Abstract: The human auditory system possesses excellent capability to analysis non-stationary signal. In auditory system, before a signal is recognized by the auditory cortex, it is sequentially processed by the basilar membrane, which can be seen as a bandpass filterbank, and other elements in auditory system. Therefore, to describe the structure features of signal in time-frequency space, an auditory model is proposed based on Wang-Brown model and the auditory nerve fiber oscillatory network with single layer. This model consists of basilar membrane, inner hair cells, middle auditory stage and auditory cortex, and the auditory cortex model is a single layer auditory nerve fiber oscillatory network. According to the characteristic of mechanical vibration signal, the random term and lateral inhibitor in Wang-Brown model are ignored, and the inner hair cells model is simplified. Furthermore, the active rule of neuron and the connection mode between neurons are designed. In proposed model, the oscillatory network synthesizes the output of the preceding submodels. The oscillation of neurons corresponding to similar time-frequency structure is synchronized. Therefore the distribution of synchronized neurons is utilized to describe the time-frequency structure feature of the analyzed signal. The proposed model is evaluated by using the run-up vibration signals of a rotor with different fault and of a gear box used on wind turbine. The results show that the proposed model can effectively describe the structure features and evolvement of a signal with low data quantity, and is sensitive to the instantaneous change of the signal. Then the model is convenient to be applied in intelligent recognition.

Key words: Auditory model, Faults diagnosis, Features extraction, Time-frequency analysis