A Migration Fault Diagnosis Method Based on Digital Twin and Generative Adversarial Network

Abstract

The present application relates to a migration fault diagnosis method based on digital twin and generative adversarial network, and relates to the technical field of industrial processes. The method includes the following steps: constructing a digital twin model of a rolling bearing based on the actual size of the rolling bearing of a target device; obtaining simulated fault data at a preset fault depth and a preset fault width based on the digital twin model; inputting the simulated fault data into an improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; judging the fault type of the rolling bearing of the target device based on a fault diagnosis model and the actual operation data of the rolling bearing, so that the fault diagnosis model trained on the virtual entity can be used for the fault diagnosis problem of the physical entity, and solves the problem of insufficient fault data in the physical entity. In this method, the parameter iteration optimization rate is adjusted in a timely manner according to the backdoor attack defense effect, so that the adjusted defense model has a better defense effect against the backdoor.

Description

Technical Field

[0001] The present application relates to the technical field of process industry, and particularly relates to a migration fault diagnosis method based on digital twin and generative adversarial network. Background Art

[0002] The shearer is a large and complex system integrating machinery, electricity and hydraulics. Its working environment is harsh. If a fault occurs, it will lead to the interruption of the entire coal mining work and cause huge economic losses. As an important component of the shearer, monitoring and diagnosing the health status of rolling bearings has important economic benefits.

[0003] The prerequisite for a data-driven fault diagnosis method to achieve optimal performance is to obtain a large amount of training data sets. However, during the operation of the shearer with high safety requirements, the fault occurrence frequency is low, and the vast majority of the monitored data are the normal operation data of the equipment, resulting in insufficient fault data. Aiming at the problems of insufficient fault data and poor model generalization ability of the intelligent diagnosis model, the fault diagnosis method based on data augmentation has attracted wide attention. However, the prerequisite for these data augmentation methods to be implemented is that there are a certain number of fault samples as the basic augmentation objects. If the number of samples to be augmented is too small, the diagnostic accuracy will be low.

[0004] Aiming at this problem, the data augmentation method based on fault mechanism has been developed. Although this method has certain advantages, there are significant differences in the characteristic distributions between the simulated fault data and the actual fault data, which limits the accuracy of fault diagnosis. Summary of the Invention

[0005] The present application provides a migration fault diagnosis method based on digital twin and generative adversarial network, constructs a digital twin model of the rolling bearing to generate virtual fault data, and eliminates the domain difference between the real fault data and the virtual fault data through the domain adaptation method of the auxiliary generative adversarial network, so that the fault diagnosis model trained on the virtual entity can be used for the fault diagnosis problem of the physical entity, and solves the problem of insufficient fault data in the physical entity.

[0006] To achieve the above object, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a migration fault diagnosis method based on digital twin and generative adversarial network, and the method includes the following steps:

[0008] Construct a digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target instrument;

[0009] Obtain simulated fault data at a preset fault depth and a preset fault width based on the digital twin model;

[0010] Input the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; the synthetic fault data is used to train the fault diagnosis model;

[0011] Based on the fault diagnosis model and the actual operation data of the rolling bearing, determine the fault type of the rolling bearing of the target device.

[0012] Further, constructing the digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target device includes the following steps:

[0013] Generate the geometric model of the rolling bearing based on the geometric dimensions, shapes of the components in the rolling bearing of the target device and the positional relationship between the components;

[0014] Based on the physical properties of the rolling bearing of the target device, obtain the physical model that maps the physical state of the rolling bearing itself;

[0015] Based on the action responses of the rolling bearing to the internal and external environments and system instructions, obtain the behavior model of the rolling bearing;

[0016] Based on the fault occurrence rules summarized from the historical operation data and the corresponding rolling bearing operation states, obtain the rule model of the rolling bearing.

[0017] Further, obtaining the simulated fault data at the preset fault depth and preset fault width based on the digital twin model includes the following steps:

[0018] Based on the geometric model, physical model, behavior model and rule model, obtain the digital twin model of the rolling bearing;

[0019] Set the preset fault depth and preset fault width for the digital twin model;

[0020] Based on the responses of the digital twin model at the preset fault depth and preset fault width, obtain the simulated fault data of the digital twin model.

[0021] Further, the improved generative adversarial network model includes a generator network and a discriminator network.

[0022] Further, the generator network consists of five convolutional layers, and the discriminator network consists of an SCAE model, three convolutional layers and two fully connected layers.

[0023] Further, the training process of the discriminator network includes:

[0024] Randomly sample Gaussian noise to obtain real samples and generated samples;

[0025] Train the discriminant network using the first cross - entropy cost function until the cost function error of the class labels of the generated samples and the real samples is minimized;

[0026] Train the generator network using the second cross - entropy cost function until the cost function error of the authenticity labels of the generated samples and the real samples is minimized, and complete the training process of the discriminant network.

[0027] Further, the method further includes the following steps:

[0028] Train a fault diagnosis model using the synthetic fault data marked with fault types to obtain a trained fault diagnosis model.

[0029] In a second aspect, the present application provides a migration fault diagnosis device based on digital twin and generative adversarial network, and the device includes:

[0030] A model construction module, which is used to construct a digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target instrument;

[0031] A data acquisition module, which is used to obtain simulated fault data at a preset fault depth and a preset fault width based on the digital twin model;

[0032] A data synthesis module, which is used to input the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; the synthetic fault data is used to train the fault diagnosis model;

[0033] A fault judgment module, which is used to judge the fault type of the rolling bearing of the target instrument based on the fault diagnosis model and the actual operation data of the rolling bearing.

[0034] Further, the model construction module includes;

[0035] A geometric model construction sub - module, which is used to generate a geometric model of the rolling bearing based on the geometric dimensions, shapes of the components in the rolling bearing of the target instrument and the positional relationship between the components;

[0036] A physical model construction sub - module, which is used to obtain a physical model mapping the physical state of the rolling bearing itself based on the physical properties of the rolling bearing of the target instrument;

[0037] A behavior model construction sub - module, which is used to obtain a behavior model of the rolling bearing based on the action responses of the rolling bearing to the internal and external environments and system instructions;

[0038] A rule model construction sub-module, which is used to obtain the rule model of the rolling bearing based on the historical operation data and the fault occurrence rules summarized from the corresponding operation status of the rolling bearing.

[0039] Further, the data acquisition module includes:

[0040] A model synthesis sub-module, which is used to obtain the digital twin model of the rolling bearing based on the geometric model, physical model, behavior model and rule model;

[0041] A fault setting sub-module, which is used to set a preset fault depth and a preset fault width for the digital twin model;

[0042] A model simulation sub-module, which is used to obtain the simulated fault data of the digital twin model based on the response of the digital twin model at the preset fault depth and preset fault width.

[0043] Further, the improved generative adversarial network model includes a generator network and a discriminator network.

[0044] Further, the generator network is composed of five convolutional layers, and the discriminator network is composed of an SCAE model, three convolutional layers and two fully connected layers.

[0045] Further, the device further includes a discriminator network training module:

[0046] A sample generation sub-module, which is used to randomly sample Gaussian noise to obtain real samples and generated samples;

[0047] A first training sub-module, which is used to train the discriminator network using the first cross-entropy cost function until the cost function error of the class labels of the generated samples and the real samples is minimized;

[0048] A second training sub-module, which is used to train the generator network using the second cross-entropy cost function until the cost function error of the authenticity labels of the generated samples and the real samples is minimized, and complete the training process of the discriminator network.

[0049] Further, the device is also used to train a fault diagnosis model using the synthetic fault data marked with fault types to obtain a trained fault diagnosis model.

[0050] The beneficial effects brought by the technical solution provided by this application include:

[0051] In this application, the server constructs a digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target device; based on the digital twin model, obtains simulated fault data under a preset fault depth and a preset fault width; inputs the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; the synthetic fault data is used to train the fault diagnosis model; based on the fault diagnosis model and the actual operation data of the rolling bearing, determines the fault type of the rolling bearing of the target device. By constructing a digital twin model of the rolling bearing to generate virtual fault data, and eliminating the domain difference between the real fault data and the virtual fault data through the domain adaptation method of the auxiliary generative adversarial network, the fault diagnosis model trained on the virtual entity can be used for the fault diagnosis problem of the physical entity, solving the problem of insufficient fault data in the physical entity. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0053] Figure 1 It is a flowchart of the steps of the migration fault diagnosis method based on digital twin and generative adversarial network provided in the embodiments of the present application;

[0054] Figure 2 It is a comparison of the bearing inner ring fault simulation data, synthetic data and real data in the embodiments of the present application;

[0055] Figure 3 It is a comparison of the bearing rolling element fault simulation data, synthetic data and real data in the embodiments of the present application;

[0056] Figure 4 It is a comparison of the bearing outer ring fault simulation data, synthetic data and real data in the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0058] The following further elaborates on the embodiments of the present application with reference to the drawings.

[0059] An embodiment of the present application provides a migration fault diagnosis method based on digital twin and generative adversarial network to solve the problem of insufficient fault data in physical entities.

[0060] To achieve the above technical effects, the general idea of the present application is as follows:

[0061] See Figure 1 As shown, a migration fault diagnosis method based on digital twin and generative adversarial network, the method includes the following steps:

[0062] S1. Based on the actual size of the rolling bearing of the target instrument, construct a digital twin model of the rolling bearing;

[0063] Among them, the digital twin model is a simulation twin model with bearing dynamic performance.

[0064] The server generates a geometric model of the rolling bearing based on the geometric dimensions, shapes of the components in the rolling bearing of the target instrument and the positional relationship between the components; obtains a physical model that maps the physical state of the rolling bearing itself based on the physical properties of the rolling bearing of the target instrument; obtains a behavior model of the rolling bearing based on the action response of the rolling bearing to the internal and external environment and system instructions; obtains a rule model of the rolling bearing based on the historical operation data and the fault occurrence rules summarized from the corresponding rolling bearing operation states.

[0065] S2. Based on the digital twin model, obtain simulated fault data under a preset fault depth and a preset fault width;

[0066] Based on the geometric model, physical model, behavior model and rule model, obtain a digital twin model of the rolling bearing; set a preset fault depth and a preset fault width for the digital twin model; obtain simulated fault data of the digital twin model based on the response of the digital twin model under the preset fault depth and the preset fault width.

[0067] S3. Input the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data;

[0068] Among them, the synthetic fault data is used to train the fault diagnosis model; the generative adversarial network model includes a generative network and a discriminative network; the generative network consists of five convolutional layers, and the discriminative network consists of an SCAE model, three convolutional layers and two fully connected layers.

[0069] Specifically, the server inputs the simulated fault data into the generative network in the generative adversarial network model to obtain generated fault data, inputs the generated fault data into the discriminator in the generative adversarial network for authenticity determination, and if the discriminator determines that the generated fault data is true, outputs the generated fault data.

[0070] Among them, the training process of the discriminator includes: randomly sampling Gaussian noise to obtain real samples and generated samples; training the discriminant network using the first cross-entropy cost function until the cost function error of the class labels of the generated samples and real samples is minimized; training the generator network using the second cross-entropy cost function until the cost function error of the authenticity labels of the generated samples and real samples is minimized, thus completing the training process of the discriminant network.

[0071] S4. Based on the fault diagnosis model and the actual operation data of the rolling bearing, determine the fault type of the rolling bearing of the target device.

[0072] Specifically, the server uses the synthetic fault data marked with fault types to train the fault diagnosis model, obtains the trained fault diagnosis model, and based on the fault diagnosis model and the actual operation data of the rolling bearing, determines the fault type of the rolling bearing of the target device.

[0073] In the embodiment of the present application, a digital twin model of a rolling bearing is constructed to generate virtual fault data, and the domain adaptation method of an auxiliary generative adversarial network is used to eliminate the domain difference between the real fault data and the virtual fault data, so that the fault diagnosis model trained on the virtual entity can be used for the fault diagnosis problem of the physical entity, and the problem of insufficient fault data in the physical entity is solved.

[0074] In one embodiment, the improved generative adversarial network model includes a generator network and a discriminator network.

[0075] The generator network consists of five one-dimensional convolutional layers. The size of the convolutional kernel of each layer is 5*5, the stride is 1, and batch normalization and LeakyReLU activation functions are added after each convolutional layer; the discriminator network consists of an SCAE model, 3 one-dimensional convolutional layers and two fully connected layers. The SCAE model is set as a four-layer network, and the number of neurons in each layer of the network is 1000, 300, 25 and 5; the size of the convolutional kernel of each layer is 5*5, the stride is 2, and the last two fully connected layers are used for judging whether the fault is real and for class judgment. Among them, the output of the fully connected layer for judging whether the fault is real is 1, and the activation function is Tanh, and the output of the fully connected layer for class judgment is 3, and the activation function is Softmax.

[0076] The premise for traditional data-driven fault diagnosis methods to achieve better performance is to obtain a large amount of training data sets. However, during the operation of devices with high safety requirements, the frequency of faults is low, and the vast majority of the monitored data are normal operation data of the devices, resulting in insufficient fault data. The present invention uses digital twin technology to establish a twin model, which can effectively solve the problem of insufficient fault samples.

[0077] In one embodiment, the training process of the discrimination network includes:

[0078] S301, randomly sample Gaussian noise to obtain real samples and generated samples;

[0079] The noise vector obtained by randomly sampling Gaussian noise The generation network maps it to a hidden layer vector Form a generated sample The corresponding class label is

[0080]

[0081]

[0082] Where: θ z , θ z ' are the parameter sets of the input layer and the output layer of the generation network respectively, and θ z = {W z , b z}, θ z ' = {W z ', b' z}; W z and W z ' are weight matrices; b z and b' z are bias vectors.

[0083] S302, train the discrimination network using the first cross-entropy cost function until the cost function error of the class labels of the generated samples and the real samples is minimized;

[0084] S303, train the generation network using the second cross-entropy cost function until the cost function error of the authenticity labels of the generated samples and the real samples is minimized, and complete the training process of the discrimination network.

[0085] Label as 0, and label the real sample as 1. Input the real sample and the generated sample into the SCAE for authenticity determination and fault identification, and give the authenticity label of the sample through Equation 1 and the class label

[0086]

[0087] Where, y m is the output class label; θ N+1 is the classification layer parameter set; s g is the classification layer activation function.

[0088] SA-SCAE-ACGAN completes the training of discriminator SCAE by minimizing the cost function error between the true / false label and the class label. Based on the cross-entropy cost function l SA-SCAE-ACGAN-D is:

[0089]

[0090] where: l c is the cross-entropy cost function error of the class label; l d is the cross-entropy cost function error of the true / false label.

[0091]

[0092]

[0093] where: Θ is the parameter set, Θ = {θ1, θ2,..., θ N+1}.

[0094] In one embodiment, the training process of the generator includes: First, label as 1. When the discriminator SCAE discriminates the generated samples and the output result is false (the true / false label result output is 0), it means that the generated samples of the generator cannot successfully deceive the discriminator, and then the generator network is feedback-regulated through the following formula.

[0095]

[0096] where: l SA-SCAE-ACGAN-G is the cost function of the generator; Θ' is the parameter set, Θ' = (θ z , θ z '); l g is the cross-entropy cost function error of the true / false label.

[0097]

[0098]

[0099] In the formula, is the smoothing coefficient; α is the variation amplitude coefficient ξ; introducing ξ is to prevent ω from being too large due to D(G(z r )) being too small; the variation amplitude coefficient α can make ω change more when the performance of the discriminator and the generator differ greatly, and accelerate the training convergence speed.

[0100] In this embodiment, aiming at the large difference in feature distribution between the simulated fault samples and the real samples, which reduces the accuracy of fault diagnosis, an improved auxiliary classification generative adversarial network is adopted to eliminate the domain difference between the real data and the simulated fault data.

[0101] In an application embodiment, as Figures 2 - 4 shown, Matlab2017 software is used to perform simulation verification on the comparison between simulated data and real data. Among them, Figure 2 is the comparison between the simulated data, synthetic data and real data of the inner ring fault of the bearing; Figure 3 is the comparison between the simulated data, synthetic data and real data of the rolling element fault of the bearing; Figure 4 is the comparison between the simulated data, synthetic data and real data of the outer ring fault of the bearing in the application embodiment; among them, the bearing model is selected as SKF6025, the number of rollers is set to 9, the pitch diameter is 38.55 mm, the rolling element diameter is 7.938 mm, the sampling rate is 12 khz, the rotational speed is 1772 rpm, the fault width is 0.07 mm, and the fault depth is 0.28 mm. Among them, the settings of the generator and discriminator in the improved auxiliary classification generative adversarial network are as follows: both the generator and discriminator adopt the Adma optimizer, and the learning rate is set to 0.0001. The smoothing coefficient ζ of the adaptive loss function in the generator takes the value of 1, and the variation amplitude coefficient α takes the value of 2. At this time, the value range of ω is 0.25 - 4.00. Under this value condition, the model training converges quickly and stably. Through experimental verification, the digital twin technology generates simulated data reflecting the bearing fault characteristics, and the improved auxiliary classification generative adversarial network can effectively eliminate the domain difference between the simulated data and the real data.

[0102] In this embodiment, in order to match the performance of the generator and discriminator in the auxiliary classification generative adversarial network, the generator adopts an adaptive loss function, and by adaptively adjusting the loss value of the generator, the training converges faster and the generated data quality is better; the discriminator adopts SCAE, and uses the good anti-data fluctuation ability of SCAE to extract effective depth features from the extended sample set and realize the determination of the authenticity and category of the samples.

[0103] It should be noted that the step numbers of each step in the application embodiment do not limit the sequence of each operation in the technical solution of the application.

[0104] In the second aspect, based on the same inventive concept as the real-time example of the transfer fault diagnosis method based on digital twin and generative adversarial network, the application embodiment provides a transfer fault diagnosis device based on digital twin and generative adversarial network. The device includes:

[0105] A model construction module, which is used to construct a digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target instrument;

[0106] A data acquisition module, which is used to acquire simulated fault data under a preset fault depth and a preset fault width based on the digital twin model;

[0107] A data synthesis module, which is used to input the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; the synthetic fault data is used to train a fault diagnosis model.

[0108] A fault judgment module, which is used to judge the fault type of the rolling bearing of the target device based on the fault diagnosis model and the actual operation data of the rolling bearing.

[0109] In this application, a digital twin model of a rolling bearing is constructed to generate virtual fault data, and the domain adaptation method of an auxiliary generative adversarial network is used to eliminate the domain difference between the real fault data and the virtual fault data, so that the fault diagnosis model trained on the virtual entity can be used for the fault diagnosis problem of the physical entity, solving the problem of insufficient fault data in the physical entity.

[0110] It should be noted that for the migration fault diagnosis device based on digital twin and generative adversarial network provided in the embodiments of this application, the corresponding technical problems, technical means, and technical effects are similar in principle to those of the migration fault diagnosis method based on digital twin and generative adversarial network.

[0111] In a second aspect, an embodiment of this application provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the migration fault diagnosis method based on digital twin and generative adversarial network mentioned in the first aspect.

[0112] In a third aspect, an embodiment of this application provides an electronic device, including a memory and a processor. A computer program is stored on the memory and runs on the processor. When the processor executes the computer program, it implements the migration fault diagnosis method based on digital twin and generative adversarial network mentioned in the first aspect.

[0113] It should be noted that in this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover a non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the element.

[0114] The above are only specific embodiments of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but rather will be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A migration fault diagnosis method based on digital twin and generative adversarial network, characterized in that, The method includes the following steps: Based on the actual size of the rolling bearing of the target device, construct a digital twin model of the rolling bearing; Based on the digital twin model, obtain simulated fault data at a preset fault depth and a preset fault width; Input the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; the synthetic fault data is used to train a fault diagnosis model; Based on the fault diagnosis model and the actual operation data of the rolling bearing, determine the fault type of the rolling bearing of the target device; the constructing the digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target device includes the following steps: Based on the geometric dimensions, shapes of the components in the rolling bearing of the target device, and the positional relationships between the components, generate a geometric model of the rolling bearing; based on the physical properties of the rolling bearing of the target device, obtain a physical model that maps the physical state of the rolling bearing itself; Based on the action responses of the rolling bearing to the internal and external environments and system instructions, obtain a behavior model of the rolling bearing; Based on the fault occurrence rules summarized from historical operation data and the corresponding operating states of the rolling bearing, obtain a rule model of the rolling bearing; The obtaining the simulated fault data at a preset fault depth and a preset fault width based on the digital twin model includes the following steps: Based on the geometric model, physical model, behavior model, and rule model, obtain the digital twin model of the rolling bearing; Set a preset fault depth and a preset fault width for the digital twin model; Based on the responses of the digital twin model at the preset fault depth and the preset fault width, obtain the simulated fault data of the digital twin model; The improved generative adversarial network model includes a generator network and a discriminator network; The training process of the discriminator network includes: S301, randomly sample Gaussian noise to obtain real samples and generated samples; The noise vector obtained by randomly sampling Gaussian noise The generation network maps it to a hidden layer vector Form a generated sample The corresponding class label is where: θ z , θ z ' are the parameter sets of the input layer and the output layer of the generation network, respectively, and θ z = {W z , b z}, θ z ' = {W z ', b' z}; W z and W z ' are weight matrices; b z and b' z are bias vectors; S302, train the discriminator network using the first cross-entropy cost function until the cost function error of the class labels of the generated samples and the real samples is minimized; S303, train the generator network using the second cross-entropy cost function until the cost function error of the authenticity labels of the generated samples and the real samples is minimized, completing the training process of the discriminator network; Label as 0, and label the real sample as 1. Input the real sample and the generated sample into SCAE for authenticity determination and fault identification, and give the authenticity label of the sample through Formula 1 and the class label Among them, y m is the output class label; θ N+1 is the parameter set of the classification layer; s g is the activation function of the classification layer; SA-SCAE-ACGAN completes the training of discriminator SCAE by minimizing the cost function error between the true / false label and the class label; based on the cross-entropy cost function l SA-SCAE-ACGAN-D is as follows: where: l c is the cross-entropy cost function error of the class label; l d is the cross-entropy cost function error of the authenticity label; where: Θ is a parameter set, Θ = {θ1, θ2,..., θ N+1}; The training process of the generator includes: First, is labeled as 1, and the discriminator SCAE discriminates the generated samples. When the output result is false, it means that the generated samples of the generator cannot successfully deceive the discriminator, and then the following formula is used to feedback and adjust the generation network; where: l SA-SCAE-ACGAN-G is the cost function of the generator; Θ' is the parameter set, Θ' = (θ z , θ′ z ); l g is the cross-entropy cost function error of the true / false label; where ξ is the smoothing coefficient; α is the variation amplitude coefficient; introducing ξ is to prevent ω from being too large due to D(G(z r )) being too small; the variation amplitude coefficient α can make ω change more when the performance of the discriminator and the generator differ greatly, accelerating the training convergence speed.

2. The migration fault diagnosis method based on digital twin and generative adversarial network according to claim 1, wherein The generator network consists of five convolutional layers, and the discriminator network consists of an SCAE model, three convolutional layers, and two fully connected layers.

3. The migration fault diagnosis method based on digital twin and generative adversarial network according to claim 1, characterized in that, The training process of the discriminator network includes: Randomly sample Gaussian noise to obtain real samples and generated samples; Train the discriminator network using the first cross-entropy cost function until the cost function error of the class labels of the generated samples and the real samples is minimized; train the generator network using the second cross-entropy cost function until the cost function error of the authenticity labels of the generated samples and the real samples is minimized, completing the training process of the discriminator network.

4. The migration fault diagnosis method based on digital twin and generative adversarial network according to claim 1, characterized in that, The method further includes the following steps: training a fault diagnosis model using the synthetic fault data marked with fault types to obtain a trained fault diagnosis model.

5. A migration fault diagnosis device based on digital twin and generative adversarial network, characterized in that, The device includes: a model construction module, which is used to construct a digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target device; a data acquisition module, which is used to obtain simulated fault data at a preset fault depth and a preset fault width based on the digital twin model; a data synthesis module, which is used to input the simulated fault data into the improved generative adversarial network model to obtain synthetic fault data corresponding to the simulated fault data; the synthetic fault data is used to train the fault diagnosis model; a fault judgment module, which is used to judge the fault type of the rolling bearing of the target device based on the fault diagnosis model and the actual operation data of the rolling bearing; The constructing of the digital twin model of the rolling bearing based on the actual size of the rolling bearing of the target device includes the following steps: generating a geometric model of the rolling bearing based on the geometric dimensions, shapes of the components in the rolling bearing of the target device and the positional relationship between the components; obtaining a physical model that maps the physical state of the rolling bearing itself based on the physical properties of the rolling bearing of the target device; obtaining a behavior model of the rolling bearing based on the action response of the rolling bearing to the internal and external environment and system instructions; obtaining a rule model of the rolling bearing based on the fault occurrence rules summarized from the historical operation data and the corresponding rolling bearing operation states; The obtaining of the simulated fault data at a preset fault depth and a preset fault width based on the digital twin model includes the following steps: obtaining the digital twin model of the rolling bearing based on the geometric model, physical model, behavior model and rule model; setting a preset fault depth and a preset fault width for the digital twin model; obtaining the simulated fault data of the digital twin model based on the response of the digital twin model at the preset fault depth and the preset fault width; The improved generative adversarial network model includes a generative network and a discriminative network; The training process of the discriminative network includes: S301, randomly sampling Gaussian noise to obtain real samples and generated samples; The noise vector obtained by randomly sampling Gaussian noise The generation network maps it to a hidden layer vector Form a generated sample The corresponding class label is Where: θ z , θ z ' are the parameter sets of the input layer and the output layer of the generation network, respectively, and θ z = {W z , b z}, θ z ' = {W z ', b' z}; W z and W z ' are weight matrices; b z and b' z are bias vectors; S302, training the discriminative network using the first cross-entropy cost function until the cost function error of the class labels of the generated samples and the real samples is minimized; S303, training the generative network using the second cross-entropy cost function until the cost function error of the authenticity labels of the generated samples and the real samples is minimized, completing the training process of the discriminative network; Label as 0 for the real sample as 1, input the real sample and the generated sample into SCAE for authenticity determination and fault identification, and give the authenticity label of the sample through Formula 1 and the class label where y m is the output class label; θ N+1 is the parameter set of the classification layer; s g is the activation function of the classification layer; SA-SCAE-ACGAN completes the training of discriminator SCAE by minimizing the cost function error between the real / fake label and the class label; based on the cross-entropy cost function l SA-SCAE-ACGAN-D is as follows: where: l c is the cross-entropy cost function error of the class label; l d is the cross-entropy cost function error of the authenticity label; where: Θ is a parameter set, Θ = {θ1, θ2,..., θ N+1}; The training process of the generator includes: First, is labeled as 1, and the discriminator SCAE discriminates the generated samples. When the output result is false, it means that the generated samples of the generator cannot successfully deceive the discriminator, and then the following formula is used to feedback and adjust the generation network; where: l SA-SCAE-ACGAN-G is the cost function of the generator; Θ' is the parameter set, Θ' = (θ z , θ′ z ); l g is the cross-entropy cost function error of the true / false label; where ξ is the smoothing coefficient; α is the variation amplitude coefficient; introducing ξ is to prevent ω from being too large due to D(G(z r )) being too small; the variation amplitude coefficient α can make ω change more when the discriminator performance and the generator performance differ greatly, and accelerate the training convergence speed.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 4.

Images