Technology / Technical guide
MTBF, Failure Rate & Availability: Different Wind Metrics
Reliability, failure rate and availability describe different statistical questions. A mean interval between failures is not a lifetime guarantee, and an available machine may still have little resource to convert. The model and time boundary determine what each number can mean.
A mean is not the time at which every unit fails.
Understand probability, event rate and recovery separately before interpreting a headline reliability statistic.
01 / Metric identities
Non-repairable life and repeated failures are different.
MTTF describes a mean time to failure for a stated population and model.
MTTF describes a mean time to failure for a stated population and model. MTBF is used for repeated failures of repairable equipment, with its time convention defined. Operating time and calendar time including repair or waiting can produce different numerical means.
NIST treats exponential life and homogeneous repeated-event models as stated assumptions. Their terminology does not turn a repairable system’s mean event interval into the physical lifetime of every component. The counted failure also needs a defined function.
02 / Hazard meaning
A conditional rate is not a finite-period probability.
A hazard rate describes instantaneous failure tendency conditional on survival to the relevant time.
A hazard rate describes instantaneous failure tendency conditional on survival to the relevant time. Its units are inverse time. A probability over a whole period is dimensionless and depends on how that rate acts across the period.
For the constant-hazard exponential model, R(t) = exp(−λt), where λ is the assumed rate and t uses the corresponding time unit. NIST gives this reliability relation and a mean waiting time of 1/λ. Other hazard behaviour needs a different model.
03 / Illustrative model
One assumed mean permits widely varying outcomes.
Hypothetical constant-hazard calculation: λ = 0.001 per operating hour gives a mean waiting time of 1,000 operating hours.
Hypothetical constant-hazard calculation: λ = 0.001 per operating hour gives a mean waiting time of 1,000 operating hours. Reliability over 100 hours is exp(−0.1), approximately 90.5%. These invented numbers belong to an ideal statistical model.
At the 1,000-hour mean, survival probability is exp(−1), approximately 36.8%; units do not all last exactly this duration. No MARSTON reliability target or life is established.
04 / Age and evidence
A constant rate does not describe every life phase.
Wear, changing environment or an evolving population can make hazard depend on age or conditions.
Wear, changing environment or an evolving population can make hazard depend on age or conditions. Combining unlike failure modes into one constant rate may conceal those differences. A mean estimated from one represented population should not silently describe another configuration.
An observation with no failures does not prove an infinite MTBF or zero underlying hazard. Finite evidence and censored observations affect what can be inferred. Uncertainty in a statistical estimate is distinct from the assumed shape of its model.
05 / Availability
Recovery changes time available for the defined function.
In a simple long-run alternating up/down model, availability is mean up duration divided by mean up duration plus mean down duration.
In a simple long-run alternating up/down model, availability is mean up duration divided by mean up duration plus mean down duration. The states, included delays and stable cycle assumptions must be declared. This ratio is different from no-failure probability over a mission.
With invented mean up duration of 90 h and mean down duration of 10 h, that model gives 90% availability. The same failure occurrence behaviour with different recovery delays can produce different availability. The result establishes neither instantaneous supply nor energy yield.
06 / Product scope
Statistical labels do not establish turbine performance.
The downtime guide addresses observed event records.
The downtime guide addresses observed event records. Here the focus is the statistical model: hardware availability, valid data and useful generation are separate quantities, even when reported as percentages.
No MARSTON reliability guarantee is given. These ideal models prescribe no tests, maintenance intervals or operating changes.
Continue the review with Fail-Safe & Fail-Operational: Wind Control Concepts.
Continue the review with Redundant Sensors: Voting, Agreement & Common Causes.
Continue the review with Wind Control State Machines: State, Trigger & Completion.
General background: NIST: Exponential reliability, constant hazard and mean waiting time · NIST: Exponential and repeated-event model assumptions for MTBF. These references explain industry concepts; they are separate from the applicable MARSTON product and project documents.
Key concepts and conditions.
Read each quantity with its reference, units and stated conditions. A general relation does not establish a model-specific value.
| Concept / quantity | Meaning / relation | Conditions / limits |
|---|---|---|
| Mean interval | MTTF or MTBF under a stated convention | Population, event and included time are defined. |
| Hazard | Inverse-time conditional rate | Not itself a finite-period probability. |
| Exponential reliability | exp(−λt) | Constant hazard and matching time units. |
| Availability | Mean up/(mean up + mean down) | Defined alternating states and long-run assumptions. |
| Useful generation | Electrical contribution over its own boundary | Availability alone does not supply resource or output. |
Engineering questions.
Is MTBF a guaranteed lifespan?
No. It is a statistical mean with a defined event and time convention.
Does a rate of 0.001/h mean 0.1% failure probability for every period?
No. Period probability depends on duration and the assumed model.
Does no observed failure prove zero hazard?
No. Finite observations leave statistical uncertainty.
Is availability the same as no-failure probability?
No. Availability includes restoration of the defined function.
Does 90% availability mean 90% rated energy?
No. Resource, output and the energy boundary remain separate.
Are the example metrics MARSTON values?
No. They are invented illustrations, not product reliability claims.
Technical enquiry
Discuss the technical question.
Reliability, failure rate and availability describe different statistical questions. Share the model and applicable operating or design information for an engineering discussion.