How to customize control limit in RunChart with alarm after 3 consecutive points beyond the target

[#1]

Hello experts,

I'm having problem to build a Run-chart for attribute (defect %) which will trigger a warning after 3 consecutive points above target defect %.

For example below are the images from Control chart builder.

I use P-chart to monitor every group defect %.

Based on JMP calculation, only Test 1 (Test beyond limit) trigger due to 1 point exceed Zone A

The calculated limit / zones A,B,C are very high against our target defect % of 5%.

Is there any way to customize the zone limit/test so that if i got 3 consecutive points >5% , it will give a warning?

Thank you

[#2 · Accepted solution]

Try right-clicking in the control chart plot and see if there is an option to customize tests or warnings.

See Help for more information.

I think that your case is not handled by the conventional tests, though. You will have to write a script that adds a reference line for the upper specification limit on the proportion of defects and applies your rule (3 subgroups beyond USL in a row) to catch such events.

[#3 · Accepted solution]

It appears to me that the customizing available might not get you where you want to go.  An alternative approach, which admittedly would require a good deal of coding.  But I do believe that by using 

     Add Graphics Script

would allow you an open ended methodology to calculating and modifying of the control chart.

This methodology is detailed in the Scripting Guide

[#4]

Thanks @Mark_Bailey @txnelson  for the suggestion,

For scripting, usually I'll try from clicking chart & save the script to tweak (since I'm a novice in scripting).

I'm able to change the limit by tuning the "Get Limit" and saved it in the script window.

I tweak the script to be something like below but it still triggering every single point out of new limit.

I change warning test 1 & test beyond limit from 1 to 3

gb = dt3 <<Control Chart Builder(
	Size( 720, 500 ),
	Show Control Panel( 0 ),
	Show Alarm Report( 1 ),
	Show Capability( 0 ),
	Class( Shewhart Attribute ),
	Variables( Subgroup( :End Lot Time ), Y( :TTV_WB ) ),
	Chart(
		Points( Statistic( "Proportion" ) ),
		Limits(
			Sigma( "Binomial" ),
			Zones( 1 ),
			Shade Zones( 1 ),
			Show Center Line( 0 )
		),
		Warnings( Test 1( 3 ), Test Beyond Limits( 3 ) )
	),
	Get Limits(
		"C:\Users\jmpuser2020c\GetLimitS3014.jmp"
	),
	SendToReport(
		Dispatch(
			{},
			"End Lot Time",
			ScaleBox,
			{in( Today()-in days(30) ), Max( Today() ), Interval( "Day" ),
			Inc( 1 ), Minor Ticks( 0 )}
		)
	)
);
gb << Bring Window to Front;
Wait (1);

[#5]

Thanks @txnelson , @Mark_Bailey , 

Just to update, I've customized the script which has to run through ~6 charts/table sequence before I got the result that I want.

[#6]

Please excuse my impulse to respond to your situation and feel free to completely ignore my post.  

I ask myself why would you want to do this?  Are you planning to react to this "signal"?  If so, you might want to read Deming's thoughts on Common/Special cause (see the funnel experiments and Deming, W. E. (1975) On probability as a basis for action, The American Statistician, 29(4), pp. 146–152).  Reacting to spec. limits as if this implies some signal regarding underlying causal structure is dangerous and can often increase variation in the process.  See Shewhart (Shewhart, W. A. (1931) Economic Control of Quality of Manufactured Product ISBN 0-87389-076-0)

A clarification as well, P-charts are charting % defective (not defects).  Perhaps P-charts that are "in-control" suggest the process producing the defectives is consistent (In other words it is consistently making bad product).  This, of course, is not ideal.  Why not try to understand what is causing the defectives?  Perhaps experimentation or directed sampling to understand the causal structure instead of developing a process to react to the defective rate?

[#7]

Hi @statman 

In my case, there is only 1 defect type and either got that 1 defect to be defective or do not have defect (good).

That's why we used p-chart.

I do agree with your input that the p-chart suggest the process producing the defectives is consistent (i.e consistently producing bad item). Anyway, there's a problem to solve.

Ideal case, we all want to achieve is 0 defect.

As of now, I'm working on control limit detection to find the pattern of existing data (exploring different type of rules which would benefit us).

I cannot investigate all 153 failed data which are too time-consuming.

Investigate the 1 failed point root cause make no different to overall defective %.

SPC/Run chart is reactive but I believe we can tighten the control limit over the time to do continuously improve the situation until it saturated at a better point rather than 5%. 

Experiment will be done by other staff, my side just need to propose a monitoring plan.