Diagnosis in Linear and Nonlinear Mixed-Signal Systems: a Parameter Identification Based Technique

In this paper, we consider the nonlinear system modelling problem for on-chip testing and diagnosis of embedded mixed-signal systems. A SituationDependent AutoRegressive model with eXogenous variable (SDARX) is introduced to approximate the conventional Nonlinear-ARX (NARX). The parameter search space is divided into a linear weight subspace and the nonlinear parameter subspace. A nonlinear parameter estimation strategy combines the Levenberg-Marquardt method (LMM) for nonlinear parameter optimization and the leastsquare method (LSM) for linear parameter estimation. The diagnosis procedure requires a recursive estimation of the model parameters corresponding to the nominal behaviour, using input-output data recorded on the system under test. Emphasis is given to the characterisation of a particular failure mode by choosing the best model structure and identification of model parameters for diagnostic purposes. For fault identification, the parameter estimation technique is associated with the fault dictionary approach.