A Study On Vibration Signals Of Hydropower Unit Based On Symmetrized Dot Pattern

As China is abundant in hydro resources,the hydro units play an important role in its grid like load and frequency modulations and emergency reserve.Therefore,it is of great significance to ensure its safety and stability.And fault diagnosis is vital for the purposes of equipment optimization,loss decreasing and service life increasing.By a comprehensive consideration of safety,stability and economic efficiency,the conditionbased maintenance is the ideal method which includes condition monitoring,fault diagnosis and maintenance instruction.The fault diagnosis in condition-based maintenance method can be divided into four stages: signal acquisition,signal processing,feature extraction and status identification.This paper extracts features from signals after signal processing and use the features to recognize different working conditions.For signal processing,variational mode decomposition is combined with wavelet threshold for denoising.The sample entropy is used to calculate IMFs to divide them by noise levels so that the wavelet is applied only on the high-noised part to avoid useful information loss.Comparing of the waveforms of signals before and after denoising and the signal to noise ratio both confirms its availability.After denoising,the signal is decomposed again by VMD and each IMFs’ variance is calculated as a feature.For each signal,its IMFs’ variances are packaged into a vector as the input of back propagation neural network for status recognition of four different working conditions of rotor acquired from a rotor test bed.Comparing to EMD-WT and WT,this method shows a superiority in both denoising and recognition.However,this variance-formed vector is too simple and cannot show the differences among different states distinctively,considering this,this paper introduces a feature extraction method based on images called the symmetrized dot pattern method,and proposes an SDP-BP diagnosis method for states identification.The denoised signal is transformed into a symmetrical picture in polar coordinates according to a corresponding formula and 6 eigenvalues in four different directions can be extracted from its grey scale image matrix,that is,totally 24 eigenvalues.To obtain the best parameters in signal-to-image transformation,the correlation between different images is calculated so that the images can present the differences among states to the maximum degree.The recognition result proves its superiority comparing with the former method.The four different working conditions of rotor acquired from a rotor test bed used in the former state identification are established conditions and its signals are steady,but practically it often needs to give diagnosis in the early development period of fault to reduce losses.Therefore,this paper also makes a fault diagnosis to the different stages of cracks on turbine impeller blades,which furtherly proved the validity of SDPBP diagnosis method.