A hierarchical neuro-fuzzy system for identification of simultaneous faults in hydraulic servovalves

This paper addresses the problem of identifying either single or simultaneous faults that may practically occur in hydraulic servovalves. A hierarchical neuro-fuzzy system is proposed to deal with the complex data and the interference between the phenomenological features of the faults. The proposed system decomposes the task of identification into three manageable subtasks. A mechanism of information abstraction from each stage is implemented. Data clustering and offline hybrid training are used to construct the rule base and the network structure. The suggested system was applied to identify simulated faults and compared with the single-stage system available in the literature. It was found that the hierarchical neuro-fuzzy system is capable of detecting the faults through the whole tested range while the single inference system cannot deal with the complex data. The proposed diagnostic scheme offers a simple design procedure and high feasibility. It was also shown that appropriate architecture and associated knowledge structure affect the accuracy of results.