Bad actors and the jack-knife diagram
A “bad actor” is a repeat offender: an asset or a failure mode that consumes a disproportionate share of downtime, cost, or attention. Finding the list is not the same as fixing the item, but it is how you decide where the next RCA or RCM hour should go.
Pareto and Jackknife answer different questions
A Pareto chart ranks items by a stated measure, such as downtime hours or cost, and may add a cumulative percentage curve. Use it to see which items dominate the total. A reliability Jackknife diagram is a scatter plot, usually on logarithmic axes, of failure count (or frequency for a stated exposure) against mean downtime or mean time to restore service per event. It distinguishes frequent short stops from infrequent long recoveries. It is not a ranked bar chart with a cumulative curve.
For one common observation period, total downtime for an item is its event count multiplied by its mean downtime. Equal-downtime contours on log axes help identify combinations with similar total impact. Set acute (long-duration) and chronic (frequent) thresholds explicitly for the plant; there is no universal cutoff or guaranteed 80/20 split.
Worked example: one observation window, different actions
Illustrative Apex example, not client data. Three assets were observed for the same year, using one definition of unplanned downtime. For this exercise only, the team flags more than five events per year as frequent and mean restoration above two hours as long-duration.
| Asset | Events | Mean downtime (h/event) | Total (h) | Signal |
|---|---|---|---|---|
| A | 10 | 1 | 10 | Frequent, short stops |
| B | 2 | 12 | 24 | Infrequent, long recovery |
| C | 6 | 3 | 18 | Frequent and long-duration |
A downtime Pareto ranks B, C, A. A Jackknife plot places A at (10, 1), B at (2, 12) and C at (6, 3). For A, examine recurring triggers and task effectiveness; for B, examine diagnosis, access, spares and restoration delays as well as the failure mechanism; C merits both lines of inquiry. Those are investigation directions, not demonstrated root causes. Count zero cannot be plotted on a logarithmic axis; retain zero-event assets separately rather than silently discarding them.
Make the comparison fair
Record the observation window, exposure, asset boundary and downtime definition. Do not compare raw counts across unequal operating hours without considering exposure. Separate waiting time from hands-on repair where records support it. A short window can make a one-off event look chronic. Estimate downtime cost consistently before adding a cost ranking. Inspect the supplied toolkit before assuming it implements a particular plotting method.
Primary reference: ESO, Reliability culture at La Silla Paranal Observatory, section 3.3 (Jack Knife). The table and interpretation above are original illustrative material; no source figure is reproduced.
Related reading
- RCFA step by step
- Expected loss
- Bad Actor Toolkit