Case 2 · Maintenance & asset reliability¶
Data used on this page
AI4I 2020 Predictive Maintenance Dataset - real public data · Maintenance work orders - synthetic. Details on the data sources page. With synthetic data the findings illustrate the method; they are not claims about a real organisation.
Context¶
A plant runs 10 machines in two shifts. The process data has one record per 4-hour production run with the machine's temperatures, speed, torque, tool wear and failure flags (10,000 runs in 2024). Every failure opens a corrective work order; each machine also has a weekly preventive maintenance (PM) routine.
Question¶
Why is availability below target - is it a reliability problem (we fail too often) or a repair problem (we take too long to fix)? And what can we plan instead of react to?
KPIs · 2024¶
Availability is the outcome operations feels. It decomposes into MTBF (how often we fail) and MTTR (how long we are down per failure): availability ≈ MTBF / (MTBF + MTTR). PM compliance and planned share are the leading side: they move before failures do.
Finding¶
Show the data behind this chart
| asset_id | availability | mtbf | mttr | pm_compliance | failures | downtime_h |
|---|---|---|---|---|---|---|
| M01 | 0.953 | 173.913 | 8.560 | 0.941 | 23 | 196.885 |
| M02 | 0.975 | 137.931 | 3.659 | 0.941 | 29 | 102.455 |
| M03 | 0.982 | 181.818 | 3.385 | 0.980 | 22 | 74.459 |
| M04 | 0.958 | 148.148 | 6.547 | 0.922 | 27 | 176.764 |
| M05 | 0.847 | 29.851 | 5.379 | 0.961 | 134 | 720.836 |
| M06 | 0.976 | 200.000 | 4.991 | 1.000 | 20 | 99.823 |
| M07 | 0.983 | 173.913 | 2.973 | 0.580 | 23 | 68.387 |
| M08 | 0.971 | 181.818 | 5.437 | 0.549 | 22 | 119.620 |
| M09 | 0.978 | 235.294 | 5.382 | 0.902 | 17 | 91.496 |
| M10 | 0.964 | 181.818 | 6.801 | 0.882 | 22 | 149.618 |
Note: this bar chart does not start at zero on purpose - all machines are between 75% and 100% and the decision is about the gap to target; the axis start is stated in the subtitle.
Reading it together. Low availability has two different causes: M05 fails far more often. Across M05 the dominant failure mode is Heat dissipation (78% of their failures). Fleet-wide, tool wear is the clearest controllable driver: failure risk jumps after 200 minutes. MTTR looks worse than the typical repair because of a few long waits for spare parts. PM lateness is concentrated in two crews (M08, M07). In the plan B data, PM lateness does not cause failures (an assumption of the generator); the page does not claim that link.
Decision¶
- Condition-based tool change at 190 min of wear. In 2024 it would have turned 116 unplanned tool-wear / overstrain failures (about 215 h of downtime at the median repair time) into planned tool changes - and raised the planned-work share.
- Root-cause heat dissipation failures on M05 before anything else: a machine-specific reliability problem that more technicians would not fix.
- Stock the critical spare parts behind the 21 long repairs; the target is MTTR closer to the 2.4 h median.
- Re-plan the PM routine of the M08, M07 crews (night shift) - same weekly PM, scheduled inside their shift.
How we keep measuring¶
| KPI | Owner | Cadence | Target | Trigger for action |
|---|---|---|---|---|
| Preventive maintenance compliance | Maintenance Planner | Weekly, per crew (leading) | ≥ 90.0% | Below 80% for a crew - re-plan the week |
| Planned maintenance share | Maintenance Manager | Monthly | ≥ 80.0% | Below 65% - review reactive work orders |
| Mean time between failures (MTBF) | Reliability Engineer | Monthly per machine, 3-month rolling | ≥ 150.0 h | Machine in the bottom 2 for 2 months |
| Mean time to repair (MTTR) | Maintenance Supervisor | Monthly, with median and longest repairs | ≤ 4.0 h | Any repair > 12 h - check spare parts |
| Availability | Maintenance Manager | Monthly per machine | ≥ 97.0% | Below 95% - open a reliability review |
Tool wear at failure is tracked as a diagnostic: after the policy change, no failure should happen above 190 min.