Provide a list of NELSON rule violations in a historian control chart set

[#1]

I need to extract the batches with nelson rule violations for any parameter that the event happens, is there any way we can generate the list directly?

[#2 · Accepted solution]

Hi @Zara,

There might be several options to do this, but here is a very simple one :

Once your control chart is build and the Nelson tests checked in the Control chart builder options, you can click on the red triangle and select "Save Summaries". It will create a table with the same observations than you had previously in your original datatable, but you have some supplementary columns like "Test Failures" and the settings of your control chart. You can then sort easily based on this column and filter out rows with no test failures.

This solution could probably be automatized with JSL scripting.

You can have a look at the example provided in the datatable Process Measurements, with the scripts "Control Chart Builder" and "Process Measurements - Process 4 Summaries" for the table summaries with the "Test Failures" column.

I hope it will help you,

[#3]

Thanks Victor for your help and for providing the example file. This shows a new summary table for each parameter and when we have too many parameters(over 100) it will be very manual to filter out every single file and put them together in a summary.

I need to make a final report with a list of all failures of all batches, any ideas that can help, I was thinking the list can be generated somehow when the failures are already in the results.  

[#4]

One way to list all the failures of all the batches would be to use the Process Screening Platform, turning on all alarms (red triangle, choose tests, enable all), then outputting a details table (red triangle, save details table). The script below uses some sample data and outputs this table that includes an Alarm column, which includes a list of any/all Nelson Rules any row violates.

Names Default To Here (1);
dt = open ("$SAMPLE_DATA\Consumer Prices.jmp");
rpt = dt << Process Screening(
	Process Variables( :Price ),
	Grouping( :Series ),
	Control Chart Type( "Indiv and MR" ),
	Time( :Date ),
	Subgroup Sample Size( 3 ),
	Test 2( 1 ),
	Test 3( 1 ),
	Test 4( 1 ),
	Test 5( 1 ),
	Test 6( 1 ),
	Test 7( 1 ),
	Test 8( 1 ),
	Range Limit Exceeded( 1 )
);
dt_final_report = rpt << Save Details Table();

[#5]

Hi @Zara,

Another option would be to use the script from "Alarm Script", that seems to do what you expect : list samples, rows and alarm tests. Some related informations can be found here : Solved: Control charts Alarm Script: identify rows into data table - JMP User Community

For several parameters to check, the platform Process Screening will be more efficient to use.

[#6]

I cannot be helpful in your request, so please ignore this post if it is not useful.  Alas, I don't actually understand why you would want this?  Batches that are correlated to specific data points that are evidence of assignable cause variation (Nelson or Western Electric rules) do not indicate they are failures.  That is what specifications are intended for and we certainly realize specifications have little to do with actual process variation (since they are derived independently).  

Control limits DO NOT equal specification limits.

The original intent of control chart method was to compare sources of variation and provide assistance in understanding causal structure (Of course for an appropriate comparison, first the basis for comparison, the within subgroup variation, should be tested for consistency, hence the range chart).  They work best when they are actively used.  For example, when a range chart displays an "out-of-control" condition, this suggests one or more of the x's in the process is assignably different at this point in time.  Shewhart suggests It would be worth while to investigate at this point in time, while the process is running, not a post mortem.  I will use an analogy.  Imagine you are a criminal detective.  When the range chart shows an OOC condition, this is when the crime is happening.  It would be a rather easy crime to solve if you were actually present when the crime was committed (vs. having to look for clues after the crime was committed and subsequently testing hypotheses).  

So for your example, you have already missed the opportunity to get a real-time look at the effects of x's in the process while they are exposing themselves, so to speak.  The batches that correlate with those OOC conditions may, in fact, be the best batches you ever made.

BTW, that does not mean you can't plot the data in some other fashion.

[#7 · Accepted solution]

Thanks for your comments, I agree with all what you are telling me and we actually using the control charts with real-time approach and I am aware that Control limits are not spec limits... 

So my request does not have anything to do with CPV or even further analytics... It is just because we need to look back in history and list all the for this case "Nelson Rules Violation" for many batches and lots as part of documentation, and thats it. I know that it is different with routine control charts and also continuous processing.