Resumo

TARIFA, Enrique Eduardo  e  MARTINEZ, Sergio Luis. Neural network fault diagnosis: Part II: flow recognition. Ing. Investig. [online]. 2007, vol.27, n.2, pp.65-71. ISSN 0120-5609.

The diagnostic system introduced in Part I is modified in this work for supervising complex processes when faults present themselves. As in Part I, it is supposed that when a fault affects a process, then each variable evolves following a trajectory. However, this time the aforementioned trajectory is not unique but belongs to a set of infinite possible trajectories named flow. Each fault in a particular flow is associated with each variable. Faults affecting a process can then be diagnosed by recognising which flow the trajectory being observed belongs to for every variable in turn. Once flows have been identified, then the fault causing them is also identified. Theory was developed after modelling fault diagnosis as being a flow recognition problem, definitions being yielded for both structure and training method for the artificial neural networks used by the new diagnostic system. The diagnostic system performed well in tests, diagnosis being exact, having high, stable resolution in the presence of noise. The theory so developed recommends networks being scaled-up for supervising more complex processes.

Palavras-chave : fault diagnosis; artificial neural network; flow recognition; optimisation; noise tolerance.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )