Research On Fault Diagnosis Method For Wind Turbine Gearbox Based On Deep Residual Shrinkage Network

In order to alleviate the contradiction between human social development,environmental pollution,and energy scarcity,wind energy has emerged as a clean and renewable energy application.In this context,the wind power industry,as the most promising technology industry for commercial development,is booming.With the increasing capacity of wind turbines and the scale of wind farm operation,the fault identification and operation maintenance of wind turbines face significant challenges.The statistical results show that the frequency of wind turbine gearbox failures is high,and the downtime caused by failures is the longest and the economic losses are enormous.Therefore,how to achieve convenient and accurate fault diagnosis of wind turbine gearbox is one of the research hotspots in the wind power industry.At the same time,the complex structure and non-stationary operating conditions of wind power gearboxes have brought great challenges to traditional fault diagnosis methods for gearboxes.With the advent of the era of industrial big data,the massive data generated by mechanical equipment greatly limits the application of traditional signal processing and shallow machine learning technology in fault diagnosis.Therefore,there is an urgent need to propose a new intelligent method for fault diagnosis of wind turbine gearbox under variable operating conditions.The main research content of this article is as follows:Firstly,the basic structure and working principle of wind turbine are deeply studied,and the typical faults of gearbox are emphatically described.On this theoretical basis,in view of the existing experimental data that is difficult to meet the off-load data requirements of the wind power gearbox fault diagnosis algorithm under off-load conditions,referring to the collection process of the open dataset,a typical gearbox fault simulation experimental platform was established,and nine different working conditions of gearbox fault data were designed and collected,providing data support for subsequent research.Secondly,for fault diagnosis of wind power gearbox under variable operating conditions,a fault diagnosis method based on reconstructed two-dimensional data and deep residual shrinkage network is proposed.This method reconstructs one-dimensional data into a two-dimensional feature map arranged in the form of a square matrix,greatly simplifying the workload of data processing,effectively preserving the timing and correlation of one-dimensional original vibration signals,and obtaining two-dimensional features consistent with model inputs.At the same time,based on the deep learning theory,a deep residual shrinkage network model is constructed,which uses its powerful feature extraction ability to learn fault features from vibration signals,and uses its pattern recognition ability to classify.This solves the problem of separation between feature extraction and classification evaluation,which is difficult to optimize as a whole.The effectiveness of this method under variable operating conditions is verified using the fault data obtained from the gearbox fault diagnosis simulation experimental platform.Finally,a fault diagnosis method based on continuous wavelet transform and deep residual shrinkage network fusion is proposed for wind power gearbox fault diagnosis under variable operating conditions.This method utilizes continuous wavelet transform to transform the original vibration signal into a two-dimensional time-frequency map.The attention mechanism and soft thresholding of deep residual shrinkage networks are utilized to eliminate the impact of redundant information and fully extract the essential features of faults.The effectiveness of this method under variable operating conditions is verified using open dataset data from a university.In addition,the results of noise experiments verify the robustness of the proposed method.This method meets the integrated and intelligent requirements of wind turbine gearbox fault diagnosis,and the robustness and generalization of the model are improved.The research in this paper provides new ideas and methods for solving the practical engineering diagnosis of wind turbine gearboxes,breaks the bottleneck of traditional signal processing and machine learning diagnosis methods,and lays a foundation for subsequent intelligent research on fault diagnosis of wind turbine gearboxes.