open access

https://hdl.handle.net/2115/100449

Quantifying the Systematic Error of Single-Weibull Approximations for Competing Failure Modes

英語

Decision Framework

2026-07-01

Proceedings of the 10th International Symposium on Reliability Engineering and Risk Management (ISRERM 2026)

Fitting a single Weibull distribution to all failure data is a questionable simplification for systems with multiple competing failure modes. This paper quantifies the systematic error of this approach to determine the boundaries within which this simplification remains acceptable compared to exact competing risks models. To establish these boundaries, Monte-Carlo simulations are employed to evaluate the coverage probability of the lower 90% confidence bound for the 𝐵10-lifetime. When all failure causes are known, the single-Weibull method only provides valid estimates if the competing modes share similar shape parameters. Otherwise, it systematically overestimates the lifetime. For masked failure data, a mathematically exact joint likelihood formulation exists. However, computational complexity and limited software support restrict its practical use. Therefore, engineers often use a standard simplification, treating masked failures as suspensions. This study compares the baseline single-Weibull approach against both the exact method and this suspension heuristic. Results show that at high masking, the heuristic introduces a severe positive bias. While single-Weibull provides more reliable coverage here, both simplifications collapse completely for fundamentally different failure modes. Finally, the paper provides a practical decision framework to establish strict validity limits for these heuristics. Based on expected failure physics, sample size, and masking, it defines where simplifications remain applicable versus where the exact approach is strictly required.

会議発表論文

Version of Record