Fault Diagnosis Method For Key Components Of Wind Turbines Gearbox Based On Time Series Complexity And Mode Decomposition

Wind energy,as a kind of renewable energy,has attracted more and more attention by virtue of its rich resources,pollution-free,strong industrial competitiveness,and large economic advantages in the later period,and has played an important role in optimizing the energy structure.In order to meet the increasing demand for energy,the installed capacity of wind turbines（WTs）has increased year by year.Gearbox is an important mechanical transmission component of WTs.Its failure will seriously affect the normal operation of WTs.Therefore,ensuring the safe and stable operation of gearbox is crucial.In this paper,fault feature extraction and diagnosis of bearings and gears,the key components of WTs gearbox,are studied.By using time series complexity and mode decomposition technology,vibration signal noise reduction method,fault feature extraction method,fault feature selection method,early fault warning and diagnosis method are studied,and the operation state information mining and fault diagnosis of WTs gearbox are realized.The main research work is as follows:（1）Focused on the disadvantages of existing noise reduction methods such as poor noise reduction effect and insufficient adaptive ability,a vibration signal noise reduction algorithm based on adaptive blind deconvolution decomposition（ABDD）is proposed.Firstly,a Finite-Impulse Response（FIR）filter method is proposed,and a FIR filter bank is constructed through Hamming window to provide direction for decomposition.Then,input the signals preliminarily processed by FIR into the improved cyclostationarity blind deconvolution to get the final filtering mode.Thirdly,the main modes are filtered according to the correlated kurtosis value of the filtering mode in descending order.Finally,the spectrum analysis of the remaining mode is carried out to realize the identification of single fault signal and mixed fault signal.The simulation and experimental data verify that the vibration signal denoising algorithm based on ABDD can effectively reduce the noise and realize the accurate diagnosis of bearing faults.（2）To solve the problems of poor fault feature extraction effect and high time complexity of existing entropy methods,a feature extraction method based on refined composite multi-scale dynamic causality diagram（RCMSDCD）is proposed.Firstly,the generalized entropy（GIE）method is proposed,which can characterize both the static and dynamic complexity of time series and reduce the noise of time series.Then,GIE is combined with Fractional calculus to construct the generalized inverse fractional order entropy（GIFOE）.GIFOE captures the dynamic evolution of complex systems by analyzing small changes in time series.Thirdly,dynamic causality diagram（DCD）is proposed by combining GIFOE and complexity entropy curve.The simulation results show that DCD can effectively represent the dynamic change of fault signals,and has the consistency of entropy and complexity,noise robustness and high computational efficiency.Finally,the single-scale DCD is extended to multiple time scales to extract cross-scale entropy and complexity features.The experimental results show that the feature extraction method based on RCMSDCD has the higher recognition accuracy and the lower time complexity.（3）To solve the problem of low accuracy and efficiency of fault diagnosis caused by feature redundancy in existing entropy-based high-dimensional feature extraction methods,a fault feature dimension selection algorithm based on refine generalized composite multiscale state joint entropy,multi-scale average euclidean divergence（MSAED）and robust spectral feature selection（RSFS）is proposed.Firstly,state joint entropy is proposed to extract static and dynamic fault features.Then,in order to improve the richness and stability of the feature space,the refined generalized composite multi-scale analysis method is proposed.Thirdly,the MSAED algorithm is proposed to adaptively select the parameters of the entropy method for different datasets to address the difficulty of parameter selection for the entropy method.Finally,RSFS is used to select multi-scale features,and a variety of machine learning algorithms are combined to recognize gear faults.The results show that the features selected by RSFS are the best distinguishable when compared with the three feature selection methods:laplace score,fisher score and and max-relevance min-redundancy,and the effective selection of high-dimensional gear fault features is realized.（4）Focused on the problems of delayed early fault warning and difficult identification of early fault characteristics of bearings,a bearing early fault warning method based on anomaly indexAI_DM and a bearing early fault diagnosis method based on successive variational mode decomposition-fast spectral correlation（SVMD-FSC）were proposed.AI_DM only needs the normal operation data of bearings to realize early abnormal warning of bearings by evaluating the irregularity of detected signals.And,it also exhibits higher sensitivity and stability in terms of warning performance than peak value,kurtosis factor,and root mean square value.The warning signal is input into the SVMD-FSC method to diagnose the fault.Firstly,SVMD continuously decomposes the warning signals to obtain a series of modal components.Then,the denoised signal is obtained through the modal component.Finally,spectrum coherence analysis is carried out to extract fault characteristic frequency to achieve fault diagnosis.Through the analysis of experimental data,it is verified that theAI_DM method can detect early anomalies in a timely and stable manner.Meanwhile,SVMD-FSC diagnoses early faults of bearings by extracting early fault features.