Capability Study on a New Machining Process - Issues

B

brahmaiah

This was in reference to the raw material, so it is not a customer concern.

However, you are right in that any time you have a length or thickness, you should check parallelism (at least max-min over the area the dimension affects) or roundness for circular features. People measure one location out of an infinite number of locations - not good odds of getting a picture of what is happening...

As far as general convention, I would say if you are using more than 30% of your tolerance in parallelism (or roundness), you should look at another process. If you can not afford a better process, you better track that dimension like a hawk - usually using the X hi/lo-R charting methodology.
Pl.mail an X Hi/lo R chart in xles.
Thanks
V.J.Brahmaiah
 
B

brahmaiah

For simplicity's sake, the parallelism is the variation in thickness of one side to the other. A part with no variation - every location has the same thickness - would have a parallelism of 0.0000" As the variation increases, so does the parallelism. So, if you take the highest thickness from across the sheet, and subtract the lowest thickness from across the sheet, the difference is parallelism. I have prepared an X hi/lo-r chart on the last page of the attachment. It is designed to display and control that type of characteristic - particularly in precision machining. The upper char it the plot of the thickest and thinnest data on the part, the range is teh difference - or parallelism. You have precision tolerances - but I am not sure if your process falls into the definition of precision machining. In any event, it is telling.

First, you are not capable. Capability is the ratio of the process variation to the spec - and since you are both using a large amount of your spec, and have out of spec points, you are not capable. Whatever calculation you used to get 1.5 does not reflect reality.

Second, you data shows a downward trend overall. Point 4,5,6 show a shift. The other point do not show the shift as dramatically - so those locations may be influenced by what happened in locations 4,5 and 6, showing a gentler trend.


Not knowing your process (is it cutting or forming?) it is difficult to tell from a distance was the issue are. Not knowing what the process is exactly, I can not judge if you have - or should expect - a true trend. If you should expect a trend - such as in tool wear - you need to eliminate your strong special causes. As an outside dimension, tool wear should exhibit increase in size over time (if it is a cutting tool). If you have heat build-up issues, however, the tool may expand from part to part, causing a slow decrease in the dimension - similar to what you are seeing. Your fixturing may also be a problem - although it generally would affect all parts equally. Again, the variation of the raw material may also contribute to your overall variation - but without more understanding of the process, it is not clear how it might.

Hope this helps....
In the graphic file I could not find the (Data)X values in figures.
V.J.Brahmaiah
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
In the graphic file I could not find the (Data)X values in figures.
V.J.Brahmaiah

The X value portion of the X hi/lo chart is attached here. The blue dotted line is an Excel linear regression of the X hi values, providing a general idea of a trend in the data.
 

Attachments

  • x values.doc
    28.5 KB · Views: 122

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Attached is the data set for the plastic thinning X hi/lo-R chart. Also I did a gross capability calculation {(USL-LSL)/(MAX DATA-MIN DATA)} and the capability is .83 for the data set. That makes sense, since there is data outside of the specifications.
 

Attachments

  • plastic thinning data sheet x hi-lo.xls
    101.5 KB · Views: 205

Bev D

Heretical Statistician
Leader
Super Moderator
The X value portion of the X hi/lo chart is attached here. The blue dotted line is an Excel linear regression of the X hi values, providing a general idea of a trend in the data.

just looking at the run - isn't there more of a shift than a trend?
 
B

brahmaiah

Attached is the data set for the plastic thinning X hi/lo-R chart. Also I did a gross capability calculation {(USL-LSL)/(MAX DATA-MIN DATA)} and the capability is .83 for the data set. That makes sense, since there is data outside of the specifications.
Process capability should be much higher than your calculated value of 0.83. You have to ignore the values which are out of spec.Because they are caused by some stray causes which are easily preventable.Only a very few readings are out of specs.You could have investigated those stray causes if you had plotted on a standard X bar R chart which gives you more detailed information like Time,operator,any changes made during process etc.
In SPC study how you collect data is more important than the data itself.
V.J.Brahmaiah
 
Last edited by a moderator:

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Process capability should be much higher than your calculated value of 0.83. You have to ignore the values which are out of spec.Because they are caused by some stray causes which are easily preventable.Only a very few readings are out of specs.

We do not know at this time if these values can be "easily preventable". We need evidence of that. If they are special causes, they need eliminated, and new data re-evaluated.


You could have investigated those stray causes if you had plotted on a standard X bar R chart which gives you more detailed information like Time,operator,any changes made during process etc.

Yes, any SPC plotting technique needs that info for evaluation. If I had it, I would have included it in the X hi/lo-R chart too. The X hi/lo-R chart ultimately provides more feedback than X bar-R, in that details are not masked by averaging.

In SPC study how you collect data is more important than the data itself.

True, and understanding the process is more important than the data collection. What would one expect the distribution to be? Why?
 

Miner

Forum Moderator
Leader
Admin
You have to ignore the values which are out of spec.Because they are caused by some stray causes which are easily preventable.
V.J.Brahmaiah
You are confusing spec limits with control limits. A process may be in a state of statistical control and half of the measurements be out of spec. You cannot simply ignore those points.

Even if you take your statement as applying to control limits, you must be careful not to confuse actual capability based on current, total variation, and the potential capability of a process that is under control. Customers will see the variation of the current, out-of-control process, not the potential, in-control process.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Customers will see the variation of the current, out-of-control process, not the potential, in-control process.

That is correct - and using the X hi/lo-R approach shows when any of the points approaches the spec. Of course for control, all points should be within the control limit - and that charting method assures that point also. The X-bar-R chart does not reflect on whether all points are in control, just the average. That is why a common question comes up: "If the process is in control, but one of my data points is out of spec, is that OK?" Gotta love averages. An average would mask that fact - which is one reason why you can not expect an X bar-R chart to help in capability (in addition to its mathematic basis). It can only display (or predict to a degree) unexpected change.
 
Top Bottom