Research On Real-Time Fault Warning Methods For Wind Turbines Based On Flink Platform

With the rapid growth of installed wind turbines in China,the wind power industry is developing rapidly.However,accelerating the advancement of the wind power industry also brings many challenges.One of the most prominent problems is the serious economic losses caused by frequent downtime due to wind t urbine failures.Therefore,it is especially important to study the fault diagnosis system,but most of the traditional fault diagnosis analysis is offline,which is difficult to perform fault warning in real time and does not meet the requirement of real-time fault warning for wind turbines.The paper selects gearbox as the main research component,combines offline and real-time data analysis methods to study and process gearbox vibration data,and finally builds a fault diagnosis model of gearbox based on o ffline data,and realizes real-time fault warning of wind turbine with the help of real-time computing platform.To address the problem of optimizing the selection of the parameters of the LSSVM algorithm,a method combining artificial fish swarm(AFSA)an d particle swarm optimization(PSO)hybrid intelligence algorithms is proposed to optimize the parameters of LSSVM.The AFSA algorithm is used to perform the global search for the initial value of the parameters,and the PSO algorithm is used to update the optimal solution locally and accelerate the jump out of the local optimum.Finally,the fault diagnosis models of PSO-LSSVM,AFSA-LSSVM and AFSA-PSO-LSSVM were established by simulating the wind turbine gearbox vibration acceleration data.The experimental results show that AFSA-PSO-LSSVM has faster convergence and higher accuracy compared with PSO-LSSVM and AFSA-LSSVM models,which verifies the effectiveness of this method.Finally,we built a real-time fault diagnosis and early warning platform for wind turbines by combining Flink streaming computing with Kafka,a high-throughput messaging system.The platform implements real-time processing of gearbox vibration data,including data filtering,data sampling,data windowing and feature engineering.The mean filtering method is used to remove a large amount of anisotropic data from the vibration data,and the PAA algorithm is implemented to compress the vibration data based on the Flink platform.However,direct feature extraction is not possible for streaming data,and windowing operations need to be performed on the data.The HHT transform based on Flink sliding window is mainly implemented in feature engineering,which contains EMD decomposition and Hilbert transform.The experiments show that the accuracy of all three hybrid intelligent algorithm models for gearbox fault diagnosis under the Flink platform can reach more than 90%,and the overall delay can be stabilized within 1500 ms,which meets the requirements of real-time gearbox fault warning and verifies the effectiveness and feasibility of the wind turbine gearbox real-time fault warning system based on the Flink platform.