A Centrifugal Pump Fault Diagnosis Method, System and Device Based on Digital Twin

Abstract

The present application discloses a centrifugal pump fault diagnosis method, system and device based on digital twin. The method includes: obtaining historical operation data of the centrifugal pump to obtain original sample data; using a feature selection method based on multi-objective optimization to determine the characteristic parameters of the original sample data; establishing a digital twin model of the centrifugal pump according to the characteristic parameters to generate first new sample data similar to the original sample data; using a WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium; performing feature extraction on the mixed sample data based on a convolutional neural network to obtain an intelligent fault diagnosis model, and realizing the intelligent fault diagnosis of the centrifugal pump. The present invention generates a sufficient number of fault sample sets with quality approaching the Nash equilibrium based on digital twin technology, overcomes the problem of limited model generalization ability caused by scarce samples in the prior art, and provides a new solution for centrifugal pump fault diagnosis under few fault samples.

Description

Technical Field

[0001] The present invention relates to the technical field of computer integrated manufacturing, and particularly relates to a centrifugal pump fault diagnosis method, system, electronic device and storage device based on digital twin. Background Art

[0002] Centrifugal pumps are common key equipment in the machinery industry and are widely used in complex product systems such as large-scale nuclear power energy, water supply and drainage, and ship power. The efficient and reliable operation of centrifugal pumps is an important guarantee for the safety and stability of complex product systems. Therefore, researching on the fault diagnosis method of centrifugal pumps is of great significance for ensuring the safe and efficient operation of complex product systems and has broad engineering application prospects.

[0003] With the popularization of deep learning technology, diagnostic models based on deep learning have been applied in the fault diagnosis of centrifugal pumps. However, the existing centrifugal pump fault diagnosis methods have certain limitations because most of the existing diagnostic models are generally supervised learning and rely on sufficient labeled data for iterative training. If the distribution of fault samples is biased or the quantity is insufficient, it will lead to the lack of sufficient generalization and diagnostic performance of supervised learning, making it difficult to meet the diagnostic requirements for abnormal types during the operation of centrifugal pumps in the real production environment.

[0004] Therefore, it is necessary to propose a centrifugal pump fault diagnosis method, system and device based on digital twin, which can solve the problems of lack of fault data and heavy dependence on expert experience faced by centrifugal pumps during the fault diagnosis process, resulting in low performance of the diagnostic model. Summary of the Invention

[0005] The present invention provides a centrifugal pump fault diagnosis method, system and device based on digital twin to solve the technical problem that the existing centrifugal pump fault diagnosis method based on supervised learning model has insufficient generalization and diagnostic performance due to the lack of fault data, making it difficult to meet the diagnostic requirements for abnormal types of centrifugal pumps in the real production environment.

[0006] To solve the above problems, the present invention provides a centrifugal pump fault diagnosis method based on digital twin, including:

[0007] Obtain the historical operation data of the centrifugal pump to obtain the original sample data;

[0008] Adopt a feature selection method based on multi-objective optimization to determine the characteristic parameters of the original sample data;

[0009] Establish a digital twin model of the centrifugal pump according to the characteristic parameters to generate first new sample data similar to the original sample data;

[0010] Using a WGAN model with gradient penalty, generate second new sample data approaching the Nash equilibrium according to the original sample data and the first new sample data;

[0011] Based on a convolutional neural network, perform feature extraction on the mixed sample data composed of the original sample data, the first new sample data, and the second new sample data to obtain an intelligent fault diagnosis model;

[0012] Obtain the operating data of the centrifugal pump in real time and input it into the intelligent fault diagnosis model to perform real-time fault diagnosis on the centrifugal pump.

[0013] Further, adopt a feature selection method based on multi-objective optimization to determine the feature parameters of the original sample data, including:

[0014] Taking the minimization of the number of feature parameters and the maximization of the fault correlation accuracy as the multi-objective function, generate the initial solution set of the multi-objective function;

[0015] Calculate the multi-objective function values corresponding to each individual in the initial solution set, select candidate individuals according to the multi-objective function values, and perform crossover and mutation operations on the candidate individuals to generate new offspring individuals;

[0016] Iteratively perform selection, crossover, and mutation operations on the new offspring individuals until the preset number of iterations is reached to obtain the best solution set;

[0017] Evaluate the dominance relationship among the individuals in the best solution set, and extract the non-dominated solution set approaching the Pareto front according to the dominance relationship. The non-dominated solution set is the feature parameters of the original sample data.

[0018] Further, establish a digital twin model of the centrifugal pump according to the feature parameters to generate first new sample data similar to the original sample data, including:

[0019] Determine the static information according to the physical structure of the centrifugal pump, and select the dynamic information of the actual operating environment of the centrifugal pump from the feature parameters;

[0020] Based on the static information and dynamic information, establish a digital twin model of the centrifugal pump, and use a finite element analysis tool to solve the response data of the digital twin model of the centrifugal pump to obtain first new sample data similar to the original sample data.

[0021] Further, based on the static information and dynamic information, establish a digital twin model of the centrifugal pump, and use a finite element analysis tool to solve the response data of the digital twin model of the centrifugal pump, including:

[0022] Create a 3D model of a centrifugal pump in a preset modeling software and import it into a finite element analysis tool to obtain a geometric model, and define attributes for each part of the geometric model;

[0023] Use the finite element analysis tool to divide the geometric model into multiple basic units;

[0024] According to the actual operating conditions of the centrifugal pump, apply boundary conditions and loads to the geometric model with defined attributes and divided to ensure compliance with the actual operating conditions of the centrifugal pump;

[0025] Solve the response data of the model according to the boundary conditions and loads to obtain first new sample data.

[0026] Furthermore, use a WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium according to the original sample data and the first new sample data, including:

[0027] Input random noise into the generator network of the WGAN model with gradient penalty to obtain generated data;

[0028] Randomly sample the original sample data and the first new sample data to obtain real data;

[0029] Input the generated data and the real data into the discriminator network of the WGAN model to obtain the classification error of the discriminator network;

[0030] Update the parameters of the discriminator network according to the classification error, and optimize the generator network based on the updated discriminator network;

[0031] When the loss functions of the discriminator network and the generator network both converge, obtain the trained WGAN model, and generate second new sample data approaching the Nash equilibrium based on the trained WGAN model.

[0032] Furthermore, the loss function of the generator network is expressed as:

[0033]

[0034] Among them, D(G(z(t))) represents the judgment probability output by the discriminator network, and E pG(z) represents the expected value of the true data distribution pG(z).

[0035] Furthermore, the loss function of the discriminator network is expressed as:

[0036]

[0037] Among them, x represents the original sample, z represents the input random noise, and L gpand λ represent the gradient penalty term and its weight, respectively, and E pG(z) represents the expected value of the discriminative network on the generated data G(z).

[0038] The present invention also provides a centrifugal pump fault diagnosis system based on digital twin, including:

[0039] A data acquisition module, configured to acquire historical operation data of the centrifugal pump to obtain original sample data;

[0040] A feature extraction module, configured to determine the characteristic parameters of the original sample data by using a feature selection method based on multi-objective optimization;

[0041] A first generation module, configured to establish a finite element model according to the characteristic parameters and generate first new sample data similar to the original sample data;

[0042] A second generation module, configured to use a WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium according to the original sample data and the new sample data;

[0043] A diagnostic model establishment module, configured to perform feature extraction on the mixed sample data composed of the original sample data, the first new sample data, and the second new sample data based on a convolutional neural network to obtain an intelligent fault diagnosis model;

[0044] An analysis module, configured to acquire the operation data of the centrifugal pump in real time and input it into the intelligent fault diagnosis model to perform intelligent fault diagnosis on the centrifugal pump.

[0045] The present invention also provides an electronic device, including a processor and a memory, where a computer program is stored on the memory, and when the computer program is executed by the processor, the method for diagnosing faults of a centrifugal pump based on digital twin according to any one of the above technical solutions is implemented.

[0046] The present invention also provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method for diagnosing faults of a centrifugal pump based on digital twin according to any one of the above technical solutions is implemented.

[0047] Compared with the prior art, the beneficial effects of the present invention include: The present invention fully considers the actual situation that most centrifugal pumps are in a normal state in actual engineering applications and lack equipment failure sample data. By using digital twin technology to generate new samples similar to the original sample data, it enhances the diversity of the training data set in the case of scarce real data, which helps to improve the generalization ability of the model; Through the WGAN model with gradient penalty, it better processes the sample generation process, ensures higher quality of the generated samples, reduces the impact of unbalanced failure samples on the failure diagnosis results, and alleviates the impact of few failure samples on the diagnosis efficiency. The present invention expands the number of failure samples of the centrifugal pump, reduces the impact of unbalanced failure samples on failure diagnosis, improves the accuracy and efficiency of centrifugal pump failure diagnosis, and has extremely strong engineering practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 FIG. is a schematic flow chart of a centrifugal pump failure diagnosis method based on digital twin provided by the present invention;

[0049] Figure 2 FIG. is a schematic diagram of the establishment process of the intelligent failure diagnosis model provided by the present invention;

[0050] Figure 3 FIG. is a schematic structural diagram of an embodiment of a centrifugal pump failure diagnosis system based on digital twin provided by the present invention;

[0051] Figure 4 FIG. is a schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The following will specifically describe the preferred embodiments of the present invention with reference to the accompanying drawings. The accompanying drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, but are not used to limit the scope of the present invention.

[0053] This embodiment provides a centrifugal pump failure diagnosis method, system, electronic device, and computer-readable storage device based on digital twin, which mainly includes steps such as collecting operation data, selecting characteristic parameters, expanding failure samples, and constructing a failure diagnosis model.

[0054] As Figure 1 shown, Figure 1 is a schematic flow chart of the centrifugal pump failure diagnosis method based on digital twin, and the method includes:

[0055] Step S101: Obtain the historical operation data of the centrifugal pump to obtain the original sample data;

[0056] Step S102: Use a feature selection method based on multi-objective optimization to determine the characteristic parameters of the original sample data;

[0057] Step S103: Establish a digital twin model of the centrifugal pump according to the characteristic parameters, and generate first new sample data similar to the original sample data;

[0058] Step S104: Use the WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium according to the original sample data and the first new sample data;

[0059] Step S105: Based on a convolutional neural network, perform feature extraction on the mixed sample data composed of the original sample data, the first new sample data, and the second new sample data to obtain an intelligent fault diagnosis model;

[0060] Step S106: Real-time obtain the operation data of the centrifugal pump and input it into the intelligent fault diagnosis model to perform real-time fault diagnosis on the centrifugal pump.

[0061] For the centrifugal pump fault diagnosis method based on digital twin provided in this embodiment, first, obtain the operation data of the centrifugal pump, such as temperature, pressure, vibration, etc.; second, adopt a feature selection method based on multi-objective optimization (considering accuracy and the number of characteristic parameters) to identify the most relevant characteristic parameters and reduce the impact of redundant data on fault diagnosis; third, in order to generate new samples similar to the original samples, use finite element analysis software to establish a finite element model. At the same time, in order to overcome problems such as gradient disappearance or mode collapse existing in traditional GAN methods, a generative adversarial network with Wasserstein gradient penalty (WGAN with GP) is proposed to effectively extract the features of the original samples and generate high-quality samples with high rationality and diversity; finally, use a convolutional neural network to perform feature extraction on the mixed samples to realize the fault diagnosis of the centrifugal pump. The method of this embodiment generates a sufficient number of fault sample sets with quality approaching the Nash equilibrium based on digital twin technology, overcomes the problem of limited model generalization ability caused by scarce samples in traditional methods, and provides a new solution idea for the centrifugal pump fault diagnosis under few fault samples.

[0062] As a preferred embodiment, in step S101, the historical operation data of the centrifugal pump is obtained as follows: on the basis of the original plant SCADA system (supervisory control and data acquisition system), vibration sensors are installed on both sides of the centrifugal pump unit, and the OPC UA technology (open platform communication unified architecture) is used to collect the operation data such as temperature, pressure, and vibration of the centrifugal pump.

[0063] As a specific embodiment, the specific method in the data acquisition process is as follows: First, install vibration displacement sensors (model: LZDSL1-930) on the rolling bearing seats on both sides of the centrifugal pump unit; then, use OPC UA technology to collect the vibration signals in the vertical and horizontal directions of the equipment and extract the pressure signals in the SCADA system; finally, collect and store once every 1 second as a sample, and each sample records 10 measured vibration data, thereby obtaining the original sample data reflecting the actual operation of the centrifugal pump.

[0064] As a preferred embodiment, in step S102, a feature selection method based on multi-objective optimization is used to determine the characteristic parameters of the original sample data, including:

[0065] Taking the minimization of the number of characteristic parameters and the maximization of the fault correlation accuracy as the multi-objective function, generate the initial solution set of the multi-objective function;

[0066] Calculate the multi-objective function values corresponding to each individual in the initial solution set, select candidate individuals according to the multi-objective function values, and perform crossover and mutation operations on the candidate individuals to generate new offspring individuals;

[0067] Iteratively perform selection, crossover, and mutation operations on the new offspring individuals until the preset number of iterations is reached to obtain the best solution set;

[0068] Evaluate the dominance relationship between individuals in the best solution set, and extract the non-dominated solution set tending to the Pareto front according to the dominance relationship. The non-dominated solution set is the characteristic parameter of the original sample data.

[0069] As a specific embodiment, first, use the initialization algorithm to obtain the initial solution set Pop of the objective function. Suppose there are w objective functions, and FS' w represents the parameter combination corresponding to the w-th objective Next, optimize the initial population and perform function evaluation. represents the function value corresponding to the w-th individual; then, use the selection operator to select the individuals participating in the evolution in the next generation; the selection operator here can be roulette wheel selection, tournament selection, or ranking selection, etc. Repeatedly execute the above steps until the number of iterations is reached, and output the best solution set of the problem; finally, through continuous optimization, the problem solution set continuously approaches the ideal Pareto front surface, and the non-dominated solution set (Non-dominated solutions, NDSs) of the problem is obtained, that is: the characteristic parameters of the original sample data.

[0070] Use the feature selection method based on multi-objective optimization to identify the most relevant characteristic parameters and reduce the impact of redundant data on fault diagnosis.

[0071] After determining the characteristic parameters affecting the faults of the centrifugal pump, a digital twin model of the centrifugal pump is established to simulate the operation of the centrifugal pump, and new sample data is generated to expand the fault samples. The Wasserstein generative adversarial network with gradient penalty algorithm is used to further extract the sample features and generate high-quality samples with high rationality and diversity.

[0072] As a preferred embodiment, in step S103, according to the characteristic parameters, a digital twin model of the centrifugal pump is established to generate first new sample data similar to the original sample data, including:

[0073] Determine the static information according to the physical structure of the centrifugal pump, and select the dynamic information of the actual operating environment of the centrifugal pump from the characteristic parameters;

[0074] Based on the static information and the dynamic information, a digital twin model of the centrifugal pump is established, and the response data of the digital twin model of the centrifugal pump is solved by using a finite element analysis tool to obtain first new sample data similar to the original sample data.

[0075] Furthermore, based on the static information and the dynamic information, a digital twin model of the centrifugal pump is established, and the response data of the digital twin model of the centrifugal pump is solved by using a finite element analysis tool, including:

[0076] Create a three-dimensional model of the centrifugal pump in a preset modeling software and import it into the finite element analysis tool to obtain a geometric model, and define the attributes for each part of the geometric model;

[0077] Use the finite element analysis tool to divide the geometric model into multiple basic units;

[0078] According to the actual operating conditions of the centrifugal pump, apply boundary conditions and loads to the geometric model with defined attributes and divided to ensure that it is consistent with the actual operating conditions of the centrifugal pump.

[0079] As a specific embodiment, the preset modeling software is Solidworks, and the finite element analysis tool is Comsol. The specific process of solving the response data of the digital twin model of the centrifugal pump by using the finite element analysis tool is as follows:

[0080] The first step is to create a geometric model: Import the three-dimensional model created in Solidworks into Comsol (finite element software), and define the material properties for each part of the geometric model, such as elastic modulus, Poisson's ratio, density, thermal conductivity, etc.

[0081] Step 2, Mesh Generation: Automatically divide the geometric model into a series of small elements (such as triangles, quadrilaterals, tetrahedrons, or hexahedrons) using Comsol software, and manually adjust the size, shape, and distribution of the mesh to ensure the accuracy and computational efficiency of the key areas.

[0082] Step 3, Boundary Conditions and Loads: According to the actual working conditions (the operating data collected in step S101), apply appropriate boundary conditions and loads to the model to ensure they conform to the actual situation and simulate the state of the analysis object as accurately as possible.

[0083] Step 4, Generate New Samples: Based on the defined model, material properties, mesh generation, boundary conditions, and loads, etc., solve the response of the model (such as displacement, stress, temperature, etc.) to generate new samples similar to the original samples, and obtain the first new sample data.

[0084] As a preferred embodiment, in step S104, in order to overcome problems such as gradient disappearance or mode collapse in traditional GAN methods, this embodiment uses a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN with GP) to generate a sufficient number of fault sample sets with quality approaching the Nash equilibrium. The specific method is as follows:

[0085] Input the random noise into the generator network of the WGAN model with gradient penalty to obtain the generated data;

[0086] Randomly sample the original sample data and the first new sample data to obtain the real data;

[0087] Input the generated data and the real data into the discriminator network of the WGAN model to obtain the classification error of the discriminator network;

[0088] Update the parameters of the discriminator network according to the classification error, and optimize the generator network based on the updated discriminator network;

[0089] When the loss functions of both the discriminator network and the generator network converge, obtain the trained WGAN model, and generate the second new sample data approaching the Nash equilibrium based on the trained WGAN model.

[0090] It should be noted here that WGAN (Wasserstein Generative Adversarial Network) is a variant of the Generative Adversarial Network (GAN). WGAN uses the Wasserstein distance (or Earth Mover's Distance) to measure the difference between the generated sample distribution and the real sample distribution. This distance has better mathematical properties and can provide a smoother loss function, which helps to improve the training process of the model.

[0091] As a specific embodiment, the sample generation process of the Generative Adversarial Network with Wasserstein Gradient Penalty (WGAN with GP) is as follows:

[0092] The first step: The generator generates "fake" samples: Randomly sample from the noise data distribution and input it into the generator G to obtain a set of fake data, denoted as G(z).

[0093] Among them, the loss function during the training process of the generator G is shown in Equation (1). Compared with the traditional GAN method, the generator introducing the constrained Wasserstein distance can obtain more similar samples.

[0094]

[0095] In Equation (1), D(G(z)) is the judgment probability output by the discriminator D.

[0096] Loss function The optimization objective is to continuously decrease, that is, to make G(z) deceive the discriminator D as much as possible, so that the discriminator thinks this is a real sample.

[0097] To solve some problems in WGAN (such as weight clipping), compared with the traditional GAN method, we introduce gradient penalty in WGAN to help ensure the Lipschitz (used to describe the smoothness of a function within its domain) continuity of the discriminator and further improve the stability of the system.

[0098] The loss function of the discriminator network is expressed as:

[0099]

[0100] In Equation (2), x represents the original sample, z represents the input random noise, L gp and λ respectively represent the gradient penalty term and its weight, E pG(z) represents the expected value of the discriminator network on the generated data G(z) and represents the ability of the discriminator to recognize "fake" samples. The optimization objective of the discriminator network is: maximize the score for real data and minimize the score for generated data. At the same time, by penalizing the gradient of the discriminator, ensure the smoothness of its change in the input space, thereby improving the stability of the model.

[0101] The second step: Discriminate between "true" and "false" samples: Randomly sample from the real data distribution as the real data. Use the previously generated data as the input of the discriminator network. Therefore, the input of the discriminator model is two types of data, "true" data and generated "false" data. The output value of the discriminator network is the probability that the input belongs to the real data, "true" is recorded as 1, and "false" is 0.

[0102] Step 3: Update the parameters of the discriminant network: Update the parameters of the discriminant network according to the classification error to improve its classification accuracy.

[0103] Step 4: Generate realistic false data: The generation network updates its own parameters according to the feedback information of the discriminant network to generate more realistic false data.

[0104] As a preferred embodiment, in step S105, a fault intelligent diagnosis model is constructed by using a Convolutional Neural Network (CNN) to extract features from the mixed sample data composed of the original sample data, the first new sample data, and the second new sample data, including extracting local features by using a convolutional kernel; reducing the feature size through a pooling layer to retain important information while reducing the computational complexity; introducing non-linearity through a ReLU activation function, etc., and finally obtaining an intelligent fault diagnosis model for realizing automatic fault diagnosis of centrifugal pumps.

[0105] As Figure 2 shown, Figure 2 The processing procedures of the above steps are shown in detail. The establishment process of the above intelligent fault diagnosis model includes four steps: collecting operation data, selecting characteristic parameters, expanding fault samples, and constructing a fault diagnosis model. By introducing digital twin technology, the number of fault samples of centrifugal pumps is expanded, the influence of unbalanced fault samples on fault diagnosis is reduced, and the reliability of centrifugal pump units is improved.

[0106] This embodiment also provides a centrifugal pump fault diagnosis system based on digital twin. As Figure 3 shown, the centrifugal pump fault diagnosis system 300 based on digital twin includes:

[0107] A data acquisition module 301, configured to acquire historical operation data of a centrifugal pump to obtain original sample data;

[0108] A feature extraction module 302, configured to determine characteristic parameters of the original sample data by using a feature selection method based on multi-objective optimization;

[0109] A first generation module 303, configured to establish a finite element model according to the characteristic parameters and generate first new sample data similar to the original sample data;

[0110] A second generation module 304, configured to use a WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium according to the original sample data and the first new sample data;

[0111] A diagnostic model establishment module 305, configured to extract features from the mixed sample data composed of the original sample data, the first new sample data, and the second new sample data based on a convolutional neural network, so as to obtain an intelligent fault diagnosis model;

[0112] An analysis module 306, configured to obtain the operation data of the centrifugal pump in real time and input it into the intelligent fault diagnosis model to perform intelligent fault diagnosis on the centrifugal pump.

[0113] As Figure 4 shown, for the above-mentioned centrifugal pump fault diagnosis method based on digital twin, the present invention also correspondingly provides an electronic device 400, which may be a computing device such as a mobile terminal, a desktop computer, a notebook, a palm computer, and a server. The electronic device includes a processor 401, a memory 402, and a display 403.

[0114] The memory 402 may be an internal storage unit of the computer device in some embodiments, such as the hard disk or memory of the computer device. The memory 402 may also be an external storage device of the computer device in other embodiments, such as a plug-in hard disk equipped on the computer device, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 402 may also include both the internal storage unit and the external storage device of the computer device. The memory 402 is used to store application software installed on the computer device and various types of data, such as program codes installed on the computer device. The memory 402 may also be used to temporarily store data that has been output or will be output. In one embodiment, a centrifugal pump fault diagnosis method program 404 based on digital twin is stored on the memory 402, and the centrifugal pump fault diagnosis method program 404 based on digital twin can be executed by the processor 401, so as to implement the centrifugal pump fault diagnosis method based on digital twin in various embodiments of the present invention.

[0115] The processor 401 may be a central processing unit (CPU), a microprocessor, or other data processing chips in some embodiments, and is used to run the program codes stored in the memory 402 or process data, such as executing a centrifugal pump fault diagnosis method program based on digital twin, etc.

[0116] The display 403 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch display, etc. in some embodiments. The display 403 is used to display information of the computer device and to display a visual user interface. Components 401-403 of the computer device communicate with each other via a system bus.

[0117] An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it implements the centrifugal pump fault diagnosis method based on digital twin as described in any of the above technical solutions.

[0118] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A centrifugal pump fault diagnosis method based on digital twin, characterized in that, including; Obtain the historical operation data of the centrifugal pump to get the original sample data; Adopt a feature selection method based on multi-objective optimization to determine the characteristic parameters of the original sample data; Establish a digital twin model of the centrifugal pump according to the characteristic parameters, and generate first new sample data similar to the original sample data, including: determining static information according to the physical structure of the centrifugal pump, and selecting dynamic information of the actual operating environment of the centrifugal pump from the characteristic parameters; establishing a digital twin model of the centrifugal pump based on the static information and dynamic information, and using a finite element analysis tool to solve the response data of the digital twin model of the centrifugal pump, including: creating a three-dimensional model of the centrifugal pump in a preset modeling software and importing it into the finite element analysis tool to obtain a geometric model, and defining attributes for each part of the geometric model; using the finite element analysis tool to divide the geometric model into multiple basic units; according to the actual operating conditions of the centrifugal pump, applying boundary conditions and loads to the geometric model with defined attributes and divided to ensure consistency with the actual operating conditions of the centrifugal pump; solving the response data of the model according to the boundary conditions and loads to obtain first new sample data similar to the original sample data; Use a WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium according to the original sample data and the first new sample data, including: inputting random noise into the generation network of the WGAN model with gradient penalty to obtain generated data; randomly sampling the original sample data and the first new sample data to obtain real data; inputting the generated data and the real data into the discriminant network of the WGAN model to obtain the classification error of the discriminant network; updating the parameters of the discriminant network according to the classification error, and optimizing the generation network based on the updated discriminant network; when the loss functions of the discriminant network and the generation network both converge, obtain the trained WGAN model, and generate second new sample data approaching the Nash equilibrium based on the trained WGAN model; Extract features from the mixed sample data composed of the original sample data, the first new sample data and the second new sample data based on a convolutional neural network to obtain an intelligent fault diagnosis model; Obtain the operation data of the centrifugal pump in real time and input it into the intelligent fault diagnosis model to perform real-time fault diagnosis on the centrifugal pump.

2. The method for diagnosing centrifugal pump faults based on digital twin according to claim 1, wherein Adopt a feature selection method based on multi-objective optimization to determine the characteristic parameters of the original sample data, including: Taking the minimization of the number of characteristic parameters and the maximization of the fault correlation accuracy as multi-objective functions, generate an initial solution set of the multi-objective functions; Calculate the multi-objective function values corresponding to each individual in the initial solution set, select candidate individuals according to the multi-objective function values, and perform crossover and mutation operations on the candidate individuals to generate new offspring individuals; Iteratively perform selection, crossover and mutation operations on the new offspring individuals until a preset number of iterations is reached to obtain the optimal solution set; Evaluate the dominance relationship among individuals in the optimal solution set, and extract the non-dominated solution set approaching the Pareto front according to the dominance relationship. The non-dominated solution set is the characteristic parameter of the original sample data.

3. The method for diagnosing centrifugal pump faults based on digital twin according to claim 1, characterized in that The loss function of the generation network is expressed as: Among them, D(G(z(t))) represents the judgment probability output by the discriminant network, and E pG(z) represents the expected value of the true data distribution pG(z).

4. The method for diagnosing centrifugal pump faults based on digital twin according to claim 1, characterized in that, The loss function of the discriminant network is expressed as: Among them, x represents the original sample, z represents the input random noise, L gp and λ represent the gradient penalty term and its weight respectively, E pG(z) represents the expected value of the discriminative network on the generated data G(z).

5. A centrifugal pump fault diagnosis system based on digital twin, characterized in that, It includes: A data acquisition module, configured to acquire historical operation data of a centrifugal pump to obtain original sample data; A feature extraction module, configured to determine the characteristic parameters of the original sample data by using a feature selection method based on multi-objective optimization; A first generation module, configured to establish a digital twin model of the centrifugal pump according to the characteristic parameters and generate first new sample data similar to the original sample data, including: determining static information according to the physical structure of the centrifugal pump, and selecting dynamic information of the actual operation environment of the centrifugal pump from the characteristic parameters; establishing a digital twin model of the centrifugal pump based on the static information and dynamic information, and using a finite element analysis tool to solve the response data of the digital twin model of the centrifugal pump, including: creating a three-dimensional model of the centrifugal pump in a preset modeling software and importing it into the finite element analysis tool to obtain a geometric model, defining attributes for each part of the geometric model; using the finite element analysis tool to divide the geometric model into multiple basic units; applying boundary conditions and loads to the geometric model with defined attributes and divided according to the actual operation conditions of the centrifugal pump to ensure compliance with the actual operation conditions of the centrifugal pump; solving the response data of the model according to the boundary conditions and loads to obtain first new sample data similar to the original sample data; A second generation module, configured to use a WGAN model with gradient penalty to generate second new sample data approaching the Nash equilibrium according to the original sample data and the first new sample data, including: inputting random noise into the generation network of the WGAN model with gradient penalty to obtain generated data; randomly sampling the original sample data and the first new sample data to obtain real data; inputting the generated data and the real data into the discriminant network of the WGAN model to obtain the classification error of the discriminant network; updating the parameters of the discriminant network according to the classification error, and optimizing the generation network based on the updated discriminant network; when the loss functions of the discriminant network and the generation network both converge, obtaining a trained WGAN model, and generating second new sample data approaching the Nash equilibrium based on the trained WGAN model; A diagnostic model establishment module, configured to extract features from the mixed sample data composed of the original sample data, the first new sample data and the second new sample data based on a convolutional neural network to obtain an intelligent fault diagnosis model; An analysis module, configured to acquire the operation data of the centrifugal pump in real time and input it into the intelligent fault diagnosis model to perform intelligent fault diagnosis on the centrifugal pump.

6. An electronic device, characterized in that, It includes a processor and a memory, and a computer program is stored on the memory. When the computer program is executed by the processor, the digital twin-based centrifugal pump fault diagnosis method according to any one of claims 1-4 is implemented.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the digital-twin-based centrifugal pump fault diagnosis method according to any one of claims 1-4.

Images