Research On On-line Condition Monitoring And Fault Warning Methods For Centrifugal Compressor Blade

Centrifugal compressors have been widely used in petrochemical,aerospace and other fields,which play an important role in heavy industry system.The blade is the core working part of a centrifugal compressor,which bears the comprehensive effect of fluid load,centrifugal force,and so on during the working process.With the continuous development of modern industry,compressors often need to work in complex and changeable conditions,which can easily lead to blade cracks,fractures and other faults.Once the blade fails,it will not only bring huge economic losses,but also bring great threats to the lives of relevant personnel.Therefore,it is urgent to carry out relevant research on online monitoring of blade status and warning work of blade fault to prevent accidents,so as to provide guarantee for long-term reliable operation of compressors.The research ideas and framework are arranged according to the changing trend of blade health status.Firstly,the characteristics of unstable flow field fluctuation affecting blade state are revealed.Then,considering that the blade will vibrate under the effect of flow field fluctuation load and other factors,the on-line monitoring methods for blade state is studied.Finally,aiming at the inevitable fatigue failure of blades under long-term abnormal vibration,further research on fault warning methods of blades is carried out.The main work of this paper includes:The unstable flow inside the centrifugal compressor has great influence on blade state,however,the current experimental study on the mechanism of flow pulsation is insufficient.Aiming at this problem,an experimental analysis method based on cyclostationary theory is proposed to study the flow field pulsation characteristics.Firstly,the theory of cyclostationary is expounded.Then,fast spectral correlation based fast implementation algorithm of cyclostationary theory is deeply studied.Finally,several flow conditions from large flow rate to near surge flow rate are designed,and experiments are carried out to collect pressure pulsation signals of flow field under different conditions.The proposed algorithm is used to analyze experimental signal,which effectively reveals the change process of unstable fluctuation characteristics of flow field.To meet the requirement of on-line monitoring of blade vibration during the operation of centrifugal compressor,an error model is established to analyze the vibration measurement error of traditional blade tip timing(BTT)technique and an improved BTT method is proposed accordingly.The establishment of error analysis model is based on basic test principle of BTT technique.And the model is used to analyze the influence of various unstable factors on vibration measurement accuracy of traditional BTT.According to the analysis results,it is pointed out that for large centrifugal compressors,sensor vibration will have great influence on the accuracy of measurement results.To ensure the validity of test results,the improved BTT method using mean timing signal as key-phase reference is proposed.Theoretical analysis and experiment prove the superiority of the improved method,which can effectively reflect the real vibration information of blades.Aiming at fault identification of blades,pressure fluctuation signal is considered for monitoring blade status.Two weak feature identification methods based on stochastic resonance(SR)theory are proposed to extract fault characteristic frequency from experimental data.Firstly,a multi-scale noise tuning adaptive bistable SR method is proposed.This method decomposes target signal with empirical mode decomposition(EMD).Then,selecting sensitive eigenmode components and modifying their coefficients to construct multi-scale noise.Genetic algorithm is used to search for optimal parameter values of SR system,so as to obtain the output signal with optimal signal-to-noise ratio.Furthermore,a method based on the combination of continuous wavelet transform(CWT)and SR with improved potential function is proposed.The basic idea of this method is to obtain time-scale spectrum by processing target signal with CWT.Selecting specific scales to reconstruct time domain signal,which realizes the filtering pretreatment.Combine Woods-Saxon and Gaussian potential as the new potential function of SR system,so as to improve its characteristic enhancement effect.The two methods are used to process single channel pressure pulsation data,and the weak fault characteristic frequency can be effectively identified.In the condition that there is no prior knowledge for blade fault,a weak feature identification method based on two-channel sparse blind source separation is proposed to analyze pressure pulsation signal first for the purpose of fusing data information from different sensors.On this basis,a fault warning method combining pressure pulsation signal and tip timing signal is proposed.The abnormal frequency component extracted from pressure pulsation signal is used as frequency domain index,and abnormal amplitude ratio extracted from tip timing signal is used as time domain index for joint diagnosis.Only the two indexes are satisfied at the same time,warning work is conducted.Then,the faulted blade can be determined by searching for maximum amplitude point.Experimental results demonstrate the effectiveness of proposed fault warning method.