Research On Fault Detection Method Of Fan Blade Based On Data Analysis

The global economy is developing rapidly,and the economic model of sustainable development continues to stabilize.Green energy such as wind energy has received more and more attention internationally and has also been widely used.The working environment of wind turbines is generally-very harsh,which also makes the failure rate of wind turbines high,and the causes of failures vary widely.Compared with traditional methods,using wind turbine operating data to detect faults has many advantages in terms of implementation process and system stability.However,the use of SCADA data during the operation of wind turbines to detect faults and evaluate their status is still facing challenges in field applications.In this paper,based on the application of SCADA data of wind turbine operating process in wind turbine blades,the following research contents are carried out:First,the random forest model is used to perform feature reduction on the SCADA data features that affect the icing of wind turbine blades,and then the power feature is converted into the power mean square error feature in the KNN algorithm to achieve feature enhancement.Combine the reduced features of the random forest and the power mean square error feature enhanced by the KNN method as the input of the fully connected neural network(FCNN).The simulation is performed based on the operating data of wind turbines in Yunnan Province,and use multiple methods to compare with the blade icing diagnosis results obtained in this paper.Secondly,through the vine-Copula model,the correlation analysis between the various state parameters in SCADA and the blade icing state of the wind turbines is carried out,and the characteristic variables,root nodes and relationship numbers are determined,then a high-dimensional vine-Copula structure is constructed.After removing the features that are not directly related to the blade icing state,the final vine-Copula model is obtained.Use the filtered features as the input of the LSTM-Autoencoding algorithm,and use the"memory" function and non-linear feature extraction capabilities of the LSTM-Autoencoding algorithm to obtain the evaluation results of the icing state of the fan blades,and compare them with various methods analysis.Finally,this chapter proposes an algorithm that uses adaptive stochastic resonance and variational mode decomposition(VMD)to modal decompose abnormal sound signals and extract multi-layer features.First,the ant colony optimization adaptive stochastic resonance method is used to extract the weak periodic characteristic signal from the abnormal sound signal,and then the abnormal sound signal of the fan blade is decomposed by VMD to obtain multiple modal functions,and the Mel frequency inversion of each modal function is calculated separately.Feature indicators such as spectral coefficients,and normalize them to form multi-layer features.The extracted features are used as input,and a deep convolutional neural network is used to recognize and classify abnormal sounds.