Reliability Growth and Life Data Analysis
Reliability Growth and Life Data Analysis
This domain tests growth models (Duane, Crow-AMSAA) and life-data fitting: Weibull plots, maximum likelihood, censoring, and B10 life with confidence.
Core rules
- Duane: cumulative failure rate C(T)=λT^(-α); a positive slope α on log-log means growth. Crow-AMSAA NHPP intensity ρ(t)=λβt^(β-1); β<1 indicates growth.
- Growth tests are Test-Analyze-Fix. Fixes must be corrective and verified, not merely delayed.
- Life data: rank failures, include suspensions, fit the distribution, check fit, then compute reliability and B_p life.
- Confidence intervals widen with few failures and heavy censoring.
Worked example
At T1=100 h, cumulative failures=20; at T2=1,000 h, cumulative failures=50. Duane slope α=ln(50/20)/ln(1,000/100)=0.916/2.303=0.398. Crow β=1-α=0.602, less than 1, indicating growth. The instantaneous MTBF at T2 is better than the cumulative average; using the cumulative curve alone understates current reliability.
Traps
- Plotting interval failures instead of cumulative failures on a Duane plot.
- Treating a growth slope as demonstrated reliability.
- Dropping suspensions from Weibull analysis.
- Extrapolating beyond the observed failure mode.