Research On Lifetime State Identification Method Of Space Rolling Bearing Based On Stacked Sparse Autoencoder Network

As an important part of the mechanical parts of the aerospace mechanism,the running state and performance of the space rolling bearing play a key role for the spacecraft in the normal operation,realizing the predetermined function and the expected life.Because the space mechanism works under harsh conditions such as high vacuum,alternating temperature,strong irradiation,atomic oxygen and so on and involves special working conditions such as microgravity,high and low speed,multiple start-up and shutdown,mechanical impact in the launch phase,which result in existing a big difference for rolling bearing between the space and the conventional environment on the ground.There is still a lack of in-depth understanding of the regularity.Because it is impossible to obtain the lifetime state information of the rolling bearing in orbit,at present,the accelerated lifetime test of the rolling bearing is carried out in the simulated space environment on the ground,the running state of the bearing is monitored and the lifetime data are collected.The lifetime state of the space rolling bearing is analyzed through the data,based on which the lifetime prediction and reliability evaluation of the space rolling bearing are carried out.Considering the single amount of bearing state information in friction torque and temperature parameters,this paper chooses the vibration signal which contains rich lifetime state information of space rolling bearing as the research object.the deep learning method is used to automatically extract data features and identify the life state of the bearing.Considering the problem of different data distribution of vibration signals of space rolling bearing under different working conditions,the transfer learning method is introduced to reduce the differences of data sets under different working conditions,and the lifetime state of space rolling bearing under different working conditions is identified.The following is the main research content of the paper:（1）In this paper,the solid lubrication Mo S₂film of rolling bearing used in space service is introduced,which has excellent lubrication performance because of its own structural characteristics,and the applicable environment is summarized.The vibration theory of space rolling bearing and the relevant factors leading to its failure behavior are described.The changes of lifetime and vibration of space rolling bearing under different rotational speeds and loads are analyzed.The relationship between the vibration signal of the space rolling bearing and different lifetime states is established,and then the change rules of the vibration signal characteristics during the change of the lifetime state of the space rolling bearing are shown.（2）Aiming at the problem of lifetime state characterization and recognition of space rolling bearings,a lifetime state recognition method of space rolling bearing based on stacked sparse autoencoder network is proposed in this paper.Taking the collected vibration signal data set as the experimental object,the feature vectors extracted by SSDAE and signal processing methods are used to characterize the characteristics of different lifetime states of the bearing.The t-SNE algorithm is used to visualize them respectively to verify the feature extraction performance of SSDAE.The SSDAE network recognition model established by adding softmax classifier is compared with SVM,BP neural network,DBN and time-frequency feature recognition methods.The experimental results show that the proposed method has higher recognition accuracy,and then verify the feasibility of this method in space rolling bearing lifetime state recognition.（3）Aiming at the problem that the characteristic distribution of source domain training data set and target domain test data set of space rolling bearing under different working conditions are different,a deep learning method combined with transfer learning method is proposed to study the lifetime state identification method of space rolling bearing under different working conditions.If the distribution of source domain data and target domain data is different but it is small,it is proposed that the training SSDAE model based on auxiliary data sets and it can still better identify the lifetime state of space rolling bearings,but the number of target samples added to the source domain will affect the final results.If the distribution difference between the source domain data and the target domain data is large,the JGSA method in the feature transfer learning method is referenced.By processing the feature samples of different distribution,the distribution difference between different domains can be reduced and compared with other domain adaptation methods.The experimental results show that,compared with other domain adaptation methods based on SSDAE,the proposed method has the highest accuracy in space rolling bearing lifetime state recognition.