Hi all,

I am new to to this industry and want to know what is appropriate control chart to use. The customer requirements are minimum point insulation thickness and average insulation thickness of stranded wires.

Thanks,
bbzen

Dear BBZen

To decide which type of control chart you need for your probuct,please give us following information about the process:
a)operation name
b)Machine employed
c)Is the operation AUTOMATIC OR SEMI-AUTOMATIC OR CNC OR MANUAL?
d)Are you carryingout variable measurment or attribute gauging?
e)Is it mass production or batch production(How long is the production run)?

In the meantime I WILL SEARCH FOR A SUITABLE FLOW CHART which should take you to the required control chart.

V.J.Brahmaiah:agree:

For the sake of starting the discussion, I will assume this is an extruded coating. The first thing you should do is a capability study. You will need to understand these things:

-What is the variation of the coating thickness around the wire? Is that variation significant versus the variation along the wire or the specification? This is important, because if you measure one thickness about the wire for each sample, then plot the sample data - you may mistake the signals you get for the process as changes along the wire, when it is actually just the variation around the wire. The difference is very significant - as you may attribute the variation to the wrong cause, and make wrong corrections, or it may mask true over-time variation and give you no direction.

-What are the variables that affect the variation? Do they shut down the machine for breaks, lunch, etc.? Is that information marked down on the chart? What would you adjust if you found an out of control condition? Temperature? Speed? Pressure? Do the gages for those variables pass gage R&R? Do they have enough resolution (so you can adjust with their information)? Are those parameters capable? Do you need to do a DOE to find out which parameters affect the output the most to focus on them? There is a possibility that you may end up doing a "report card chart" for the thickness (for evidence to the customer of aggregate control), but actually control the process using a process parameter, such as speed or pressure.

-Do you coat one wire at a time or several? If more than one, study each wire individually - DO NOT COMBINE DATA (you will get multimodal distributions)

Excellent recommendations from bobdoering!

Another consideration if this is an extrusion process is whether there is any autocorrelation. Autocorrelation exists when one measurement depends on the measurement immediately prior to it in time.

This is a link (http://elsmar.com/Forums/showthread.php?t=29301&highlight=autocorrelation) to a discussion on autocorrelation.

When autocorrelation exists, do not use SPC methods based on subgroups. You will need to determine the duration of the autocorrelation, then use an individuals based chart with a measurement frequency greater than the period of autocorrelation.

Another consideration if this is an extrusion process is whether there is any autocorrelation. Autocorrelation exists when one measurement depends on the measurement immediately prior to it in time.


Interesting point. If you do discover autocorrelation, take some time to find out a logical reason why it exists. Do you have cycling or pulsing from bad feed bearings or material loading variation? What its cause is has a lot to do with how you handle the data.

Actually, the things that you mention make it less likely to find autocorrelation because they will introduce significant part to part variation. If you perform an autocorrelation study on your precision machining processes, you will probably find that they are also autocorrelated until a significant amount of tool wear occurs.

My first experience with SPC was with an extrusion process. Ford came in and mandated Xbar & R control charts (subgroup size of 5; one subgroup per hour). This was also before MSA. The control limits were extremely tight and every subgroup was out of control on either side without making an adjustment.

We finally realized that the control limits were based entirely on the measurement error. There was no part to part variation due to autocorrelation. Studies revealed that part to part differences did not show up for 20 minutes.

We converted to an I-MR chart with samples taken every 20 minutes and it worked beautifully.

Actually, the things that you mention make it less likely to find autocorrelation because they will introduce significant part to part variation.

Depending on the feed issues, some materials, in some extruders, can slowly change over time as the back pressure from the feed lessens as the material level lowers. When you drop in more material, it can boost back up, and continue on a downward change until the next load. Dropping in chunks of butyl rubber in an extruder may do this more than a plastics extruder feeding pellets. This will give you a rough autocorrelation, and a rough sawtooth curve. Quite frankly, it has been a long time since I dealt with extruders, but now that I recall that problem, it makes me think there may be another uniform distribution hiding there.

It all depends, and unlike machining, there are many more variables that can become significant.

Actually, the things that you mention make it less likely to find autocorrelation because they will introduce significant part to part variation. If you perform an autocorrelation study on your precision machining processes, you will probably find that they are also autocorrelated until a significant amount of tool wear occurs.


Yes, they are autocorrelated due to tool wear. That time function is what makes the distribution a continuous one, rather than a discrete one. But, we can control it rather readily thanks to the continuous uniform distribution.

My first experience with SPC was with an extrusion process. Ford came in and mandated Xbar & R control charts (subgroup size of 5; one subgroup per hour). This was also before MSA. The control limits were extremely tight and every subgroup was out of control on either side without making an adjustment.

We finally realized that the control limits were based entirely on the measurement error. There was no part to part variation due to autocorrelation. Studies revealed that part to part differences did not show up for 20 minutes.

We converted to an I-MR chart with samples taken every 20 minutes and it worked beautifully.

This is an AWESOME story! Just goes to show you can't rubber stamp those X-bar R Charts! Control limits based entirely on the measurement error are way more common that people realize. So are processes that have become normal due to overcontrol as the operators have become the process. One of the line items of TS16949 is that in implementing SPC, your people must understand overcontrol! C'mon! Most of the people looking at the process do not recognize it!

Good stuff!

:)Thanks a lot.

Our's is a single coating extrusion process continous in nature and will stop only when the parts produced are out of specification beyond on line adjustment. The process is semi-automatic and controlled by an OD scanner which is the reference point of the speed of the extruder, when the OD fails the machine will automatically scrap and the machine compensate by increasing or reducing extruder speed.

But the operator will manually check the min. point insulation thickness when it fails he will adjust the 4 bolt to re-center the conductor (multi strand, the smallest is 7 strand and 65 for the biggest wires) with thickness ranging from 0.24 to 0.60mm.


I'd tried to gather data on the min point insulation and plot to have the a better view of the process but the sad thing is that the minimum point is not always on the same spot (quadrant) because the wire twist during the process and besides when the bunched wire (multi strand wire) is not that good it will artificially shift the min. point to other location w/ no trend/pattern.

Honestly I am new to the SPC concept and very eager to learn.

thanks,
bbzen

But the operator will manually check the min. point insulation thickness when it fails he will adjust the 4 bolt to re-center the conductor (multi strand, the smallest is 7 strand and 65 for the biggest wires) with thickness ranging from 0.24 to 0.60mm.

I'd tried to gather data on the min point insulation and plot to have the a better view of the process but the sad thing is that the minimum point is not always on the same spot (quadrant) because the wire twist during the process and besides when the bunched wire (multi strand wire) is not that good it will artificially shift the min. point to other location w/ no trend/pattern.


Oh, boy. This has to be fun! I bet another reason is that there is some natural vibration (for lack of a better word) of the wire as it is going through the extruder head, which will randomize the min location. It will be like trying to sit on the opposite side of a campfire to get away from the smoke, only to have it shift and blow over you anyway.

SPC works best when the process is in control. It can be handy when the process is not to help get the process into control. But, with the causes of variation you have, and the adjustment you wish to make not being the only variation to control the effect, you may not have a process that SPC will work for. If you could imagine an automated compensation device that could adjust based on scanning real time around the wire, by the time such device measured the part and sent a signal to the adjusting mechanism, the twist would move, and you would be adjusting to the wrong location. Now, slow that process down to an operator making the adjustment. Wow. Poor guy (ok...or gal.)

Fortunately, you are taking the time to understand your process, and possible controls - which is admirable! You may have to try to work with the process to gain more control before SPC will give you solid signals. I am just blue skying here, but something to better control the wire location prior to going through the head, for example. I am not sure what your current tool design is, so it is hard to tell, but you might ponder it conceptually.

More blue sky: What is the frequency of the twist? You might want to try to sample at a rate (3 to 5 per quadrant) that will include the lowest point in the twist in the data, plot the data in an Excel radar chart and adjust - or SPC I-MR chart - based on the lowest point. You might end up with a 12-20 point sampling plan per check. But, if it works, you may be able to determine a reasonable check frequency to temper such a test.

Whew! No easy answers here. :eek:

It sounds like you may need to move your SPC further upstream in the process. Or add SPC upstream first before you can successfully implement it here.

You need to determine the process factors that control the wire twist and bunch quality then determine whether SPC (or another method of control) is feasible for these upstream factors. Once these factors are under control, SPC may be viable in this location.

Your process may vary enough to determine this graphically rather than using more complex statistical tools such as DOE. I recommend learning how to use a Multi-vari Chart (http://en.wikipedia.org/wiki/Multi-vari_chart). ASQ also has a good article (http://www.asq.org/data/subscriptions/qp/1995/1095/qp1095zaciewski.pdf) on this that you may access if you are an ASQ member. Here is another reference (http://www.opensourcesixsigma.com/v/vspfiles/files/multi-vari%20analysis.pdf) fom Open Source Six Sigma.