Deep Manifold Learning And Its Application In Predictive Maintenance Of Rotating Machinery

With the rapid development of artificial intelligence technology,modern industry is undergoing a transition from traditional manufacturing to intelligent manufacturing.The development level of intelligent manufacturing technology could represent the core competitiveness level of a country’s manufacturing industry.Predictive maintenance technology is a core technology in the intelligent manufacturing technology system,which could make traditional maintenance technology more intelligent.Rotating machinery is one of the most important industrial equipment.In the long-term operation process,local or distributed faults are inevitable.If the fault symptoms cannot be detected in time,it may cause the entire machinery system to shut down,resulting in economic losses and even casualties.Therefore,the research on predictive maintenance technology of rotating machinery has important scientific significance and engineering application value.The main work of this article is as follows:(1)In order to deal with the disturbance of the noise and weak fault features,the unified discriminative manifold learning algorithm is proposed.In the unified discriminative manifold learning algorithm,the local linear reconstruction relationship,the distance between neighboring points,the variance between classes and the variance within classes are constrained.The local features,global features and label information are preserved so that the fault features are effectively extracted.The weighted neighbor graph based on the q-Renyi function is designed,and the interference of noise and outliers could be reduced.A fault diagnosis model based on the unified discriminant manifold learning algorithm is established,and the effectiveness of the proposed method is verified by bearing fault diagnosis experiments and gear fault diagnosis.(2)Considering that different fault types may have similar features and are difficult to distinguish,the multi-kernel supervised manifold learning algorithm is proposed.Inspired by the idea of supervised learning,the multi-kernel supervised manifold learning algorithm uses the label information of the samples to increase the aggregation of the same state feature samples and the difference between different types of feature samples,and the accuracy of fault diagnosis is improved.A weighted neighbor graph based on a multi-kernel function is established,the distance information and angle information between neighbor points are preserved,and the interference of outliers and noise is suppressed.The fault diagnosis model based on the multi-core supervised manifold learning algorithm is constructed,and the bearing fault diagnosis experiment and the gear fault diagnosis experiment were carried out.Fault features could be effectively extracted by multi-kernel supervised manifold learning algorithms,which is proved by the experimental results.(3)In order to extract local information of fault features under complex working conditions,the hypergraph robust multi-manifold learning algorithm is proposed.In the hypergraph robust multi-manifold learning algorithm.the traditional Euclidean distance is replaced by the dynamic time-warping distance.so that the interference caused by the change of working conditions is suppressed.The hypergraph is constructed to replace the traditional neighbor graph,accurately representing complex high-order relationships among samples.and preserving the local information of original data more effectively.A multi-manifold learning framework is constructed,and the accuracy of fault diagnosis is improved by using label information.An electromechanical fault diagnosis model based on the hypergraph robust multi-manifold algorithm was established,and the superiority of the hypergraph robust multi-manifold algorithm was verified through motor fault diagnosis experiments and bearing fault diagnosis experiments.(4)In order to exact weak fault features and simultaneously suppress abundant redundant information under complex working conditions,the hypergraph sparse multi-manifold learning algorithm is proposed.Hypergraph and multi-manifold learning frameworks are constructed so that the local relationships of samples are described more accurately.The interference caused by the change in working conditions is suppressed by the dynamic time-warping algorithm.The sparse representation reconstruction algorithm is combined.The mapping matrix is sparsely constrained by the L2,1 structured norm,redundant features are removed,and the robustness of the algorithm is improved.The manufacturing process monitoring model based on the hypergraph manifold learning algorithm was established,and the additive manufacturing process monitoring experiment and the tool wear state monitoring experiment is carried out.The experimental results show that the feature information of the additive/subtractive manufacturing process could be effectively extracted by the hypergraph manifold learning algorithm,thereby accurately determining the abnormal state of the additive/subtractive manufacturing process is realized.(5)In contrast to traditional deep learning algorithms for which the geometric manifold structure is ignored,the manifold learning algorithm and the deep learning algorithm are combined,and a rotating machinery fault diagnosis model based on the deep discriminative manifold learning algorithm is established.The deep discriminative manifold learning algorithm is proposed based on a symmetric positive definite matrix neural network.The intra-class information and inter-class similarity information are encoded by the Riemannian batch regularization layer in the deep discriminative manifold learning algorithm,and the transformation matrix obtains label information and manifold distribution features.The robustness of the algorithm is improved by the pooling layer based on Riemannian manifolds.A mechanical fault diagnosis model based on a deep discriminative manifold learning algorithm is established,and motor fault diagnosis is completed.The deep discriminative manifold learning algorithm exhibits superior fault feature extraction capabilities and achieves the high fault diagnosis accuracy.Finally,the research content of this article is summarized,and future research work has prospected.