Capability of Inherently Non-Normal Process - Plating Process Thickness Distribution

[jefnik3201028]

Hello guys,
I'm a newbie here so was actually reluctant to post any questions but I have this lingering issue on my head for quite a while now so I hope you guys can give me some wonderful advise.

Im working with a plating process where in the thickness distribution is dictated by the current distribution. Based on previous studies, since current distribution is directly related to plated area -- inverse relation to plating thickness so the problem now is a product will not have a uniform plating thickness on different plating locations, some will have thicker plating and some points have thinner plating. When a random sampling is done, data will reveal that there is 2 sets of distribution present within the process, an inherent behavior due to technology limitations and probably poor process design.
If I have this kind of process, what is the best way to show process capability. Can I consider bias sampling in this case ? I'm thinking of identifying the location points that belong to the thicker and thinner plating. Any advise ?

Thanks,
Jefnik

 

[jefnik3201028]

No takers ?

 

[Al Rosen]

Re: Capability of Inherently Non-Normal Process - Plating Process Thickness Distribut

jefnik3201028,

Welcome to the cove! You may have not gotten anyone to respond since it is the weekend here in the states and usually slow in the cove. Check back later in the week.

 

[jefnik3201028]

Hello Al,
Thanks for welcoming me!

 

[Wesley Richardson]

Hi Jefnik,

Plating thickness is a function of several factors, including: time and current density. The current density is affected by the current, geometry of the plated part, distance to the electrode, shape of the electrode, and concentration of electrolyte.

Surface preparation of the plated part is also a factor. Oils or other contaminants on the surface can mean that virtually no plating occurs in those areas.

When parts are manually placed in the plating bath, even on fixtures, the technician can have a considerable influence on the plating thickness.

For the above reasons, plating is considered a special process. The best method is to establish the process parameters, set-up configuration, and relationship between the electrode shape and the plated part, and the distances. Through trial an error of varying current and time, measure the plating thickness in the various types of areas: flat surface, interior locations, convex corners, concave corners, etc. For each of these types of locations, expect a different plating thickness.

For a process capability, you could repeat this a number of times, with the same parameters, and then determine mean and standard deviation for the sample.

Wes R.

 

[jefnik3201028]

Hi Wes,
Thanks for your reply. The situation is... the current process shows plating thickness difference within unit , unit-to-unit (in one part there are several units) and part-to-part. If a random sampling is done and once a frequency histogram is plotted, the graph shows non-normal distribution. In most cases, the distribution manifest a bi-modal distribution causing the calculated mean (Xbar) to represent a "dead zone" (an imaginary process center) and sometimes multi peaks. Subsequently, the standard deviation is over estimated. Bottomline is process capability can not be established but customers insists in using Ppk as measurement.
Is non parametric calculation applicable for this situation ? The only samples given before was for skewed or flat distribution. I am now contemplating of using a bias sampling, meaning to group thickness based on actual process output, one capability each for thick and thin plated locations, will this be acceptable to the customer? The idea is to identify the extreme points for each group, same point measurements with standard deviation taken from unit to unit and part-to-part variation, between subgroup fluctuation is represented by change of averages thru time.
Does this makes sense ? I just think that that is the best way for us to show how well we satisfy the plating thickness tolerance window.

Regards,
Jefnik:

 

[Wesley Richardson]

Hi Wes,
Thanks for your reply. The situation is... the current process shows plating thickness difference within unit , unit-to-unit (in one part there are several units) and part-to-part. If a random sampling is done and once a frequency histogram is plotted, the graph shows non-normal distribution. In most cases, the distribution manifest a bi-modal distribution causing the calculated mean (Xbar) to represent a "dead zone" (an imaginary process center) and sometimes multi peaks. Subsequently, the standard deviation is over estimated. Bottomline is process capability can not be established but customers insists in using Ppk as measurement.
Is non parametric calculation applicable for this situation ? The only samples given before was for skewed or flat distribution. I am now contemplating of using a bias sampling, meaning to group thickness based on actual process output, one capability each for thick and thin plated locations, will this be acceptable to the customer? The idea is to identify the extreme points for each group, same point measurements with standard deviation taken from unit to unit and part-to-part variation, between subgroup fluctuation is represented by change of averages thru time.
Does this makes sense ? I just think that that is the best way for us to show how well we satisfy the plating thickness tolerance window.

Regards,
Jefnik:

Hi Jefnik,

I cannot tell you if this will be acceptable to your customer, only your customer can tell you that. It does make sense to me to separate the data between the thick and thin types.

An assumption that is being made when a Ppk, Cpk, Pp, or Cp calculation is performed, is that the distribution is normal. If you take averages of sample averages, this is probably a good assumption. For non-normal data, a Ppk calculation is not a good predictor of the fraction nonconforming. For example, with a skewed or bimodal distribution, the fraction nonconforming could be significantly higher that predicted, at a given Ppk value.

Wes R.

 

[triner]

I think you are on the right track with trying to separate the data by location. If you were to define a few critical points and measured the thickness in those defined areas only, is the measurement data for each critical point normally distributed?

If so, I suggest sharing this data with your customer along with the argument that if these critical points are in control, the entire surface is under control. You and the customer can work together to define the critical points.

As far as obtaining random samples, randomly select many parts and measure the same points on each part.

 

[cyberjyothi]

Re: Capability of Inherently Non-Normal Process - Plating Process Thickness Distribut

Dear Sirs,

we are also having similar problem in our process.

We are basically automotive battery manufacturers. For making battery we use grids.
For the process of grid manufacturing, casters with gravity casting are used.

For gravity cast grids, mould cleaning (the mold is brushed with a wire until the old cork is removed from the surface of both halves of the mold) and corking (Cork solution is wood fibre based & used to coat the mold surface) is periodically done for maintaining Grid weights.

Mold cleaning & corking adjustment is an integral part of the process & these adjustments are inherent to the Grid Casting process.

In between 2 mold cleanings, we apply cork spraying about 15 minutes once.

We were measuring grid weights at a frequency of once in 1 hour.

An increasing trend in grid weight is observed up to about 4 hrs. At this instant, mould cleaning and corking brings the weight of Grid to lower warning limit .The weight of the grid is controlled by the size of the cast grid wires. The size of the cast grid weight is controlled by the cork sprayed into the wires on the mold.

At present we are using Run charts (i.e., Pre-control chart with warning limits) for monitoring of the grid weight.

Hence we conclude that Grid casting process is inherently unstable.

From the above, don’t we say that the Grid casting process is unstable?

1. If this is unstable, don’t we need to calculate Capability indices?
2. If we observe the grid wt data between mold cleanings, the data follows non normal pattern. Are we using transformations for this?

And suggest us which chart is suitable for our process and applicability of the SPC 2nd edition with respect to said process.

Thanks in advance

Jyothiswar.M

 

[ccostello4]

Re: Capability of Inherently Non-Normal Process - Plating Process Thickness Distribut

Hello,
Being in the plating industry for a good 25 years this is a very common problem.
Achieving process control in the process depends on the thickness window in which you need to accomplish. Some part configurations even using special anodes or fixturing will still have a variance of thickness all over the part. If the window lets say in +/- .000020 on a part that has a complex configuration it is very hard to control that process to consistantly achieve thickness within tolerance. Back in the early 90's when Motorola was going gung ho with its 6 sigma program, plating was a big issue to achieve that 6 sigma process with the tight tolerances they were demanding. If the part can still function and still have the properties that are accpetable then the tolerance or window should be as open as possible to achieve good process control. If not its going to be a long hard journey. Plating in either a barrel or a racking the current density varies so much is the main problem for achieving process control with tight tolerances. If you are plating a part one at a time then its doable but not practical in the production world.
Open the window and you'll breath easier.
CC