Study On Feature Extraction Technique In Fault Signal Of Mechanical System

Mechanical fault diagnosis consists of signal measurement, feature extraction, faultanalysis and diagnosis. The crucial one of which is fault feature extraction. In mechanicalengineering field, rotary mechanism is widely used and hence it is the pivot of conditionmonitoring and fault diagnosis. This thesis will focus on gearbox drive system and take anintensive study on the fault feature extraction technique of gears and rolling bearings. Itscontributions list as follows:1. Setup a mathematical vibration signal model of gears and rolling bearings by analyzingthe feature of vibration signals caused by typical faults in gearbox. Study how the gearboxtypical fault affects the vibrations displacement, velocity, acceleration and high ordersacceleration signals. Present an improved time-domain average technique to extract thegear vibration signals from the general vibrations signals of gearbox. In this way, thesignals reflecting gear faults can be effectively separated. Also analyze and compare therelative metrics of the applications of different methods such as envelop-thinningdemodulation and cepstrum analysis technique in fault feature extraction.2. According to the analysis of how wavelet basis function impacts on signal featureextraction, bring about a new enveloping algorithm on basis of complex morlet waveletand wavelet transform. The new approach adopts FFT to compute wavelet coefficient andhence gets the scale envelope spectrum. Using this spectrum, the new algorithm can dothe envelope processing in all time-frequency plane, both impulse fault components andtheir frequency band can be extracted from the signal. This process can be automaticallyachieved without any prior knowledge or manual intervention.3. Give a specific description of singularity detection theory basing on wavelet transform,especially dyadic wavelet transform basing on B-spline and fast algorithm-¨¤ torusalgorithm ,and the theory of signal singularity detection basing on modulus maxima ofwavelet transform. According to the modulus maxima multi-scale transform character ofsignal and noise, a method which use the multiply of the adjacent two-stage specificsignals as detection signal to get the multi-scale modulus maxima map is brought about.The new technique can effectively enhance signal while restraining noises and can get thelocation of singularity point accurately through the location of modulus maxima. And ithas been successfully applied into the fault signal singularity detection of rolling bearings,as a new fault diagnosis technique, and provides a new way for mechanical faultdiagnosis.4. Study on the application of non-linear time-frequency distribution in fault diagnosis. Todetect transient signal caused by complex mechanical abnormal vibration, presents a novelalias-free exponential bilinear time-frequency distribution, gives its new kernel function,the new method can avoid the frequency aliasing and information loss in traditionalbilinear distributions;In addition, it can reduce the cross-terms effectively and possessvery high time-frequency resolution simultaneously. By digital simulation and gearboxfault signal detection, the effect of bilinear time-frequency transform is validated fortransient signal detection.5. Employ fractal theory to analyze non-stationary vibration signal in complicatedmechanical equipment which the traditional signal analysis approach cannot workeffectively. Bring about a reliable correlation dimension improved computing methodwhich has low amount of calculation and reliable result. Study the relationship betweencorrelation dimension and fault model of rolling bearings, also analyze the affecting factorin course of computing the correlation dimension, The result shows that the size ofcorrelation dimensions directly relate to whether or not fault exists.