Bearing Fault Diagnosis Of Permanent Magnet Synchronous Generator In Direct-Drive Wind Turbine Under Variable-Speed Condition

Wind energy is one kind of renewable energies.Wind power generation decreases the burning of fossil fuels,and hence promotes the economic development,protects the environment,and achieves sustainable development.Wind turbine converts the mechanical energy into the electrical energy,and its condition motoring and fault diagnosis are important to guarantee the safety operations,reduce the shutdown losses and avoid catastrophic failures.Direct-drive permanent magnet synchronous generator(PMSG)is widely used in wind turbines due to its distinct merits such as simple structure and high efficiency.Bearing which supports the rotating shaft is a key component in a PMSG,and its reliability affects the efficiency and safety of the wind turbine.This dissertation investigates PMSG bearing fault diagnosis in wind turbine.When a bearing rotates at constant-speed condition,the time-domain or frequency-domain methods can be used to analyze the bearing signals for fault diagnosis.When a fault occurs in a bearing component,the fault characteristic frequency will appear in the envelope spectrum of the bearing signal.The PMSG bearing works at variable-speed condition because the wind direction and magnitude always change with the weather.The variable-speed induces spectrum smearing and affects accurate diagnosis of bearing fault.Order analysis(OA)converts a time-domain signal into an angular-domain one and it is suitable to be used for variable-speed bearing fault diagnosis.OA requires accurate rotating speed as a reference signal for angular-domain signal resampling.In some cases,the wind turbine manufacturer or the wind farm cannot provide accurate rotating speed information.In this case,the PMSG rotating speed should be estimated from other sensors to realize OA-based bearing fault diagnosis.Focusing on this issue,this dissertation investigates several methods to estimate the rotating speeds from the PMSG phase current and bearing vibration signals.The estimated rotating speed is used to resample the original vibration signal to diagnose bearing fault under variable-speed condition.First,an adaptive PMSG bearing fault diagnosis method is proposed in considering of the noise interference.In this method,the PMSG phase current and bearing vibration signals are synchronously acquired.An adaptive filter is designed to purify the current signal and then the rotating phase is estimated from the filtered signal.The vibration signal is resampled according to the estimated rotating phase.An adaptive stochastic resonance-based filter is used to enhance the order signal.Finally,the fault characteristic order can be recognized in the envelope order spectrum for fault diagnosis.Specially,all the procedures in the proposed method are totally adaptive without human intervention.The proposed method improves the accuracy and efficiency for PMSG bearing fault diagnosis under variable-speed condition.Second,a method based on single vibration signal analysis is proposed for PMSG bearing fault diagnosis.Synchro-Squeezing Wavelet Transform is applied to process the vibration signal,and the instantaneous frequency(IF)of the rotating component is extracted from the two-dimensional time-frequency representation.The rotating component is then reconstructed from the extracted IF curve.The vibration signal is resampled according to the calculated phase of the rotating component to realize OA and bearing fault diagnosis.This method requires only one channel of vibration signal,and hence reduces the requirement of sensor and decreases the complexity and cost of the diagnosis system.Third,a two-step method is proposed for pattern recognition of multiple bearing faults.The PMSG phase current and bearing vibration signals are synchronously sampled.Peak detection and numerical interpolation techniques are applied to the phase current signal to obtain the upper envelope of the current signal.The vibration signal is then compensated and adjusted according to the upper envelope of the current signal.Subsequently,the rotating phase is estimated from the current signal to resample the vibration signal.Multiple dimensions of features are extracted from the resampled signal and a neural network model is used to fuse the features for bearing fault recognition.If the features are directly extracted from the time-varying vibration signals,their statistical characteristics are not stable as these features change with time.By pre-processing the vibration signal using the proposed method,the extracted features become stable respect to time.Hence,the accuracy of fault recognition is improved by using the proposed method.The methods proposed have several distinct merits such as simple algorithm structure,easy to be implemented,high computation efficiency and high diagnosis accuracy,and hence they are suitable for PMSG bearing fault diagnosis under variable-speed condition.Specially,if the rotating speed cannot be provided or the accuracy of rotating speed is not high enough,the proposed methods have good engineering application prospects.The principles of the proposed methods can also be extended to diagnose other rotating components in wind turbine under variable-speed condition.