A Condition Monitoring Methodology Based on a Non-Parametric System Identification Approach

Abstract

Abstract The paper presents a non-parametric approach to a general model-based scheme of the condition monitoring of a technology process. Within such an approach, the total inference on a technology process status is implemented by current comparing preliminary nominal indicators obtained, which correspond to the nominal technology process model with corresponding characteristics obtained by current observations. A consistent measure of stochastic dependence of random processes is proposed to be used as a technology process behavior indicator. Such a measure is the maximal correlation function. It consistently captures the actual nonlinear dependence between random processes, while conventional measures of dependence based on the dispersion and, all the more so, ordinary product correlation functions do not. Meanwhile, there are known examples demonstrating that actual dependence between model variables may be nonlinear even if the regression of a variable onto another one is linear. Within the case, such dependence can be properly described by maximal correlation ultimately. Such a consideration justifies applying consistent measures of dependence when applied to the model-based condition monitoring methodology.