Interpretation summary for Accelerated Life Testing

Accelerated life testing performs a regression with one predictor that is used to model failure times at extreme stress levels and to extrapolate back to normal use conditions. The predictor in the regression is an accelerating variable; its levels are more extreme than those normally used in the field. Use engineering knowledge to find a model that, when estimated under more extreme stress levels, can be used to extrapolate back to normal-use conditions.

Accelerated life testing consists of the following output:

A regression table, which describes the model that predicts failure times.
Goodness-of-fit measures, which help you assess the fit of different models.
Tables of percentiles and survival probabilities, and a relation plot, which help you assess the reliability of the product.
Probability plots of the residuals, which help you assess whether the model assumptions are appropriate.

Data description

Some electrical current will leak between transistors inside an electronic device. If the leakage reaches a certain threshold, the device will short. Electrical current leakage increases under increased temperatures. A manufacturer tests electronic devices to estimate the B5 life at the design temperature of 55° C and at the worst-case temperature of 85° C.

Because the devices should last for several years under normal operating conditions, running them until they fail is not a practical method to test them. To make the devices fail more rapidly, the manufacturer tests them under much higher than normal temperatures. A device fails when its leakage reaches a specified threshold value. The devices are inspected for failure every two days.

The manufacturer chooses a Weibull distribution with an Arrhenius transformation to model the data.

Data: CurrentLeakage.MTW