We have a process that is being transferred into our facility.

This process has been running at our sister facility for a long long time.
Some time back in 1997 they had a huge push to use SPC. Seems that they went hog wild and are charting anything and everything.

We have a process coming in that makes a part similar to a washer, the part print lists the Inner Diameter and the outer diameter for this part as a Key characteristic.

The problem is that the parts are not perfectly round, so at some point you have a max and minimum diameter..its not very consistent all the way across the part.

After looking at the current data collected by our other plant I have determined that the charts they are currently plotting are completely useless. They have 2 charts one for ID and one for OD. The problem is that they take a min and max of each and plot it into the same chart. This basically shows up as a chart thats running pretty much right down the middle, and the histograms for capability look like the letter M totally bimodal split towards each end of the spec.

My first thought is to take each dimension and seperate the Min and Max into seperate charts. So there would be Min ID Max ID Min OD Max OD. My question is this though.. would I need to treat each of these like it was a one sided specification? If i use the total tolerace for the ID and OD on each chart the data is still going to be skewed towards one side.

Is there a better way to monitor such dimensions on parts that are not round?

If the issue is "roundness" - I'd plot two charts - one for the difference between the min and max of the inner diameter, and one for the difference between the min and max of the outer diameter. You then are working to minimize those differences, or at least detect if they increase.

If you also have specifications for the minimum or maximum diameter of the hole or the outside, then I'd plot those charts as appropriate to the other specs.

Actually the issue isnt roundness. Thats not even on the print.

The Inner and outer diameter of the parts are what are a concern, but because the parts are not round we are unsure of how to measure it. Because of the "egg" shape you can have a different diameter based on the two points of contact made by the gage.

right now the parts are going through a 200% inspection process with Jo Blocks in order to protect the customer.

Dont even get me started on who ever it was in the organization that agreed on a tolerance of +/- .025 mm :bonk:

You breathe on these parts too hard and the dimension changes.

My fear was that if we are charting say Minimum ID on one chart and Maximum ID on the other using the same +/- .025mm spec that our Cpk is still going to look crappy because the the minimum diameter readings are going to lean towards the smaller side of the tolerance and the Max readings will all be grouped towards the positive side.

Would plotting the average of the min and max per part be any better than how they are doing it now?

6576

You touched on my specialty - SPC of precision round dimensions!! You have determine the key reason why X-Bar R charts do not work - they cannot capture the roundness variation (it is not on the print, but it is the issue!) - and they are useless! Averages - or even individuals - mask the roundness 'error". The best chart is the X hi/low - R chart. Measure around the diameter and determine the largest and smallest diameter measurement. Plot both of them for that part. The range is the difference between them - or the roundness. The beauty of this chart is that you now have a visual indication of ALL diameters of the characteristic. One of the problems in machining round parts is that people tend to measure one diameter per part and report it. But wait, there are an infinite number of diameters in a circle!! So, you have to be a lottery winner to think you can predict all of the diameters with one measurement!! Also, a rule of life is that the operator who measures one diameter for a characteristic will measure a good one...and the customer will measure a bad one (and reject the part back to a befuddled quality manager). But, the area between X hi and X lo (by definition) describes every diameter in that characteristic! Perform a capability study, plotting 30 consecutive part OD's in that manner on one chart, and 30 consecutive part ID's in that manner on another chart. If the OD gradually increases, and the ID gradually increases, you are in luck! You have a process that is the uniform distribution - not the normal distribution. All you have to do is set up your control limits to 75% of your spec and run. For your OD, when you X hi measurement hits the upper control limit, adjust it down until your X low hits the lower control limit, then allow the dimension to grow larger as the tools wear. The ID should be treated similarly, except the dimension should get smaller over time. If you keep both the largest and smallest diameter within the 75% upper and lower control limits, you will never make a bad part (except for setup and tool change, etc.) Also, if it is a uniform (or rectangular distribution) you will have a 1.33 capability no problem. You just CAN NOT use the normal distribution calculations - they are wrong for that distribution!!!

I attached a sample blank X hi/low-R chart and a sample completed chart to get you started. Good luck! Using this charting process we were able to control a similar steel ring part to +/- .013 mm (that's right, total 23 micron tolerance!) single pointing on a CNC! Just don't tell anybody...or they will expect you to do it all of the time! :notme:

Oh, and by the way, using this technique you will be able to chuck that 200% inspection...unless your process is truly out of control. Whew!!

Sorry for the late long response I started it and left it for a while. I know that others have responded in the mean time and this may not be relevant to their comments...

The diameters shown in the picture look very large in comparison to the small tolerance for diameter size +/-0.025mm. I do not know what the function of the inner and outer diameters is but that must be considered in choosing the gauging method.

Let me say that all inspection methods make assumptions and are accomplished with an small subset of an infinite set of data points.

If you had chosen a gauging method that derived the circle size from equally spaced points around the circumferences, i.e. a CMM “best fit” diameter, you would end up with one average value for each diameter size and typically you would disregard the form error, at least until functional problems forced you to take a closer look.

Say for instance that this is a bronze thrust washer that has its function location constrained by the inner diameter and the outer diameter is in clearance to mating parts. Furthermore consider that the smallest radius of outer diameter was effectively reduced by its size and location error relative to the inner, and that this effective diameter must not breech the boundary needed for the mating part’s thrust surface.

Your gauging method should insure minimum fit of the ID so that the washer will assemble and it should likewise insure the minimum boundary of the outer diameter derived from the size, form and location of the OD fulfills its functional requirement.

These functional considerations should be guarded by the gauging method. They can be accomplished in many different ways, with the inspection tools readily available but there application or scrutiny of given assumptions should match the functional liabilities. These gauges would protect the buyer / customer.

There are other considerations in gauging that aid the process owner or producer. Say that the OD of the thrust washer was squeezed in a three-jaw chuck to process the ID and when the pressure on the OD was released the ID would assume some triangular lobbing.

There could be little (if any) variation for size detected from a two point caliper measurement because of the odd lobbing but the functional gauge “possibly an attribute go-gauge” might indicate that fit is compromised. My point is that process gauges should be complementary to process variation vulnerabilities and assist the process owner to control his process (jaw squeeze, registry variation, drift, tool signature variation, etc.) and if the two gauging outcomes (in-process and end-process) can be combined into one 'all the better' but it always must be proven that in-process measurements are not compromised by subsequent process influences.

Paul

I agree with Paul's statements on gaging. As far as the performance of the product, MMC will determine the fit characteristics. However to control the process, you will need to determine the roundness. Ignoring it will provide unwarranted variation in the data. See attached sketch - the black line is diameter measurement ignoring the roundness, and the two red lines are the X hi (largest measured diameter) and X lo (smallest measured diameter). The yellow area between them represents roundness. To control a round dimension, it contains all of the data you need. This is a much more powerful predictive tool – which is the point of SPC. The slope of a linear regression of the X hi or X lo line will give you a more accurate representation of the tool wear than the black line.

Thanks for your help guys.

Yeah I realize that roundness is an issue, getting a tooling engineer to do something about it when its not on the print is another story.


Yeah the part is large compared to the tolerances we are having to keep. Like I said I don't know who it was that agreed to the tolerance, but we seem to think it was a case of "we will say we can do anything in order to secure new business" Anyways thats what we agreed to, and the customer (Toyota ultimately) doesnt care what we have to do to meet it...like I said right now we are doing a 2-300% inspection on every part that comes off the press.

There could be little (if any) variation for size detected from a two point caliper measurement because of the odd lobbing but the functional gauge “possibly an attribute go-gauge” might indicate that fit is compromised. My point is that process gauges should be complementary to process variation vulnerabilities and assist the process owner to control his process (jaw squeeze, registry variation, drift, tool signature variation, etc.) and if the two gauging outcomes (in-process and end-process) can be combined into one 'all the better' but it always must be proven that in-process measurements are not compromised by subsequent process influences.

Paul

For these parts they use a 3 point diameter gage, custom built for them. They had done some experiments using a experimental laser diameter gage, but they couldnt get the gage to repeat at all because it was so precise, you could never hit the same spots on the diameter.

I attached a sample blank X hi/low-R chart and a sample completed chart to get you started. Good luck! Using this charting process we were able to control a similar steel ring part to +/- .013 mm (that's right, total 23 micron tolerance!) single pointing on a CNC! Just don't tell anybody...or they will expect you to do it all of the time! :notme:

Oh, and by the way, using this technique you will be able to chuck that 200% inspection...unless your process is truly out of control. Whew!!

Bob, what does the capability on these charts look like? is the histogram skewed? you said that it uses a uniform distribution correct?

The distribution is a rectangle, centered over the nominal (do not like to use the term mean, since it plays no role in this control). Should be shown on slide 9 or so of the presentation. So, as long as the LCL and UCL are 75% of the specification, the capability is the tolerance divided by the process spread - or (USL-LSL)/(UCL-LCL), an at 75% the capability is 1.33.

If the concern is the washer fitting over the mating part, why not simply use go/no go pin gages? A pin controls for your egg shape condition and custom pins aren't too expensive. For the OD issue, you could use ring gages.

Fit really is not the issue - control is the issue. To use fit, you would have to inspect 100%, but with control, you might reduce that down to between 1 every 5 pcs, or 1 every 30 pcs (or more), depending on the tool wear rate. Cuts down inspection, provides variable data for making adjustments, more information about the process, etc. Win-win.

Fit really is not the issue - control is the issue. To use fit, you would have to inspect 100%, but with control, you might reduce that down to between 1 every 5 pcs, or 1 every 30 pcs (or more), depending on the tool wear rate. Cuts down inspection, provides variable data for making adjustments, more information about the process, etc. Win-win.

Yeah, basically we are doing a 300 percent inspection right now with 2 sets of Jo Blocks and a 3 point diameter gage.

I am going to have the operators get me some sequential sample data off 30 pcs. and plug them into the spreadsheet you gave me to see what things look like. I know for a fact that we are not capable, otherwise we wouldnt be 300% inspecting.

Another thing is... these charts, do they assume that all special causes of variation have been identified leaving only tool wear as a concern? The Tooling engineers are telling me that this isn't a tool wear issue, they just cant hold the dimension wanted by the customer.

This die that runs these parts is really a pain in the butt! It crashes constantly. I cant wait until its in our facility full time :bonk:

Although the uniform distribution assumes the primary variable is tool wear, the charting will still prevent reject parts from being made if the total variation is consistant and predictable. Not sure if it is at this point - but this chart is a clearer picture of the process than measuring one diameter per part, or averaging the measurement mess! If it is not predictable, then you really do not have a controllable process, and you are left to sort for life. I hope that is not the case!!

And, if it is not a tool wear issue, then what are they offering up as the root cause? Material variation? Machine capability? Poor tool design? There are only so many things...

Well, if it is not a tool wear issue, then what are they offering up as the root cause? Material variation? Machine capability? Poor tool design? There are only so many things...

I would say a combo of Tool Design and Machine Capability. I know they are in the process of putting together a huge Capital Appropriation request.

Well, don't jump to conclusions until the data shows it. And, if you do go for a new process, use this charting system to verify it is an improvement!

Well, don't jump to conclusions until the data shows it. And, if you do go for a new process, use this charting system to verify it is an improvement!


Oh totally agree there.

Well, don't jump to conclusions until the data shows it. And, if you do go for a new process, use this charting system to verify it is an improvement!


heres some data, as you can see... one Diameter is better than the other, but still out of control.

6590

Yes, the OD tolerance is clearly outside of the capability of the process. The out-of-roundness is eating up over 50% of your tolerance!! The ID would have been OK, except for that blip at the end. Was it real? Anyway, now you can say without exception that you are stuck with sorting the OD - unless someone that knows your process can make some improvements. If that ID had not had the one bad data point, you could have SPC'ed the ID one very 10 parts or so and been done with it. Any theories why the ID is so much better than the OD? Any lessons learned there?

Do you feel as if you really know more about your process with this charting technique than the ones you have tried before?

Yes, the OD tolerance is clearly outside of the capability of the process. The out-of-roundness is eating up over 50% of your tolerance!! The ID would have been OK, except for that blip at the end. Was it real? Anyway, now you can say without exception that you are stuck with sorting the OD - unless someone that knows your process can make some improvements. If that ID had not had the one bad data point, you could have SPC'ed the ID one very 10 parts or so and been done with it. Any theories why the ID is so much better than the OD? Any lessons learned there?

Do you feel as if you really know more about your process with this charting technique than the ones you have tried before?

Dont know, Im going to show this to a couple of the tooling engineers and see if it helps them. They are the ones trying to improve the process.

Yeah that flyer was real on the ID, at least as far as i know. I asked for confirmation of the part, but like I said before this process is currently in another facility 40 miles from here, I havent gotten a reply to my inquiry yet.

Above all else you have helped me understand the how/why its not really possible to get a true (normal) capability analysis on this.

All you have to do is set up your control limits to 75% of your spec and run. [/B]

I haven't made a study of SPC and roundness, or all of the posts in this thread, so maybe I missed something.

But that statement up there sounds a whole lot like something that is NOT SPC as far as I've ever seen it.

Setting control limits based on a specification? What????

That is because everything you have seen so far is based on the normal distribution, and precision machining is not. The issue is not roundness, it is precision machining that makes the difference. So yes, the rules are different - and it is true you likely have not seen this before. But, it is statistical...and it is process control. And, most importantly, it is correct. If you are in precision machining, I suggest you go through all of the postings in this thread. If not, then it may not apply to your processes at all.

The point I was making is that control limits are established based the amount of variation the process would be expected to exhibit if it were subject only to chance causes. This has nothing to do with specification limits.

So, setting control limits based on specifications...doesn't sound like statisitical process control. It might be a useful technique - I can't judge that. I'm not sure that the normal distribution has anything to do with it because control charts work with non-normal data.

Can you suggest a reference that explains the math behind setting a control limit based on a specification, and not on the natural variation of a process?

That reference should be out by the end of the month. Until then, you need to know that in precision machining the most significant common cause is tool wear, it creates the sawtooth curve, that is the uniform distribution, and its controls are based on the probability of that rectangular distribution - which is much different than the normal distribution. The basics and the math have all been described in previous posts. "Chance causes only" only works for processes that exhibit natural variation - such as heat treat or the height of baked loaves of bread - and those processes can be controlled by normal distribution statistics. Precision machining does not fall into that category. In fact, if you get a chart that shows the normal distribution in precision machining, it is typically evidence of incorrect charting or a process out of control.

I assume here that you are using some type of chart that is based on the theory of a uniform distribution...

...in the same way that a p-chart uses binomial distribution theory, and a c-chart uses the poisson distribution theory for setting control limits.

"control limits are established based the amount of variation the process would be expected to exhibit if it were subject only to chance causes"


I think it is more accurately put as: "control limits are established based the amount of variation the process would be expected to exhibit if it were subject only to common causes." Again, the "chance cause" is based on phenomena that have natural deviation, as described above. Practically all SPC texts are based on this, and it works for those types of processes. The particular issue of the correct SPC for precision machining and its unique distribution has not been dealt with well at all in the texts. We can not blame Dr. Shewhart for missing this condition, since the state of precision machining in the early 1930’s is not the same as today. Those that have applied these techniques understand clearly why this technique is correct, and that it works.

I assume here that you are using some type of chart that is based on the theory of a uniform distribution...

...in the same way that a p-chart uses binomial distribution theory, and a c-chart uses the poisson distribution theory for setting control limits.

Yes, the X hi/lo -R chart works very nicely. It is described in previous posts.