| The core components,such as bearings and gears,within rotating machinery are susceptible to damage,and therefore effective fault diagnosis and condition monitoring are of great significance to ensure the normal operation of equipment and prevent major accidents.Vibration signals collected from mechanical equipment usually contain important information reflecting its operating status,but there is a large amount of redundant information and noise in the original data,resulting in poor performance of the fault diagnosis system,so an effective method is needed to extract valuable discriminative information from the high-dimensional fault data.Based on the above problems,this paper takes the feature extraction algorithm of fault data as the specific research object,and conducts research and discussion with the goal of improving the accuracy of fault diagnosis.The main work of this dissertation is as follows:(1)Fault diagnosis methods are usually sensitive to outliers,and it is difficult to extract global and local discriminant information at the same time,resulting in poor separation between low-dimensional discriminant feature subsets.To solve this problem,a fault diagnosis method of rotating machinery was proposed based on global-local Euler elastic discriminant projection.This method maps the high-dimensional fault features to the Euler representation space through the cosine metrics,and expands the differences between heterogeneous fault samples.Then,an optimization model based on three objective functions of global,local and within-class scatter is constructed in this space,which further improves the local intra class aggregation and global inter class separation of low dimensional discriminant feature subsets on the basis of maintaining the overall structure.The experimental results on two mechanical fault datasets of bearing and gearbox show that the proposed method can effectively explore the fault discrimination information and has superior fault diagnosis performance.(2)Aiming at the problem that the global and local discriminative information of Euler fault data is not fully captured and difficult to balance,a Euler representation-based structural balance discriminant projection is further proposed.First,the high-dimensional fault features are mapped to the Euler representation space.Then,a unified objective model combining four objective functions with different structural and class information is constructed in this space.Finally,an adaptive balance strategy is given for optimizing the unified objective model of the algorithm,which achieves an elastic balance between global and local features in the projection subspace.The diagnosis performance of the algorithm is analyzed and verified by two mechanical fault cases.Encouraging experimental results show that the algorithm is able to capture effective fault discriminative features.(3)Aiming at the problem that it is difficult for subspace learning algorithms to obtain and balance the adaptive global and local intrinsic graph structure information of fault data,a new elastic subspace diagnosis algorithm based on graph balancing discriminant projection is proposed.The algorithm constructs adaptive global geometric graphs and adaptive local geometric graphs through inter-sample collaborative representation and supervised information,and further gives a graph balancing strategy for constructing an elastic subspace learning optimization model.The algorithm improves the intra-class aggregation and inter-class differentiability of the elastic subspace based on capturing the inherent discriminative structure of fault data,and achieves automatic learning corresponding to the global and local information weights.The results on two different experimental cases show that the proposed algorithm has strong robustness and can extract more discriminative low-dimensional sensitive features to improve the accuracy of fault diagnosis.Figure [38] Table [9] Reference [75]... |
PDF Full Text Request