Reliability growth tracking measures whether an asset or fleet is actually getting more reliable over time by plotting cumulative failures against cumulative operating time. On a Crow-AMSAA (Duane) log-log plot, a line that bends downward, a growth slope below one, means failures are slowing and reliability is improving. A line bending upward means it is getting worse.
Most plants track reliability with a single snapshot number and fool themselves. This month's MTBF is up, so things are better, except a good month can hide a worsening trend, and a bad month can hide real progress. Reliability growth tracking replaces the snapshot with a trend line that tells you the direction of travel, not just the current position.
What is reliability growth?
Reliability growth is the measurable improvement in an asset's or fleet's reliability over time as failures are found, understood, and eliminated. It is the difference between fixing failures and fixing the causes of failures: a plant doing real reliability work should see the interval between failures lengthen, year over year, as the same problems stop recurring. The concept comes from reliability engineering and defense testing, where programs formally track whether design and process changes are actually raising reliability.
The key word is tracking. Reliability growth does not happen on its own; it is the visible result of defect elimination, root cause analysis, and better maintenance strategy. The chart does not create the improvement, it tells you whether the improvement is real, which is exactly what a snapshot metric cannot do.
It is worth being clear about what "more reliable" means here. It does not mean fewer repairs this week or a faster mean time to repair, those measure how well you react to failure. Reliability growth measures how often the equipment fails in the first place, corrected for how much it ran. A fleet that runs twice as many hours will naturally log more failures, so raw failure counts lie; plotting against cumulative operating time removes that distortion and lets you compare a busy year against a quiet one on equal terms.
How do you tell if reliability is improving?
You tell by looking at the trend in failures over time, not a single period's number. A snapshot metric like this month's MTBF is noisy: with a handful of failures, one lucky or unlucky month swings it wildly, and you cannot tell signal from randomness. The fix is to accumulate: plot every failure against the total operating time at which it occurred, and read the shape of the resulting line.
This matters because the two most common reliability metrics answer different questions. MTBF and MTTF tell you the current average time between failures, a position. Reliability growth tells you whether that position is improving or decaying, a direction. A plant can have a decent MTBF that is quietly trending worse, and only a growth plot exposes it before it becomes a crisis. It also cleanly separates real improvement from the natural difference between availability and reliability which snapshot dashboards routinely blur.
What is a Crow-AMSAA / Duane plot, and how do you read it?
A Crow-AMSAA plot is a graph of cumulative failures against cumulative operating time on logarithmic axes, where the data falls on a straight line whose slope reveals the reliability trend. The model began with James Duane's 1960s observation that plotting cumulative failure rate against time on log-log paper produced a straight line; Larry Crow later gave it a rigorous statistical basis as a non-homogeneous Poisson process, and it became the reliability growth model in the U.S. defense standard MIL-HDBK-189.
Reading it comes down to one parameter, the slope beta:
| Slope (β) | What it means | What to do |
|---|---|---|
| β < 1 | Reliability is improving, time between failures is growing | Keep doing what you are doing; the defect elimination is working. |
| β = 1 | Reliability is steady, constant failure rate, no change | Your work is holding the line but not gaining. Look for new causes to attack. |
| β > 1 | Reliability is degrading, failures are speeding up | Investigate now. Something is wearing out, drifting, or being introduced by your own work. |
One more virtue: because the plot is cumulative, it is far more stable than a monthly metric. A single bad month barely bends the line, so you argue about the real trend instead of chasing noise. The plot can also flag change points, a place where the slope shifts, which often lines up with a real event: a rebuild, a new operator, a design change.
How do you build a reliability growth chart from CMMS data?
You can build one from the failure records you already have in the CMMS. The inputs are modest: a defined asset or fleet, the failures, and the operating time. The steps:
- Define the population and clock. Decide what you are tracking, one critical asset, a class of pumps, a whole line, and what counts as "time": calendar time, running hours, or cycles.
- Pull the failure events. Extract the failures from the CMMS work-order history. Be consistent about what qualifies as a failure versus routine maintenance; clean failure coding makes this far easier.
- Accumulate failures and time. Order the failures by the operating time at which each occurred, and build a running count: failure 1 at time t1, failure 2 at t2, and so on.
- Plot on log-log axes. Put cumulative time on the x-axis and cumulative failures on the y-axis, both logarithmic. The points should fall roughly on a line.
- Fit the line and read beta. The slope of the best-fit line is your growth parameter. Below one is improving; above one is degrading. Most reliability software and even a spreadsheet can do the fit.
- Watch for change points. A kink in the line marks a shift in behavior. Tie it back to what changed on the floor, and you have turned data into a story you can act on.
What actions actually grow reliability?
Reliability grows when you permanently remove causes of failure, not when you get faster at repairs. The plot is only a scoreboard; the growth comes from the work behind it:
- Root cause analysis on repeat failures. Every recurring failure is a cause you have not yet eliminated. A disciplined root cause analysis program is the single biggest driver of a downward-bending line.
- Reliability-centered maintenance. Matching each failure mode to the right task removes the failures that fixed schedules were missing. See reliability-centered maintenance.
- Condition monitoring. Catching degradation early with condition-based and predictive maintenance converts catastrophic failures into planned ones and reduces the total.
- Precision practices. Better installation, alignment, balancing, and lubrication remove the early-life failures that dominate real failure data.
All of this sits inside the broader equipment reliability strategy. The growth plot simply keeps that program honest by showing whether the effort is paying off.
What are the limits of these models?
Reliability growth models are powerful but not magic, and reading them naively causes trouble. The main limits:
- They need enough failures. With only two or three events, the fit is meaningless. The method rewards fleets and longer histories over single new assets.
- They assume consistent conditions. If the duty, product mix, or environment changes mid-history, the trend blends two different worlds and can mislead.
- The clock has to be right. Plotting against calendar time when the real driver is running hours or cycles distorts the slope. Match the time base to what actually wears the asset.
- The plot describes, it does not explain. A worsening slope tells you to investigate; it does not tell you why. That is still human work, usually root cause analysis.
What does reliability growth tracking pay back?
The payoff is early, honest feedback on whether your reliability spend is working. The supporting economics come from primary maintenance research:
- A well-run predictive program, a major driver of reliability growth, saves 8–12% over preventive alone and preventive saves 12–18% over reactive (PNNL, Maintenance Approaches).
- The failure data that reliability work targets is mostly not age-related: age-related failures are under 20% of the total (Nowlan & Heap, 1978), which is why removing specific causes, not overhauling on a calendar, is what bends the growth curve.
- The Crow-AMSAA model itself is codified in the U.S. defense reliability standard MIL-HDBK-189 giving the method a rigorous, public statistical basis rather than a vendor's chart.
Harmony's agents make the loop turn faster: they watch failure signals across the fleet, surface the recurring causes worth attacking, and track whether the fixes actually move the trend, so reliability growth is something you steer, not something you hope for. See it on a real floor in the CLS case study and pair growth tracking with disciplined reliability-centered spares so the right part is ready when a predicted failure comes due.