Industrial production fault intelligent diagnosis method and system based on multi-modal data

Abstract

The invention relates to the technical field of industrial management, and provides an industrial production fault intelligent diagnosis method and system based on multi-modal data. Real-time data in the industrial production process are obtained through an intelligent sensor; projecting the real-time data of different modalities to a high-dimensional space, determining cross-modal distance measurement, and constructing decoupled spatial-temporal characteristics through a spatial-temporal alignment matrix generated by the distance measurement; constructing a state evolution process through an attention mechanism according to industrial nodes and the spatial-temporal characteristics in industrial production, and generating a dynamic topological graph according to the state evolution process; based on the dynamic topological graph and node features in the dynamic topological graph, generating a causal path of a fault through a causal discovery model, and screening a fault root node from industrial nodes according to the causal path; and feeding back the causal path and the fault root node to a fault management terminal, thereby realizing efficient, accurate and reliable fault diagnosis and processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial management, in particular to an industrial production fault intelligent diagnosis method and system based on multi-modal data. BACKGROUND

[0002] In modern industrial production, the equipment system is increasingly complex, often composed of multiple subsystems and numerous components, and single type of sensor data is difficult to fully reflect the running state of the equipment. At the same time, there are many types of faults in industrial production, not only mechanical faults such as equipment wear and tear, fracture, but also electrical faults, logic errors of control system, etc. Some faults are hidden in the early stage and are not easy to be detected. Some faults may be the result of the joint action of multiple factors, and it is difficult to accurately determine the fault cause by relying on a single data source. Therefore, the existing technology has the problems of low accuracy and reliability of fault diagnosis in industrial production process. SUMMARY

[0003] The present application provides an industrial production fault intelligent diagnosis method and system based on multi-modal data, thereby at least to some extent can solve the problem of low accuracy and reliability of fault diagnosis in industrial production process.

[0004] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the present application.

[0005] According to one aspect of the present application, an industrial production fault intelligent diagnosis method based on multi-modal data is provided, comprising: acquiring real-time data in industrial production process through intelligent sensors; projecting the real-time data of different modalities into a high-dimensional space to determine a cross-modal distance measure, generating a spatio-temporal alignment matrix through the distance measure, and constructing decoupled spatio-temporal features; constructing a state evolution process through an attention mechanism according to the industrial nodes in industrial production and the spatio-temporal features, generating a dynamic topology graph according to the state evolution process; generating a causal path of the fault based on the dynamic topology graph and the node features therein through a causal discovery model, and screening out a fault root node from the industrial nodes according to the causal path; feeding back the causal path and the fault root node to a fault management terminal.

[0006] In the present application, based on the foregoing scheme, the projecting the real-time data of different modalities into a high-dimensional space, determining a distance metric across modalities, generating a spatio-temporal alignment matrix through the distance metric, and constructing decoupled spatio-temporal features include: determining a kernel function based on the data points corresponding to the real-time data; projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a distance metric across modalities; determining a spatio-temporal alignment matrix between different modalities according to the distance metric, and generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and a graph attention mechanism.

[0007] In the present application, based on the foregoing scheme, the determining a kernel function based on the data points corresponding to the real-time data includes:

[0008]

[0009] wherein, represents a data point corresponding to real-time data, k and K represent the identification and the number of kernels respectively, respectively represent the hyperparameter and the kernel width corresponding to the kth kernel.

[0010] In the present application, based on the foregoing scheme, the projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a distance metric across modalities include:

[0011]

[0012] wherein, represents the ith data of the mth modality, represents the jth data of the nth modality, represents the data point corresponding to the kernel function.

[0013] In the present application, based on the foregoing scheme, the determining a spatio-temporal alignment matrix between different modalities according to the distance metric, and generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and a graph attention mechanism include: determining a spatio-temporal alignment matrix between different modalities according to the distance metric; and generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and a graph attention mechanism

[0014]

[0015] wherein, ​​​​denotes a graph attention mechanism, X denotes multimodal real-time data, A denotes a spatio-temporal alignment matrix, and W denotes a learnable weight, denotes an inverse of a modal covariance matrix, denotes a Hadamard product and a tensor product of matrices, respectively.

[0016] In the present application, based on the foregoing scheme, the state evolution process is constructed by an attention mechanism according to the industrial nodes in industrial production and the spatio-temporal features, and a dynamic topology graph is generated according to the state evolution process, comprising: determining the real-time state of the industrial nodes in industrial production based on the spatio-temporal features, calculating the time-varying coefficient of the state evolution process through the attention mechanism; constructing the state evolution process according to the industrial nodes in industrial production and the time-varying coefficient; predicting the evolution state of the industrial nodes at the next moment according to the state evolution process, and generating a dynamic topology graph through the evolution state.

[0017] In the present application, based on the foregoing scheme, the causal path of the fault is generated by a causal discovery model based on the dynamic topology graph and the node features therein, and the fault root node is screened out from the industrial nodes according to the causal path, comprising: determining the direct causal neighbors related to the industrial nodes through a causal discovery model according to the dynamic topology graph and the node features therein, generating a causal path of the fault; combining the causal path to generate a global causal graph, and screening out a fault root node from the industrial nodes according to the global causal graph.

[0018] In the present application, based on the foregoing scheme, the causal path and the fault root node are fed back to the fault management terminal, comprising: generating a fault report based on the information of the causal path and the fault root node; feeding back the fault report to the fault management terminal.

[0019] In the present application, based on the foregoing scheme, after generating the fault report based on the information of the causal path and the fault root node, the fault report is further encrypted.

[0020] According to one aspect of the present application, an industrial production fault intelligent diagnosis system based on multimodal data is provided, comprising:

[0021] An acquisition module is configured to acquire real-time data in an industrial production process through an intelligent sensor, wherein the real-time data includes vibration data, temperature data, pressure data, and image data;

[0022] A feature module is configured to project the real-time data of different modalities into a high-dimensional space, determine a cross-modal distance measure, generate a spatio-temporal alignment matrix through the distance measure, and construct decoupled spatio-temporal features;

[0023] a topology module, configured to construct a state evolution process according to the industrial nodes and the spatio-temporal features in the industrial production by an attention mechanism, and generate a dynamic topology graph according to the state evolution process;

[0024] a fault module, configured to generate a causal path of a fault based on the dynamic topology graph and the node features therein by a causal discovery model, and screen a root cause node of the fault from the nodes in the dynamic topology according to a length of the causal path;

[0025] a feedback module, configured to feed back the causal path and the root cause node of the fault to a fault management terminal.

[0026] According to an aspect of the present application, there is provided a computer readable medium having stored thereon a computer program, the computer program being executed by a processor to implement the industrial production fault intelligent diagnosis method based on multi-modal data as described in the above embodiments.

[0027] According to an aspect of the present application, there is provided an electronic device, comprising: one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the industrial production fault intelligent diagnosis method based on multi-modal data as described in the above embodiments.

[0028] According to an aspect of the present application, there is provided a computer program product or computer program, comprising computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the industrial production fault intelligent diagnosis method based on multi-modal data provided in the above various optional implementation manners.

[0029] In the technical solution of the present application, real-time data in the industrial production process is acquired through intelligent sensors; the real-time data of different modalities is projected into a high-dimensional space to determine a cross-modality distance metric, a spatio-temporal alignment matrix generated by the distance metric is used to construct decoupled spatio-temporal features; a state evolution process is constructed through an attention mechanism according to the industrial nodes and the spatio-temporal features in the industrial production, a dynamic topology graph is generated according to the state evolution process; a causal path of a fault is generated through a causal discovery model based on the dynamic topology graph and the node features therein, a fault root node is screened out from the industrial nodes according to the causal path; the causal path and the fault root node are fed back to a fault management terminal. The multi-modal real-time data is acquired through intelligent sensors to ensure comprehensive information. The decoupled spatio-temporal features are constructed by projection, which can accurately capture fault features. The dynamic topology graph is generated by means of the attention mechanism, which clearly presents the node relationship. The causal path and the fault root are found out through the causal model, and finally fed back to the terminal to realize efficient, accurate and reliable fault diagnosis and processing.

[0030] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0031] The drawings incorporated into the specification and forming part of the specification, show embodiments consistent with the present application, and together with the specification, serve to explain the principles of the present application. It is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0032] Figure 1 A flowchart of an industrial production fault intelligent diagnosis method based on multi-modal data in an embodiment of the present application is schematically shown.

[0033] Figure 2 A flowchart of generating spatio-temporal features in an embodiment of the present application is schematically shown.

[0034] Figure 3 A schematic diagram of an industrial production fault intelligent diagnosis system based on multi-modal data in an embodiment of the present application is schematically shown.

[0035] Figure 4 The structural schematic diagram of a computer system of an electronic device suitable for implementing embodiments of the present application is shown. DETAILED DESCRIPTION

[0036] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art.

[0037] Moreover, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that the

[0038] The block diagrams in the drawings show only the functionality of the example implementations and do not imply any particular physical or architectural arrangement of the example implementations. No inference should be made regarding the architecture or configuration of a hardware implementation from the description of the example implementations provided herein. Also, the functionality provided by the example implementations can be split into additional components not explicitly described, or combined with other example implementations.

[0039] The flow diagrams depicted in the drawings show example implementations only, and are not necessarily to scale. Also, the flow diagrams can not include all of the steps that are necessary for the practice of the example implementations. For example, one or more of the steps can be performed in parallel, or can be omitted in other embodiments. Furthermore, the flow diagrams can not necessarily be executed in the order shown, or in any particular order.

[0040] The following describes in detail the implementation details of the technical solutions of the present application:

[0041] Figure 1 A flowchart of an intelligent diagnosis method for industrial production faults based on multi-modal data according to an embodiment of the present application is shown. Referring to Figure 1 As shown, the intelligent diagnosis method for industrial production faults based on multi-modal data includes at least steps S110 to S150, which are described in detail as follows:

[0042] S110, acquiring real-time data in an industrial production process through an intelligent sensor.

[0043] In an embodiment of the present application, the real-time data includes vibration data, temperature data, pressure data, and image data, etc. These data have different sampling rates, time offsets, and spatial distributions, resulting in data heterogeneity.

[0044] In this embodiment, the device system is increasingly complex, often composed of multiple subsystems and numerous components. Single type of sensor data is difficult to fully reflect the running state of the device, while multi-modal data, such as vibration data, can reflect the mechanical operation of the device, temperature data can reflect the thermal state of the device, pressure data is related to the operation of the fluid system, and image data is helpful to observe the appearance of the device and the material situation in the process flow. Integrating these multi-modal data can more accurately diagnose the failure of complex equipment.

[0045] In industrial production sites, various intelligent sensors are deployed according to the layout of production equipment, key process steps and possible failure points. For example, vibration sensors are installed at key parts of rotating equipment such as motors and pumps to capture vibration signals generated during equipment operation; temperature sensors are installed in areas prone to temperature changes such as reaction kettles and pipelines; pressure sensors are installed in areas involving fluid delivery and pressure control; image sensors are installed at key visual inspection points on the production line to obtain image information such as product appearance and equipment operating status. After deployment, all sensors are initialized and set, including setting sampling frequency, range and communication parameters, to ensure that the sensors can start normally and work according to the predetermined requirements.

[0046] After the sensors collect raw data, preliminary processing is performed locally. For example, simple filtering is performed on vibration data to remove some high-frequency noise interference and make the data more accurate and reliable; temperature data is calibrated and compensated to eliminate the influence of environmental factors on measurement results; range checks are performed on pressure data to ensure that the data is within a reasonable range; image data is compressed and encoded to reduce data volume for more efficient transmission. After processing, the processed data is packaged in a specific format, adding timestamp, sensor number and other identification information to make the data have clear source and time attributes, facilitating subsequent data management and analysis.

[0047] The packaged real-time data is transmitted to the data receiving end through wired or wireless communication network. If it is wired transmission, the sensor sends data to the data acquisition server or industrial control system through the cable; if it is wireless transmission, the data is sent to the nearby wireless receiver using wireless communication protocols such as Wi-Fi, Bluetooth, ZigBee, etc., and then forwarded to the target system by the receiver. During transmission, the system will use certain data check and error correction mechanisms to ensure data integrity and accuracy. Once the data is successfully transmitted to the receiving end, it can provide real-time data support for subsequent industrial production monitoring, fault diagnosis and other applications.

[0048] The above process can capture various information in real time and accurately, such as vibration, temperature, pressure and image, etc. in the industrial production site through intelligent sensors. The rich real-time data provide a comprehensive and true basis for subsequent fault diagnosis, and the diagnosis work has a reliable basis, and subtle changes in the production process can be found in time.

[0049] S120, projecting the real-time data of different modalities into a high-dimensional space, determining a cross-modality distance measure, generating a spatio-temporal alignment matrix through the distance measure, and constructing decoupled spatio-temporal features.

[0050] In an embodiment of the present application, after obtaining the real-time data in the industrial production process, the data of different modalities are first projected into a high-dimensional space, so that the data has a more suitable expression form in the new space. Then a cross-modality distance measure is determined to measure the difference degree of data of different modalities in the high-dimensional space. Then a spatio-temporal alignment matrix is generated according to the distance measure, and the data is processed through the matrix, and finally the decoupled spatio-temporal features which can better reflect the industrial production state and fault characteristics are constructed.

[0051] As shown in Figure 2 In an embodiment of the present application, the real-time data of different modalities are projected into a high-dimensional space, a cross-modality distance measure is determined, a spatio-temporal alignment matrix is generated through the distance measure, and decoupled spatio-temporal features are constructed, comprising:

[0052] S210, determining a kernel function based on the data points corresponding to the real-time data;

[0053] S220, projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a cross-modality distance measure;

[0054] S230, determining a spatio-temporal alignment matrix between different modalities according to the distance measure, and generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and a graph attention mechanism.

[0055] In practical application, the real-time data in the embodiment include vibration signals, temperature sequences, pressure waveforms and visual image frame sequences, etc. They come from different types of sensors, and there are significant differences in sampling rate, time offset and spatial distribution. This leads to the problem of data heterogeneity, which brings challenges to subsequent analysis and processing. Before starting processing, the data needs to be preliminarily arranged and preprocessed, such as normalizing the data to unify the data of different dimensions to the same scale range, so as to facilitate subsequent unified analysis and processing. The embodiment enables the data of different modalities to be aligned in time and space, eliminates the influence of data dimension difference on subsequent calculation and comparison, lays a foundation for alignment and feature decoupling of multi-modal data, and facilitates subsequent feature extraction and analysis.

[0056] In an embodiment of the present application, by selecting a suitable kernel function, the data of different modalities is converted so that the features that are difficult to distinguish in the low-dimensional space become clear in the high-dimensional space. Based on the data points corresponding to the real-time data , the kernel function is determined as:

[0057]

[0058] wherein, represents the data points corresponding to the real-time data, k and K represent the identity and number of kernels respectively, represents the hyperparameter and kernel width corresponding to the kth kernel respectively.

[0059] In an embodiment of the present application, the data of different modalities is projected to a high-dimensional reproducing kernel Hilbert space by kernel function mapping. In the high-dimensional reproducing kernel Hilbert space, the features of the data can be better separated and represented, so as to more accurately define the cross-modality distance metric. According to the kernel function, the real-time data of different modalities is projected to a high-dimensional space, and the cross-modality distance metric is determined as:

[0060]

[0061] wherein, represents the ith data of the mth modality, represents the jth data of the nth modality. By calculating the distance between two data points in the high-dimensional space through the kernel function, mapping the data to the high-dimensional space and defining a suitable distance metric, the similarity between the data of different modalities can be more accurately measured, the complex relationship between the multi-modality data can be captured, and a basis for subsequent time alignment is provided. In this high-dimensional space, the features of the data can be better separated and represented. This is like in the low-dimensional space, the data of different modalities may be mixed together, but after being mapped to the high-dimensional space by the kernel function, they can be distributed according to their own characteristics, so as to more accurately measure the similarity between the data of different modalities and provide a more reliable basis for subsequent time alignment.

[0062] In an embodiment of the present application, since the data of different modalities can have a time offset, the distance metric across modalities calculated can find the spatio-temporal alignment matrix between the data of different modalities. This process is like finding corresponding points on different time tracks, and recording the relationship of these corresponding points to form a spatio-temporal alignment matrix. The spatio-temporal alignment matrix in the embodiment describes the alignment of data of different modalities in time, reflecting the correlation of multi-modal data in the time dimension. By constructing the spatio-temporal alignment matrix, the data of different modalities can be sequentially arranged and aligned in time, so that the subsequent feature extraction and analysis can be based on accurate time correspondence.

[0063] In an embodiment of the present application, the data aligned in space and time is decoupled by using a graph attention mechanism (GAT). The graph attention mechanism can learn the spatial dependency between modalities, which is like an intelligent analyzer that can identify which features in different modalities are correlated and which are independent. Through comprehensive analysis of the original multi-modal data and the spatio-temporal alignment matrix, GAT can extract more discriminative spatio-temporal features and decouple them. The decoupled features can better reflect the essential information of the data, removing the interference and redundancy between different modalities. Finally, a cross-modal feature matrix aligned in space and time is output, which contains high-quality feature information after alignment and decoupling, providing a reliable data basis for subsequent tasks such as fault diagnosis. Specifically, according to the spatio-temporal alignment matrix and the graph attention mechanism, the decoupled spatio-temporal features are generated as follows:

[0064]

[0065] wherein, represents the graph attention mechanism, X represents real-time multi-modal data, A represents the spatio-temporal alignment matrix, and W represents a learnable weight, represents the inverse of the modal covariance matrix, respectively represent the Hadamard product and the tensor product of matrices. By attention calculation on real-time data, the spatial dependency between modalities is learned, and feature decoupling is performed through the Kronecker product with the inverse of the covariance matrix to obtain the final cross-modal feature matrix aligned in space and time

[0066] In the embodiment, different modal data is projected into a high-dimensional space, which can better reveal the internal relationship between the data. Determining the cross-modal distance metric can accurately measure the difference and correlation degree of different modal data. By constructing the spatio-temporal alignment matrix and using the graph attention mechanism to learn the dependency between modalities, the data of different modalities can be aligned in space and time, and more discriminative spatio-temporal features can be decoupled, making the data features clearer and independent, facilitating subsequent analysis and improving the accuracy of subsequent fault diagnosis.​

[0067] S130, constructing a state evolution process according to the industrial nodes and the spatio-temporal features in the industrial production by using an attention mechanism, and generating a dynamic topology graph according to the state evolution process.

[0068] In the industrial production scenario of the embodiment, based on the existing industrial node information and the extracted spatio-temporal features, an attention mechanism is used to focus on the correlation degree of different industrial nodes with the spatio-temporal features at different times. Through this mechanism, the evolution process of the state of the industrial nodes over time can be clearly constructed, and the dynamic changes of the states of the nodes can be presented. According to the relationship and change rule between the nodes presented by the state evolution process, a dynamic topology graph reflecting the dynamic connection and interaction between the nodes in the industrial production system can be generated.

[0069] In an embodiment of the present application, according to the industrial nodes and the spatio-temporal features in the industrial production, a state evolution process is constructed by using an attention mechanism, and a dynamic topology graph is generated according to the state evolution process, comprising:

[0070] Based on the spatio-temporal features, the real-time state of the industrial nodes in the industrial production is determined, and the time-varying coefficient of the state evolution process is calculated by using an attention mechanism;

[0071] According to the industrial nodes and the time-varying coefficient in the industrial production, a state evolution process is constructed;

[0072] According to the state evolution process, the evolution state of the industrial nodes at the next time is predicted, and a dynamic topology graph is generated by using the evolution state.

[0073] In actual application, the traditional static graph model cannot reflect the dynamic changes of faults in time and space. In the embodiment, the dynamic topology graph can capture the path and intensity changes of fault propagation in real time, capture the dynamic topology relationship of fault propagation in industrial production, break through the limitations of the traditional static graph model, and provide support for more accurate fault diagnosis.

[0074] In an embodiment of the present application, first, a node set composed of industrial nodes is determined, and these industrial nodes can represent different devices, sensors or feature dimensions in industrial production. For each industrial node, at each time step, the time-varying coefficient of the time-varying graph neural network is calculated according to its neighbor node set by using an attention mechanism is:

[0075]

[0076] wherein, respectively represent the state of the industrial nodes v, u and k at time step t, denote a learnable weight matrix, b denotes a learnable parameter vector, and denotes a linear transformation in the multi-head attention mechanism; denote a set of neighbor nodes k of node v at time step t; k, v, and u denote the identities of industrial nodes, denote an activation function. In fault propagation, the fault correlation between certain nodes can be closer, and the multi-head attention mechanism can capture the correlation between nodes from multiple angles, more comprehensively and meticulously capture such complex dependency relationships, provide more reasonable weight information for the update of node states, and enable nodes to more accurately reflect the impact of faults on them, thereby improving the expression ability of the model.

[0077] In an embodiment of the present application, the time-varying graph neural network is a model designed to process dynamic graph structure data. In industrial production fault diagnosis, the fault propagation relationship is dynamically changing, and the time-varying graph neural network can update the node state and the connection relationship of the edge according to the time change. According to the industrial nodes and the spatio-temporal features in industrial production, a time-varying graph neural network is constructed. The state evolution process of the industrial nodes is represented as:

[0078]

[0079] wherein, denote the state of industrial node v at time steps t and t+1, respectively, denote an activation function, denote a learnable weight matrix, denote a set of neighbor nodes k of node v at time step t, denote the state of neighbor node u of industrial node v at time step t, denote the time-varying coefficient of node v to neighbor node u at time step t.

[0080] Through the node state evolution equation, the node state can be dynamically updated to reflect the propagation and influence of faults between different times and nodes, thereby providing a basis for constructing a dynamic topology graph. The above state evolution process of the industrial nodes describes how the state of node v at time t+1 is updated according to the state of itself and the neighbor nodes at time t. The evolution state of node v at time t+1 is updated according to the node state evolution equation, and this process is repeated on all nodes and time steps to gradually update the state of the entire dynamic topology graph.

[0081] ​In this embodiment, the dynamic topology graph includes information such as node sets, time-varying edge sets, and edge weights. It intuitively illustrates the dynamic relationships of fault propagation in industrial production. For example, the graph shows how a fault propagates from one node to another at different times, and how the intensity of propagation changes. This dynamic topology graph provides important structural information for subsequent fault diagnosis, helping us better understand the propagation paths and patterns of faults, thereby enabling us to take more effective measures for fault early warning and handling.

[0082] The above process, utilizing an attention mechanism combined with industrial nodes and spatiotemporal characteristics to construct a state evolution process, can focus on key information and dynamically reflect the changes in the state of industrial nodes over time. The resulting dynamic topology map can intuitively display the dynamic connections and interactions between nodes in the industrial production system, providing a clear diagram for analyzing fault propagation paths and facilitating the rapid location of the fault's impact range.

[0083] S140, based on the dynamic topology graph and the node characteristics therein, a causal path of the fault is generated through a causal discovery model, and the root cause node of the fault is selected from the nodes in the dynamic topology according to the length of the causal path.

[0084] In one embodiment of this application, a causal discovery model is used to deeply analyze the causal relationships between nodes based on the generated dynamic topology graph and the characteristics of each node in the graph. This model extracts logical clues about the generation and propagation of faults from a complex network of node relationships, thereby generating a causal path that clearly illustrates the development of the fault. Subsequently, tracing and analyzing along this causal path, the root cause node of the fault is accurately selected from numerous industrial nodes, providing crucial directional guidance for subsequent fault handling.

[0085] In one embodiment of this application, based on the dynamic topology graph and the node characteristics therein, a causal path for the fault is generated through a causal discovery model, and the root cause node of the fault is screened from the industrial nodes according to the causal path, including:

[0086] Based on the dynamic topology graph and the node characteristics therein, the direct causal neighbors related to the industrial node are determined by the causal discovery model, and the causal path of the fault is generated.

[0087] The causal paths are combined to generate a global causal graph, and the root cause nodes of the faults are selected from the industrial nodes based on the global causal graph.

[0088] In practical applications, the problem that the traditional method can only diagnose the fault type but cannot locate the root cause is solved, and support is provided for explainable maintenance. The traditional fault diagnosis method can only determine what kind of fault occurs in the system, but it is difficult to determine the starting point of the fault, and the step can accurately locate the fault root cause, thereby providing a clear direction for subsequent fault handling and maintenance.

[0089] Each node in the dynamic topology graph is traversed, and a Markov blanket of the node is calculated according to the structure and node characteristics of the dynamic topology graph. The Markov blanket includes a set of nodes that are directly related to the node v and can shield the influence of other nodes on v. By calculating the Markov blanket, the range of finding the causal relationship can be reduced.

[0090] For example, conditional independence test is performed on the node v and other nodes in the Markov blanket of the node v, to determine whether the node v and the node u are conditionally independent. If the value is approximately 0, it indicates that the node u is not a direct causal neighbor of the node v. In this way, the direct causal neighbors of the node v are screened out. By calculating the Markov blanket and performing the conditional independence test, the direct causal neighbors of each node can be accurately found, thereby laying a foundation for subsequent construction of a global causal graph, and improving the efficiency and accuracy of causal relationship discovery.

[0091] In an embodiment of the present application, the local causal relationships of the nodes are integrated to construct a preliminary global causal graph. Then, the structure of the global causal graph is optimized, and the structure parameter is minimized by adjusting the structure of the graph. Specifically, the structure parameter is :

[0092]

[0093] Wherein, v and V represent the identification and set of industrial nodes respectively, represent the event of failure of the industrial node v and the set of parent nodes of the industrial node v respectively represent the event of failure of the industrial node v and the set of parent nodes of the industrial node v respectively

[0094] In this embodiment, the structure of the global causal graph is optimized, and the optimal global causal graph that satisfies the minimum structure parameter is found by continuously adjusting the graph structure, so that the graph structure is more reasonable and concise, and the real causal relationship between the nodes is better reflected, thereby improving the reliability of causal reasoning.

[0095] In this embodiment, in order to determine the fault root node, the causal path length and the node abnormality degree are combined to determine the fault root score. The causal path length reflects the distance from the fault root node to other nodes, and the node abnormality degree represents the abnormal situation of the node.

[0096] Specifically, for each node in the global causal graph, a causal path from the node to the root node is determined, and an anomaly score of each node on the causal path is calculated. Then, according to the path length of the causal path and a decay factor, a contribution value of each node u to node v is calculated. Finally, the contribution values of all nodes on the path are added to obtain a fault root cause score of node v.

[0097] The embodiment can comprehensively consider the causal path and the anomaly degree of the node by the fault root cause score function, more accurately evaluate the possibility of each node as a fault root cause, and accurately locate the fault root cause node.

[0098] In the embodiment, the causal discovery model can deeply analyze the dynamic topology graph and the node features, sort out the logical relationship of fault generation and propagation, and generate a causal path. Along the causal path, the fault root cause node can be screened to accurately find the starting point of the fault, provide a clear target for the complete solution of the fault, and avoid misjudgment and processing of irrelevant nodes.

[0099] S150, feeding back the causal path and the fault root cause node to a fault management terminal.

[0100] In the embodiment, after determining the causal path and the fault root cause node, the two key information are arranged in a form convenient for transmission and understanding. Through a network communication channel, information containing the causal path and the fault root cause node is sent to the fault management terminal. After receiving the information, the relevant personnel can understand the propagation process of the fault according to the causal path, and determine the starting point of the fault according to the fault root cause node, so as to carry out fault troubleshooting and repair work.

[0101] In an embodiment of the present application, feeding back the causal path and the fault root cause node to the fault management terminal comprises:

[0102] generating a fault report based on the information of the causal path and the fault root cause node;

[0103] feeding back the fault report to the fault management terminal.

[0104] In an embodiment of the present application, after completing the cross-modal causal reasoning and fault root cause positioning, the key contents are extracted from the module storing the relevant information. The causal path details the propagation path and impact range of the fault in the industrial production system from the root cause node, and the fault root cause node accurately points out the location where the fault initially occurs. Then, according to the preset report template, the causal path and fault root cause node information are filled into the corresponding positions. The report template usually includes fault overview, root cause node description, causal path analysis, etc. These information is organized in logical order to form a clear structure and complete content of the fault report draft. At the same time, the information in the report is preliminarily checked for logic to ensure that the expression of the causal path and root cause node information is accurate and free of contradictions or ambiguities.

[0105] Before feeding back the fault report to the fault management terminal, the report is converted in format to ensure that the format of the report is compatible with the format supported by the fault management terminal. For example, if the terminal supports PDF format, the generated report is converted to PDF format. Then, the report is encrypted to protect the security and confidentiality of the fault information and prevent information leakage. The encryption process uses a specific encryption algorithm, and only authorized fault management terminals can decrypt and view the report content. At the same time, relevant identification information such as report generation time and report number is added to the report to facilitate subsequent management and query of the report.

[0106] Through the network communication module, the encrypted report is sent to the designated fault management terminal address. During transmission, the transmission status is monitored in real time to ensure that the report can be accurately and correctly delivered to the terminal. Once the terminal successfully receives the report, a feedback confirmation information is sent. If no confirmation information is received within a certain time, the report is re-sent to ensure that the report can be delivered in time. After the fault management terminal receives the report, relevant personnel can decrypt and view it, and according to the causal path and fault root cause node information in the report, develop specific fault handling schemes and carry out subsequent fault repair work.

[0107] The above process, by feeding back the key fault information to the fault management terminal, enables relevant personnel to timely understand the fault situation, grasp the fault development context according to the causal path, and develop targeted solutions according to the fault root cause node, so as to efficiently carry out fault troubleshooting and repair work and reduce the impact of the fault on industrial production.

[0108] The technical scheme of the application obtains real-time data in an industrial production process through an intelligent sensor; projects the real-time data of different modalities into a high-dimensional space to determine a cross-modality distance metric, generates a spatio-temporal alignment matrix through the distance metric, and constructs decoupled spatio-temporal features; constructs a state evolution process through an attention mechanism according to the industrial nodes and the spatio-temporal features in the industrial production, generates a dynamic topology graph according to the state evolution process; generates a causal path of a fault through a causal discovery model based on the dynamic topology graph and the node features therein, filters out a fault root node from the industrial nodes according to the causal path; and feeds back the causal path and the fault root node to a fault management terminal. The multi-modality real-time data is obtained through the intelligent sensor, ensuring comprehensive information. The real-time data is projected to construct decoupled spatio-temporal features, which can accurately capture fault features. The dynamic topology graph is generated through the attention mechanism, clearly presenting the node relationship. The causal path and the fault root are found out through the causal model, and finally fed back to the terminal, realizing efficient, accurate and reliable fault diagnosis and processing.

[0109] The following describes an embodiment of an industrial production fault intelligent diagnosis system based on multi-modality data of the application, which can be used to execute the industrial production fault intelligent diagnosis method based on multi-modality data in the above-mentioned embodiments of the application. It can be understood that the industrial production fault intelligent diagnosis system based on multi-modality data can be a computer program (including program code) running in a computer device, for example, the industrial production fault intelligent diagnosis system based on multi-modality data is an application software; the industrial production fault intelligent diagnosis system based on multi-modality data can be used to execute the corresponding steps in the method provided by the embodiments of the application. For details not disclosed in the embodiment of the industrial production fault intelligent diagnosis system based on multi-modality data of the application, please refer to the above-mentioned embodiments of the industrial production fault intelligent diagnosis method based on multi-modality data of the application.

[0110] Figure 3 A block diagram of an industrial production fault intelligent diagnosis system based on multi-modality data according to an embodiment of the application is shown.

[0111] Referring to Figure 3 According to an embodiment of the application, the industrial production fault intelligent diagnosis system based on multi-modality data includes:

[0112] The acquisition module 310 is configured to obtain real-time data in an industrial production process through an intelligent sensor, wherein the real-time data includes vibration data, temperature data, pressure data, and image data.

[0113] The feature module 320 is configured to project the real-time data of different modalities into a high-dimensional space to determine a cross-modality distance metric, generate a spatio-temporal alignment matrix through the distance metric, and construct decoupled spatio-temporal features.

[0114] a topology module 330, configured to construct a state evolution process according to the industrial nodes and the spatio-temporal features in the industrial production by an attention mechanism, and generate a dynamic topology graph according to the state evolution process;

[0115] a fault module 340, configured to generate a causal path of a fault based on the dynamic topology graph and the node features therein by a causal discovery model, and screen out a fault root node from the nodes in the dynamic topology according to a length of the causal path;

[0116] a feedback module 350, configured to feed back the causal path and the fault root node to a fault management terminal.

[0117] In the present application, based on the foregoing scheme, the projecting of the real-time data of different modalities into a high-dimensional space, the determination of a distance metric across modalities, the generation of a spatio-temporal alignment matrix by the distance metric, and the construction of decoupled spatio-temporal features include: determining a kernel function based on data points corresponding to the real-time data; projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a distance metric across modalities; determining a spatio-temporal alignment matrix between different modalities according to the distance metric, and generating decoupled spatio-temporal features by the spatio-temporal alignment matrix and a graph attention mechanism.

[0118] In the present application, based on the foregoing scheme, the determining of a kernel function based on data points corresponding to the real-time data includes: determining a kernel function based on data points corresponding to the real-time data . The kernel function is as follows:

[0119]

[0120] wherein, represents a data point corresponding to real-time data, k and K represent the identification and the number of kernels respectively, represents a hyperparameter and a kernel width corresponding to the kth kernel respectively.

[0121] In the present application, based on the foregoing scheme, the projecting of the real-time data of different modalities into a high-dimensional space according to the kernel function, and the determination of a distance metric across modalities include: projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a distance metric across modalities as follows:

[0122]

[0123] wherein, represents the ith data of the mth modality, represents the jth data of the nth modality, represents a data point corresponding to a kernel function.

[0124] In the present application, based on the foregoing scheme, the determination of the spatio-temporal alignment matrix between different modalities according to the distance metric, the generation of the decoupled spatio-temporal features through the spatio-temporal alignment matrix and the graph attention mechanism, comprises: determining the spatio-temporal alignment matrix between different modalities according to the distance metric; generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and the graph attention mechanism For:

[0125]

[0126] Among them, denotes the graph attention mechanism, X denotes real-time data of multi-modal, A denotes the spatio-temporal alignment matrix, W denotes the learnable weight, denotes the inverse of the modal covariance matrix, denote the Hadamard product and tensor product of the matrix respectively.

[0127] In the present application, based on the foregoing scheme, the construction of the state evolution process through the attention mechanism based on the industrial nodes in industrial production and the spatio-temporal features, the generation of the dynamic topology graph according to the state evolution process, comprises: determining the real-time state of the industrial nodes in industrial production based on the spatio-temporal features, calculating the time-varying coefficient of the state evolution process through the attention mechanism; constructing the state evolution process according to the industrial nodes in industrial production and the time-varying coefficient; predicting the evolution state of the industrial nodes at the next time according to the state evolution process, and generating a dynamic topology graph through the evolution state.

[0128] In the present application, based on the foregoing scheme, the generation of the causal path of the fault through the causal discovery model based on the dynamic topology graph and the node features therein, the screening of the fault root node from the industrial nodes according to the causal path, comprises: determining the direct causal neighbor related to the industrial nodes through the causal discovery model according to the dynamic topology graph and the node features therein, generating the causal path of the fault; combining the causal path to generate a global causal graph, and screening the fault root node from the industrial nodes according to the global causal graph.

[0129] In the present application, based on the foregoing scheme, the feedback of the causal path and the fault root node to the fault management terminal comprises: generating a fault report based on the information of the causal path and the fault root node; feeding back the fault report to the fault management terminal.

[0130] In the present application, based on the foregoing scheme, after generating the fault report based on the information of the causal path and the fault root node, it further comprises: encrypting the fault report.

[0131] The technical scheme of the application obtains real-time data in an industrial production process through an intelligent sensor; projects the real-time data of different modalities into a high-dimensional space to determine a cross-modality distance measure, generates a spatio-temporal alignment matrix through the distance measure, and constructs decoupled spatio-temporal features; constructs a state evolution process through an attention mechanism according to the industrial nodes and the spatio-temporal features in the industrial production, generates a dynamic topology graph according to the state evolution process; generates a causal path of a fault through a causal discovery model based on the dynamic topology graph and the node features therein, screens out a fault root node from the industrial nodes according to the causal path; and feeds back the causal path and the fault root node to a fault management terminal. The multi-modal real-time data is obtained through the intelligent sensor, ensuring comprehensive information. The decoupled spatio-temporal features are constructed by projection, which can accurately capture fault features. The dynamic topology graph is generated through the attention mechanism, clearly presenting the node relationship. The causal path and the fault root are found out through the causal model, and finally fed back to the terminal, realizing efficient, accurate and reliable fault diagnosis and processing.

[0132] Figure 4 A structural schematic diagram of a computer system of an electronic device suitable for implementing the embodiments of the application is shown.

[0133] It should be noted that the computer system of the electronic device in this embodiment is only an example, and should not impose any limitation on the functions and use range of the embodiments of the application.

[0134] The computer system in this embodiment includes a central processing unit 401, which can perform various appropriate actions and processes according to programs stored in a read-only memory 402 or loaded into a random access memory 403 from a storage part 408, such as the industrial production fault intelligent diagnosis method based on multi-modal data described in the above embodiments. Various programs and data required for system operation are also stored in the random access memory 403. The central processing unit 401, the read-only memory 402, and the random access memory 403 are connected to each other through a bus 404. An input / output interface 405 is also connected to the bus 404.

[0135] The following components are connected to the input / output interface 405: an input portion 406 including input devices such as a keyboard and a mouse; an output portion 407 including output devices such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage portion 408 including a hard disk; and a communication portion 409 including a network interface card such as a LAN (Local Area Network) card, a modem, and the like. The communication portion 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the input / output interface 405 as necessary. A removable media 411 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is attached to the drive 410 as necessary, so that a computer program read therefrom is installed in the storage portion 408 as necessary.

[0136] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present application. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing computer programs for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication portion 409, and / or installed from the removable media 411. When the computer program is executed by the central processing unit 401, various functions defined in the system of the present application are executed.

[0137] It should be noted that the computer-readable medium in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, the 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, device or apparatus. In the present application, the computer-readable signal medium can include a data signal carrying computer-readable computer programs in a baseband or as a part of a carrier wave. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit programs for use by or in conjunction with an instruction execution system, device or apparatus. The computer programs contained in the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, or the like, or any suitable combination thereof.

[0138] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In the flowcharts or block diagrams, each block can represent a module, a program segment or a part of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different orders than that shown in the drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, and they can also be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0139] The units described in the embodiments of the present application can be implemented by software, or can be implemented by hardware, and the units described can also be arranged in a processor. In some cases, the names of the units do not constitute a limitation on the units themselves.

[0140] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in the various optional implementation manners described above.

[0141] As another aspect, the present application also provides a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist separately without being assembled into the electronic device. The computer readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the intelligent diagnosis method of industrial production faults based on multi-modal data described in the above embodiments.

[0142] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, such division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into several modules or units.

[0143] From the above description of the embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk, etc.) or network, and includes several instructions to make a computing device (which can be a personal computer, server, touch terminal, or network device, etc.) execute the methods according to the embodiments of the present application.

[0144] Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the embodiments disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the application following the general principles thereof and including such departures from the present disclosure as come within known use or custom in the art.

[0145] It is to be understood that the application is not limited to the precise construction already described above and shown in the drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application should only be limited by the claims appended hereto.

Claims

1. An intelligent diagnosis method for industrial production faults based on multi-modal data, characterized in that, The method comprises the following steps: obtaining real-time data in an industrial production process through intelligent sensors; projecting the real-time data of different modalities into a high-dimensional space, determining a distance metric across modalities, generating a spatio-temporal alignment matrix through the distance metric, and constructing decoupled spatio-temporal features; constructing a state evolution process through an attention mechanism according to industrial nodes in the industrial production and the spatio-temporal features, and generating a dynamic topology graph according to the state evolution process; generating a causal path of a fault through a causal discovery model based on the dynamic topology graph and node features therein, and screening out a fault root node from the industrial nodes according to the causal path; feeding back the causal path and the fault root node to a fault management terminal.

2. The intelligent diagnosis method for industrial production faults based on multi-modal data according to claim 1, characterized in that, The method of projecting the real-time data of different modalities into a high-dimensional space, determining a distance metric across modalities, generating a spatio-temporal alignment matrix through the distance metric, and constructing decoupled spatio-temporal features comprises the following steps: determining a kernel function based on data points corresponding to the real-time data; projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a distance metric across modalities; determining a spatio-temporal alignment matrix between different modalities according to the distance metric, and generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and a graph attention mechanism.

3. The intelligent diagnosis method for industrial production faults based on multi-modal data according to claim 2, characterized in that, The method of determining a kernel function based on data points corresponding to the real-time data comprises the following steps: based on the data point corresponding to the real-time data , determine a kernel function is: ； wherein, represents a data point corresponding to real-time data, k and K represent the identity and number of kernels, respectively, represents a hyperparameter and kernel width corresponding to the kth kernel, respectively.

4. The intelligent diagnosis method for industrial production faults based on multi-modal data according to claim 3, characterized in that, The method of projecting the real-time data of different modalities into a high-dimensional space according to the kernel function, and determining a distance metric across modalities comprises the following steps: projecting the real-time data of different modalities to a high dimensional space according to the kernel function, determining a distance measure across modalities is: ； wherein, denotes the i-th data of the m-th modality, denotes the j-th data of the n-th modality, denotes a data point a corresponding kernel function.

5. The intelligent diagnosis method for industrial production faults based on multi-modal data according to claim 4, characterized in that, The method of determining a spatio-temporal alignment matrix between different modalities according to the distance metric, and generating decoupled spatio-temporal features through the spatio-temporal alignment matrix and a graph attention mechanism comprises the following steps: determining a spatio-temporal alignment matrix between different modalities according to the distance metric; The spatio-temporal features are generated through a spatio-temporal alignment matrix and a graph attention mechanism is: ； wherein, denotes the graph attention mechanism, X denotes the multimodal real-time data, A denotes the spatio-temporal alignment matrix, W denotes the learnable weights, denotes the inverse of the modal covariance matrix, denote the Hadamard product and the tensor product of matrices, respectively.

6. The intelligent diagnosis method for industrial production faults based on multi-modal data according to claim 1, characterized in that, The method of constructing a state evolution process through an attention mechanism according to industrial nodes in the industrial production and the spatio-temporal features, and generating a dynamic topology graph according to the state evolution process comprises the following steps: determining real-time states of the industrial nodes in the industrial production based on the spatio-temporal features, and calculating time-varying coefficients of the state evolution process through an attention mechanism; constructing a state evolution process according to the industrial nodes in the industrial production and the time-varying coefficients; predicting evolution states of the industrial nodes at the next moment according to the state evolution process, and generating a dynamic topology graph through the evolution states.

7. The intelligent diagnosis method for industrial production faults based on multi-modal data according to claim 1, characterized in that, The method of generating a causal path of a fault through a causal discovery model based on the dynamic topology graph and node features therein, and screening out a fault root node from the industrial nodes according to the causal path comprises the following steps: determining direct causal neighbors related to the industrial nodes through a causal discovery model based on the dynamic topology graph and node features therein, and generating a causal path of a fault; combining the causal path to generate a global causal graph, and screening out a fault root node from the industrial nodes according to the global causal graph. 8.The intelligent diagnosis method of industrial production fault based on multi-modal data according to claim 1, characterized in that, The method of feeding back the causal path and the fault root node to a fault management terminal comprises the following steps: generating a fault report based on information of the causal path and the fault root node; feeding back the fault report to a fault management terminal.

9. The intelligent diagnosis method of industrial production faults based on multi-modal data according to any one of claims 1-8, characterized in that, After generating a fault report based on information of the causal path and the fault root node, the method further comprises the following steps: encrypting the fault report.

10. An intelligent diagnosis system for industrial production faults based on multi-modal data, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire real-time data in an industrial production process through an intelligent sensor, wherein the real-time data includes vibration data, temperature data, pressure data, and image data; A feature module is configured to project the real-time data of different modalities into a high-dimensional space, determine a distance metric across modalities, generate a spatio-temporal alignment matrix through the distance metric, and construct decoupled spatio-temporal features; A topology module is configured to construct a state evolution process through an attention mechanism according to industrial nodes in the industrial production and the spatio-temporal features, and generate a dynamic topology graph according to the state evolution process; A fault module is configured to generate a causal path of a fault through a causal discovery model based on the dynamic topology graph and node features therein, and screen out a fault root node from nodes in the dynamic topology according to a length of the causal path; A feedback module is configured to feed back the causal path and the fault root node to a fault management terminal.

Images