Gearbox fault diagnosis method and device based on digital twinborn and composite adversarial domain discrimination network

Abstract

The invention provides a gearbox fault diagnosis method and device based on a digital twinborn and composite adversarial domain discrimination network, and the method comprises the steps: inputting a target actual measurement operation parameter of a target gearbox into a feature extraction module of a fault diagnosis model, and obtaining a target feature vector of the target actual measurement operation parameter; inputting the target feature vector into a classification module of a fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; wherein the fault diagnosis model is obtained by performing iterative training on a composite adversarial domain discrimination network based on sample actual measurement operation parameters and sample simulation operation parameters of the sample gearbox and fault labels corresponding to the sample simulation operation parameters; the sample simulation operation parameters are obtained by simulating a dynamic model of the digital twinning dynamic simulation model of the sample gearbox. According to the invention, effective and accurate fault detection can be carried out on the gearbox under the condition that the effectively marked fault sample data is limited.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a gearbox fault diagnosis method and device based on digital twin and composite adversarial domain discrimination network. Background Art

[0002] In the development of modern industry, the use of mechanical equipment is indispensable. Gearboxes are widely used in various industrial mechanical equipment due to their high transmission efficiency, strong load-bearing capacity and good stability. Gearboxes usually operate in various complex environments, and long-term operation will inevitably cause various failures and wear. In actual production, failures, wear, etc. will have an adverse effect on the reliable operation of mechanical equipment, which may lead to production interruptions, losses, and even huge safety accidents. Therefore, analyzing and diagnosing the fault status of gearboxes is crucial to the operation of mechanical equipment.

[0003] In recent years, machine learning has been widely used in mechanical equipment fault diagnosis technology and has achieved good performance due to its strong nonlinear representation ability and ability to adaptively extract features from large amounts of data. Although machine learning has outstanding performance, it relies heavily on a large number of labeled samples for model training and has weak capabilities in testing data for new tasks.

[0004] However, under actual working conditions, the amount of fault data collected by the sensor is far less than the normal data, so it is difficult to obtain a large amount of fault data. Training using only limited labeled data will lead to overfitting and poor generalization of the model, making it difficult to ensure the accuracy of the gearbox fault diagnosis results. Summary of the invention

[0005] The present invention provides a gearbox fault diagnosis method and device based on digital twin and composite adversarial domain discrimination network, so as to solve the defect that effectively labeled fault sample data in the prior art is relatively scarce and it is difficult to ensure the accuracy of the gearbox fault diagnosis results, and to improve the accuracy of gearbox fault diagnosis when effectively labeled fault sample data is scarce.

[0006] The present invention provides a gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network, comprising: Obtaining target measured operating parameters of a target gearbox; Inputting the target measured operating parameters into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameters; Inputting the target feature vector into the classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; Among them, the fault diagnosis model is obtained by iteratively training the composite adversarial domain discriminant network based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

[0007] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the present invention, the sample simulation operation parameters of the sample gearbox are obtained based on the following steps: Constructing a digital twin dynamic simulation model of the sample gearbox according to the nonlinear motion characteristics of the sample gearbox; Based on a lumped parameter algorithm and according to the nonlinear motion characteristics of the sample gearbox, a dynamic model of the digital twin dynamic simulation model is established; The dynamic model is simulated using a preset fault state and a preset simulation condition as input parameters to obtain sample simulation operation parameters of the sample gearbox; Among them, the preset simulation conditions are used to constrain multiple items of contact force configuration information, kinematic pair configuration information, material configuration information, driving force configuration information, load configuration information, and integrator configuration information in the digital twin dynamic simulation model.

[0008] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the present invention, the lumped parameter algorithm is based on the nonlinear motion characteristics of the sample gearbox to establish a dynamic model of the digital twin dynamic simulation model, including: According to the nonlinear motion characteristics of the sample gearbox, the rotational freedom equation of the driving wheel and the translational freedom equations in different directions in the digital twin dynamic simulation model, as well as the rotational freedom equation of the driven wheel and the translational freedom equations in different directions in the digital twin dynamic simulation model are constructed; Based on the lumped parameter algorithm, the dynamic model is constructed by combining the rotational freedom equation of the driving wheel and the translational freedom equations in different directions, as well as the rotational freedom equation of the driven wheel and the translational freedom equations in different directions.

[0009] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network provided by the present invention, the composite adversarial domain discrimination network includes an initial feature extraction module, an initial classification module, an initial domain recognition module and an initial dynamic weighted joint distribution adaptive module; The fault diagnosis model is trained based on the following steps: Inputting sample measured operating parameters and sample simulated operating parameters of the sample gearbox into the initial feature extraction module to obtain a first sample feature vector of the sample simulated operating parameters and a second sample feature vector of the sample measured operating parameters; Inputting the first sample feature vector into the initial classification module to obtain fault diagnosis prediction information corresponding to the sample simulation operation parameter, and obtaining a first loss function according to the classification difference between the fault diagnosis prediction information corresponding to the sample simulation operation parameter and the fault label; Inputting the first sample feature vector and the second sample feature vector into the initial domain identification module, obtaining a first domain spatial feature corresponding to the sample simulation operation parameter and a second domain spatial feature corresponding to the sample measured operation parameter, and obtaining a second loss function according to a domain distribution difference between the first domain spatial feature and the second domain spatial feature; Inputting the first sample feature vector and the second sample feature vector into the initial dynamic weighted joint distribution adaptive module to obtain the spatial distribution difference between the sample measured operation parameters and the sample simulated operation parameters, and obtaining a third loss function according to the spatial distribution difference; Combining the first loss function, the second loss function and the third loss function, iteratively training the composite adversarial domain discriminant network; The fault diagnosis model is constructed according to the trained initial feature extraction module and the trained initial classification module in the trained composite adversarial domain discrimination network.

[0010] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the present invention, the first sample feature vector and the second sample feature vector are input into the initial dynamic weighted joint distribution adaptive module to obtain the spatial distribution difference between the sample measured operating parameters and the sample simulated operating parameters, and according to the spatial distribution difference, a third loss function is obtained, including: The first sample feature vector and the second sample feature vector are input into the initial dynamic weighted joint distribution adaptive module, and the initial dynamic weighted joint distribution adaptive module calculates the local maximum mean difference between the first sample feature vector and the second sample feature vector, and the conditional distribution alignment difference between the first sample feature vector and the second sample feature vector, and obtains the third loss function according to the local maximum mean difference and the conditional distribution alignment difference.

[0011] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the present invention, the conditional distribution alignment difference is calculated based on the following formula: ; in, Align the differences for the conditional distributions; is the distribution difference measurement function; is the number of fault categories; The sample simulation run parameters belong to the fault category The probability value of The parameters for the sample simulation runs are given in Belongs to the fault category Under the condition of The conditional distribution of The measured operating parameters of the sample belong to the fault category The probability value of For the sample measured operating parameters belong to the fault category Under the conditions of The first sample feature vector of The conditional distribution of .

[0012] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the present invention, the third loss function is obtained according to the local maximum mean difference and the conditional distribution alignment difference, including: Obtaining a dynamic adaptation factor according to an exponential coefficient corresponding to the local maximum mean difference and an exponential coefficient corresponding to the conditional distribution alignment difference; According to the dynamic adaptation factor, weighted addition is performed on the local maximum mean difference and the conditional distribution alignment difference to obtain the third loss function.

[0013] According to a gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the present invention, the initial feature extraction module is constructed based on a multi-layer spatial channel reconstruction convolution layer and a residual layer; The spatial channel reconstruction convolutional layer is constructed based on a spatial reconstruction unit and a channel reconstruction unit. The spatial reconstruction unit is used to obtain spatial refinement features, and the channel reconstruction unit is used to obtain channel refinement features.

[0014] The present invention also provides a gearbox fault diagnosis device based on digital twin and composite adversarial domain discrimination network, comprising: A data acquisition unit, used to acquire target measured operating parameters of a target gearbox; A feature extraction unit, used for inputting the target measured operating parameter into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameter; A fault diagnosis unit, used for inputting the target feature vector into the classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; Among them, the fault diagnosis model is obtained by iteratively training the composite adversarial domain discriminant network based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

[0015] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a gearbox fault diagnosis method based on digital twins and a composite adversarial domain discrimination network as described above is implemented.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the gearbox fault diagnosis method based on digital twins and composite adversarial domain discrimination networks as described in any one of the above is implemented.

[0017] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned gearbox fault diagnosis methods based on digital twins and composite adversarial domain discrimination networks.

[0018] The gearbox fault diagnosis method and device based on digital twin and composite adversarial domain discrimination network provided by the present invention generate high-fidelity dynamic response data of various health states through digital twin simulation, which solves the problem that deep learning algorithms require a large amount of data for training while there is insufficient fault data in actual industrial scenarios. A data set based on the combination of sample gearbox actual measurement and digital twin simulation is constructed, and the composite adversarial domain discrimination network is iteratively trained to obtain a fault diagnosis model that can transfer knowledge from simulation data to measurement data and has higher generalization and accuracy. This effectively makes up for the limitation of scarce fault data in the actual working environment, avoids the overfitting problem caused by relying solely on limited labeled data, and significantly improves the generalization ability of the model, thereby ensuring the accuracy of the gearbox fault diagnosis results. This achieves effective and accurate fault detection of the gearbox even when the effectively labeled fault sample data is limited. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is one of the flow charts of the gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network provided by the present invention.

[0021] Figure 2 This is the second flow chart of the gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network provided by the present invention.

[0022] Figure 3 It is a schematic diagram of the digital twin dynamic simulation model provided by the present invention.

[0023] Figure 4 It is a structural schematic diagram of the spatial channel reconstruction convolutional layer provided by the present invention.

[0024] Figure 5 It is a structural schematic diagram of the space reconstruction unit provided by the present invention.

[0025] Figure 6 It is a structural schematic diagram of the channel reconstruction unit provided by the present invention.

[0026] Figure 7 It is a comparative simulation diagram of the accuracy of the fault diagnosis method provided by the present invention.

[0027] Figure 8 It is a structural schematic diagram of a gearbox fault diagnosis device based on digital twin and composite adversarial domain discrimination network provided by the present invention.

[0028] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] In recent years, machine learning has been widely used in mechanical equipment fault diagnosis technology and has achieved good performance due to its strong nonlinear representation ability and ability to adaptively extract features from large amounts of data. Although machine learning has outstanding performance, it relies heavily on a large number of labeled samples for model training and has weak capabilities in testing data for new tasks.

[0031] However, under actual working conditions, the amount of fault data collected by the sensor is far less than the normal data, so it is difficult to obtain a large amount of fault data. Training using only limited labeled data will lead to overfitting and poor generalization of the model, making it difficult to ensure the accuracy of the gearbox fault diagnosis results.

[0032] With the rapid development of intelligent manufacturing technology, the digital twin method has become an important intelligent operation and maintenance method. Its application mainly focuses on extracting fault data from virtual space to provide necessary supplementary information for training intelligent diagnosis models. However, gearboxes usually exhibit multiple fault conditions. When comparing data from different fields, there are obvious differences in the characteristic distribution of each health condition. Transfer learning technology has been proven to effectively solve the problem of significant distribution differences in sample data in the field of intelligent diagnosis. It learns feature knowledge from a source domain (simulated operating parameters) with sufficient labeled data, and then transfers this knowledge to a target domain (measured operating parameters) with limited labeled data. Therefore, advanced intelligent diagnosis technology has important practical value for equipment monitoring and maintenance.

[0033] In this context, an embodiment of the present application provides a gearbox fault diagnosis method based on digital twins and a composite adversarial domain discriminant network. The method is based on the construction and simulation of a digital twin model of the gearbox to generate high-fidelity dynamic response data of various health states, solving the problem that deep learning algorithms require a large amount of data for training while there is insufficient fault data in actual industrial scenarios; then, a composite adversarial domain discriminant network is used to transfer knowledge from simulation data to measurement data, thereby achieving effective and accurate fault detection of the gearbox when the effectively labeled fault sample data is limited.

[0034] Figure 1 is one of the flow charts of the gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network provided by the present invention, such as Figure 1 As shown, the method includes the following steps 110, 120 and 130.

[0035] Step 110, obtaining target measured operating parameters of the target gearbox.

[0036] Among them, the target gearbox is the gearbox required for fault diagnosis, and the so-called target measured operating parameters include multi-dimensional time series parameters formed by collecting the parameters generated by the target gearbox during actual operation at each sampling point in the current cycle, including but not limited to vibration response signals in the horizontal and vertical directions. The embodiment of the present application does not make specific limitations on this.

[0037] Optionally, when it is necessary to perform fault diagnosis on the target gearbox, the original measured operating parameters of the target gearbox can be obtained first, so as to obtain the target measured operating parameters based on the original measured operating parameters; here, the method of obtaining the target measured operating parameters can be to preprocess the original measured operating parameters, including but not limited to denoising, normalization, and feature conversion, thereby obtaining high-quality target measured operating parameters, so as to subsequently improve the performance of fault diagnosis.

[0038] Here, the target measured operating parameters of the target gearbox may be input by a user, or may be input through sharing of an acquisition device, and this embodiment of the present application does not specifically limit this.

[0039] Step 120, inputting the target measured operating parameter into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameter; Step 130, inputting the target feature vector into the classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; Among them, the fault diagnosis model is obtained by iteratively training the composite adversarial domain discriminant network based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

[0040] The fault diagnosis model here is constructed based on the feature extraction module and the classification module connected in sequence, wherein the feature extraction module is used to extract the high-dimensional nonlinear features associated with the fault diagnosis information in the input information, and the classification module is used to classify and output the corresponding fault diagnosis prediction results based on the high-dimensional nonlinear features. The fault diagnosis prediction information here includes information such as the presence or absence of a fault, the severity of the fault, and the specific type of the fault.

[0041] Figure 2 2 is a flow chart of a gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network provided by the present invention; Figure 2 As shown, before executing step 120, it is necessary to pre-train a fault diagnosis model, and the specific steps include: Firstly, a digital twin dynamic simulation model of a large number of sample gearboxes under multi-source fault categories and a dynamic model of the digital twin dynamic simulation model are established to effectively simulate the operating characteristics of the sample gearbox, and then simulate and generate the operating parameters of the sample gearbox under different health states, such as vibration response signals, thereby obtaining high-fidelity simulated fault data, that is, the sample simulation operating parameters.

[0042] Next, based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox, the composite adversarial domain discriminant network is subjected to dynamic weighted joint distribution adaptive adversarial training to construct a fault diagnosis model that can accurately derive fault diagnosis prediction information from the measured operating parameters.

[0043] The Hybrid Adversarial Domain Discriminative Networks (HADDN) herein is a deep learning model for domain adaptation, which can transfer knowledge from virtual space to physical space, so as to reduce the distribution difference between simulated operating parameters and measured operating parameters, so that the model can mine valuable fault knowledge from virtual data, thereby improving the generalization ability of the model on measured operating parameters. The model parameter initialization of the network can be achieved by pre-training based on the operating parameters and fault diagnosis labels of equipment in general fields, or by random initialization, which is not specifically limited in the embodiments of the present application.

[0044] Optionally, after the fault diagnosis model is trained, the target measured operating parameters of the target gearbox may be input into a feature extraction module of the fault diagnosis model, so that the feature extraction module of the fault diagnosis model can perform high-dimensional nonlinear feature extraction on the target measured operating parameters to obtain a target feature vector of the target measured operating parameters.

[0045] Subsequently, the target feature vector is input into a classification module of a fault diagnosis model, so that the classification module of the fault diagnosis model classifies and outputs fault diagnosis prediction information of the target gearbox according to the target feature vector.

[0046] The method provided in the embodiment of the present application generates high-fidelity dynamic response data of various health states through digital twin simulation, which solves the problem that deep learning algorithms require a large amount of data for training while there is insufficient fault data in actual industrial scenarios. A data set based on the combination of sample gearbox actual measurement and digital twin simulation is constructed, and the composite adversarial domain discriminant network is iteratively trained to obtain a fault diagnosis model that can transfer knowledge from simulation data to measurement data and has higher generalization and accuracy. This effectively makes up for the limitation of scarce fault data in the actual working environment, avoids the overfitting problem caused by relying solely on limited labeled data, and significantly improves the generalization ability of the model, thereby ensuring the accuracy of the gearbox fault diagnosis results. This achieves effective and accurate fault detection of the gearbox even when the effectively labeled fault sample data is limited.

[0047] In some embodiments, the sample simulated operating parameters of the sample gearbox are obtained based on the following steps: Constructing a digital twin dynamic simulation model of the sample gearbox according to the nonlinear motion characteristics of the sample gearbox; Based on a lumped parameter algorithm and according to the nonlinear motion characteristics of the sample gearbox, a dynamic model of the digital twin dynamic simulation model is established; The dynamic model is simulated using a preset fault state and a preset simulation condition as input parameters to obtain sample simulation operation parameters of the sample gearbox; Among them, the preset simulation conditions are used to constrain multiple items of contact force configuration information, kinematic pair configuration information, material configuration information, driving force configuration information, load configuration information, and integrator configuration information in the digital twin dynamic simulation model.

[0048] Optionally, because the amount of fault data collected by the sensor under actual working conditions is far less than the amount of normal data, it is difficult to obtain a large amount of fault data, which results in limited model performance; in this regard, the embodiment of the present application generates sample simulation operation parameters under different health states through simulation, so as to enrich the training data set and alleviate the problem of sample scarcity. Here, the steps for obtaining the sample simulation operation parameters specifically include: Step 210, constructing a digital twin: constructing a digital twin dynamic simulation model of the sample gearbox according to the nonlinear motion characteristics of the sample gearbox, and using the lumped parameter method to analyze the nonlinear motion characteristics of the sample gearbox, so as to establish a dynamic model of the digital twin dynamic simulation model based on this, so as to discretely simplify the gearbox, optimize the calculation, and accurately simulate and generate sample simulation operating parameters of the gearbox.

[0049] The dynamic model of the digital twin dynamic simulation model here can be constructed based on the degree of freedom dynamic model of each component of the virtual gearbox corresponding to the digital twin dynamic simulation model; wherein, the degree of freedom dynamic model of each component of the virtual gearbox includes but is not limited to the rotational degree of freedom equation of the driving wheel of the virtual gearbox and the degree of freedom equation of the translational motion in different directions, as well as the rotational degree of freedom equation of the driven wheel of the virtual gearbox and the degree of freedom equation of the translational motion in different directions.

[0050] Exemplarily, the step of constructing the kinetic model further comprises: According to the nonlinear motion characteristics of the sample gearbox, the rotational freedom equation of the driving wheel and the translational freedom equations in different directions in the digital twin dynamic simulation model, as well as the rotational freedom equation of the driven wheel and the translational freedom equations in different directions in the digital twin dynamic simulation model are constructed; Based on the lumped parameter algorithm, the dynamic model is constructed by combining the rotational freedom equation of the driving wheel and the translational freedom equations in different directions, as well as the rotational freedom equation of the driven wheel and the translational freedom equations in different directions.

[0051] Figure 3 is a schematic diagram of the digital twin dynamic simulation model provided by the present invention; Figure 3 As shown, according to the nonlinear motion characteristics of the sample gearbox, six degree-of-freedom equations can be constructed, including the rotational degree-of-freedom equation of the driving wheel, the translational degree-of-freedom equation in the x-axis and y-axis directions, and the rotational degree-of-freedom equation of the driven wheel, and the translational degree-of-freedom equation in the x-axis and y-axis directions.

[0052] Next, assuming that the sliding friction between the teeth is perpendicular to the meshing line, the driving and driven wheels are regarded as rigid plates with small holes, and the spring damper is used to simulate the joint of the meshing gears. On this basis, based on the lumped parameter algorithm, the rotational freedom equation of the driving wheel, the translational freedom equation of the x-axis and y-axis directions, and the rotational freedom equation of the driven wheel, the translational freedom equation of the x-axis and y-axis directions are combined to construct the dynamic model shown below: ; ; ; ; ; ; in, and are the translation displacements of the driving wheel in the x and y directions respectively; and are the first-order derivative and the second-order derivative of the translation displacement of the driving wheel in the x direction respectively; and are the first-order derivative and the second-order derivative of the translation displacement of the driving wheel in the y direction respectively; and They are Direction Radial support stiffness and damping of the impeller, , representing different directions and the master and driven wheels; is the sliding friction between teeth; , , , are the mass, moment of inertia, friction moment and torque of the driving wheel respectively. and are the translation displacements of the driven wheel in the x and y directions respectively; , , , are the mass, moment of inertia, friction moment and torque of the driven wheel respectively; is the inter-tooth meshing damping, is the inter-tooth support stiffness, is the first-order derivative of the relative displacement between gears, yes; is the angular acceleration of the driving wheel, is the base radius of the driving wheel, is the angular acceleration of the driven wheel, is the base circle radius of the driven wheel.

[0053] Step 220, dynamic model simulation: Use Automatic Dynamic Analysis of Mechanical Systems (ADAMS) software to perform dynamic simulation, inject multiple preset simulation conditions, and set multiple items of contact force configuration information, kinematic pair configuration information, material configuration information, driving force configuration information, load configuration information, and integrator configuration information to form simulation conditions, thereby simulating the dynamic model to obtain vibration response signals of the gearbox under different fault modes through simulation, thereby obtaining high-fidelity simulated fault data.

[0054] Among them, for the material configuration information, the model material can be selected as steel material; for the kinematic pair configuration information, it is used to set the connection relationship between various components, mainly including: the fixed pair between the gear box and the earth, the rotating pair of the pinion, the rotating pair of the large gear, the fixed pair between the pinion and the pinion shaft, the fixed pair between the large gear and the large gear shaft, and the fixed pair between the four shafts and the gear box. For the driving force configuration information, load configuration information and integrator configuration information, the driving force can be set on the rotating pair of the large gear, the load is added to the rotating pair of the pinion shaft, the speed is 1500RPM, the integrator of the dynamic model is GSTIFF, and the integral format is I3.

[0055] From the above, it can be seen that by constructing a digital twin dynamic simulation model based on the nonlinear motion characteristics of the sample gearbox, the automatic degree equations of the driving wheel and the driven wheel in the rotational freedom and multiple translational motion directions are systematically established. Subsequently, these equations are combined using the concentrated parameter algorithm to finely construct a comprehensive dynamic model that reflects the dynamic behavior characteristics of the gearbox, aiming to ensure that the simulation process can accurately simulate the parameter changes in actual operation, thereby capturing the operating characteristics of the gearbox, and then improving the accuracy and reliability of the simulated operating parameters, so that the fault diagnosis model can be trained more accurately and efficiently based on these parameters in the future.

[0056] In some embodiments, the composite adversarial domain discrimination network includes an initial feature extraction module, an initial classification module, an initial domain recognition module, and an initial dynamic weighted joint distribution adaptation module; The fault diagnosis model is trained based on the following steps: Inputting sample measured operating parameters and sample simulated operating parameters of the sample gearbox into the initial feature extraction module to obtain a first sample feature vector of the sample simulated operating parameters and a second sample feature vector of the sample measured operating parameters; Inputting the first sample feature vector into the initial classification module to obtain fault diagnosis prediction information corresponding to the sample simulation operation parameter, and obtaining a first loss function according to the classification difference between the fault diagnosis prediction information corresponding to the sample simulation operation parameter and the fault label; Inputting the first sample feature vector and the second sample feature vector into the initial domain identification module, obtaining a first domain spatial feature corresponding to the sample simulation operation parameter and a second domain spatial feature corresponding to the sample measured operation parameter, and obtaining a second loss function according to a domain distribution difference between the first domain spatial feature and the second domain spatial feature; Inputting the first sample feature vector and the second sample feature vector into the initial dynamic weighted joint distribution adaptive module to obtain the spatial distribution difference between the sample measured operation parameters and the sample simulated operation parameters, and obtaining a third loss function according to the spatial distribution difference; Combining the first loss function, the second loss function and the third loss function, iteratively training the composite adversarial domain discriminant network; The fault diagnosis model is constructed according to the trained initial feature extraction module and the trained initial classification module in the trained composite adversarial domain discrimination network.

[0057] like Figure 2 As shown, the training steps of the fault diagnosis model further include: Step 230, construct a composite adversarial domain discrimination network; the composite adversarial domain discrimination network includes an initial feature extraction module , initial classification module , Initial domain identification module and the initial Dynamic Weighted Joint Distribution Adaptation (DWJDA) module; Among them, the initial feature extraction module It is used to extract the high-dimensional nonlinear features of the signal, which are then input into the initial classifier module, the initial domain identification module and the DWJDA module.

[0058] The initial feature extraction module here It can be constructed based on one or more neural network models, such as multiple constructions in a convolutional network, a residual network, and a spatial channel reconstruction network, to achieve high-dimensional nonlinear feature extraction of signals.

[0059] Exemplarily, the initial feature extraction module is constructed based on a multi-layer spatial channel reconstruction convolution layer and a residual layer; The spatial channel reconstruction convolutional layer is constructed based on a spatial reconstruction unit and a channel reconstruction unit. The spatial reconstruction unit is used to obtain spatial refinement features, and the channel reconstruction unit is used to obtain channel refinement features.

[0060] Optionally, in order to reduce redundant features in the data, the embodiment of the present application uses a multi-layer spatial and channel reconstruction convolution (SCConv) layer to replace the ordinary convolution layer in the deep residual network to effectively improve the efficiency of network learning.

[0061] Figure 4Schematic diagram of the structure of the spatial channel reconstruction convolutional layer provided by the present invention. Figure 4 As shown in the figure, the SCConv layer is a module that jointly reduces the spatial and channel redundant feature capabilities in the convolution layer. It consists of a spatial reconstruction unit (SRU) and a channel reconstruction unit (CRU). The SCConv layer first obtains spatial refinement features through SRU, and then obtains channel refinement features through CRU to reduce the redundancy between intermediate feature maps and improve the feature representation capability of the network. The residual layer can jump connect multiple SCConv layers to transmit shallow features losslessly to the deep convolution layer through jump connections, avoiding the loss of shallow features during the transmission of the deep network.

[0062] Figure 5 is a schematic diagram of the structure of the space reconstruction unit provided by the present invention; Figure 5 As shown, the spatial reconstruction unit is constructed based on a separation layer (or Separate) and a reconstruction layer (or Reconstruct) to output a spatial reconstruction feature (or Spatial-Refined feature).

[0063] Figure 6 is a schematic diagram of the structure of the channel reconstruction unit provided by the present invention; Figure 6 As shown in the figure, the channel reconstruction unit is used to convert the spatial reconstruction features output by the spatial reconstruction unit into channel-reconstructed features. After convolutional segmentation by the segmentation layer (or Split), the features are input into the feature transformation layer (or Transform) for global weighted cross-channel (or GWC) feature extraction and parallel weighted cross-channel (or PWC) feature extraction. The features are then input into the fusion layer for feature fusion and the channel-reconstructed features (or Channel-Refined feature) are output.

[0064] From the above, it can be seen that by constructing a spatial channel reconstruction convolutional layer containing a spatial reconstruction unit and a channel reconstruction unit in the initial feature extraction module of the composite adversarial domain discriminant network, data features can be effectively extracted from the channel and spatial dimensions, achieving an effective reduction in spatial and channel redundancy in the feature graph, improving the feature representation ability of the network, and losslessly transmitting shallow features through the jump connection of the residual layer, avoiding the loss of shallow features in the deep network, thereby improving the efficiency and accuracy of the model in feature extraction and representation, thereby effectively improving the fault diagnosis performance.

[0065] Among them, the initial classification module It can be constructed by the Softmax function, the initial domain recognition module It can be constructed from binary classifiers.

[0066] Step 240, after the composite adversarial domain discrimination network is constructed, the sample measured operating parameters and the sample simulated operating parameters of the sample gearbox can be input into the initial feature extraction module, so that the initial feature extraction module can extract high-dimensional features from the sample measured operating parameters and the sample simulated operating parameters, thereby obtaining a first sample feature vector of the sample simulated operating parameters and a second sample feature vector of the sample measured operating parameters.

[0067] Step 250, input the first sample feature vector extracted by the initial feature extraction module into the initial classification module to realize fault diagnosis of the sample gearbox, thereby obtaining fault diagnosis prediction information corresponding to the sample simulation operation parameters, and obtaining the first loss function according to the classification difference between the fault diagnosis prediction information corresponding to the sample simulation operation parameters and the fault label to minimize the fault classification error; the specific calculation formula of the first loss function is as follows: ; in, is the first loss function; The number of simulation runs for the labeled samples; For the Sample simulation operation parameters; It is the calculation function of the initial feature extraction module; is the calculation function of the initial classification module; is a classification loss function, such as a cross entropy loss function.

[0068] Step 260: Input the first sample feature vector and the second sample feature vector into the initial domain identification module, so that the initial domain identification module extracts domain space features from the input feature vectors to obtain first domain space features corresponding to the sample simulation operation parameters and second domain space features corresponding to the sample measured operation parameters; and calculate the second loss function based on the domain distribution difference between the first domain space features and the second domain space features to maximize the domain classification error; wherein the calculation formula of the second loss function is: ; in, is the second loss function; The number of measured operating parameters for unlabeled samples; is the calculation function of the initial domain recognition module; For the Sample simulation run parameters.

[0069] Step 270, input the first sample feature vector and the second sample feature vector into an initial dynamic weighted joint distribution adaptation (Dynamic Weighted Joint Distribution Adaptation, DWJDA) module, so that the DWJDA module calculates the spatial distribution difference between the sample measured operating parameters and the sample simulated operating parameters, and obtains a third loss function based on the spatial distribution difference to reduce the distribution difference between the virtual data and the real data, thereby improving the domain confusion capability of the network.

[0070] The spatial distribution difference here includes the local maximum mean difference and / or the conditional distribution alignment difference, and the third loss function here can be obtained by fusing the local maximum mean difference and / or the conditional distribution alignment difference.

[0071] Exemplarily, the third loss function can be obtained based on the following steps: The first sample feature vector and the second sample feature vector are input into the initial dynamic weighted joint distribution adaptive module, and the initial dynamic weighted joint distribution adaptive module calculates the local maximum mean difference between the first sample feature vector and the second sample feature vector, and the conditional distribution alignment difference between the first sample feature vector and the second sample feature vector, and obtains the third loss function according to the local maximum mean difference and the conditional distribution alignment difference.

[0072] Optionally, the conditional distribution alignment difference can be obtained by the following steps: Through Bayes' theorem, the conditional probability distribution is converted into the form of quasi-conditional probability distribution, which is expressed as follows: ; And based on the class conditional probability distribution, the conditional distribution alignment algorithm is improved to obtain the improved conditional distribution alignment (or ICDA) algorithm, which is specifically expressed as follows: ; Where C represents the number of fault categories.

[0073] The correlation alignment (CORrelation Alignmen, CORAL) and the maximum mean difference (Maximum Mean Discrepanc, MMD) are combined to form a new distribution difference measure, which is substituted into the improved ICDA mechanism to obtain the conditional distribution alignment difference.

[0074] Exemplarily, the conditional distribution alignment difference is calculated based on the following formula: ; in, Align the differences for the conditional distributions; is the distribution difference measurement function; is the number of fault categories; The sample simulation run parameters belong to the fault category The probability value of The parameters for the sample simulation runs are given in Belongs to the fault category Under the condition of The conditional distribution of The measured operating parameters of the sample belong to the fault category The probability value of For the sample measured operating parameters belong to the fault category Under the conditions of The first sample feature vector of The conditional distribution of .

[0075] The local maximum mean difference may be obtained by calculating the local maximum mean difference between the first sample feature vector and the second sample feature vector based on a Local Maximum Mean Discrepancy (LMMD) algorithm. The LMMD algorithm is a local domain adaptive algorithm that divides the sample space into multiple local regions, calculates the MMD of each local region, and uses the maximum value as the final metric.

[0076] After obtaining the local maximum mean difference and the conditional distribution alignment difference, the local maximum mean difference and the conditional distribution alignment difference may be fused to obtain a third loss function.

[0077] The fusion here can be obtained by weighted addition, addition or model fusion.

[0078] Exemplarily, the third loss function can be obtained by integrating the following steps: Obtaining a dynamic adaptation factor according to an exponential coefficient corresponding to the local maximum mean difference and an exponential coefficient corresponding to the conditional distribution alignment difference; According to the dynamic adaptation factor, weighted addition is performed on the local maximum mean difference and the conditional distribution alignment difference to obtain the third loss function.

[0079] Optionally, during training, there may be or The extreme case close to 0 will have an adverse effect on the overall calculation. Therefore, the embodiment of the present application performs exponential optimization expression according to the exponential coefficient corresponding to the local maximum mean difference and the exponential coefficient corresponding to the conditional distribution alignment difference to obtain the dynamic adaptation factor. The specific calculation formula is as follows: ; in, is a dynamic adaptation factor used to balance the importance of marginal and conditional probability distributions; is the exponential coefficient corresponding to the local maximum mean difference, is the exponential coefficient corresponding to the difference in conditional distribution alignment.

[0080] Next, the weight coefficients corresponding to the local maximum mean difference and the conditional distribution alignment difference are determined according to the dynamic adaptation factor, and the local maximum mean difference and the conditional distribution alignment difference are weightedly added based on the weight coefficient to obtain the third loss function. The specific calculation formula is as follows: ; in, is the third loss function.

[0081] Step 280, combining the first loss function, the second loss function and the third loss function to construct a total loss function, so as to iteratively train the composite adversarial domain discriminant network based on the total loss function.

[0082] Here, the method of constructing the total loss function includes weighted addition of the first loss function, the second loss function and the third loss function, or direct addition and fusion.

[0083] Exemplarily, the calculation formula of the total loss function is as follows: ; in, is the total loss function.

[0084] For example, during the training process, the model parameters of the composite adversarial domain discrimination network can be updated based on the following formula: ; in, are the model parameters of the composite adversarial domain discriminant network; is the learning rate.

[0085] Step 290, using the trained initial feature extraction module in the trained composite adversarial domain discrimination network as the feature extraction module of the fault diagnosis model, and using the trained initial classification module in the trained composite adversarial domain discrimination network as the classification module of the fault diagnosis model, to construct a fault diagnosis model.

[0086] The method provided in this embodiment establishes a composite adversarial domain discrimination network through modules such as spatial channel reconstruction convolution, domain adversarial, and dynamic weighted joint distribution adaptation, uses a deep residual SCConv network to extract deep features, and minimizes the subdomain features between simulation data and real data through DWJDA to reduce the distribution between the two types of data, so as to use a composite adversarial domain discrimination network to transfer knowledge from simulation data to measurement data, thereby realizing knowledge transfer from simulation to measured data, thereby allowing effective fault detection of gearboxes under limited information, thereby improving the generalization ability and accuracy of the fault diagnosis model.

[0087] In some embodiments, the embodiments of the present application use gearbox data under different working conditions as the target domain, and compare with other models and methods to verify the effectiveness of the method provided by the embodiments of the present application.

[0088] During the verification process, various health status data generated in the digital twin simulation model were used as the source domain, and the measured gearbox data under two working conditions were used as the target domain. Two groups of migration tasks were designed for simulation to verify the accuracy and transferability of the entire algorithm. An experimental result graph after parameter adjustment was given to verify the effectiveness of the algorithm.

[0089] Figure 7 It is a comparative simulation diagram of the accuracy of the fault diagnosis method provided by the present invention.

[0090] like Figure 7 As shown, a variety of conventional methods, such as the baseline method (referred to as Baseline), deep adaptation network (referred to as DAN), domain adversarial network (referred to as DANN), dynamic adversarial adaptive network (referred to as DAAN), conditional domain adversarial network (referred to as CDAN), partial domain adaptation network (referred to as PADA), and importance weighted adversarial network (referred to as IWAN), are compared with the accuracy (or Accuracy) of the method provided in the embodiment of the present application. After multiple verifications, the accuracy of the method provided in the embodiment of the present application (or Proposed) in the fault diagnosis task can reach more than 80%.

[0091] To further demonstrate the effectiveness of adversarial adaptive training in the method provided in the embodiment of the present application, the features between the source domain and the target domain are reduced to two dimensions using the t-Stochastic Neighbor Embedding (TSNE) method. It can be seen from the visualization that after the training is completed, the feature distribution distance between the source domain and the target domain is reduced, and feature alignment is effectively achieved.

[0092] In summary, the embodiments of the present application address the problem that data-driven fault diagnosis technology relies on sufficiently high-quality labeled data as training data, which is difficult to obtain in actual industrial scenarios. A gearbox fault diagnosis method based on digital twins and composite adversarial domain discriminant networks is proposed, and simulation experiments are carried out based on university gearbox data. Through experiments, the effectiveness of the algorithm model provided in the embodiments of the present application is effectively demonstrated.

[0093] The gearbox fault diagnosis device based on digital twin and composite adversarial domain discrimination network provided by the present invention is described below. The gearbox fault diagnosis device based on digital twin and composite adversarial domain discrimination network described below and the gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network described above can be referenced to each other.

[0094] Figure 8 is a structural schematic diagram of a gearbox fault diagnosis device based on digital twin and composite adversarial domain discrimination network provided by the present invention; Figure 8 As shown, the device comprises: The data acquisition unit 810 is used to acquire the target measured operating parameters of the target gearbox; The feature extraction unit 820 is used to input the target measured operating parameters into the feature extraction module of the fault diagnosis model to obtain the target feature vector of the target measured operating parameters; The fault diagnosis unit 830 is used to input the target feature vector into the classification module of the fault diagnosis model to obtain the fault diagnosis prediction information of the target gearbox; Among them, the fault diagnosis model is obtained by iteratively training the composite adversarial domain discriminant network based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

[0095] The device provided in the embodiment of the present application generates high-fidelity dynamic response data of various health states through digital twin simulation, which solves the problem that deep learning algorithms require a large amount of data for training while there is insufficient fault data in actual industrial scenarios. A data set based on the combination of sample gearbox actual measurement and digital twin simulation is constructed, and the composite adversarial domain discriminant network is iteratively trained to obtain a fault diagnosis model that can transfer knowledge from simulation data to measurement data and has higher generalization and accuracy. This effectively makes up for the limitation of scarce fault data in the actual working environment, avoids the overfitting problem caused by relying solely on limited labeled data, and significantly improves the generalization ability of the model, thereby ensuring the accuracy of the gearbox fault diagnosis results. This achieves effective and accurate fault detection of the gearbox even when the effectively labeled fault sample data is limited.

[0096] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930 and a communication bus 940, wherein the processor 910, the communication interface 920 and the memory 930 communicate with each other through the communication bus 940. The processor 910 may call the logic instructions in the memory 930 to execute a gearbox fault diagnosis method based on a digital twin and a composite adversarial domain discriminant network, the method comprising: obtaining a target measured operating parameter of a target gearbox; inputting the target measured operating parameter into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameter; inputting the target feature vector into a classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; wherein the fault diagnosis model is obtained by iteratively training a composite adversarial domain discriminant network based on sample measured operating parameters, sample simulated operating parameters and fault labels corresponding to the sample simulated operating parameters of a sample gearbox; the sample simulated operating parameters are obtained by simulating a dynamic model of a digital twin dynamic simulation model of the sample gearbox.

[0097] In addition, the logic instructions in the above-mentioned memory 930 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0098] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the gearbox fault diagnosis method based on digital twins and composite adversarial domain discriminant networks provided by the above methods, the method including: obtaining target measured operating parameters of the target gearbox; inputting the target measured operating parameters into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameters; inputting the target feature vector into a classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; wherein the fault diagnosis model is obtained by iteratively training a composite adversarial domain discriminant network based on sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of a sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

[0099] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network provided by the above-mentioned methods, the method comprising: obtaining target measured operating parameters of the target gearbox; inputting the target measured operating parameters into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameters; inputting the target feature vector into a classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; wherein the fault diagnosis model is obtained by iteratively training a composite adversarial domain discriminant network based on sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of a sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

[0100] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0101] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A gearbox fault diagnosis method based on digital twin and composite adversarial domain discriminant network, characterized in that: include: Obtaining target measured operating parameters of a target gearbox; Inputting the target measured operating parameters into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameters; Inputting the target feature vector into the classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; Among them, the fault diagnosis model is obtained by iteratively training the composite adversarial domain discriminant network based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

2. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to claim 1 is characterized in that: The sample simulation operating parameters of the sample gearbox are obtained based on the following steps: Constructing a digital twin dynamic simulation model of the sample gearbox according to the nonlinear motion characteristics of the sample gearbox; Based on a lumped parameter algorithm and according to the nonlinear motion characteristics of the sample gearbox, a dynamic model of the digital twin dynamic simulation model is established; The dynamic model is simulated using a preset fault state and a preset simulation condition as input parameters to obtain sample simulation operation parameters of the sample gearbox; Among them, the preset simulation conditions are used to constrain multiple items of contact force configuration information, kinematic pair configuration information, material configuration information, driving force configuration information, load configuration information, and integrator configuration information in the digital twin dynamic simulation model.

3. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to claim 2 is characterized in that: The method of establishing a dynamic model of the digital twin dynamic simulation model based on the lumped parameter algorithm and the nonlinear motion characteristics of the sample gearbox includes: According to the nonlinear motion characteristics of the sample gearbox, the rotational freedom equation of the driving wheel and the translational freedom equations in different directions in the digital twin dynamic simulation model, as well as the rotational freedom equation of the driven wheel and the translational freedom equations in different directions in the digital twin dynamic simulation model are constructed; Based on the lumped parameter algorithm, the dynamic model is constructed by combining the rotational freedom equation of the driving wheel and the translational freedom equations in different directions, as well as the rotational freedom equation of the driven wheel and the translational freedom equations in different directions.

4. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to any one of claims 1 to 3, characterized in that: The composite adversarial domain discrimination network includes an initial feature extraction module, an initial classification module, an initial domain recognition module and an initial dynamic weighted joint distribution adaptive module; The fault diagnosis model is trained based on the following steps: Inputting sample measured operating parameters and sample simulated operating parameters of the sample gearbox into the initial feature extraction module to obtain a first sample feature vector of the sample simulated operating parameters and a second sample feature vector of the sample measured operating parameters; Inputting the first sample feature vector into the initial classification module to obtain fault diagnosis prediction information corresponding to the sample simulation operation parameter, and obtaining a first loss function according to the classification difference between the fault diagnosis prediction information corresponding to the sample simulation operation parameter and the fault label; Inputting the first sample feature vector and the second sample feature vector into the initial domain identification module, obtaining a first domain spatial feature corresponding to the sample simulation operation parameter and a second domain spatial feature corresponding to the sample measured operation parameter, and obtaining a second loss function according to a domain distribution difference between the first domain spatial feature and the second domain spatial feature; Inputting the first sample feature vector and the second sample feature vector into the initial dynamic weighted joint distribution adaptive module to obtain the spatial distribution difference between the sample measured operation parameters and the sample simulated operation parameters, and obtaining a third loss function according to the spatial distribution difference; Combining the first loss function, the second loss function and the third loss function, iteratively training the composite adversarial domain discriminant network; The fault diagnosis model is constructed according to the trained initial feature extraction module and the trained initial classification module in the trained composite adversarial domain discrimination network.

5. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to claim 4 is characterized in that: The step of inputting the first sample feature vector and the second sample feature vector into the initial dynamic weighted joint distribution adaptive module to obtain a spatial distribution difference between the sample measured operating parameters and the sample simulated operating parameters, and obtaining a third loss function according to the spatial distribution difference includes: The first sample feature vector and the second sample feature vector are input into the initial dynamic weighted joint distribution adaptive module, and the initial dynamic weighted joint distribution adaptive module calculates the local maximum mean difference between the first sample feature vector and the second sample feature vector, and the conditional distribution alignment difference between the first sample feature vector and the second sample feature vector, and obtains the third loss function according to the local maximum mean difference and the conditional distribution alignment difference.

6. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to claim 5 is characterized in that: The conditional distribution alignment difference is calculated based on the following formula: ； in, Align the differences for the conditional distributions; is the distribution difference measurement function; is the number of fault categories; The sample simulation run parameters belong to the fault category The probability value of The parameters for the sample simulation runs are given in Belongs to the fault category Under the condition of The conditional distribution of The measured operating parameters of the sample belong to the fault category The probability value of For the sample measured operating parameters belong to the fault category Under the conditions of The first sample feature vector of The conditional distribution of .

7. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to claim 5 is characterized in that: The obtaining the third loss function according to the local maximum mean difference and the conditional distribution alignment difference includes: Obtaining a dynamic adaptation factor according to an exponential coefficient corresponding to the local maximum mean difference and an exponential coefficient corresponding to the conditional distribution alignment difference; According to the dynamic adaptation factor, weighted addition is performed on the local maximum mean difference and the conditional distribution alignment difference to obtain the third loss function.

8. The gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network according to claim 4 is characterized in that: The initial feature extraction module is constructed based on a multi-layer spatial channel reconstruction convolution layer and a residual layer; The spatial channel reconstruction convolutional layer is constructed based on a spatial reconstruction unit and a channel reconstruction unit. The spatial reconstruction unit is used to obtain spatial refinement features, and the channel reconstruction unit is used to obtain channel refinement features.

9. A gearbox fault diagnosis device based on digital twin and composite adversarial domain discrimination network, characterized in that: include: A data acquisition unit, used to acquire target measured operating parameters of a target gearbox; A feature extraction unit, used for inputting the target measured operating parameter into a feature extraction module of a fault diagnosis model to obtain a target feature vector of the target measured operating parameter; A fault diagnosis unit, used for inputting the target feature vector into the classification module of the fault diagnosis model to obtain fault diagnosis prediction information of the target gearbox; Among them, the fault diagnosis model is obtained by iteratively training the composite adversarial domain discriminant network based on the sample measured operating parameters, sample simulated operating parameters, and fault labels corresponding to the sample simulated operating parameters of the sample gearbox; the sample simulated operating parameters are obtained by simulating the dynamic model of the digital twin dynamic simulation model of the sample gearbox.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the gearbox fault diagnosis method based on digital twin and composite adversarial domain discrimination network as described in any one of claims 1 to 8.

Images