Looking for some guidance from our statistical analysis experts...

My situation is my company has been monitoring the % of poles that we have had to straighten and we have had an increase lately. 30-40 ft length steel poles are dipped in a hot metal bath to be galvanized at one of our suppliers. I had a suspicion that things were getting worse when the weather turns colder, but was told that poles are allowed to reach ambient indoors and outside temperature is negligible. Regardless of the cause, looking at the %'s below or attached, statistically can I say that something in the process has changed?

month #bent #received %bent
May-04 18 1358 1.33%
Jun-04 30 2282 1.31%
Jul-04 31 2200 1.41%
Aug-04 10 1570 0.64%
Sep-04 13 1131 1.15%

Nov-04 53 1363 3.89%

Andy,

I am not a statistician but if you do a t test, you will find that the two periods of production are statistically different. You had already suspected that but I think you were wanting more than just a mathematical proof of the difference. Just looking at your data, I have the following comments.

Not only is the % bent different in the warm and cold months but the average quantity run is about 600 higher during the warm months. I also noticed that 73% of the bent poles are Type 1T. Are you making the same products in the summer as the winter?

The histogram does show the beginning of a two hump process but there is not enough data to verify it. If you had another set of data from the prior year you may be able to draw a better conclusion.

Even if there is a winter temperature effect, you noted that the poles "are allowed to reach ambient indoors and outside temperature is negligible". Does that mean that someone is actually verifying the indoor ambient temperature is constant year round and that the poles have reached that ambient temperature before the dipping process?

Are the Type 1T poles more susceptible to bending if the temperatures are not at ambient before dipping?


Bill Pflanz

Definitely a trend is happening, based upon the p-chart (http://www.hanford.gov/safety/vpp/pchart.htm)

Besides Steve's p-charts :agree1: and Bill's insightful comments about different size runs and potentially different types of poles in each month :agree1: , I tried a scatter plot of % bent vs average monthly temps for central Iowa this year. The correlation is R-Squared = 60%, which would also support the supposition that there is some correlation. The equation is
% bent = 4.71% - (0.0483%) x (monthly average temperature)

I just got ambitious and modified Steve's P-chart spreadsheet (by the way, nice spreadsheet) to predict each month's % bend based on the regression line. If you take as given that the temperature really does affect the % bent, then you can recalculate the baseline every month based on the changing prediction. With this, then no months are out of control (although March and August are close). I don't know that I have seen control charts done this way, but it seems to make sense.


Of course, you have to be careful about assigning a causal relationship. It you look at enough variables, there is a good chance of finding a correlation when none is truly present. There are also any number of factors that could be confounded with the seasonal change. As Bill suggests, if you had similar data back over the course of several years it would be much more convincing.

Tim F :agree1:

Thanks to all for the help.
I will take a crack at the t test tomorrow, but the P-chart helps show that something has changed.

I wish we did have more historical data, but that was my first task when I started last year, to get them to start collecting data. At that time all they could tell me was there seemed to be a bent pole problem.

As far as the 1T poles that Bill had asked about, yes 1T poles are the most suseptable. The are the thinnest poles. We also ship more 1T poles than any other size, so the Pareto wasn't too surprising. Summer is also our busiest time, it is a seasonal business so that is why I had to focus on % bent.

The supplier doesn't seem to think outdoor temperature is a factor. I'm not sure why, but they are the experts and so I can at least ask them to consider what other factors may have changed since the data is showing something has. The supplier seems to focus on stresses inherent to the steel when the poles are formed. The galvanization process then relieves those stresses, causing the poles to warp. Perhaps I'll ask them about the raw steel before the poles are formed, and maybe the colder raw steel will have more inherent stresses.

Thanks again.

The supplier seems to focus on stresses inherent to the steel when the poles are formed. The galvanization process then relieves those stresses, causing the poles to warp. Perhaps I'll ask them about the raw steel before the poles are formed, and maybe the colder raw steel will have more inherent stresses.

Thanks again.

Andy - two general areas of question.

1) What is the "form" of the pole? Is it asymmetrical? If so, then there may be some inherent distortions that happen from heating, even if the heating is very uniform.

2) what kind of fabrication processes are done in the manufacture of
the poles? It is true that you will get internal stresses from many processes; look in particular for operations that might heat the material asymmetrically,
e.g. welding. You won't be able to eliminate the internal stresses but
you can improve the situation by reducing the welding heat to the
minimum possible consistent with a good weld.

If you can explain more about the fabrication processes we may be
able to help specifically.

Regards,
Brad

The supplier doesn't seem to think outdoor temperature is a factor. I'm not sure why, but they are the experts and so I can at least ask them to consider what other factors may have changed since the data is showing something has. The supplier seems to focus on stresses inherent to the steel when the poles are formed. The galvanization process then relieves those stresses, causing the poles to warp. Perhaps I'll ask them about the raw steel before the poles are formed, and maybe the colder raw steel will have more inherent stresses.

The supplier may be the expert but thinking temperature is not a factor is not as good as knowing. It sounds like they could do some DOE work for you. I don't know what kind of analysis is done on your raw materials but you may want to check that data for the same time period to rule out any changes that may have caused the problem.

Based on your comments about T1 versus the other types, you may want to separate out the defect data and other analysis for T1. It may change the histogram and control chart analysis.

If you can be successful on your analysis on this problem, it will make it easier to get them to collect other data for you. The more data you can collect the more you can use it to identify problems that even they don't know exist by using control charting.

Good luck.

Bill Pflanz

I tend to vote with Brad. My "gut" tells me if nothing is changing in your operation, something is different in the manufacture, handling, or storage of the bars before they get to you.

In the machining trade, I have seen 12 foot random length hex bars (cold drawn) have a 180 degree to 360 degree torsion twist from one end to the other as delivered from the mill. Sometimes a round bar will look like a strung bow because of bending. All are due from internal stress introduced during the bar forming process.

Are you buying from a distributor or a mill? This is a situation which should be referred back to mill metallurgist.

If these "poles" are welded tubing, certainly the forming and welding process can be causing this. If they are extruded, there are myriad opportunities for introducing stress.

I can't think of anything in your operation which would account for this condition.

The poles start out as flat sheet, coiled steel. So one form of stress is in the raw coils. On the outside of the coils the steel is wound less tight. The steel is unrolled and cut into trapezoidal shapes. Two trapezoids are cut in opposite directions (end-to-end) from the same section of the coil. They do try to stay in the middle section of the coil and away from the outside ends.

Each trapezoid is then rolled around to form the tube, (one end will have a larger diameter than the other), the seam is welded along the length of the tube and then I believe the last operation is cold-working. The tube is slipped over a solid form and rotated as rollers work along the length.

So there a lot of opportunities for stress to be induced, and then relieved in the galvanization process. Wes is correct in that when the poles bend, the take the shape of a bow. However, we are only talking about .5" - 1.0" of bow across a 30 ft tube.

The process is the same throughout the year, so it still seems strange to me that the amount of bow would double during the winter months as the trend shows.

Thanks for the input.

Interesting side note. I was informed that the Bismuth content of the galvanizing agent was increased from .080 to .100 around 10/01/04. I'm not a chemical engineer but I'm beginning to also wonder about the impact of the chemical make-up of the galvanizer.
I've asked for previous dates and chemical composition changes from our supplier.

Could it be something as simple as improper handling?
A 30' pipe being lifted in the center without support? I don't know just asking.
I now help is hard to find around hunting and Christmas maybe there are some lesser trained individuals getting involved at certain times.

Mark

Could it be something as simple as improper handling?
A 30' pipe being lifted in the center without support? I don't know just asking.
I now help is hard to find around hunting and Christmas maybe there are some lesser trained individuals getting involved at certain times.

MarkMark essentially cites "Occam's Razor" (google for enlightenment.)

Usually, it's best to look for simple explanations FIRST!

Could it be something as simple as improper handling?
A 30' pipe being lifted in the center without support? I don't know just asking.
I now help is hard to find around hunting and Christmas maybe there are some lesser trained individuals getting involved at certain times.

Mark
It is a good question and you are correct in that it could make a big difference. This question was asked up front and we were assured that nothing had changed in their hanging process.

As a side note, I found out that their last chemical change in the galvanization process prior to September was October '03. So the change from last winter to summer from 4% to 1.5% cannot be explained by a chemical change.

As there is more likely to be a larger temperature difference between the processing environment and the storage/transport environment in the winter months could the bending be caused by the zinc outer layer contracting more than the steel?

As there is more likely to be a larger temperature difference between the processing environment and the storage/transport environment in the winter months could the bending be caused by the zinc outer layer contracting more than the steel?
I wondered the same thing. As a test for this, if I brought a pole that was bent indoors to reach room temperature could I expect it to return to an unbowed condition? Or would the steel be bent in this case?

Interesting side note. I was informed that the Bismuth content of the galvanizing agent was increased from .080 to .100 around 10/01/04. I'm not a chemical engineer but I'm beginning to also wonder about the impact of the chemical make-up of the galvanizer.
I've asked for previous dates and chemical composition changes from our supplier.

Andy,

I am a chemical engineer even though it has been some time since I practiced it. I did want to comment on the galvanizer now that you have brought it up.

If the chemical analysis is a critical process parameter in the galvanizing process than it should be monitored but it may not impact a bending problem. I believe the purpose of galvanizing is to prevent oxidation and corrosion. As the galvanizing agent gets "old" it probably loses some of its effect and makes the galvanizing less effective. A galvanizing agent that has the wrong specifications could also be ineffective. I would also bet that galvanizing occurs at some controlled temperature that effects the process.

My concern would be that the temperatures may be adjusted as the galvanizing agent characteristics change. If you change the temperature to improve than galvanizing than it could impact the bending characteristics if the temperature adjustments are significant. I would also be concerned that they control the temperature the same in the winter as the summer since ambient temperature effects heating (or cooling if needed) of the galvanizing bath.

I am merely speculating on the galvanizing process but you may want to check if they control process temperatures as the agent ages or if replaced and then determine if it could impact bending.

Bill Pflanz

...
I am merely speculating on the galvanizing process but you may want to check if they control process temperatures as the agent ages or if replaced and then determine if it could impact bending.

Bill Pflanz
This is an excellent suggestion, I will ask our supplier and look into it. Thanks.

It is a good question and you are correct in that it could make a big difference. This question was asked up front and we were assured that nothing had changed in their hanging process.
I sometimes get more concerned about what is not said than I do about what is said. Are their other times that these are handled beside the "hanging process" for example setting the poles supported on both ends but no center support? just asking.
Also, once the poles bend I would be very doubtful they would ever return to their previous condition (assuming they did not start out bent). They may move to another condition but i doubt the same condition would ever be reachieved.
JMO,
Mark

If this problem is enough of a concern, give the supplier a visit for an explanation of the process and a review of the in-process data. A new perspective and set of eyes may be revealing.

One thing here is we are only using go-no go data - is the pole bent or not. Are there any continuous measure data available - such as on the amount of offset from the center line? This might be able to tell us if there are two distinct populations - very straight poles, and very bad poles, or is there a continuous shift in the poles in the winter months (even the "good" ones are more bent than the "good" ones in the summer).

One thing here is we are only using go-no go data - is the pole bent or not. Are there any continuous measure data available - such as on the amount of offset from the center line? This might be able to tell us if there are two distinct populations - very straight poles, and very bad poles, or is there a continuous shift in the poles in the winter months (even the "good" ones are more bent than the "good" ones in the summer).
Attached is the spreadsheet with raw data on sheet2. Comparing Nov data with data over the summer does not appear to show any significant difference at first glance. Perhaps others might see another trend.
Note: data from good poles is not available since we only record data from poles that required straightening.

Also - I did visit the supplier late last year to review their processes. There is a lot of pole handling going on, but the poles are usually supported in 3-4 places. I would expect handling type issues to show up in summer months as well, but you never know. It could be something to look at as well.

Andy,

I took a quick look at the data. Could you give one other bit of info? The "AmountBent" is presumably the amount of warping (to the nearest 1/16" perhaps). What is "AngleBent"? Is it the direction of the bend from some feature on the pole - perhaps the side with the seam? I don't know if it matters, but it is nice to know what the numbers might mean before playing around too much. Most are 0, some are 45,90, or 180. (Interestingly, none are 135, which seems like it should be a perfectly reasonable option.)

Tim F

Note: data from good poles is not available since we only record data from poles that required straightening.



Hmmm. This is a common mistake folks make. One consultant tells the story of a factory investigating bearing failures. They discovered that all of the failed bearings had a certain contaminant in their lubricating oil. They embarked on an expensive effort to remove the contaminant. Somehow thought finally to look at the "good" bearings. They all had exactly the same contaminant.

Moral of the story - if you only collect data from the "failures", you may be missing a significant chunk of the picture.

I think you are wasting time and effort on collecting statistical data on this situation. I would expend more effort at looking at a root cause for bending. Some of the things that come to mind:

More bending apparent in winter than in summer (are poles stored in cold and then brought to temperature abruptly in galvanizing bath? uneven heating/cooling creates stresses.)
Is there any difference in handling between seasons?
When are the tubes/posts checked for bending? How about a pre-ship check at supplier? laser levels are wonderful tools!
I'm sure some of our colleagues here can think of more avenues for investigation.

I think you are wasting time and effort on collecting statistical data on this situation. I would expend more effort at looking at a root cause for bending.

For Andy to look at the root cause he will need to collect data to verify if it is a cause. Even the suggestions that Wes provided wiill result in collecting data.

Steve noted what should be obvious but is seldom done. Separating out your bad parts from your good parts data does not allow you to look at your entire process. I also have heard of a story similar to Steve's with the same result. It is a good reminder for anyone looking at process failures.

Keep looking at the processes, Andy. If it was easy to solve your problem then you would not have asked the question here.

Bill Pflanz

Andy,

I took a quick look at the data. Could you give one other bit of info? The "AmountBent" is presumably the amount of warping (to the nearest 1/16" perhaps). What is "AngleBent"? Is it the direction of the bend from some feature on the pole - perhaps the side with the seam? I don't know if it matters, but it is nice to know what the numbers might mean before playing around too much. Most are 0, some are 45,90, or 180. (Interestingly, none are 135, which seems like it should be a perfectly reasonable option.)

Tim F
You are correct, Tim. 0 degrees represents the weld seam, and the angles represent the direction of bend in relation to the weld seam. You are also correct in that 135 would be a reasonable choice. I checked the database and found that I did not give them an option button for 135...oops... :o
So I'm guessing that some of those 45's are really 135's.

I also found out today from the supplier that their burner controls are set to 830F all year.

Hmmm. This is a common mistake folks make. One consultant tells the story of a factory investigating bearing failures. They discovered that all of the failed bearings had a certain contaminant in their lubricating oil. They embarked on an expensive effort to remove the contaminant. Somehow thought finally to look at the "good" bearings. They all had exactly the same contaminant.

Moral of the story - if you only collect data from the "failures", you may be missing a significant chunk of the picture.
It would be nice to see some histogram shapes showing the distributions of ALL of the data at different times. I agree, this may indicate a potential direction to look in. I think it will be a while, unfortunately, before I can get a data collection process fast enough to make this feasible for our pole checkers.

At this rate I may not have to worry, my friends at the Cove are going to have this one solved before long. :)

:thanx:

Hey Andy,

Let me make two "psychic" predictions and see if I am right.

1) There has been an improvement in training or equipment used in the measurements - perhaps during September or October. :magic:
2) Poles of the type "5BT" are less sturdy and/or more expensive than the more common "1T". :magic:


Now the rational...

I) There seems to be a definite trend toward "round numbers". Here is a list of the amount of bend (in 1/16" increments) and the number of poles in that category:
Bend #
7 74
8 161
9 79
10 39
11 11
12 19
13 0
14 4
15 0
16 2

Much of that is human nature. If the number is somewhere between 15/16" and 1", most people will just call it 1".


II) There seems to be a definite trend away from 0.4375 = 7/16". This can partially be attributed to the trend toward round numbers, but it can also be attributed to the fact that 6/16" = 3/8" appears to be the last "in spec" measurement. Consciously or subconsciously, the people doing the measurements seem to avoid calling the poles slightly out of spec. If fact, I wouldn't be surprised if there was a little "persuasion" applied :whip: to force some of the slightly out of spec parts back in spec.

This is where Steve's suggestion about having all the data would come in. We could check and see if there is an unusual amount of parts at 3/8" that might include a few bad parts that somehow got labeled good. :ca:


III) These trends seem to disappear :magic: in Nov & Dec! During the summer only 0-10% of the OOS parts were 7/16"; in Oct, Nov & Dec it is more like 35-40%. My interpretation is that something in the measurement process (not the actual production) changed this fall. Much (but not all) of the increase in defects can be attributed to accurately recording the slightly OOS parts as OOS.


IV) Type 5BT has no parts at 7/16" or 8/16", but a lot slightly higher. In fact, two of the most bent poles are 5TB. How do you get parts that are either in spec or badly out of spec but not borderline? Well, it could be:
* The processes really does produce two very distinct populations of parts.
* The workers could be under pressure to produce and hence either fudge the numbers, or force them into spec, or make the bad parts "disappear" (although these seem too big & expensive to have a worker drop on in the garbage.) More forcing could be expected if the parts are more easily straightened and more fudging could be expected if the parts are more valuable (and hence more pressure to have it look right).


So, how close were my psychic predictions?

Tim F

Well Tim, I'm afraid you were not too close, but there were some unforeseen circumstances throwing you off.
1) There has been no training or upgrade in the equipment used to take measurements. I found another database slip-up in that there were no hot buttons available for them to select 13/16 or 15/16. The method of measurement is to stretch a string along the length of the pole, and then use a ruler to measure from the string to the pole in the middle where the bow is.

2) The 5BT poles are larger diameter and longer poles than the 1T. They are less likely to bend, but when they do there is more length and so the bend in the middle is usually larger.

I've attached another study for those inclined. This was a 50 piece 1T pole study done last year in an attempt to study jacking bracket location effect. Jacking brackets are steel ear-shaped brackets welded near either end of the pole in pairs (180 deg apart). We determined at that time that it was best to keep the brackets welded at or near the seam of the weld in the pole, and we've kept it that way ever since.

The 0 deg tabs in the spreadsheet contain data where the brackets were welded at the seam, and the histogram tab is most interesting to me. It shows how the measurement data increased in variation from the time our supplier measured the poles after galvanizing, to the time we received them at our plant.

It is a lot of data to look at but my initial review indicates that the galvanizing process does have an impact but it also appears that something is happening during the shipping unless you are just seeing measurement error.

You may want to verify that the way the bend is being measured is consistent both at the supplier and internally. The measurement method you are describing could be introducing variation since it depends on how tight you pull the string and with what precision you read the ruler. Over time, the measurer and the technique could have been changing. I guess Tim and I were thinking of more sophisticated tools than you use.

I had suggested separating the pole type data and you seem to have verified that bend will be different depending on the type of pole. Since you have so much data, now is a good time to hypothesize the root causes and use the data to verify it. It would be a good analysis technique to show unusual or unexpected patterns and ask the process operators and suppliers if they know anything that could have caused it.

The welding has introduced another variable that needs to be explored. The more variables that you have the more difficult it will be to determine the cause if all are changing in a random way. Now you don't know if the variation is the effect of addititive changes from more than one cause or cancelling each other out.

You have definitely introduced a difficult problem but it does provide an opportunity to learn how to do process control analysis. As you can see, it is not as simple as the text books describe.

Bill Pflanz

Well,

It looks like my Sherlock Holmes imitation didn't pan out so well :o , but I still say something happened between summer and fall to cause a change in the measurements. There is a definite bias away from readings of 0.4375 (just barely out of spec) during the summer:

During Jul-Sep, 1/46 (2.2%) measured 0.4375
During Oct-Dec, 65/179 (36%) measured 0.4375



Tim F

P.S. I'll attach the worksheet with my analysis in case you're interested.

I'd lose the string and invest in a laser level and you also need at least 40 feet of flat (not necessarily "optical flat") surface to lay the post on for inspection. I fear the current inspection method has R&R features which would make the hair stand up on most metrologists' necks.

Another thing to consider in welded pipe is whether the roundness of the pipe has any relationship to reported bending.

Andy,
Earlier in this thread you said that it was a seasonal business. I am wondering, do your batch sizes change through out the year? If so then your processes likely change from batch to batch. Some processes take time to "warm up". For example, your poles get formed, welded and galvanized and these type processes almost assuredly need to be brought to temprature. Dies run different when they are cold then when they are hot, welding is different depending on temp, and I am guessing that galvanizing probably should have temp controlled also. Have you found anything in regard to these?
Mark

Andy,
Earlier in this thread you said that it was a seasonal business. I am wondering, do your batch sizes change through out the year? If so then your processes likely change from batch to batch. Some processes take time to "warm up". For example, your poles get formed, welded and galvanized and these type processes almost assuredly need to be brought to temprature. Dies run different when they are cold then when they are hot, welding is different depending on temp, and I am guessing that galvanizing probably should have temp controlled also. Have you found anything in regard to these?
Mark
Mark,
I think you are on to something. We are a seasonal business and our pole volumes just about double during the summer months, (we also see our bent pole % double in the winter). I was more focused on how outside temperature may be affecting the poles. Our supplier is more focused on the inherent stresses induced during the forming and welding processes, but they are at a loss to explain why it would be different during the winter months. It may be because they are running lower volumes, and/or it is cooler. Would poles at the beginning of a run have a higher % of bending if this were the case?

Would poles at the beginning of a run have a higher % of bending if this were the case?

Possible - this is why it would be good to have all of the curvature measurement data in the sequence the poles were produced, and if there is batches of production, and also using that as the sample subsets for xbar-R (even if batch size varies).