Fault monitoring method and system, electronic equipment and storage medium

Abstract

The invention provides a fault monitoring method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining a target operation signal of a to-be-detected device; inputting the target operation signal into a pre-deployed fault monitoring model for fault monitoring to obtain a fault monitoring result; wherein the fault monitoring result comprises probabilities that the to-be-monitored device is in a normal state and different fault types, the fault monitoring model is obtained by performing antagonism training on fused fault features generated by a generator and a discriminator, and the fused fault features are extracted from a variational mode decomposition result of a sample data set; the data sample set comprises a label operation signal set and a label-free operation signal set, and labels comprise normal state labels and fault type labels. Based on variational mode decomposition and adversarial training, an adversarial semi-supervised framework based on variational mode decomposition is formed, cooperative information of labeled and unlabeled data is effectively utilized to realize fault monitoring, and the reliability of fault monitoring is ensured.

Description

Technical Field

[0001] The present application relates to the technical field of industrial automation equipment fault diagnosis, and in particular to a fault monitoring method, system, electronic device and storage medium. Background Art

[0002] Industrial automation equipment plays a vital role in intelligent manufacturing. It represents a sustainable development solution for global manufacturing automation and efficient production, and has become a major development direction for future industrial production. However, in actual operation, various components within automation equipment are subject to factors such as long-term loads, environmental changes, and mechanical wear, leading to various failures that can affect operational stability and production efficiency. Therefore, fault monitoring of components within industrial automation equipment has become a critical issue in the field of industrial equipment technology, crucial for ensuring stable operation and production efficiency.

[0003] In related technologies, fault monitoring for components in industrial automation equipment primarily relies on rule-based or model-based techniques. While these methods can achieve fault monitoring in certain simulations, their reliance on predefined rules or known fault modes often prevents them from effectively identifying and processing complex, unknown, or novel fault modes. Some methods utilize artificial intelligence and machine learning to achieve fault monitoring, but deep learning relies on large amounts of labeled data. In industrial scenarios, data labeling is expensive and difficult to obtain, resulting in a lack of labeled data. This lack of labeled data reduces the accuracy and robustness of deep learning, impacting the reliability of fault monitoring. Summary of the Invention

[0004] In view of the above shortcomings, the present application discloses a fault monitoring method, system, electronic device and storage medium to solve the technical problem of low reliability of fault monitoring.

[0005] In a first aspect, the present application provides a fault monitoring method, the method comprising: obtaining a target operating signal of a device under test; inputting the target operating signal into a pre-deployed fault monitoring model for fault monitoring to obtain a fault monitoring result; wherein the fault monitoring result includes the probability of the device under test being in a normal state and different fault types, the fault monitoring model is obtained by generating a fused fault feature through a generator and performing adversarial training with a discriminator, the fused fault feature is obtained by extracting it from a variational mode decomposition result of a sample data set, the data sample set includes a labeled operating signal set and an unlabeled operating signal set, and the labels include a normal state label and a fault type label.

[0006] In one embodiment of the present application, the method for obtaining the fused fault feature includes: performing energy integration on the first eigenmodal component set to obtain a first total energy, and performing energy integration on the second eigenmodal component set to obtain a second total energy, the variational modal decomposition result includes the first eigenmodal component set corresponding to the labeled operation signal set, the first center frequency set corresponding to the first eigenmodal component set, the second eigenmodal component set corresponding to the unlabeled operation signal set, and the second center frequency set corresponding to the second eigenmodal component set; calculating the energy entropy of each eigenmodal component in the first eigenmodal component set to obtain a first energy entropy set, and calculating the energy entropy of each eigenmodal component in the second eigenmodal component set to obtain a second energy entropy set; performing feature fusion on the first total energy and the second total energy according to the first center frequency set, the second center frequency set, the first energy entropy set, and the second energy entropy set to obtain the fused fault feature.

[0007] In one embodiment of the present application, the feature fusion of the first total energy and the second total energy is performed according to the first center frequency set, the second center frequency set, the first energy entropy set and the second energy entropy set to obtain the fused fault feature, including: fusing the first center frequency set and the second center frequency set to obtain a center frequency fusion matrix, and fusing the first energy entropy set and the second energy entropy set to obtain an energy distribution fusion matrix; performing a convolution operation on the first total energy according to the center frequency fusion matrix and the first convolution weight matrix to obtain a first aggregate feature, performing a convolution operation on the second total energy according to the energy distribution fusion matrix and the second convolution weight matrix to obtain a second aggregate feature, the first convolution weight matrix and the second convolution weight matrix being learning parameters; fusing the first aggregate feature and the second aggregate feature to obtain the fused fault feature.

[0008] In one embodiment of the present application, the fusing of the first aggregate feature and the second aggregate feature to obtain the fused fault feature includes: fusing the first center frequency set with the first energy entropy set to obtain a first energy-frequency fusion matrix, fusing the second center frequency set with the second energy entropy set to obtain a second energy-frequency fusion matrix; performing a convolution operation on the first total energy according to the first energy-frequency fusion matrix and the third convolution weight matrix to obtain a third aggregate feature, performing a convolution operation on the second total energy according to the second energy-frequency fusion matrix and the fourth convolution weight matrix to obtain a fourth aggregate feature, the third convolution weight matrix and the fourth convolution weight matrix being learning parameters; calculating the similarity between the third aggregate feature and the fourth aggregate feature to obtain a similarity constraint matrix; and fusing the first aggregate feature and the second aggregate feature according to the similarity constraint matrix to obtain the fused fault feature.

[0009] In one embodiment of the present application, the adversarial training method includes: controlling the fusion of the third aggregate feature extracted from the labeled operation signal set according to the fused fault feature and a preset first control fusion ratio to obtain the first target fault feature corresponding to the labeled operation signal set, and controlling the fusion of the fourth aggregate feature extracted from the unlabeled operation signal set according to the fused fault feature and the first control fusion ratio to obtain the second target fault feature corresponding to the unlabeled operation signal set; inputting the first target fault feature and the second target fault feature into the discriminator for true or false discrimination, and performing adversarial training on the generator and the discriminator based on the discrimination results.

[0010] In one embodiment of the present application, the adversarial training of the generator and the discriminator based on the discrimination result includes: setting the training goal of the generator to maximize the misjudgment rate of the discriminator for the second target fault feature, and setting the training goal of the discriminator to minimize the discrimination error between the first target fault feature and the second target fault feature; based on the discrimination result, the training goal of the generator and the training goal of the discriminator, the adversarial training of the generator and the discriminator is performed until a preset stopping condition is reached, thereby completing the adversarial training of the generator and the discriminator.

[0011] In one embodiment of the present application, the training method of the fault monitoring model also includes: controlling the fusion of the third aggregated features extracted from the labeled operating signal set according to the fused fault features and the preset second fusion control ratio to obtain a fault classification feature; inputting the fault classification feature into the classifier to perform the learning of the fault type identification task to complete the training of the fault monitoring model.

[0012] In a second aspect, the present application provides a fault monitoring system, which includes: a signal acquisition module for acquiring a target operating signal of a device under test; a fault monitoring module for inputting the target operating signal into a pre-deployed fault monitoring model for fault monitoring to obtain a fault monitoring result; wherein the fault monitoring result includes the probability of the device under test being in a normal state and different fault types, and the fault monitoring model is obtained by generating a fused fault feature through a generator and performing adversarial training with a discriminator, and the fused fault feature is obtained by extracting it from a variational modal decomposition result of a sample data set, and the data sample set includes a labeled operating signal set and an unlabeled operating signal set, and the label includes a normal state label and a fault type label.

[0013] In a third aspect, the present application provides an electronic device comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device implements the fault monitoring method as described in the first aspect.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor of a computer, the computer is caused to execute the fault monitoring method as described in the first aspect.

[0015] As described above, the embodiments of the present application provide a fault monitoring method, system, electronic device, and storage medium, which have the following beneficial effects:

[0016] First, the target operating signal of the device under test is obtained, and then the target operating signal is input into a pre-deployed fault monitoring model for fault monitoring to obtain a fault monitoring result. The fault monitoring result includes the probability of the device under test being in a normal state and different fault types. The fault monitoring model is obtained by generating a fused fault feature through a generator and performing adversarial training with a discriminator. The fused fault feature is extracted from the variational modal decomposition result of the sample data set. The data sample set includes a labeled operating signal set and an unlabeled operating signal set. The labels include normal state labels and fault type labels. The fused fault feature is extracted from the labeled operating signal set and the unlabeled operating signal set through variational modal decomposition, and the generator and discriminator are adversarially trained based on the fused fault feature, forming an adversarial semi-supervised framework based on variational modal decomposition, which effectively utilizes the collaborative information of labeled and unlabeled data, solves the problem of poor deep learning effect caused by insufficient fault feature discrimination and lack of labeled data, improves the accuracy and robustness of the fault monitoring model, and thus ensures the reliability of fault monitoring.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are incorporated into and constitute a part of the specification, illustrating embodiments consistent with the present application and, together with the specification, serving to explain the principles of the present application. It is obvious that the drawings described below are merely some embodiments of the present application, and a person of ordinary skill in the art can derive other drawings based on these drawings without inventive effort. In the drawings:

[0019] Figure 1 is a schematic diagram of an implementation environment of a fault monitoring system shown in an exemplary embodiment of the present application;

[0020] Figure 2 is a flow chart of a fault monitoring method shown in an exemplary embodiment of the present application;

[0021] Figure 3 This is a flow chart of an energy distribution-center frequency dual fusion mechanism shown in an exemplary embodiment of the present application;

[0022] Figure 4 This is a flowchart of a similarity constraint for fusion of fault features shown in an exemplary embodiment of the present application;

[0023] Figure 5 is a structural diagram of an adversarial semi-supervised framework based on variational modal decomposition, shown in an exemplary embodiment of the present application;

[0024] Figure 6 is a block diagram of a fault monitoring system shown in an exemplary embodiment of the present application;

[0025] Figure 7 This is a structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will describe the embodiments of the present application with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand the other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for the purpose of illustrating the present application and are not intended to limit the scope of protection of the present application.

[0027] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application. Therefore, the illustrations only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the shape, quantity and proportion of each component may be changed at will, and the component layout may also be more complicated.

[0028] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application. However, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present application difficult to understand.

[0029] Industrial automation equipment plays a vital role in intelligent manufacturing. It is a sustainable development solution for global manufacturing automation and efficient production, and has become a major development direction for future industrial production. For example, with the continuous advancement of industrial robotics technology, the global penetration of industrial robots has increased dramatically. However, in actual operation, various components within automation equipment are subject to various factors, such as long-term loads, environmental changes, and mechanical wear, which can lead to various failures, thus affecting the equipment's operational stability and production efficiency. For example, the reducer in an industrial robot, as a key component in the drive system, plays an extremely important role. It regulates the motor's torque and speed to ensure smooth operation and high efficiency. Due to the high speed and high torque of the industrial robot's electric drive system, the reducer bearings operate in a relatively harsh environment, making them prone to failure. Therefore, fault monitoring of components in industrial automation equipment to ensure rapid repair is key to ensuring stable equipment operation and production efficiency.

[0030] Typically, fault monitoring of components in industrial automation equipment relies on rule-based or model-based techniques. While these methods can achieve fault monitoring in certain simulations, their reliance on predefined rules or known fault modes often makes them ineffective in identifying and processing complex, unknown, or novel fault modes. Some methods utilize artificial intelligence and machine learning to achieve fault monitoring, but the inventors have discovered that deep learning relies on large amounts of labeled data. In industrial scenarios, data labeling is expensive and difficult to obtain, resulting in a lack of labeled data. This lack of labeled data results in poor accuracy and robustness in deep learning, which in turn impacts the reliability of fault monitoring.

[0031] Therefore, see Figure 1 , Figure 1 FIG. 1 is a schematic diagram of an implementation environment of a fault monitoring system according to an exemplary embodiment of the present application. Figure 1 As shown, the implementation environment may include a fault monitoring system 110 and a computer device 120. The fault monitoring system 110 may be set in the computer device 120 and used to perform fault monitoring of components in industrial automation equipment. The computer device 120 may be at least one of a desktop graphics processing unit (GPU) computer, a GPU computing cluster, a neural network computer, etc. The fault monitoring system 110 extracts fused fault features from a labeled operating signal set and an unlabeled operating signal set through variational mode decomposition, and conducts adversarial training on a generator and a discriminator based on the fused fault features, thereby forming an adversarial semi-supervised framework based on variational mode decomposition. This framework effectively utilizes the collaborative information of labeled and unlabeled data, solves the problem of poor deep learning results caused by insufficient fault feature discrimination and lack of labeled data, and improves the accuracy and robustness of the fault monitoring model, thereby ensuring the reliability of fault monitoring.

[0032] See Figure 2 , Figure 2 This is a flowchart of a fault monitoring method shown in an exemplary embodiment of the present application. This method can be applied to Figure 1 The implementation environment shown is specifically implemented by the fault monitoring system 110 in the implementation environment. It should be understood that the method can also be applied to other exemplary implementation environments and specifically implemented by devices in other implementation environments. This embodiment does not limit the implementation environment to which the method is applicable.

[0033] like Figure 2 As shown, in an exemplary embodiment, the fault monitoring method includes at least steps S210 to S220, which are described in detail as follows:

[0034] Step S210, obtaining a target operating signal of the device under test;

[0035] Step S220: input the target operation signal into a pre-deployed fault monitoring model to perform fault monitoring and obtain a fault monitoring result;

[0036] Among them, the fault monitoring results include the probability of the device under test being in a normal state and different fault types; the fault monitoring model is obtained by generating fused fault features through a generator and conducting adversarial training with a discriminator. The fused fault features are extracted from the variational modal decomposition results of the sample data set. The data sample set includes a labeled operating signal set and an unlabeled operating signal set. The labels include normal state labels and fault type labels.

[0037] In this embodiment, the device under test may be a reducer, motor, sensor, controller, actuator, etc. in industrial automation equipment; the target operating signal, the operating signals in the tagged operating signal set, and the operating signals in the untagged operating signal set may all be vibration signals, current signals, voltage signals, temperature signals, etc. The operating signals in the tagged operating signals are operating signals of multiple devices under different operating conditions in various application scenarios, each operating signal having a normal state label and a fault type label. The untagged operating signal set is operating signals of multiple devices under different operating conditions in various application scenarios, each operating signal having no normal state label or fault type label. The multiple devices and the device under test are of the same type, for example, all being reducers, i.e., the multiple devices are multiple reducers, the device under test is a reducer under test, and the multiple devices do not include the device under test.

[0038] In addition, the set of labeled running signals is recorded as is the signal for the i-th labeled run, where is the label corresponding to the i-th labeled operation signal, which is a probability distribution vector representing the probability of being in a normal state and different fault types. The fault type is based on the specific settings of different devices under test. For example, the fault types of reducers include fatigue peeling, wear, fracture, and bonding. The unlabeled operation signal set is recorded as For the i-th unlabeled running signal, in general, n< <m。

[0039] In this embodiment, fused fault features are extracted from the labeled operating signal set and the unlabeled operating signal set through variational mode decomposition, and the generator and discriminator are adversarially trained based on the fused fault features, forming an adversarial semi-supervised framework based on variational mode decomposition. This framework effectively utilizes the collaborative information of labeled and unlabeled data, solves the problems of poor deep learning effect caused by insufficient fault feature discrimination and lack of labeled data, improves the accuracy and robustness of the fault monitoring model, and thus ensures the reliability of fault monitoring.

[0040] Exemplarily, the variational modal decomposition result includes a first set of inherent modal components corresponding to the labeled operating signal set, a first set of center frequencies corresponding to the first set of inherent modal components, a second set of inherent modal components corresponding to the unlabeled operating signal set, and a second set of center frequencies corresponding to the second set of inherent modal components.

[0041] Among them, the first intrinsic modal component set is the intrinsic modal component set obtained by performing variational modal decomposition on the labeled running signal set, which is recorded as The second intrinsic mode component set is the intrinsic mode component set obtained by performing variational mode decomposition on the unlabeled running signal set, denoted as The first center frequency set is composed of the center frequency corresponding to each natural mode component in the first natural mode component set, which is recorded as The second center frequency set is composed of the center frequency corresponding to each natural mode component in the second natural mode component set, which is recorded as

[0042] In a possible embodiment, a method for generating variational modal decomposition results includes: performing variational modal decomposition on a labeled operating signal set, determining multiple first natural modal components and the center frequency of each first natural modal component, and generating a first natural modal component set and a first center frequency set; performing variational modal decomposition on an unlabeled operating signal set, determining multiple second natural modal components and the center frequency of each second natural modal component, and generating a second natural modal component set and a second center frequency set.

[0043] As a possible embodiment, the variational model of VMD (Variational modal decomposition) is:

[0044]

[0045] Among them: U k represents the intrinsic mode set; W k represents the center frequency set; K represents the number of preset natural mode components or center frequencies; t represents time; δ(t) represents the Dirac impulse function; j represents the imaginary unit; * represents convolution; u k (t) represents the decomposed sub-signal, that is, the kth natural mode component; w represents the frequency; st represents the constraint condition; f(t) represents the original signal, that is, the set of running signals to be decomposed.

[0046] Then the Lagrangian function is introduced to solve the optimal solution of the constrained variational problem. The Lagrangian function is:

[0047]

[0048] Among them, U k represents the intrinsic mode set; W k represents the center frequency set; λ represents the Lagrangian operator; α represents the secondary penalty factor; t represents time; j represents the imaginary unit; u k (t) represents the decomposed sub-signal, i.e., the kth intrinsic modal component; w represents the frequency; f(t) represents the original signal, i.e., the set of running signals to be decomposed; i represents the i-th data; u i (t) represents the i-th natural mode component after decomposition. In addition, i and k have the same physical meaning, both representing the i-th or k-th natural mode component. Different symbols are used to distinguish different parts in the formula.

[0049] Then, the saddle point of the Lagrangian function is obtained by the alternating direction multiplier algorithm, that is, the optimal solution of the constrained variational model, the inherent modal component set U k and the center frequency set W k In this process, the frequency domain update formula of the modal function is:

[0050]

[0051] Where w represents frequency; represents the updated value of the kth natural mode at frequency w in the n+1th iteration; represents the Fourier transform of the original signal f(t); represents the sum of the current estimated values ​​of other natural modes excluding the current mode k; is the frequency domain representation of the Lagrangian operator, used to enforce constraints w k represents the center frequency corresponding to the kth mode at the nth iteration; α represents the secondary penalty factor.

[0052] The update formula of the center frequency is:

[0053]

[0054] Where w represents frequency; represents the updated value of the kth mode at frequency w at the n+1th iteration; represents the intrinsic mode u k The power spectral density of (t).

[0055] For all w ≥ 0, update λ k :

[0056]

[0057] Where w represents frequency; represents the Fourier transform of the updated Lagrangian operator; represents the Fourier transform of the Lagrangian operator of the nth iteration; τ represents the penalty factor; represents the Fourier transform of the original signal f(t); represents the sum of all mode functions in the frequency domain.

[0058] When the relative error e is less than the preset convergence accuracy ε, the VMD process stops and K intrinsic modal components are obtained.

[0059]

[0060] Where, e represents relative error; w represents frequency; is the frequency domain representation of the kth mode function at the n+1th iteration; is the frequency domain representation of the kth mode function at the nth iteration; ε represents the preset convergence accuracy.

[0061] Based on the above variational mode decomposition process, the labeled running signal set and the unlabeled running signal set are solved to obtain the first intrinsic mode component set The second natural mode component set The first center frequency set With the second center frequency set

[0062] In one embodiment, a method for obtaining a fused fault feature includes: performing energy integration on a first set of inherent modal components to obtain a first total energy, and performing energy integration on a second set of inherent modal components to obtain a second total energy; calculating the energy entropy of each inherent modal component in the first set of inherent modal components to obtain a first energy entropy set, and calculating the energy entropy of each inherent modal component in the second set of inherent modal components to obtain a second energy entropy set; performing feature fusion on the first total energy and the second total energy according to the first center frequency set, the second center frequency set, the first energy entropy set, and the second energy entropy set to obtain a fused fault feature.

[0063] Among them, energy integral provides a global energy benchmark, energy entropy reveals the local energy distribution law, and center frequency supplements frequency domain information. The fusion of the three can form a fused fault feature that comprehensively characterizes the fault characteristics.

[0064] In this embodiment, the energy distribution complexity of each modal component is quantified by energy entropy, which overcomes the limitation of traditional methods that only focus on total energy. Combined with the center frequency characteristics, complementary enhancement of frequency domain features is achieved. The multi-feature fusion strategy effectively improves the characterization ability of features. Through the coordinated fusion of multi-dimensional signal features, the problem of insufficient single feature extraction is solved, providing higher quality feature input for subsequent model training.

[0065] Exemplarily, energy integration is performed on the first natural modal component set to obtain a first total energy, which is calculated as follows:

[0066]

[0067] in, represents the first total energy; t represents time; represents the first set of intrinsic modal components.

[0068] Perform energy integration on the second natural mode component set to obtain the second total energy, which is calculated as follows:

[0069]

[0070] in, represents the second total energy; t represents time; represents the second set of intrinsic modal components.

[0071] In this way, by time-integrating the square of the signal amplitude, the energy accumulation of the quantified signal in the entire time domain directly reflects the strength of the operating signal. When a fault occurs, the amplitude of the signal change usually increases, resulting in E k The value increases, and the E k The values ​​are different.

[0072] In obtaining and Finally, the input is sent to the generator for fusion feature extraction. First, the energy entropy under each total energy is obtained. This is because when a fault occurs, the frequency distribution of the operating signal will change and the energy value of the modal component will change.

[0073] Exemplarily, the energy entropy of each intrinsic modal component in the first eigenmodal component set is calculated using the following formula:

[0074]

[0075] Where, j = 1, 2, 3, …, k; represents the energy entropy of the jth intrinsic modal component in the first intrinsic modal component set; It represents the percentage of the energy of the jth natural modal component in the first natural modal component set to the total energy of the entire first natural modal component set.

[0076] Calculate the energy entropy of each natural mode component in the second natural mode component set. The calculation formula is:

[0077]

[0078] Where, j = 1, 2, 3, …, k; represents the energy entropy of the jth intrinsic modal component in the second intrinsic modal component set; It represents the percentage of the energy of the jth natural modal component in the second natural modal component set to the total energy of the entire second natural modal component set.

[0079] in addition, The calculation formula is:

[0080]

[0081] in, represents the percentage of the energy of the jth natural modal component in the first natural modal component set to the total energy of the entire first natural modal component set; represents the energy of the jth natural mode component in the first natural mode component set; represents the total energy of the entire first natural mode component set.

[0082] The calculation formula is:

[0083]

[0084] in, represents the percentage of the energy of the jth natural modal component in the second natural modal component set to the total energy of the entire second natural modal component set; represents the energy of the jth natural mode component in the second natural mode component set; represents the total energy of the entire second natural mode component set.

[0085] Based on the above, the energy entropy of each intrinsic modal component in the first intrinsic modal component set can be calculated to form a first energy entropy set, and the energy entropy of each intrinsic modal component in the second intrinsic modal component set can be calculated to form a second energy entropy set. In the above process, for the labeled operating signal set, the corresponding first center frequency set and first energy entropy set are obtained, and for the unlabeled operating signal set, the corresponding second center frequency set and second energy entropy set are obtained. Thus, based on the first center frequency set, the first energy entropy set, the second center frequency set, and the second energy entropy set, the feature fusion of the first total energy and the second total energy is achieved to obtain a fused fault feature.

[0086] In one embodiment, according to the first center frequency set, the second center frequency set, the first energy entropy set and the second energy entropy set, feature fusion is performed on the first total energy and the second total energy to obtain a fused fault feature, including: fusing the first center frequency set and the second center frequency set to obtain a center frequency fusion matrix, and fusing the first energy entropy set and the second energy entropy set to obtain an energy distribution fusion matrix; according to the center frequency fusion matrix and the first convolution weight matrix, a convolution operation is performed on the first total energy to obtain a first aggregate feature; according to the energy distribution fusion matrix and the second convolution weight matrix, a convolution operation is performed on the second total energy to obtain a second aggregate feature, and the first convolution weight matrix and the second convolution weight matrix are learning parameters; the first aggregate feature and the second aggregate feature are fused to obtain a fused fault feature.

[0087] Among them, the center frequency fusion matrix unifies the frequency domain representations of different signal sources, the energy distribution fusion matrix retains the statistical characteristics of energy entropy, and the introduction of the convolution weight matrix enables the model to adaptively adjust the weight distribution of different features, thereby improving the discriminative ability of the fused features.

[0088] In this embodiment, an energy distribution-center frequency cross-fusion mechanism is designed. The energy distribution provides the overall energy situation of the frequency components, while the center frequency helps identify the most representative frequency points. By combining the energy distribution and center frequency information, the model can analyze the time-frequency characteristics of the signal from multiple dimensions, thereby improving the model's ability to identify different types of faults.

[0089] In this way, by establishing a dual fusion mechanism of frequency and energy and combining it with learnable convolution operations, deep integration of cross-modal features is achieved. Through the synergistic effect of matrix fusion and convolution operations, the expressive ability of fault features is significantly improved.

[0090] In a possible embodiment, a convolution operation is performed on the first total energy according to the center frequency fusion matrix and the first convolution weight matrix to obtain a first aggregate feature, and the calculation process is as follows:

[0091]

[0092] in, represents the first aggregate feature; Represents the first center frequency set, which needs to be normalized; Represents the second center frequency set, which needs to be normalized; represents the transposed matrix of the second center frequency set after normalization; Indicates the symmetric normalization operation on the center frequency fusion matrix; represents the first total energy; Represents the first convolution weight matrix; ReLU represents the activation function;

[0093] According to the energy distribution fusion matrix and the second convolution weight matrix, the second total energy is convolved to obtain the second aggregate feature. The calculation process is:

[0094]

[0095] in, represents the second aggregate feature; Represents the first energy entropy set, which needs to be normalized; Represents the second energy entropy set, which needs to be normalized; represents the transposed matrix of the second energy entropy set after normalization; Indicates the symmetric normalization operation on the energy distribution fusion matrix; represents the second total energy; Represents the second convolution weight matrix; ReLU represents the activation function.

[0096] As a possible embodiment, Represents the aggregated features of the total energy output of the K intrinsic modes in the first intrinsic mode component concentration based on the center frequency fusion matrix after graph convolution. The first center frequency set after normalization (of size k×1) and The second center frequency set after normalization (size is k×1) and the transposed matrix is ​​multiplied to obtain a center frequency fusion matrix (size is K×K), and then the center frequency fusion matrix is ​​used as the first total energy corresponding to the labeled running signal set The connection matrix between the two models is constructed, and the features of the total energy of K intrinsic modes are fused through a graph convolutional network. The graph structure generated by the center frequency crossover can effectively integrate the features of different modal components. The prior information based on frequency domain knowledge enables the model to focus on modalities that are closely related in frequency during the feature extraction process, thereby achieving feature fusion of labeled and unlabeled data in the same category, improving the performance of subsequent classification tasks.

[0097] in addition, Represents the aggregated features of the total energy output of K intrinsic modes in the second intrinsic mode component set based on the energy distribution fusion matrix after graph convolution. The first total energy entropy set after normalization (of size k×1) and The second total energy entropy set after normalization is The transposed matrix (size is k×1) is multiplied to obtain an energy distribution fusion matrix (size is K×K), and then the energy distribution fusion matrix is ​​used as the second energy entropy of the K intrinsic modal components of the unlabeled dataset The connection relationship matrix between them is used to fuse the energy entropy features of K inherent modal components through a graph convolutional network. The graph structure generated by the energy distribution fusion matrix is ​​used to explicitly store the characteristic relationships such as energy and frequency between the modes in a matrix. With the help of the powerful representation ability of the graph model, these inter-modal associations and the single modality’s own characteristics are incorporated into the learning process, thereby effectively improving the performance of feature extraction and downstream tasks.

[0098] Figure 3 This is a flow chart of an energy distribution-center frequency dual fusion mechanism shown in an exemplary embodiment of the present application. Figure 3 As shown, first, based on the first total energy, the first energy entropy is calculated to determine the first energy entropy set, and based on the second total energy, the second energy entropy is calculated to determine the second energy entropy set, and then based on the first energy entropy set, the second energy entropy set and the first center frequency set and the second center frequency set obtained by variational modal decomposition, feature fusion is performed to obtain a fused fault feature.

[0099] In one embodiment, the first aggregate feature and the second aggregate feature are fused to obtain a fused fault feature, including: fusing the first center frequency set with the first energy entropy set to obtain a first energy-frequency fusion matrix, fusing the second center frequency set with the second energy entropy set to obtain a second energy-frequency fusion matrix; performing a convolution operation on the first total energy according to the first energy-frequency fusion matrix and the third convolution weight matrix to obtain a third aggregate feature, performing a convolution operation on the second total energy according to the second energy-frequency fusion matrix and the fourth convolution weight matrix to obtain a fourth aggregate feature, the third convolution weight matrix and the fourth convolution weight matrix being learning parameters; calculating the similarity between the third aggregate feature and the fourth aggregate feature to obtain a similarity constraint matrix; and fusing the first aggregate feature and the second aggregate feature according to the similarity constraint matrix to obtain a fused fault feature.

[0100] The similarity calculation may adopt cosine similarity or Euclidean distance measurement.

[0101] In this embodiment, considering that cross-fusion is performed separately from the perspectives of energy distribution and center frequency, there will be a problem of excessive cross-fusion generating confusing features. Therefore, a similarity constraint mechanism is designed. The feature space of the labeled running signal set and the unlabeled running signal set is optimized based on their respective energy distribution and center frequency fusion, and then a similarity constraint matrix is ​​generated based on their respective fusion features to prevent excessive cross-fusion from generating confusing features, thereby improving the fault monitoring capability.

[0102] In a possible embodiment, a convolution operation is performed on the first total energy according to the first energy-frequency fusion matrix and the third convolution weight matrix to obtain a third aggregate feature, which is calculated as follows:

[0103]

[0104] in, represents the third aggregate feature; Represents the first center frequency set, which needs to be normalized; Represents the first energy entropy set, which needs to be normalized; represents the transposed matrix of the first energy entropy set after normalization; Indicates that a symmetric normalization operation is performed on the first energy-frequency fusion matrix; represents the first total energy; Represents the third convolution weight matrix; ReLU represents the activation function;

[0105] According to the second energy-frequency fusion matrix and the fourth convolution weight matrix, the second total energy is convolved to obtain the fourth aggregate feature, which is calculated as follows:

[0106]

[0107] in, represents the fourth aggregate feature; Represents the second center frequency set, which needs to be normalized; Represents the second energy entropy set, which needs to be normalized; represents the transposed matrix of the second energy entropy set after normalization; Indicates that a symmetric normalization operation is performed on the second energy-frequency fusion matrix; represents the second total energy; Represents the fourth convolution weight matrix; ReLU represents the activation function.

[0108] As a possible embodiment, The energy-frequency fusion matrix representing the set of labeled running signals aggregates the total energy of K intrinsic modes through graph convolution The output aggregate feature is the total energy of the K intrinsic modes of the signal set with labels Based on its own energy distribution and center frequency To perform energy distribution-center frequency fusion, first construct the first energy frequency fusion matrix, that is, the center frequency and energy distribution The transposed matrix of is multiplied (size is K×K), and then the total energy of the K natural modes of the labeled running signal set is calculated based on this matrix. The connection matrix of is used to perform K feature fusion based on the energy entropy of modal components through the graph convolutional network.

[0109] in addition, The energy-frequency fusion matrix representing the unlabeled running signal set aggregates the total energy of K intrinsic modes through graph convolution The output aggregate feature is the total energy of the K intrinsic modes of the signal set running on the unlabeled Based on its own energy distribution and center frequency Perform energy distribution-center frequency fusion. Similarly, construct the energy frequency fusion matrix of the unlabeled running signal set, that is, the center frequency and energy distribution The transposed matrix of is multiplied (size is K×K), and then the total energy of the K natural modes of the unlabeled running signal set is calculated based on this matrix. The connection matrix of is used to perform K feature fusion based on the energy entropy of modal components through the graph convolutional network.

[0110] In a possible embodiment, calculating the similarity between the third aggregate feature and the fourth aggregate feature to obtain a similarity constraint matrix includes: calculating the Euclidean distance between the third aggregate feature and the fourth aggregate feature in K dimensions to form a Euclidean distance matrix; and normalizing the Euclidean distance matrix to obtain a similarity constraint matrix.

[0111] As a possible embodiment, the calculation formula of the similarity constraint matrix is:

[0112]

[0113] Where i=1,2,…,K represents the i-th dimension; j represents the j-th data value; L represents the length of the data value in each dimension; Represents the similarity matrix.

[0114] In this possible embodiment, a K×1 Euclidean distance matrix is ​​obtained, representing the similarity of the total energy of the K intrinsic modes of the labeled and unlabeled operating signal sets, after aggregation by their own energy distribution and center frequency, along the corresponding dimension. This Euclidean distance matrix is ​​then normalized and used as the similarity constraint matrix for the first and second aggregated features output by the dual fusion mechanism in the energy distribution and center frequency domains, denoted as R1.

[0115] In a possible embodiment, the first aggregated feature and the second aggregated feature are fused according to the similarity constraint matrix to obtain a fused fault feature, which is calculated as follows:

[0116]

[0117] Among them, H r Indicates the fusion fault characteristics; represents the first aggregated feature; R1 represents the similarity constraint matrix; Represents the second aggregate feature.

[0118] As a possible embodiment, adaptive fusion of the first aggregated features and the second aggregated features is achieved based on a similarity constraint matrix.

[0119] Figure 4 This is a flowchart of a fault feature fusion similarity constraint shown in an exemplary embodiment of the present application. Figure 4 As shown, first, the third aggregate feature is determined based on the first energy entropy set and the first center frequency set, and the fourth aggregate feature is determined based on the second energy entropy set and the second center frequency set. Then, a similarity constraint matrix is ​​formed by the third aggregate feature and the fourth aggregate feature to constrain the fusion of the first aggregate feature and the second aggregate feature, thereby obtaining a fused fault feature.

[0120] In one embodiment, an adversarial training method includes: performing controlled fusion on a third aggregate feature extracted from a labeled operation signal set according to a fused fault feature and a preset first controlled fusion ratio to obtain a first target fault feature corresponding to the labeled operation signal set; and performing controlled fusion on a fourth aggregate feature extracted from an unlabeled operation signal set according to the fused fault feature and the first controlled fusion ratio to obtain a second target fault feature corresponding to the unlabeled operation signal set; inputting the first target fault feature and the second target fault feature into a discriminator for true or false discrimination, and performing adversarial training on the generator and the discriminator based on the discrimination results.

[0121] The first control fusion ratio is adjustable, establishing a dynamic balance mechanism between the labeled and unlabeled running signal sets. Setting the first control fusion ratio enables the model to adaptively adjust the contribution of the labeled and unlabeled running signal sets based on the actual data distribution, avoiding the problem of mode collapse caused by imbalance in the ratio in traditional adversarial training. For example, the first control fusion ratio is set to 0.2.

[0122] In this embodiment, the design of the dual feature fusion path retains the discriminative feature boundaries of the labeled running signal set, while enhancing the coverage of the feature space through the unlabeled running signal set. Adversarial training is performed through quantization control to achieve an organic combination of supervised learning and unsupervised learning, solving the problem of performance degradation of traditional methods when data is insufficiently labeled.

[0123] Exemplarily, the first target fault feature is calculated as follows:

[0124]

[0125] in, represents the first target fault feature; β represents the first control fusion ratio; H r Indicates the fusion fault characteristics; Represents the third aggregate feature.

[0126] The second target fault feature is calculated as follows:

[0127]

[0128] in, represents; β represents the first control fusion ratio; H r Indicates the fusion fault characteristics; Represents the fourth aggregate feature.

[0129] In one embodiment, adversarial training is performed on the generator and the discriminator based on the discrimination result, including: setting the training goal of the generator to maximize the misjudgment rate of the discriminator for the second target fault feature, and setting the training goal of the discriminator to minimize the discrimination error between the first target fault feature and the second target fault feature; based on the discrimination result, the training goal of the generator and the training goal of the discriminator, the generator and the discriminator are adversarially trained until a preset stopping condition is reached, thereby completing the adversarial training of the generator and the discriminator.

[0130] Among them, maximizing the misjudgment rate can be achieved by constructing a generator loss function, such as using the cross-entropy loss function to calculate the probability of the discriminator misjudging the generated features, and updating the generator parameters through the gradient ascent algorithm; minimizing the discrimination error can be achieved through a binary classification loss function, such as using a focal loss function with weight adjustment to deal with the category imbalance problem; the stopping condition can be set as the training round reaches a threshold or the discriminator accuracy enters a stable range, etc.

[0131] In this embodiment, a bidirectionally optimized closed-loop feedback mechanism is established by quantifying the adversarial objectives of the generator and the discriminator. The generator deceives the discriminator by improving the authenticity of the forged features, while the discriminator forces the generator to improve its feature generation capability through precise discrimination, thereby improving the accuracy and robustness of the fault monitoring model.

[0132] For example, the objective function of adversarial training between the generator and the discriminator is:

[0133]

[0134] in, represents the first target fault feature corresponding to the labeled operating signal set; represents the second target fault feature corresponding to the unlabeled running signal set; while the logarithmic function does not change the monotonicity of the function, represents the probability of classifying true as true; D(G(x i )) represents the probability of classifying a false positive as a true positive; 1-D(G(x i )) can be used to describe the probability of classifying a false as false.

[0135] For the discriminator, the higher the ability to classify true as true and false as false, the better the discriminator effect is. Therefore, the parameter optimization process of the discriminator is a process of maximizing the objective function, that is, using When the discriminator parameters are fixed, the first term of the objective function becomes a constant. When the generator parameters are adjusted, only the second term changes. The second term represents the ability to classify fake as fake. For the generator, this ability should be as small as possible, so that the generator can deceive the discriminator. Therefore, the optimization of the generator is a process of minimizing the objective function, that is, using In this way, the discrimination error between the first target fault feature and the second target fault feature of the identifier is minimized and the misjudgment rate of the second target fault feature of the identifier is maximized.

[0136] In this way, refining the fusion operation and combining it with the generative adversarial network not only enhances the model's controllability over different features, but also achieves stronger discrimination ability and better generalization performance through the effective use of unlabeled data.

[0137] In one embodiment, the training method of the fault monitoring model also includes: controlling the fusion of the third aggregated features extracted from the labeled operating signal set according to the fused fault features and the preset second fusion control ratio to obtain the fault classification features; inputting the fault classification features into the classifier to perform the learning of the fault type identification task to complete the training of the fault monitoring model.

[0138] The second control fusion ratio is adjustable, for example, the second control fusion ratio is set to 0.2.

[0139] In this embodiment, the fused fault signature is controlled and fused with the first target fault signature corresponding to the labeled operating signal set. This is then input into a classifier to classify device faults, including normal states and multiple fault types. Introducing the classifier for model training helps the model learn more representative fault modes, thereby maintaining stable performance when faced with new samples or changing operating conditions.

[0140] For example, the fault classification feature is calculated as follows:

[0141]

[0142] Among them, H 分类 represents the fault classification feature; β2 represents the second control fusion ratio; H r Indicates the fusion fault characteristics; Represents the third aggregate feature; softmax represents the normalization operation.

[0143] See Figure 5 , Figure 5 This is a structural diagram of an adversarial semi-supervised framework based on variational mode decomposition, as shown in an exemplary embodiment of the present application. Figure 5As shown, in the adversarial semi-supervised framework based on variational mode decomposition, variational mode decomposition is first performed on the labeled operating signal set and the unlabeled operating signal set respectively, and the energy of the decomposed multiple inherent mode components is integrated to obtain the corresponding first total energy and second total energy. Then, the corresponding first energy entropy set is extracted from the first total energy by the generator, and the corresponding second energy entropy set is extracted from the second total energy. Based on the first energy entropy set, the second energy entropy set and the first center frequency set and the second center frequency set obtained by variational mode decomposition, feature fusion is performed to first obtain the first aggregate feature, the second aggregate feature, the third aggregate feature and the fourth aggregate feature, and then feature fusion is performed to obtain the fused fault feature. After that, the fused fault feature is used to perform adversarial training on the generator and the discriminator, and the fused fault feature is used to perform classification task learning on the classifier, thereby training to obtain a fault monitoring model.

[0144] In one possible embodiment, the fault monitoring result includes probability distribution vectors of the device under test being in a normal state and different fault types. After obtaining the fault monitoring result, the method further includes: determining the normal state or fault type corresponding to the maximum probability item as the final result, for example, the normal state or a specific fault type.

[0145] In a possible embodiment, after obtaining the fault monitoring result, the method further includes: formulating a fault avoidance strategy according to the probability distribution vector in the fault monitoring result.

[0146] For example, if the probability value of a certain fault in the probability distribution vector reaches a threshold, a corresponding fault avoidance strategy is formulated to avoid the fault in advance.

[0147] The above-mentioned fault monitoring method first obtains the target operating signal of the device under test, and then inputs the target operating signal into a pre-deployed fault monitoring model for fault monitoring to obtain a fault monitoring result. The fault monitoring result includes the probability of the device under test being in a normal state and different fault types. The fault monitoring model is obtained by generating a fused fault feature through a generator and performing adversarial training with a discriminator. The fused fault feature is obtained by extracting it from the variational modal decomposition result of a sample data set. The data sample set includes a labeled operating signal set and an unlabeled operating signal set. The labels include normal state labels and fault type labels. The fused fault feature is extracted from the labeled operating signal set and the unlabeled operating signal set through variational modal decomposition, and the generator and discriminator are adversarially trained based on the fused fault feature, forming an adversarial semi-supervised framework based on variational modal decomposition, which effectively utilizes the collaborative information of labeled and unlabeled data, solves the problem of poor deep learning effect caused by insufficient fault feature discrimination and lack of labeled data, improves the accuracy and robustness of the fault monitoring model, and thus ensures the reliability of fault monitoring.

[0148] See Figure 6 , Figure 6 This is a block diagram of a fault monitoring system shown in an exemplary embodiment of the present application. The system can be applied to Figure 1 The implementation environment shown is shown. It should be understood that the system can also be applied to other exemplary implementation environments, and this embodiment does not limit the implementation environment to which the system is applicable.

[0149] like Figure 6 As shown, in an exemplary embodiment, the fault monitoring system 600 includes at least a signal acquisition module 610 and a fault monitoring module 620, which are described in detail as follows:

[0150] The signal acquisition module 610 is used to obtain the target operating signal of the device under test;

[0151] The fault monitoring module 620 is used to input the target operation signal into the pre-deployed fault monitoring model to perform fault monitoring and obtain the fault monitoring result;

[0152] Among them, the fault monitoring results include the probability of the device under test being in a normal state and different fault types. The fault monitoring model is obtained by generating fused fault features through a generator and conducting adversarial training with the discriminator. The fused fault features are extracted from the variational modal decomposition results of the sample data set. The data sample set includes a labeled operating signal set and an unlabeled operating signal set. The labels include normal state labels and fault type labels.

[0153] It should be noted that the fault monitoring system provided in the above embodiment and the fault monitoring method provided in the above embodiment belong to the same concept, wherein the contents of the operations performed by each module have been described in detail in the method embodiment and will not be repeated here.

[0154] See Figure 7 , Figure 7 This is a structural diagram of an electronic device provided by an embodiment of the present application. Figure 7 The following is a schematic diagram showing the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application. Figure 7 The computer system 700 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0155] like Figure 7As shown, computer system 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes according to the program stored in read-only memory (ROM) 702 or the program loaded from storage portion 708 into random access memory (RAM) 703, such as executing the method in the above embodiment. Various programs and data required for system operation are also stored in RAM 703. CPU 701, ROM 702 and RAM 703 are connected to each other via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0156] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, and the like; an output section 707 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 708 including a hard disk and the like; and a communication section 709 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. Removable media 711, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 710 as needed, so that computer programs read therefrom can be installed into the storage section 708 as needed.

[0157] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from a removable medium 711. When the computer program is executed by the central processing unit (CPU) 701, the various functions defined in the system of the present application are executed.

[0158] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a computer processor, the computer executes the fault monitoring method described above. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist independently and not be incorporated into the electronic device.

[0159] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable computer program. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer program embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0160] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0161] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. In some cases, the names of these units do not constitute limitations on the units themselves.

[0162] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, any equivalent modifications or alterations accomplished by a person of ordinary skill in the art without departing from the spirit and technical concepts disclosed in this application shall be covered by the claims of this application.

Claims

1. A fault monitoring method, characterized in that: The method comprises: Obtain target operating signals of the device under test; Inputting the target operation signal into a pre-deployed fault monitoring model to perform fault monitoring and obtain a fault monitoring result; Among them, the fault monitoring results include the probability of the device under test being in a normal state and different fault types. The fault monitoring model is obtained by generating fused fault features through a generator and performing adversarial training with a discriminator. The fused fault features are obtained by extracting them from the variational modal decomposition results of the sample data set. The data sample set includes a labeled operating signal set and an unlabeled operating signal set, and the labels include normal state labels and fault type labels.

2. The fault monitoring method according to claim 1, characterized in that: The method for obtaining the fusion fault feature includes: Performing energy integration on the first eigenmodal component set to obtain a first total energy, and performing energy integration on the second eigenmodal component set to obtain a second total energy, wherein the variational modal decomposition result includes the first eigenmodal component set corresponding to the labeled operation signal set, a first center frequency set corresponding to the first eigenmodal component set, a second eigenmodal component set corresponding to the unlabeled operation signal set, and a second center frequency set corresponding to the second eigenmodal component set; Calculating the energy entropy of each inherent modal component in the first inherent modal component set to obtain a first energy entropy set, and calculating the energy entropy of each inherent modal component in the second inherent modal component set to obtain a second energy entropy set; According to the first center frequency set, the second center frequency set, the first energy entropy set, and the second energy entropy set, feature fusion is performed on the first total energy and the second total energy to obtain the fused fault feature.

3. The fault monitoring method according to claim 2, characterized in that: The performing feature fusion on the first total energy and the second total energy according to the first center frequency set, the second center frequency set, the first energy entropy set, and the second energy entropy set to obtain the fused fault feature includes: fusing the first center frequency set and the second center frequency set to obtain a center frequency fusion matrix, and fusing the first energy entropy set and the second energy entropy set to obtain an energy distribution fusion matrix; Performing a convolution operation on the first total energy according to the center frequency fusion matrix and a first convolution weight matrix to obtain a first aggregate feature, and performing a convolution operation on the second total energy according to the energy distribution fusion matrix and a second convolution weight matrix to obtain a second aggregate feature, where the first convolution weight matrix and the second convolution weight matrix are learning parameters; The first aggregated feature and the second aggregated feature are fused to obtain the fused fault feature.

4. The fault monitoring method according to claim 3, characterized in that: The fusing the first aggregated feature and the second aggregated feature to obtain the fused fault feature includes: fusing the first center frequency set with the first energy entropy set to obtain a first energy-frequency fusion matrix, and fusing the second center frequency set with the second energy entropy set to obtain a second energy-frequency fusion matrix; Performing a convolution operation on the first total energy according to the first energy-frequency fusion matrix and a third convolution weight matrix to obtain a third aggregate feature; performing a convolution operation on the second total energy according to the second energy-frequency fusion matrix and a fourth convolution weight matrix to obtain a fourth aggregate feature, where the third convolution weight matrix and the fourth convolution weight matrix are learning parameters; Calculating the similarity between the third aggregate feature and the fourth aggregate feature to obtain a similarity constraint matrix; The first aggregated features and the second aggregated features are fused according to the similarity constraint matrix to obtain the fused fault features.

5. The fault monitoring method according to claim 1, characterized in that: The adversarial training method includes: performing controlled fusion on a third aggregated feature extracted from the labeled operation signal set based on the fused fault feature and a preset first control fusion ratio to obtain a first target fault feature corresponding to the labeled operation signal set; and performing controlled fusion on a fourth aggregated feature extracted from the unlabeled operation signal set based on the fused fault feature and the first control fusion ratio to obtain a second target fault feature corresponding to the unlabeled operation signal set; The first target fault feature and the second target fault feature are input into the discriminator for true or false discrimination, and adversarial training is performed on the generator and the discriminator according to the discrimination results.

6. The fault monitoring method according to claim 5, characterized in that: The performing adversarial training on the generator and the discriminator according to the discrimination result includes: Setting the training objective of the generator to maximize the misjudgment rate of the discriminator for the second target fault feature, and setting the training objective of the discriminator to minimize the discrimination error between the first target fault feature and the second target fault feature; According to the discrimination result, the training target of the generator and the training target of the discriminator, the generator and the discriminator are subjected to adversarial training until a preset stopping condition is reached, thereby completing the adversarial training of the generator and the discriminator.

7. The fault monitoring method according to any one of claims 1 to 6, characterized in that: The training method of the fault monitoring model further includes: According to the fused fault feature and a preset second fusion control ratio, a third aggregate feature extracted from the labeled operation signal set is controlled and fused to obtain a fault classification feature; The fault classification features are input into a classifier to perform fault type identification task learning, thereby completing the training of the fault monitoring model.

8. A fault monitoring system, characterized in that: The system comprises: A signal acquisition module is used to obtain the target operating signal of the device under test; a fault monitoring module, configured to input the target operation signal into a pre-deployed fault monitoring model to perform fault monitoring and obtain a fault monitoring result; Among them, the fault monitoring results include the probability of the device under test being in a normal state and different fault types. The fault monitoring model is obtained by generating fused fault features through a generator and performing adversarial training with a discriminator. The fused fault features are obtained by extracting them from the variational modal decomposition results of the sample data set. The data sample set includes a labeled operating signal set and an unlabeled operating signal set, and the labels include normal state labels and fault type labels.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the fault monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the fault monitoring method according to any one of claims 1 to 7.

Images