Steam Turbine Condition Monitoring Based On Big Data

Steam turbines are a type of rotary steam power unit with a wide range of applications.They are often used in the main equipment of modern power plant generator sets,and at the same time,there are a large number of applications in the field of metallurgical,chemical,and naval ship generator set.As a common rotary machine,turbine units often encounter abnormal operating conditions or even major failures that result in the forced shutdown of the entire plant.Causing a lot of economic losses,and also pose a threat to the safety of workers.Therefore,equipment often requires condition monitoring to ensure the normal operation of the unit,but also requires periodic maintenance and maintenance to avoid damaging accidents,and the development of condition monitoring and abnormal analysis technology provides a more effective and more versatile method for real-time monitoring and predictive maintenance of rotating machinery.Condition monitoring technology has become an important tool for preventing potential equipment failures,reducing unplanned forced outages,further reducing maintenance costs,and improving the reliability,availability,and maintainability of rotating machinery systems.This article describes in detail the development status of the mechanical system condition monitoring technology at home and abroad.It can be seen that for the complex structure and special operating environment of steam turbine units,as well as different types of equipment and various application conditions,the signal processing technology and the prediction algorithm play a key role in the condition monitoring of rotating machinery.This study takes into account the uncertainties and multivariate correlations of the measured data of the mechanical system.The multivariate time series data collected from different locations of the rotating machine are used as research objects.A probabilistic signal processing method and a data predictive model based on artificial neural network are introduced and used.In addition,these methods have been established and tested examples of abnormal condition monitoring and analysis algorithms for turbine generator sets.The established signal processing algorithm integrates three advanced data mining techniques: Discrete Wavelet Packet Transform,Bayesian Hypothesis Test and Probabilistic Principal Component Analysis.Based on this algorithm,the reconstructed data is used to establish a fuzzy neural network model,for equipment future data through predictive analysis.In this paper,the following work is done for turbine generator condition monitoring based on data mining algorithms:Firstly,the multi-resolution wavelet analysis method is used to decompose the time-series signal into wavelet coefficients of different levels.The decomposed coefficients represent multiple time-frequency resolutions of the signal.Then the Bayesian hypothesis test theory is used to eliminate the defect data of wavelet coefficients at each level for avoiding the problem of excessive noise removal.And then used the power spectral density of the Welch method to evaluate the effectiveness of the Bayesian Wavelet method.Next,using the probabilistic principal component analysis,reducing the dimension of the multi-dimensional time series data,solving the multivariable correlation and uncertainty of data.Afterwards,using the established algorithm and process framework,an example analysis based on actual operating data including 21-dimensional variable of the steam turbine was conducted.In addition,a dynamic fuzzy neural network model based on Nonlinear Auto-Regressive Moving Average method is constructed.The fuzzy clustering algorithm is integrated into radial basis function neural network,and the natural gradient descent method is used to maximize the log-likelihood function for training.Similarly,the proposed model and program are validated by the time series data of the 21-dimensional variables of the steam turbine.