Developer Tutorial: Constant Stress Accelerated Life Testing
calking
| Updated on
Accelerated Life Testing is useful to evaluate product reliability for parts that need to perform for long periods and can’t afford to wait for failures to occur at their normal rate. The new JMP 18 Constant Stress Accelerated Lifetime Testing (CSALT) special-purpose DOE capability lets you create and evaluate multi-factor designs for up to 3 acceleration factors and then use a JMP script to launch the Fit Life by X model after the data is collected.
See how to interactively identify factors and test conditions, specify planning values for the underlying acceleration model and automatically create three test plans: Optimal, Compromise, and Balanced. See how to interpret design diganostics for each plan and choose your design(s) before creating the design with a simple button click.
Questions answered by Caleb @calking and Jacob Rhyne @jacob_rhyne during the live webinar demo. Additional questions are at the end of the video, beginning at time ~ 46:53.
Q: Usage Temp conditions allow you to select Low/High; can we use a Temp distribution instead?
A: The CSALT Design platform does not currently allow you to specify the distribution of factors, just the low/high usage and test conditions.
Q: Didn't see power law in the factor transformations. How would we handle that?
A: For power law models, you could use the Log transformation in CSALT. JMP Systems Engineer Charlie Whitman discusses this around the 25 minute mark in the Diving Into Accelerated Life Testing video.
Q: 0.1 (10%) Probability. What Confidence Integral is used in this case?
A: By default, CSALT Design sets alpha=0.05. You can change this using the 'Plan Comparisons' red-triangle menu.
Q: Where does the Expected Failures come from? Is it something the user needs to a priori have estimate for in their system?
A: If you're asking about setting the probability in the Acceleration Model settings, yes this would need to be an estimate based on prior knowledge. If you are asking about the Expected Failures in the diagnostics, these are computed using Unit Allocation * Failure Probability.
Q: Is CSALT useful for semiconductor lifetime modeling?
A: The models being assumed here are not degradation models. The types of tests you are running here are pass/fail. If you are looking to monitor a response over time as it ages or degrades, this does not handle that. We're looking to that for the future, so it is a good item to add to the JMP Wish List.
Q: Model Parameter says with 1-sigma, thus I guess it might be 68% CL, or alpha = 0.32?
A: In this case, JMP Is not thinking in terms of sigma used in a Control Chart where it is representing sigma as a statistical measurement of variability in the result. It is not saying it is 1-Sigma away from the mean. Instead, when you see Sigma of 1 here, JMP assumes the value of Sigma is 1. If you change it to 2, then it accounts for little bit more uncertainty because the value of
Sigma is wider. For Response Distribution type lognormal, as in the example, this value option is Sigma. If you change Response Distribution to Weibull, the option is no longer Sigma, but rather Weibull beta.
Resources
- Planning Accelerated Life Tests With Two or More Experimental Factors , Luis A. Escobar and William Q. Meeker, American Statistical Association and the American Society for Quality Control ,TECHNOMETRICS, NOVEMBER 1995, VOL. 37, NO. 4
- Register for the May 17 Diving into Accelerated Life Test Mastering JMP webinar.
