APPLICATION OF GEOMETRICAL FEATURE SELECTION METHOD TO MACHINERY STATE RECOGNITION
The non-invasive diagnosis (recognition of an object?ˉs state) could be treated as a process of detection and discrimination of the object?ˉs defects as a result of collection, processing and analysis of the features of registered diagnostic signals (that is of such signals whose features depend on the technical state of the diagnosed object). In connection with this the construction of a diagnostic system calls for selection of these features of registered signals which are defect-oriented (the diagnostic parameters). The paper will present a geometrical feature selection method which relies on two criteria of separation of classes: the criterion of average scatters and the criterion of the number of prototypes of classes. This method, which determines the significance of the features depending on the assumed classification task, is a strong tool for diagnostic information selection and it enables delivering the relevant set of diagnostic parameters to the classifier. The paper will also present the example of practical use of geometrical feature selection method in machinery state diagnosis.
Artificial Intelligence Pattern Recognition Feature Selection Vibroacoustic Diagnostics
Jacek DYBAA
Institute of Machine Designs Fundamentals, Warsaw University of Technology ul. Narbutta 84, 02-524 Warsaw, Poland
国际会议
北京
英文
442-447
2008-10-27（万方平台首次上网日期，不代表论文的发表时间）