If you have a process that runs on 4 machines...

2 sets of 2 power presses

We use SPC to check a particular dimension on the finished part.

When we had out Paper SPC system we would use one chart for each pair of machines.

Now with our new computerized system there is only one chart. Both machines are now plotting on the same chart.

Should we react the exact same way now as when we had the 2 paper ones?

say if we get a 7 point run above the center line..

would we adjust one set of presses to bring its AVG down? or would we adj both?

all of the parts coming off of the presses all feed into the same machine at the end of the assembly lines.

When we had out Paper SPC system we would use one chart for each pair of machines.

Now with our new computerized system there is only one chart. Both machines are now plotting on the same chart.

Should we react the exact same way now as when we had the 2 paper ones?

You don't say if you were involved with the decision when you computerized but what reason do you have for doing both machines on one chart? I would treat each machine as a separate process and control chart them separately since it is unlikely they would behave exactly the same. More than likely, the original reason for 2 separate charts was to identify which machine needed attention or was causing defective parts.

say if we get a 7 point run above the center line..

would we adjust one set of presses to bring its AVG down? or would we adj both?

all of the parts coming off of the presses all feed into the same machine at the end of the assembly lines.

When you plot two or more separate processes on a single control chart, the out of control limit rules would no longer apply. By adjusting one press, I agree you can get the average down but it could be at the cost of making both presses controlled improperly and you could increase your defects in the next process. Many times a histogram of the data can show the two processes in a very visible way (a two humped normal distribution) and could be used as the argument for going back to separate control charts.

Bill Pflanz

I'm with Bill. I'd hope your SPC software requires some type of inputs to start the measuring process. One of those inputs should be "Machine #" and you should be able to sort data/graphs by that variable. Doing it the way you are sounds like it could very easily lead to "tampering" (which causes more variation).

Bill

I'm with Bill. I'd hope your SPC software requires some type of inputs to start the measuring process. One of those inputs should be "Machine #" and you should be able to sort data/graphs by that variable. Doing it the way you are sounds like it could very easily lead to "tampering" (which causes more variation).

Bill

Its not so much not being able to separate the data thats the issue... I can do that easily with the software, yes I can have them enter a machine number before they gauge and it will tag that data with the machine number... the issue is more in terms of reaction.

Right now we are testing the software so we are using both paper charts and the computer system. The two paper charts are saying that each process is currently running in control, however in the computer because all of the data is collected together it is telling the operators that they have a 7 point run and need to react, which is confusing them.

I was looking for more of a feeling of how others collect data in thier facilities when they have multiple machines all making the same component.

My thoughts on this were that we have 3 sets of Dies for the presses and to number the Dies somehow that they are easily identifiable, then create a chart in the computer for each set. So the operators would then have to choose which chart to measure on .

When we had out Paper SPC system we would use one chart for each pair of machines.

Now with our new computerized system there is only one chart. Both machines are now plotting on the same chart. I agree with the Bill's. The two processes may be as similar as you like, but no matter how you look at it you still have two processes. They should be on separate charts. As it is you are trying to control two processes with just one set of controls.

/Claes

Thanks for the opinion.

please feel free to offer any more advice that you have.

I suppose I would vote for plotting both the consolidated chart and individual machine charts. The consolidated chart can quickly react to something that is in common across the machines that perhaps the individual charts would not catch so quickly. The individual charts can of course capture a trend specific to an individual machine.

Similar situation at work here. I do hundreds of injury rate charts, at three different levels. Sometimes I see things at the higher levels I can't see on the lower levels (since they are so infrequent) and vice versa.

I agree with Steve's comment about one chart for a high level view of the process. My only concern was that Mr. Medrano wanted to change one press to get the total average in control which could be risky. He would not know which press to change without the individual charts.

He also provided additional information in his last posting that indicated that there were 3 sets of dies that could be used interchangeably in any press. If I am understanding the process correctly then he will need control charts that monitor the effect of each set on each press. That is a more complicated system to monitor since there is normal variation in each press plus variation caused by using different dies.

Let me know if I am interpreting it correctly.

Bill Pflanz

A fundamental concept of process control charts is that a single chart describes the behaviour of a single process. I just pulled some books at random off my shelf and found these:

VARIOUS CONTROL CHARTING ERRORS TO AVOID
Control charts can be misplotted for seveal different reasons. Plotting errors on a control chart will cause poor decisons to be made on the operation of the process. some specific errors to avoid are:
Measurement errors, ...
Miscalculations, ...
Misplotting, such as using one chart for two different processes, ...
Data entry erors (using SPC software), ...

(Statistical Process Control Methods for Long and Short Runs, 2nd Edition. Gary K. Griffith. ASQ Quality Press, 1996. Pages 201-202.)
Regardless of the scheme used to obtain the subgroup, the lots from which the subgroups are chosen must be homogenous. By homogenous is meant that the pieces in the lot are as alike as possible -- same machine, same operator, same mold cavity and so on.

(Quality Control, 3rd Edition. Dale H. Besterfield. Prentice-Hall, 1990. Page 74.)

2.3.3 On the basis of the analysis of the process (2.2), decide how the observed results are to be grouped so that the articles in any subgroup are produced under the same essential conditions. ...

(ANSI/ASQC B3-1996, Control chart method of controlling quality during production.)
Juran's Quality Handbook also has information on this, but I don't have my copy handy for a reference. As I recall, there is even a diagram showing two process streams being mixed.

For each process stream (set of presses) the interest is normally variation within that process over time. That monitors the process and allows making judgements on tool wear, maintenance requirements, process adjustment and so on. Even two "identical" machines (or sets of machines) will have enough differences that they are separate production processes.

Plotting the output of several processes on one chart may be an indicator of variation between the processes (Steve?) that may be useful on a higher level -- but not for evaluation of the individual process streams.

In my opinion, if the software is not capable of analyzing the process streams separately then the problem is the software. To agree with what Steve said, producing one combined chart may be useful for some purposes, but producing a chart for each process stream is essential for knowledge of how the process is working.

Plotting the output of several processes on one chart may be an indicator of variation between the processes (Steve?) that may be useful on a higher level -- but not for evaluation of the individual process streams.



I would agree. If I HAD TO choose between individual charts or a consolidated chart, I would choose individual charts. But in these days of automated data gathering and Excel spreadsheet, I don't think you have to choose one or the other.

Dr. Deming in Out of the Crisis on pages 354 - 355 points out the hazards of mixing product from three sources into one chart.

I agree with using one chart for two processes. In the past, I have had managers tell me "work on this machine here, but not that machine over there,that does the same thing". Politics.

The other reason is more statistical and human factors. It is easier to look at one chart and react to it. Look at the cockpit of of the older jets. There are hundreds of gauges. If one of them is off it probably means something, but the chances of the pilot seeing it? Now the statistical part; if one machine mean drifts, the s chart will light up like a Christmas tree! On the other hand, if there is something affecting both processes, it probably won't be detected rapidly. In some cases this could be a good thing. Another good thing about one chart is there is more statistical power with the sampling. This paragraph is valid if looking at each individually is part of the Corrective Action Guidelines (CAGs).

Decades ago I used to teach a course in "Funtional Analysis Based SPC", where the number of control charts were minimized, but I guess the concept never caught on.

Best Wishes,

Tom

I like numbers, so I agree with Steve - do both!

One composite chart for al the machines will be best for catching a special cause which occurs on all machines (e.g. a bad shipment of raw material).

A separate chart for each machine will allow you to spot problems with individual machines.

I can see you concern of "data overload". If someone has to hand check lots and lots of charts for the out-of-control rules, then it would be easy to get bored/tired/inattententive. However, since most software is pretty good at automatically flagging these rules, I don't thing that is a problems.

One other slight concern with lots of charts in that it gives more opportunites for "false positives" (kind of like multiple t-tests vs ANOVA). If you plot every combination of every set of data then there are more opportunities for finding an example of the out-of-control rules just by chance. Still, as long you are reasonable about the charts you choose to generate, I think that the benefits of the extra knowledge would outweigh the risks of false positives.


Tim F

I was under the impression that Hottelling's T Square Charts can be used for situations like this:
a. Parts from multiple cavities of a mould
b. Parts coming from a group of machines (similar to posting #1)
c. Different features from the same part.

I have not done a lot of study in this approach. Not used so far.

I did some material search:
http://www.itl.nist.gov/div898/handbook/pmc/section3/pmc34.htm

Like to hear experiences from anyone who has taken this approach.
First of all, would it be statistically valid to use this approach for the applications "a to c" mentioned above ?

Govind.

Folks,
My Dad used to tell a story about two farmer that were argueing about the age of horse. A third farmer came along, opened the horse's mouth and counted its teeth (a quantitative method of determinings a horse's age). :lmao:

My point is someone should do a simple Monte Carlo simulation. Try shifting the mean and variance etc. and see which method detects it first. The answer could be determined within an hour, and still have time for a write-up, coffee and a sticky bun.

I suppose I should be sending out my resume and get back to this type of work.

Happy 4th,

Tom

BTW, Govind, that is the best explanation of multivariate control charts I have ever seen!

This has been sort of mentioned, but I'd use individual charts to monitor each machine-process and then use a standardized short-run chart as the composite view to monitor bad raw material events or a common instrument failure. The standardized short run has the advantage of normalizing (I believe the equation is [process mean(process Y,part X) - X(i) ]/sigma(process Y,part X)) all the various machine-part-test combinations into the same reference frame. This helps th einterpretation considerably.

Hi Guys,
just registered for this forum and am well impressed with posts I've read so far. Relative newbie to SPC But have a vested interest in this thread. To kick start my contribution, I have the following questions:

I am currently attempting to implement an SPC project. I have 3 machines electrically testing same product type with same software. In essence I have 3 machines and 1 process (or so I believe) I wish to monitor equipment and process perf. Should I plot a control chart for each machine to monitor equipment. Can I plot one chart to control process?? If so how should I choose it?
I understand the concepts of xbar and range charts well. How should I decide on subgroup size? what should determine this or what would be the optimal choice??
thanks

Hi Guys,
just registered for this forum and am well impressed with posts I've read so far. Relative newbie to SPC But have a vested interest in this thread. To kick start my contribution, I have the following questions:

I am currently attempting to implement an SPC project. I have 3 machines electrically testing same product type with same software. In essence I have 3 machines and 1 process (or so I believe) I wish to monitor equipment and process perf. Should I plot a control chart for each machine to monitor equipment. Can I plot one chart to control process?? If so how should I choose it?
I understand the concepts of xbar and range charts well. How should I decide on subgroup size? what should determine this or what would be the optimal choice??
thanks

I would start off by being cautious. When you have three m/cs you have three processes. Just as you can have three process operators all supposedly making the same product, the same way with the same equipment. Regardless of the "same software", each machine may possess its own little foibles, just as might three micrometers allegedly capable of measuring 0-25mm, delivered by the same maker. I would want to know. (A little R&R, seems appropriate.) You may find it useful to test the same batch of parts on all three at first to compare results, before pursuing your SPC. Who knows how effective has been the initial set-up?

Hi Mull1,

Welcome to the Cove! First I would like you to expand on 'implement an SPC project'. What are your goals? Second I would like you to not think in terms of 'implement', but I would like you to think in terms of 'pursue'.

If you would do the above, I can make further recommendations.

Best Wishes,

Tom