Detecção e diagnostico de falhas em sistemas dinamicos utilizando redes neurais e logica nebulosa

AUTOR(ES)
DATA DE PUBLICAÇÃO

1999

RESUMO

Fault detection and diagnosis methods have been intensively studied lately, as a result of the demand for systems of greater reliability. In this work, computational intelligence methods were adopted, in a configuration that uses artificial neural networks and fuzzy logic for monitoring dynamic systems represented by state-space models of adequate dimension. The parameters of the model are recursively estimated using a Hopfield-type recurrent neural network, endowed with additional optimization mechanisms. The outputs of the estimator are taken as symptoms submitted to a pre-diagnostic module, in order to establish the nominal reference parameter values. A third module is used to implement the diagnosis itself, based on a fuzzy inference method. The knowledge base is determined by means of a signed directed graph, that represents the relations among the physical parameters and the parameters of the state-space model. Besides several simulations using time-invariant linear systems, the configuration was also tested in the presence of time-varying dynamics and smooth nonlinearities including artificial and real world systems. The results guide to the conclusion that the developed methods can be successfully applied to a wide range of dynamic systems

ASSUNTO(S)

localização de falhas identificação de sistemas redes neurais (computação) estimativa de parametro sistemas difusos

ACESSO AO ARTIGO

Documentos Relacionados