Transformer Core Engineering Notes · CF-N299
A reliability percentage is incomplete without a population, observation period and definition of failure. A claim based on a small new fleet cannot be compared directly with one based on decades of mixed equipment service. Ask how the data were collected and what was counted before using the number to rank transformer suppliers.
Define the population and exposure
Record the number of units, their ages, design families and relevant operating conditions. Unit-years of exposure can help describe a fleet, but the underlying age distribution and duty still matter. Excluding retired units or losing track of installations can bias the result. A supplier should state which equipment is included, how follow-up is maintained and whether the reported population represents the design currently being offered.
State what counts as a failure
A major tank failure, an accessory replacement and an unplanned outage are different event definitions. Repairs may allow a unit to continue operating, so repeated events need careful treatment. NIST distinguishes repairable-system analysis from lifetime models for non-repairable populations. The comparison should preserve that distinction rather than mixing first failures, repeated repairs and warranty claims into one unexplained percentage.
Show the denominator and uncertainty
For a hypothetical fleet with two counted events over 1,000 unit-years, the observed event rate is 0.002 events per unit-year. That arithmetic is not automatically an annual failure probability, and the example is not Chenfan fleet data. Small event counts can produce substantial uncertainty. Zero observed events also does not prove zero risk; the exposure and confidence assessment remain necessary.
Compare evidence quality, not marketing precision
Ask for the data source, exclusions, observation cutoff and independent verification where available. Separate manufacturer reports from audited or independently collected records. Consider differences in service environment and maintenance before attributing outcomes entirely to design quality. A useful procurement conclusion states what the evidence supports and what remains unknown, rather than treating extra decimal places as proof of superior reliability.
| Review item | Evidence to request | Boundary to preserve |
|---|---|---|
| Population | Units, ages, design families and exposure | A new small fleet is not comparable with a mature mixed fleet |
| Event definition | First failure, repair, accessory event or outage | Different definitions produce different statistics |
| Statistical meaning | Denominator, exclusions and uncertainty | Zero observed failures does not mean zero risk |
References and scope
- NIST/SEMATECH e-Handbook: Non-repairable populations and repairable systemsGovernment statistical reference · Population definitions, failures and the distinction between repairable-system events and non-repairable lifetimes.
- IEC 60076-1:2011 — Power transformers: GeneralPublic standard abstract · General transformer scope, interfaces and controlled equipment specification.