Research On Wind Turbine Gearbox Fault Diagnosis Method Based On Single Parameter EEMD And Adaptive Feature Selection Strategy

Under the direction of the ’Dual Carbon’ and ’14th Five-Year Plan’ policies,the wind energy sector of China is in the midst of a large-scale,high-quality growth phase.As a crucial part of wind turbines,the wind turbine gearbox is prone to failure from long-term working in challenging conditions like erratic wind speeds.Gearbox failures will result in long maintenance downtime,which reduces wind energy output and increases the risk of potential accidents as well as certain financial losses.Therefore,intelligent diagnosis and online diagnosis of wind turbine gearboxes is an inevitable requirement to ensure the sustainable development of the wind power industry.In order to accurately and effectively discover and identify the fault characteristics of wind turbine gearboxes,it is necessary to use a non-stationary signal processing method with noise processing capability to extract features from the health state signals of wind turbine gearboxes,and then deep learning techniques to learn and identify their health states.This technical system,however,suffers from the issue of insufficient adaptivity in fault diagnosis,particularly online diagnosis,which is primarily manifested in the challenge of adaptive determination of parameter selection methods for specific health state signals and the lack of precise adaptive screening of decomposition result feature components.Such non-adaptive character will directly impact the reliability and accuracy of health state identification in fault diagnosis,particularly online diagnosis.Therefore,the thesis investigates the adaptive noise addition of EEMD(Ensemble Empirical Mode Decomposition)decomposition process and the adaptive accurate selection of decomposition result feature components respectively,proposes an adaptive fault diagnosis strategy based on deep learning and establishes a lightweight diagnosis model to achieve a highly accurate and portable online diagnosis of wind turbine gearbox online diagnosis.The main research and work of the thesis are as follows:(1)A single-parameter EEMD method with better adaptivity is proposed.To address the problem that EEMD is difficult to obtain accurate decomposition results under different working conditions and health state signals adaptively,based on the review and discussion of existing solutions to the parameter selection problem,beginning with the principle that EEMD introduces noise to improve the signal extreme point distribution,the noise addition effect evaluation index is constructed,and the random noise generation and screening mechanism and spEEMD(single parameter EEMD)method.The proposed method transforms the noise with a certain amplitude into noise with random amplitude to improve the signal extreme point distribution,thus making the spEEMD more adaptive in practical applications.At the same time,the stability of the spEEMD decomposition result is also ensured by adding dynamic indicators and selecting screening methods.Applying spEEMD to the decomposition of simulation signals and wind turbine gearbox fault signals,it is verified that the proposed spEEMD method has greater adaptiveness in feature extraction.(2)A two-step selection method of health status feature components and an adaptive intelligent diagnosis strategy based on fusion layer are proposed.To address the problem that the selection of IMF(Intrinsic Mode Function)components obtained from EEMD decomposition depend on prior knowledge and subjective judgment,a two-step selection method and an intelligent diagnosis strategy based on the fusion layer are proposed in order to realize the adaptive and accurate selection of feature components in the online diagnosis process.Firstly,the power spectral density(PSD)is used to select and merge the components with similar features to avoid the influence of mode mixing and redundant irrelevant components on feature component evaluation;then the gray correlation analysis is used as the skeleton to organically combine the multidimensional feature assessment indexes and screen the health status feature components and feature-interference co-mingled components;finally,the fusion layer-based fault intelligent diagnosis method is proposed to adaptively realize feature extraction of health status.The comparison results with seven other selection methods demonstrate that the proposed two-step selection method and adaptive fault diagnosis strategy can adaptively select and extract more accurate health status features.(3)A Lightweigh model and method based on frequency domain transform and large size convolution kernel are proposed.To address the problems of complex structure,large parameters number,and high requirements for deployment platform of diagnostic models,a lightweight adaptive online diagnostic method and model for wind turbine gearbox is proposed,which improves the feature recognition effect of the model and reduces the number of parameters of the model by adopting frequency domain feature transformation and large size convolution kernel to realize the lightweight model.The examination of wind turbine offline data and simulated online data revealed that the established lightweight model has a stronger feature recognition effect and requires fewer computing resources,which is suitable for online fault diagnosis of wind turbine gearboxes.