- \n
- Two machines can lose the same minutes for opposite reasons.<\/li>\n
- A Pareto by minutes hides frequent 30-second stops.<\/li>\n
- Count, duration and restart losses each tell part of the story.<\/li>\n
- Frequent stops need robustness, long ones need faster recovery.<\/li>\n<\/ul>\n<\/div>\n
Same downtime, very different machines<\/h2>\n
Ray Harkins, who teaches through The Manufacturing Academy and is general manager of Lexington Technologies, makes this point in “Downtime Frequency and Duration”, chapter 6 of his book Measuring Manufacturing Effectiveness<\/a> (2026), also available as an article on Accendo Reliability<\/a>. Comparing equipment with the same total downtime, one interrupted often and briefly, the other rarely but for long periods, he writes: “Total downtime does not distinguish between these realities. Downtime behavior does.”<\/p><\/blockquote>\n
His chapter frames the two dimensions in reliability terms: how often interruptions occur relates to MTBF, how long recovery takes relates to MTTR, and availability on its own does not tell you which pattern you are facing. He also observes that frequent short interruptions tend to become normal for operators and, without automated data collection, are recorded inconsistently. And he argues that the two patterns call for different improvement work: reliability and robustness for frequent interruptions, maintainability and recovery capability for long ones. This article takes that framework to the tool most teams use to decide what to fix first: the OEE Pareto.<\/p>\n
Why a Pareto sorted by minutes buries 30-second stops<\/h2>\n
A downtime Pareto is usually built on total minutes lost per reason. That is a sensible default, because minutes convert directly into lost output. But it has three blind spots when stops are short and frequent.<\/p>\n
- \n
- The logging threshold.<\/strong> If stops are written down by hand, or only counted above a minimum duration, the 30-second stop is never recorded. It does not appear low in the Pareto; it does not appear at all. Its time ends up inside the performance loss, as an unexplained gap between the theoretical and the actual rate.<\/li>\n
- Dilution across reasons.<\/strong> Even when short stops are captured, they are often grouped under vague labels (jam, adjustment, sensor) or spread over many reasons. Each bar stays short, while a single breakdown with a clear label and a long duration takes the top spot.<\/li>\n
- The cost of a stop is not its duration.<\/strong> The recorded 30 seconds is the time the machine stood still. It does not include what follows: slower cycles after the restart, the first parts to check, the operator pulled away from another task to clear the jam. Harkins describes the same mechanism when he explains how frequent interruptions fragment production and create repeated ramp-up losses.<\/li>\n<\/ul>\n
The result is a Pareto that is accurate about long stops and silent about short ones. Improvement effort then goes to the breakdowns everyone can see, while the line keeps losing capacity in increments too small to notice one at a time.<\/p>\n
![\"Four-step](\"https:\/\/teeptrak.com\/wp-content\/uploads\/2026\/11\/teeptrak-harkins-en.png\")
<\/figure>\nAn illustrative example: one week, three Paretos<\/h2>\n
Take, for example, a packaging machine over one week of production (the figures below are invented for illustration). The stop data contains four reasons. Sorted three different ways, the same data tells three different stories.<\/p>\n
- Dilution across reasons.<\/strong> Even when short stops are captured, they are often grouped under vague labels (jam, adjustment, sensor) or spread over many reasons. Each bar stays short, while a single breakdown with a clear label and a long duration takes the top spot.<\/li>\n
- The logging threshold.<\/strong> If stops are written down by hand, or only counted above a minimum duration, the 30-second stop is never recorded. It does not appear low in the Pareto; it does not appear at all. Its time ends up inside the performance loss, as an unexplained gap between the theoretical and the actual rate.<\/li>\n