Session: 05-06 PHM Systems & Comparative Studies

Paper Number: 129597

129597 - A Comparison of Flat and Hierarchical Structures in Aircraft Engine Fault Classification Algorithms 

Gas turbine engine diagnostic algorithms using gas path measurements present an important component of an engine monitoring system. Pattern recognition and machine learning techniques are increasingly involved in this gas path diagnostics. The necessary fault classifications are formed following the principles of either flat or hierarchical structures. With the flat classification formed from all elemental fault classes, a classifier makes its decision at once. In contrast, the hierarchical classification implies grouping elemental classes in “mega-classes”, and the classification process includes different stages. The recent studies indicate that the hierarchical classification principle has some advantages, however these studies and our preliminary calculations show that the difference between these two principles can be small.

Comparing the flat classification with some variations of the  hierarchical classifications, the present study aims to ascertain which of the classifications is more accurate and why and provide a monitoring system designer with clear recommendations about a fault classification process. The comparison criteria are based on a confusion matrix that unites the probabilities of a correct and wrong fault recognition. For an accurate comparison, we compare diagnostic algorithms that differ from each other only by a classification configuration. To draw solid conclusions the comparison is repeated under different diagnostic conditions. Propulsion Diagnostic Method Evaluation Strategy (ProDiMES) is used as a benchmarking platform because it is designated for comparing diagnostic solutions and provides realistic simulation of an engine fleet based on a turbofan nonlinear thermodynamic model.

Presenting Author: Igor Loboda Instituto Politecnico Nacional

Presenting Author Biography: Igor Loboda received the M.S. and Ph.D. degrees in aircraft engine engineering from the Kharkov Aviation Institute (Ukraine) in 1979 and 1994, respectively. He was an investigator, lecture and assistant professor at the Kharkov Aviation Institute in 1992-2001. Since 2001 his has been an assistant professor and investigator at the National Polytechnic Institute of Mexico. His research interests are in the areas of modelling, simulation and condition monitoring of gas turbines and common theory of pattern recognition. Particular issues of interest are gas turbine thermodynamic models (static and dynamic), model identification, analysis of real data (gas path variables), fault identification techniques, and neural network application to the gas path diagnostics.

Authors:

Igor Loboda Instituto Politecnico Nacional
Juan Luis Pérez-Ruiz Universidad de Ciencias y Artes de Chiapas
Iván González Castillo Secretaría de Marina Armada de México
Sergiy Yepifanov National Aerospace University “Kharkiv Aviation Institute”