Estrategias de detecção e diagnostico de falhas em sistemas dinamicos

AUTOR(ES)
DATA DE PUBLICAÇÃO

1997

RESUMO

Analytical redundancy for fault detection and diagnosis of dynamic systems, FDD, has been approached by several methodologies, state estimation, parameter estimation, expert systems, and pattern classification and recognition being typical examples. In particular, pattern classification and recognition methods adopt probabilistic, heuristic, neural, and fuzzy set based techniques as a solution framework. This work introduces the FDD as a pattern classification and recognition problem. Two neural approaches for pattern classific ation and recognition are introduced and compared: a neurofuzzy network composed by logical and and or neurons with a competitive learning, and a variation of a supervised Learning Vector Quantization (LVQ) network with ?N IND. 1?,? N IND. 2? and Noo norrns and a pruning scheme, respectively. In addition, three schemes are proposed to solve FDD problems. The first scheme uses input and output signals to classify the operational condition of the dynamic system. In this case mathematical models are not needed to detect and to diagnose faults. The second uses residues generated by comparisons among the states of several sliding mode observers. The third uses the parameters of a model of the system. The model parameters are the weights of a neurofuzzy network. The last two schemes constitute hybrid strategies to solve FDD problems. The schemes developed herein were tested in three different dynamic systems: a DC motor drive; an AC motor drive, and an interactive tank system. Simulation results are reported for these three example problems

ASSUNTO(S)

falha de sistema (engenharia) sistemas dinamicos diferenciais conjuntos difusos redes neurais (computação)

Documentos Relacionados