Looking for some opinions on the way a particular process is being monitored.

For the sake of simplicity, Im going to use an example of something similar to the actuall process in question, but wont take me as long to explain.


Lets say you making Ladders, and you are trying to monitor / control the amount of variation in the width of each rung of the ladder.

Right now as it is set up the process is measuring All the rungs (30) on one Ladder every hour and is putting it on an X-Bar and R chart.

Now first off, I would have them use an X-Bar and S due to the relativly huge sample size. My main concern though is that this isnt really capturing any part to part or in this case ladder to ladder variation, only within part variation.

Would it make more sense to say, measure a sample of rungs (top, middle, bottom) from different ladders (the last 5 produced maybe?) every hour?

Or does it not matter which way it is done?

To answer your question intelligently, we would need more information on the process. How are the rungs made?

If for example. the rungs were aluminum extrusions made at one time, cut to length and later assembled to the ladder as pulled randomly from a container, SPC on the rung after assembly would make no sense at all. The proper method would be to perform SPC at the point where the rungs were made, that is, at extrusion.

What is your process relative to the rungs and where is the width characteristic controlled?

To answer your question intelligently, we would need more information on the process. How are the rungs made?

If for example. the rungs were aluminum extrusions made at one time, cut to length and later assembled to the ladder as pulled randomly from a container, SPC on the rung after assembly would make no sense at all. The proper method would be to perform SPC at the point where the rungs were made, that is, at extrusion.

What is your process relative to the rungs and where is the width characteristic controlled?

Totally agree with you if the rungs were made in a seperate process.

In this "example" the ladder starts as a solid piece of steel, and the rungs are made by notching or punching out the material its all one solid piece nothing is assembled.

Think of it as a power press that punches out windows in a piece of steel, the material that remains is what makes the "rungs" of the ladder.

In this situation, each rung is a unique characteristic independent of the others. Look at this a multi-cavity mold/die, or multiple process streams.

Perform a statistical study of each rung over time to verify stability and to quantify capability. Make any necessary adjustments to the target value, then select one representative rung to track for process control.

You may want to periodically check all rungs to verify that the one rung is still representative, but not at process control frequencies.

In this situation, each rung is a unique characteristic independent of the others. Look at this a multi-cavity mold/die, or multiple process streams.

Perform a statistical study of each rung over time to verify stability and to quantify capability. Make any necessary adjustments to the target value, then select one representative rung to track for process control.

You may want to periodically check all rungs to verify that the one rung is still representative, but not at process control frequencies.

Make sure I make this clear.

All the windows that are punched into the part are made by the same tool.
there are not 30 Die cavities. Would doing a statistical study of each of the "rungs" still make sense?

It depends. if your assumption that each rung within the ladder is 'the same' (since each are made with the same tool) then it makes no sense to measure more than one rung per ladder. (if you assumption is correct, rung to rung is a small component of the variaiton and ladder is much larger) In this case you could consider measuring a random rung on 3-5 sequential ladders. the range chart is composed of the min rung to max rung but also measures ladder to ladder...

of course depending on volumes and what can go wrong, you might want to measure rung to rung within a ladder and ladder to ladder. I woudl stronlgy recommend gettign a copy of "innovative control charting" by steven wise. it covers the situations where mulitple components of variation need to be monitored...particularly when within piece variation must be monitored in a ddition to piece to piece over time.

It depends. if your assumption that each rung within the ladder is 'the same' (since each are made with the same tool) then it makes no sense to measure more than one rung per ladder. (if you assumption is correct, rung to rung is a small component of the variaiton and ladder is much larger) In this case you could consider measuring a random rung on 3-5 sequential ladders. the range chart is composed of the min rung to max rung but also measures ladder to ladder...

of course depending on volumes and what can go wrong, you might want to measure rung to rung within a ladder and ladder to ladder. I woudl stronlgy recommend gettign a copy of "innovative control charting" by steven wise. it covers the situations where mulitple components of variation need to be monitored...particularly when within piece variation must be monitored in a ddition to piece to piece over time.


ok I will look into that book! Thanks for your help and opinion it is appreciated.

Make sure I make this clear.

All the windows that are punched into the part are made by the same tool.
there are not 30 Die cavities. Would doing a statistical study of each of the "rungs" still make sense?

So the "ladder" is indexed through a single punch 30x? This definitely makes a difference. For future reference, the more information that you can provide in your original post, the better answer you will get.

I would recommend the following:

1. Measure all 30 rungs on 5-10 consecutive ladders. Plot the data as individual points in the exact sequence that it was punched. Look for repeating patterns, and for clues to the major source of variation.

2a. Run an Xbar/R chart using subgroups of rung widths within a single ladder. However do not use all 30, use a more reasonable subgroup size.
2b. Run an Xbar/R chart using subgroups of rung widths from consecutive ladders. Again use a reasonable subgroup size.

Collect subgroups for both studies at the same time interval. Calculate control limits and plot the data for each. Typically, one of the charts will either hug the central line or will appear wildly out-of-control compared to the control limits while the other chart will appear relatively normal. The subgrouping scheme that provides the latter chart is the one to use.

Note that while this will usually work, it does not work 100% of the time. When it does not work, look at the usual suspects (e.g., high gage error, inadequate gage resolution, form variation in the part itself, etc.).

I would recommend the following:

1. Measure all 30 rungs on 5-10 consecutive ladders. Plot the data as individual points in the exact sequence that it was punched. Look for repeating patterns, and for clues to the major source of variation.

2a. Run an Xbar/R chart using subgroups of rung widths within a single ladder. However do not use all 30, use a more reasonable subgroup size.
2b. Run an Xbar/R chart using subgroups of rung widths from consecutive ladders. Again use a reasonable subgroup size.

Collect subgroups for both studies at the same time interval. Calculate control limits and plot the data for each. Typically, one of the charts will either hug the central line or will appear wildly out-of-control compared to the control limits while the other chart will appear relatively normal. The subgrouping scheme that provides the latter chart is the one to use.


For 2a, im not sure Im understanding, the entire chart is being generated from just one ladder? so with 30 rungs, using 5 of them as a subgroup id then have a chart with 6 points, or are we talking about doing that on 5-10 ladders on a single chart?

I think you mean to do that on say 5 consecutive ladders so then i have 30 SG of data correct?

tracking the order in which they are punched may be difficult based on how they are processed, but Im going to talk to the Mfg engineer and see what we can do.

Thanks for all your input.

Right now like I said they are only measuring one part evert 2 hours measuring every "rung" so they have these huge SG's

I was always taught anything with more than 10 in a SG should be using Xbar and S at a minimum.

For 2a, im not sure Im understanding, the entire chart is being generated from just one ladder? so with 30 rungs, using 5 of them as a subgroup id then have a chart with 6 points, or are we talking about doing that on 5-10 ladders on a single chart?

I think you mean to do that on say 5 consecutive ladders so then i have 30 SG of data correct?

tracking the order in which they are punched may be difficult based on how they are processed, but Im going to talk to the Mfg engineer and see what we can do.

No. For example, Subgroup 1 = measure 5 consecutive rungs on 1 ladder, wait a specified period of time (or quantity of ladders), then repeat Subgroup 2 = 5 consecutive rungs on the next ladder, etc.

I guess I am old fashioned and use multiple charts. I would keep the Xbar/S chart to monitor within ladder variability and then use an x/movR to monitor the between ladder variabilty. Now having said that, the best approach is to do variance components to identify your sources of variation. If within ladder variabililty is small, you pool the rungs and use the x/movR chart. If the ladder to ladder variability is small then the Xbar/S chart is fine.

I guess I am old fashioned and use multiple charts. I would keep the Xbar/S chart to monitor within ladder variability and then use an x/movR to monitor the between ladder variabilty. Now having said that, the best approach is to do variance components to identify your sources of variation. If within ladder variabililty is small, you pool the rungs and use the x/movR chart. If the ladder to ladder variability is small then the Xbar/S chart is fine.

I wouldnt call that old fasioned, sometimes its the only way to do things correctly. Hell I have one machine that has 14 individual SPC charts on it : )


I think when whoever it was that set up the SPC in this are just didnt have a firm grasp on exactly what it was they were going to measure or why.

So like alot of times i see people just measuring for measurements sake, I guess it makes them feel better about thier process.

No. For example, Subgroup 1 = measure 5 consecutive rungs on 1 ladder, wait a specified period of time (or quantity of ladders), then repeat Subgroup 2 = 5 consecutive rungs on the next ladder, etc.

ok thanks for clearing that up!

I guess I am old fashioned and use multiple charts. I would keep the Xbar/S chart to monitor within ladder variability and then use an x/movR to monitor the between ladder variabilty. Now having said that, the best approach is to do variance components to identify your sources of variation. If within ladder variabililty is small, you pool the rungs and use the x/movR chart. If the ladder to ladder variability is small then the Xbar/S chart is fine.

This exactly what I would do and is how steve wise covers it in his book. both componenets of variation need to monitored in some fashion.

This exactly what I would do and is how steve wise covers it in his book. both componenets of variation need to monitored in some fashion.

Hey Bev...we agree on something :):):D

Hey Bev...we agree on something :):):D


Maybe the thread should be made a sticky LOL :lmao:

Hey Bev...we agree on something :):):D

more often than not actually

I would recommend the following:

1. Measure all 30 rungs on 5-10 consecutive ladders. Plot the data as individual points in the exact sequence that it was punched. Look for repeating patterns, and for clues to the major source of variation.

2a. Run an Xbar/R chart using subgroups of rung widths within a single ladder. However do not use all 30, use a more reasonable subgroup size.
2b. Run an Xbar/R chart using subgroups of rung widths from consecutive ladders. Again use a reasonable subgroup size.

I whole-heartedly agree with step 1, but I would not jump to 2a or 2b until I was satisfied that I understood 1. It might tell me to look at both ends and middle of each sample ladder - the locations towards the end may affect the hole because of less material nearby. It may indicate one hole per ladder is fine, because there is not statistically significant difference for the holes within each ladder, so measuring a sample within the ladder would be a waste of time. The sample may be 1 hole per ladder for 5 ladders in a row per sampling period. The correct answer is - as usual - it depends.

whatsa sticky?

A permanent post