Research On Equipment Fault Prediction Method Based On Full Vector Spectrum

During the operation of the mechanical equipment,various types of faults will occur due to different reasons.These faults will affect the operation at times and cause serious accidents with enormous economic losses.The correct fault prediction can remind the equipment managers to take measures in advance to avoid accidents and provide technical guarantee for safe and efficient operation of the equipment.The current prediction researches still focus on the numerical prediction of a single parameter,and cannot make sound judgments on the nature and location of possible faults through prediction.Aiming at overcoming the shortcomings in the existing fault prediction,the fault prediction method based on the full vector spectrum was studied from the aspects of information fusion,data acquisition,predictive modeling and product development.The research solves the problem that the traditional prediction method cannot predict the nature and location of the fault.The use of vibration big data for early pre-warning and evaluation of equipment health provides a basic guarantee for the promotion and application of prediction technology.The main research work and achievements made in this dissertation are as follows:(1)The method of equipment fault prediction based on full vector information fusion is studied.One-dimensional full vector vibration magnitude prediction model(FVMP),two-dimensional full vector spectrum structure prediction model(FVSP)and equipment failure multidimensional prediction model(FVEP)are elaborated and constructed.The prediction model established on the basis of analyzing the fundamental principles,methods and numerical calculations of the full vector spectrum technology can address the problem that the vibration characteristics of the equipment reflected by the data of traditional testing methods are not comprehensive enough.The prediction model can predict the changes of the frequency spectrum structure based on the vibration signal after the full vector information fusion,judge the nature of the fault and evaluate the health status of the equipment.(2)The method of full vector data acquisition and feature extraction is studied.The acquisition circuit and sampling control strategy to support full vector data acquisition are analyzed.Using the vibration data of the power plant equipment,the correctness and reliability of the full vector collection are verified.The concept of virtual sensor is proposed and defined.The virtual sensor is a logically existing sensor whose output is the result of the fusion of two homogenous sensor signals.The research shows that the diagnosis results based on the single sensor data in the same section of the rotor are one-sided and the data after the full vector fusion can reflect the running status of the equipment more comprehensively.The fusion vibration spectrum structure and characteristic values are unique and can be used for vibration intensity and spectral structure predictions.(3)The full vector autoregressive time series prediction model(FAR)is studied.The FAR prediction model combines the technique of full vector spectrum information fusion with the method of AR model analysis.Theoretical research and data validation were performed on modeling steps,model identification,parameter estimation and recursive calculation.FAR model can be used for trend prediction of stationary time series data,and it also provides an effective method for prediction of vibration magnitude,fault property and equipment status.However,for the non-stationary operation process,the use of AR prediction has some limitations because of the relatively large prediction error.(4)The full vector grey prediction model(FGM)is studied.The FGM prediction model is based on the full vector spectrum homology information fusion.Combining with the grey theory,it not only guarantees the reliability of the prediction data,but also realizes the ability to predict the trend data of small samples.Full vector grey models include full vector GM(1,1)and full vector MGM(1,m).The full vector MGM(1,m)is the extension of the GM(1,1)model under the m variable.The research shows that the full vector grey model has a good prediction effect on "small sample" and monotonic growth sequence,and can predict vibration amplitude and spectrum structure.(5)The comprehensive prediction model(FVCP)and the big data early-warning based on the full vector spectrum are studied.The full vector comprehensive prediction model organically combines the full vector FAR model with the full vector FGM model.Using empirical mode decomposition(EMD),the random and trend terms in the data sequence are separated,fed into the FAR model and FGM model respectively for prediction,and the results are organically integrated.The FVCP model makes full use of the advantages of the two numerical prediction models.Research shows that the FVCP full-vector comprehensive prediction model takes into account the characteristics of randomness and tendency of vibration signals,and has better applicability.The full-vector spectrum based big data early warning extracts vibration characteristics from massive historical data and predicts the trend of these characteristics.The method combines other process parameters such as temperature and current in the process of equipment operation.Based on fuzzy production rules,device health evaluation and fault early pre-warning based on big data can be used to better guide maintenance decisions.(6)The research on the full-vector fault prediction technology is summarized,and the related software and hardware products based on the full vector fault prediction are developed and used in the engineering practice.The SDC series intelligent data collector is developed,and its acquisition strategy follows the requirements of the full vector spectrum analysis technology,and supports the information fusion based on full vector spectrum technology.The intelligent early pre-warning function module based on full vector prediction is developed,which uses homologous information fusion technology to improve signal comprehensiveness,and also provides reliability for evaluation and diagnosis in prediction domain.The equipment state health optimization and analysis platform based on big data pre-warning for equipment is able to predict and evaluate the health status of equipment well,provide decision-making basis for equipment maintenance,and verify the correctness and practicability of theoretical research.