Reliability Modeling And Analyzing Of Momentum Wheel Based On Bayesian Network

Momentum Wheel is a crucial electromechanical component in the posture controlling system of long-life satellite, which is characterized by small sample, long life, high reliability and Zero-Failure Data. Thus, it is very difficult to assess its reliability by adopting traditional statistical inference methods for life data of large sample. So, how to conduct the reliability modeling and analysis for Momentum Wheel is an urgent problem that is paid much attention by domestic spaceflight model units. On the other hand, although with relative small sample size and being lack of Failure Data in the spot test of Momentum Wheel, there is a great deal of accumulated prior information in the process of development, production and application of Momentum Wheel, including function test, environmental test as well as telemetry data of on-orbit operation and engineering experience. Therefore, aiming at the characteristics of its reliability, and with both learning & inference technologies on Bayesian network and rich prior information of Momentum Wheel, this paper establishes reliability model of momentum, by which analysises and assesses the reliability of Momentum Wheel XX-3 to solve the problems of reliability modeling and analysis for Momentum Wheel. And the main content of this paper is as follows:1. Construct the structure of Bayesian network through failure analysis on Momentum Wheel, on the basis of which establish the discrete Bayesian network of momentum wheel with learning technology of Bayesian network and various test data. And then assess the ground reliability level of Momentum Wheel with such model and inference technology, put up diagnosis methods for single fault and mixed fault of Momentum Wheel based on Bayesian network, and research the factors that affect the reliability of momentum wheel by sensitivity analysis.2. Based on the model of discrete Bayesian network, and integrating the capability data, such as bearing temperature and electric current etc, and orbit data of momentum wheel in experimental environment, build the Hybrid Bayesian Network model of Momentum Wheel. With this model, assess the orbit reliability level of Momentum Wheel and present a fault diagnosis method for the on-orbit status of Momentum Wheel. Besides, the paper studies the relation between the reliability and performance of momentum wheel in quantity through sensitivity analysis.3. Based on Hybrid Bayesian Network, combining with time-series model for performance data of Momentum Wheel, build the Dynamic Bayesian Network. The model solves the prediction problem of life and reliability of Momentum Wheel. Meanwhile, the paper also puts forward the conception of Dynamic Sensitivity Analysis (DSA), and analyze of Dynamic Sensitivity (DS) in the typical fault model of Momentum Wheel.At last, it is the summary and perspective of future work.