# SPC

Six Sigma – iSixSigma › Forums › Old Forums › General › SPC

- This topic has 13 replies, 7 voices, and was last updated 12 years, 10 months ago by vijaya.

- AuthorPosts
- April 17, 2007 at 8:06 am #46751
Hi,

I would like to know ,how to fix the optimal control limit for employing spc in software?

thanks,

Vijaya0April 17, 2007 at 9:27 am #154869

accringtonParticipant@accrington**Include @accrington in your post and this person will**

be notified via email.Have you tried switching it off and switching it on again?

0April 17, 2007 at 11:21 am #154870Hi Vijaya,

Please explain the process, for which you want to find out the control limits, for better understanding. What do you mean by optimal control limit? I understand product specification limits are set by designer and control limits are set by the process employed to achieve the design specifications.

– Samir.0April 18, 2007 at 4:06 am #154913Samir,

Actually i am expecting the answer for the software process.Two control limits are set for any control charts which are used to find the stability of the system which is developed.Those limits are called UCL and LCL if the points fall within this limit then we say that the process is stable.I would like to set the optimum limits.so that i can avoid the high threat.How to set the optimum limits?

Vijaya

0April 18, 2007 at 6:36 am #154914UCL and LCL are calculated from the data that you have calculated.

USL and LSL will be your design specifications.

What do you mean by optimal control limits ? if its UCL and LCL then its calculated +- 3sigma distance from the mean.

regards

joe0April 18, 2007 at 7:34 am #154915Hi Vijaya,

it is a nice idea to try SPC for a software process! The control limits will come from the process you are charting – the trick is to figure out the precise way to calculate them. It will be always +/- 3 stddev from the mean, but the formula for the stddev will depend on the type of distribution you are using to model the process.For SW I suppose you should definitely NOT use a simple normal distribution unless your subgroups are quite large – about 10 to 30 I’d say. I would try to look at a binomial distribution – if you’re interested write me an e-mail at rogojel_at_index.hu and we can discuss the details.Regards

Sandor0April 18, 2007 at 7:46 am #154916

Allthingsidiot OParticipant@Allthingsidiot-O**Include @Allthingsidiot-O in your post and this person will**

be notified via email.USL & LSL should not be shown in the Control Chart

0April 18, 2007 at 9:56 am #154921Hi sandor,

I am considering normal distribution now yoyr are suggesting not to use Normal distribution what else you suggest? and how to calculate the shifts due to assignable cause?Kindly guide me in this regard

Vijaya

0April 18, 2007 at 12:25 pm #154928I am not the person from IT industry and hence I am not able to understand what you mean by SOFTWARE process. My understanding of process approach is “INPUTS converted to saleable OUTPUT through some transformation mechanism”. Here TRANSFORMATION MECHANISM is process.

By SOFTWARE process, do you mean

1. writing the software for SPC applications

2. writing the general software

3. finding the bugs in the program

4. do you mean any other process

– Samir0April 18, 2007 at 12:49 pm #154930He said not to use normal distribution UNLESS your sub-group sizes are large. So If you have a large sub-group size, you can go ahead with the normal distribution.

Your control limits will be automatically set according to your process behaviour. +/- 3 sigma (3 times standard deviation) would determine your process control limits.

Once any of your data points go beyond the control limits (either UCL or LCL), you have an assignable cause. To determine exactly what caused this unusual variation, you would need to analyse the data and do a root cause analysis.

0April 18, 2007 at 1:37 pm #154932“how to calculate the shifts due to assignable cause?”

Are you asking about Mean shift due to assignable cause? Mean shifts happen over time, and are a result of a variation that becomes inherent

to the process. Therefore mean shifts are caused by Inherent causes (common causes) which become a constant in the process itself (unlike

special causes which are one-off kinds).

You don’t calculate shifts due to assignable or inherent causes. You gague shifts by analysing your control chart and the behaviour of data

points. When your data points indicate a mean shift, calculate the new mean and you have your mean shift. You could use various means for that; if you use a good SPC software, it should automatically show you the new mean and the mean shift.0April 18, 2007 at 3:10 pm #154936Hi Vijaya,

it will depend a lot on your sampling. For instance, are you plotting individual values or do you have some kind of subgrouping? If you have subgroups, do you have a constant number in a subgroup – like every five measurement points, or a variable number – like every point that occurred this week?Do you use Minitab BTW?

Regards

Sandor0April 18, 2007 at 3:17 pm #154937Hi Vijaya,

it will depend a lot on your sampling. For instance, are you plotting individual values or do you have some kind of subgrouping? If you have subgroups, do you have a constant number in a subgroup – like every five measurement points, or a variable number – like every point that occurred this week?Do you use Minitab BTW?

Regards

Sandor0April 19, 2007 at 5:50 am #154960Thank you! rpaul for your information and if i relate cost with the control then how to calculate the false alarm cost?i am doing my PhD in the area how to use the SPC for software process and i am searching problem to do in this area if any body has the problem kindly post to me .

Vijaya

0 - AuthorPosts

The forum ‘General’ is closed to new topics and replies.