Исследование диагностических рисков параметрически неопределенной квадратичной дискриминантной функции при диагностике промышленных объектов

Abstract

A probabilistic model of quadratic discriminating transformation of input random vector measuring signals in the information system of technical diagnostics is considered, when the coefficients of the transformation function are random. It is shown that such a model is a superposition of conditional (by classes of diagnosed states) central and no central Fisher distributions, the number of degrees of which depends both on the volumes of training samples of the input vector components and on the classes of states. A method for estimating the reliability of diagnostics as a function of not only the dimensionality of the input vector of measurable informative features but also the discriminate function is proposed.