Intelligent Fault Diagnosis Technique For Varying Speed Bearings Using Convolutional Neural Network

As one of the core components of rotating machinery,rolling bearing plays an indispensable role.Bearings working in the practical environment often subject to varying working condition,which can lead to various forms of bearing failures.A bearing failure can affect the overall performance of a machine,and result in huge economic losses.Therefore,it is of great practical significance to implement a continuing condition monitoring and fault diagnosis program for rolling bearings.The running state of rolling bearings is often under changing load or speed conditions,and the condition monitoring signals often demonstrate strong non-stationarity and non-linearity characteristics.This then poses a great challenge for the fault feature extraction and diagnosis.Traditional fault diagnosis methods need to extract fault features for fault diagnosis,and largely rely on prior knowledge and expert experience.Deep learning algorithm,on the other hand,has strong feature extraction ability,and can build a deep neural network to automatically learn features from a large number of data samples.Therefore,this study aims to fully utilize with the capacity of deep learning for automated rolling bearing fault diagnosis.The main contributions of this thesis are summarized as follows:(1)This thesis firstly introduced the time-frequency analysis methods of non-stationary signal,introduced multisynchrosqueezing transform(MSST)technology,then introduced the structure of convolution neural network(CNN)and recurrent neural network(RNN)and the principle of extracting feature information.Finally,the fault simulation experimental platform and signal acquisition process used in this thesis were introduced,which provides theoretical and data support for the follow-up research.(2)Aiming at the problem that the vibration signal of variable speed rolling bearing has non-stationary characteristics and the fault characteristics are difficult to effectively extract.An intelligent diagnosis method for rolling bearing faults based on multisynchrosqueezing transform(MSST)and two-channel convolutional neural network(TCNN)was proposed.Firstly,the vibration signal under variable speed was pre-processed by Peak-Hold-DownSample(PHDS)algorithm.Then,performed multisynchrosqueezing transform on the vibration signal under variable speed conditions to obtain an energy-concentrated time-frequency representation.Finally,combined with the two-channel small convolution kernel CNN,the fault feature information in the time-frequency image was extracted and fused,and the classification results are output.It is shown in the study that the proposed technique can achieve a high classification accuracy of 99.67% using the experimental bearing data acquired from a proposed built machine fault simulator.(3)For one-dimensional original time-series vibration signal.a MSPCNN-GRU model is also proposed in the study aiming to achieve an "end-to-end" fault diagnosis of bearings operating under varying speed and load condition.The MSPCNN-GRU model optimizes the idea of multi-scale convolutional layer for a better extraction of the spatial fault features in the signal and combines with a GRU algorithm to retain the temporal information contained in the time series for a more accurate fault diagnosis.It is shown that the technique can effectively capture the fault characteristics contained in the time series of one-dimensional data to yield a high accurate fault recognition result.The test results of vibration signals under different loads in the CWRU bearing data set showed that the test accuracy of the model was 99.83%,99.96%and 99.92% respectively,which verifies that the model has a strong generalization ability.