Diagnosis And Monitoring Of The Compressor Surge Based On Pressure, Vibration Sound Signals

Gas turbine is playing an increasingly important role in the energy, power, military field. However, because of its poor working conditions, complicated mechanical structure and other factors gas turbines will exist various failures in the running projects. No planned downtime will bring huge economic losses and the extension of time.So it will be significent if we could monitor some abnormal phenomenon before the serious accident. Compressor surge is a common failure in the turbine operation. the compressor surge is a flow oscillation in low-frequency (usually only a few Hz or more than a dozen Hz), high amplitude (the strong pressure and flow fluctuations).When surge occured, the engine sounds boring, and accompanied by very strong mechanical vibration. Therefore monitoring and fault diagnosis of the compressor surge is very important.In this paper,our major research is the compressor surge monitoring and fault diagnosis and compare the different sensors signal of the compressor surge in time domain and frequency domain signals.This compressor surge test-bed is consist of centrifugal compressor, change gear box, composed of three-phase induction motor. In the experiment,we adjust the compressor speed by changing the motor frequency and adjust the compressor working conditions. So that the compressor is closed to compressor surge curve at various operating conditions.In this paper, we select five different conditions of the surge signal for data analysis.In this paper,we use FFT and wavelet analysis as the basic method to analyse the pressure of collection, vibration, acoustic signals that we collect. In this paper, we use FFT on the frequency domain as well as power spectrum, energy spectrum analysis and wavelet transform to the local time-domain signal data noise reduction processing.And we get the conclusion that it is best for the signal analysize by use db8 wavelet.We compare the pressure, vibration, acoustic signals in the compressor surge monitoring and diagnosis.We found that vibration signal for the surge response is not obvious and pressure signal is related to the compressor operating parameters both in the time domain or frequency domain. Moreover the increasing of sound signal in the time domain is more than the pressure signal.But the characteristic frequency of sound signal is not related to the compressor operating parameters.So we choose the sound signal as the diagnosis signal.At last we use the neural networks to establish the diagnostic systems for compressor surge with sound signal. When the compressor is running in normal condition area, the network's output is 1.But when the compressor into surge zone, the network is up to 0 to establish an effective monitoring for the sound signal monitoring of the compressor surge.