Real-time reliability assessment method based on performance degradation tracked by dynamic probability model

In the process of real-time reliability assessment, acquisition of prior information is difficult and distribution hypothesis does not always conform to the actual situation, thus a real-time reliability assessment method based on dynamic probability model is presented. Taking nonparametric kernel estimation method, a moving probability neural networking is constructed, and the sliding time-window technique is used to pick statistical samples respectively and then conditional probability distribution of performance degradation data is estimated. The distribution value of performance degradation data more than failure threshold is regarded as the reliability indicator. The individual equipment reliability assessment can be accomplished without any prior information. The analysis of data from high pressure water descaling pump and heating furnace fan in the process of failure verifies the feasibility and practicability.