A novel multi-stage precision reliability assessment method for mechanical system by Bayesian fusion

Abstract

Precision reliability is crucial for evaluating mechanical system performance. However, limited data makes reliability assessment challenging due to difficult data collection and small sample sizes. Currently, little research has focused on using initial theoretical models as prior information for reliability assessment. This paper proposes a multi-stage precision reliability assessment method for a mechanical system by Bayesian fusion, which can effectively integrate design phase models with usage phase data under limited data conditions to carry out reliability assessment. First, the mechanical system is divided into meta-action units for precision modeling during the design phase. Then, an initial theoretical precision model is developed by incorporating operational error sources. Next, initial theoretical precision model is used to fit the Wiener process-driven model as Bayesian prior information, and reliability assessment is evaluated under different distribution assumptions for both the prior information and experimental data. This approach combines the prior advantages of the theoretical model with the data processing ability of the data-driven model under no prior data and small sample sizes, improving assessment precision and interpretability. Finally, a case study on a machine tool rotary table system validates the effectiveness of this method.