Parallel asynchronous structure-based fluid system digital twin system establishment method, program, equipment and storage medium

Abstract

The invention relates to a fluid system digital twin system establishment method based on a parallel asynchronous structure, a program, equipment and a storage medium. The system takes a time sequence database as a core, and a data acquisition module, a state monitoring module, a fault diagnosis module, a fluid simulation module, a data assimilation module and a visualization module are connected in parallel to realize data sharing and asynchronous communication. The method comprises the following steps: acquiring and preprocessing sensor data; performing signal noise reduction and time-frequency feature extraction to calculate a state index; performing fluid system parameter prediction based on the indexes; marking the to-be-corrected component by comparing the difference between the predicted value and the measured value, and correcting the to-be-corrected component; and finally, result visualization is realized. Through module decoupling, the problems that a series structure is poor in stability and wastes resources are solved, the real-time performance, fault tolerance and expandability of the system are remarkably improved, and the system is suitable for state monitoring and fault diagnosis of fluid systems such as reactors.

Description

Technical Field

[0001] This invention relates to the field of reactor engineering technology, and in particular to a method, program, equipment and storage medium for establishing a digital twin system of a fluid system based on a parallel asynchronous structure. Background Technology

[0002] With the application and development of digital twin technology in various disciplines, and the maturity of full-range simulation for power equipment, relevant experts have proposed applying the concept of digital twins to power plants.

[0003] Traditional digital twins are essentially online real-time simulation systems of actual equipment. This traditional form of "digital twin" has limited engineering significance and contributes little to the intelligent operation and maintenance of reactors during actual operation. Therefore, researchers have proposed incorporating intelligent operation and maintenance modules into online simulations, using the simulation results to judge and verify intelligent operation and maintenance decisions.

[0004] However, as intelligent operation and maintenance modules are added to the online real-time simulation model in a sequential (serial) manner, the model needs to run through the following modules in a single operation process: data acquisition, data preprocessing, online data transmission, status monitoring and fault diagnosis, online simulation model, data assimilation, and visualization interface. If the previous module has not provided a result, the subsequent modules will have difficulty running, resulting in poor stability of the operating framework, waste of computing resources, and extremely high requirements for the interfaces between modules and the data transmission between modules, which cannot meet the real-time simulation requirements in digital twins.

[0005] Therefore, this invention proposes a "parallel asynchronous" structure, which uses a time-series database to store the preprocessed data of the actual device's sensor data and the calculation results of each module. At the same time, it transmits the required data to each module in real time according to the input required by each module. Each module performs calculations in a "parallel" manner, interconnected but not interfering with each other. Even if some modules fail, it still has a certain operating capability and practical function.

[0006] Based on this, this invention researches and develops a method, program, equipment, and storage medium for establishing a digital twin system for fluid systems based on a parallel asynchronous structure, which can improve the real-time performance and stability of the digital twin system. This invention uses a real-time database as the core hub to achieve parallel operation and data sharing among various functional modules. The system features module decoupling, asynchronous communication, and strong fault tolerance, significantly improving the system's real-time performance, stability, and scalability. It is particularly suitable for applications such as condition monitoring, fault diagnosis, and simulation verification of reactor fluid systems. Summary of the Invention

[0007] The purpose of this invention is to provide a method, program, device, and storage medium for establishing a digital twin system for a fluid system based on a parallel asynchronous structure. This system can achieve real-time status monitoring and fault diagnosis of the fluid system, verification of status monitoring and fault diagnosis results, digital navigation of fluid system equipment, refined simulation of faulty equipment, and real-time display of fluid system simulation results and equipment status.

[0008] This invention proposes a method, program, equipment, and storage medium for establishing a digital twin system for a fluid system based on a parallel asynchronous structure. The core technical solutions include the following:

[0009] A method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure includes the following steps:

[0010] Step 1: Obtain the raw measurement value dataset and control signal data; initialize the time series database; and preprocess the raw measurement value dataset to obtain the preprocessed measurement value dataset, and store it in the time series database.

[0011] Step 2: Denoise the preprocessed measurement dataset and extract time-frequency domain features to obtain time-frequency domain feature vectors.

[0012] Step 3: Calculate the state index based on the time-frequency domain feature vector.

[0013] Step 4: Based on the state indicators, combined with the control signal data and the preprocessed measurement dataset, predict the fluid system parameters to obtain a set of predicted values.

[0014] Step 5: Calculate the absolute difference between each predicted value in the predicted value set and the original measurement data corresponding to each predicted value, and determine whether the absolute difference exceeds the absolute difference threshold. If so, mark the component corresponding to the predicted value as the component to be corrected, and combine all the components to be corrected into a list of components to be corrected.

[0015] Step 6: Based on the original measurement dataset and control signal data, select the hybrid drive structure and correct the components in the component list to be corrected to obtain the corrected prediction value set.

[0016] Step 7: Visualize the corrected set of predicted values, the state index vector, and the original measurement dataset.

[0017] Furthermore, the preprocessing of the original measurement dataset described in step 1 specifically includes: cleaning the original measurement dataset, filling in gaps using interpolation or mean methods, and making the data continuous.

[0018] Furthermore, step 2 specifically includes:

[0019] Step 2.1: Divide the preprocessed measurement dataset into sliding windows. Based on frequency conversion mode decomposition, decompose the signal into multiple intrinsic mode functions. Solve the objective function by minimizing the objective function to obtain the noise reduction result for each window.

[0020] Step 2.2: Calculate the time-frequency feature vector for the noise reduction result of each window.

[0021] Further, the objective function described in step 2.1 is:

[0022]

[0023] in, For the first One eigenmode function For the first The center frequencies of the eigenmode functions For the delta function, This is the Hilbert transform operator.

[0024] Furthermore, the time-frequency feature vector in step 2.2 includes the average absolute value. Maximum value Minimum value Peak Mean square value Root amplitude ,variance Peak-to-peak value skewness , cliff Peak factor, impulse factor, waveform factor, margin factor and coefficient of variation ;

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040] in, For the length of time, The signal amplitude, For differential operators, For the number of data points, For discrete signal values, The mean of the signal. Standard deviation The mean, For valid values, For expectation operator, The standard deviation is denoted as .

[0041] Furthermore, step 4 specifically includes:

[0042] Establish a fluid network simulation model that includes components under normal operating conditions and components under fault conditions.

[0043] The fluid path is selected based on the status indicators. If the status is normal, the fluid flows into the normal component; otherwise, the fluid flows into the faulty component.

[0044] Data is read according to the time step and the predicted value is calculated.

[0045] Furthermore, the absolute difference described in step 5 The specific calculation methods include:

[0046]

[0047] in, For predicted values, These are measured values.

[0048] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described above.

[0049] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0050] A computer program product includes computer instructions that, when executed by a processor, implement the steps of the method described above.

[0051] The beneficial effects of this invention are as follows:

[0052] Compared with existing technologies, this invention can effectively achieve real-time status monitoring and fault diagnosis of fluid systems, verification of status monitoring and fault diagnosis results, digital navigation of fluid system equipment, refined simulation of faulty equipment, and real-time display of fluid system simulation results and equipment status. At the same time, all modules of this invention are connected in parallel, which can reduce the overall system running time and reduce the impact of disturbances in a single module on the overall system. It also reduces the coupling between system modules, facilitates independent upgrades and maintenance of individual modules, improves system resource utilization, and allows each module to asynchronously schedule computing resources according to its own computing needs, increasing the system's scalability. Adding new functional modules only requires access to the real-time database without modifying the existing architecture. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the specific structure of the present invention.

[0054] Figure 2 This is a flowchart of the fault diagnosis operation for condition monitoring according to the present invention.

[0055] Figure 3 This is a schematic diagram of the series hybrid model of fluid components in an embodiment of the present invention.

[0056] Figure 4 This is a schematic diagram of the parallel hybrid model of fluid components in an embodiment of the present invention.

[0057] Figure 5 This is a schematic diagram of the embedded hybrid model of the fluid component in an embodiment of the present invention.

[0058] Figure 6 This is a schematic diagram of the visual interface of an embodiment of the present invention. Detailed Implementation

[0059] The following is in conjunction with the appendix Figure 1 The present invention will be further described below.

[0060] The technical solution of the present invention for establishing a digital twin system of a fluid system based on a parallel asynchronous structure is as follows:

[0061] Step 1: Obtain the raw measurement value dataset and control signal data; initialize the time series database; and preprocess the raw measurement value dataset to obtain the preprocessed measurement value dataset, and store it in the time series database.

[0062] The preprocessing of the original measurement dataset specifically includes: cleaning the original measurement dataset, filling in gaps using interpolation or mean methods, and making the data continuous.

[0063] Step 2: Denoise the preprocessed measurement dataset and extract time-frequency domain features to obtain time-frequency domain feature vectors.

[0064] Step 2 specifically includes:

[0065] Step 2.1: Divide the preprocessed measurement dataset into sliding windows, decompose the signal into multiple intrinsic mode functions according to frequency conversion mode decomposition, and solve the objective function by minimizing the objective function to obtain the noise reduction result of each window;

[0066]

[0067] in, For the first One eigenmode function For the first The center frequencies of the eigenmode functions For the delta function, This is the Hilbert transform operator.

[0068] Step 2.2: Calculate the time-frequency feature vector for the noise reduction result of each window.

[0069] The time-frequency feature vector includes the mean absolute value. Maximum value Minimum value Peak Mean square value Root amplitude ,variance Peak-to-peak value skewness , cliff Peak factor, impulse factor, waveform factor, margin factor and coefficient of variation ;

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085] in, For the length of time, The signal amplitude, For differential operators, For the number of data points, For discrete signal values, The mean of the signal. Standard deviation The mean, For valid values, For expectation operator, The standard deviation is denoted as .

[0086] Step 3: Calculate the state index based on the time-frequency domain feature vector.

[0087] Step 4: Based on the state indicators, combined with the control signal data and the preprocessed measurement dataset, predict the fluid system parameters to obtain a set of predicted values.

[0088] Step 4 specifically includes:

[0089] Establish a fluid network simulation model that includes components under normal operating conditions and components under fault conditions.

[0090] The fluid path is selected based on the status indicators. If the status is normal, the fluid flows into the normal component; otherwise, the fluid flows into the faulty component.

[0091] Data is read according to the time step and the predicted value is calculated.

[0092] Step 5: Calculate the absolute difference between each predicted value in the predicted value set and the original measurement data corresponding to each predicted value, and determine whether the absolute difference exceeds the absolute difference threshold. If so, mark the component corresponding to the predicted value as the component to be corrected, and combine all the components to be corrected into a list of components to be corrected.

[0093] The absolute difference The specific calculation methods include:

[0094]

[0095] in, For predicted values, These are measured values.

[0096] Step 6: Based on the original measurement dataset and control signal data, select the hybrid drive structure and correct the components in the component list to be corrected to obtain the corrected prediction value set.

[0097] Step 7: Visualize the corrected set of predicted values, the state index vector, and the original measurement dataset.

[0098] Example

[0099] See appendix Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for establishing a digital parallel system based on a fluid system, including the following steps:

[0100] S1. Transmit fluid system sensor data and fluid system control signal data to the time series database.

[0101] Preferably, in step S1, the fluid system sensor signals and control signal data are transmitted via TCP / UDP communication protocol to a real-time database, such as InfluxDB, capable of processing rapidly changing, constantly updated data. Simultaneously, data preprocessing is performed during transmission, including filtering, outlier removal, and missing value imputation. Finally, the real-time database stores the processed data in both a real-time data storage module and a historical data storage module for easy access by other modules.

[0102] See appendix Figure 2 As shown in Figure S2, the fault diagnosis system for the condition monitoring of the fluid system selects the sensor signals to be read from the time series database and transmits the calculation results to the time series database.

[0103] Preferably, in step S2, the condition monitoring and fault diagnosis module reads data from the real-time database, including low-frequency data from temperature, pressure, and flow sensors of various components of the fluid system, as well as high-frequency data from acoustic and vibration sensors installed for specific components. The condition monitoring and fault diagnosis module uses deep learning algorithms and historical data from the sensors to train a black-box model. The trained model reads the required data in real time, outputs the calculation results at a frequency of once per second, and transmits the calculation results to the real-time database in real time via the UDP communication protocol. The operation flow of the condition monitoring and fault diagnosis module is as follows: Figure 2 As shown.

[0104] S3. The fluid network model established for the fluid system reads the initial values ​​of fluid components and the real-time control signals of the fluid system from the time-series database in real time, and calls the normal or fault model of the corresponding operating condition based on the calculation results of the read condition monitoring and fault diagnosis module. After real-time calculation, the calculation results are transmitted to the time-series database in real time.

[0105] Preferably, in step S3, a fluid network model of the fluid system is established in MWORKS.sysplorer using the Modelica language, and the initial simulation values ​​of each fluid component and the real-time control signals of components such as pumps and valves are transmitted to the corresponding fluid components using the UDP communication protocol. Simultaneously, the fluid network model includes fault modeling, creating parallel fluid components under normal and different fault conditions using shut-off valves from the official standard library, and changing the connected components by reading the calculation results from the condition monitoring and fault diagnosis modules. During the simulation, the platform's simulation speed adjustment function ensures real-time simulation operation, and the simulation results are output to the real-time database via the UDP communication protocol.

[0106] The schematic diagrams of fluid network systems incorporating data-driven models in series, parallel, and embedded structures are shown below. Figure 3 , Figure 4 , Figure 5 As shown.

[0107] S4. Transmit the calculation results of part of the mechanism model in the fluid network to the data assimilation module, and transmit the data assimilated results back to the fluid network model instead of the calculation results of the mechanism model, and transmit the corrected simulation results to the time series database in real time;

[0108] Preferably, in step S4, the data assimilation section targets complex fluid components or fluid components where simulation results differ significantly from experimental results. The data assimilation component trains a data-driven model using deep learning algorithms and couples the data-driven model with the fluid network using the PythonIO component in the MWORKS.sysplorer platform. The data assimilation component uses historical data from a real-time database as a training set, and for different mechanism-data hybrid driven model development approaches (serial, parallel, embedded), it uses different independent variables, targeting outlet flow or key characteristic parameters of the fluid component, to train the data-driven model. Whenever the fluid component has been running for a certain period (e.g., 1 day or 1 week), historical data is collected again to retrain the data-driven model, replacing the original data-driven model in the fluid network.

[0109] S5. Transmit the real-time experimental data, real-time running results of condition monitoring and fault diagnosis, and simulation results after data assimilation from the database to the visualization interface.

[0110] Preferably, in step S5, the visualization interface uses a schematic diagram of the fluid system as its homepage, displays sensor data and simulation data of the fluid system near each fluid component, and prominently displays whether the experimental rig has malfunctioned (condition monitoring results) and which component of the experimental rig has malfunctioned and what type of malfunction (fault diagnosis results) in a prominent position on the homepage. The visualization interface reads real-time data from the real-time database via the UDP communication protocol.

[0111] Visual interface display examples Figure 6 As shown.

[0112] Preferably, the deep learning algorithms used for condition monitoring and fault diagnosis are graph neural networks, convolutional neural networks, etc.

[0113] Furthermore, the data-driven model in the series, parallel, and embedded hybrid models is built using deep learning algorithms and all preprocessed historical data. Here, deep learning refers to algorithms with time-series prediction capabilities, such as Long Short-Term Memory Neural Networks (LSTM) and Temporal Neural Networks (RNN). Meanwhile, the embedded hybrid driven model can use parameter estimation algorithms, such as genetic algorithms, to estimate the correction coefficients of the characteristic parameters of fluid components. For example, for the total heat transfer coefficient of a heat exchanger: a correction coefficient for the total heat transfer coefficient is introduced into the mechanistic model. When not estimated by parameters, this correction coefficient is 1. The correction coefficient obtained after parameter estimation is multiplied by the total heat transfer coefficient in the mechanistic model to obtain the corrected total heat transfer coefficient.

[0114] Preferably, the normal model in the fluid system is a mechanistic model written in Modelica, while complex components can be reduced-order models obtained from CFD calculations. The fault model is a mechanistic model written in Modelica. If the simulation results of the fault model differ significantly from the experimental results, a reduced-order model obtained from CFD calculations is used. The normal and fault models of the same fluid component are connected in parallel. Components in different states are selected for connection to the fluid loop based on the output results of the condition monitoring and fault diagnosis modules.

[0115] Preferably, the fluid system in this embodiment uses water as the flowing medium.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure, characterized in that, Includes the following steps: Step 1: Obtain the raw measurement data set and control signal data; Initialize the time-series database; The original measurement dataset is preprocessed to obtain a preprocessed measurement dataset, which is then stored in a time series database. Step 2: Denoise the preprocessed measurement dataset and extract time-frequency domain features to obtain time-frequency domain feature vectors; Step 3: Calculate the state index based on the time-frequency domain feature vector; Step 4: Based on the state indicators, combined with the control signal data and the preprocessed measurement dataset, predict the fluid system parameters to obtain a set of predicted values; Step 5: Calculate the absolute difference between each predicted value in the predicted value set and the original measurement data corresponding to each predicted value, and determine whether the absolute difference exceeds the absolute difference threshold. If so, mark the component corresponding to the predicted value as the component to be corrected, and combine all the components to be corrected into a list of components to be corrected. Step 6: Based on the original measurement dataset and control signal data, select the hybrid drive structure and correct the components in the component list to be corrected to obtain the corrected prediction value set; Step 7: Visualize the corrected set of predicted values, the state index vector, and the original measurement dataset.

2. The method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure according to claim 1, characterized in that, Step 1, which involves preprocessing the original measurement dataset, specifically includes cleaning the original measurement dataset and filling in gaps using interpolation or mean methods to ensure data continuity.

3. The method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure according to claim 2, characterized in that, Step 2 specifically includes: Step 2.1: Divide the preprocessed measurement dataset into sliding windows, decompose the signal into multiple intrinsic mode functions according to frequency conversion mode decomposition, and solve the objective function by minimizing the objective function to obtain the noise reduction result of each window; Step 2.2: Calculate the time-frequency feature vector for the noise reduction result of each window.

4. The method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure according to claim 3, characterized in that, The objective function described in step 2.1 is: in, For the first One eigenmode function For the first The center frequencies of the eigenmode functions For the delta function, This is the Hilbert transform operator.

5. The method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure according to claim 4, characterized in that: The time-frequency feature vector mentioned in step 2.2 includes the mean absolute value. Maximum value Minimum value Peak Mean square value Root amplitude ,variance Peak-to-peak value skewness , cliff Peak factor, impulse factor, waveform factor, margin factor and coefficient of variation ; in, For the length of time, The signal amplitude, For differential operators, For the number of data points, For discrete signal values, The mean of the signal. Standard deviation The mean, For valid values, For expectation operator, The standard deviation is denoted as .

6. The method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure according to claim 5, characterized in that, Step 4 specifically includes: Establish a fluid network simulation model that includes components under normal operating conditions and components under fault conditions; The fluid path is selected based on the status indicators. If the status is normal, the fluid flows into the normal component; otherwise, the fluid flows into the faulty component. Data is read according to the time step and the predicted value is calculated.

7. The method for establishing a digital twin system for a fluid system based on a parallel asynchronous structure according to claim 6, characterized in that, The absolute difference described in step 5 The specific calculation methods include: in, For predicted values, These are measured values.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method of claim 7.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by the processor, they implement the steps of the method of claim 7.

Images