Motor fault diagnosis method and system
By establishing a motor dynamic model in virtual space and using generative adversarial networks and space-time graph convolutional neural networks, the problem of scarcity of motor failure data is solved, efficient fault diagnosis under limited data situations is achieved, and motor fault diagnosis is improved Accuracy and efficiency.
Patent Information
- Application Number
- CN202411283117.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2025-01-03
AI Technical Summary
In actual industrial scenarios, the collection of motor fault data is limited, resulting in limited application of fault diagnosis methods based on artificial intelligence in the real world, especially under the requirements of safe and stable operation of equipment, it is difficult to obtain sufficient fault data.
Through digital twin technology, a dynamic model of motor rotation machinery is established in virtual space, fault simulation is carried out to generate virtual fault data, and a generative adversarial network and space-time graph convolution neural network are used to improve the quality and accuracy of virtual data, and finally through space-time Graph convolutional neural network for troubleshooting.
It significantly improves the fault diagnosis performance under limited or no fault data, improves the accuracy and efficiency of diagnosis, and can effectively and accurately solve the problems of virtual data quality and multi-time series space-time dependence in motor fault diagnosis.
Smart Images
Figure CN119249875A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of motor rotating equipment, and in particular to a motor fault diagnosis method and system. Background Art
[0002] Fault diagnosis of motors has always been a hot research topic in the field of industrial equipment. Due to their complex structure and operation in harsh environments, they are prone to various faults such as imbalance, asymmetry, bearing wear and mechanical damage. This can lead to production interruptions, economic losses, and even safety risks. Therefore, reliable fault diagnosis is crucial. And in the actual operation of rotating machinery, it is common for multiple sensors to collect data simultaneously, so multivariate time series (MTS) has become more common in actual industrial scenarios. Multivariate time series classification (MTSC) involves capturing the characteristics of each time series while considering the complex dependencies between different variables, and has a wide range of practical applications.
[0003] At present, most fault diagnosis methods based on motor rotating machinery can be divided into signal processing-based methods and artificial intelligence-based methods. In particular, the prerequisite for artificial intelligence-based fault diagnosis is that the quantity and quality of training data are sufficient to establish an effective mapping relationship between the measurement signal and the health status of the equipment. However, in actual industrial scenarios, the requirements for safe and stable operation of equipment make the collection under fault conditions particularly limited. It is quite difficult to collect a large amount of fault data. Therefore, insufficient data for model training may hinder the widespread and successful application of artificial intelligence-based methods in the real world. Summary of the invention
[0004] To this end, the present invention provides a motor fault diagnosis method and system, which can extract fault information from the virtual space through digital twin technology. By combining deep learning and feature extraction technology, accurate fault diagnosis can be achieved, which significantly improves the diagnostic performance in cases with limited or no fault data.
[0005] In order to solve the above technical problems, the present invention provides a motor fault diagnosis method, comprising:
[0006] Establish the dynamic model of the motor rotating mechanical device;
[0007] Performing fault simulation on the dynamic model in a virtual space to obtain first virtual fault data corresponding to the fault type of the motor rotating mechanical device;
[0008] Collecting real fault data of the motor rotating mechanical device corresponding to the first virtual fault data;
[0009] Based on a generative adversarial network, reducing the difference between the first virtual fault data and the real fault data to obtain second virtual fault data with improved quality;
[0010] Based on the cosine similarity, calculating the similarity between the virtual vibration response corresponding to the second virtual fault data and the physical vibration response corresponding to the real fault data to verify the validity of the second virtual fault data;
[0011] Based on a spatiotemporal graph convolutional neural network, the second virtual fault data is modeled as a graph structure, each variable in the graph structure is represented as a node, and each node in a node set is randomly initialized to an embedding vector composed of trainable parameters;
[0012] Calculating the cosine similarity between the embedding vectors in the node set, and constructing an adjacency matrix based on the cosine similarity between the embedding vectors;
[0013] Based on the adjacency matrix, a fault diagnosis result is obtained.
[0014] In one embodiment of the present invention, the fault type includes rotor imbalance, asymmetry in the shaft, bearing inner ring fault and / or bearing outer ring fault.
[0015] In one embodiment of the present invention, based on a generative adversarial network, reducing the difference between the first virtual fault data and the real fault data to obtain second virtual fault data with improved quality includes:
[0016] Using the first virtual fault data as input of a generator to generate new virtual data;
[0017] Inputting the new virtual data and the real fault data into a discriminator, and judging whether the input data is the real fault data or the new virtual data by the discriminator;
[0018] The generator and the discriminator are trained adversarially so that the new virtual data generated by the generator gradually approaches the real fault data.
[0019] In one embodiment of the present invention, the cosine similarity is calculated as follows:
[0020]
[0021] Where P rν and V rν They represent the physical vibration response and virtual vibration response in the time domain under the same operating conditions and in the normal state, respectively.
[0022] In one embodiment of the present invention, the cosine similarity between the embedding vectors in the node set is calculated using the following formula:
[0023]
[0024] Among them, the embedding vector The embedding vector of the i-th node is represented as and They represent two different time series.
[0025] In one embodiment of the present invention, the adjacency matrix is expressed as follows:
[0026]
[0027] Among them, N represents the number of nodes, and the element e in the adjacency matrix ij Represents the cosine similarity between node i and node j.
[0028] In one embodiment of the present invention, obtaining a fault diagnosis result based on the adjacency matrix includes:
[0029] The adjacency matrix is input into the GCN layer in the spatiotemporal graph convolutional neural network, and the fault diagnosis result is obtained through the Softmax layer in the spatiotemporal graph convolutional neural network.
[0030] The present invention also provides a motor fault diagnosis system, comprising:
[0031] A modeling unit, used to establish a dynamic model of the motor rotating mechanical device;
[0032] A first virtual fault data generating unit, configured to perform fault simulation on the dynamic model in a virtual space to obtain first virtual fault data corresponding to the fault type of the motor rotating mechanical device;
[0033] A real fault data collection unit, used for collecting real fault data of the motor rotating mechanical device corresponding to the first virtual fault data;
[0034] A second virtual fault data generating unit, configured to reduce the difference between the first virtual fault data and the real fault data based on a generative adversarial network, so as to obtain second virtual fault data with improved quality;
[0035] a verification unit, configured to calculate, based on cosine similarity, a similarity between a virtual vibration response corresponding to the second virtual fault data and a physical vibration response corresponding to the real fault data, so as to verify the validity of the second virtual fault data;
[0036] an embedding vector acquisition unit, configured to model the second virtual fault data into a graph structure based on a spatiotemporal graph convolutional neural network, wherein each variable in the graph structure is represented as a node, and each node in a node set is randomly initialized to an embedding vector composed of trainable parameters;
[0037] An adjacency matrix construction unit, used to calculate the cosine similarity between the embedding vectors in the node set, and construct an adjacency matrix based on the cosine similarity between the embedding vectors;
[0038] The fault diagnosis result acquisition module is used to obtain the fault diagnosis result based on the adjacency matrix.
[0039] The above technical solution of the present invention has the following advantages compared with the prior art:
[0040] A motor fault diagnosis method and system described in the present invention generates virtual fault data of different fault types by constructing a digital twin model, further improving the quality of virtual data, and then adopts a spatiotemporal graph convolutional network to solve the problem of spatiotemporal dependency in multivariate time series. Finally, softmax is used for classification to obtain the final motor fault diagnosis result. By combining a generative adversarial network (WGAN) with a spatiotemporal graph convolutional neural network (ST-GCN), it is possible to efficiently and accurately solve the common virtual data quality problems in motor fault diagnosis and the spatiotemporal dependency problems in multivariate time series (MTS), thereby greatly improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.
[0042] Figure 1 This is a flow chart of the motor fault diagnosis method in Example 1 of the present invention.
[0043] Figure 2 It is a structural diagram of the dynamic model (virtual model) of the motor rotating mechanical device in Example 1 of the present invention.
[0044] Figure 3 This is a schematic diagram of simulated fault types in Example 1 of the present invention.
[0045] Figure 4 Schematic diagram of the WGAN network structure in Example 1 of the present invention.
[0046] Figure 5 It is a comparison diagram between physical data and virtual data in a normal state in Example 1 of the present invention.
[0047] Figure 6(a)-(f) are diagrams showing the results of motor fault diagnosis in Example 1 of the present invention. DETAILED DESCRIPTION
[0048] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
[0049] In the present invention, if directions (up, down, left, right, front and back) are described, it is only for the convenience of describing the technical solution of the present invention, and does not indicate or imply that the technical features referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention.
[0050] In the present invention, "several" means one or more, "multiple" means more than two, "greater than", "less than", "exceed" and the like are understood to exclude the number itself; "above", "below", "within" and the like are understood to include the number itself. In the description of the present invention, if there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.
[0051] In the present invention, unless otherwise clearly defined, the words "set", "install", "connect" and the like should be understood in a broad sense, for example, they can be directly connected or indirectly connected through an intermediate medium; they can be fixedly connected or detachably connected or integrally formed; they can be mechanically connected or electrically connected or able to communicate with each other; they can be the internal connection of two elements or the interaction relationship between two elements. Those skilled in the art can reasonably determine the specific meanings of the above words in the present invention in combination with the specific content of the technical solution.
[0052] Example 1
[0053] Reference Figure 1 As shown, a motor fault diagnosis method according to an embodiment of the present invention includes:
[0054] S1. Establish the dynamic model (virtual model) of the motor rotating mechanical device. Figure 2 as shown.
[0055] As the core component of the virtual space, the dynamic model is designed to accurately replicate the vibration characteristics of the physical entity. In this embodiment, the dynamic model of the motor rotating mechanical device is established by using the automatic dynamic analysis software for mechanical systems (ADMAS). The components involved in the dynamic model include motors, bearings, rotors, shafts, etc. Among them, bearings, rotors, and shafts are regarded as rigid bodies, and their performance is mainly determined by geometric and material parameters. The failure mode is directly reflected by the geometric shape, and the speed, workload, and input speed are set according to the data actually collected on site to simulate the real working conditions.
[0056] S2. Perform fault simulation on the dynamic model in a virtual space to obtain first virtual fault data corresponding to the fault type of the motor rotating mechanical device.
[0057] Reference Figure 3 As shown, the fault types include rotor imbalance, asymmetry in the shaft, bearing inner race fault and / or bearing outer race fault.
[0058] ADMAS is used to establish a virtual model of the motor rotating machinery, and then a fault simulation is performed in a virtual space based on the fault state description to generate and store first virtual fault data corresponding to the fault type.
[0059] S3. Collecting real fault data of the motor rotating mechanical device corresponding to the first virtual fault data.
[0060] S4. Based on a generative adversarial network (WGAN), reduce the difference between the first virtual fault data and the real fault data to obtain second virtual fault data with improved quality.
[0061] Since the requirements for safe and stable operation of equipment in real industrial scenarios make the acquisition of fault conditions particularly limited, thus affecting diagnosis, digital twins are used to solve the problem of scarce fault data. Digital twins refer to the creation of virtual models in virtual space that are highly consistent with real physical entities (such as equipment, systems or processes) through digital technology, and can simulate, analyze and predict the behavior, status and performance of entities in real time or near real time.
[0062] In order to reduce the difference between virtual data and the real data collected, and further improve the usability of virtual data in fault diagnosis, a generative adversarial network (WGAN) is used to solve the above problems. Its overall structure is as follows: Figure 4 It encourages the generator to explore the entire data distribution and make the virtual data closer to the real data. At the same time, WGAN adopts an adaptive stopping criterion. During the training process, the generator does not need to continue training to the maximum number of iterations. When the difference between the generated virtual data and the real data reaches a certain level, the training can be automatically stopped to avoid overfitting.
[0063] It should be noted that the Generative Adversarial Network (GAN) is a deep learning architecture that includes:
[0064] Generator: Responsible for generating "virtual" data. It generates new data from random noise (or some initial input, such as virtual data) that is as close to the real data as possible.
[0065] Discriminator: Responsible for distinguishing whether the input data is "real" data (i.e. data collected from the real world) or "virtual" data generated by the generator. These two models are trained in an "adversarial" process:
[0066] The goal of the generator is to deceive the discriminator into thinking that the virtual data it generates is real data. The goal of the discriminator is to distinguish between real data and virtual data as accurately as possible. Through this adversarial training, the generator gradually learns to generate more realistic data, making its output closer and closer to real data.
[0067] Through adversarial training of the generator and the discriminator, the generated virtual fault data gradually approaches the real data, thereby improving the accuracy of the model.
[0068] The specific process includes:
[0069] Using the first virtual fault data as input of a generator to generate new virtual data;
[0070] Inputting the new virtual data and the real fault data into a discriminator, and judging whether the input data is the real fault data or the new virtual data by the discriminator;
[0071] The generator and the discriminator are trained adversarially so that the new virtual data generated by the generator gradually approaches the real fault data.
[0072] To avoid overfitting or overtraining, WGAN introduces an adaptive stopping criterion. When the difference between the generated virtual data and the real data reaches a certain threshold, the training process can be automatically stopped. This not only saves computing resources, but also ensures that the quality of the generated virtual data is high enough.
[0073] After multiple iterations and adversarial training, the generator is finally able to generate second virtual fault data that is very close to the real fault data and can better reflect the real fault situation.
[0074] S5. Based on cosine similarity, calculate the similarity between the virtual vibration response corresponding to the second virtual fault data and the physical vibration response corresponding to the real fault data to verify the validity of the second virtual fault data.
[0075] Through the above method, the feasibility of the constructed virtual model and the usability of WGAN can be verified. By introducing cosine similarity, the virtual model effectively reflects the vibration characteristics under different health states, and then the information interaction between the physical vibration response and the virtual vibration response is used to verify the validity of the virtual model.
[0076] Since the data collected in the physical space are mainly vibration acceleration signals, and the operating state in the virtual model is consistent with the operating state of the physical entity, verification is performed by utilizing the information interaction between the physical vibration response and the virtual vibration response.
[0077] The cosine similarity is calculated as follows:
[0078]
[0079] Where P rν and V rν They represent the physical vibration response and virtual vibration response in the time domain under the same operating conditions and in the normal state, respectively.
[0080] Reference Figure 5 As shown in Figure 2, the cosine similarity of these data within a certain time range is calculated to be 0.732, which can quantitatively illustrate that the two sets of data have good similarity. Therefore, the constructed virtual model can be said to be very accurate in generating virtual data.
[0081] S6. Based on a spatiotemporal graph convolutional neural network (ST-GCN), the second virtual fault data is modeled as a graph structure, each variable in the graph structure is represented as a node, and each node in the node set is randomly initialized to an embedding vector composed of trainable parameters.
[0082] ST-GCN combines the characteristics of convolutional neural network (CNN) and graph convolutional neural network (GCN), and can process information in both time and space dimensions. Each spatiotemporal convolution module consists of three submodules:
[0083] CNN layer: used to extract local time features from time series data. Convolutional neural networks perform sliding operations on the time dimension and can capture signal change trends within a short time range, such as sudden vibrations or current changes when a fault occurs.
[0084] GCN layer: used to extract spatial features, that is, the dependencies between different signals (nodes). Graph convolutional neural networks operate on graph structures and can capture the mutual influence between different signals.
[0085] Batch Normalization (BN) layer: Using a BN layer after each spatiotemporal convolution block can improve the convergence speed of the model and the stability of training, and prevent the gradient vanishing or exploding problem.
[0086] The second virtual fault data (virtual multivariate time series data) with improved quality is modeled as a graph structure, where each variable is represented as a node, the dependencies between variables are represented as edges, and then the node set is randomly initialized with an embedding vector consisting of trainable parameters The embedding vector of the i-th node is represented as k represents the dimension of node embedding.
[0087] S7. Calculate the cosine similarity between the embedding vectors in the node set, and construct an adjacency matrix based on the cosine similarity between the embedding vectors.
[0088] In order to obtain the dependency relationship between nodes, cosine similarity is used to construct the adjacency matrix between nodes. Cosine similarity measures the similarity between two embedding vectors. The closer the value is to 1, the more similar the features of the two nodes are, and the closer it is to -1, the less similar their features are.
[0089] The cosine similarity between the embedding vectors in the node set is calculated using the following formula:
[0090]
[0091] Among them, the embedding vector The embedding vector of the i-th node is represented as and They represent two different time series.
[0092] The adjacency matrix is represented as follows:
[0093]
[0094] Among them, N represents the number of nodes, and the element e in the adjacency matrix ij Represents the cosine similarity between node i and node j.
[0095] The adjacency matrix constructed by cosine similarity can effectively obtain the dependency relationship between different time series variables.
[0096] S8. Obtaining a fault diagnosis result based on the adjacency matrix. Including:
[0097] The adjacency matrix is input into the GCN layer in the spatiotemporal graph convolutional neural network to perform a convolution operation, thereby extracting the features of each node. After multiple layers of GCN convolution, the features of the nodes are passed through the Softmax layer in the spatiotemporal graph convolutional neural network to obtain the fault diagnosis results.
[0098] The Softmax layer converts the feature vector output by the last GCN layer into a probability distribution. The probability value of each category represents the probability that the sample belongs to a certain fault type. The final fault diagnosis result is determined by the category corresponding to the maximum probability.
[0099] This example uses a motor dataset from the Human Laboratory of the Noise and Vibration Control Center of the Department of Mechanical Engineering of the Korea Advanced Institute of Science and Technology for experimental testing, and compares it with other models and methods to verify the effectiveness of this method.
[0100] The final motor fault diagnosis results are as follows: Figure 6 As shown in the figure, after comparison with other baseline methods, the accuracy of this experiment can reach more than 90%.
[0101] In response to the problem of motor fault diagnosis, the present invention proposes a motor fault diagnosis method that combines digital twins with spatiotemporal graph convolutional networks, and implements simulation experiments based on relevant data. Through experiments, the effectiveness of the method model in the present invention is proved.
[0102] Example 2
[0103] Based on the same inventive concept, this embodiment provides a motor fault diagnosis system, the principle of solving the problem is similar to that of the motor fault diagnosis method, and the repeated parts are not repeated here.
[0104] This embodiment provides a motor fault diagnosis system, including:
[0105] A modeling unit, used to establish a dynamic model of the motor rotating mechanical device;
[0106] A first virtual fault data generating unit, configured to perform fault simulation on the dynamic model in a virtual space to obtain first virtual fault data corresponding to the fault type of the motor rotating mechanical device;
[0107] A real fault data collection unit, used for collecting real fault data of the motor rotating mechanical device corresponding to the first virtual fault data;
[0108] A second virtual fault data generating unit, configured to reduce the difference between the first virtual fault data and the real fault data based on a generative adversarial network, so as to obtain second virtual fault data with improved quality;
[0109] a verification unit, configured to calculate, based on cosine similarity, a similarity between a virtual vibration response corresponding to the second virtual fault data and a physical vibration response corresponding to the real fault data, so as to verify the validity of the second virtual fault data;
[0110] an embedding vector acquisition unit, configured to model the second virtual fault data into a graph structure based on a spatiotemporal graph convolutional neural network, wherein each variable in the graph structure is represented as a node, and each node in a node set is randomly initialized to an embedding vector composed of trainable parameters;
[0111] An adjacency matrix construction unit, used to calculate the cosine similarity between the embedding vectors in the node set, and construct an adjacency matrix based on the cosine similarity between the embedding vectors;
[0112] The fault diagnosis result acquisition module is used to obtain the fault diagnosis result based on the adjacency matrix.
[0113] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0114] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0115] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0117] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.
Claims
1. A motor fault diagnosis method, characterized in that: include: Establish the dynamic model of the motor rotating mechanical device; Performing fault simulation on the dynamic model in a virtual space to obtain first virtual fault data corresponding to the fault type of the motor rotating mechanical device; Collecting real fault data of the motor rotating mechanical device corresponding to the first virtual fault data; Based on a generative adversarial network, reducing the difference between the first virtual fault data and the real fault data to obtain second virtual fault data with improved quality; Based on the cosine similarity, calculating the similarity between the virtual vibration response corresponding to the second virtual fault data and the physical vibration response corresponding to the real fault data to verify the validity of the second virtual fault data; Based on a spatiotemporal graph convolutional neural network, the second virtual fault data is modeled as a graph structure, each variable in the graph structure is represented as a node, and each node in a node set is randomly initialized to an embedding vector composed of trainable parameters; Calculating the cosine similarity between the embedding vectors in the node set, and constructing an adjacency matrix based on the cosine similarity between the embedding vectors; Based on the adjacency matrix, a fault diagnosis result is obtained.
2. A motor fault diagnosis method according to claim 1, characterized in that: The fault types include rotor imbalance, asymmetry in the shaft, bearing inner race fault and / or bearing outer race fault.
3. A motor fault diagnosis method according to claim 1, characterized in that: Based on the generative adversarial network, reducing the difference between the first virtual fault data and the real fault data to obtain second virtual fault data with improved quality, including: Using the first virtual fault data as input of a generator to generate new virtual data; Inputting the new virtual data and the real fault data into a discriminator, and judging whether the input data is the real fault data or the new virtual data by the discriminator; The generator and the discriminator are trained adversarially so that the new virtual data generated by the generator gradually approaches the real fault data.
4. A motor fault diagnosis method according to claim 1, characterized in that: The cosine similarity is calculated as follows: Where P rν and V rν They represent the physical vibration response and virtual vibration response in the time domain under the same operating conditions and in the normal state, respectively.
5. A motor fault diagnosis method according to claim 1, characterized in that: The cosine similarity between the embedding vectors in the node set is calculated using the following formula: Among them, the embedding vector The embedding vector of the i-th node is represented as and They represent two different time series.
6. A motor fault diagnosis method according to claim 1, characterized in that: The adjacency matrix is represented as follows: Among them, N represents the number of nodes, and the element e in the adjacency matrix ij Represents the cosine similarity between node i and node j.
7. A motor fault diagnosis method according to claim 1, characterized in that: Based on the adjacency matrix, a fault diagnosis result is obtained, including: The adjacency matrix is input into the GCN layer in the spatiotemporal graph convolutional neural network, and the fault diagnosis result is obtained through the Softmax layer in the spatiotemporal graph convolutional neural network.
8. A motor fault diagnosis system, characterized in that: include: A modeling unit, used to establish a dynamic model of the motor rotating mechanical device; A first virtual fault data generating unit, configured to perform fault simulation on the dynamic model in a virtual space to obtain first virtual fault data corresponding to the fault type of the motor rotating mechanical device; A real fault data collection unit, used for collecting real fault data of the motor rotating mechanical device corresponding to the first virtual fault data; A second virtual fault data generating unit, configured to reduce the difference between the first virtual fault data and the real fault data based on a generative adversarial network, so as to obtain second virtual fault data with improved quality; a verification unit, configured to calculate, based on cosine similarity, a similarity between a virtual vibration response corresponding to the second virtual fault data and a physical vibration response corresponding to the real fault data, so as to verify the validity of the second virtual fault data; an embedding vector acquisition unit, configured to model the second virtual fault data into a graph structure based on a spatiotemporal graph convolutional neural network, wherein each variable in the graph structure is represented as a node, and each node in a node set is randomly initialized to an embedding vector composed of trainable parameters; An adjacency matrix construction unit, used to calculate the cosine similarity between the embedding vectors in the node set, and construct an adjacency matrix based on the cosine similarity between the embedding vectors; The fault diagnosis result acquisition module is used to obtain the fault diagnosis result based on the adjacency matrix.