Condition classification and tendency prediction for prognostics using feature extraction and reconstruction

As the development of Condition Based Maintenance (CBM), condition monitoring and prognostics are playing increasingly important parts in maintenance plans which can extract the fault onsets in running process and make maintenance strategy before component failure. Condition classification applied in condition monitoring with a significant deterioration process can classify the equipments' conditions into three categories: normal state, abnormality and fault state which are beneficial for the determination of maintenance plans. Using the data measured in field, three faults represented by three parameters in some type of engine are studied in this paper. Firstly, condition classification under original feature vector is studied and the classification effects are evaluated using Learning Vector Quantization (LVQ) neural network. Secondly, condition classification in low dimensional space by feature reconstruction is studied as well as the adaptability of reducing dimensions. The parameter transforms and feature mapping methods considered in this paper can preserve the separability while the feature mapping method is more robust. Thirdly, tendency prediction is studied for prognostics when the engine's state is abnormal. The general prediction algorithm is presented based on the low dimensional space reduced by feature mapping method. Good separability is acquired in low dimensional space with the definitions of departure angle and distance. Finally, the possibility of fault occurrence can be established for prognostics based on the algorithm.

[1]  C.S. Byington,et al.  A model-based approach to prognostics and health management for flight control actuators , 2004, 2004 IEEE Aerospace Conference Proceedings (IEEE Cat. No.04TH8720).

[2]  Stephan Staudacher,et al.  Probabilistic Engine Performance Scatter and Deterioration Modeling , 2008 .

[3]  Glen Karlsons,et al.  Performance Deterioration Modeling in Aircraft Gas Turbine Engines , 1998 .

[4]  Jay Lee,et al.  Feature signature prediction of a boring process using neural network modeling with confidence bounds , 2006 .

[5]  R. Brereton,et al.  Comparison of performance of five common classifiers represented as boundary methods: Euclidean Distance to Centroids, Linear Discriminant Analysis, Quadratic Discriminant Analysis, Learning Vector Quantization and Support Vector Machines, as dependent on data structure , 2009 .

[6]  Bo-Suk Yang,et al.  Support vector machine in machine condition monitoring and fault diagnosis , 2007 .

[7]  Kai Goebel,et al.  A Survey of Artificial Intelligence for Prognostics , 2007, AAAI Fall Symposium: Artificial Intelligence for Prognostics.

[8]  Daming Lin,et al.  A review on machinery diagnostics and prognostics implementing condition-based maintenance , 2006 .

[9]  Carl S. Byington,et al.  An Overview of Selected Prognostic Technologies With Application to Engine Health Management , 2006 .

[10]  A. M. Birk,et al.  Development of Fault Diagnosis and Failure Prediction Techniques for Small Gas Turbine Engines , 2001 .