A device fault feature extraction method and system based on graph cognitive convolutional network

Abstract

The application relates to a device fault judgment method and system based on a graph cognitive convolution network, the method comprising the following steps: positioning a fault device according to system data and sensor data to obtain a primary diagnosis graph; introducing expert diagnosis data into an eye movement tracking algorithm to simulate the diagnosis process of expert diagnosis and diagnose the positioned fault device to obtain a high-level diagnosis graph; fusing the primary diagnosis graph and the high-level diagnosis graph according to a set weight value, using a graph cognitive convolution network to extract fault features of the fused graph; and inputting the fault features into a classifier to obtain a fault judgment result. Compared with the prior art, the application has the advantages that fault detection is more accurate.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis of large equipment in smart factories, and in particular to a method and system for extracting equipment fault features based on graph cognitive convolutional networks. Background Technology

[0002] In the field of fault diagnosis for large equipment in smart factories, real-time monitoring of the factory's operation is necessary. While various system and sensor data provide support, the number of potentially faulty components is vast, necessitating fault diagnosis by maintenance engineers. Current diagnostic methods primarily rely on the maintenance engineers' experience, which fails to provide comprehensive and accurate fault diagnosis, resulting in limited practicality. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method and system for extracting device fault features based on graph cognitive convolutional networks.

[0004] The objective of this invention can be achieved through the following technical solutions:

[0005] A device fault diagnosis method based on graph cognitive convolutional networks includes the following steps:

[0006] S1. Locate the faulty equipment based on system data and sensor data to obtain a preliminary diagnostic diagram;

[0007] S2. Import expert diagnostic data into the eye-tracking algorithm, simulate the expert diagnostic process to diagnose the faulty device and obtain an advanced diagnostic map.

[0008] S3. Based on the set weight values, fuse the primary diagnostic map and the advanced diagnostic map, and use a graph cognitive convolutional network to extract the fault features of the fused map;

[0009] S4. Input the fault features into the classifier to obtain the fault judgment result.

[0010] Furthermore, after obtaining the fault judgment result, the domain knowledge graph is updated according to the fault judgment result, and the domain knowledge graph is imported into the localization in step S1 and the eye tracking in step S2, so as to directly output the result when the fault is detected again.

[0011] Furthermore, the classifier applies Dropout to randomly discard neurons, and the number of neurons in the last layer is equal to the number of fault categories.

[0012] Furthermore, the graph cognitive convolutional network includes multiple GCN layers, which use ReLU as the activation function.

[0013] Furthermore, after obtaining the fault features of the image, principal component analysis is used to reduce the dimensionality of the fault features.

[0014] A device fault diagnosis system based on graph cognitive convolutional networks includes a data perception subsystem, a graph generation subsystem, and a feature analysis subsystem.

[0015] The data sensing subsystem is used to acquire system data, sensor data, and expert diagnostic data;

[0016] The graph generation subsystem is used to locate faulty equipment based on system data and sensor data, and generate a primary diagnostic graph; to perform eye tracking on the located equipment fault based on expert diagnostic data to obtain an advanced diagnostic graph; and to fuse the primary and advanced diagnostic graphs according to set weight values.

[0017] The feature analysis subsystem uses a graph cognitive convolutional network to extract fault features from the fused graph and inputs the fault features into a classifier to obtain the fault judgment result.

[0018] Compared with the prior art, the present invention has the following advantages:

[0019] 1. This invention primarily acquires primary and advanced diagnostic images using methods such as system data localization and eye tracking, and then uses graph convolutional neural networks to extract features from the fused images. This hierarchical acquisition of device diagnostic images makes the information more complete, and the use of graph convolutional neural networks with good classification and clustering effects to extract features ensures that the extracted features more accurately reflect device faults, thereby improving the accuracy of fault diagnosis.

[0020] 2. After obtaining the fault information, the present invention uses principal component analysis to reduce the dimensionality, thereby reducing the computational complexity and making the fault diagnosis more efficient.

[0021] 3. After each fault diagnosis, the present invention updates the knowledge graph based on the diagnosis results and imports it into the previous graph generation step to continuously update and optimize the fault diagnosis process, so that when the same fault is encountered in the future, it can be resolved quickly and accurately. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the process of the present invention.

[0023] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0025] This embodiment provides a method for extracting equipment fault features based on graph cognitive convolutional networks, which can be applied to equipment inspection in various production lines, such as cold rolling production lines. The method specifically includes the following steps:

[0026] Step S1: Locate the faulty equipment based on system data and sensor data to obtain a preliminary diagnostic diagram;

[0027] Step S2: Based on expert diagnostic data, simulate the focus and diagnostic process of expert diagnosis for the located device faults and perform eye tracking to obtain an advanced diagnostic map.

[0028] Step S3: Based on the set weight function values, fuse the primary diagnostic map and the advanced diagnostic map, and use a graph cognitive convolutional network based on frequency domain convolution theory to extract the fault features of the fused map;

[0029] Step S4: Input the fault features into the classifier to obtain the fault judgment result.

[0030] In step S1, the sensor data mainly includes IoT data, which refers to data generated primarily by equipment and sensors on the production line. This data can intuitively reflect certain parameters and states of the equipment or environment, such as tension, torque, and speed. System data refers to data from various information systems, such as production line output and product quality. First, the sensor data and system data are fused to locate the fault. Then, the fault cause is matched based on the domain knowledge graph. After fusing the located fault and its cause, a preliminary diagnostic diagram is obtained.

[0031] In step S2, expert diagnostic data mainly refers to the data stored in the database after extracting key information from the expert's observations and thoughts during on-site diagnosis. Using an eye-tracking algorithm, the expert diagnostic data is imported into the eye-tracking reference values ​​to simulate the expert's focus and diagnostic process, obtaining gaze maps and heatmaps. Finally, through graph calculations, an advanced diagnostic map can be obtained.

[0032] After obtaining the primary and advanced diagnostic maps, the node attribute data is normalized.

[0033] In step S3, the fusion algorithm for the primary diagnostic map and the advanced diagnostic map specifically includes:

[0034] First, in multi-graph fusion algorithms, a single graph can be represented as G. n (V n E n ), where V represents a node in the graph, E represents an edge between two nodes, and G represents an edge between two nodes. f Represents the fusion diagram, W n (V n E n ) represents graph Gn The weight function. The formula for multi-graph fusion is as follows:

[0035] G f = ∑[G n (V n , E n )W n (V n , E n )]

[0036] The above multi-graph fusion algorithm will, on the one hand, superimpose the weights of common vertices and edges, and on the other hand, will not ignore the weights in the original graph, and can preserve the basic information of the graph before and after fusion.

[0037] The graph cognitive convolutional network includes two GCN layers. The GCN layer uses ReLU as the activation function to map the output of neurons. The output y output of the GCN layer is shown according to the following expression:

[0038] y output = σ(U g θ (λ)U T x)

[0039] In the formula, x is the input feature, σ(*) is the activation function, and g θ (λ) is the convolution kernel.

[0040] Then the data is normalized. The graph cognitive convolutional network model is to find a feature space and learn to map the source domain data and the target domain data in the feature space, so as to reduce the data distribution difference between the source domain data and the target domain data.

[0041] After obtaining the fault features, the principal component analysis method is used to reduce the dimension of the fault features. P simplifies the n-dimensional input data to r dimensions, where r < n. Simply put, the principal component analysis method (PCA) is a basic transformation. By rotating the coordinate axes and translating the coordinate origin, the difference between any axis (main axis) and the data points is minimized, and the high-variance orthogonal axes are removed after coordinate transformation, so that the transformed data has the maximum variance, and a reduced-dimension data set is obtained.

[0042] In step S4, the classifier can map the reduced-dimension data set to the sample space, and finally obtain the result of fault judgment. Among them, Dropout is applied to randomly discard some neurons. The number of neurons in the last layer (softmax layer) is the number of fault categories, and its result represents the fault label. The goal on the classifier is to reduce the fault classification error on the source domain data, which is manifested as minimizing the nll_loss loss function (based on pytorch) on the source domain data set to minimize the classification loss.

[0043] After obtaining the fault diagnosis results, these results can be imported into the domain knowledge graph, which in turn can be used for localization in step S1 and eye tracking in step S2. This way, when encountering the same fault in the future, the diagnosis results can be directly output, quickly and accurately identifying the fault without requiring recalculation.

[0044] This embodiment also provides a device fault diagnosis system based on graph cognitive convolutional networks, such as... Figure 2 As shown, it includes a data-aware subsystem, a graph generation subsystem, and a feature analysis subsystem:

[0045] The data sensing subsystem is used to acquire system data, sensor data, and expert diagnostic data;

[0046] The graph generation subsystem is used to locate faulty equipment based on system data and sensor data, and generate a primary diagnostic graph; to perform eye tracking on the located equipment faults based on expert diagnostic data to obtain an advanced diagnostic graph; and to fuse the primary and advanced diagnostic graphs according to set weight values.

[0047] The feature analysis subsystem uses a graph cognitive convolutional network to extract fault features from the fused graph and inputs these fault features into a classifier to obtain fault judgment results.

[0048] After obtaining the fault diagnosis results, the fault diagnosis results can be imported into the domain knowledge graph, and the domain knowledge graph can be imported into the data perception subsystem to affect the graph generation subsystem.

[0049] This embodiment further provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, this program implements the device fault diagnosis method based on graph cognitive convolutional networks mentioned in this embodiment of the invention. It can employ any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of this invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of this invention through logical analysis, reasoning, or limited experimentation on the basis of the prior art should be within the scope of protection defined by the claims.

Claims

1. A device fault diagnosis method based on graph cognitive convolutional networks, characterized in that, Includes the following steps: S1. Locate the faulty equipment based on system data and sensor data to obtain a preliminary diagnostic diagram; S2. Import expert diagnostic data into the eye-tracking algorithm, simulate the expert diagnostic process to diagnose the faulty device and obtain an advanced diagnostic map. S3. Based on the set weight values, fuse the primary diagnostic map and the advanced diagnostic map, and use a graph cognitive convolutional network to extract the fault features of the fused map; S4. Input the fault characteristics into the classifier to obtain the fault judgment result; The expert diagnostic data refers to the data stored in the database after extracting key information from what the experts see and think during on-site diagnosis; step S2 includes using an eye-tracking algorithm to import the expert diagnostic data into the reference values ​​of eye-tracking, simulating the focus and diagnostic process of expert diagnosis, obtaining a gaze map and a heat map, and finally obtaining an advanced diagnostic map through graph calculation. The graph cognitive convolutional network includes multiple GCN layers, which use ReLU as the activation function.

2. The device fault diagnosis method based on graph cognitive convolutional networks according to claim 1, characterized in that, After obtaining the fault judgment result, the domain knowledge graph is updated according to the fault judgment result. The domain knowledge graph is then imported into the localization in step S1 and the eye tracking in step S2, so that the result can be directly output when the fault is detected again.

3. The device fault diagnosis method based on graph cognitive convolutional networks according to claim 1, characterized in that, The classifier uses Dropout to randomly discard neurons, and the number of neurons in the last layer is the number of fault categories.

4. The device fault diagnosis method based on graph cognitive convolutional networks according to claim 1, characterized in that, After obtaining the fault features of the image, principal component analysis is used to reduce the dimensionality of the fault features.

5. A device fault diagnosis system based on graph cognitive convolutional networks, characterized in that, It includes a data perception subsystem, a graph generation subsystem, and a feature analysis subsystem: The data sensing subsystem is used to acquire system data, sensor data, and expert diagnostic data; The graph generation subsystem is used to locate faulty equipment based on system data and sensor data, and generate a primary diagnostic map; it imports expert diagnostic data into an eye-tracking algorithm, simulates the expert diagnostic process to diagnose the located faulty equipment, obtains a high-level diagnostic map, and merges the primary diagnostic map and the high-level diagnostic map according to the set weight values. The feature analysis subsystem uses a graph cognitive convolutional network to extract fault features from the fused graph and inputs the fault features into a classifier to obtain fault judgment results. The expert diagnostic data refers to the data stored in the database after extracting key information from what the experts see and think during on-site diagnosis; step S2 includes using an eye-tracking algorithm to import the expert diagnostic data into the reference values ​​of eye-tracking, simulating the focus and diagnostic process of expert diagnosis, obtaining a gaze map and a heat map, and finally obtaining an advanced diagnostic map through graph calculation. The graph cognitive convolutional network includes multiple GCN layers, which use ReLU as the activation function.

6. The device fault diagnosis system based on graph cognitive convolutional networks according to claim 5, characterized in that, After obtaining the fault judgment result, the feature analysis subsystem updates the domain knowledge graph based on the fault judgment result and imports the domain knowledge graph into the localization and eye tracking of the graph generation subsystem, so as to directly output the result when the fault is detected again.

7. A device fault diagnosis system based on graph cognitive convolutional networks according to claim 5, characterized in that, The classifier uses Dropout to randomly discard neurons, and the number of neurons in the last layer is the number of fault categories.

8. A device fault diagnosis system based on graph cognitive convolutional networks according to claim 5, characterized in that, After obtaining the fault features of the image, the feature analysis subsystem uses principal component analysis to reduce the dimensionality of the fault features.

Images