Manufacturing, process, operations
Industry and quality
Control charts and capability indices for people who have to answer to a customer about a process, not to an examiner about a thesis.
Two questions get confused constantly on a factory floor: is the process stable, and is it good enough. A process can be perfectly stable and consistently out of specification; it can also be capable on average while drifting towards a limit. ZStat answers them separately and refuses to answer the second before the first.
What you can do
X-bar and R for subgrouped measurements, individuals and moving range when they come one at a time, with the Western Electric rules applied so a sustained small shift is caught rather than only a dramatic one.
Cp, Cpk, Pp and Ppk with expected defects per million - and a warning when the spread would fit but the process is aimed off centre, which is usually far cheaper to fix than variation.
ANOVA, factorial designs and non-parametric equivalents for deciding whether an observed difference between lines or operators is real.
Survival analysis for time to failure, and forecasting for demand and throughput.
Before and after comparisons with effect sizes, and non-inferiority testing for when a cheaper input has to be shown to be no worse.
What it refuses
- Capability indices on a process that is not in statistical control. The index would describe a distribution the process does not have.
- Control limits drawn from the specification instead of from the process. That chart cannot show drift inside the tolerance, which is the whole point of drawing it.
- Declaring a process unstable on the sensitive zone rules alone. On forty points, one stable process in three trips one of them, and a chart that cries wolf gets ignored.