Abstract:A composite fault diagnosis approach is proposed, combining the real-valued negative selection algorithm and the support vector machine, after researching the shortcoming about the conventional classification algorithm in the fault diagnosis. In the new method, the RNS algorithm is used to generate the detector (non-self) as the unknown fault samples, which are used as input to SVM algorithm for training purpose. The difficult problem, lacking the training samples, is solved using the new method on the conventional classification algorithm. When the axial piston pump have fault because vibration signal generated from slipper/swashplate impact is modulated by high frequency resonance signal, the modulation signal has to be demodulated using the wavelet cluster envelope demodulation method. Where after, the envelope signal is decomposed using wavelet packet and signal eigenvectors are extracted. At last, the loose slipper and the wearing valve plate fault samples of axial piston pump are tested using the composite approach. The classification right rate of this method reaches to 90%, so the composite approach is valid for the fault diagnosis.