The application of blind source separation (BSS) to the mechanical vib ration signal processing provides a new technique for mechanical fault diagnosi s. Mechanical vibration signals in practice can be viewed as sums of differently convolved source. For this characteristic, based on the convolution model and b asic theories of blind deconvolution (BD), a BD method online real time proc es sing technique using the extended HJ network framework was proposed. The effec t iveness and accuracy of the BD algorithm was verified by the numerical simulatio n data. Applied to the actually measuring data, the method provides more informa tion about the vibration and diagnoses faults more accurately than the conventio nal Fourier transformation techniques.