Fault diagnosis method and device based on graph attention convolution auto-encoder, and medium

Abstract

The invention relates to a fault diagnosis method and device based on a graph attention convolution auto-encoder and a medium, and the method comprises the steps: S1, obtaining an original fault data set, and carrying out the standardization operation of the fault data set obtained after slicing; s2, calculating the distance between variable feature vectors in the standardized fault data set, and constructing an adjacency matrix by adopting a K adjacency algorithm; s3, inputting the adjacent matrix and the standardized fault data set into a graph attention convolution auto-encoder to extract spatial correlation features and fault information features; s4, inputting the fault information features into a double-layer full-connection layer, and outputting a classification prediction category; s5, calculating a loss function of the graph attention auto-encoder, and training the model through back propagation of the loss function; and S6, performing fault diagnosis on a fault data set to be diagnosed by using the trained graph attention convolution auto-encoder model, and outputting a classification prediction category. Compared with the prior art, the method has the advantages of high interpretability and high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of product fault diagnosis, and in particular to a fault diagnosis method, device and medium based on a graph attention convolutional autoencoder. Background Art

[0002] With the expansion of industrial scale and the continuous progress of technological innovation, fault diagnosis in the manufacturing process has become crucial. An efficient fault diagnosis system can identify potential manufacturing defects at the initial stage of production, thereby optimizing the production process. In the manufacturing industry, fault diagnosis is not only a key link to ensure product reliability, but also with the development of intelligent manufacturing technology, the requirements for the timeliness and accuracy of fault information are getting higher and higher. The application of cutting-edge technologies such as industrial big data and artificial intelligence provides strong support for real-time fault diagnosis in the manufacturing process. By analyzing production data, these technologies can predict and identify fault patterns, thus realizing early warning and rapid response to faults, significantly improving production efficiency and product quality.

[0003] In the manufacturing field, traditional fault diagnosis methods mainly rely on statistical process control (SPC) technology. This method predicts potential faults by monitoring the fluctuation trend of quality parameters in the production process. These SPC-based fault diagnosis methods largely rely on the experience and knowledge of quality control experts, and the response speed is slow, which limits their applicability in different scenarios. To improve the universality and accuracy of fault diagnosis, modern manufacturing is exploring the use of big data and artificial intelligence technologies to reduce the dependence on expert experience and improve the objectivity and automation level of diagnosis. Through these technologies, fault patterns can be identified and predicted more accurately, thus realizing early warning and rapid response to faults, significantly improving production efficiency and product reliability.

[0004] Driven by Internet technology and the Internet of Things (IoT), the collection of manufacturing process data has reached an unprecedented scale. Facing this challenge, traditional SPC-based fault diagnosis methods are no longer applicable due to parameter and speed limitations. Therefore, artificial intelligence-based fault diagnosis methods have been proposed. These methods specifically model the characteristics of manufacturing data and predict potential faults by analyzing the model output. These methods automatically extract key features from the data through deep learning technology and establish the relationship between these features and faults, effectively coping with the explosion of data volume. Although these methods have improved in feature extraction and fault prediction, they often ignore the interconnections between data and fail to fully utilize the high-level information of the data. At the same time, the "black box" nature of deep learning models makes the internal learning process of the model opaque, and it is difficult to understand the specific impact of model parameters on the results, which limits the interpretability and reliability of the model. Summary of the Invention

[0005] The object of the present invention is to overcome the defects of the above-mentioned existing technologies, and provide a fault diagnosis method, device and medium based on a graph attention convolutional autoencoder. This method uses a graph attention autoencoder to extract spatial correlation features and fault information, so as to make more full use of the correlation relationship between process variables, improve the accuracy of fault diagnosis, and at the same time use the attention coefficient of the graph neural network to inversely deduce several production parameters with the largest weight for the fault diagnosis of the final product, providing interpretability analysis and research for the prediction results of the model.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] According to the first aspect of the present invention, there is provided a fault diagnosis method based on a graph attention convolutional autoencoder, including:

[0008] S1. Obtain the original fault data set D, perform a slicing operation on the original fault data set D, and perform a normalization operation on the sliced fault data set X and the original fault data set D respectively to obtain a normalized fault data set S(X) and a normalized original fault data set S(D);

[0009] S2. Calculate the distance between each variable feature vector in the normalized fault data set S(D), and use the K-nearest neighbor algorithm to construct an adjacency matrix A;

[0010] S3. Input the adjacency matrix A and the normalized fault data set S(X) into the graph attention convolutional autoencoder model to extract spatial correlation features and fault information features Z;

[0011] S4. Input the fault information features Z into a two-layer fully connected layer, and use the softmax algorithm to calculate the classification prediction category

[0012] S5. Calculate the loss function of the reconstructed spatial correlation features and the loss function of the classification prediction category and combine them into the loss function of the graph attention autoencoder, and train the graph attention convolutional neural network model through backpropagation of the loss function;

[0013] S6. Use the trained graph attention convolutional autoencoder model to perform fault diagnosis on the fault data set to be diagnosed, and output the classification prediction category.

[0014] Preferably, the slicing of the original fault data set D is specifically: performing a sliding window edge cutting operation on the original fault data set D with a step size of 1 and a window size of 20.

[0015] Preferably, the normalization operation is specifically:

[0016]

[0017] Where: x is the current data variable of the current item, x ∈ X; u is the mean value of the current data variable; σ is the standard deviation of the current data variable.

[0018] Preferably, the distance between each variable feature vector in the calculated standardized fault data set S(D) is calculated by the following expression:

[0019]

[0020] Where: d :,i is the i-th variable vector of matrix D; dist(d :,i , d :,j ) is the distance between the i-th variable feature data and the j-th variable feature data in matrix D.

[0021] Preferably, the distance between each variable feature vector is the cosine distance, and the calculation expression is

[0022]

[0023] Where: d :,i is the i-th variable vector of matrix D; dist(d :,i , d :,j ) is the distance between the i-th variable feature data and the j-th variable feature data in D; m is the number of variable vectors in matrix D.

[0024] Preferably, the adjacency matrix A and the standardized fault data set S(X) are input into the graph attention autoencoder to extract spatial association features and fault information feature Z. The specific mathematical expression is:

[0025] Z = GAT(GAT(S(X))) (4)

[0026]

[0027] Where: GAT is the graph attention convolutional network, which is used as an encoder to extract the fault information feature Z of the input data; σ is the sigmoid activation function.

[0028] Preferably, the fault information feature Z is input into a two-layer fully connected layer, and the softmax algorithm is used to calculate the final classification prediction category of each sample The mathematical expression is:

[0029]

[0030] Where: FC is the fully connected layer.

[0031] Preferably, the loss function of calculating the reconstructed spatial correlation features and the loss function of the classification prediction category are combined into the loss function of the graph attention autoencoder, specifically:

[0032] The loss function L of the reconstructed spatial correlation features a :

[0033]

[0034] In the formula: n is the number of sample points, a represents the element in the adjacency matrix, and its value is 0 or 1; is the element in the reconstructed spatial correlation features ;

[0035] The loss function L of the classification prediction category b :

[0036]

[0037] In the formula: n is the number of sample points, c is the number of categories, y ij is the original sample label, is the predicted sample label probability distribution;

[0038] The loss function L of the graph attention autoencoder:

[0039]

[0040] According to the second aspect of the present invention, there is provided an electronic device, including a memory and a processor, wherein a computer program is stored on the memory, and when the processor executes the program, any one of the above methods is implemented.

[0041] According to the third aspect of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, any one of the above methods is implemented.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) The present invention extracts fault information features through a graph attention autoencoder, and also reconstructs a spatial correlation feature matrix based on the fault information features, so that the network can adaptively obtain the spatial correlation of sample features, and extract more fault information features according to the correlation, improving the accuracy of fault diagnosis.

[0044] (2) The present invention utilizes the attention coefficients of the graph neural network to inversely deduce several production parameters with the greatest weight for the fault diagnosis of the final product, providing an interpretability analysis study for the prediction results of the model. Description of the Drawings

[0045] Figure 1 is the flowchart of the method of the present invention;

[0046] Figure 2 is the flowchart of the manufacturing process fault diagnosis based on the graph attention autoencoder of the present invention.

[0047] Figure 3 is the architecture diagram of the graph attention autoencoder model provided by the present invention. Detailed Embodiments

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Embodiment

[0050] As Figure 1 shown, this embodiment provides a fault diagnosis method based on a graph attention convolutional autoencoder, and the method includes the following steps:

[0051] S1. Obtain the original fault data set D, perform a slicing operation on the original fault data set D, and perform a normalization operation on the sliced fault data set X1 and the original fault data set D respectively to obtain a normalized fault data set S(X) and a normalized original fault data set S(D);

[0052] S2. Calculate the distances between the variable feature vectors in the normalized fault data set S(D), and use the K-nearest neighbor algorithm to construct an adjacency matrix A;

[0053] S3. Input the adjacency matrix A and the normalized fault data set S(X) into the graph attention convolutional neural network model to extract spatial correlation features and fault information features Z;

[0054] S4. Input the fault information features Z into a two-layer fully connected layer, and use the softmax algorithm to calculate the classification prediction category

[0055] S5. Calculate the loss function of the reconstructed spatial correlation features and the classification prediction category The loss functions are combined into the loss function of the graph attention autoencoder, and the graph attention convolutional neural network model is trained through backpropagation of the loss function;

[0056] S6. Use the trained graph attention convolutional neural network model to perform fault diagnosis on the fault data set to be diagnosed, and output the classification prediction category.

[0057] Next, combined with Figure 2 and Figure 3 , the method of this embodiment will be introduced in detail.

[0058] Experimental environment: The hardware configuration used in the experiment is an Intel(R) Core(TM) i7-12700H processor and a GTX 3050 graphics card. The software environment is CUDA11.3 and cuDNN8.0, and the development environment is Windows 11. The graph attention autoencoder model is completed through Pycharm and the open-source deep learning framework Pytorch1.12.1.

[0059] S1. Obtain the original fault data set D, perform slicing operations on the original fault data set D, and perform standardization operations on the sliced fault data set X1 and the original fault data set D respectively to obtain the standardized fault data set S(X) and the standardized original fault data set S(D).

[0060] In this embodiment, the Tennessee-Eastman process (TEP) public data set is used. This data set mainly contains 41 measurement variables and 12 operating variables. A total of 21 different faults are set. Two fault data sets and a normal data set are randomly selected to form the overall data set D, with a total of 2400 data. Perform sliding window slicing operations on data sets of the same category, with a window size S = 20 and a step length L = 1. The divided data set is X m*n*S , that is, m = 2340, n = 53, and a 7:3 data division is adopted, that is, 1638 training data, and the remaining 702 are test data. For

[0061] D and X are standardized to obtain S(D) and S(X), and the conversion function is:

[0062]

[0063] where: x ∈ X, which is each data in X, u is the mean of the variables of this data, and σ is the standard deviation of the variables of this data.

[0064] S2. Calculate the distance between each variable feature vector in the standardized fault data set S(D), and use the K-nearest neighbor algorithm to construct the adjacency matrix A.

[0065] Calculate the distance between each variable feature vector in S(D):

[0066]

[0067] Where: d :,i is the i-th variable vector of matrix D, and dist(d :,i , d :,j ) is the distance between the i-th variable feature data and the j-th variable feature data in D. For each variable feature, the K nearest sample features are selected as adjacent samples to construct an adjacent matrix A, where A ∈ R n*n is a 0-1 sparse matrix.

[0068] In addition, the cosine distance can also be used for the distance between variable feature vectors:

[0069]

[0070] The cosine distance is not affected by the vector dimension and can better handle the direction information between vectors, and can extract more spatial feature information.

[0071] S3. Input the adjacent matrix A and the standardized fault data set S(X) into the graph attention convolutional neural network model to extract spatial correlation features and fault information features Z. The specific mathematical expression is:

[0072] Z = GAT(GAT(S(X))) (4)

[0073]

[0074] Where: GAT is the graph attention convolutional network, which is used as an encoder to extract the fault information feature Z of the input data; σ is the sigmoid activation function.

[0075] S4. Input the fault information feature Z into the double-layer fully connected layer and use the softmax algorithm to calculate the final classification prediction category of each sample The expression formula is as follows:

[0076]

[0077] Where: FC is the fully connected layer, and Z is the fault information feature extracted using the graph attention convolutional network.

[0078] S5. Calculate the loss function of the reconstructed spatial correlation feature and the loss function of the classification prediction category and merge them into the loss function of the graph attention autoencoder. Through the backpropagation of the loss function, the graph attention convolutional neural network model is trained to improve the quality prediction ability of the model.

[0079] In this embodiment, the reconstructed spatial association feature of the loss function L a :

[0080]

[0081] In the formula: n is the number of sample points, a represents the element in the adjacency matrix, and its value is 0 or 1; is the element in the reconstructed spatial association feature ;

[0082] Classification prediction category of the loss function L b :

[0083]

[0084] In the formula: n is the number of sample points, c is the number of categories, y ij is the original sample label, is the predicted sample label probability distribution;

[0085] The loss function L of the graph attention autoencoder:

[0086]

[0087] S6. Use the trained graph attention convolutional neural network model to perform fault diagnosis on the fault data set to be diagnosed, and output the classification prediction category.

[0088] The fault diagnosis method based on graph attention convolutional autoencoder proposed by the present invention abandons the traditional model that relies on manual parameter tuning, and instead deeply mines industrial data to effectively capture the mutual influence relationship between data. By fusing the spatial features of manufacturing parameters, this method improves the accuracy of industrial process fault diagnosis, thereby ensuring the smooth progress of the manufacturing process. In addition, in order to enhance the interpretability of the model, this method uses the attention coefficient of the graph neural network to inversely deduce several production parameters with the largest weights for the final product fault diagnosis, providing an interpretability analysis for the prediction results of the model. The advantage of this method is that it can automatically extract and abstract features from process data, and then directly establish the relationship between the learned features and the target output, effectively solving the problem of data explosion.

[0089] The electronic device of the present invention includes a central processing unit (CPU), which can execute various appropriate actions and processes according to the computer program instructions stored in the read-only memory (ROM) or the computer program instructions loaded from the storage unit to the random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus.

[0090] Multiple components in the device are connected to the I / O interface, including: an input unit, such as a keyboard, a mouse, etc.; an output unit, such as various types of displays, speakers, etc.; a storage unit, such as a disk, an optical disc, etc.; and a communication unit, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0091] The processing unit executes the various methods and processes described above, such as methods S1 - S6. For example, in some embodiments, methods S1 - S6 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 - S6 described above can be executed. Alternatively, in other embodiments, the CPU can be configured to execute methods S1 - S6 by any other suitable means (e.g., by means of firmware).

[0092] The functions described above herein can be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: Field Programmable Gate Arrays (FPGA), Application Specific Integrated Circuits (ASIC), Application Specific Standard Products (ASSP), Systems on Chip (SOC), Complex Programmable Logic Devices (CPLD), and so on.

[0093] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or a controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program codes are executed by the processor or the controller, the functions / operations specified in the flowchart and / or the block diagram are implemented. The program codes can be executed entirely on the machine, partially on the machine, executed partially on the machine and partially on a remote machine as an independent software package, or executed entirely on a remote machine or a server.

[0094] In the context of the present invention, a machine-readable medium may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0095] As described above, the foregoing are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily conceive of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A fault diagnosis method based on graph attention convolutional autoencoder, characterized in that: include: S1. Obtain the original fault data set D, perform a slicing operation on the original fault data set D, and perform standardization operations on the fault data set X and the original fault data set D obtained after slicing, respectively, to obtain a standardized fault data set S(X) and a standardized original fault data set S(D); S2, calculate the distance between the feature vectors of each variable in the standardized fault data set S(D), and use the K-adjacent algorithm to construct the adjacency matrix A; S3. Input the adjacency matrix A and the standardized fault dataset S(X) into the graph attention convolutional autoencoder model to extract spatial correlation features. and fault information feature Z; S4. Input the fault information feature Z into the double-layer fully connected layer and use the softmax algorithm to calculate the classification prediction category S5. Calculate and reconstruct spatial correlation features Loss function and classification prediction category The loss function is combined into the loss function of the graph attention autoencoder, and the graph attention convolutional neural network model is trained through back propagation of the loss function; S6. Use the trained graph attention convolutional autoencoder model to perform fault diagnosis on the fault dataset to be diagnosed and output the classification prediction category.

2. According to claim 1, a fault diagnosis method based on graph attention convolutional autoencoder is characterized in that: The slicing of the original fault data set D specifically includes: performing a sliding window slicing operation on the original fault data set D with a step size of 1 and a window size of 20.

3. According to the fault diagnosis method based on graph attention convolutional autoencoder according to claim 1, it is characterized in that: The standardized operation is specifically as follows: In the formula: x is the current data variable, x∈X; u is the mean of the current data variable; σ is the standard deviation of the current data variable.

4. A fault diagnosis method based on graph attention convolutional autoencoder according to claim 1, characterized in that: The distance between each variable feature vector in the standardized fault data set S(D) is calculated by the following expression: Where: d :,i is the i-th variable vector of matrix D; dist(d :,i ,d :,j ) is the distance between the i-th variable feature data and the j-th variable feature data in the matrix D.

5. A fault diagnosis method based on graph attention convolutional autoencoder according to claim 1, characterized in that: The distance between each variable feature vector is the cosine distance, and the calculation expression is: Where: d :，i is the i-th variable vector of matrix D; dist(d :，i , d ：，j ) is the distance between the i-th variable feature data and the j-th variable feature data in D; m is the number of variable vectors in matrix D.

6. A fault diagnosis method based on graph attention convolutional autoencoder according to claim 1, characterized in that: The adjacency matrix A and the standardized fault dataset S(X) are input into the graph attention autoencoder to extract spatial correlation features And the fault information feature Z, the mathematical expression is: Z=GAT(GAT(S(X))) (4) Where: GAT is the graph attention convolutional network, which is used as an encoder to extract the fault information feature Z of the input data; σ is the sigmoid activation function.

7. A fault diagnosis method based on graph attention convolutional autoencoder according to claim 1, characterized in that: The fault information feature Z is input into the double-layer fully connected layer, and the softmax algorithm is used to calculate the final classification prediction category of each sample The mathematical expression is: Where: FC is the fully connected layer.

8. A fault diagnosis method based on graph attention convolutional autoencoder according to claim 1, characterized in that: The calculation reconstructs the spatial correlation features Loss function and classification prediction category The loss function of the graph attention autoencoder is merged into the loss function of the graph attention autoencoder, which is: Reconstructing spatial association features The loss function L a : Where: n is the number of sample points, a represents the element of the adjacency matrix, and its value is 0 or 1; The spatial correlation features for reconstruction Elements in Classification prediction category The loss function L b : Where: n is the number of sample points, c is the number of categories, y ij is the original sample label, To predict the probability distribution of sample labels; The loss function L of the graph attention autoencoder is:

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images