Sampling - Measuring and Inspecting a part with both Variable & Attribute Data

[alongtain]

I've got a real-world sampling problem I need some help with.

I'm in the process of checking, both, variable & attribute dimensions on a part I'm manufacturing for my customer. There are about 12 variable dimensions and 4 attribute (GO/NO GO) dimensions. I'm running a 200 piece lot and am checking at a sampling rate of every 4th part (per Z1.4, Level II, AQL=1). This is taking quite a while.

My customer said that if I can show a Cpk of 1.33 or better on the variable data, I can reduce my sample size for those attributes to 13 parts (per Z1.4, Level I, AQL=1). I have performed my Cpk checks and my process is in a tight state of control, so I'm going to reduce my inspection frequency on those dimensions.

The problem I'm having is coming to an agreement with the customer in regard to reducing my attribute checks to z1.4, Level I, AQL=1. They initially stated that I need to check at least 59 pieces out of a 500 piece lot at 95% confidence level and then switched and said they wanted me to check the first 5 lots according to Z1.4, Level II, AQL=1 before reducing my inspection frequency.

I know that my process is in control after checking the first lot and checking 5 lots at 50 pieces per lot vs. 13 is going to require a lot of extra hours of inspection. Is there another standard rationale I can use/reference to convince my customer to let me reduce sampling frequency on the attribute dimensions? Also, my customer said that the 5 lot attribute inspection rule comes from the ANSI/ASQC Z1.4 standard. I don't have a copy of this standard and was wondering if this is correct?

Any help in regard to onducting sampling on variable and attribute data and determining a reasonable rationale for reducing sampling frequency on both data types would be greatly appreciated. I typically run lots from 200 to 2000 pieces and want to put an easy-to-follow system in place for determining that initial lots are in control and reducing sampling on subsequent lots, once control has been proven.

I know this is long-winded, but, I couldn't explain it any quicker.

Thanks,

 

[bobdoering]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

I am not sure why you are using acceptance sampling for a manufacturing process - it is really for incoming inspection (unless you are doing a final inspection as in the medical industry, which is designed to emulate incoming inspection). Sounds to me like you need evidence of process control, such as a control chart. With that data you can begin to develop sound rational for sampling frequency based on the process. What kind of process is this? Is it really a batch process?

 

[alongtain]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

The process is swiss machining. We're making bone screws for a medical device customer and our inspection data is the final inspection. They perform one initial 1st article inspection and then rely upon our data thereafter, so my final results serve as their incoming inspection. Also, we typically run jobs (batches) of 200 to 400 pieces. What I want to be able to do is reduce my inspection frequency for future jobs, based on the solid results I've gotten thus far and am wondering what the rationale is for reducing inspection frequency on attribute data?

 

[Bev D]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

yes Bob - believe it or not, many industries (medical and aerospace among them) use AQL acceptance sampling for FINAL acceptance inspection prior to shipping. It's common for 100% final inspection that can be reduced to a sample plan if oen can demonstrate process control and capability...

 

[bobdoering]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

yes Bob - believe it or not, many industries (medical and aerospace among them) use AQL acceptance sampling for FINAL acceptance inspection prior to shipping. It's common for 100% final inspection that can be reduced to a sample plan if one can demonstrate process control and capability...

Yes, having been in both industries I am aware of that. But, they were also prone to use it as a basis of process inspection, too. I had to sort that notion out for them.

I have also battled the urge to reduce inspection to save time and resources. Process control and capability work out nice in variable data...well, nicer. Attribute is always a dicey thing to sort out - especially things like surface defects and other visuals. Raw frequency of occurrence is interesting, but frequency of the cause is the real problem. How often does someone forget to change a tool until the ridges show up in the self-tapping flutes causing rejection, for example? It is typically not a simple roll of the dice probability function.

 

[alongtain]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

Everything you mention makes sense and when dealing with quick visual checks (which are easy to execute quickly), but, I'm checking part profiles on overlays, which take time to set up and execute. Really, I'm checking dimensions that could be saved as variable data if I had a high-end vision system, but, we're using GO/NO GO profiles on a comparator right now and I haven't had one bad part check out of the 3 lots I've checked thus far.

 

[bobdoering]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

Everything you mention makes sense and when dealing with quick visual checks (which are easy to execute quickly), but, I'm checking part profiles on overlays, which take time to set up and execute. Really, I'm checking dimensions that could be saved as variable data if I had a high-end vision system, but, we're using GO/NO GO profiles on a comparator right now and I haven't had one bad part check out of the 3 lots I've checked thus far.

Ah! Then maybe a capability study of those characteristics done the hard way, maybe even measuring with the comparator or sent out to a vision system vendor, just to show the true capability may provide the evidence you need to reduce the attribute verification frequency. It will take time - might even have to send it out. But, it the evidence is compelling the savings will be significant.

There is one other option. If you are controlling a characteristic on the tool path to a tighter tolerance, with variable data (and maybe SPC?), that should be evidence enough that the other dimensions on that tool path will be capable, too. I have used that logic with some customers.

 

[alongtain]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

I like that logic! Especially since we've been holding all the variable dims on that tool path much tighter. I'll try to get the customer to go with it.

Thanks,

 

[Bev D]

Re: Sampling - Measuring and Inspecting a part with both Variable & Attribute Date

Well, not to burst the enthusiasm for process capability which I heartily promote, but the dilmma that Bob alluded to earlier about mistakes (forgetting to change a tool, etc.) still exists. And a feature that can measured with continuous data is still susceptable to mistakes that will not be predicted by continuous data. So continuous data capability studies are not a comprehensive model of your defect rate.

Even mistakes can be modeled and predicted by probability theory - not exactly when it will happen, but how often it will happen in a given period of opportunity. Poisson was correct. However, for inspection purposes rare failures - and most mistakes failure modes are rare events - drive the sample to 100% if your desired maximum acceptable defect escape rate is also very small...

How to avoid large sample size? mistake proof. And then perform categorical data capability studies...

 

[alongtain]

Again, this all applies to long production runs at large organizations with big budgets and big quality departments full of people getting paid to study and prove theories related to SPC, but I manage a small operation that sets up and runs several jobs per week at relatively small production quantities. I'd be out of business if I spent this much time evaluating how tightly my process is being controlled. I'm just looking for easy-to-implement methods that will enable me to make my customer happy and reduce the inspection burden on my end. So, it looks like sticking to Z1.4, Level II, AQL=1.0 is probably my best option for now.

One final question: Does the Z1.4 standard state anything about running at least 5 lots when checking attribute data?