Stable parametric identification of vibratory diagnostics objects

Abstract

A common model of vibratory diagnostics objects is the stochastic difference schemes, and theirs parametrical identification is carried out least squares and least absolute deviations techniques. It is well known that these techniques are unstable under stochastic heterogeneity of observable process, specifically, in the presence of outliers. One way to make the stable parametrical identification of vibratory diagnostics objects is implementation of generalized least absolute deviations method based on concave loss function. Obtained requirements to the loss function guaranteeing the steadiness evaluation, algorithms of identification and examples are presented. © Vibromechanika.