Bearing fault diagnosis based on pseudo-tag semi-supervised kernel local Fisher discriminant analysis is provided. The method is characterized in comprising the following steps: (1) collecting vibration signals of bearings under different working conditions to form a training sample; (2) performing feature extraction on the training sample obtained in (1); (3) performing normalization processing on features obtained in (2); (4) obtaining a clustering tag set by using density peak clustering for the entire feature set obtained in (3); (5) using clustering pseudo-tags obtained in (4) to construct local inter-cluster divergence and intra-cluster divergence regularization terms, and combining the regularization terms with the inter-class divergence and intra-class divergence with tag samples in the FDA to determine a final projection vector; (6) using the final projection vector obtained in (5) to solve a projection set of the tagged feature set in the dimensionality reduction space; (7) using the projection set obtained in (6) to train an extreme learning machine; and (8) performing processing of (2), (3) and (5) on the collected vibration signals, and inputting the processed vibration signals to determine the working conditions. The technical scheme of the present invention is applied to the problem of fault identification of bearing equipment.

Bearing fault diagnosis based on pseudo-tag semi-supervised kernel local Fisher discriminant analysis

The invention relates to a state recognition method for high-end numerical control equipment based on multi-source information fusion, and the method comprises the steps: collecting multi-source sensor information of the operation state of the high-end numerical control equipment through an onsite collection sensor, and carrying out the information fusion; obtaining features of the time domain, frequency domain and wavelet domain of a fusion signal, and carrying out the fusion of the extracted time domain features, frequency domain features and wavelet domain features; reconstructing a multi-source multi-domain high-dimensional phase space of the initial features, employing a local linear embedded structure for the dimension reduction, optimizing an intrinsic dimension number, constructingan neighbourhood which can reflect the local linear features of a low-dimensional manifold, and obtaining the low-dimensional manifold changes in a high-dimensional space of the operation state; obtaining low-dimensional sensitive features through the distance criteria, constructing a multiplication similarity of the low-dimensional manifold features, and achieving the recognition of different states. The method can achieve the effective recognition and judgment of the operation state of the numerical control equipment based on real-time monitoring and effective sensing information, and accurately recognizes different stats of the numerical control equipment.

State recognition method for high-end numerical control equipment based on multi-source information fusion

The invention discloses a fault diagnosis method oriented to the empirical mode and wavelet decomposition and feature selection of the steam turbine main engine of an intelligent power plant. The invention obtains a plurality of eigenmode components (IMF) through decomposing the original signal by EMD, and overcomes the mode aliasing problem of EMD decomposition by using wavelet decomposition. Atthat same time, the stationary part and the non-stationary part are selected respectively, the key feature is extracted, which not only reduces the dimension of the original feature and the redundancyof data, but also overcomes the problem that the non-stationary signal feature is easily concealed. At the same time, the fault diagnosis efficiency of the turbine main engine is greatly improved, which is beneficial for the maintainers to control and repair the equipment, and ensures the safety of the power generation process.

A fault diagnosis method oriented to the empirical mode and wavelet decomposition and feature selection of the steam turbine main engine of an intelligent power plant

The present invention provides a batch process fault detection method based on a time sequence extension and neighbourhood maintenance extreme learning machine. The method mainly comprises the following steps of: (1) collecting batch process data in a plurality of batches of normal conditions; (2) developing the collected three-dimensional data to two-dimensional data and performing standardization of the two-dimensional data; (3) establishing a time sequence extension and neighbourhood maintenance extreme learning machine model; (4) establishing statistics of an SPE (Squared Prediction Error)and Hotelling T2 in the normal data to obtain control limits; (5) collecting online batch process data; (6) performing projection of the online data through the established time sequence extension and neighbourhood maintenance extreme learning machine model; and (7) solving the SPE and T2 statistics of the online data to determine whether faults are generated or not. The batch process fault detection method based on the time sequence extension and neighbourhood maintenance extreme learning machine maintains the space local neighbourhood structure and the dynamic time sequence structure of thedata in the projection data at the same time so as to avoid that the process data meets the hypothesis of the Gaussian distribution and more meet the actual industrial process.

Batch process fault detection method based on time sequence extension and neighbourhood maintenance extreme learning machine

The invention discloses a fault diagnosis method based on non-equal-weighted local preserving embedding, and the method aims at solving a problem of how to build a dynamic directed graph structure according to the complex dynamic characteristics of data in a modern industrial process, so as to preserve the local neighbor relation of original samples, to preserve the local structural relation of the original feature variables, and to enable the embedded low-dimensional data structural relation to be consistent with an original space. According to the invention, the method comprises the steps: constructing a directed network through non-equal-weighted connecting sides of the graph, calculating a probability distance guide sample similarity matrix, forming a non-equal-weighted local preserving embedded model, so as to effectively represent the local neighbor relation of samples in a dynamic process; introducing a feature variable to the graph construction, preserving the local relation information of the feature variable, selecting a feature variable which severely affects a process fault, and further improving the classification precision of a diagnosis model. Compared with a neighbor preserving embedding method, the method can represent the topological structure relation of process data, also achieves the construction of the non-equal-weighted local preserving directed graph, achieves the obtaining of the neighbor relation between the samples, represents the local manifold structure of the feature variable in a better way, and reflects the dynamic change conditions of the process. Therefore, the non-equal-weighted local preserving embedded model can obtain a better fault diagnosis result of the dynamic process.

Fault diagnosis method based on non-equal-weighted local preserving embedding

The invention relates to a time series data manifold learning-based mechanical equipment fault diagnosis method The method comprises the following steps of (1) obtaining a training sample under each fault type, wherein the training sample comprises multiple groups of vibration signal time series; (2) performing data preprocessing, multi-feature construction, feature selection and feature fusion on each group of the vibration signal time series under each fault type in sequence, finally obtaining a low-dimensional feature set of the sample, and inputting the low-dimensional feature set to a classifier; (3) obtaining test data, performing data preprocessing, multi-feature construction, feature selection and feature fusion on the test data in sequence, and finally obtaining a low-dimensional feature set of the test data; and (4) inputting the low-dimensional feature set of the test data to the classifier, and performing fault diagnosis by the classifier to obtain the corresponding fault type Compared with the prior art, the method has the advantage that a fault diagnosis result is accurate and reliable

Chen Zhangli

Papers

Time series data manifold learning-based mechanical equipment fault diagnosis method