Research on three-state reliability evaluation method of high reliability system based on multi-source prior information

Abstract

A high reliability system has the characteristics of complexity, modularization, high cost and small sample size. Throughout the entire lifecycle of system development, storage and use, the high reliability requirements and the risk analysis form a direct contradiction with the testing expenses. In order to ensure the system, module or component maintains good reliability status and effectively reduces the cost of sampling tests, it is necessary to make full use of multi-source prior information to evaluate its reliability. Therefore, in order to evaluate the reliability of highly reliable equipment under the condition of a small sample size correctly, the equipment reliability evaluation model should be built based on multi-source prior information and form scientific computing methods to meet the needs of condition evaluation and fund assurance of high reliability system. In engineering practice, high reliability system or module gradually develops from normal state to failure state, generally going through three working states of “safety-potential failure-functional failure”. Firstly, the historical test data under the three states can be used for the data source for the reliability evaluation of the system at the current stage, which supplements the deficiency of the field data; secondly, due to the lack of accurate judgment on the working state of a high reliability system or modules and analysis of the health status, the unnecessary maintenance may aggravate the evolution speed from potential failure to functional failure; thirdly, when high reliability system or module operates under overload or harsh conditions, the potential failure will be worsened to a certain extent. Aiming at the difficulty of multi-state system reliability evaluation, a reliability evaluation method based on non-information prior distribution is proposed by fusing multi-source prior information, which provides ideas and methods for reliability evaluation and optimization analysis of high reliability system or module. The results show that the three-state reliability evaluation method proposed in this article is consistent with the actual engineering situation, providing a scientific theoretical basis for preventive maintenance of high reliability system. At the same time, the research method not only helps evaluate the reliability state of a high reliability system accurately, but also achieves the goal of effectively reducing test costs with good economic benefits and engineering application value.

Unlocking the potential of LSTM for accurate salary prediction with MLE, Jeffreys prior, and advanced risk functions

“…First, as a non-informative prior, Jeffreys prior does not rely on subjective prior information, consistent with MLE’s “no need for priors” viewpoint. Secondly, it has good invariance properties, i.e ., it maintains its form under parameter transformations ( Clarke & Barron, 1994 ; Huang, Huang & Zhan, 2023 ), which is crucial for multi-parameter and complex models. Feasibility and robustness: Bayesian methods using Jeffreys prior are computationally feasible and can generally be efficiently computed using numerical methods like MCMC.…”