First I have to thank all the people here to answer question. It's a very nice and friendly place for me to ask any question I faced.

Back to the problem, currently our company wants to implement SPC to control the IPQC performance in the plastic part which produced by multi-cavities mold.

Our sampling plan is choosing 1 pc from each cavity every 8 hrs. Currently we have 25 batches data to calculate the history control limits for the control chart we want. But I have some questions to ask : (assuming we use 8 cavities molding)

Problem 1 :

Background :

We already know that different cavity have different performance. Some cavity produce parts with larger inspection results than other cavities. So we decide to use Xbar-S chart to calculate mean of all the 8pcs in a single inspection. We plan to use Xbar-S chart to detect mean shift in the production. The sigma here I use the formula comes from mintab. However, I found the control limit is very narrow that nearly every Xbar from different batch OOCL. I realize it may because the calculation use to estimate sigma only consider within batches variation instead of Between batch variation.

Question :

1. In this situation, Control Limit seems not very reasonable. Would it be better to use other control chart? (For example, 3-way control chart.)

2. Is using historical data to monitoring future performance reasonable? Or is it better to control only within batches variation by each batches (calculate control limit for each batches every time)?

Problem 2 :

Background :

We already know that QC sometimes may misuse their measuring equipment. We want to use I-MR chart to detect this variation every time QC do the inspection. I gather all 25 batches data to calculate the control limit by using sigma of I chart = MR-bar/1.128. 1.128 comes from the constant d2 with moving range every 2 data point. The calculation of MR use the all data points arrange by their inspection period and cavity. For example, 8 cavities molding, 2 batches with 2 inspection period in each batch. So totally 8*2*2 = 32 data points. I calculate the mean of these 31 MR as MR-bar in the formula. By implement this process, I calculating all the control limits by 25 batches we have. However the control limit seems too narrow that make a lot OOCL point when mapping back to the historical data.

Question :

1. Instead of monitor Xbar of 8 cavities, if I want to control the each individual to the real time detection for QC misuse. Is I-MR chart reasonable?

2. The same with problem 1, the batches variation seems not well consider in the sigma calculation. Any other way to have a better estimate that contain batches variation? The calculation I can come up with is the total standard deviation of all the data.

Between batches variation seems a big resource in our data. So SPC suffer us from implementing. In this situation, It's better to dig into it and make the process more stable, then use SPC to detect future out of control. However, I still wondering about is there any method that can include between batches variation in the calculation.

Thank you for reading the post in advanced. I am not very familiar with SPC control. So the post may make no sense. Hope I can learn something and experience with all of you.