Semi-supervised learning of decision making for parts faults to system-level failures diagnosis in avionics system

Supervised fault detection and fault diagnosis are the techniques for recognizing small faults with abrupt or incipient time behavior in closed loops. Thus the acquired data scale and software scale became more and more huge that active fault diagnosis treats with the data hardly. After decades of Artificial Intelligence development, AI technology has achieved significant results. Machine learning methods in AI have been widely used and developed in the field of fault diagnosis and prognosis. This paper discusses and demonstrates a complete machine learning fault diagnosis structure based on support vector regression, neural gas clustering, multiple-classes support vector machine, and Bayesian fuzzy fault tree, which are semi-supervised to isolate and predict faults from a component to a system/subsystem when there are partly uncertainty faults and finally provide a decision for the maintenance. It is crucial that machine learning methods are applied in the fault detection and prediction. Furthermore, the diagnostic intelligence can be found in multi-dimension empirical data and from granularity partition of avionics system based on the knowledge found and representation. Therefore the symptom-knowledge-information is suitable for representing the faults or failures in a system. The presented structure is generic and can be extended to the verification and validation of other diagnosis and prognostic algorithms on different platforms. It has been successfully preventing aircraft system/subsystem failures, identifying and predicting failures that will occur, which provides real application on making health management information and decisions.