Deriving bayesian classifiers from flight data to enhance aircraft diagnosis models

Abstract

Online fault diagnosis is critical for detecting the onset and hence the mitigation of adverse events that arise in complex systems, such as aircraft and industrial processes. A typical fault diagnosis system consists of: (1) a reference model that provides a mathematical representation for various diagnostic monitors that provide partial evidence towards active failure modes, and (2) a reasoning algorithm that combines set-covering and probabilistic computation to establish fault candidates and their rankings. However, for complex systems reference models are typically incomplete, and simplifying assumptions are made to make the reasoning algorithms tractable. Incompleteness in the reference models can take several forms, such as absence of discriminating evidence, and errors and incompleteness in the mapping between evidence and failure modes. Inaccuracies in the reasoning algorithm arise from the use of simplified noise models and independence assumptions about the evidence and the faults. Recently, data mining approaches have been proposed to help mitigate some of the problems with the reference models and reasoning schemes. This paper describes a Tree Augmented Näive Bayesian Classifier (TAN) that forms the basis for systematically extending aircraft diagnosis reference models using flight data from systems operating with and without faults. The performance of the TAN models is investigated by comparing them against an expert supplied reference model. The results demonstrate that the generated TAN structures can be used by human experts to identify improvements to the reference model, by adding (1) new causal links that relate evidence to faults, and different pieces of evidence, and (2) updated thresholds and new monitors that facilitate the derivation of more precise evidence from the sensor data. A case study shows that this improves overall reasoner performance.