What reliability growth means.
Reliability growth is improvement produced by finding failure modes, identifying causes, implementing changes, and verifying that those changes work. More test time alone does not cause growth. A favorable curve without a credible corrective-action record should be treated as a data-quality question.
Read cumulative and instantaneous MTBF separately.
Cumulative MTBF is total test time divided by all included failures through that time. It averages early and recent performance. Instantaneous MTBF is the power-law model's fitted performance at a point in time. Under the improving Duane model, instantaneous MTBF is higher than cumulative MTBF because recent modeled performance is better than the historical average.
Build a traceable test clock.
Define whether exposure is equipment-hours, calendar hours, cycles, miles, or another unit. Keep duty, configuration, failure definition, and suspended time rules consistent. Record each failure time in order, the disposition, the implemented fix, and the configuration at return to test. A cumulative plot cannot repair missing lineage.
Use the four calculations in sequence.
Estimate growth slope as a transparent early check, then prefer a full-data fit. Use MTBF projection to define a conditional milestone and growth test time to expose the required test budget. At a completed milestone, the achieved MTBF calculator distinguishes observed cumulative MTBF from fitted end-of-test instantaneous MTBF.
Know when one slope is not enough.
Separate phases when hardware, software, duty, test strategy, or corrective-action policy changes materially. A two-point slope can hide a plateau or a late cluster of failures. Delayed corrective actions can also make calendar sequence and effective configuration sequence diverge. Review the plotted data and engineering chronology before extending a line.
Do not turn projection into demonstration.
The tools in this cluster do not calculate confidence bounds or acceptance probability. Reliability demonstration requires a separately selected statistical test plan and explicit confidence, discrimination, censoring, and distribution assumptions. A growth projection is a planning model, not evidence that a requirement has been met.
Primary technical basis.
The formulas and terminology follow the U.S. National Institute of Standards and Technology Engineering Statistics Handbook sections on NHPP power-law models, Duane plots, reliability growth planning, and power-law model relationships. See the formula reference for a compact lookup.