Research On Condition Monitoring And Early Fault Warning Of Coal Mill In Thermal Power Unit

With the increasing scale of grid integration of renewable energy,traditional thermal power units as flexible power sources need to frequently participate in grid peaking and frequency regulation.This makes the units often operate under variable operating conditions,and the correlation between the main and auxiliary equipment may lead to potential failures of the units.With its complex structure,harsh working environment and long-term high temperature and high pressure variable load operation,the coal mill has a high failure rate among the auxiliary equipment of thermal power units.When the status parameters of the coal mill exceed the limit alarm,the failure of the coal mill has often developed to a certain extent.If an abnormality can be detected at the early stage of a coal mill fault,the operating personnel will have enough time to maintain the mill,avoiding further deterioration of the fault and resulting in an abnormal shutdown of the unit.The safe and stable operation of the coal mill directly affects the stability and economy of the boiler combustion,which in turn affects the safety and reliability of the entire power production.Therefore,the research on condition monitoring and early fault warning of coal mills in thermal power units has important academic significance and application value,which can reduce the failure rate and operation and maintenance costs of coal mills and improve the safe and stable operation of thermal power units.With the widespread use of Distributed Control System(DCS)in thermal power plants,a large amount of coal mill operation data is stored in the DCS,which provides the data basis for a data-driven modelling approach.Since most of the coal mill operation data is normal state data and fault samples are scarce,how to make use of the large amount of normal operation data to achieve condition monitoring of coal mills and early warning of faults is the main problem to be solved in this paper.This paper takes the MPS-type medium-speed coal mill as the research object,and draws on the latest theoretical techniques in the field of deep learning and data mining to study a deep learning model based on the autoencoder theory to realise intelligent condition monitoring and early fault warning of the coal mill.The main research elements are as follows:(1)Aiming at the high-dimensional nonlinear characteristics of coal mill operation data,a deep feature extraction method based on Stacked Autoencoder(SAE)is studied based on the analysis of its operation characteristics,and compared with other commonly used feature extraction methods.The two-dimensional scatter plot is used to characterise the ability of SAE to extract deep-level features from coal mill operation data,and to lay the foundation for the establishment of a multi-dimensional data reconstruction model for coal mills.(2)To address the characteristics of coal mill operation data containing noise,a Stacked Denoising Auto-Encoder(SDAE)based multivariate data reconstruction model for coal mills is established to extract the deep-level robust features of multivariate data layer by layer and to fully learn the non-linear mapping relationship between data and operation status.By analysing the residuals of the multivariate data reconstruction model,the monitoring of the coal mill operating condition is achieved.The validity of the proposed method is verified by the actual operation data of the coal mill.The results show that the proposed multivariate data reconstruction model has high accuracy and can be used for monitoring the operation status of the coal mill.(3)To address the problem that most of the existing studies do not consider the temporal correlation of coal mill operation data,the LSTM-based early fault warning model for coal mills is proposed by replacing ordinary neurons in the autoencoder with LSTM neurons using the processing capability of Long Short-Term Memory(LSTM)for temporal data.For the flexible and variable operating conditions of coal mills,an adaptive dynamic threshold strategy based on Chebyshev theory is proposed,which overcomes the shortcomings of the traditional fixed threshold detection method with high false alarm rate and poor adaptability.Through the simulation of coal mill blockage abnormalities,it is shown that the output of the fault warning model can give early warning of actual faults,which can further advance the early fault warning time compared with the traditional fixed threshold method.(4)In view of the lack of fault samples in the actual coal mill operation data and the lack of abundant samples in the data set,the method of generating fault data using the coal mill mechanism model is used to establish an SAE-based coal mill fault diagnosis model.In order to meet the practical needs of coal mill fault diagnosis under variable operating conditions of thermal power units,the SAE fault diagnosis model introduces the idea of migration learning to achieve variable operating conditions fault diagnosis of coal mills.When the operating conditions of the unit change,the fault diagnosis model does not need to be retrained,but only needs to migrate and fine-tune the parameters on the basis of the pre-trained model to achieve satisfactory fault diagnosis results.Considering the actual application requirements of the project,the actual coal mill blockage fault data is used to complete the coal mill migration fault diagnosis,and the model diagnosis results are consistent with the occurrence mechanism of coal mill blockage slow change fault.This study aims to explore the problem of intelligent condition monitoring and early fault warning based on coal mill operation data.The research results can provide some scientific basis for the intelligent operation and maintenance of coal mills,and provide new research methods and ideas for the monitoring and evaluation of the operating conditions of other large power equipment.