1,850 PM Routines, 71% Compliance, and Three Failures That Had Just Been "PM'd"

Not as the sole first move. Compliance measures whether the work was done, not whether it should have been done, and this programme has a quality problem before it has a discipline problem.

The evidence that compliance is not the constraint:

• Three of five critical failures occurred within 30 days of a completed PM. If the PM cannot prevent the failure, doing it more reliably at 95% buys nothing and may cost money.
• 60% of routines have no acceptance criteria, so "inspect and lubricate" is not a task-it is an instruction to look. A completed work order against it tells you nothing.
• Vibration coverage is duplicated on 25 assets and split across two contractors, which is a scope problem, not a compliance problem.

Where the maintenance manager is right: compliance discipline does matter, and there is a legal, safety, and warranty subset for which 95%+ is non-negotiable regardless of analysis.

The position I would take: split the programme. Carve out statutory, safety-critical, and warranty-obligated routines today and drive their compliance to near 100% immediately, with a visible dashboard. Simultaneously start the review of the remaining routines, which is where the 40% reduction lives. Then, as reviewed routines are reissued with acceptance criteria, compliance becomes a meaningful number and the dashboard can be extended to the whole programme.

Locking the schedule before the content is fixed simply locks in waste and intrusive maintenance. Say that to the plant manager in exactly those terms.

Step 1: criticality ranking (weeks 1–4). Rank the 340 assets on consequence (safety, environment, production, cost) and likelihood. Expect roughly 20–30 assets to carry most of the risk. Do not start with FMEA on 340 assets; you will not finish.

Step 2: FMEA/RCM on the top tier only. For each significant failure mode ask the four questions: what is the function, what is the failure mode, what causes it, what happens when it fails? Then apply task selection:

• Is there a detectable potential failure with a definable P-F interval? If yes, an on-condition task is applicable.
• Is there a demonstrated age-reliability relationship with a life long enough to act on? If yes, scheduled restoration or discard is applicable.
• Is the consequence worth the cost of the task? If not, run to failure deliberately and say so.
• If no applicable task exists and the consequence is unacceptable, redesign or change operating practice-do not invent a PM.

Step 3: triage the remaining routines. Delete duplicates (the 25 dual-covered assets are the first win), delete routines with no failure mode behind them, and rewrite the 60% vague routines so each has a measurable acceptance criterion and a defined action on failure. Consolidate by interval to reduce mobilisation cost.

Step 4: the PM-then-failure cases. Treat these as modification candidates, not deletions, and interrogate them: Was the task technically applicable to the failure mode? Was it executed to standard, or was it signed off in five minutes? Did it find a defect and the defect was not actioned? Each of the three cases should produce a specific answer, and a task is only deleted if the answer is that the task could never have prevented the mode. Deleting a task because it "did not work" without that analysis is how you lose the next one.

Short term (0–3 months), use what already exists:

• Findings per route, normalised per asset per month. A route that finds nothing on 60 assets for six months is either perfectly healthy or, more likely, not being read.
• Alarm-to-action time: the days between a contractor alarm and a work order being raised. This is the most commonly broken link and it is measurable today.
• Condition-generated work as a share of corrective work: the percentage of corrective work orders that originated from a PdM finding rather than from an operator complaint or a failure.
• Unexpected failures on assets with full on-condition coverage. The target is zero; any occurrence is either a coverage gap or a detection failure, and both are actionable.

Medium term (3–9 months), fix the data at source:

• Implement a standard failure code set (ISO 14224 is the usual reference) as mandatory fields on corrective work orders, with problem and cause codes.
• The analyst's first job is to recode 12 months of free-text history into a consistent Pareto; it is imperfect but far better than nothing.
• Track PM effectiveness per routine: failures of the mode the routine was designed to prevent, per asset-year.

The pilot. Pick one system-20 to 30 pumps or the compressor train-review its tasks under the criticality process, and run for 90 days. Compare failures, condition-generated work, and PM hours before and after. A credible pilot result is what unlocks budget for the remaining 300 assets; trying to change everything at once with one reliability engineer guarantees that nothing is finished.

What I would not use as a success measure: PM compliance. It remains a compliance measure, not an effectiveness measure, and it rises when you delete work.