100% Measurement Shift to Sample Inspection? PCB Substrates

I have a supplier providing PCB substrates. The PCBs are manufactured in panels. There are 12 functional strips of substrates on the panel. Alignment of the panel is measured in all 4 corners of every panel. No measurement is made on the strip. The only measurement is on the panel.
The recorded measurement is the difference of top layer pattern to bottom layer pattern referenced to a drilled thru hole. The spec is 0 - .004" acceptable.
We are plotting the alignment CpK or CpU (one sided spec) and finding a capability of CpK= 0.86 at best. The supplier tracks this as attribute and I track capability as a variable.

The supplier says all panels pass spec and therefore the process is capable. I feel the process capability is below CpK=1.0 therefore not capable without 100% inspection. I can not afford the value add cost of using poorly aligned substrates.

When can a supplier be allowed shift to sample inspection from 100%

To quote a colorful post in another recent thread:

Jim Wynne said:

If your customer tenders a contract that includes the requirement to not only measure each part, but to have it blessed by a priest in a Pittsburgh Steelers uniform, and your company agrees to it, then that's what you have to do. It's a matter of contract review. If the customer is making unreasonable demands, don't agree to them, and if the customer tries to invoke new requirements that weren't figured into the original quote, then your company has the right to reopen the negotiations.

Click to expand...

The short answer is that whatever agreement you made with your supplier - that's what determines the process for accepting materials.


Now my humble opinion about your specific situation....
Ideally you want products with a small variation, but I'm assuming that the contractual requirements relate to the specs, not Cpk.

Cpk only gives an estimate of the capabilty to meet a requirement. Values like Cpk = 1.33 or Cpk = 1.0 are just guidelines for good capability. A value around 0.8 is relatively poor for Cpk -- about 2.5 sigma from the closer spec limit. For a normal distribution, that would be ~ 1% nonconforming. But it is quite possible with other distributions to have no nonconforming parts with such a Cpk, or to have way more than 1% nonconforming. Sure, it would give you more confidence to have a better Cpk, but Cpk is just giving you one particular measure of how good the parts are.

If the contract says 0 - .004", and all parts are found to be within that spec, that tha would seem to trump any Cpk value. It would be hard to reject the parts simply because your aren't comfortable with the Cpk values. On the other hand, if you can show that there are actual defects, then you would have grounds to go back to the supplier.


Tim F

In addition to Tim's good advice, a few observations:

Magus said:

The supplier tracks this as attribute and I track capability as a variable.

Click to expand...

Seems like you and the supplier might be speaking different languages right from the git. When measurement is important, and especially when close tolerances are involved, it's important for customer and supplier to be in sync with regard to measurement and data collection requirements.

Magus said:

The supplier says all panels pass spec and therefore the process is capable.

Click to expand...

There you go--you and the supplier are applying different meanings to "capable." He's saying, in essence, "Don't holler 'til you're hurt," and you're trying to make predictions regarding future performance in an attempt to prevent or minimize defects.

Magus said:

I can not afford the value add cost of using poorly aligned substrates.

Click to expand...

Can you afford the added cost of 100% inspection, whatever that might be? As Tim suggested, unless there is specific language in the purchase agreement with regard to Cpk and inspection requirements, the supplier isn't obligated to do any inspection--he only has to provide parts that meet the specifications.


At this point you need to sit down with the supplier and 'splain the situation. You need to explain the potential costs involved for you in the event that nonconforming material is received, and the need for him to collect variables data (if he does this, then maybe you won't have to, so cost might be a wash). If he suggests that a price adjustment might be necessary, then you can objectively determine value--will the extra cost of parts be offset by the likelihood of nonconformance costs down the line? Is one nonconforming panel in 1000 enough to cause you to pay some value x more for the parts and still come out ahead? You can't make this decision without good data.

and you should check the data for normality befoer that discussion. It is quite possible (since the Normal distribution is a model, not a physical truth or certainty) that the distribution is not Normal. this woudl provide a low Cpk and no defects. If the distribution is fairly narrow but has "thick shoulders" you may be getting very minimal variation but the thick shoulders provide a large standard deviation even tho the max to min range is small. This will give you the small cpk. of course the opposite is true. the product may span the full range of the tolerance but not have any actual values out of spec.

This is why I distrust Cpk indices and always go to the individual data points - plotted in time sequence, of couse. then I don't even need to calculate a single value to quanitfy the variation I can see it with my own eyes.