Moving all of our SPC charts to an Electronic System - Software to flag OOC Condition

[RMedrano]

We are in the process of moving all of our SPC charts to an Electronic System.

One of our processes that stamps out brackets on a progressive die is now the big focus of getting hooked up to the electronic system because of some recent customer concerns.

On the current SPC chart, the operator takes one measurement from the left and right side of a bracket, finds the average of those two readings and writes that number on the chart as one of the 5 samples in the subgroup. This is repeated on 4 other brackets to get 5 sample readings. The Average and Range of the 5 Samples is then calculated and plotted on an X-Bar & R chart.

Our Electronic software can do this, but in a very round about way.

The program takes the individual left and right readings and finds the Average and Range of them, plots the values and then sends the Averages of the left and right side to an equasion which finds the average of the lefta nd right side.

The average value is plotted on a seperate chart, but because this chart only has a subgroup of one (limitation of the SPC software, any variable that gets it value from an equasion can only have a subgroup size of 1) The software only lets me chose X&MR as the default chart type.

My question is this basically. Does it make sense to have the software flag any out of control situation on any of the three charts? and force the operators to react?

An engineer here made the suggestion to have the operators react if the Range goes out on Individual L or R charts goes out and ignore the Xbar and only react if the Average on the X&MR chart goes out, and ignore the MR

does this make sense? I may not have explained the situation very well LOL

[Jim Wynne]

It sounds like you're allowing the tail to wag the dog. If X/MR is a viable alternative, why weren't you doing it before? Without knowing anything about the part other than what you've described, I also think that the practice of making two measurements and averaging the results is a little questionable unless the resulting number can be related to the real-world application of the part. There is a possible simple solution, which would be to use your current sampling and charting strategies and randomly measure one end of the part or the other, which would, in the long run, give basically the same results--if the decision as to which end of the part to measure each time is actually random (i.e., each end has an equal chance of being measured each time).

[D.Scott]

I would like to ask what software you are using that only allows single sample subgroups based on results of an equation. If you are using script files to do the original entries and subsequent calculations, simply repeat the process on the next line of the script file. This can be done for as many subgroup samples as you determine are needed in the final data file.

If your software won't allow you to write multiple lines in your script files, I would suggest you look into another program. There are some free stand alone programs here at the Cove available for downloads that will allow the calculations you want.

Dave

[RMedrano]

In Reply to Parent Post by D.Scott

I would like to ask what software you are using that only allows single sample subgroups based on results of an equation. If you are using script files to do the original entries and subsequent calculations, simply repeat the process on the next line of the script file. This can be done for as many subgroup samples as you determine are needed in the final data file.

If your software won't allow you to write multiple lines in your script files, I would suggest you look into another program. There are some free stand alone programs here at the Cove available for downloads that will allow the calculations you want.

Dave

We are using WinSPC from Datanet Quality Systems.

Basically what I have discovered is that because of the way WiNSPC stores and writes information to the SQL database is why it does not allow you to make calacualtions on anything other than certain values. (this is what i have been led to believe from thier tech support)

Basically there is no way to manipulate or use any of the individual samples that make up a subgroup, only the results of that subgroup.

I may have not made it clear about the part, so i included a little picture of the gauge. you can see the two black probes where the measurement is taken.

LIke i said before we use the average of those two measurements as a sample on our X bar & R chart.

so if you have 5 Brackets

Bracket 1 = AVG(L1+R1)
Bracket 2 = AVG(L2+R2)
Bracket 3 = AVG(L3+R3)
Bracket 4 = AVG(L4+R4)
Bracket 5 = AVG(L5+R5)

all 5 averages are then added up and then the average of those is taken and plotted on the SPC chart, Range is calculated using the High and low averages of the brackets.

With the electronic system I basically have to calculate it like this There is an XBar and R chart for each side where the measurement is taken.

Left side = AVG(L1+L2++L3+L4+L5) Right Side = AVG(R1+R2+R3+R4+R5)

The results of each of those is plugged into a third chart that is plotting this average value
Strut Seat =AVG(Left side)+(Right Side)

Because that equasion only gives me 1 value, I am limited to using XBar&MR as my only choice in the system.

I can tell it that the equated value will consiste of 5 numbers in the subgroup, but the system will get the one value from the equaion and plug in o's for the rest of the sub group.

[Steve Prevette]

I would suggest against automating SPC rules. It removes the human element. We need a person to look at the chart. Dr. Deming warned against the "unmanned computer". If you have a person do the SPC rules, then at least you are guaranteed that at least one person has looked at the chart.

Besides, it only takes about 10 seconds to evaluate a chart for the SPC rules. And you need a qualified statistician to look anyway, as there is also the decision of how and when to set the baseline averages and the control limits.

[Tim Folkerts]

I like sticking to charting X&R for the L & R ends separately and skipping the charts the combined charts, for basically the same reasons JSW05 gave. I have a hard time thinking how the average plays into the quality of the part separately from the two ends.

As to having the operators react, I understand Steve's point about having real, live people involved, but I don't see the problem with having the computer flag errors. Human inspection is notoriosly unreliable. Most people in most situtation consider visual inspection a step backwards.

How the operators might react needs to be thought out. Being OOC means the process (probably) isn't performing as well as it should, but it doesn't necessarily mean the process is out of spec. The reaction needs to consider both OOC and OOS when deciding how to react.


Just my $0.02

Tim F

[bpritts]

[QUOTE=Tim Folkerts]I like sticking to charting X&R for the L & R ends separately and skipping the charts the combined charts, for basically the same reasons JSW05 gave. I have a hard time thinking how the average plays into the quality of the part separately from the two ends.[quote]

I'm with Tim and JSW05. I have a real hard time understanding why you want to chart an average of two different features. One of two things is true:

A) The two ends are statistically independent

B) The two ends are correlated somehow.

If (A) is true, then you are losing data by averaging them.

If (B) is true, then you don't need to track both of them. Also, I am
suspicious that the statistical assumptions may be violated, altho
I would need to think about it a little more to be sure.


Can you explain why you are using this approach?

Thanks!

Brad

[RMedrano]

Thanks guys for the replies,

I couldnt give you a reason as to why its gauged that way, only thing I can think of is that the two points that they are measureing are where alot of the work is being done with the tooling in the die. This is really the only part gauged this way, and the gauge itself is specific to just that particular part.

all of the other parts in our plant are measured using standard XBar&R or Median&R



And the data collection isnt automatic, Its just being done in a computer rather than a paper chart, So there is always a person interacting with it.
THe operator has to phsically measure the 5 brackets with the gauge that is pictured above. There are gauge cables that run from the Mitutoyo indicators to the computer. The computer is simply applying the SPC rules and it flags the operator for information on Assignable cause, Corrective action etc if a rule is violated.

It also sends out an email to me and the 3 other technicians in the quality department if an out of spec part is entered into the system so we can follow up on that.

Its my job to transfer all of this charting to electronic methods throughout the plant. I have nothing to do with the setup of the original charts or charting methods.

I would love to just use a standard 5 sample X-bar and R chart it would make my job much simpler!

I will ask the program engineer if there is a particualr reason why this part is beaing measured the way it is, and see what he says.

Again thanks for all the input.