DIAGNOSTIC FEATURE’S AND PATTERN SPACES’ informativity ESTIMATION SYSTEM

T.P. Gryzlova, A.O. Balykina

A methodical solution is proposed for the problem of feature search based on informativity estimations of characteristics of digital signals. The diagnostic feature’s and pattern spaces’ informativity estimation system makes possible an objective test of feature set. Informativity for evaluation of feature set that is the ratio of the mean interclass Euclidean distance in a feature space to the mean intraclass distance is used. The result is a pattern space set with proper clustering. In the example of the problem of system diagnostics of transmission bearings in a gas_turbine engine, the advisability of change to a new approach for choice of features’set is shown. Informativity helps in choice of spectrum coefficients with high informativity. Also it helps to choose signal processing method to solve diagnostics problems. There was shown that informativity estimations of features are helpful for correct formalization of classification problem and for choice of vibration signal processing method and decision-making behavior.

Key words: diagnostics, features, informativity, feature set, interclass Euclidean distance, intraclass distance, distance function, automatic feature search, vibration, transmission bearings, gas_turbine engine.