Precision Machining and Compressed Control Limits

Brad Gover

Involved In Discussions
Hi all, my company mills high purity graphite. Our customer has a tolerance window of +/- 0.0002". I read an article on this site "Statistical Process Control for Precision Machining Part 1", dated 12/2/08, by Bob Doering. In the article he describes how precision machining does not follow a normal distribution but due to tool wear more resembles a uniform distribution. Bob recommends calculating the control limits
[nominal +/- 0.75(tolerance/2)]. When I plot the data on a histogram, the data looks more like a normal distribution than a uniform. It does fail the anderson darling test for normality though. When I put it on a RbarS chart the control limits are 0.000009" wide. The operators do make adjustments as follows. When a part is on the mill, they will machine 9 pockets and then measure those nine pockets to determine if they need to make an adjustment. If they are on target they will machine the rest of the pockets (54 total pockets). When that part is finished, we inspect all pockets on that part. If they are on target, we will machine another part.
If not, we make an adjustment and do the 9 pocket inspection again. Question, what would normally be a uniform distribution without operator adjustment be a normal distribution with so much operator adjustment. We go through one tool for about every 2 parts. Graphite is very abrasive. We know our process is not capable to the customer specs at a 1.33 Cpk. Run charts arn't working as I am unable to determine stability due to tightly compressed control limits. Artificial control limits arn't accurate in determing stability as we change tooling often (one tool per 2 parts). Any suggustions on how to moniter process?
Thanks for the help!
Brad:cool:
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Hi all, my company mills high purity graphite. Our customer has a tolerance window of +/- 0.0002". I read an article on this site "Statistical Process Control for Precision Machining Part 1", dated 12/2/08, by Bob Doering. In the article he describes how precision machining does not follow a normal distribution but due to tool wear more resembles a uniform distribution. Bob recommends calculating the control limits
[nominal +/- 0.75(tolerance/2)]. When I plot the data on a histogram, the data looks more like a normal distribution than a uniform. It does fail the anderson darling test for normality though.

Your data might look like a normal distribution if you have other variances that contribute to your total variation equation that are significant and are also normal. Typically, that would be measurement error or gage error. Even so, the true process variation is still tool wear (unless your machine is sloppy from loose bearings, etc.) That being the case, you should have a sawtooth curve from tool wear and adjustment.

But, you threw a new variable into the equation that is usually not an issue with machining - and that is tool life. By changing the tools every two parts, your most significant process variation is actually a function of the tool changes. The variation caused by tool changes over time will likely be normally distributed. They will also be linked to other process variation such as start-up/warm-up and the resulting instability of not being at steady state during that time. That being the case, you may never really reach a truly "stable" process. You would have to have a tool that has not been invented yet - or one not costed into the part.

Very interesting problem!!

With as much tool wear as you are getting, the sampling becomes an issue. Are all of the pockets the same size (width, length, diameter, etc.)? If so, there may be some logical sampling that can be done.

You need to ensure that the measurement and gage error have no significant contribution, then focus on the tool wear and tool change effects. I doubt you have a true normal distribution, so Cpk is really a non issue. At a minimum you are multi-modal. In reality, most processes are.

For your process I still recommend the same limits.
 

Wes Bucey

Prophet of Profit
Your data might look like a normal distribution if you have other variances that contribute to your total variation equation that are significant and are also normal. Typically, that would be measurement error or gage error. Even so, the true process variation is still tool wear (unless your machine is sloppy from loose bearings, etc.) That being the case, you should have a sawtooth curve from tool wear and adjustment.

But, you threw a new variable into the equation that is usually not an issue with machining - and that is tool life. By changing the tools every two parts, your most significant process variation is actually a function of the tool changes. The variation caused by tool changes over time will likely be normally distributed. They will also be linked to other process variation such as start-up/warm-up and the resulting instability of not being at steady state during that time. That being the case, you may never really reach a truly "stable" process. You would have to have a tool that has not been invented yet - or one not costed into the part.

Very interesting problem!!

With as much tool wear as you are getting, the sampling becomes an issue. Are all of the pockets the same size (width, length, diameter, etc.)? If so, there may be some logical sampling that can be done.

You need to ensure that the measurement and gage error have no significant contribution, then focus on the tool wear and tool change effects. I doubt you have a true normal distribution, so Cpk is really a non issue. At a minimum you are multi-modal. In reality, most processes are.

For your process I still recommend the same limits.
I like Bob's response here. I'd like to add a few with questions:


  1. You mention "pockets" - I presume you are using either a single axis or multi-axis machining center. Am I correct?
  2. I'm curious about your inspection method - hand instruments? machine vision? or CMM? (especially since you are measuring 9 pockets part-way through the entire process of 54 pockets)
  3. Is the tool wear uniform? Do you inspect the tool at the 9 pocket mark or only the pockets?
  4. Have you had any tool breakage?
  5. Is the total production run large enough to economically justify a few DOE to determine optimum tool material and design, feeds, and speeds for this project?
  6. Given the machine tools we worked with, our results tallied with Bob's observation: once optimum tool material and design, feeds, and speeds were established, tool wear was uniform enough to predict and adjust settings until tool needed replacement. On long runs, we hardly ever had a tool failure and resulting nonconformance because we regularly switched out tools BEFORE wear or fatigue set in causing slow cutting or tool breakage. We found it extremely beneficial to microscopically examine tool bits regularly for anomalous breakage or wear, especially with highly abrasive piece part material.
  7. What kind of nonconformances are you finding when you examine these pockets? Are they the same? different? Is it due to tool wear or failure?
 

Brad Gover

Involved In Discussions
Thanks Bob, Wes, and my old six sigma instructor Bev! I have gathered some more information about this unique process. The mill is a multi-axis CNC Mazak. We use a CMM to measure the pockets. Our Gage R&R is

Gage R&R

%Contribution
Source VarComp (of VarComp)
Total Gage R&R 0.0000001 1.86
Repeatability 0.0000001 1.85
Reproducibility 0.0000000 0.01
Operators 0.0000000 0.01
Part-To-Part 0.0000044 98.14
Total Variation 0.0000045 100.00


Process tolerance = 0.01016

Study Var %Study Var %Tolerance
Source StdDev (SD) (5.15 * SD) (%SV) (SV/Toler)
Total Gage R&R 0.0002891 0.0014887 13.64 14.65
Repeatability 0.0002887 0.0014867 13.62 14.63
Reproducibility 0.0000152 0.0000782 0.72 0.77
Operators 0.0000152 0.0000782 0.72 0.77
Part-To-Part 0.0020998 0.0108142 99.07 106.44
Total Variation 0.0021197 0.0109162 100.00 107.44


Number of Distinct Categories = 10

We don't inspect the tool, only the pockets. Maybe we should. We change inserts out every two parts unless the tool chips. Chipped tools happen; but, not often. Our process engineers have studied speeds and feeds and I am told the program is set at optimal settings for maximizing tool wear. When I got assigned to moniter this process reciently, I first set my sub group size to be n = 4 due to the four points in the pocket that are machined at that depth. That puts my range to be just one pocket. From there I went to a subgroup size of n = 216. (54 pockets x 4points/pocket). Quality inspects every part 100% per costomer requirements. With n = 216 I'm thinking I am looking at the variation within part. When I chart an XbarS chart (see attachment chart 1) I don't see indication of your saw tooth pattern. When I chart it with n = 4 to see the variation across a part (see attachment 2) I start to see repeating patterns of first 9 pockets usually are shallow, pockets 17 thru 24 are shallow and the last nine pockets do both. Repeats from part to part in some variation. Pockets 25 thru 45 usually stay put. I did an Individuals plot (chart 3) and we have an interesting pattern within part. A "Wave" that shows up most of the time. There is always a distinct shift between pocket 24 and 25 which correlates to a hemispheric change in a round flat shape part that has 54 pockets machined across it. Could this be bearings?
Appriciate your help!
Thanks,
Brad
 

Attachments

  • Pockets.xls
    96.5 KB · Views: 174

Wes Bucey

Prophet of Profit
The results you report lead me to suspect something amiss with the positioning of the tool as it works on pockets. This could be a mechanical problem within the fixturing of the workpiece, within the Mazak, or within the program itself. The repeatability is the clue. Once a tool bit starts to go, it gets continually worse, not variable.

You are on site, you have to make a judgment call what to examine/test first to narrow the cause and implement corrective action.

I was blessed with duplicates of many of my machines and could simply transfer the work and program to a duplicate to see whether I could replicate the effect. Perhaps you could talk to your machine dealer if you don't have an in-house duplicate?

GOOD LUCK!
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Remember this with a high tool wear rate - you have the phenomena that as a tool wears you are crossing the continuum from cutting to burnishing. Cutting cuts...but as you wear the edge and it starts to have more rubbing it picks up the machine vibration variation and starts to duplicate it in the part. In longer term wear rates, you can see this in the X hi-lo/R charts as an increase in roundness (for turning, parallelism for lengths) usually over a significant period of time.

But, with your very high tool wear rate, you dash through this continuum – so that the latter cuts will have more variation (maybe even much more!) than the former. With your tool wear rate, you may even notice this pocket to pocket! This variation can only be controlled by replacing the tools even more often or ensuring the machine is in pristine shape as far as maintenance of any parts that can contribute to vibration (locking in tool holders, bearings, etc.)

As far as the shifts - does anything occur at that point in time? Anything that could affect the temerature of the tool (long wait between parts or pockets?) If it is consistent, the should be a somewhat obvious shift in the process. Look for thinks that affect temperature, material or tool shifting, etc.

I agree with Wes - GOOD LUCK!!
 
Last edited:

Wes Bucey

Prophet of Profit
Remember this with a high tool wear rate - you have the phenomena that as a tool wears you are crossing the continuum from cutting to burnishing. Cutting cuts...but as you wear the edge and it starts to have more rubbing it picks up the machine vibration variation and starts to duplicate it in the part. In longer term wear rates, you can see this in the X hi-lo/R charts as an increase in roundness (for turning, parallelism for lengths) usually over a significant period of time.

But, with your very high tool wear rate, you dash through this continuum – so that the latter cuts will have more variation (maybe even much more!) than the former. With your tool wear rate, you may even notice this pocket to pocket! This variation can only be controlled by replacing the tools even more often or ensuring the machine is in pristine shape as far as maintenance of any parts that can contribute to vibration (locking in tool holders, bearings, etc.)

As far as the shifts - does anything occur at that point in time? Anything that could affect the temerature of the tool (long wait between parts or pockets?) If it is consistent, the should be a somewhat obvious shift in the process. Look for thinks that affect temperature, material or tool shifting, etc.

I agree with Wes - GOOD LUCK!!
One of the ways you can determine whether you are "burnishing" or "cutting" is by looking at the chips.

The situation with machining is there are a lot of variables which can act independently or in concert to affect the finished product. Bob has mentioned the tool wear allowing natural vibration to become more prominent. I was lucky (smart? insightful? overcautious?) in having an entire shop of experienced operators empowered to work together to resolve problems like yours and, often, those problems never reached me except "solved" as a throwaway comment in a weekly meeting or a one or two line note in a monthly "shop events" and "lessons learned" summary.

As I have written often here in the Cove, my guys were empowered to reach out to engineers at material, tooling, and machine manufacturers and suppliers for help in resolving issues without seeking our management approval or intercession.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
One of the ways you can determine whether you are "burnishing" or "cutting" is by looking at the chips.

True - in some metals you will even see color variation or length change. Graphite "chips" might be a bit trickier!
 
Top Bottom