MTBF (Mean Time Between Failures) and MTTR (Mean Time to Repair) are the two reliability metrics that most maintenance departments report upward. They're also the two metrics that predictive maintenance most directly affects — but in different mechanisms, and over different time horizons. Understanding which mechanism each metric responds to helps maintenance directors set realistic expectations and measure the right things when evaluating a condition monitoring program.
This article walks through both metrics, the specific ways predictive maintenance data moves them, and what a realistic improvement picture looks like over a 6-month monitoring period on a fleet of about 50 rotating assets.
MTBF: The Failure Rate Equation
MTBF is the average time between successive failures on a given asset or asset class. It's calculated as total operating time divided by number of failures in that period. If a pump ran for 8,760 hours in a year and failed three times, its MTBF is 2,920 hours.
Predictive maintenance improves MTBF by reducing the number of failures that occur — specifically, unplanned failures caused by detectable degradation. The mechanism is: catch the degradation before it reaches the failure state, intervene with a planned repair, restart the asset in good health. If that degradation would otherwise have continued to failure, you've removed a failure event from the count. Over time, as more degradation events are caught and addressed before failure, the failure count per operating hour drops and MTBF rises.
Two important nuances: first, MTBF improvement from predictive maintenance requires that the failures being prevented are from detectable-degradation failure modes. Sudden failures from external causes (process upsets, foreign object ingestion, operator error) aren't prevented by vibration and temperature monitoring — they happen faster than the monitoring window can capture. A realistic estimate is that predictive monitoring addresses 60-75% of rotating equipment failure modes in typical industrial service; the remaining 25-40% are not preventable through condition monitoring alone.
Second, MTBF calculations are sensitive to how you count "failures." If your CMMS counts a planned maintenance work order triggered by a health score as a failure event, your MTBF calculation will look worse after implementing predictive monitoring even if the asset health program is working perfectly. The correct approach is to track failure types separately: unplanned failures (asset reached failure state) vs. planned interventions (asset removed from service before failure, based on condition data). Only unplanned failures count against MTBF; planned interventions don't.
MTTR: The Repair Duration Equation
MTTR is the average time to restore a failed asset to operating condition. It includes diagnosis time, parts sourcing time, repair execution time, and return-to-service testing. For unplanned failures — the kind that happen on a Friday night with no parts on the shelf — MTTR can be 24-72+ hours on complex rotating equipment. For planned interventions with pre-staged parts and scheduled technician time, MTTR is typically 2-8 hours for the same repair.
Predictive maintenance improves MTTR through two mechanisms:
Mechanism 1: Converting unplanned to planned. A bearing replacement triggered by a health score with 3 weeks of lead time has a very different MTTR than the same bearing replacement triggered by equipment failure. The planned event has parts on-site, a scheduled maintenance window, and a technician who knows exactly what they're replacing before they arrive. The unplanned event involves identifying the failure cause (which is itself time-consuming when you don't have sensor data leading up to the failure), sourcing emergency parts, and executing the repair under pressure. Converting even a fraction of unplanned events to planned interventions has a significant effect on average MTTR across the fleet.
Mechanism 2: Faster root-cause diagnosis when failures do occur. Not all failures are preventable, and some will reach failure state despite monitoring. But for assets with continuous sensor data, the diagnostic phase of MTTR is dramatically shorter. Instead of a technician spending 2-4 hours disassembling equipment to figure out what failed, the sensor data from the hours before failure tells them exactly which component degraded and how. A clear BPFO trend over three weeks ending in failure tells the technician it's an outer race bearing defect before they open the housing. That diagnosis-time reduction is real and compounds across all failures, planned or not.
What 6 Months Looks Like on a 50-Asset Fleet
Here's a plausible but illustrative picture based on what we'd expect from a growing industrial facility with 50 monitored rotating assets, transitioning from a primarily time-based PM program.
In the first two months, the monitoring baseline is established. Health scores are calibrated to each asset. You'll likely see several assets enter the watch band almost immediately — assets whose degradation was already in progress but invisible under time-based PM. These represent backlog failures that were coming regardless; catching them in months 1-2 and scheduling planned interventions is the first visible win.
By month 3-4, the PM schedule typically starts shifting. Assets that were getting quarterly rebuilds are now receiving them based on health score thresholds rather than calendar. Some assets don't hit the intervention threshold at 3 months. Some hit it at 5 weeks. The PM labor costs start to reflect this: fewer unnecessary rebuilds on healthy assets, but more total interventions because previously-invisible degradation is now being caught. At this stage, total maintenance events may be similar to or slightly higher than before monitoring; the difference is most of them are planned rather than unplanned.
By month 5-6, the unplanned failure count starts declining measurably. Assets that are being caught at health score 65-70 are being repaired before they reach the failure state. MTBF, measured correctly with unplanned failures only, starts to improve. MTTR on all failure events (planned + unplanned) drops because even the remaining unplanned events benefit from the sensor history. A 6-month review on a 50-asset fleet should show a reduction of 2-4 unplanned failure events compared to a comparable prior period — which, at a conservative $15,000-$40,000 per unplanned event including parts, labor, and production loss, represents a clear ROI on the monitoring program.
The CMMS Integration Is Critical for Measurement
None of the MTBF and MTTR improvement tracking described above is possible without clean CMMS data that distinguishes planned condition-based interventions from unplanned failures. This is why Fleetpio's direct CMMS integration matters beyond convenience: when health-score-triggered work orders are automatically written to the CMMS with the trigger type, fault classification, and sensor data summary, those events are categorized correctly from the start.
Manual data entry creates a classification problem. A technician who replaces a bearing because the health score said to might record it as "bearing failure" in the CMMS because that's the closest matching code, which counts against MTBF. An automated work order that captures "health-score-triggered intervention, bearing defect frequency BPFO elevated, fault type: bearing outer race degradation" is categorized correctly and doesn't inflate the failure count.
This seems like a data management detail, but it's the mechanism that makes the MTBF improvement visible to plant management and finance. If condition-monitoring interventions are misclassified as failures, the MTBF metric appears flat or worsening even as the actual reliability of the fleet is improving. The measurement infrastructure needs to match the maintenance strategy.
Setting Realistic Expectations
We want to be direct about what 6 months of condition monitoring does not produce on a 50-asset fleet. It doesn't eliminate all failures — sudden-onset failure modes, external-cause failures, and failure modes outside the sensor coverage will continue to occur. It doesn't immediately justify eliminating time-based PM intervals on all assets — some consumable components (elastomeric seals in aggressive service, grease-lubricated bearings in high-contamination environments) are best replaced on interval regardless of vibration condition. And it doesn't produce reliable MTBF improvement numbers until the baseline period is complete and the first full round of condition-triggered interventions has been executed — expecting measurable MTBF data at month 2 is unrealistic.
What 6 months does produce: a clear picture of which assets are degrading faster than expected (candidates for root cause investigation — is it a process issue, a mounting issue, or the expected wear rate for that service?), a baseline for distinguishing planned from unplanned events going forward, and the first cohort of averted failures that become the ROI case for the maintenance director's budget review. The improvement trajectory is real; the timeline requires patience through the calibration period before the signal becomes statistically significant.
MTBF and MTTR are lagging indicators — they measure the result of decisions made over the preceding months. Condition monitoring changes the decisions; the metrics follow, with a lag that roughly equals the P-F window length of the failure modes you're catching. For most rotating equipment in industrial service, that's a 3-6 month lag between deploying monitoring and seeing it in the numbers. That's the timeline to set with plant management from the beginning.