CN-122020246-A - Open set fault diagnosis method and system based on graph small wavelength short-term memory network

CN122020246ACN 122020246 ACN122020246 ACN 122020246ACN-122020246-A

Abstract

The invention discloses an open-set fault diagnosis method based on a graph small-wavelength short-term memory network, which comprises the steps of obtaining multi-source sensor data in different states of a gearbox, carrying out normalization processing on the obtained multi-source sensor data, regarding each sensor as a graph node, constructing an inter-node adjacency matrix, simultaneously generating a node characteristic matrix of each state mode by using a sliding window, constructing GWLSTM memory network units, carrying out multi-scale coding on the node characteristic matrix through the stacked GWLSTM memory network units, fusing time sequence information and graph structure dependence to obtain node representation, constructing a graph small-wave energy pooling module, adaptively aggregating the node representation into a global graph representation based on graph small-wave energy pooling, inputting the global graph representation into two layers of fully-connected classifiers, and carrying out open-set discrimination by combining an energy function, so that the invention realizes classification of known fault types and reliable identification of unknown faults.

Inventors

LI TIANFU
HE JIANG
Hou Bingchang
CHEN JUNFAN
LIU TAO

Assignees

昆明理工大学
重庆大学

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (7)

1. An open set fault diagnosis method based on a graph small wavelength short-term memory network is characterized by comprising the following steps: step S1, acquiring multi-source sensor data of a gearbox in different states, carrying out normalization processing on the acquired multi-source sensor data, regarding each sensor as a graph node, and constructing an inter-node adjacency matrix; S2, constructing GWLSTM memory network units, performing multi-scale coding on the node characteristic matrix through the stacked GWLSTM memory network units, and fusing time sequence information and graph structure dependence to obtain node representation; s3, constructing a graph wavelet energy pooling module, and adaptively aggregating node representations into global graph representations based on graph wavelet energy pooling; and S4, inputting the global diagram representation into a two-layer fully-connected classifier, and carrying out open set judgment by combining an energy function to realize classification of known fault types and reliable identification of unknown faults.
2. The open set fault diagnosis method based on a graph small wavelength short term memory network of claim 1, wherein the stacked GWLSTM memory network units are represented as: ; ; In the formula, Is a first layer GWLSTM memory network element For a pair of The node representation that is learned after encoding, Is a second layer GWLSTM memory network element Based on The node representation that is learned after further encoding, The characteristic matrix of the node is represented, Is an adjacency matrix.
3. The open set fault diagnosis method based on graph wavelet short-term memory network according to claim 1, wherein said GWLSTM memory network unit organically combines discrete graph wavelet convolution with a long-term memory network gating mechanism, and the updating process is formally expressed as: ; ; ; ; Wherein, the , , , Respectively representing a forgetting gate, an input gate, a cell input and output gate; , , And The respective representation being applied to the input instant Node feature matrix of (a) The frequency domain filter matrix of (a) is used for respectively adjusting the operations of the forgetting gate, the input gate and the cell input and output gates; , , And Respectively representing frequency domain filter matrices applied to hidden states, respectively, to be applied to Hidden state of time Connected to the forget gate, input gate, cell input and output gate; Is a trainable parameter that controls the filtering behavior; , , And Deviation terms respectively representing a forgetting gate, an input gate, a cell input and output gate; Representing a discrete diagram wavelet convolution operator, Representing a discrete graph wavelet inverse transformation operator; is a soft contraction function that can be learned.
4. The open set fault diagnosis method based on the graph wavelet short-term memory network according to claim 1, wherein the graph wavelet energy pooling module is expressed as: ; Wherein, the Representing a global graph representation obtained by a graph wavelet energy pooling module; is shown in the first Scale, first On each characteristic channel, vectors composed of wavelet coefficients after soft thresholding of all nodes.
5. The open set fault diagnosis method based on the graph small wavelength short-term memory network according to claim 1, wherein the process of realizing classification of known fault categories and reliable identification of unknown faults is divided into a training phase and a testing phase: Training the proposed GWLSTM memory network model with a standard cross entropy loss function by using the marked known state in a training stage, wherein the GWLSTM memory network model comprises a GWLSTM memory network unit, a graph wavelet energy pooling module and a full-connection classifier; In the test stage, an open set classifier based on graph energy is introduced, the open set classifier constructs an energy function, the graph features are mapped onto a scale energy, and the energy function is used as a scoring function of the open set classifier for a given energy threshold value: ; Wherein, the For open-set classifier, if the energy value of the fault Above the energy threshold It is considered an open set fault and marked-1, otherwise the label corresponding to the highest probability will be used Marking is carried out.
6. An open set fault diagnosis system based on a graph small wavelength short-term memory network, characterized by comprising the module of the open set fault diagnosis method based on a graph small wavelength short-term memory network as set forth in any one of claims 1 to 5.
7. A terminal comprising a processor, a memory and a computer program stored in the memory and executable on the processor, the processor being configured to perform the steps of the open set fault diagnosis method based on a graph-small wavelength short-term memory network of any one of claims 1-5.

Description

Open set fault diagnosis method and system based on graph small wavelength short-term memory network Technical Field The invention relates to an open-set fault diagnosis method and system based on a graph small-wavelength short-term memory network, and belongs to the field of mechanical fault diagnosis. Background In recent years, the intelligent fault diagnosis method based on deep learning is widely applied in the fields of prediction and health management due to strong data driving feature learning capability, but the existing method still faces two key problems, namely, how to extract robust and working condition-independent features from complex multivariable signals so as to adapt to different running conditions and environmental changes, and secondly, when unknown or unseen fault types occur, the recognition performance of the traditional depth model is obviously reduced, and the open-set diagnosis requirement under complex actual working conditions is difficult to meet. Although in recent years graph neural networks have been available to model spatial correlation of multi-source signals, existing approaches still suffer from deficiencies in capturing spatio-temporal dependencies and achieving robust open-set fault identification at the same time. The above information disclosed in the background section is only for enhancement of understanding of the background of the invention and therefore may contain information that does not form the prior art that is already known in the country to a person of ordinary skill in the art. Disclosure of Invention The invention provides an open-set fault diagnosis method based on a graph wavelet short-term memory network, which is characterized in that discrete graph wavelet convolution is embedded in an LSTM unit to realize the self-adaptive extraction of multi-scale space-time characteristics; while introducing a learnable soft-shrink function to suppress noise sensitive components and enhance feature robustness. Meanwhile, the operation of the design GWPool is carried out, the node characteristic matrix is adaptively aggregated into a global graph representation, and the global graph representation is used for calculating graph energy in test trial, so that the reliable identification of unknown faults is realized. The technical scheme of the invention is as follows: according to a first aspect of the present invention, there is provided an open set fault diagnosis method based on a graph small wavelength short-term memory network, comprising: step S1, acquiring multi-source sensor data of a gearbox in different states, carrying out normalization processing on the acquired multi-source sensor data, regarding each sensor as a graph node, and constructing an inter-node adjacency matrix; S2, constructing GWLSTM memory network units, performing multi-scale coding on the node characteristic matrix through the stacked GWLSTM memory network units, and fusing time sequence information and graph structure dependence to obtain node representation; s3, constructing a graph wavelet energy pooling module, and adaptively aggregating node representations into global graph representations based on graph wavelet energy pooling; and S4, inputting the global diagram representation into a two-layer fully-connected classifier, and carrying out open set judgment by combining an energy function to realize classification of known fault types and reliable identification of unknown faults. Further, the stacked GWLSTM memory network elements are represented as: ; ; In the formula, Is a first layer GWLSTM memory network elementFor the node representation learned after encoding X,Is a second layer GWLSTM memory network elementBased onThe node representation that is learned after further encoding,The characteristic matrix of the node is represented,Is an adjacency matrix. Further, the GWLSTM memory network unit organically combines the discrete graph wavelet convolution with a long-short-term memory network gating mechanism, and the updating process is formally expressed as: ; ; ; ; Wherein, the ,,,Respectively representing a forgetting gate, an input gate, a cell input and output gate;,, And Respectively representing node characteristic matrices applied to input time tThe frequency domain filter matrix of (a) is used for respectively adjusting the operations of the forgetting gate, the input gate and the cell input and output gates;,, And Respectively representing frequency domain filter matrixes applied to hidden states, and respectively representing hidden states at time t-1Connected to the forget gate, input gate, cell input and output gate; Is a trainable parameter that controls the filtering behavior; ,, And Deviation terms respectively representing a forgetting gate, an input gate, a cell input and output gate; Representing a discrete diagram wavelet convolution operator, Representing a discrete graph wavelet inverse transformation operator; is a soft contraction function that can b