Estimation, compression and classification of Volterra kernels with application to process diagnosis

The identification of non-linear systems whose input-output relationship can be summarised as the Volterra functional expansions of the system output with their inputs is considered in this paper. The Volterra series-like input-output expansion exists for an important large class of non-linear systems. Such expansions are fully specified by a set of Volterra kernels that can be estimated from the I-O measurements of the systems under study. Also, is studied the possibilities for Volterra kernel´s estimation of non-linearity by means of neural networks, for direct applications in pattern recognition and diagnosis problems. The neural methods presented can be further developed to study more complicated models, and will therefore have future potential for modelling and identifying highly complex multi-input multi-output nonlinear systems. The feasibility of these methods is demonstrated using simulated examples