Research On Intelligent Fault Diagnosis Method For Large Wind Turbine Gearbox

With the increasing capacity and complexity of wind turbines,a series of problems,such as high failure rate,low reliability,high production and maintenance costs,have seriously affected the survival space of wind power enterprises.In order to realize the steady and rapid development of wind turbines,reduce costs and increase efficiency,the condition monitoring and fault diagnosis technology of wind turbines have become the key technical mean to solve these problems.This paper starts from the original vibration data collected by the gearbox experimental platform and the measured data of the gearbox coming from the wind farm Supervisory Control and Data Acquisition（SCADA）system.Aiming at the current difficulties in extracting fault characteristic signals of wind turbine gearbox and the deficiencies in traditional fault diagnosis methods,combining with the frontier theory in artificial intelligence field,such as Extreme Learning Machine（ELM）and its variant Kernel Extreme Learning Machine（KELM）,deep learning（DP）,multi-sensor information fusion and Pearson related technology,the deep fault feature information hidden in equipment is excavated.It realizes the condition monitoring and fault diagnosis of wind turbine gearbox.The main work is as follows:1)For the problem that the wind farm owner’s confidentiality of the fault data and the different fault modes of the gearbox are difficult to obtain,a dynamic experimental platform for the transmission system of wind power gearbox is established and the original vibration fault data is collected.Aiming at the problem of large amount of information and weak fault characteristics of original vibration data,it adopts time domain analysis method to process the original vibration signal,which plays an important role in improving signal-to-noise ratio and reducing the information dimension of signals.Without the ability to acquire all target information using a single sensor,multi-sensor information fusion based on parallel layer superposition method using decision layer and feature layer fusion is adopted to obtain target information,so as to improve data validity.On this basis,the multi-sensor information fusion model of wind turbine gearbox based on time domain feature analysis is innovatively established.The application and algorithm simulation verify the efficiency and superiority of the model.2)Aiming at the problem of large noise pollution,high complexity and large data volume information of wind turbine fault data,combined with multi-sensor information fusion model of time domain analysis,a new improved ELM method based on Particle Swarm Optimization（PSO）is proposed.Firstly,the multi-sensor information fusion model based on time-domain analysis is used to extract fault feature components,which reduces the dimension of data information and fault diagnosis time.Then,the PSO is used to improve the optimization process of ELM to solve the influence of initial parametershidden layer deviation and input weight random setting on the stability of ELM model.Furthermore,the time domain characteristic index value is taken as the model input parameter,and the fault category is used as the model output parameter.The new fault classification recognition model based on PSO-ELM is established to realize the condition monitoring and evaluation of the wind turbine gearbox.The PSO-ELM method proposed in this paper and the improving ELM methods by Grasshopper Optimization Algorithm（GOA）and Artificial Fish Swarm Algorithm（AFSA）are compared and analyzed.The example application and algorithm simulation verify the advantages and effectiveness of the new PSO-ELM fault diagnosis method proposed in this paper.3)In order to further improve the fault diagnosis rate of wind turbines and better meet the actual needs of engineering,it uses KELM,a variant of ELM,to study the condition monitoring and fault diagnosis of wind turbine gearbox in this paper.Combining with the multi-sensor information fusion model of time domain analysis,a new improved fault diagnosis KELM method based on Cloud Bat Algorithm（CBA）is proposed.Compared with ELM,KELM replaces the matrix H in ELM model with kernel matrix?_ELM,which improves the generalization ability and stability of ELM model.However,the existence of the kernel function will cause the model to be very sensitive to the setting of?parameter and C parameter.In order to solve this problem,this paper combines the CBA method to optimize the key parameters?and C of the KELM model,and establishes a new fault classification and recognition model based on CBA-KELM,which realizes the fault diagnosis of wind turbine gearbox.The CBA-KELM method and the Grey Wolf Optimization（GWO）method for improving KELM are compared and analyzed.The advantagesand practicability of the new CBA-KELM fault diagnosis method proposed in this paper are verified by practical application and algorithm simulation.4)In order to reduce the high cost of equipment deployment,condition monitoring and fault diagnosis for wind turbine gearbox are further studied.In this paper,based on the measured data of wind turbine gearbox coming from the SCADA system,combined with Pearson correlation coefficient,domain expert knowledge and exponential weighted average threshold method,a new improved method of Deep Belief Networks（DBN）based on Whale Optimization Algorithms（WOA）is proposed for fault diagnosis of wind turbine gearbox.In view of the problem that different sample data will have a great impact on the accuracy of model diagnosis,combining Pearson correlation coefficient and domain expert knowledge to select the input condition parameters of fault diagnosis model,the insufficiency of human experience in choosing the condition parameters is improved.Aiming at the problem that the selection of initial parameters will lead to large fluctuation of DBN model,the whale optimization algorithm is used to optimize the initial parameters of deep belief network,and a new fault diagnosis model of wind turbine based on WOA-DBN is established.Based on the reconstruction error,the exponential weighted average threshold method is used to realize the fault of wind turbine gearbox.Case verification and algorithm simulation results show the superiority and effectiveness of the method,and the economic benefit can be as high as800,000 yuan if only the condition monitoring and fault diagnosis of the field gearbox is taken as an example.