Data mining–based intelligent fault diagnostics for integrated system health management to avionics

Abstract

Space avionics are the essential capabilities of a spacecraft that guarantee space flight safety and mission success. One of the most important elements developed to deal with the health of the space avionics is the integrated system health management. Fault diagnostics, a safety-critical process in the integrated system health management, has become more complex as the number of avionics systems within the spacecraft has grown, so failure data are now multidimensional, often incomplete, and have cumulatively acquired uncertainties. Therefore, an accurate fault diagnostics model is needed to handle these types of data and ensure information is adequately adapted and efficiently updated. To date, there has been little research focused on efficient and effective space avionics fault diagnostics. This article presents a novel integrated system health management–oriented intelligent diagnostics methodology based on data mining. A numerical example is provided to illustrate the methodology, which demonstrates the significant benefits of data mining for the efficient processing of massive, incomplete data, and the ability of using a robust diagnostic Bayesian network to identify faults with uncertainty in a dynamic environment. The combined approach shows how some limitations can be overcome with an improved diagnostic performance. For application, sensory information must initially be discretized to Boolean values. Data mining is then used to mine for useful association rules and to learn the dynamic Bayesian network structure. After parameter training, the diagnostics is conducted. This methodology can be applied to systems of varying sizes and is flexible enough to accommodate other efficient diagnostic methods.