Lognormal distribution
A lognormal distribution models a positive quantity whose natural logarithm, expressed relative to a fixed unit, is normal.
For an unshifted lifetime T > 0, ln(T/t0) is normal with mean mu and standard deviation sigma > 0; t0 is a declared time unit. The median is m = t0 exp(mu), whereas the mean is m exp(sigma^2/2). Mu and sigma describe log time, not the arithmetic mean and spread of lifetime. Changing the time unit changes mu, not sigma; the physical distribution is unchanged. [1]
For t > 0, let z = ln(t/m)/sigma. Survival is S(t) = Phi(-z) and hazard is h(t) = phi(z)/(sigma t Phi(-z)), where Phi and phi are the standard normal CDF and density. The hazard has inverse-time units and rises before eventually falling. That model shape does not establish physical rejuvenation of individual devices. [2]
Lognormal and Weibull curves can both fit a small, censored sample well yet give very different extrapolations. Select a distribution using the failure mechanism, event definition and fit diagnostics. Taking logarithms does not remove censoring: a still-working device remains an incomplete lifetime observation. A lognormal fit is not a universal law for HDD or SSD aging. [3] [1]