part's height, thickness? diameter? What type of product are you blow-molding?

I think the first step is to take some samples and measure the variable that you need to control. Look at the distribution ... is it a normal distribution or the target is off center.

In the first case you can use Cpk in the second case you can use Cpkm which allows for an off center process mean.

ppk is a measure mainly of dispersion... I would use Cpk or Cpkm depending on how the distribution curve looks.

Before doing cpk etc, it is important to know the tolerance.

You did not provide a tolerance.

My personal practice is to take thirty consecutive pieces and measure the dimension(s) of concern, then use excel to calculate average (Important) and standard deviation (CRITICAL!!!)

I then divide the total tolerance by the standard deviation and if it is six or less i'm screwed, as I will be merely compliant when i have the average perfectly centered.

So I work hard to create very small standard deviations. THEY are the true determinant of prcess capability, parking the average is a machine adjustment. Its pretty flippin tough to adjust your way to a smaller std. deviation.

After this, I make a histogram to look for normalcy, then I calculate cpk, and ppk. I do this to please the fastidious.

by the way, calculating cpk without doing a x-bar-r chart is a great way to commit statistical malpractice. How do you know you are in statistical control and allowed to use the data for cpk?

But since most people dont really do X-bar r charts, I say the heck with Cpk, JUst do the statistics for average and standard deviation and histogram, and if your tolerance doesnt divide by the standard deviation morethan six times, you need to work on your process. You won't be capable.

This has worked for me, and though I've had a few of the ultraorthodox ASQ people challenge me, when they see my calculations and data they move on to other things...

first things first. What is being checked, what is the tolerance?, and by the way if you check all of these variables on each of the thirty things in sequence and put them in rows and columns, in excel, you can then correlate them and then use the variable(s) with the biggest standard deviations as proxy indicators for all of the others.

Practitioner practicum, not orthodoxy.

Not to offend, but there really are many ways to skin a cat.

sorry del.

milo

Milo,

Thank you for your insight, very helpful! Now, to take this a step further, let's say that this single cavity mold is being used as a precursor for the larger multi-cavity production mold(s). Does this change how I need to look at the single cavity mold data and analysis?

My need is to prove that the package produced from the single cavity mold is representative of what I will get out of a multi-cavity production mold. Will only using the standard deviation give me that confidence? Can cpk be used as 'proof' or can it be more of a predictive tool of what I will see in a production scenario (identify 'watch outs')? I can see a nice balance using both std. dev. and the cpk.... but would like your thoughts.

In regards to what is being checked... your typical dimensions for bottle and neck finish. Critical (cpk >1.33) and non-critical (cpk>1.00) items identified. Tolerances generally follow SPI, unless it is a very critical dimension that needs a tighter tolerance.

Two things- Dimensions data is variables- std deviation is valid on variables data. Finish is attributes data, This will require different approach.

This single cavity mold data will be a predictor of the future toolings capability, but it is not a guarantee. If the single cavity mold is made by same vendor using same machines etc etc etc itwill be better indicator than if the multi is made by someone else using different tooling materials and gaging . Also assumptions about feeding each cavity will need to be proven, its one thing to over provide to a single cavity, vs providing to many. cooling issue too.

Each cavity of the new multiple cavity mold will be a separate process. this means that you have a mixture problem, and cannot use standard assumptions about normality of distribution etc.

In the screw machine business, our machines have six spindles, and we treat each of these as if it were a separate "mini lathe" Once we identify the spindle with the worst capability, we use it as the "screaming indicator" for our monitoring.

In my opinion you will need to qualify each cavity as capable, and again, my std. deviation shortcut will be a great first pass thru the data to identify issues. As there will be subtle differences in how each cavity is machined, and fed, and cools, you really need to prove each out. And this proof will be short term, not reflective of process over time and much thermal cycle on a shift /day basis.

Pointy headed theoretical geeky side of me wants to say that 6 is no longer the proper divisor, knowing that you will have multiple cavities each of which will have slightly different centered "averages". Now the spread of the averages is of interest, as their distribution could eat up much of your tolerance band... and bias to one side or the other...

I'm still travelling, and away from resources. will get back to you friday after i do some thinking.

My initial reaction is that the standard deviations of each cavity are likely to be about the same assuming similar workmanship and process of providing resin is consistent; but the differences between the averages of each cavity will be what controls your ability to hit the tolearance with safety margin.

milo