1;3409;0c Hierarchical representation and machine learning from faulty jet engine behavioral examples to detect real time abnormal conditions

Hierarchical representation and machine learning from faulty jet engine behavioral examples to detect real time abnormal conditions

1st international conference on Industrial and engineering applications of artificial intelligence and expert systems - Volume 2, 1988
Pages: 710-720DOI: 10.1145/55674.55684

IEA/AIE

bibtex

Jet engine behavior can be described by four major engine parameters and by their temporal and qualitative relationships. These parameters are rotational speeds of the low and high pressure turbine assemblies referred to as N1 and N2 respectively, exhaust gas temperature EGT, and combustion temperature COMBT. Normally, these parameters show stable readings. Faulty conditions like fuel interruption or bearing loss cause significant variations in the values of these parameters. These variations have been transformed into structural representations involving instances and events. Every instance node is associated with a moment in time when a significant change occurs in the value of a parameter, and the node stores information related to this moment. An event stores information related to the happenings between two instances. Our analysis of engine behavior under normal and faulty conditions has illustrated several important features called descriptors which contribute significantly to diagnosing and predicting faults. These descriptors are generated from temporal and qualitative features derived from sensor data and from relationships between these features. For example, a