Informational Control Chart Interpretation - General "Rules"

The fact that the data doesn't fit a normal curve doesn't mean that the process isn't stable. You might be dealing with a different distribution. In statistical analysis, "stable" and "in control" mean the same thing. They both mean that the process output follows a reasonably predictable path, and sources of special-cause variation have been identified and eliminated. A process can be statistically stable and produce nothing but nonconforming output. A process that's stable and the mean of which is centered between the specification limits is said to be "capable."

The fact that your overall distribution is non-normal can mean a number of things:

• The process MAY indeed be unstable
• You may be combining multiple process streams. The mixture of many stable normally distributed process stream with differing averages will appear non-normal. This can arise if you combine the output of multiple cavities, machines, etc.
• The process may inherently be non-normal, yet stable.

'stability' can only be determined by a 'control' chart, not a histogram.

Miner said:

The fact that your overall distribution is non-normal can mean a number of things:

• ...The process may inherently be non-normal, yet stable.

Click to expand...

Indeed, temperature control and a stable U-shaped distribution.

Any suggestion on which control chart should i use for monitoring reliability/mttf of a device?

Romvill2002 said:

Any suggestion on which control chart should i use for monitoring reliability/mttf of a device?

Click to expand...

Can you provide additional detail and context on the device? Is this under development, fielded? Will you be using reliability testing or field data?

Your use of MTTF would normally indicate that these are non-repairable devices, but MTBF/MTTF is so misused, that I will not make that assumption.

Medical device/equipment, fielded data used.

An SPC chart is probably not the best approach to this. There are two approaches that I would recommend.

The first is a Broom chart developed by Larry George.
the second is a Lloyd diagram developed by Robert D. Lloyd.

I will try to locate some references to these tomorrow.

Miner will provide some excellent guidance...

I am in the diagnostic medical device / instrument field and have developed many control charts on our field data. In general, the I-MR, p, c, p' charts work well for complaint, failure and service events. (the inverse of MTBF/MTTF, etc.)

One of the more overlooked issues with these control charts is the base (denominator) and the subgroup.

Subgroup: do you use the calendar time period (month or week), elapsed time since manufacture/shipment or the lot / time or serial number grouping of manufacture?

Base value: do you use install base, lot quantity or actual usage rate within the subgroup?


Do you use actual failure rates or reported failure rates? Or are these complaint or return/service rates? Many people say they are calculating MTBF/MTTF when they are really plotting only those events that are reported by their Customers...

Miner said:

An SPC chart is probably not the best approach to this. There are two approaches that I would recommend.

The first is a Broom chart developed by Larry George.
the second is a Lloyd diagram developed by Robert D. Lloyd.

I will try to locate some references to these tomorrow.

Click to expand...

I was unable to locate more information on the broom diagrams. Larry George's old website is down, and his new one is not fully up and running. I recommend contacting him directly. He is usually quick to respond.

There is a good overview of Lloyd diagrams here: Field Failure Analysis using Root Cause Pattern Diagrams.