Statistical Process Control for Precision Machining - Part 1

[bobdoering]

Actually, I now have a chart prepared that can control taper on a circular feature - three "slices". Shows taper left to right, right to left, bow tie or barrel. Gives the operator some idea what might need controlled in such a case! Uses the same X hi/lo-R concepts.

 

[Guest]

I am trying to incorporate this method of SPC into my company's quality system. I've already reviewed this concept with several engineers and managers, and while everyone finds this method logical, I have to present a business case for altering methods. For some reason the poor Xbar-R charts aren't cause enough. I was wondering if you know of a company who has implemented this which I could use as a real-world example.

Also, would you foresee any issues with incorporating elements of Pre-control charting into this method? It seems like it would be fairly straightforward, but perhaps I am overlooking something.

 

[bobdoering]

I am trying to incorporate this method of SPC into my company's quality system. I've already reviewed this concept with several engineers and managers, and while everyone finds this method logical, I have to present a business case for altering methods. For some reason the poor Xbar-R charts aren't cause enough. I was wondering if you know of a company who has implemented this which I could use as a real-world example.

It is not so much that the X-bar R charts are poor, they are statistically incorrect and provide no useful information. That should be adequate to discontinue confusing operators and wasting resources. I do not say this to be dramatic, but simply to state the fact of the matter.
The x hi/lo-R charts provide tool wear rate, controls the dimension completely, identifies when to make adjustment, and may even tell you when to change tools before they break. X-bar R chart cannot tell you that. They look for signals buried in measurement error.
Although companies have told me they are implementing this methodology, I do not have their specific information. The companies I had implemented the technique in the past may not still be doing it after I left due to consolidation or hiring of people responsible for quality that are not familiar with the correct technique – resorting back to what they know….X bar –R.

Also, would you foresee any issues with incorporating elements of Pre-control charting into this method? It seems like it would be fairly straightforward, but perhaps I am overlooking something.

You could use pre-control zones if you do NOT put on a mean line and you are not comfortable with 75% of tolerance control limits. The reason why you do not want a mean line is you want to AVOID the mentality of running to the mean. It is an invitation to overcontrol.

 

[bobdoering]

I am trying to incorporate this method of SPC into my company's quality system. I've already reviewed this concept with several engineers and managers, and while everyone finds this method logical, I have to present a business case for altering methods.

Here is the basis of the business case as I see it:

•Precision diameters and lengths should be primarily affected by tool wear
•Tool wear and associated adjustment for tool wear generates the “sawtooth curve”
•The sawtooth curve’s distribution is the uniform or rectangular distribution
•The uniform distribution is non-normal, and does not follow the rules of normality, such as Cpk calculations or the ‘Western Electric Rules” for control chart evaluation
•The X bar chart from the X bar – R charts represent the average of a statistically insignificant sample of measurements for a of a circular feature
•R charts from the X bar – R charts represent the range of measurement error
•Control limits are calculated using statistics for the wrong distribution – the normal distribution
•The end result of using the X bar – R chart is to provide statistically meaningless information - very little specifically about the process variation - to use to make decisions about the process on the shop floor (at a significant cost in resources.)

•X hi/lo – R charts represent the GD&T characteristics of a circular feature: diameter and the zone represented by roundness (or length and parallelism for a linear dimension)
•X hi/lo – R charts provide more valuable data, such as tool wear rate
•X hi/lo – R charts use the correct uniform distribution for precision machining
•X hi/lo – R charting techniques can be expanded to control taper
•X hi/lo – R charting techniques are easier to understand by the operators, relate directly to process variation, take less effort and resources to maintain.

Also:
•Automated tool wear compensation is not statistical process control, it is automated overcontrol.
•The algorithm for compensation becomes the process, not tool wear
•Lose some benefits of SPC because the constant adjustment masks the information
•Would help if the compensation was tracked

Not sure what business case supports the use of company resources maintaining a data analysis system that is incorrect.

 

[carleycar]

How can I define whether my process is a precision machining or not ?? I work for a company that does deep drawing on metal sheets, but I can´t define whether this is a precision machining process... Any advice ??

Carlos Carranza

 

[bobdoering]

How can I define whether my process is a precision machining or not ?? I work for a company that does deep drawing on metal sheets, but I can´t define whether this is a precision machining process... Any advice ??

Here is my starting point, from my presentation:

DEFINITION OF PRECISION MACHINING
A process where material is removed by a cutting surface – such as grinding, honing, turning, milling, etc.
The process must be controlled in a manner that all variation (vibration, bearings, gage error) is statistically insignificant except tool wear.

That being said, there are some processes that exhibit very similar variation. Stamping can be one if the effect of lot to lot material variation (a special cause) is relatively small compared to the tolerance. You can tell if it is a candidate when you make the tool to one end of the spec to allow for wear over time. For CNC turning operations, the tool life may be one shift. For a stamping die (whether forming or punching) can have a very long life. But, the distribution is still the same - a rectangle. It is hard to prove with short term data - long term data is much more convincing.

Hope that helps!