A major goal of intelligent control systems is to achieve high perform
ance with increased reliability, availability, and automation of maint
enance procedures. In order to achieve fault tolerance in dynamical sy
stems many algorithms have been developed during the past two decades.
Fault diagnosis and accommodation methods have traditionally been bas
ed on linear modeling techniques, which restricts the type of practica
l failure situations that can be modeled. This article presents a lear
ning methodology for failure detection and accommodation. The main ide
a behind this approach is to monitor the physical system for any off-n
ominal behavior in its dynamics using nonlinear modeling techniques. T
he principal design tool used is a generic function approximator with
adjustable parameters, referred to as on-line approximator. Examples o
f such structures include traditional approximation models such as pol
ynomials and splines as well as neural networks topologies such as sig
moidal multi-layer networks and radial basis function networks. Stable
learning methods are developed for monitoring the dynamical system. T
he non-linear modeling nature and learning capability of the estimator
allow the output of the on-line approximator to be used not only for
detection but also for identification and accommodation of system fail
ures. Simulation studies are used to illustrate the learning methodolo
gy and to gain intuition into the effect of modeling uncertainties on
the performance of the fault diagnosis scheme.