Hello everyone, all the textbooks that I have come across show examples of equal distance of upper and lower control limits from the process average for Xbar charts. Does this apply to the 1-sided processes like wire bond pull strength (higher is better)? Would it be okay to set 3-sigma for lower control limit and 6-sigma for upper control limits or maybe just go without the upper limit? Putting in the upper control limit would only cause unnecessary false alarms.

Hello everyone, all the textbooks that I have come across show examples of equal distance of upper and lower control limits from the process average for Xbar charts. Does this apply to the 1-sided processes like wire bond pull strength (higher is better)? Would it be okay to set 3-sigma for lower control limit and 6-sigma for upper control limits or maybe just go without the upper limit? Putting in the upper control limit would only cause unnecessary false alarms.

I think you are right to just use the one spec limit you have. Sounds like you have a MIN spec, so just calculate how far your process average is from your spec. Forget the upper spec, it has no physical meaning in the real world.

Our software does capability on unilateral specs wrong. I won't name the vendor, cause to their credit they are fixing it.

Our software said we had good Cp but poor Cpk. We were averaging almost zero on a MAX spec. Sure enough, the software was taking Cpk to be the lesser of 0 and MAX, so we got punished for being too good.

Plot the chart and look at if to see if the number the software is giving makes sense.

Good luck, it is a real bear to try to prove capability on tensiles in the foundry industry. I suspect wire has more consistency.

Most SPC software have the ability to enable or disable the rules, enable the rules that mean something to you, there is no "one size fit all" on rules.

I doud that is there a way to set 6 sigma for upper and 3 for lower rules in your software and IMHO has no meaning for the upper (if the bigger the better).

Also SPC work fine with almost any kind of distributions but seems to me that most unilateral control make a skewed distribution, is such is the case you can use a transformation to make look like a normal distribution (so the rules could have more meanning) or use non normal SPC (using persentile for SPC limits), or take my first advice "use what you need".

Keep in mind that control charts are used to detect changes in processes. Because of this, the upper limit really isn't meaningless. By ignoring the the upper control limit, you may miss "special cause variation" that could show improvement. Out-of-control points on the high side should be investigated to see what went right. Same thing with trends toward the high side. If the process has shifted, you may need to recalculate the limits if your software doesn't do it automatically.

On more thing to consider - if you really have a one-sided specification (one with a physical limit), then you probably have a skewed distribution. The central limit theorem allows control charts to work with skewed distributions, but they are more valid with larger sample sizes. Try plotting the distribution of sample averages. If the distribution is not normal, then the sample size is too small.

Agreed that you shouldn't 'mess with' the limits. The key is in how you structure the response flow to any out-of control condition....teh response to out of control high may not be to shut the process down until you figure it out.

Also agree on the sample size issue: do check the 'rough' distribution of your sample averages. if you have a large skew to the high side your control limit on the lower side will too large and you can potentially miss true process degradation signals...