This paper develops an adaptive approximation
based approach for distributed fault diagnosis for a class of interconnected
continuous-time nonlinear systems with modeling
uncertainties and measurement noise. The proposed approach
integrates learning with filtering techniques and allows the
derivation of tight detection thresholds. This is accomplished
in two ways: at first by learning the modeling uncertainty
through adaptive approximation methods, so that the learned
function is used for the derivation of the residual signal, and
then by using filtering for dampening measurement noise. The
required signals for both tasks are derived through a two-stage
filtering process, by exploiting the properties of the filtering
framework. Finally, simulation results are used to demonstrate
the effectiveness of the proposed approach
