Vibration Characteristic Analysis And Intelligent Fault Diagnosis Of Gearboxes

With the increasing demand for capacity in society and the development of technology,mechanical equipment tends to be high-speed,high-efficiency and high-precision.As one of the key components used in mechanical equipment to transmit power,the gearbox plays an important role in the running of mechanical equipment.Therefore,in order to ensure the safe and reliable operation of the mechanical system and avoid the occurrence of major accidents,it is of great significance to study fault diagnosis technology of gearbox.This thesis focuses on the key issues of intelligent fault diagnosis of gearbox,including fault feature extraction,fault feature evaluation and selection,diagnosis model training and multi-sensor information fusion.The main contents of this thesis includes:(1)Study on fault feature extraction method of rolling bearing vibration signal.Considering the strong non-stationary characteristics of rolling bearing vibration signal and the phenomenon that the fault characteristics are not obvious,a new method for feature extraction of rolling bearing vibration signals based on synchronous compression wavelet transform(SST)and non-negative matrix factorization(NMF)is proposed.First,the time-frequency analysis of the rolling bearing vibration signal is carried out by SST,and the feature space that can characterize the different fault states of the rolling bearing is extracted.Then,considering that the time-frequency analysis may lead to slight pattern aliasing and high-dimension feature space,the feature space is optimized by NMF to obtain feature sets with moderate dimensions and high sensitivity to various fault patterns.Finally,the effectiveness of the presented fault feature extraction method is validated by experiment.(2)Study on unsupervised intelligent fault diagnosis method of gearbox based on deep belief network(DBN).First,the problems existing in the traditional intelligent diagnosis method based on supervised feature learning mechanism are analyzed.Second,an unsupervised intelligent fault diagnosis method based on DBN is proposed for the fault diagnosis of rolling bearing inside the gearbox.Then,the adaptive feature learning ability of the DBN is studied,and the influence of model parameter and sample number on its performance are considered.Finally,the experimental results show that the proposed method has better classification performance and robustness than the traditional intelligent diagnosis methods.(3)Study on the vibration characteristic analysis methods of gear faults.First,a simplified dynamic model of gear transmission is established to study the vibration characteristics of gears under random excitation.Besides,the vibration signal model of gear under different fault conditions is established and its main characteristics are analyzed.Finally,the simulation signals are used to train the diagnostic model,and a variety of gear faults are identified accurately,which provides a new solution to the problem due to the lack of fault data samples or insufficient training of diagnostic models in actual engineering.(4)Study on the intelligent fault diagnosis of gearboxes based on multi-sensor information fusion.Condisering the shortcomings that individual sensor data may not provide sufficient diagnostic information,and the potential problems that the features selected based on a single evaluation criterion may not be optimal,a new method for fault diagnosis of gearboxes based on multi-sensor data fusion is presented.First,a multi-sensor vibration signal feature extraction.method based on energy operator and time synchronous averaging is proposed to fully extract fault information from multi-sensor and multi-feature domains.Then,a two-stage fault feature selection and information fusion method based on distance evaluation technique and max-relevance&min-redundancy is proposed,which effectively selects features that are sensitive to faults and contain less redundant information.The experimental data analysis results show that the proposed method is superior to the traditional fault diagnosis method based on single sensor information.