Research Of Rotor Fault Measure Based On DSP And Wavelet Analysis

The rotor fault measure is very important as the core components of rotating machinery. It is a comprehensive multi-disciplinary technology. Signals acquisition, processing and features extraction are necessary for rotor fault measure. It is mainly dependent on the rotor vibration signals, which contain plenty of informations generated in the running rotor. Denoise of rotor vibration signal must be carried out due to noises such as mechanical equipment periodic vibration and other interference signals. In the paper, selecting the rotor as the research object, taking typical fault of rotor as example, the signals acquisition, processing and features extraction were carried on.Accordingly the research was carried out as follow:A signal acquisition system was designed based on DSP TMS320F2812processor chip. TMS320F2812and AD7606-4ADC functional overviews and GPIO various modes of operation were described, The connection methods as well as functions of them, and the sampling system software functions were introduced in detail. Timing diagram of AD7606Simultaneous Sampling on Channel-Parallel Mode were shown. The CPU Timer0interrupt was adopted to start the A/D conversion device. The sampling frequency was changed by change the values of the timer divide-down register and prescale counter register.Bayes sample estimation method was adopted to estimate wavelet threshold, the denoised signal was obtained via processing each layer wavelet decomposition detail coefficients by using the soft threshold function and reconstructing the coefficients layer by layer, then a new practical and adaptive rotor denoise way came out. Comparing it with the results of Donoho threshold algorithm, Penalty threshold algorithm, Birge-Massart threshold algorithm showed that the signals disposed via the method mentioned above were better to retain some details of the original signals, and the SNR was improved significantly.Feature extraction is an important part of the rotor measure. Two feature vector extraction methods were studied which were the wavelet frequency band energy and wavelet envelope.The formula of calculating the frequency range which signals were decomposed via Mallat algorithm was proved on the basis of the simulation. The rotor vibration signals were decomposed by wavelet, then the energy of the frequency band was calculated and normalized as the feature vectors. The fault feature frequency values and their frequency spectrum were extracted to obtain the feature in the fault frequency band of wavelet decomposition signals via Hilbert envelope spectrum. The rotor fault was diagnosed according to the feature vectors.