Research On Bearing Fault Diagnosis Method Based On The Combination Of Acoustic And Vibration Signals

With the current rapid development of big data and artificial intelligence technology,the gradual transformation of the traditional machinery manufacturing industry is also an inevitable trend.The bearing is one of the most typical components in rotating machinery equipment,and its operating state has a great impact on the normal operation of rotating machinery equipment.Therefore,it has important theoretical and practical significance to research the fault diagnosis of bearings.Traditional bearing fault diagnosis research is mostly based on the time-frequency domain feature extraction of vibration signals to complete the diagnosis task.However,in the actual factory workshop,due to the harsh environment such as high corrosion and high temperature,the vibration signal collected only through contact cannot satisfy the requirements of the bearing fault diagnosis.Thus,the vibration signal and the sound signal collected through the non-contact such as a microphone are complementary to each other,and jointly reflect the working state of the bearing.In addition,in bearing fault diagnosis,most of the researches are based on single signal source and single model,and how to realize bearing fault diagnosis through multi-model fusion under the condition of multiple signal sources is the focus of this thesis.The main research work of this thesis is as follows.(1)Aiming at the nonlinear and non-stationary characteristics of rolling bearing fault vibration signals,an adaptive bearing fault diagnosis method based on the fusion of one-dimensional convolutional neural network and long-short-term memory network is proposed.Firstly,the original one-dimensional vibration signal is input the 1-DCNN and LSTM channels through overlapping sampling,and then the dimensional splicing method is used to merge the feature information in the space and time dimensions.Finally,the classification output of the fault category is performed through the softmax classifier.This method can directly extract features from the original vibration signal adaptively,and realize "end-to-end" fault diagnosis.Using the fault data of the CUT-2 experimental platform,through the experimental analysis of different fault types,different sensor acquisition orientations,and different fault diameters of rolling bearings,the results show that this method has higher recognition accuracy than other methods in identifying bearing fault categories,and has good generalization and robustness.(2)In the bearing fault diagnosis task,a single vibration signal source cannot fully reflect the working state of the bearing due to the influence of the equipment environment and other factors.This paper proposes an end-to-end bearing fault diagnosis method based on multi-channel convolutional neural network and attention mechanism combined analysis of sound and vibration signals.This method collects bearing fault information by arranging vibration sensors and sound sensors in different orientations.First,it uses the powerful self-learning ability of the convolutional neural network to adaptively extract the fault characteristic information of each vibration and sound sensor.Secondly,the attention mechanism is used to adaptively assign the feature weight of each channel,and fuse the fault feature information.Finally,the multi-classification function is used to identify various bearing faults to realize the bearing fault diagnosis task.The effectiveness of the method is verified by the fault data of the CUT-2 experimental platform.The experimental results show that the method proposed in this thesis has a higher fault recognition rate than the diagnosis accuracy of a single sensor.