CN-116304617-B - Electromechanical equipment fault diagnosis method and system based on neural network

CN116304617BCN 116304617 BCN116304617 BCN 116304617BCN-116304617-B

Abstract

The application discloses an electromechanical equipment fault diagnosis method and system based on a neural network, wherein the method comprises the steps of acquiring normal data of electromechanical equipment and fault data of different fault types, and carrying out normalization operation; the method comprises the steps of carrying out Fourier change on normal data and fault data, converting the normal data and the fault data into frequency domain data, converting the frequency domain data into a graph structure, dividing the data into a training set and a testing set according to a preset proportion, training the training set and the testing set through a graph Chebyshev network model, obtaining and splicing features with different fine granularity, importing the features into a trained model, checking a prediction result, and visualizing the prediction result. The method solves the problems of feature extraction and noise reduction of long-term and short-term time sequences, can learn local features through a plurality of surrounding nodes, combines the local features with long-term features through pyramid attention, and realizes the extraction of the long-term and short-term features.

Inventors

CHENG LIANGLUN
MAO DONG
CHEN LI
WANG TAO

Assignees

广东工业大学
广东能哥知识科技有限公司

Dates

Publication Date: 20260505
Application Date: 20230215

Claims (8)

1. An electromechanical device fault diagnosis method based on a neural network, which is characterized by comprising the following steps: Acquiring normal data of the electromechanical equipment and fault data of different fault types, and performing normalization operation; Performing Fourier transform on the normal data and the fault data, and converting the normal data and the fault data into frequency domain data; Converting the frequency domain data into a graph structure, constructing a directed weighted graph, and dividing the data into a training set and a testing set according to a preset proportion; Training the training set and the testing set through a graphic Chebyshev network model to obtain features with different fine granularity and splicing the features; importing the characteristics into a trained model to view a predicted result, and visualizing the predicted result; after obtaining and splicing the features with different fine granularity, the method further comprises the following steps: The method comprises the steps of splicing features with different fine granularity, using the spliced features with different fine granularity as input of a pyramid attention mechanism, and processing the features with different fine granularity through the pyramid attention mechanism to obtain a prediction result of training data; Updating the weight, and back-propagating until training is completed; The step of processing the features with different fine granularity through the pyramid attention mechanism after the features with different fine granularity are spliced as input of the pyramid attention mechanism comprises the following steps: pyramid attention input is obtained by splicing convolution features with different fine granularity: Different fine granularity features are input as a pyramid attention mechanism, and C node features are adopted to form a directed graph The node only needs to pay attention to the adjacent points with the same scale C child node using it as parent node And its parent node And finally form a group of characteristics The scale is represented by a scale and, Representing the first at that scale A plurality of nodes; Wherein, the In the form of an edge matrix, Is a laplace matrix.
2. The neural network-based electromechanical device fault diagnosis method according to claim 1, wherein performing the normalization operation comprises: And carrying out normalization operation on the normal data and the fault data, and controlling the result after the normalization operation in a [0,1] interval.
3. The neural network-based electromechanical device fault diagnosis method according to claim 1, wherein the normal data and the fault data are fourier-transformed into frequency domain data, the fourier transform formula comprising: Wherein, the half output of the original data is The original data is x (t), and the length of the original data is L.
4. The neural network-based electromechanical device fault diagnosis method according to claim 1, wherein the graph chebyshev network model adopts a graph laplace matrix.
5. The neural network-based electromechanical device fault diagnosis method according to claim 1, wherein training the training set and the test set through a chebyshev network model to obtain and splice features with different fine granularity, comprises: taking fault data as input of multi-scale convolution; features between nodes are extracted by a Laplace matrix.
6. An electromechanical device fault diagnosis system based on a neural network, characterized in that the system comprises: the first module is used for acquiring normal data of the electromechanical equipment and fault data of different fault types and carrying out normalization operation; The second module is used for carrying out Fourier change on the normal data and the fault data and converting the normal data and the fault data into frequency domain data; The third module is used for converting the frequency domain data into a graph structure and dividing the data into a training set and a testing set according to a preset proportion; The fourth module is used for training the training set and the testing set through a Tochebyshev network model to obtain and splice the characteristics of different fine granularity; a fifth module for importing the features into the trained model to view the predicted results and visualizing the predicted results; a sixth module, configured to input features with different fine granularity as a pyramid attention mechanism, and process the features with different fine granularity through the pyramid attention mechanism to obtain a prediction result of training data; A seventh module for updating the weights, back-propagating until training is completed; The step of processing the features with different fine granularity through the pyramid attention mechanism after the features with different fine granularity are spliced as input of the pyramid attention mechanism comprises the following steps: pyramid attention input is obtained by splicing convolution features with different fine granularity: Different fine granularity features are input as a pyramid attention mechanism, and C node features are adopted to form a directed graph The node only needs to pay attention to the adjacent points with the same scale C child node using it as parent node And its parent node And finally form a group of characteristics The scale is represented by a scale and, Representing the first at that scale A plurality of nodes; Wherein, the In the form of an edge matrix, Is a laplace matrix.
7. The neural network-based electromechanical device fault diagnosis system according to claim 6, wherein the system comprises: and an eighth module, configured to perform normalization operation on the normal data and the fault data, and control a result after the normalization operation to be within a [0,1] interval.
8. An electromechanical device fault diagnosis system based on a neural network, characterized in that the system comprises: At least one processor; At least one memory for storing at least one program; the neural network-based electromechanical device fault diagnosis method of any one of claims 1-5, when at least one of the programs is executed by at least one of the processors.

Description

Electromechanical equipment fault diagnosis method and system based on neural network Technical Field The application relates to the field of electromechanical equipment fault diagnosis, in particular to an electromechanical equipment fault diagnosis method and system based on a neural network. Background Currently, the manufacturing industry is rapidly developing, and electromechanical devices have become the primary devices in the manufacturing industry. As the length of service of the electromechanical device increases, the failure rate increases, resulting in increased maintenance costs, repair cycles, and miswork costs. In order to solve the problem of production efficiency reduction and even shutdown caused by sudden equipment failure, it is imperative to study fault diagnosis of electromechanical equipment. If the traditional maintenance method is adopted, the maintenance is carried out after the machine is in a fault state, and secondly, as the machine equipment is more and more complex and fine, the time cost and labor cost for checking and detecting the maintenance are greatly increased, and even in most automatic production lines, the machine equipment is huge in quantity, and the cost cannot be measured if the traditional passive maintenance method is relied on. The existing partial time sequence fault diagnosis method has the problems that 1) short-term time sequence features are extracted through local data, so that feature extraction is incomplete, and the effect is unsatisfactory. 2) The characteristics are extracted by adopting long-term and short-term memory through multi-layer convolution or full-connection attention recycling or a circulating neural network, and the characteristics can be obtained from relatively comprehensive data, but the cost is relatively high, and the characteristic extraction is low in efficiency. In addition, the existing method is more flexible in selecting different attentions and has good noise immunity, but has the defects that the processing and denoising method for time series data is lacked, dot Product (DP) and single-layer neural network, mixing and the like are simultaneously used when an attention model is used, all adjacent nodes are required to be combined for the feature extraction of the nodes, so that the space and time complexity is too great, particularly for time series data, excessive data is not required to be combined, and the practice for long-term and short-term prediction of fault detection is lacked. Accordingly, the above-mentioned technical problems of the related art are to be solved. Disclosure of Invention The present application is directed to solving one of the technical problems in the related art. Therefore, the embodiment of the application provides an electromechanical equipment fault diagnosis method and system based on a neural network, which can carry out fault diagnosis on electromechanical equipment based on the neural network. According to an aspect of the embodiment of the present application, there is provided an electromechanical device fault diagnosis method based on a neural network, the method including: Acquiring normal data of the electromechanical equipment and fault data of different fault types, and performing normalization operation; Performing Fourier transform on the normal data and the fault data, and converting the normal data and the fault data into frequency domain data; converting the frequency domain data into a graph structure, and dividing the data into a training set and a testing set according to a preset proportion; Training the training set and the testing set through a graphic Chebyshev network model to obtain features with different fine granularity and splicing the features; And importing the characteristics into a trained model to view the predicted result, and visualizing the predicted result. In one embodiment, after obtaining and stitching features of different fine granularity, the method further comprises: Processing the features with different fine granularity through a pyramid attention mechanism to obtain a prediction result of training data; the weights are updated and back propagated until training is completed. In one embodiment, performing the normalization operation includes: And carrying out normalization operation on the normal data and the fault data, and controlling the result after the normalization operation in a [0,1] interval. In one embodiment, the normal data and the fault data are fourier-transformed into frequency domain data, and the fourier transform formula includes: the half output of the original data is x' (t), the original data is x (t), and the length of the original data is L. In one embodiment, the Tochebyshev network model employs a Tochebyshev matrix. In one embodiment, training the training set and the test set through a chebyshev network model to obtain and splice features with different fine granularity, including: taking fault data as input of multi-sc