Study On Detection And Feature Extraction Of Weak Fault Signal Of Mechanical System

The crucial issue in mechanical fault diagnosis field is fault feature extraction, while a noise disturbance is an important factor which affects fault extraction. The abnormal signal, generated by early fault, is usually weak and covered with strong noise. Therefore a high detection technique is requested. Based on vibration signal features, this paper researches a series of methods for signal detection and extraction, which is combined with the theories of modern signal process. And at the same time, lots of simulation and test work are done.The meaning of this paper and the method for weak signal detection are first described. Then the usually ways to extract fault and its developing are introduced.Considering non-stationary vibration signal in complicated mechanical equipment, Time-Frequency analysis is employed to detect and extract fault. The results of simulation and test in gearwheel fault diagnosis prove that Time-Frequency analysis can reduce noise and detect weak abnormal signal effectively by designing reasonable kernel function.Wavelet basis function impact on signal feature extraction is discussed. According to singularity detection theory, a method which uses the multiply of more than three-stage specific signals as detection signal to get the multi-scale modulus maxima map is brought about in order to detect singularity point in noise and get the location of singularity point. A method based on wavelet transform is studied and applied to extract weak abnormal vibration signal.At last, used its good performance in suppressing Gaussian noise, high-order statistics is studied as a method for fault signal detection. More emphasis is placed on bispectrum, which has been applied into the fault signal detection of gearwheel.