Blind system identification is a signal processing technology

which obtains the function of system characteristics from its output data only.The system characteristics are associated with its own structure

peroperating mode corresponding to a particular working state.Gearbox fault diagnosis can be conducted by combining the system structure and working condition.Firstly

using independent component analysis to preprocess the output data

the fault frequency of signal as the responding signal of a systemmodel is extracted.Secondly

the time series model is built by using high order cumulants of eliminating and attenuating characteristics of Gaussian noise.Finally

the ARMA bispectrum qualitatively analysis isobtained according to the coefficients of the model.In the meanwhile

the quantitative criteria is obtained by using quantum self-organizing feature map neural network.The results show that

the method can provide some valuable conclusions for presence and type of the fault diagnosis of the gearbox.