Industrial fault diagnosis method and system based on graph data set distillation

Abstract

The invention discloses an industrial fault diagnosis method and system based on graph data set distillation. The method comprises the steps of obtaining operation and maintenance data of an industrial robot; performing distillation processing on the operation and maintenance data through a graph data set to obtain graph distillation data; wherein the graph data set distillation processing is used for compressing the operation and maintenance data; performing fault prediction on the industrial robot through a graph attention network according to the graph distillation data to obtain a fault detection result; wherein the graph attention network is obtained based on graph distillation data sample training. According to the embodiment of the invention, the operation and maintenance data of the industrial robot is processed by combining the graph data set distillation with the graph attention network, privacy protection of related data of the industrial robot can be realized, and meanwhile, the fault detection accuracy is improved. The method can be widely applied to the technical field of artificial intelligence detection.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence detection technology, and in particular to an industrial fault diagnosis method and system based on graph data set distillation. Background Art

[0002] Reducers and motors, as core components of electromechanical equipment such as industrial robots, have a significant impact on production due to their health status. Therefore, timely and accurate prediction of reducer and motor fault diagnosis is crucial to avoid the negative consequences of component failure. However, the monitoring data of these components contains a large amount of industrial robot privacy information. In order to protect privacy data, companies are reluctant to share data, resulting in data silos. In related technologies, fault detection is performed through a privacy data protection scheme of federated learning. However, this scheme has the risk of inferring sensitive information in the process of sharing model parameters. In addition, the federated learning system requires an efficient communication mechanism to achieve coordination among many participants, which may lead to network delays and synchronization problems. In addition, the global consistency of the model among all participants is maintained while maintaining an efficient training process. The above problems all affect the accuracy of fault detection. Summary of the invention

[0003] The purpose of the present invention is to solve one of the technical problems existing in the prior art to at least a certain extent.

[0004] To this end, an object of the present invention is to provide a highly accurate industrial fault diagnosis method and system based on graph dataset distillation.

[0005] In order to achieve the above technical objectives, one aspect of an embodiment of the present invention provides an industrial fault diagnosis method based on graph data set distillation, comprising the following steps: obtaining operation and maintenance data of an industrial robot; processing the operation and maintenance data through graph data set distillation to obtain graph distillation data; wherein the graph data set distillation processing is used to compress the operation and maintenance data; passing the graph distillation data through a graph attention network to perform fault prediction on the industrial robot to obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples. The embodiment of the present application processes the operation and maintenance data of an industrial robot through graph data set distillation combined with a graph attention network, which can achieve privacy protection for relevant data of the industrial robot and is conducive to improving the accuracy of fault detection.

[0006] In some embodiments, the industrial fault diagnosis method based on graph data set distillation of an embodiment of the present invention, wherein the operation and maintenance data is processed through graph data set distillation to obtain graph distillation data, comprises:

[0007] Build a graph dataset distillation model;

[0008] Fitting the gradient loss of the distilled data with a dynamic learning rate, and updating the graph dataset distillation model according to the gradient loss;

[0009] The operation and maintenance data is subjected to graph data set distillation processing through the updated graph data set distillation model to obtain graph distilled data.

[0010] In some embodiments, in one embodiment of the present invention, the method comprises:

[0011] Graph-structuring the operation and maintenance data to obtain an original graph data set;

[0012] Performing graph dataset distillation processing on the original graph dataset to obtain a synthetic graph dataset of graph distilled data;

[0013] The parameter matching process of the graph dataset distillation model is as follows:

[0014]

[0015] Where D(·,·) is the distance function, E is the number of training rounds, and opt θ is the update rule for the model parameters, They represent the model parameters trained on S and T at time step t respectively; L() is the loss function, GNN is the graph dataset distillation model, and θ is the parameter of the graph dataset distillation model; is the original graph dataset; S = {A′, X′, Y′} is the synthetic graph dataset.

[0016] In some embodiments, in one embodiment of the present invention, the update rule is performed in a step-down manner through the following steps:

[0017]

[0018] Where η is the learning rate of gradient descent, represents the gradient of the parameter θ.

[0019] In some embodiments, in one embodiment of the present invention, the graph-structuring the operation and maintenance data to obtain the original graph data set includes:

[0020] The operation and maintenance data is normalized, and then cut based on a sliding window to obtain a graph structure containing nodes and data graph features;

[0021] For any node, select other nodes whose similarity with the node is greater than or equal to a threshold to establish links, and obtain the original graph data set.

[0022] In some embodiments, in one embodiment of the present invention, the graph attention network is trained by the following steps:

[0023] Creating a weight for each neighboring node of the graph distillation data; the weight is used to characterize the degree of association between nodes;

[0024] According to the training results of the graph attention network, the weights are updated, and then the graph attention network is updated.

[0025] In some embodiments, in one embodiment of the present invention, the method further comprises:

[0026] According to the weights, determining the attention coefficients between the nodes, and then determining the node attention features;

[0027] Aggregate the attention features of all nodes to obtain time series features;

[0028] The temporal features of multiple graph attention networks are concatenated and linearly transformed to obtain intermediate features; the intermediate features are used as data input for the distillation of the next layer of graph data set.

[0029] On the other hand, an embodiment of the present invention proposes an industrial fault diagnosis system based on graph dataset distillation, comprising:

[0030] The first module is used to obtain the operation and maintenance data of industrial robots;

[0031] The second module is used to process the operation and maintenance data through a graph data set distillation process to obtain graph distillation data; wherein the graph data set distillation process is used to compress the operation and maintenance data;

[0032] The third module is used to pass the graph distillation data through a graph attention network to perform fault prediction on the industrial robot and obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples.

[0033] On the other hand, an embodiment of the present invention provides an industrial fault diagnosis device based on graph dataset distillation, comprising:

[0034] at least one processor;

[0035] at least one memory for storing at least one program;

[0036] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned industrial fault diagnosis method based on graph data set distillation.

[0037] On the other hand, an embodiment of the present invention provides a storage medium storing a program executable by a processor, wherein the program executable by the processor is used to implement the above-mentioned industrial fault diagnosis method based on graph data set distillation when executed by the processor.

[0038] The embodiments of the present application include at least the following beneficial effects: The method provided by the embodiments of the present invention includes: obtaining the operation and maintenance data of the industrial robot; processing the operation and maintenance data through a graph data set distillation to obtain graph distillation data; wherein the graph data set distillation is used to compress the operation and maintenance data; the graph distillation data is passed through a graph attention network to perform fault prediction on the industrial robot to obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples. The embodiments of the present application process the operation and maintenance data of the industrial robot through graph data set distillation combined with a graph attention network, which can achieve privacy protection for the relevant data of the industrial robot and is conducive to improving the accuracy of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 A schematic flow chart of an embodiment of an industrial fault diagnosis method based on graph data set distillation provided by the present invention;

[0041] Figure 2 A schematic flow chart of another embodiment of the industrial fault diagnosis method based on graph data set distillation provided by the present invention;

[0042] Figure 3 A schematic diagram of the structure of an embodiment of an industrial fault diagnosis process based on graph data set distillation provided by the present invention;

[0043] Figure 4 A schematic diagram of the structure of an embodiment of an industrial fault diagnosis system based on graph data set distillation provided by the present invention;

[0044] Figure 5 A schematic structural diagram of an embodiment of an industrial fault diagnosis device based on graph dataset distillation provided by the present invention. DETAILED DESCRIPTION

[0045] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0046] With the continuous development of modern manufacturing, the production of high-precision and complex products has put forward higher requirements for equipment. Reducers and motors, as core components of electromechanical equipment such as industrial robots, have a significant impact on production due to their health status. Therefore, timely and accurate prediction of reducer and motor fault diagnosis is crucial to avoid the negative consequences of component failure. However, the monitoring data of these components contains a large amount of industrial robot privacy information. Due to the protection of privacy data, enterprises are reluctant to share data, resulting in the phenomenon of data islands.

[0047] At present, the privacy data protection methods for monitoring data of industrial robots are mainly divided into two categories: privacy data protection methods based on federated learning and privacy data protection methods based on data set distillation. In privacy protection based on federated learning, although individual data remains private, there may be a risk of inferring sensitive information in the process of sharing model parameters. In addition, the federated learning system requires an efficient communication mechanism to achieve coordination among many participants, which may cause network delays and synchronization problems. In addition, maintaining the global consistency of the model among all participants while maintaining an efficient training process is also a major challenge. Dataset distillation compresses large training sets into small synthetic sets without compromising the training effect of deep neural networks, and the generated data also contributes to more efficient training. Therefore, the present invention protects the privacy of monitoring data of industrial robots by adopting a method based on graph dataset distillation and GAT.

[0048] Specifically, a federated learning framework in the related art is used to improve privacy protection and efficiency in industrial artificial intelligence. The method achieves efficient and privacy-enhanced federated learning through a non-interactive secure aggregation protocol and a distributed Gaussian mechanism. The framework uses an improved homomorphic encryption scheme to process private gradients, which improves processing speed and increases plaintext space. The effectiveness of the framework in post-quantum security and aggregator-agnostic security is demonstrated experimentally. Performance evaluation demonstrates its advantages in actual training accuracy and computational and communication overhead. The main workflow of the method is as follows:

[0049] 1) First, the improved homomorphic encryption technology is used to encrypt the private gradients in the federated learning process to protect data privacy.

[0050] 2) Apply a non-interactive secure aggregation protocol that allows model updates from multiple participants to be securely merged without leaking the original data.

[0051] 3) Finally, the distributed Gaussian mechanism is used to further enhance privacy protection while ensuring the effectiveness and accuracy of the learning model.

[0052] This approach has made progress in privacy protection and computational efficiency, but it still faces challenges in dealing with data heterogeneity and maintaining model privacy. Especially in industrial AI environments, data heterogeneity may lead to inconsistent model training results. At the same time, maintaining model privacy in a distributed environment while ensuring data utility and training efficiency is a complex balancing act. These challenges require further research and optimization to ensure that federated learning systems can effectively handle diverse and distributed data sources while protecting privacy.

[0053] Another related technology is the FedAR framework, which is designed for federated learning of distributed mobile robots. The core of this framework is real-time data collection and building trust and resource awareness, as well as applying the FedAR algorithm for client training, model quality assessment, and asynchronous learning processing. In addition, FedAR optimizes the performance of federated learning by improving the efficiency of data processing and ensuring data privacy. The design concept of the FedAR framework is to utilize distributed resources while reducing dependence on central servers and maintaining local processing of data, thereby achieving a safer and more efficient federated learning environment.

[0054] Although the FedAR framework has advantages in improving data processing efficiency and protecting privacy, it still faces challenges in dealing with data heterogeneity and maintaining model privacy. Data heterogeneity may lead to different model training effects, and maintaining efficient model training and privacy protection in a resource-constrained environment is a complex task. These problems require further research and innovative solutions to achieve comprehensive, efficient and secure applications of federated learning.

[0055] The present invention discloses a data set distillation method for industrial robots based on graph neural networks, and the steps include: 1) collecting multivariate operation sensor data of the robot; 2) using graph data set distillation technology to convert large original graphs into small, high-information synthetic graphs. This step includes matching the graph neural network training trajectories of the original graph and the synthetic graph, and optimizing the gradient matching loss. 3) Building and using a GAT network to facilitate subsequent learning of the features of the distilled graph data; 4) using the synthetic graph to train the GAT model, classifying through test data, and obtaining the final prediction results. The present invention generates an effective synthetic graph by using a graph data set distillation method, and uses a graph neural network to perform fault diagnosis on industrial robots; this method can effectively reduce the possibility of privacy data leakage while ensuring classification accuracy.

[0056] The industrial fault diagnosis method and system based on graph dataset distillation according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings. First, the industrial fault diagnosis method based on graph dataset distillation according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0057] Reference Figure 1 In an embodiment of the present invention, an industrial fault diagnosis method based on graph data set distillation is provided. The industrial fault diagnosis method based on graph data set distillation in the embodiment of the present invention can be applied to a terminal, or to a server, or can be software running in a terminal or a server, etc. The terminal can be a tablet computer, a laptop computer, a desktop computer, etc., but is not limited to this. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The industrial fault diagnosis method based on graph data set distillation in the embodiment of the present invention mainly includes the following steps:

[0058] Obtain operation and maintenance data of industrial robots;

[0059] Processing the operation and maintenance data through graph data set distillation to obtain graph distilled data; wherein the graph data set distillation processing is used to compress the operation and maintenance data;

[0060] The graph distillation data is passed through a graph attention network to perform fault prediction on the industrial robot to obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples.

[0061] The purpose of this invention is to propose a fault diagnosis method for industrial robots combined with graph dataset distillation, which ensures the accuracy of diagnosis while encrypting the privacy information of the data to reduce the risk of privacy data leakage:

[0062] 1) Solve the problems that may be caused by privacy data leakage. The monitoring data of industrial robots usually contains various privacy information. Enterprises are unwilling to share monitoring data due to concerns about the leakage of privacy information, which is likely to lead to data silos. Therefore, this phenomenon is circumvented by introducing graph dataset distillation.

[0063] 2) GAT is introduced to learn the local multi-scale features of distilled graph data to achieve accurate fault diagnosis of industrial robots. The key advantage of GAT is that it can focus on the most important parts of the graph and highlight important nodes by assigning different weights, thereby achieving a deep understanding of complex network structures. This enables the model to diagnose industrial robot faults more accurately, especially when dealing with data with complex relationships and multi-level features. Therefore, this method not only improves the accuracy of fault diagnosis, but also provides a more efficient technical means for the maintenance and fault prevention of industrial robots.

[0064] The complete technical solution of the present invention is as follows:

[0065] Figure 2 This is a flow chart of an industrial robot fault diagnosis method based on graph dataset distillation. The method mainly includes the following steps:

[0066] (1) Collect multiple operation and maintenance data such as feedback current and feedback torque of the motor and reducer of the industrial robot during normal operation and degradation to build a data set;

[0067] (2) The newly generated dataset is shuffled and divided into training and test sets in a fixed ratio of 8:2, and then labeled;

[0068] (3) Construct a graph attention network (GAT) and a graph dataset distillation algorithm to perform distillation operations on the training set;

[0069] (4) Inputting the graph distillation dataset generated in step (3) into GAT for training to obtain a trained model;

[0070] (5) Input the test set data into the GAT network trained in step (4) to evaluate the fault diagnosis effect of the model.

[0071] Figure 3 Architecture diagram of industrial robot fault diagnosis using graph attention network and graph dataset distillation.

[0072] Optionally, in an embodiment of the present invention, the step of processing the operation and maintenance data through graph data set distillation to obtain graph distillation data includes:

[0073] Build a graph dataset distillation model;

[0074] Fitting the gradient loss of the distilled data with a dynamic learning rate, and updating the graph dataset distillation model according to the gradient loss;

[0075] The operation and maintenance data is subjected to graph data set distillation processing through the updated graph data set distillation model to obtain graph distilled data.

[0076] In some possible implementations, the embodiments of the present application update the graph dataset distillation model through gradient loss.

[0077] Optionally, in one embodiment of the present invention, the method further comprises:

[0078] Graph-structuring the operation and maintenance data to obtain an original graph data set;

[0079] Performing graph dataset distillation processing on the original graph dataset to obtain a synthetic graph dataset of graph distilled data;

[0080] The parameter matching process of the graph dataset distillation model is as follows:

[0081]

[0082] Where D(·,·) is the distance function, E is the number of training rounds, and opt θ is the update rule for the model parameters, They represent the model parameters trained on S and T at time step t respectively; L() is the loss function, GNN is the graph dataset distillation model, and θ is the parameter of the graph dataset distillation model; is the original graph dataset; S = {A′, X′, Y′} is the synthetic graph dataset.

[0083] Optionally, in one embodiment of the present invention, the update rule is stepped down through the following steps:

[0084]

[0085] Where η is the learning rate of gradient descent, represents the gradient of the parameter θ.

[0086] Optionally, in one embodiment of the present invention, the graph-structuring the operation and maintenance data to obtain an original graph data set includes:

[0087] The operation and maintenance data is normalized, and then cut based on a sliding window to obtain a graph structure containing nodes and data graph features;

[0088] For any node, select other nodes whose similarity with the node is greater than or equal to a threshold to establish links, and obtain the original graph data set.

[0089] Optionally, in an embodiment of the present invention, the graph attention network is trained through the following steps:

[0090] Create weights for each adjacent node of the graph distilled data; the weights are used to characterize the degree of association between nodes;

[0091] According to the training result of the graph attention network, update the weights, and further update the graph attention network.

[0092] Optionally, in an embodiment of the present invention, the method further includes:

[0093] According to the weights, determine the attention coefficients between nodes, and further determine the node attention features;

[0094] Aggregate the attention features of all nodes to obtain temporal features;

[0095] Perform splicing and linear transformation processing on the temporal features of multiple graph attention networks to obtain intermediate features; the intermediate features are used as the data input for the next layer of graph dataset distillation.

[0096] The following introduces the relevant algorithm formulas used in the present invention:

[0097] Graph dataset distillation: Convert a large-scale graph into a smaller, information-rich synthetic graph. This step involves screening important nodes and learning a compressed representation of node features.

[0098] Graph dataset where A ∈ R N×N is the adjacency matrix, N is the number of nodes, X ∈ R N×d is the d-dimensional node feature matrix, Y ∈ {0,..., C - 1} N represents the node label, and there are C categories in total. The goal of graph compression is to learn a small synthetic graph dataset where A' ∈ R N×N′ , x' ∈ R N×D , Y' ∈ {0,..., C - 1} N′ , and N' << N, so that the GNN trained on can achieve performance comparable to that of the GNN trained on a larger . Therefore, this goal can be formulated as the following bilevel problem. The outer goal is to minimize the test loss of the GNN parameters trained on the final synthetic graph on the original graph. The inner goal is to make its gradient match the gradient of the original graph by training the GNN parameters on the synthetic graph. See Formulas (1) and (2) below:

[0099]

[0100] Among them, GNNθ represents the parameterized θGNN model, θ S represents the parameters of the model trained on S, and L represents the loss function used to measure the difference between the model prediction and the actual result, namely the cross entropy loss. In the graph dataset distillation, the parameter matching process of GNN is shown in the following formulas (3) to (5):

[0101]

[0102] where D(·,·) is a distance function, T is the number of steps in the entire training trajectory, and opt θ is the update rule for the model parameters, denote the model parameters trained on S and T at time step t, respectively. Since our goal is to match the parameters step by step, we consider the update rule opt θ Perform a single-step gradient descent, as shown in formula (6) and formula (7):

[0103]

[0104] Where η is the learning rate of gradient descent, represents the gradient of the parameter θ. And and The distance between them is usually small, so we can simplify the goal to a gradient matching process, which aims to measure the difference between the two gradient matrices in order to find similarities between the gradients of the synthetic and original data. This formula measures whether the direction of the gradient is consistent by calculating the cosine similarity between each column. As shown in formula (8):

[0105]

[0106] in and By θ t Instead, it is trained on small synthetic images. The distance D is further defined as the sum of the distances dis at each layer. Given two gradients at a certain layer and The distance dis(·,·) used for distillation is defined as follows:

[0107]

[0108] in is the i-th column vector of the gradient matrix. This formula ensures that the model trained on the synthetic graph can effectively imitate the model trained on the original graph by comparing the gradients of the synthetic data and the original data, so that the synthetic data can be used as a substitute. Through the above formula, we can achieve parameter matching through an effective strategy of gradient matching.

[0109] Through the above process, the final generated synthetic graph can greatly reduce the size of the original graph, but still maintain high information for the GNN model. In experiments, this method can significantly reduce the data size on datasets such as Reddit and Cora while maintaining classification performance similar to the original data.

[0110] Regarding the graph attention network: First, the data is normalized to prevent abnormal data from causing the model training to fail to converge. Then, a sliding window is used to cut out n local time domain data of size D. n will be used as the total number of nodes in subsequent processing, and D will be used as the number of time features contained in each node. Its purpose is to extract local fault features in this time segment.

[0111] The data obtained through the above steps is further cut with a sliding window of size N and constructed into a graph structure to obtain multiple graphs. Each graph contains L nodes, and each node contains D feature data. In the process of constructing the graph structure, cosine similarity is used to evaluate the correlation between two points, and a link is established with other nodes whose similarity reaches the threshold as the out-degree. The out-degree operation is completed in sequence to form a directed graph. GAT will implicitly create a W weight matrix to assign the association strength between each neighboring node as a reference for subsequent training, and W can flexibly adjust the weight size according to the model training results. Using W to implement self-attention on the node is shown in Formula 10, where Represents the characteristics of the i node, e ij is the attention coefficient of the edge between nodes i and j, see formula (10):

[0112]

[0113] The significance of the attention coefficient is to express the importance of node j to node i. In GAT, each node forms an attention coefficient with other nodes. This experiment only considers the attention coefficient between each node and its first-order neighbor. The softmax function is used to normalize it, and e ij The LeakyReLU activation function is used, see formula (11):

[0114]

[0115] In order to make the fault feature extraction process more stable and effective, a multi-head attention mechanism is used in GAT. On the one hand, multi-head attention can prevent overfitting problems. Each attention is independent of each other and generates prediction results of its own attention. On the other hand, each independent attention feature can be combined at the output to learn the information extracted by each attention. This paper uses the average value to aggregate the feature information of multi-head attention, as shown in Formula 12, where K represents the number of multi-head attentions, and the final output result is Aggregate all node attention features to form iGAT n , see formula (12) to formula (13):

[0116]

[0117]

[0118] By combining multiple GAT modules and extracting features at the scale of C multiple times, we can obtain time series features of different fine-grained levels. The time series features of different fine-grained levels are concatenated and converted into the input of the next layer of distilled graph dataset through a linear layer. Since multiple fine-grained feature information needs to be extracted, the data is output G after passing through GAT at different scales. in , where B is the batch size, D K is the feature dimension, which is used as the input of the distilled data, see formula (14):

[0119]

[0120] The embodiments of the present application achieve effective protection of the privacy information of monitoring data: by generating information-rich graph data from the monitoring data through graph data set distillation, only the key information used for fault diagnosis is retained in the generated data, and most of the remaining privacy data is protected.

[0121] The embodiment of the present application uses a graph attention network to achieve accurate fault diagnosis: by using a distilled graph dataset to train a GAT network, efficient fault diagnosis of the data can be achieved while the GAT network learns most of the information in the distilled dataset.

[0122] The embodiments of the present application can also use blockchain technology to achieve a distributed and tamper-proof recording method for monitoring data of industrial robot reducers and motors. This method can be used to ensure the security and privacy of industrial data, and in the industrial field, people are actively studying how to use blockchain to protect the integrity of data and trace the source of data.

[0123] The embodiment of the present application uses a graph data set distillation method to encrypt the monitoring data of the industrial robot reducer and motor to generate graph data that removes most of the key privacy features;

[0124] Utilize distilled datasets to train graph attention networks and achieve accurate fault diagnosis results.

[0125] In summary, the method provided in the embodiment of the present application includes: obtaining the operation and maintenance data of the industrial robot; processing the operation and maintenance data through a graph data set distillation to obtain graph distillation data; wherein the graph data set distillation is used to compress the operation and maintenance data; and performing fault prediction on the industrial robot through a graph attention network to obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples. The embodiment of the present application processes the operation and maintenance data of the industrial robot through graph data set distillation combined with a graph attention network, which can achieve privacy protection for the relevant data of the industrial robot and is conducive to improving the accuracy of fault detection.

[0126] Secondly, refer to the attached Figure 4 An industrial fault diagnosis system based on graph dataset distillation proposed according to an embodiment of the present invention is described.

[0127] Figure 4 1 is a schematic diagram of the structure of an industrial fault diagnosis system based on graph data set distillation according to an embodiment of the present invention. The system specifically includes:

[0128] The first module 410 is used to obtain operation and maintenance data of the industrial robot;

[0129] The second module 420 is used to process the operation and maintenance data through a graph data set distillation process to obtain graph distillation data; wherein the graph data set distillation process is used to compress the operation and maintenance data;

[0130] The third module 430 is used to pass the graph distillation data through a graph attention network to perform fault prediction on the industrial robot to obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples.

[0131] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0132] Reference Figure 5 , an embodiment of the present invention provides an industrial fault diagnosis device based on graph dataset distillation, comprising:

[0133] at least one processor 510;

[0134] At least one memory 520, used to store at least one program;

[0135] When the at least one program is executed by the at least one processor 510, the at least one processor 510 implements the industrial fault diagnosis method based on graph dataset distillation.

[0136] Similarly, the contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0137] An embodiment of the present invention also provides a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to execute the above-mentioned industrial fault diagnosis method based on graph data set distillation.

[0138] Similarly, the contents of the above method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0139] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0140] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0141] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several programs to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable programs for implementing the logical functions, and may be embodied in any computer-readable medium for use by a program execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch and execute a program from a program execution system, device or apparatus), or in conjunction with such program execution systems, devices or apparatuses. For purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by a program execution system, device or apparatus, or in conjunction with such program execution systems, devices or apparatuses.

[0143] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0144] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable program execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0145] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0146] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0147] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. An industrial fault diagnosis method based on graph dataset distillation, characterized in that: The following steps are involved: Obtain operation and maintenance data of industrial robots; Processing the operation and maintenance data through graph data set distillation to obtain graph distilled data; wherein the graph data set distillation processing is used to compress the operation and maintenance data; The graph distillation data is passed through a graph attention network to perform fault prediction on the industrial robot to obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples.

2. The industrial fault diagnosis method based on graph dataset distillation according to claim 1, characterized in that: The step of processing the operation and maintenance data through graph data set distillation to obtain graph distillation data includes: Build a graph dataset distillation model; Fitting the gradient loss of the distilled data with a dynamic learning rate, and updating the graph dataset distillation model according to the gradient loss; The operation and maintenance data is subjected to graph data set distillation processing through the updated graph data set distillation model to obtain graph distilled data.

3. The industrial fault diagnosis method based on graph dataset distillation according to claim 2 is characterized in that: The method comprises: Graph-structuring the operation and maintenance data to obtain an original graph data set; Performing graph dataset distillation processing on the original graph dataset to obtain a synthetic graph dataset of graph distilled data; The parameter matching process of the graph dataset distillation model is as follows: Where D(·,·) is the distance function, E is the number of training rounds, and opt θ is the update rule for the model parameters, They represent the model parameters trained on S and T at time step t respectively; L() is the loss function, GNN is the graph dataset distillation model, and θ is the parameter of the graph dataset distillation model; is the original graph dataset; S = {A′, X′, Y′} is the synthetic graph dataset.

4. The industrial fault diagnosis method based on graph dataset distillation according to claim 3 is characterized in that: The update rule performs ladder descent through the following steps: Where η is the learning rate of gradient descent, represents the gradient of the parameter θ.

5. The industrial fault diagnosis method based on graph dataset distillation according to claim 3 is characterized in that: The graph-structuring the operation and maintenance data to obtain an original graph data set includes: The operation and maintenance data is normalized, and then cut based on a sliding window to obtain a graph structure containing nodes and data graph features; For any node, select other nodes whose similarity with the node is greater than or equal to a threshold to establish links, and obtain the original graph data set.

6. The industrial fault diagnosis method based on graph dataset distillation according to claim 1, characterized in that: The graph attention network is trained by the following steps: Creating a weight for each neighboring node of the graph distillation data; the weight is used to characterize the degree of association between nodes; According to the training results of the graph attention network, the weights are updated, and then the graph attention network is updated.

7. The industrial fault diagnosis method based on graph dataset distillation according to claim 6, characterized in that: The method further comprises: According to the weights, determining the attention coefficients between the nodes, and then determining the node attention features; Aggregate the attention features of all nodes to obtain time series features; The temporal features of multiple graph attention networks are concatenated and linearly transformed to obtain intermediate features; the intermediate features are used as data input for the distillation of the next layer of graph data set.

8. An industrial fault diagnosis system based on graph dataset distillation, characterized in that: include: The first module is used to obtain the operation and maintenance data of industrial robots; The second module is used to process the operation and maintenance data through a graph data set distillation process to obtain graph distillation data; wherein the graph data set distillation process is used to compress the operation and maintenance data; The third module is used to pass the graph distillation data through a graph attention network to perform fault prediction on the industrial robot and obtain a fault detection result; wherein the graph attention network is trained based on graph distillation data samples.

9. An industrial fault diagnosis device based on graph dataset distillation, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the industrial fault diagnosis method based on graph dataset distillation as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to implement the industrial fault diagnosis method based on graph dataset distillation as described in any one of claims 1 to 7 when executed by the processor.

Images