Augos Factory / Learn / Variation and standard work

Learn · variation, standard work and Six Sigma

Variation is a process signal, not a verdict on people.

In short

Process variation is how much the same task's duration or result changes from one time to the next. When a standard task such as a changeover takes seven minutes one day and forty the next, something is not settled. It is usually the method, tooling or materials. Standard work fixes the best known method so it can be repeated, then improved.

3 kinds
within a task, between teams, between sites
7.6 h
of CIP overrun on one beverage line in one month
Measured
IQR
the spread measure behind every target
R4,950
per line per month to measure it, ex VAT

Reading variation

Where the variation is decides who acts

Where the variation isWhat it usually signalsWho acts
Within one task on one lineNo settled method, tooling or material for the taskEngineering: standard work, quicker changeovers
Between operators on the same taskThe best operator's method has not been passed onSupervisors: the best operator trains, trainees close the gap. Best operator ›
Between teams or shifts on the same taskThe best team's method has not spreadTraining: copy what the best team does
Between sites on the same taskOne site has solved what the others have notGroup operations: move the best practice across sites

A task that varies wildly usually lacks a method, a tool or a part, rarely effort. That is why each operator is compared with the line's own best task by task, and coaching starts with the method and the tools. The shop-floor charter ›

Measuring it

How do you measure process variation?

For task durations, the median and the quartiles are more useful than the average and the standard deviation. A few very long events skew the average.

typical = median duration of the events in the calendar year to date
recommended = median − median absolute deviation, from at least ten events
spread = interquartile range = 75th percentile − 25th percentile
outliers = events beyond 1.5 × the interquartile range

A wide spread with a fast recommended time is the clearest opportunity in a plant: the fast version already exists, it just isn't repeated.

An early Augos study

One task, three sites, a tenfold spread

At a three-site steel processor, the typical best coil change ranged from 6.6 to 11.2 minutes between sites. The spread at the least consistent site was 12.5 times that at the most consistent. The fastest site was not the steadiest, so each had something the others could copy. Coil set-up remained that client's largest addressable loss.

Coil changes by site: the spread

Illustrative dots

The same coil change at three sites of one steel processor, each centred on its own typical time. The band shows how widely each site's changes scatter, relative to the steadiest site.

12.5×the spread at the least steady site, against the steadiest. The fastest site was not the steadiest: speed and consistency each have a site to learn from.

An early Augos study · one anonymised steel processor, three sites, 2,934 coil changes. Band widths are each site's standard deviation relative to the steadiest site; the dots are illustrative. Site, products and people removed.

Standard work

Three elements of standard work

  • Takt time: the pace customer demand sets.
  • Work sequence: the order of steps the best method uses.
  • Standard work in process: the materials and parts that must be ready for the sequence to run.

Six Sigma

Six Sigma is the reduction of variation

Six Sigma projects follow DMAIC: define, measure, analyse, improve, control. The measure and control steps need the same thing as kaizen: the same task, measured the same way, every time it happens.

Kaizen with facts ›

Questions

Variation, answered

Isn't variation just people working at different speeds?

Rarely. When the same task varies widely, look first at method, tooling and materials. Effort is the least common cause.

Why the median rather than the average?

One very long stop can drag the average up for weeks. The median shows the typical event, and the quartiles show the spread.

How many events do you need?

At least ten for a recommended time. Both benchmarks use every event in the calendar year to date, outliers set aside.

Find your widest spread.

Thirty minutes with a manufacturing specialist: which tasks vary most on a line like yours, and the fastest version your team already does.

  • Every task timed
  • Typical, recommended and spread
  • Prices in rand, ex VAT

A manufacturing specialist replies within two business hours to set a 30-minute call.