The table estimates the best fitting model for failure times. The
accelerated life testing model takes the form of:
Prediction = intercept + coefficient(predictor) + scale (quantile function),
or
Yp = β0 + β1(x) + σΦ-1(p)
Where:
- Prediction (Yp):
log failure time (Weibull, exponential, lognormal, and loglogistic) and failure
time for extreme value, normal and logistic distributions.
- Intercept (β0): log
failure time or failure time (depending on distribution) when the transformed
accelerating variable and the percentile of the quantile function are 0.
- Coefficient ( β1):
regression coefficient associated with x.
- Predictor (x): transformed
accelerating variable.
- Scale (σ): scale parameter.
For the Weibull distribution scale = 1.0/shape.
- Quantile function
(Φ-1(p)): The pth quantile of the standardized life distribution.
Verify that the model assumptions, such as the distribution, equal shape
(for the Weibull distribution and the exponential distribution), equal scale
(for other distributions), and the transformation, are appropriate for your
data. Use probability plots to check the assumptions of the model. These
diagnostic plots assess the appropriateness of the model at accelerated levels
of temperature. However, engineering knowledge is the only way to verify that
the model is appropriate at design temperatures.
Because of the uncertainty in the prediction of failure time at design
conditions, evaluate the model periodically as more information, such as field
data, becomes available.
Example output
Response Variable Start: StartTime End: EndTime
Frequency: Count
Censoring
Censoring Information Count
Right censored value 95
Interval censored value 58
Estimation Method: Maximum Likelihood
Distribution: Weibull
Relationship with accelerating variable(s): Arrhenius
Regression Table
Standard 95.0% Normal CI
Predictor Coef Error Z P Lower Upper
Intercept -17.0990 4.13633 -4.13 0.000 -25.2061 -8.99195
Temp 0.755405 0.157076 4.81 0.000 0.447542 1.06327
Shape 0.996225 0.136187 0.762071 1.30232
Log-Likelihood = -191.130
Interpretation
For the electronic device data, the table provides estimates of the
best-fitting model, assuming a Weibull distribution with an Arrhenius
transformation. The estimated model is:
log(Yp) = −17.0990 + 0.755405 x + (1.0/0.996225) *
Φ-1(p)
Where:
- Yp: failure time
for the electronic devices
- x: [11604.83/(Temp +
273.16)] (Arrhenius transformation)
- Φ-1(p): the
quantile function (for more information, go to
Methods
and formulas for equations and click "Quantile function".)