Research On Fault Feature Extraction For Wind Turbine Drive Train

With the objective of fault diagnosis for windturbine derivetrain under complex conditions,this thesis is based on the nonlinear and nonstationary charectristics of the fault signals of ratating parts,analyze state of the art and challenges of mathematical morphology and variational mode decomposition,conduct the studies and develompents of them under the existing theoretical framework and complete the feature extraction and detection of the incipent faults of windturbine drivetrain successfully.Three key parts are sperately investigated deeply as follows.The first part is that the fault diagnosis for rolling bearings is based on iterative mathematical morphology.The performace of mathematical morphology?IMM?depends on the defined structuring elements significantly,and the current studies need compressively considering the balance between denoising and retaining feature formation,producing the difficulties in defined SE,weak extraction to fault features and redundant noise.As a result,a novel morphology termed IMM is presented.Differeting from the independent computation in conventional morphology,IMM is correlative in the iterated process,resulting in that a shortest unit SE is applied.Therefore,it can avoid the complex defintions to SE in conventional morphology.Meanwhile,improved defference operator is designed to meet the need in IMM to a non-idempotent operator.In single scale morphological analysis,the combined method based on IMM and simplified sensitive factor?SSF?deal with the vibrations of motor rolling bearings and the fault bearings in winturbine drive train,the results show that the proposed method can isolate the fault features.Compared with the two methods presented by Nikolaou and Raj,it has a simpler definition to SE,more effective removal to noise and superior performance in extracting fault features.In multiscale morphology,IMM is used to process the results by CMM for filtering the residual noise,thereby emphasizing the feature information.The optimized approach has an adaptive performance,better denoising and fault diagnosis abilities when compared with CMM under strong noise background.The second part is a method based on asymmetric multiscale morphological filtering in fault diagnosis for motor rolling bearings.Taking the modulating phenomena of vibration signals of rolling bearings as the research object,deriving the relationship between the amplitudes of the extracted impulses and the scales of structuring elements and analyszing the weak demodulation to the amplitude modulation following the scale change of structuring element for comibined operators,an asymmetric form is considered.Aiming to the partial extraction of impulsive signals and applying the translation characteristics of flat structuring element,adaptive extended calculation is used.Compared with CMM,it indicates that the presented method is sensitive to amplitude modulation,and has the better demodulation performance.The last one is the application of variational mode decomposition to diagnosing the gearbox faults of windturbine drivetrain.The key in Variational mode decomposition is how the number of band-limited sub-signals?i.e.??is defined adaptively.Firstly,considering the relation between?and the spectrum structure of a raw signal and proposing feasibility that the parameter?can be defined directly through simplifying the original spectrum,a mixed method based on VMD and l₁ trend filtering is employed to extract the time-domain impulse signals for the defect gearbox of windturbine.Additionally,the current studies on define?adaptively is based on an assumption that the change of feature infromaiton is monotonous when a given signal is decomposited by variational mode decomposition.In fact,it can not be satisfied in real applications.Consequently,Tentative varitional mode decomposition?TVMD?and Dynamic time warping?DTW?is considered here.The first one is to handle the dependence to that assumption in defineing this parameter?adaptively,and DTW is to process the mode mixing in heavy noise background.Finally,the proposed approaches are applied to the fault diagnosis of gearbox of windturbine drivetrain.Compared with empirical mode decomposition,it can detect the gearbox failure more efficiently.The above studies are all based on simulation and the vibration signals of rotating parts.The results show that the proposed methods can enhance the featur extraction of the weak and incipient faults and are significant to diagnose windturbine drivetrain faults.