SPC for a Changing Process such as Tool Adjustments

[Coleman Donnelly]

Hello,

I am trying to implement SPC through a preliminary stage of Statistical process monitoring as a sort of beta trial to gain momentum. The problem I am running into is that a lot of the product we make is difficult to track without adjustments due to the fact that tools tend to break and adjustments are constantly necessary.

For the adjustments I want to incorporate a standard adjustment as a part of the process... i.e. change the offset by 0.002" every third part...

After about 8 pcs (give or take) the tool will break and need to be replaced. At this point a new part is made but intentionally off. The operator will measure the piece in the setup and determine the appropriate offset to use and then go back to making product (until the tool breaks again...)


It seems to me that there are 2 processes occurring here. One for regular production, and one for the first piece to bring in a new tool.

However my X-Bar will be constantly moving... It will trend up until an adjustment is made and then continue to trend up from the new point of adjustment, but does this mean that my process is out of control? Are there tools to allow for data transformation so that we can still control our process?

 

[Steve Prevette]

Re: SPC for a changing process

You should take a look at Bob Doering's posts here on the Cove. He has some adjustments to SPC for tool wear.

Seems to me if the tool is breaking as often as every 8 pieces you may want to plot the rate of tool failure.

Also, are you sure the new tool should be set "intentionally off" or at the best estimate based upon past new tools (may imply SPC analysis of the first results only).

My parents live in Macedonia, brother in Twinsburg, so I get up to Cleveland area now and then . . .

 

[bobdoering]

Re: SPC for a changing process

A little more information may be handy to help out. Is this a long path milling application, a tough material application or some other type of application?

In any event, for tool wear I recommend the X-hi/lo R chart, not X bar-R chart.

 

[Coleman Donnelly]

Re: SPC for a changing process

Bob,

Thankyou for the reply - I think I am starting to understand the things going on here. To answer your question the material is cobalt-chrome and the parts can be small and intricate, making the small tools break at an accelerated rate.

I understand the importance of roundness relative to a diameter - but what about a true position? or a profile of a non normal surface? Should I just evaluate form of a feature with my X-hi/lo R chart rubber stamp or am I missing something here? I know that SPC and GD&T don't get along in the first place but now your changing the rules on me and I want to make sure I dont miss something in the exchange...

 

[bobdoering]

Re: SPC for a changing process

Thank you for the reply - I think I am starting to understand the things going on here. To answer your question the material is cobalt-chrome and the parts can be small and intricate, making the small tools break at an accelerated rate.

I understand the importance of roundness relative to a diameter - but what about a true position? or a profile of a non normal surface? Should I just evaluate form of a feature with my X-hi/lo R chart rubber stamp or am I missing something here? I know that SPC and GD&T don't get along in the first place but now your changing the rules on me and I want to make sure I dont miss something in the exchange...

Yes, you up the ante with true position and profile. If you have an AIAG SPC second edition book, you will see they tend towards multivariate distribution analysis for GD&T (pages 116 and 144.) That analysis goes well beyond tool wear. In either event - tool wear or GD&T - X bar-R is not going to cut it.

Generally, if you can control the process feature (diameter or length) the position become less of an issue. If you can get to that point, life is a lot easier. Unfortunately, another problem with short-lived tools is by the time the process becomes stable after a tool change (start up/warm up) the tool breaks. During start up and warm up, the process is unstable, too. So, trying to generate a stable process in control is badly restricted. You might not even have enough data to predict tool breakage - something the X hi/lo -R chart can do better than X bar-R.

 

[Coleman Donnelly]

Unfortunately no - I don't have access to the AIAG manual. We are not an automotive company... Is there anywhere else I can look for guidance?

 

[Coleman Donnelly]

How do I determine sample size and frequency for this type of charting?

Also I am wondering... We are also making some parts out of plastic which (I am told) does not really get effected by tool wear - however they do have problems with chip wrap which can break tools but that is more of a special cause.... Would I then just dial in the process and let it run with a standard X-Bar R chart?

 

[bobdoering]

How do I determine sample size and frequency for this type of charting?

Also I am wondering... We are also making some parts out of plastic which (I am told) does not really get effected by tool wear - however they do have problems with chip wrap which can break tools but that is more of a special cause.... Would I then just dial in the process and let it run with a standard X-Bar R chart?

I agree, tool breakage is a special cause.

If you are machining an unfilled plastic, you are probably correct - tool wear will be negligible in the short run. But, if that is true you will see no trend - especially in the long run.


As far a frequency and sample size, to consider the issue correctly would take some thought. How long does a tool usually last? Is it ever changed before it breaks? Are there any dimensional issues (out of spec or very near out of spec condition) before it breaks? Those clues will be important to developing a sane sampling plan. As a rough rule of thumb, I expect to see 5-7 samples per tool change. As far as the sample size, it depends on the short term variation. If there is no short term part-to-part variation, I would use I-MR. If there is, some data as to what that variation looks like is needed. Remember, that variation can not include roundness or parallelism error. That is measurement error, not process variation over time (which is what the charting is to analyze.)

 

[Coleman Donnelly]

Consultants who Audit on behalf of CB's - An ISO 17021:2011 violation?

Material is CO-Cr (~47hrc) Adjustments are made typicaly every 2-3 parts to try and normalize the process... do i need to demonstrate non normality by forcing the machinist to run till the tool breaks (~15pcs) without adjustments? The concern is creating scrap at high material costs...

 

[bobdoering]

Material is CO-C2 (~47HRC) Tool breaks after ~15 pcs...

The question that I am faced with is how do I get the operator to make enough parts to prove that there is rectangular distribution with out making adjustments... that will cause them to make scrap?! As a result the operator becomes the process who is trying to produce a normal distribution centered around nominal through random adjustments by checking every part...

Is it possible to prove rectangular distribution with a small sample size?

The rectangular distribution occurs when the process is adjusted. So, for example, for an external feature you set up below the spec midpoint. You allow the process to continue without adjustment until you reach 75% of the tolerance. Then you adjust to the lower control limit at 75% of the tolerance. If this continues, the sawtooth curve you get is evidence of the rectangular distribution. No scrap is made. The 75% of spec control limits should be plenty of buffer, as long as your initial frequency is generous. Once you see your toolwear slope, you can determine a more realistic check frequency.

You will know if you have a potential for a rectangular distribution prior to adjustment if you have a constant trend - upward for outside dimension, downward for inside dimension. It might not take many parts to show that, depending on the tool wear rate. For very low tool wear rates, it may take a lot of parts, but your risk is minimal because there should be virtually no discernible variation.

Again, your biggest dilemma is reaching steady state with only 15 parts. Without steady state, you are in a "state of chaos" and that would compel your operators to constantly adjust. Machining during warm-up is notorious for wandering as tool wear and temperature variation affect the dimension. You may want to use 60% of the spec as a starting point if that much buffer is needed. But, you still do not adjust until you hit the control limit - you let the machine do its job.

Without real data to analyze, it is hard to see what is going on, and what to recommend for such a case.