Graphical Analysis of Accelerated Life Test Data with a Mix of Failure Modes

An accelerated test of a product under high stress yields life data that are extrapolated to estimate the life distribution at low stress. When the data contain competing failure modes, the dominant modes under high stress may not dominate under low stress. This has made experimenters doubt the validity of tests yielding such data. This expository paper presents new graphical methods for estimating: 1) a separate relationship between life and stress for each failure mode, 2) the life distribution at low stress when all modes act, and 3) the life distribution that would result if certain modes are eliminated. Data from a temperature-accelerated life test of electrical insulation illustrate the methods.