Research On Fault Diagnosis Methods Of Rotating Machinery Under Variable Operating Conditions Based On Convolutional Neural Networks

Fault diagnosis of rotating machinery can detect,identify and locate the fault before it happens,which is critical to ensure the safety and reliable operation of large rotating machinery.Faced with the increasing demand for real-time intelligent diagnosis of rotating machinery,this thesis makes a deep research on the intelligent fault diagnosis method of rotating machinery under variable operating conditions.Considering the influence of variable operating conditions on vibration signals and different fault characteristics of different components,the fault diagnosis methods for shaft faults,rolling bearing faults and gearbox coupling faults are proposed by improving the network structure,network input and training method of convolution neural networks(CNN).The on-line vibration monitoring and fault diagnosis system of rotating machinery is designed and developed,which makes the proposed diagnosis methods successfully applied.Firstly,for the shafting,bearing and gearbox,which are commonly used components in rotating machinery,the fault simulation experiments are described in detail,including the structure of the test-bed,fault introduction and data acquisition and storage methods.The experiments provide data sets for fault diagnosis methods based on deep learning.Secondly,a fault diagnosis method based on continuous wavelet transform scalogram(CWTS)and CNN is proposed for shafting faults at variable speed.The cropped twodimensional CWTS is directly used as the input of CNN to realize the intelligent fault diagnosis of rotating machinery.This method can effectively utilize the ability of CNN to process multi-dimensional input and CWTS can retain the characteristics of complete fault information in time-frequency domain.It avoids the complex feature extraction steps and possible omissions of fault features.The effectiveness of the method is verified by the experiment data of a rotor testbed.Thirdly,a fault diagnosis method for variable speed bearing based on Pythagorean Spatial Pyramid Pooling(PSPP)CNN is proposed.Considering the distribution of fault characteristics of rolling bearings in the CWTS under variable rotating speed,the vibration signals are transformed by wavelet transform in different scales according to rotating speed,and CWTS in different sizes are obtained.PSPP layer improves the existing spatial pyramid pooling(SPP)layer by reorganizing the feature map after the SPP layer.Then CNN with PSPP layer can process CWTS in different sizes to realize the intelligent fault diagnosis under variable speed condition.The case studied on the bearing experiment data at constant and variable speeds show the effectiveness of the method.Then considering the influence of operating conditions on bearing vibration signals,a rolling element bearing fault diagnosis and localization approach based on multi-task CNN with information fusion is proposed.The domain knowledge of bearing fault characteristic frequency,operating conditions and vibration signals of multi-sensors are fused into 3-dimensional heterogeneous information as input of CNN.The results of two multi-bearing fault experiments show that the method has higher diagnostic accuracy than the existing state-of-the-art approaches.At the same time,the fault localization of bearing will be conducive to the maintenance and replacement of the fault bearing.Furthermore,considering the fault coupling problem of rotating machinery,a diagnosis method for gearbox coupling faults based on multi-task parallel CNN is proposed.The method uses wavelet packet transform to obtain the time-frequency domain information of vibration signals.The fault characteristic map of gears and bearing,operating conditions,and wavelet packet transform of vibration signals are fused to form heterogeneous information of gearbox,which are used as the input of multi-task parallel CNN.The idea of parallel multi-task CNN realizes the diagnosis of coupling faults of gearbox.It can diagnose the faults of gears and bearing simultaneously.The validity of the method is proved by the analysis of the fault coupling experiment data.Finally,an on-line vibration monitoring and fault diagnosis system for rotating machinery is designed and developed.The system can monitor the vibration of rotating machinery with various components in real time and manage data efficiently.The system provides abundant vibration analysis functions.The fault diagnosis algorithm proposed in this thesis is implemented in the system through joint programming.In application,real-time and accurate fault diagnosis can be carried out for each component of rotating machinery.At the same time,the system provides a fault diagnosis expert system module,which enables operators to synthesize a variety of information and conduct more accurate state assessment of rotating machinery.