I want to know which process follows normal distribution. Some articles mention " Targeted process follows Normal distribution" What does it mean?

What it means depends upon the context, but essentially a 'Targeted Process' is any process you are focusing on.

Any process may, or may not, output a normal distribution (http://Elsmar.com/Forums/search.php?do=process&titleonly=1&query=normal+distribution) (normal distribution (http://elsmar.com/wiki/index.php/Normal_Distribution)) with regard to any specific dimension or other measurement criteria for that process parameter(s) {or part(s) of}, or the product {or part(s) of} at that stage of the entire process as a whole - That is what you have to determine by evaluating the raw data before further processing and evaluating the data (statistical analysis (http://elsmar.com/wiki/index.php/Statistical_Analysis)).

In your context ("...some articles mention...") it probably means there is an assumption that the process is producing data with a normal distribution.

There are several other types of data distributions, including the bi-modal distribution. How data is analyzed depends upon the type of distribution of the data.

Agree with Marc, the data will follow it's path (or distribution in this case).:D

But your quote looks simpler
Some articles mention " Targeted process follows Normal distribution" What does it mean?

Means that ; "IF AND ONLY IF" the data could have a normal distribution and you don't have a target that data (a clear objective) can show even a uniform distribution or as multimodal distribution. One example could be that someone tell you to tear 100 paper sheets, you could tear wherever you like, so it hardly could show a normal distribution (the lenght of the resulting pieces), but if he tells you to tear the 100 paper sheets on the middle (a target ), it certanly will show a normal distribution.

I want to know which process follows normal distribution.

You could use chi-square test, but: what are you going to do if no normal distribution is detected?, SPC works even for non normal distributions.

This question is not clearly enough. But I guess it may mearn to asume the studied process as a normal distribution. When on the assumption that a process is normal distribution, the process will be easy to be calculated for capibility.

This question is not clearly enough. But I guess it may mearn to asume the studied process as a normal distribution. When on the assumption that a process is normal distribution, the process will be easy to be calculated for process capability.

You could use chi-square test, but: what are you going to do if no normal distribution is detected?, SPC works even for non normal distributions.

Yes, SPC works for non normal distributions, but X bar-R does not. It is the worst possible tool to attempt to control a uniform distribution, especially in precision machining. It will drive overcontrol by design. One good reason to verify the distribution prior to rubber stamping a chart to a process.:cool:

For the uniform distribution, it is also easy to calculate process capability: (USL-LSL)/(UCL-LCL)

Yes, SPC works for non normal distributions, but X bar-R does not. ...
For the uniform distribution, it is also easy to calculate process capability: (USL-LSL)/(UCL-LCL)

Elaborate please: If SPC works and X bar-R doesn't, how do you calculate UCL and LCL? (also included in your capability index), percentile ?:confused:

Elaborate please: If SPC works and X bar-R doesn't, how do you calculate UCL and LCL? (also included in your capability index), percentile ?:confused:

The details are laid out in the thread: SPC for Diameters - Parts not Perfectly round so charts are useless (http://elsmar.com/Forums/showthread.php?t=20935)

Just hate to rehash the whole thread here. :cool:

It is the worst possible tool to attempt to control a uniform distribution, especially in precision machining.
While precision machining is a specialised application and uniform distributions are uncommon, I'm most curious about this and your web site's statement:

"The x data is a statistically insignificant amount of data for a diameter or a length, and the average of a sample set of insignificant data is of even less value. The range ends up being a value of the measurement error. As a result, using the average of insignificant data, and its related measurement error has no value in decision making for precision machining. Also, the calculations for the control limits are for the wrong distribution. ... revolutionary new SPC charting methodology"

Could you elaborate on this please and perhaps provide a graph explaining the problem you perceive and your suggested approach.

While precision machining is a specialised application and uniform distributions are uncommon, I'm most curious about this and your web site's statement:

"The x data is a statistically insignificant amount of data for a diameter or a length, and the average of a sample set of insignificant data is of even less value. The range ends up being a value of the measurement error. As a result, using the average of insignificant data, and its related measurement error has no value in decision making for precision machining. Also, the calculations for the control limits are for the wrong distribution. ... revolutionary new SPC charting methodology"

Could you elaborate on this please and perhaps provide a graph explaining the problem you perceive and your suggested approach.

Most of the details are laid out previously in:
SPC for Diameters - Parts not Perfectly round so charts are useless (http://elsmar.com/Forums/showthread.php?t=20935)

But consider this simple notion for a round part: How many diameters are in a circle? An infinite number. When people are typically measuring a part for charting they measure one diameter on each of, lets say, a five part sample. They calculate the average and plot it. So, they take one out of an infinite number of diameters (an insignificant sample), and take the average. The average of an insignificant sample is...well, insignificant. The error comes from not capturing the roundness variation. Now, precision machining is the result of eliminating all statistically significant variation, except tool wear. No chance of eliminating that. As the tool wears on an OD (for example) the part gets larger and larger. As it hits an upper limit, it is adjusted down to a lower limit and allowed to continue to grow. The resulting curve is not random variation around a mean, but a sawtooth curve with a uniform (rectangular) distribution. Using normal distribution statistics to control it (as in calculating control limits using normal statistics) does nothing but encourage overcontrol - which is a waste of time and effort, as well as introducing more variation than the process requires. That's the thumbnail...