Artificial neural networks can be trained to represent a relationship between sets of data, such as the input-output relationship of a non-linear dynamic system. To go back to my LinkedIn page click here.

Artificial neural networks loosely mimics biological neurons and could be trained to represent an arbitrary relationship between sets of data. The topology of the network is closely bound to the type of relationship that can be represented.

The figure on the left shows a feed-forward artificial network capable of representing a non-linear dynamic system.

The training of artificial neural networks require sets of data representing the relationship to be mimicked so that the trained neural network truly represents the relationship within the required range of frequency and amplitude. To achieve this, and be able to design the neural network topology (as opposed to using trial and error methods), special excitation waveforms have been developed.

In the figure on the left, the top trace shows such a excitation waveform, and the bottom trace shows the resultant output of a non-linear dynamic system.