How to implement SPC in a CNC m/c shop processing small batches of 'one timer' jobs?

O.k., first of all, we are a small business that is currently seeking QS9000 certification. We are a machine shop that offers CNC and conventional machining for all types of applications. Getting to the point, we do a lot of jobs that are "one" timers and we may never do them again. This is mostly true in our conventional machining area. On the other hand, the CNC does run production "type" jobs. We may machine 200-300 parts here and there but not on a repeatative basis. I mean the same CNC machine may run 4-5 different parts a week! I am just about frustrated with the whole SPC thing! How in the world do you apply SPC to this type of a business? I can see where the SPC methods would apply to a machine shop that made the same parts over and over, but that's simply not the case here. What can I do to ensure that we will meet the standard? How can I apply statistical methods here? Please help!

------------------
Joe W. Guy,
QS9000 Administrator

[Don Winton]

How in the world do you apply SPC to this type of a business?

In a nutshell, don't! Seriously, any SPC program must be value added. If the program does not add value to your operation, do not use it. If a job is a one shot wonder, do not use it. My personal opinion: Short run SPC is for decoration purposes and, IMHO, serves with minimal benefit. From your description above, we have eliminated the need for SPC on part your processes, let's look at the other part.

We may machine 200-300 parts here and there...

Most generally regard SPC as measuring and plotting 5 of every 20 widgets and plotting the data on a chart. This is a very limited view and is more correctly SQC. Now, if you are determined to use SPC on the machines, use them for the processes. A CNC machine that runs '4-5 different parts a week' has other opportunities for statistical techniques.

You could use Pareto to observe faults in the system (which dimensions are off-quality by category, monthly downtime by reason, etc.). You could use process capability to determine if the machine is capable of producing to specification. You can use process capability to determine if the rate of production is to cost. You could use Xbar-R for spindle speeds, feed rates, etc. And on and on. Or, you could simply use none.

Examine the processes to determine where SPC will add value. Collect data to determine which areas would benefit from the application of statistical techniques. If you can, get a copy of "Handbook of Statistical Methods in Manufacturing" by Richard Barrett Clements published by Prentice Hall. It gives many opportunities for statistical techniques that most would not realize. But remember, any technique selected must add value to the operation.

In regards to QS 9000, I believe the standard requires you identify need in your control plans. We have just done that. There is no need. In the control plan state:

Statistical Techniques Needed: None

Perhaps those more knowledgeable in the QS field can elaborate on this. I am using the 29JUL98 version of ISO 16949 as a reference.

Regards,

Don

------------------
Just the ramblings of an Old Wizard Warrior.

[in_cr_ove]

SPC is statistics, is mathamatics. if it requires a given number of values, so be it. nothing really can be done to circumvent it.
the real issue is why SPC?
preventive focus to reduce variation & improve reliability!
alternatives:
an old tool "pre-control" is my favourate for such situations. it requires 5 values to qualify the process, 2 values to check routinely and the frequency of checks varies based on actual data. it is literaly an "on-line" tool.
try it and you will get addicted.

[sonflowerinwales]

We joined an aerospace company training scheme for SPC, and the same problem was discussed, at length, and "enthusiastically"! The outcome was you set up a control chart for a size/tolerance/material combination. From this, you could determine upper/lower limits for that combination. As a result, if you had a 5/8"dia +/-0.0005" in bronze, you used the chart for that combination, regardless of the component drawing number or batch size. If the next item was 3/8" +/-0.001" steel, you would use a different chart. It would take a long time to collect all the data, but in theory you could apply SPC to any item/quantity/material.
Paul

[Evan J Miller]

Joe - a couple of comments.

One obvious question is can you chart a process variable that is consistent across all products? On a welder or molding machine, for example, you might track power consumption or pressure. Do you know the relationship between these factors and the outcome?

The other thing is that I've seen Short Run SPC techniques work really well in a machining environment - as long as the feature you're measuring is similar across the chart. You apply some clever coding to strip out the expected variation and you get a very meaningful chart. Full disclosure: my company produces software to do this kind of thing, and I'd be happy to talk with you off-line to see if it would support your application or not.

[bobdoering]

It is not that hard to use SPC in a precision machining job shop. It has a bad rap from folks trying to use X-bar -R charts...incorrectly...and being frustrated to death. The correct SPC chart is similar, but more statistically correct and powerful than pre-control. If you are properly controlling the process, you will generate a non-normal distribution - the uniform or rectangular distribution. Transformation is not recommended! Preparing control limits and calculating capability can be found at: Statistical process control for precision machining

[Wes Bucey]

Originally Posted by Evan J Miller

Joe - a couple of comments.

One obvious question is can you chart a process variable that is consistent across all products? On a welder or molding machine, for example, you might track power consumption or pressure. Do you know the relationship between these factors and the outcome?

The other thing is that I've seen Short Run SPC techniques work really well in a machining environment - as long as the feature you're measuring is similar across the chart. You apply some clever coding to strip out the expected variation and you get a very meaningful chart. Full disclosure: my company produces software to do this kind of thing, and I'd be happy to talk with you off-line to see if it would support your application or not.

Originally Posted by bobdoering

It is not that hard to use SPC in a precision machining job shop. It has a bad rap from folks trying to use X-bar -R charts...incorrectly...and being frustrated to death. The correct SPC chart is similar, but more statistically correct and powerful than pre-control. If you are properly controlling the process, you will generate a non-normal distribution - the uniform or rectangular distribution. Transformation is not recommended! Preparing control limits and calculating capability can be found at: Statistical process control for precision machining

Say, Bob, I notice you've been dredging up old threads to post a link to your article. Is there something more to the story we should be aware of?

[bobdoering]

Originally Posted by Wes Bucey

Say, Bob, I notice you've been dredging up old threads to post a link to your article. Is there something more to the story we should be aware of?

Ah, you are an observant student, Wes! I have found people that do searches may end up in old threads - a lot! If they leave those threads thinking that their questions are answered - and it does not include this newer information, they may walk away either misinformed or uninformed. I am trying to reduce their misery by preventing that from occurring.

No more, no less.