Predicting Attribute Rejects at end Customers

[#1]

We are performing manual inspection and removing rejects found (e.g. damaged wire) at each manufacturing process steps. Since manual inspection is not 100% effective, what statistical methods that can be used for ATTRIBUTE data to predict the number of rejects that a customer could get?

[#2 · Accepted solution]

This question is not so easily answered since you state that '...manual inspection is not 100% effective.' you are acknowledging that some rejects pass on to customers. And you want to predict that number. The only way I know how to do this is to have some sort of sampling plan AT customers...either you do it...or they do it and report to you. Then a simple control chart is one way to 'predict' the number of rejects that customers 'could see'. Neither of these methods will give you truth...but they can give you some information and data on which to evaluate the volume of rejects and stability of that volume over time.

[#3]

First, welcome to the community.

In addition to Pete's comments, here are my thoughts:

1. Why are you getting rejects at each step of the manufacturing process?  Wouldn't your efforts be better served to understand what is causing the rejects and work to manage those inputs to prevent the rejects.

2. I believe you mean visual inspection rather than manual (though your point may be it is not automated visual inspection). It would be beneficial (both efficiency and effectiveness) to be able to quantify the defects with a continuous measure rather than pass-fail criteria. You might need to get creative (e.g., ask yourself how to measure the wire damage, locational metrics may also be useful, sometimes defect densities can shed light).  Defects or defectives as a response pose multiple issues.  Not only does visual inspection have the challenges of effectiveness, but it can also be easily biased and aggregates possible failure mechanisms making causal relationships difficult to ascertain.

3. You can use P/NP-charts (Defectives) or C/U-charts (Defects) to assess consistency, but again you really don't want to consistently make defects.

[#4]

Thank for you for response.

Our reject rate is below the set target and we know the causes of it as well as the probability of out of specification for variable data (e.g. product height).  We have a new product that will go through the same manufacturing processes but the customer requires an RQL of 1PPM.  Given that the order volume is less than 1 million per year and our visual inspection is manual there is a high probability (escape) in my view that the customer could get more than 1PPM per lot that we will ship thus I am inquiring about predicting attribute reject at end customer.

Have you encountered the same situation? How the situation should be tackled? 

[#5]

As @P_Bartell mentioned you could use a sampling plan approach.

I wrote a JMP Acceptance Sampling Plan Add-In to assist with such questions.

See if this helps with your situation.

cheers,

Stan

[#6]

Thank you for sending the acceptance sampling plan add-in, this is very helpful.  To be honest, years ago I did try to search on JMP help the accepatance sampling plan but unfortunately I did not find it thus I missed the statistical software that I used before.

I play a bit on the add-in that you sent, I'll give my feedback/inquires when I get a chace to explore it all.