Weibull and Crow-AMSSA
Key Points
On the Use of Weibull vs. Crow AMSAA (Crow-AMSAA) for Repairable Systems
– Philip Sage introduced the topic after seeing a LinkedIn discussion emphasizing Crow AMSAA over Weibull for repairable systems.
– The post initially appeared to strongly advocate against Weibull, but on a closer read, it functioned as an introduction to Crow AMSAA techniques rather than a condemnation of Weibull.
– Fred Schenkelberg shared practical experiences where using Weibull on a factory bottling line provided accurate and actionable predictions, even before learning that statisticians often discourage its use for whole repairable systems.
Practical Application and Lessons
– Fred Schenkelberg noted that while Weibull modeling worked well and matched observed data, he later learned from statisticians about its theoretical limitations for repairable systems.
– The use of mean cumulative function (MCF) plots subsequently helped in communicating and modeling system repairs more intuitively.
– The decision for modeling approaches can pivot on whether it is more important for the client or team to easily understand and act on the analysis.
Insights from Research
– Philip Sage’s research at Central Queensland University uses synthetic data to test how well Weibull and Crow AMSAA techniques recover the “ground truth” of known distributions.
– Weibull analysis works especially well with individual failure modes or time-to-first failure but may not represent the actual mix of failure mechanisms in complex or repairable systems.
– Crow AMSAA is built for repairable systems, especially those with exponential (random) or electronic parts’ failure data, but it too has documented limitations.
– For heavily wear-out-dominated failure data, Crow AMSAA may be less appropriate.
Data and Methodological Considerations
– The key is understanding the underlying limitations and assumptions of whichever method is chosen.
– Crow AMSAA and Weibull both have their strengths depending on the failure modes present and the type/quality of the available data.
– Synthetic data experiments clarify where these methods succeed or fail, especially in the presence of right-censored (still-functioning) or left-truncated (unknown start time) data.
– Practical repair realities—such as imperfect repairs or repairs that worsen performance—can further complicate modeling.
Conceptual Distinctions
– Mean Time Between Failure (MTBF) versus Mean Time To Failure (MTTF) remains a source of confusion; both are sometimes calculated with similar formulas but are conceptually distinct and tied to repairable versus non-repairable systems.
– The notion of data being “independent and identically distributed” (IID) falls apart as soon as there are mixed failure modes, different assets, or changing environments.
Guidance for Practitioners
– No single method—Weibull, Crow AMSAA, MCF, etc.—is universally “correct” or “best” in all scenarios.
– Understanding limits and proper application is crucial; sometimes combining multiple analytic methods and comparing results can signal issues in data quality or appropriateness of methods.
– Be cautious about claims or memes seen online; use them as catalysts for deeper research and validation.
Concluding Thoughts
– Reliability analysis is a journey; diverse experience levels and perspectives are part of the professional landscape.
– Ongoing research is needed to refine methods and their applications as asset management and failure data complexity increases.
– Listeners are encouraged to critically evaluate methodology based on system details, data characteristics, and intended use of the results.
Enjoy an episode of Speaking of Reliability. Where you can join friends as they discuss reliability topics. Join us as we discuss topics ranging from design for reliability techniques to field data analysis approaches.
Show Notes
Philip Sage and Fred Schenkelberg discuss the differences between Weibull analysis and Crow-AMSAA (Crow AMSA) methods for assessing reliability in repairable systems, inspired by a LinkedIn post. They share practical experiences, including the application of Weibull to large factory lines and the challenges around statistically modeling complex, repairable systems. The conversation examines the strengths and weaknesses of both Weibull and Crow-AMSAA techniques based on their research using synthetic data at Central Queensland University, highlighting issues like failure mode identification, censored data, and the assumptions behind each model. They stress the importance of understanding the limits and appropriate contexts for each method, the impact of data quality, and communicating results in terms clients can understand. The episode wraps with encouragement to question online advice, keep learning, and reach out with reliability questions.
Leave a Reply