Does the type of out control point indicate root cause?

[davis007]

OK I have been collecting data on our production of a binary blend of two materials. A single lot is ~1800 lbs and consists of ~6 drums. For every drum we take ~3 samples of ~20 grams each. These three samples are blended and used to run an analysis in duplicate. The data are then averaged to give an estimate of the composition of the blend in that drum.

After all tests are complete I have a data point for each drum in each lot of material. I have been preparing a control chart (x-bar/S) on this data. My question is when I have an out of control point, does the reason it is out of control give me any insights to the possible special cause. For example I have two lots that are out of control by virtue of being greater than 3 sigma below the average on the X-bar chart, these same to lots are also out of control on the S chart by virtue of being greater than the control limit.

Does this tell me anything useful about where I should start looking in the process?

Again I am still new to a lot of this so any help is appreciated.

Thanks in advance

[Statistical Steven]

In Reply to Parent Post by davis007

OK I have been collecting data on our production of a binary blend of two materials. A single lot is ~1800 lbs and consists of ~6 drums. For every drum we take ~3 samples of ~20 grams each. These three samples are blended and used to run an analysis in duplicate. The data are then averaged to give an estimate of the composition of the blend in that drum.

After all tests are complete I have a data point for each drum in each lot of material. I have been preparing a control chart (x-bar/S) on this data. My question is when I have an out of control point, does the reason it is out of control give me any insights to the possible special cause. For example I have two lots that are out of control by virtue of being greater than 3 sigma below the average on the X-bar chart, these same to lots are also out of control on the S chart by virtue of being greater than the control limit.

Does this tell me anything useful about where I should start looking in the process?

Again I am still new to a lot of this so any help is appreciated.

Thanks in advance

The fact that a point is out of control does not usually point to a reason for out of control. Since you stated that the point is both low on the xbar chart, and high on the S chart would indicate that there is some significant drum to drum variability. Look at the average of the two points for each drum and see if a specific drum is low. The cause of this could be measurement error, poor blend, poor sampling, etc. Hard to say without some root cause analysis (I do not know what the 6S folks call root cause analysis).

[D.Scott]

I agree with Steven.

In addition to his suggestions, I would look at your incoming material data. Are the weights for each of the incoming materials consistent with your prior data? The sudden change in your incoming data may signify a change in the material.

Dave

[Rob Nix]

And another obvious place to look is the "recording" of the data. Perhaps one reading in the subgroup has the decimal place wrong, or some other such clerical error.

You can easily use the process of elimination, such as causes usually associated with trends, mean shifts, or cycles.

[davis007]

I have checked the data and the records are good. The out of control data point is 0.66 while the data usually falls aroung 0.75%

What I was hoping for, foolishly I expect, was that there was sort of a common knowledge when reading the control charts. Sort of a most likely hit list when you see a particular shape in the charts.

As I said this seems to have been wishfull thinking.

[D.Scott]

In Reply to Parent Post by davis007

I have checked the data and the records are good. The out of control data point is 0.66 while the data usually falls aroung 0.75%

What I was hoping for, foolishly I expect, was that there was sort of a common knowledge when reading the control charts. Sort of a most likely hit list when you see a particular shape in the charts.

As I said this seems to have been wishfull thinking.

Don't give up on your theory too quickly Davis. I think each of us learns from our own process what the probable cause is for changes. The control chart can only show you that something has changed in the normal performance of the process. Once a change is detected, we need to first verify that there was a change (rechecking calculations, measurements, etc) then investigate why the process changed. We obviously want to eliminate the special causes of variation so hopefully we don't arrive at the same root cause on every investigation. If we did, we should work on fixing that problem to keep it from interfering with our process.

Having said that, we each have learned over time that some root causes might not be fixable or maybe don't need to be fixed. These are often what would likely cause a change in our process and it sometimes is an easy fix. An example on my process might be low material level. For a number of reasons we have decided eliminating that root cause is impractical and it will sometimes slow us down but is not a big deal. We put up with it. If we see a reading OOC or a trend, our first investigation (after verification) is to check the material level.

My point is, your process may have some areas that your control charts might suggest as a root cause. It could develop into an indicator as you describe. Document what you find on your investigations and see if something shows up.

Dave

[dfirka]

I agree with all the suggestions, and let me add two comments:

1) Have you "quality" control limits? I mean, limits that clearly reflect the variability of the process during stable conditions (common causes).
Usually to calculate the limits you use a set of data taken during stable conditions that has adequate performance for the process (Cp/Cpk)

2) If there are two sources of variation (drums and lots) then make sure that the limits are calculated using both. For instance, if the limits are calculated using within sample variation, and the sample is showing only variation between drums, then the control chart can send spurious signals when being use to control lot-to-lot variation.
If that's the case (similar to a process to multiple streams) sometimes a Group charts, or 3-Way charts (Wheeler) can be used to detect sources of special causes (drums or lots).

[davis007]

I thought that the "limits" on a control chart were calculated directly from the data and indicated how the process should perform under normal circumstances subject only to its inherent randomness and that the limits said nothing about whether or not the product was "acceptable" to the customer.

In answer to your question. We do have specifications and any drum produced outside those specifications is scraped. The cpk using these specifications (Ppk as Minitab reports it is on the order of 0.75, Cpk as Minitab reports is ~1.8) is acceptable (?), and the control charts tend to indicate the process is out of control, many points outside the 3 sigma limit and I have not even tried to look at the other tests for control.

The problem is that while we currently produce material to SPEC the material is used in another process in house. This secondary process produces the finished product and is having significant quality issues. They indicate that the issue is the "variability" of the material my department produces. I have asked repeatedly for a new SPEC from the other department and they seem unwilling and probably unable to give me one. I have asked if the issue is drum-to-drum variability or lot-to-lot variability or within drum variability and they again do not seem to be able to answer that question. Thus I am trying to accomplish two things generate information that indicates our performance so that we can have a rational discussion of what would define acceptable and locate and eliminate or reduce any sources of variation to improve our product with the expectation that the specifications will be tightened.

I started by control charting all the material produced last year (2005) the control charts indicate the process is far from being in control but the Cpk and Ppk are acceptable. I have tried to attach the Minitab summary graph to this email. Because there are so many out of control points investigating each seems a bit over whelming so I repeated the process for the lots produce in Jan 2006. Again the summary graph is attached. On this control chart there are still several points that are indicated to be out of control in the Xbar but only two that are out of control in the S chart. These two also happen to be the lots that have the lowest values in the Xbar chart. So that was the reason for my original question.

Again thank you all for your insight.