OK here's the question an easy question but one that we have been debating amongst ourselves for sometime now, it is your job to make hero's of some and zero's of others.

Please see the crude diagram below that represents the nuts:

0 (nut 1) 0 (nut 2)


0 (nut 3) 0 (nut 4)

1}All nuts are identical
2}All bolts are identical
3}The air tool used to put the bolts in is same
4}The person that puts in all four bolts is the same
5}The torques are all the same
6}All nuts are on the same part

The question: How many spc charts do we need for this application?
Do we need one for each nut/bolt?
Do we combine all four nuts/bolts into one subgroup?

Input greatly appreciated


[This message has been edited by pdboilermaker (edited 06 December 1999).]

<FONT COLOR="BLUE"><BLOCKQUOTE>How many spc charts do we need for this application?</BLOCKQUOTE></FONT></P>

Need is a relative term. Your need would depend upon how the data is to be used and what it is to be used for. If you want the data to control the inputs for process improvement and quality improvement, apply the chart that best suits those needs. If the data are to be used just to monitor the outputs, that is another thing altogether.

<FONT COLOR="BLUE"><BLOCKQUOTE>Do we need one for each nut/bolt?</BLOCKQUOTE></FONT></P>

No.

<FONT COLOR="BLUE"><BLOCKQUOTE>Do we combine all four nuts/bolts into one subgroup?</BLOCKQUOTE></FONT></P>

You can if you are so inclined.

I realize the answers are vague, but I would need more details to address specific issues.

<FONT FACE="Veranda,Arial">Regards,
Don</FONT>

Speaking from experience, I'd use a sample size of four and put up a visual aid so that 1,2,3,4 are always recorded in the same box on your control chart.
Assuming that there is not much variation due to the effect of each bolt, you can happily plot the average of the four readings to chart the output from your air gun.
If, however, there is an unusual effect from one bolt, it should be evident by looking at the data for that bolt number.
I've no idea of the pure statistical validity of this approach, but it was good enough to get us through Q1 a few years ago.

Cheers

Brian

Not really knowing much about what you guys are discussing I'm going to inject a little about torque that I learned as an A&P Mechanic.

Even when using a calibrated instument or tool for tightening fasteners there are variables which can effect the holding or retention abilities of those fasteners. Some of these are known as:
* Tare torque
* Dry torque
* Wet torque
* Drag torque

Also identical fasteners may be put together and have a very wide range of holding strength between them depending on machining, forging or casting variables.

An air gun may be set for a specific installation torque value, be operating correctly and still nor achieve the desired results unless other factors are identified and stipulated in the requirement.

If I'm off base let me know. But I have seen items installed on aircraft to all the proper specs and still fail because of what I said above.

IMHO;
This process doesn't appear to be an application suited for SPC. I don't see any useful info that can be gained by recording the bolt torque in subgroups.
Now, if somewhere down the line we checked the bolts to determine if they were properly torqued, then I could see the need for charting.

Brian Said:

<font COLOR=RED><BLOCKQUOTE>I'd use a sample size of four and put up a visual aid so that 1,2,3,4 are always recorded in the same box on your control chart.</BLOCKQUOTE></font>

That is a good idea. Thus there would be traceability to any data that are not in control to the bolt/nut in question.

<font COLOR=RED><BLOCKQUOTE>I've no idea of the pure statistical validity of this approach, but it was good enough to get us through Q1 a few years ago.</BLOCKQUOTE></font>

Strictly speaking, subgroup selection is based on your particular need. From Grant and Leavenworth:

<CITE>Generally speaking, subgroups should be selected in a way that makes each subgroup as homogeneous as possible and that gives the maximum opportunity for variation from one subgroup to another.</CITE>

I used to get all kinds of questions regarding subgroup selection, and it usually boiled down to what works for your application.

Randy Said:

<font COLOR=RED><BLOCKQUOTE>An air gun may be set for a specific installation torque value, be operating correctly and still not achieve the desired results unless other factors are identified and stipulated in the requirement.</BLOCKQUOTE></font>

True enough, thus the idea of monitoring the process to determine when it is not performing according to the quality plan. Under the assumption the key inputs are monitored, the results of the key inputs are directly correlated to key outputs and these KOV's values are known, the <font COLOR=BLUE>desired results</FONT> are predictable within a confidence interval. Thus, the requirement for the KOV should be determined as a result of analyzed and acceptable (specified) KIV's.

Sam Said:

<font COLOR=RED><BLOCKQUOTE>This process doesn't appear to be an application suited for SPC. I don't see any useful info that can be gained by recording the bolt torque in subgroups. Now, if somewhere down the line we checked the bolts to determine if they were properly torqued, then I could see the need for charting.</BLOCKQUOTE></font>

Sam,

I disagree here. The entire concept of Statistical Process Control is just that: Control the process. Control the inputs and the outputs will take care of themselves. As Deming stated:

<CITE>Putting out the fire after it has started is not quality.</CITE>

Paraphrased of course, but you see my point. Checking the torque <font COLOR=BLUE>somewhere down the line</FONT> and using a chart there is nothing more than fancy inspection with bells and whistles. Thus we have detected the fire after it has started and not prevented the fire from occurring in the first place (up the line where the bolts are actually torqued).

And this goes somewhat back to Randy's point, but with a twist. Sure the KIV's were to specification, but were they correlated to the KOV's. I suspect not, thus torque to specification up the line serves no purpose if what happens down the line is not correlated to it.

Process control is OK as a tool, but it is just a tool. How effective it is depends on how the tool is used. Just as I do not use a hammer to drive screws, I do not use SPC as inspection.

Just the ramblings of an Old Wizard Warrior.

Regards,

Don

"Putting out fires is not improvement. Finding a point out of control, finding the special cause and removing it, is only putting the process back to where it was in the first place. It is not improvement of the process. You are in a hotel. You hear someone yell fire. He runs for the fire extinguisher and pulls the alarm to call the fire department. We all get out. Extinguishing the fire does not improve the hotel. That is not improvement of quality. That is putting out fires."

W. Edwards Deming
1900-1993

Regards,

Don

Maybe I missed it, but what's the verdict (summary) of this one. I have been involved in a similar situation. 4 bolts, same pattern, same operator. The bolts are not controlled, I think we got them from a hardware store, the gun is not calibrated. A minumum amount of torque seems to be needed in order to make acceptable product, although we haven't figured out what that minimum is. (It's a new process)

I dunno. I have been waiting for pdboilermaker to let us know.

Don, I agree, putting out fires does not improve quality; nor does collecting SPC data at a point not relevant to the process.

In the example given, the torque reading would be that of the air tool output and not necessarily the torque applied to the nut.

In order to check for proper torque you would need to un-torque (or is it De-torque) the nuts, at some point in the process and record the results. There is a prescribed method for doing this, found in most engineering text books.
I would then use these results to determine the stability of the process.

Not checking the actual applied torque would be similar to using the digital readout for setting a machine to cut a certain diameter and then not measuring the diameter to verify the result.

The ideal solution; "Mistake Proofing" then we don't need SPC.

<FONT COLOR="BLUE"><BLOCKQUOTE>The ideal solution; "Mistake Proofing" then we don't need SPC.</BLOCKQUOTE></FONT>

Agreed.

Regards,

Don

It is my opinion (I don't know if it is right or wrong) but if the nut/bolt torques are taken and charted seperatly we are checking our subcontractors process of welding on the nuts. Whereas if we take the torques on all four nut/bolt combonations we are checking our process

------------------

<FONT COLOR="BLUE"><BLOCKQUOTE>Whereas if we take the torques on all four nut/bolt combonations we are checking our process </BLOCKQUOTE></FONT>

I would tend to agree, in part. I beleive, as I have stated before, that controlling the process is the key.

But, we need to control vendor process as well as our own. Thus, it may be suitable to monitor both.

Any statistical application must be suited to the purpose. Thus, if we choose to apply charts to all processes, we may be defeating the purpose.

Identify key processes, control those key processes and realize what may bear fruit from them.

Originally posted by pdboilermaker:
OK here's the question an easy question but one that we have been debating amongst ourselves for sometime now, it is your job to make hero's of some and zero's of others.

Please see the crude diagram below that represents the nuts:

0 (nut 1) 0 (nut 2)


0 (nut 3) 0 (nut 4)

1}All nuts are identical
2}All bolts are identical
3}The air tool used to put the bolts in is same
4}The person that puts in all four bolts is the same
5}The torques are all the same
6}All nuts are on the same part

The question: How many spc charts do we need for this application?
Do we need one for each nut/bolt?
Do we combine all four nuts/bolts into one subgroup?What about sequence ... (the old "...hand tighten and then torque every other stud (in one direction or the other) sequentially..." as in tightening wheels on an auto. Am I beyond the pale bringing up this possible 'influence'?

(Just trying to provoke an arguement http://www.qs9000.com/ubb/wink.gif )

Sharing my experience, you'd best measure the applied torque from the gun by attaching a torque transducer between the gun and the bolt head. You'll then measure torque as applied by the gun. You'll find that the final toque registered as the gun stalls will vary quite drastically from bolt to bolt if there is any unusual influence at play, such as large variation from your nut supplier, or a difference in surface finish on your bolts. Again, recording each bolt torque in a unique place on the chart will highlight these differences. Another (more)important point though is to ask the operator to record if anything changes. He/She will be used to the noise the bolt makes as it goes in, the speed at which the bolt rotates etc. etc. This will give you another indication of things changing.
All this waff pre-assumes that you've established that your chosen applied torque gives you a good clamping force on your component.
In reply to the previous thought starter, depending on the state of your mating surface, the tightening sequence can have a huge effect on the overall result.
Last point, if the component is very critical, there are transducers on the market which record either time or bolt rotation between "snug" torque (as the gun slows down) and "final" torque (as the gun stalls). This can tell you a lot about the clamping process. I've even seen equipment which relies on a big initial sample to establish an electronic "signature" for torque , angle, time etc.. then compares each tightening to this signature and sets off an alarm if anything changes.
Hope this is of some help.

Good luck

Brian

[This message has been edited by Brian Dowsett (edited 09 February 2000).]