Dynamic Bayesian Evaluation Method for System Reliability Growth Based on In-Time Correction

Abstract

Small sample size is often a limitation in reliability field tests for expensive, destructive systems, which makes it difficult to evaluate system reliability accurately. A new system's development process is usually comprised of several stages within which the system experiences design updating, and prototype field testing, thus enabling the improvement of system reliability. A dynamic Bayesian evaluation method is presented in this paper to cope with this problem for binomial systems whose reliability grows along with stages of the system's development process. To implement the proposed method, a new discrete reliability growth model, derived from the learning-curve property, is introduced to describe system reliability growth through the stages of the development process, which is also capable of predicting the reliability at the next stage, based on the data acquired in the current stage. The prediction result, which acts as the prior knowledge for reliability at the next stage, is applied to determine the prior distribution of system reliability through the Maximum Entropy method. The advantage of the method is that it employs field test data from various stages of the system's development process to evaluate system reliability, comparing with traditional methods. The given example demonstrates that the proposed method adapts to evaluate system reliability when the sample size of the field test is small.