Support for Reliability DOE Available Only in DOE++
Overcoming Difficulties with Applying Traditional DOE to Reliability Data
Traditionally, there have been some difficulties with employing DOE analysis methods to properly analyze the response information that is often of interest to reliability engineers — product lifetime data.
Specifically, traditional DOE analysis techniques such as ANOVA and standard linear regression are based on the assumptions that the data set will be complete and normally distributed. However, the life data sets commonly used for reliability-related analyses may include information about items that did not fail (right censored data) and/or uncertainty about the time of the failure (interval data). In addition, life data sets often follow a skewed distribution such as the Weibull, lognormal or exponential. Therefore, in order to apply traditional analysis techniques to life data that do not fit the assumptions, analysts have had to adjust the interval and right censored points in their data sets (usually by treating the suspension times as failures and using the mid-points of the intervals as the failure times). Then if the adjusted data set was not normally distributed, they also had to apply a logarithmic transformation before analyzing the data. Unfortunately, the adjustments and transformations required to force the life data to fit the assumptions could result in erroneous conclusions from the analysis.
Now there is an alternative approach available only in DOE++ — reliability DOE!
Reliability DOE Provides Proper Treatment for Life Data
Reliability DOE is a special category of DOE where traditional experiment design types are combined with reliability analysis methods to investigate the effects of different factors on the life of a unit. Instead of using the F ratio (as in traditional DOE techniques), reliability DOE uses the likelihood ratio, which provides the appropriate treatment for interval and right censored data and supports analysis with skewed distributions, including the Weibull, lognormal and exponential — no adjustments or transformations required!
The reliability DOE designs available in DOE++ include:
- One Factor Reliability Designs
- Two Level Full Factorial Reliability Designs
- Two Level Fractional Factorial Reliability Designs
- Plackett-Burman Reliability Designs
