Key takeaways
See our roundup of predictive maintenance software built around these metrics.
Walk into a typical maintenance team's quarterly review and the MTBF and MTTR numbers on the wall have three properties: they are plant-wide aggregates, they were set 18+ months ago by someone no longer in the role, and they bear little relationship to the actual cost of failure on any specific asset. The targets are checked, the numbers are debated, and nothing useful happens.
This is not a malicious pattern. It is what happens when no one has a defensible method for setting the targets in the first place.
In the absence of a method, the team falls back on the only inputs they have: vendor benchmarks ("world-class is 1,000 hours"), round-number improvements ("we should aim for a 10% improvement over last year"), and budget-cycle negotiation ("what number can we commit to without getting in trouble"). All three are wrong.
Every asset is built around manufacturer-stated wear limits, bearings rated for an L10 life of around 20,000 hours, valves for roughly a million actuations, drive belts for a defined service interval. For a single dominant wear item, that rated life is the practical ceiling on how long the part should run between replacements.
For the asset as a whole, MTBF, how often it fails, is a different quantity from its service life, and it normally sits well below that life, because an asset is failed and repaired many times before it is ever retired.
The target is to push MTBF as high as the limiting components allow given the actual duty cycle, not to expect it to equal a single design-life figure.
The honest starting point: pull the design spec for the limiting components, adjust for actual duty cycle (hours per day, shifts per week, load profile), and that gives the wear-limited ceiling to aim toward. The article on the preventive maintenance schedule covers how PM cadence interacts with this ceiling.
MTTR is bounded by what the maintenance team can physically achieve given their tools, parts, and access.
Diagnose 10 minutes, source the part 30 minutes, walk to the asset 5 minutes, perform the swap 25 minutes, test 10 minutes, document 10 minutes, that is 90 minutes of actual MTTR for a routine task, even with a competent team.
Setting a target of 30 minutes for the same task without changing tools, parts location, or access is a target that will never be met.
The defensible MTTR target is the floor of "what could be achieved with the asset's current process," which is usually 60-80% of the current MTTR. Going below that requires changing the process, adding a tool drop near the asset, pre-staging a spare, redesigning the access panel. The article on root cause analysis covers the process-change side of MTTR.
The third input is the cost the plant pays each time the asset fails. Not in dollars, in production minutes lost, which is the currency that ties to OEE. An asset whose failure costs 4 minutes is in a different conversation than one whose failure costs 90 minutes. The first deserves a relaxed MTBF target; the second deserves an aggressive one.
The math is straightforward. If the asset fails 10 times a year and each failure costs 90 minutes, the asset is consuming 15 hours of production capacity per year.
If reducing the failure rate by half is achievable through PM frequency or parts upgrade, the target is "MTBF up by 100%, no MTTR change." If reducing the MTTR by half is achievable through process change, the target is "MTBF flat, MTTR down 50%." Often both are pursuable but in different timeframes.
The framing connects directly to the manufacturing KPIs family that drives OEE.
The single biggest mistake in setting targets is rolling them up to plant level. Plant-wide MTBF averages across assets whose individual MTBF varies by two orders of magnitude, a critical packaging head with 800-hour MTBF in the same bucket as an auxiliary fan with 30,000-hour MTBF. The aggregate moves with whichever class dominates by mass, and the actual interesting story, which assets are degrading, which are improving, disappears.
A working structure: targets are set per asset class (packaging heads, drive motors, fillers, etc.) and tracked separately. The plant-wide number is a roll-up for management awareness; it is not the operational metric. Asset classes with the worst gap between target and actual become the focus of the next quarter's work. Without this granularity, the target-setting exercise produces a number that means nothing.
Targets that are set once and never reviewed become fiction within two years. The mechanism is simple: the underlying conditions change. New equipment arrives, the duty cycle shifts, the maintenance team rotates, a key spare-part supplier discontinues a line. The conditions that justified the original target no longer apply, but the target is still on the wall.
The review cadence is quarterly. Each quarter, for each asset class:
This is also the mechanism that catches drift before it becomes a year-end surprise. The piece on work order management systems covers the data structure that makes quarterly review possible without a manual data pull.
A mid-market plant with 50 critical asset classes, properly set up, has 50 MTBF targets and 50 MTTR targets, each justified by the three inputs and reviewed quarterly. The targets vary widely: critical packaging heads at 600 hours MTBF, drive motors at 8,000, conveyor belts at 4,000.
Each target has a documented rationale (design spec + duty cycle + cost-of-failure analysis). Each is owned by a named engineer who answers for the variance.
That structure costs a one-time setup of about a week per asset class, plus an hour a quarter per class to review. The return is that the target-setting exercise stops being theatre and starts driving maintenance investment.
The method works in any CMMS that tracks failures and repair times against assets. Where a unified OEE + CMMS platform helps is that the cost-of-failure input, production minutes lost per failure, is already in the OEE event stream, so the per-asset-class analysis is a query rather than a quarterly project.
Fabrico is built for that workflow. To see how target-setting would look against your live asset data, book a demo .
The honest answer is "look up the design spec." If the design spec is not available, common with older assets, start with the current 12-month rolling actual and revisit after 90 days. Avoid the "industry benchmark" path; the variance across plants is wider than the benchmark range.
Estimate from the steps: diagnosis time + parts time + repair time + test + documentation. Each step is bounded by what is physically achievable given the current process. The estimate gets refined as actual repairs accumulate. A wrong estimate that is reviewed quarterly is better than a "correct" estimate that is never tested.
The maintenance manager owns achieving them. The reliability engineer, where the role exists, owns setting them, pulling design specs, analysing failure histories, computing the cost-of-failure inputs. In plants without a reliability engineer role, the maintenance manager owns both, with input from the asset's vendor or original installer.
Setting the target as a percentage improvement over the current actual without examining whether the current actual is reasonable. If the current MTBF is far below what the asset's limiting components are rated to deliver, the right target is anchored to that rated capability, not to 10% above the current actual.
The current actual may be telling you the asset is being run wrong, not that 10% better is the goal.
For specific asset classes, yes. For plant-wide aggregates, no. The aggregates obscure more than they reveal. Plants that report only plant-wide MTBF and MTTR usually find that the numbers do not drive any operational decision, which is the working signal that the metric needs to be unwound to per-asset-class.
Login
Blog