An Attempt to Unify Reliability and Confidence Level in Reliability-Based Design

Abstract

In reliability-based design and reliability testing, the reliability and confidence level are used together when distribution parameters are unknown. Determination of the design allowable stress in the MIL Handbook is a typical example. This conventional method of using the two-dimensional measure is somewhat complicated, and makes it difficult to compare allowable stress based on different confidence levels. In the present paper, the idea of using the mean reliability as a one-dimensional measure was proposed. This idea was applied to the determination of the design allowable stress S_a for the case when the strength follows a normal distribution N(μ, σ²). The factor k in the expessions of S_a= μ-kσfor the case of unknown μ, and S_a= μ-kσ for the case of unknown μand σ was derived based on this idea, where μand σare estimated values. A method for specifying the design goal of the mean reliability was also proposed. Furthermore, the meaning of the product of reliability and confidence level was clarified.