Performance Degradation Trend Prediction For Rolling Bearing Based On Manifold Learning

The space rolling bearing is a key component of space equipment,and its performance directly affects the normal operation of the machine. The extreme operating environment of space rolling bearing can easily produce various defects of bearing in operation after a period of time. With the further expansion of the initial defect to operation,The running state and performance of the bearing gradually degraded. However,the space bearing cannot replace and cannot use the backup to guarantee the reliability because the application of bearing in the space condition is limited. Currently, we just only through simulated space environment to carry out degradation trend prediction of rolling bearing. Therefore, degradation trend prediction for space rolling bearing is necessitated to study in the vacuum. The analysis and process of rolling bearing vibration signal can predict the degradation trend and accurately judge the status about equipment. This result can be as the guidance for equipment design and maintenance so that it can avoid the major accident.The degradation prediction method that based on vibration signal can provide the theory for reliability evaluation and remaining life prediction. The harsh working environments, variable operating conditions and complex structure lead to strong noise interference and strongly nonlinear characteristics of vibration signals. The early faults features of rolling bearing are very weak due to they are conceived in the normal state and the fault features have non-linear characteristics, so the early fault features are difficult to extract. But the key point about trend prediction is that extract the features that characterize the operational status so that it can fully and accurately reflect the operational status. Therefore, the trend method with non-linear noise reduction and sensitive indicator has a important significance.Currently, the degradation trend prediction method based on vibration signal can’t effectively reflect the rolling bearing operation status due to their own limitations. In recent years, the rise manifold learning is a data dimensionality reduction method, which has a strong nonlinear data mining capabilities and dynamic learning characteristic, has a very important significance for rolling bearing trend prediction. Based on the manifold learning theory, combined with other signal analysis methods, the paper have been deeply studied the degradation trend prediction method. The specific contents are as follows:① Aiming at the problem that the vibration signals for space rolling bearing has strong noise with strong nonlinear characteristic, a manifold learning noise reduction method based on adaptive neighborhood is proposed. Firstly, according to space reconstruction method, the one-dimensional nonlinear time series reconstructed to a high-dimensional space, and then the inherent nonlinear sequence information can be revealed. Secondly, with the use of local tangent space alignment algorithm, high-dimensional space can be mapped to the intrinsic dimensional space of useful signal and eliminate the noise distributed in high-dimensional space. Then, study the adaptive neighborhood method and the maximum likelihood method to estimate the intrinsic dimension about the signal. The manifold learning methods with intrinsic dimension estimation and adaptive neighborhood can improve the nonlinear noise reduction effect.② Aiming at the problem that it is difficult to establish the feature indicator for predicting the bearing degradation trend, the paper proposes a feature extraction method based on adaptive neighborhood LPP manifold learning. Time domain features, frequency domain features, time-frequency domain features and so on is extracted to form the multi-domain feature set, and the adaptive neighborhood manifold learning method LPP is used to merge the original features and reduce the dimension, so it can solve the conflict and redundancy problem between the feature set and improve the accuracy of degradation prediction.③ Aiming at the problem that the traditional prediction method model can’t be used for space rolling bearing, the paper proposes a new prediction model based on LS-SVM. Based on sensitive indicator extraction, the sensitive characteristic were inputs to the least squares support vector machine to train and construct a model, so as to accomplish the trend prediction.④ On the basis of the above theory, degradation trend prediction module is developed with C# as the development platform. According to experiments and applications, the effectiveness and engineering application of paper method is verified.