In recent years, neural networks have been exploited successfully in
pattern recognition as well as function approximation theory and they

have been proposed as a possible technique for fault diagnosis, too.
Neural networks can handle nonlinear behavior and partially known

process because they learn the diagnostic requirements by means
of the information of the training data. They are noise tolerant and their

ability to generalize the knowledge as well as to adapt during use
are extremely interesting properties.

The major motivation of this research is the use of artificial static
neural networks, *capable of approximating a large class of functions*,

for fault diagnosis of an industrial plant, too. The ultimate goal
is to develop a general method which can be applied to a broad spectrum
of

processes. In particular, in the detection and diagnosis problem, neural
networks are exploited to estimate the relationship between

symptoms and faults. In such a way, residuals generated by means of
state estimation techniques are independent of the dynamic

characteristics of the plant and dependent only on sensors faults.
Therefore, the neural network evaluates static patterns of residuals,

which are uniquely related to particular fault conditions independently
from the plant dynamics.

** Dynamic neural networks** **in system identification**

This research concerns model identification by linear and nonlinear
dynamic neural networks. Linear networks may be used to model real

systems. If the real system is linear or near linear then the linear
network can act as a zero, or low, error model. In the real system is nonlinear

linear network models the system with minimum sum-squared error. Nonlinear
networks can be used to identify a nonlinear system.

Two networks are commonly used: *Elman and Hopfield networks*.
Elman networks are two-layer backpropagation networks, with the

addition of a feedback connection from the output of the hidden layer
to its input. This feedback path allows Elman networks to learn to

recognize and generate temporal patterns. The Hopfield network is used
to store one or more stable target vectors. These stable vectors can

be viewed as memories which the network recall when provided with similar
vector which act as a cue to the network memory.

Previous Page | Simani Home Page | Dipartimento di Ingegneria |