How to sample in a real time release process?

[IHaveNoSpaceBar]

Regards all,

I am having a hard time with creating a quality inspection plan for a medical device that involves several high volume continuous process steps(cells) working in series. Essentially cell 1 feeds product into cell 2, 3, ... etc, all the way to the final step of packaging the finished good.

Customer does not have any requirements for lot sizes, but they have a fairly large amount of required CQAs. Not all CQAs are inspected automatically on the line (by machine) - about 10% of CQAs (dimension and attribute) require manual human inspections, some of which have very stringent AQL requirements (very high sample sizes is also a concern).

My main challenge is that with a continuous production cycle I don't really have any set production batch sizes - product is simply going down the conveyor to the next production step on continuous 24/7 basis.

I want to avoid doing all testing at the final step (packaging) because if we find a defect from the early steps (such as cell 1) this means that I will have to reject and possibly scrap the entire lot.

The customer expects us to release/ship product on daily basis (multiple shipments per day). I am extremely concerned with the large amount of paper work involved (medical) and can already see this as a large burden on the shop floor employees for having to create and close shop orders & device history records on daily basis. I see this as a paradox - customer wants frequent shipments but internally, we want to minimize DHR open/closure activities. There is over 40 machines that will each have a separate DHR!!!

Does anyone have any experience or recommendations regarding how to perform real time release in a high volume? I understand that this fairly new for the FDA and not much guidance material is available on this topic from what I have searched. I have seen continuous production inspection plans but they don't seem to be very popular in the medical industry - not sure if due to risk or what gives?

I have read that one approach is SPC for all critical parameters - but is that really sufficient from a regulatory perspective? I find myself having a hard time convincing the customer in performing SPC in lieu of product testing.

Sorry for the long winded topic - I am very thankful for any responses or ideas on how to proceed. Looking forward to hearing from anyone.

 

[Steve Prevette]

I would suggest that no one does SPC "instead of sampling". You've still got to have some measurement system to collect the data.

If you can decide on a rate of sampling you want (say each item should have a 5% chance of being inspected) then you could establish some sort of random number generator that as a completed item passes the inspection station, the equivalent of dice are rolled, and if the result is in a certain range, you inspect it.

 

[Statistical Steven]

Regards all,

I am having a hard time with creating a quality inspection plan for a medical device that involves several high volume continuous process steps(cells) working in series. Essentially cell 1 feeds product into cell 2, 3, ... etc, all the way to the final step of packaging the finished good.

Customer does not have any requirements for lot sizes, but they have a fairly large amount of required CQAs. Not all CQAs are inspected automatically on the line (by machine) - about 10% of CQAs (dimension and attribute) require manual human inspections, some of which have very stringent AQL requirements (very high sample sizes is also a concern).

My main challenge is that with a continuous production cycle I don't really have any set production batch sizes - product is simply going down the conveyor to the next production step on continuous 24/7 basis.

I want to avoid doing all testing at the final step (packaging) because if we find a defect from the early steps (such as cell 1) this means that I will have to reject and possibly scrap the entire lot.

The customer expects us to release/ship product on daily basis (multiple shipments per day). I am extremely concerned with the large amount of paper work involved (medical) and can already see this as a large burden on the shop floor employees for having to create and close shop orders & device history records on daily basis. I see this as a paradox - customer wants frequent shipments but internally, we want to minimize DHR open/closure activities. There is over 40 machines that will each have a separate DHR!!!

Does anyone have any experience or recommendations regarding how to perform real time release in a high volume? I understand that this fairly new for the FDA and not much guidance material is available on this topic from what I have searched. I have seen continuous production inspection plans but they don't seem to be very popular in the medical industry - not sure if due to risk or what gives?

I have read that one approach is SPC for all critical parameters - but is that really sufficient from a regulatory perspective? I find myself having a hard time convincing the customer in performing SPC in lieu of product testing.

Sorry for the long winded topic - I am very thankful for any responses or ideas on how to proceed. Looking forward to hearing from anyone.

SPC in a FDA regulated industry solves one problem while creating many additional issues. I steer clear of SPC in FDA regulated processes because of an in spec out of control or out of control in spec process. The prevailing guidance from the FDA is that the manufacturer defines a batch for continuous manufacturing processes. Can be based on time (shift), number of units, etc. What complicates your DHR is the number of parent/child relationships that can be present in the process. I would suggest you flow out the process and determine natural places where you can sample. Increase sample size and decrease sample frequency at those places to get a good balance between AQL and LQ. I would define a batch as the product manufactured between sampling points. Of course, these are just suggestions based on the limited information you provided.

 

[Mike S.]

I would suggest that no one does SPC "instead of sampling".

Maybe not in the medical device world.

"The purpose of (SAE) ARP9013/4 is to ensure conformance for each product characteristic beingverified under these methods. ARP9013/4 provides the requirements for continuous sampling, skip-lot sampling, and for statistical methods for product acceptance that do not derive from industry standard tables."

 

[Tony_C]

I am new to this forum and I am not sure this discussion is still active, but I offer the following.

- Sampling plans which require a choice of AQL actually have the underlying assumption that there is a continuous process in operation 24/7, and that from time to time a lot is withdrawn from the process and submitted for inspection. From these lots, a sample is drawn, and based upon the "Acceptable Quality Level" and the sample size, an acceptable number of nonconformances is given. (perhaps c=0 for smaller sample sizes) - and a level is given for rejection (perhaps 1 for small sample sizes).

- In addition to the "AQL", there is inherently a "LTPD" (Lot tolerance percent defective). The sampling plan will accept lots submitted for inspection which contain defect rates at the AQL with high probability - say 95%. The same sampling plan will only accept the lot with say 10% probability when the defect rate is at the LTPD. In this way, the sampling plan is actually a control chart (without the chart) which uses the binomial distribution as the underlying mathematics (p-Chart). i.e. the sampling plan accepts the process operating at the AQL defect rate but will reject when the process deteriorates to the LTPD rate.

- Most organizations would like to put blinders on because they can not convince themselves that there is any defect level which is acceptable. Mathematically, the process producing zero defects requires 100% inspection - but given the many types of human and inspection errors, Joseph Juran and others say that 100% inspection is only 80% effective.

- A more practical approach is to adjust the sample size so that "c" is always equal to zero, and the lot will be rejected upon 1 nonconformance. Ignorance can be bliss, and these plans avoid the unpleasant situation of dealing with nonconformances in inspection which come from "normally" operating processes. These c=0 sampling plans are usually based upon the hypergeometric distribution and so not assume that there is a continuous process running in the background 24/7.

Hope that helps