Research On Weak Signature Extraction Methods For Machinery Fault Diagnosis

With the development of modern industrial technology, mechanical systems become more and more sophisticated. Developing advanced maintenance technologies for them become an urgent issue. Bearing and gears are the most common components of mechanical systems, and a small defect in them would cause the failure of the whole system, even catastrophic accident. Therefore, design effective methods for gears and bearings health monitoring, especially incipient fault detection is of great significance in the reduction of system down time and production losses. Targeting with this object, this paper conducts researches in the development of advanced signal processing algorithms for machine incipient fault feature detection.First, this paper proposes a novel frequency demodulation algorithm, where signals are first processed by autocorrelation analysis to eliminate random noise, and then implemented envelope demodulation to extract original modulation frequencies. Since correlation is powerful tool for random noise elimination, this method is especially effective in frequency demodulation under low signal-to-noise circumstance.In addition, this paper proposed novel gearbox fault diagnostic approaches based on Hilbert-Huang transform. To solve the problem that the empirical mode decomposition (EMD) is prone to be affected by noise, two schemes are proposed in this paper. In the first scheme, delayed-autocorrelation is applied prior to the EMD, so that some random noises can be removed and the performance of EMD would be improved greatly. In the second scheme, the Hilbert time-frequency matrix is realigned, in order to obtain a clearer time-frequency representation with better physical meaning, at the price of sacrificing some frequency resolution.Besides, to extract fault-generated weak impulses from signals under strong noises and interferences, an effective algorithm based on optimal Morlet wavelet filter is developed. In this algorithm, the optimal wavelet filter is constructed by differential evolution using a kurtosis maximum rule. After wavelet filtering, the sparse code shrinkage is employed to eliminate the noises that are in the pass-band of wavelet filter. Simulation study and experiment validation proved the effectiveness of the proposed method.Finally, this paper introduces an gearbox fault detection and diagnostic approach based mathematical morphology. This method uses the morphological closing operation with a flat structuring element (SE) to extract impulsive features from vibration signals. To optimize the flat SE, firstly, a theoretical study is carried out to investigate the effects of the length of flat SEs. Then, an adaptive algorithm for the flat SE optimization is proposed based on the theoretical findings. The proposed method is tested by the simulated and bearing vibration signals. The test results show that this method is very effective and robust in extracting impulsive features.