| Wind turbine is complicated mechanical and electrical integration equipment,which needs efficient monitoring and maintenance system to ensure its service safety,reliability and economy.The complex service condition and the harsh environment of the wind turbine make it difficult to maintain.The existing fault diagnosis and prediction theory cannot meet the demand that the near-zero fault running in wind power industry.The intelligent maintenance strategy based on equipment state,real-time evaluation and prediction of the operation state of the mechanical equipment can be realized by using the computer technology,and the low-cost and high-efficiency preventive maintenance can be obtainted.However,it needs a large amount of data and data analysis methodology,such as efficient data acquisition,signal feature extraction under complex operating conditions,representation of state performance and intelligent decision-making of maintenance schemes.In this paper,a new model for fault diagnosis and prediction is presented,which integrates physical wind field,mirror wind field and experimental wind field to realize real-time in-loop modeling and simulation.The method for detecting and predicting the abnormal state of the wind turbine under the mutual evaluation and group diagnosis theory is studied.A structural discriminant sparse coding method for the feature extraction of vibration signals is proposed.The state estimation and forecasting system is established by using the deep belief neural network and hidden Markov model.Then,we build a complete intelligent maintenance prototype platform.This study focuses on the key theories and methods of health monitoring and assessment of wind turbine.The main contents are as follows:(1)To figure out the modeled problem of wind turbine,with complex structure,limited structural parameters and coupling of multi-structure interaction,an efficient three-mass equivalent elastic damping model is proposed.In the model,the systems above the tower are equivalent to a single mass block,and the tower is simplified as an Euler beam.The interaction between the foundation and the tower is replaced by the horizontal elastic stiffness and the rotational elastic stiffness.The dancing of the blade effect is regarded as a part of external force.The dynamic equations and boundary conditions of the model are given.The dynamic response characteristics of the system are solved by numerical analysis,and the relationship between the natural frequencies of the system and the stiffness coefficients among the modules is obtained.(2)To solve the problem that the abnormal state of the wind turbine is difficult to be accurately detected and predicted under complex operating conditions,an anomaly detection and prediction method based on mutual evaluation and group diagnosis is proposed.This method evaluates the state of the wind turbine based on the power curve of the wind turbine and the time information of the shutdown event.The quartile unilateral normalized power distribution of the power curve is extracted by using the graded statistical method.A method for calculating the distance among wind turbine shutdown time interval information is proposed.The feature of the wind turbine is classified by the linear mixture self-organizing map neural network for states classification.A feature variation cumulative trend difference method is proposed,which is used to predict the wind turbine’s anomaly and failure.(3)In order to solve the problem of vibration signal feature extraction and performance characterization in mechanical systems,a structured Fisher discriminant sparse coding model is proposed.The structured signal dictionary is obtained by the structured Fisher discriminant sparse coding method.Then the characteristic of the vibration signal under different states is obtained by using the sparse decomposition method of tree structure.The results show that the proposed algorithm has better performance,efficiency and generalization ability compared with the existing algorithms in the fault experiments of bearing and worm gear.(4)To realize the analysis of large data in wind turbine health assessment and prediction,a deep belief neural network Hidden Markov Model(HMM)model based on event self-adaptation is proposed.Using the pre-training depth neural network to express the wind turbine state data,the wind turbine state probability distribution is used as the output target of the learning network,and the hidden-Markov model is used to estimate and predict the wind turbine state.Because the state and the degradation process of the machine will be changed by the maintenance or updating of components,the method of state initialization updating with event self-adaptation is introduced.The classification precision of the health states of the wind turbine is 78.98%,and the precision of single step prediction is 94.55%.(5)In order to improve the current situation that the lack of wind turbine historical accumulation data and difficult in data acquisition,this paper proposes a new method of information synergistic acquisition of physical wind field,mirror wind field and experimental wind field.The Matlab-Simulink platform,the experimental data and the collected data of real wind turbine are used to build the virtual mirror wind turbine,which realizes the information complementation and the acceleration of the historical database completed.The system of wind turbine condition assessment and prediction verification is constructed,and the theoretical system of wind turbine’s fault prediction and health assessment is established,which lays a theoretical and technical foundation for the safe,healthy and stable operation of large-scale wind turbines. |
