In process statistics help

[rsm7400]

We are a custom job shop with thousands of part numbers that rotate in and out over several years. Some products we might make 50 of in a year others we might make 1000s. I would like to implement some SPC to the floor to help reduce the amount of scrap that makes it to inspection.

I am trying to figure out how to approach this. I know what key dimensions should be checked but I dont know how often to develop good statistics within a run or from one run to the next.

We are a small company (little more 100 employees) so there is no budget for big statistical software like minitab either.

If you have some pointers or have implemented in process inspection in your facility please help me get off on the right foot.

Ryan

 

[IRstuff]

The issue is generally getting the data SOMEWHERE, rather than the program to process the data. I think that Excel can do a pretty tolerable job of processing the data, once you've collected it.

Process control is about looking for trends and excessive variability. Getting the data into a viewable form is more important than the program. Much of the seminal work in SPC was done even before the availability of computers, much less SPC programs.

Don't forget that the key is NOT inspecting to weed out bad parts; it's to make sure that bad parts aren't made in the first place. That's usually a discipline problem, not a measurement or analysis problem.

TTFN
faq731-376

 

[GregLocock]

Agree with IRstuff, in fact much of our production SPC is done in red wax pencil on plastic boards.

The key issue is to understand the difference between the process potential, and what it is actually doing.

So typically we'd run 30 parts without touching the machine setup.

Plot the data on a histogram. Is it normal? hooray. Is it bimodal or uniform - ooer. Then plot the values in the order they were taken - is there a trend? Can you explain it?

Then set the machine up again and run some more parts. Do they fit in the original distribution? great. If not, you have a setup problem.

There is an enormous industry in SPC, but if you find a little book called "The Seven Tools of TQC" by McConnell or similar then you probably have enough to get you going, and is enough to run an 80 vehicle per hour production line.

Incidentally you should probably read The Goal by Goldratt.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[rsm7400]

Thanks for your input. I have ordered the book Seven Tools of TQC for reference on this and future things. SPC will most likely be recorded on the fly on some paper form.

There is a large area of the plant where we plan on implementing this. I want to start in a small area to test the waters and see about getting the system in place to see what works and what doesnt work. I also have a few people that will be willing to work with me on the floor more so than others. So I'd rather start with some of them..

Does anyone have a recommendation on starting in an area where the process is pretty consistant (like punch and die) or not consistent (lathes and other manual)?

 

[GregLocock]

You'll probably learn more initially from processes that are continuous. However in a sensible organisation they will already have been under the microscope and the scrap rates will be low anyway.

SPC on very short runs is not as useful (but there are many other things you can do).

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[IRstuff]

Getting your feet wet is better with a stable process, since it will be more obvious that something has gone awry. Inconsistent processes are, well, inconsistent, so trying to figure out the trend is much more hit and miss. However, if you can tame such a process, that'll be a bigger feather in your cap compared with bringing something that was mostly stable back into stability.

TTFN
faq731-376