Abstract

BARRERO VICIEDO, Egly; FERNANDEZ DIAZ, Beatriz  and  LLANES SANTIAGO, Orestes. Proposal of a procedure for setting an artificial neural network of Radial Basis with applications in fault diagnosis.. EAC [online]. 2014, vol.35, n.3, pp. 60-75. ISSN 1815-5928.

In this article a procedure is shown up for allowing the configuration of parameters of artificial neuronal network architecture of Radial Basis for failure diagnosis tasks after stablishing a logical order for their selection. Such procedure guarantees the procurance of a necessary number of hidden neurons starting from fixing the diagnosis error desired by the experts for each process, also allowing the selection of the method for esteeming the widths of the hidden neurons and the equation of distance for the propagation of the inlets vector. The procedure applies to the testing process of «Reactor Tank Constantly Agitated» searching to demonstrate its effectiveness. As a result of the realized experiments, it is confirmed that within the topology of the Radial Basis net, the calculation of the centers and widths of the hidden neurons intervene decisively in its execution, as well as the fixed-distance equation, but the election of the function of distance does not act on the selection of the method for esteeming the widths. It is verified that this architecture, with the adequate training, shows good properties of sensibility and robustness.

Keywords : radial Basic artificial neural networks; fault diagnosis; robustness.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )