A digital twin evaluation system for hydrodynamic performance of a liquid storage tank and a method thereof

Abstract

The application discloses a kind of liquid storage tank hydrodynamic performance digital twinborn evaluation systems, including based on real object data acquisition module, similar physical model test module and liquid storage tank numerical simulation module composition digital twinborn model module, hydrodynamic performance calculation module, data acquisition and interaction module and performance evaluation and diagnostic prediction module, the digital twinborn model module, hydrodynamic performance calculation module, data acquisition and interaction module and performance evaluation and diagnostic prediction module are sequentially linked, and data acquisition and interaction module are also linked with twinborn model module.The application also discloses a kind of liquid storage tank hydrodynamic performance digital twinborn evaluation method.The application effectively improves the accuracy of using digital twinborn method to calculate the hydrodynamic performance of storage tank, and enriches the function of hydrodynamic performance evaluation system.

Description

Technical Field

[0001] This invention relates to the technical field of a solution for hydrodynamic performance analysis of storage tanks, specifically a digital twin evaluation system and method for hydrodynamic performance of liquid storage tanks. Background Technology

[0002] Currently, the solutions for analyzing the hydrodynamic performance of liquid tanks or storage tanks generally include the following:

[0003] 1. Current methods for analyzing the hydrodynamic performance of liquid tanks or storage vessels often rely on theoretical approaches or empirical formulas, remaining largely theoretical and difficult to apply in practical engineering. While these methods are typically fast and reasonable, they often fail to encompass all possible variations and issues, especially as storage tanks may age, corrode, and undergo changes in structural design parameters over time. Predictive analysis based on raw data cannot accurately predict actual results. Therefore, existing methods for analyzing the hydrodynamic performance of storage tanks lack the ability to learn from actual changes and provide feedback to correct the results.

[0004] 2. While mathematical modeling, physical model testing, and specialized software calculations are used for processing, these models may no longer be applicable to containers whose parameters change. In such cases, numerical models or new models built using software are typically employed for evaluation. However, existing analytical methods, including numerical and experimental methods, are based on scaled-down models of the original tanks. As simulation time increases and actual conditions change, the simulation results gradually deviate from reality, failing to accurately predict risks and performance. Furthermore, these methods usually require independent design, modeling, and analysis for each scenario, consuming significant time, effort, and money, thus hindering rapid analysis of engineering problems.

[0005] 3. Processing through digital twin models: For some existing methods of predicting industrial production and physical performance and status based on digital twins, the measured data of the original physical model is analyzed, and the relevant results are classified and used to train machine learning artificial intelligence models. However, machine learning model training is time-consuming, and the output results are mostly single model performance feature values. Although it can guide the design, operation and performance evaluation of the structure to a certain extent, due to the lack of relevant mathematical analysis and mechanism changes, the results cannot systematically analyze the hydrodynamic performance of the storage tank, and it is difficult to infer the cause of the above results from the results. Therefore, it is not conducive to the optimization design and transformation of the structure.

[0006] In summary, the current methods for analyzing the hydrodynamic performance of liquid tanks or storage tanks, whether using theoretical methods or empirical formulas, mathematical model simulations, physical model experiments, or professional software calculations or digital twin models, all suffer from three problems that need to be addressed: the inability to learn autonomously based on changes in actual conditions, the inability to feed back to the original method for result correction, the hindrance to rapid analysis of engineering problems, and the hindrance to structural optimization design and modification. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the prior art by proposing a digital twin evaluation system and method for the hydrodynamic performance of liquid storage tanks. This system solves the problems of existing technologies, such as the inability to learn autonomously based on changes in actual conditions, the inability to feed back to the original method for result correction, the difficulty in rapid analysis of engineering problems, and the difficulty in optimizing and modifying the structure.

[0008] To achieve the above objectives, the present invention provides a digital twin evaluation system for the hydrodynamic performance of a liquid storage tank, comprising the following modules:

[0009] The digital twin model module consists of a physical data acquisition module, a similar physical model testing module, and a liquid storage tank numerical simulation module. The physical data acquisition module is used to acquire the changes in four parameters—pressure, displacement, liquid level change, and temperature—of the actual service-ready engineering liquid storage container under the influence of the external environment. The similar physical model testing module is used to perform similarity processing based on the geometric dimensions of the physical storage tank and using the Froude gravity similarity criterion. The liquid storage tank numerical simulation module uses the open-source computational fluid dynamics software OpenFOAM for modeling, ensuring that all parameters and conditions in the modeling are consistent with those of the physical liquid storage tank.

[0010] The system employs a combination of numerical simulation based on OpenFOAM open-source computational fluid dynamics code, physical model experimental methods, and machine learning prediction methods for joint calculation and analysis. The calculation can be stopped at any time, and the pressure, flow velocity, and liquid level data in the tank at the stop time point are obtained. The above results are compared with the measured results. If data updates are needed, the measured results are used as the initial conditions for the next calculation of the numerical calculation model as input to continue the calculation of the hydrodynamic performance of the pressure, flow velocity, and liquid level data in the tank.

[0011] The four main parameters of pressure, displacement, liquid level change and temperature output from physical objects, numerical models, experimental models and machine learning models are collected and classified. The hydrodynamic performance calculation module integrates the output results of each model to organize and store the data. Then, the collected and classified data are interactively mapped at the same time. The data of different models at the same time are compared and corrected. After completing the above interaction process, the data is fed back and redundant and inaccurate data is cleaned up before the data collection and interaction module for feature extraction is completed.

[0012] The system plots and visualizes the performance curves of various parameters affecting the safety performance of the storage tank, further analyzes the variation law of the hydrodynamic performance of the storage tank under different environmental loads, obtains the vibration mode and stress concentration location of the storage tank, and then predicts the risks and accidents that may be caused by the performance of the storage tank under load changes. It also performs retrospective analysis on the failures that have occurred due to the passage of time and external loads, thereby clarifying the cause of the failure. The performance evaluation and diagnosis prediction module is composed of the digital twin model module, the hydrodynamic performance calculation module, the data acquisition and interaction module, and the performance evaluation and diagnosis prediction module, which are linked in sequence. The data acquisition and interaction module is also linked to the digital twin model module.

[0013] This invention also discloses a digital twin evaluation method for the hydrodynamic performance of a liquid storage tank, including the aforementioned digital twin evaluation system for the hydrodynamic performance of a liquid storage tank, which specifically includes the following steps:

[0014] S1. Twin model construction steps, wherein the twin model construction includes data acquisition steps for the physical liquid storage tank, similar physical model construction steps, and numerical modeling steps for the liquid storage tank;

[0015] S2. Hydrodynamic performance calculation steps, which include numerical calculation steps based on OpenFOAM open-source computational fluid dynamics code, physical model test analysis steps based on hydrodynamic experiments, and neural network model calculation steps based on machine learning.

[0016] S3. The step of using data acquisition results from different types of sensors as measured data to interact with hydrodynamic performance calculation results;

[0017] S4. Input the data collected in step S3 into the internal pre-built program. The relevant analysis algorithm performs post-processing and displays the relevant data in the form of charts in a window interface after statistical analysis. Time domain or frequency domain analysis is used as the analysis content. The data after interactive comparison of relevant parameters is analyzed to analyze the possible differences. Then, the obtained results are used for performance evaluation, risk prediction and fault diagnosis.

[0018] Preferably, in step S1, the data acquisition steps for the physical storage tank are as follows: by collecting the changes in four parameters of the engineering structure in actual service under the influence of the external environment, the sampling frequency and total sampling amount should have a sufficient time span. Generally, the sample data period should not be less than 6 months, the sampling frequency should not exceed 30 minutes, and the basic materials of the physical storage tank need to be collected once before the collection process.

[0019] Preferably, in step S1, the similar physical model construction step is to perform similarity processing based on the geometric dimensions of the physical storage tank. The main steps are as follows: using the Froude gravity similarity criterion, the relevant parameters in the model are scaled using the Froude number ratio. At the same time, the model material and the liquid properties inside the storage tank are similar as a premise during the modeling process to ensure that the parameters in the numerical model modeling are consistent with the physical liquid storage tank. Finally, OpenFOAM is used for modeling.

[0020] Preferably, in step S2, the hydrodynamic performance calculation and prediction is performed using a combination of methods, including numerical calculation using the OpenFOAM open-source computational fluid dynamics code, physical model experimental methods, and machine learning prediction methods.

[0021] As a preferred approach, the numerical calculation steps based on the OpenFOAM open-source computational fluid dynamics code are as follows: the initial values ​​of the model, the fluid computational domain mesh, boundary conditions, numerical algorithm, and environmental conditions are determined sequentially using the numerical simulation process; the calculation process can be stopped at any time, and then the pressure, flow velocity, and liquid level data in the tank at the stopped time point are obtained. The above results are compared with the measured results. If data updates are needed, the measured results are used as the initial conditions for the next calculation of the numerical calculation model as input quantities to continue the calculation.

[0022] Preferably, the numerical simulation process uses the finite volume method to decompose the computational domain and mesh it. Within each computational cell, the Reynolds-averaged Navier-Stokes equations are solved to obtain the pressure-velocity coupling relationship at each point within the tank. Initial values ​​are obtained from data collected on a physical storage tank, ensuring that the boundary conditions maintain the same properties as the physical tank. If the gradient of the impermeable flow along the wall is zero, the tank moves under external environmental loads, ensuring that the wall position remains consistent with the movement of the physical tank. The experimental platform for the numerical simulation process needs to meet the environmental factors faced by the tank. Model similarity uses the Frode criterion with gravity as the main factor to scale down the force, characteristic length, velocity, mass, and time parameters. Simultaneously, detection sensors should be set up at corresponding locations for model testing. Different loads are simulated on the experimental platform for different problems, and relevant hydrodynamic performance calculations and analyses are performed.

[0023] As a preferred option, the calculation steps of the neural network model based on machine learning in step S2 are as follows: First, the results obtained from the physical object, experimental model, and numerical model should be classified and organized, and the total sample size should be divided into two parts of 50% and 50%. One part is used as the sample data for training the neural network model, and the other part is used as the verification data for verifying the reliability and accuracy of the model. The neural network model is divided into two parts: feedforward and feedback. By repeatedly calculating the training sample data, the accuracy of the neural network model is improved, and finally, reliable feature quantity output is achieved.

[0024] As a preferred embodiment, the step S3, which involves using data acquisition results from different types of sensors as measured data to interact with hydrodynamic performance calculation results, is as follows:

[0025] S3-1. Collect the four main parameters of pressure, displacement, liquid level, and temperature from the output of physical objects, experimental models, experimental model data, and machine learning model data.

[0026] S3-2. Organize and store the data based on the differences in sampling frequencies of different models;

[0027] S3-3. The data interaction process is divided into four stages: data mapping, verification, assimilation, and feedback. First, the collected and categorized data are mapped simultaneously, and data from different models at the same time are compared. The sampling frequency, from highest to lowest, is: numerical model, experimental model, physical model, and machine learning model. The data accuracy weights, from highest to lowest, are: physical model, experimental model, numerical model, and machine learning model. The hydrodynamic performance calculation speeds, from highest to lowest, are: machine learning model, numerical model, experimental model, and physical model. In the verification process, physical data is used to correct experimental model data, and the corrected experimental model data is used to correct numerical model data. The corrected data model data is then used to train the machine learning model, which is then verified against the physical data. Finally, different models are validated. Data assimilation is performed, with a verification error requirement of ≤5%. After completing the above interaction process, data feedback is provided, and redundant and inaccurate data is cleaned up to complete feature extraction. Simultaneously, during the data acquisition process, measured data and numerical calculation results can be collected at the same time, and the two results are compared. There are errors in the pressure, velocity, and liquid level results. When the error is greater than 5% of the measured result, the error calculation formula is (measured data - calculated data) / measured data > 5%. Then, the input conditions for numerical calculation data are updated and replaced. At this time, numerical calculation is stopped, and the measured data is used to replace the numerical calculation result at the corresponding time as the initial value for the next calculation step. The above data interaction process is automatically judged by a pre-written program to achieve unattended automatic operation.

[0028] Preferably, the performance evaluation in step S4 mainly involves evaluating the various parameter data that affect the safety performance of the storage tank, and plotting and visualizing the performance curves to analyze the variation law of the hydrodynamic performance of the storage tank under different environmental loads.

[0029] Compared with existing technologies, the technical advantages of the digital twin evaluation system and method for hydrodynamic performance of liquid storage tanks obtained by this invention are as follows: This method fully leverages the advantages of traditional theoretical analysis / numerical simulation methods combined with the latest digital twin and machine learning methods to achieve accurate and rapid analysis of the hydrodynamic performance of liquid storage tanks in actual engineering projects, as well as risk prediction and failure process inversion, providing a full life-cycle assessment of tank performance. It replaces the initial parameters of the original numerical model with measured data from the actual storage tank as the input parameters of the numerical model, avoiding inaccuracies caused by changes in the structural parameters of the actual storage tank over time and improving the accuracy of the numerical model. Simultaneously, it replaces the process of training an artificial intelligence model using machine learning in the digital twin method with a numerical model that considers the actual physical processes, improving model prediction efficiency while providing more comprehensive and complete data on the fluid movement process within the storage tank. This enables hydrodynamic performance analysis, risk warning, and accident cause tracing. Attached Figure Description

[0030] Figure 1 This is a reference diagram showing the external structure of each storage tank in the detection objects of this invention;

[0031] Figure 2 This embodiment uses the placement of sensors within a vertical cylinder as a reference for detection. Figure 1 ;

[0032] Figure 3 This embodiment uses the placement of sensors within a vertical cylinder as a reference for detection. Figure 2 ;

[0033] Figure 4 This is a schematic diagram of the connection structure of a digital twin evaluation system for the hydrodynamic performance of a liquid storage tank in this embodiment;

[0034] Figure 5 This is a schematic diagram of the internal logic and data interaction of the digital twin model module for the storage tank in this embodiment;

[0035] Figure 6 This is a flowchart illustrating the specific steps of the OpenFOAM numerical model in the hydrodynamic performance calculation module of this embodiment.

[0036] In the attached diagram: Labels 1-12 are all pressure sensors; Labels 13-14 are liquid level sensors; Physical data acquisition module 1-1; Similar physical model construction module 1-2; Numerical modeling module for liquid storage tank 1-3; Digital twin model module 15; Hydrodynamic performance calculation module 16; Data acquisition and interaction module 17; Performance evaluation and diagnostic prediction module 18. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] Example:

[0039] like Figure 1 As shown, the object that this invention can detect is Figure 1 The external structure of each storage tank can be a vertical cylinder, a cube, a horizontal cylinder, a sphere, or other different types of liquid storage tanks. The vertical cylindrical storage tank model can be replaced by a liquid storage container of any shape, and the sensor installation position is determined accordingly based on the container characteristics. The numerical software solves the Navier-Stokes equations based on the finite volume method to calculate characteristic results such as pressure, liquid level, and flow velocity. Other numerical solution methods can also be used to solve the pressure-velocity coupling equations of fluid motion, but it needs to be able to stop at any time, replace the input parameters at a specific moment, and restart the calculation.

[0040] like Figures 2-6 As shown in this embodiment, a digital twin evaluation system for the hydrodynamic performance of a liquid storage tank is provided, wherein... Figure 2 The components labeled 1-11 are pressure sensors; the components labeled 13-14 are level sensors, including the following modules:

[0041] The digital twin model module 15 consists of a physical data acquisition module 1-1, a similar physical model testing module 1-2, and a liquid storage tank numerical simulation module 1-3. The physical data acquisition module 1-1 is used to acquire the changes in four parameters of the engineering liquid storage container in actual service under the influence of the external environment. The similar physical model testing module 1-2 is used to perform similarity processing based on the geometric dimensions of the physical storage tank and using the Froude gravity similarity criterion. The liquid storage tank numerical simulation module 1-3 uses the open-source computational fluid dynamics software OpenFOAM for modeling, and ensures that all parameters and conditions in the modeling are consistent with those of the physical liquid storage tank.

[0042] The calculation and analysis are jointly performed using numerical simulation calculation method based on OpenFOAM open source computational fluid dynamics code, physical model experimental method and machine learning prediction method. The calculation results can be stopped at any time, and the pressure, flow velocity and liquid level data in the tank at the stop time point are obtained. The above results are compared with the measured results. If the data needs to be updated, the measured results are used as the initial conditions for the next calculation of the numerical calculation model as input to continue the calculation of the hydrodynamic performance of the pressure, flow velocity and liquid level data in the tank.

[0043] The four main parameters of pressure, displacement, liquid level change and temperature output from physical objects, numerical models, experimental models and machine learning models are collected and classified. The data is organized and stored by the hydrodynamic performance calculation module 16, which integrates the output results of each model. Then, the collected and classified data are interactively mapped at the same time, and the data of different models at the same time are compared and corrected. After completing the above interaction process, the data is fed back, and redundant and inaccurate data is cleaned up before the feature extraction data collection and interaction module 17 is completed.

[0044] The performance curves of various parameters affecting the safety performance of the storage tank are plotted and visualized. The variation law of the hydrodynamic performance of the storage tank under different environmental loads is further analyzed to obtain the vibration mode and stress concentration location of the storage tank. Then, the risks and accidents that may be caused by the performance of the storage tank under load changes are predicted, and the failures that have occurred due to the passage of time and external loads are retrospectively analyzed to clarify the cause of failure. The performance evaluation and diagnosis prediction module 18 is composed of the digital twin model module 15, the hydrodynamic performance calculation module 16, the data acquisition and interaction module 17, and the performance evaluation and diagnosis prediction module 18, which are linked in sequence. The data acquisition and interaction module 17 is also linked to the digital twin model module 15.

[0045] This invention also discloses a digital twin evaluation method for the hydrodynamic performance of a liquid storage tank, including the aforementioned digital twin evaluation system for the hydrodynamic performance of a liquid storage tank, which specifically includes the following steps:

[0046] S1. Twin model construction steps, wherein the twin model construction includes data acquisition steps for the physical liquid storage tank, similar physical model construction steps, and numerical modeling steps for the liquid storage tank;

[0047] S2. Hydrodynamic performance calculation steps, which include numerical calculation steps based on OpenFOAM open-source computational fluid dynamics code, physical model test analysis steps based on hydrodynamic experiments, and neural network model calculation steps based on machine learning.

[0048] S3. The step of using data acquisition results from different types of sensors as measured data to interact with hydrodynamic performance calculation results;

[0049] S4. Input the data collected in step S3 into the internal pre-built program. The relevant analysis algorithm performs post-processing and displays the relevant data in the form of charts in a window interface after statistical analysis. Time domain or frequency domain analysis is used as the analysis content. The data after interactive comparison of relevant parameters is analyzed to analyze the possible differences. Then, the obtained results are used for performance evaluation, risk prediction and fault diagnosis.

[0050] In this embodiment, in step S1, the data acquisition steps for the physical storage tank are as follows: by collecting the changes in four parameters of the engineering structure in actual service under the influence of the external environment, the sampling frequency and total sampling amount should have a sufficient time span. Generally, the sample data period should be no less than 6 months, the sampling frequency should be no less than 30 minutes, and the basic materials of the physical storage tank need to be collected at one time during the collection process.

[0051] In this embodiment, in step S1, the similar physical model construction step is to perform similarity processing based on the geometric dimensions of the physical storage tank. The main steps are as follows: using the Froude gravity similarity criterion, the relevant parameters in the model are scaled using the Froude number ratio. At the same time, during the modeling process, the model material and the liquid properties inside the storage tank are similar as a premise to ensure that the parameters in the numerical model modeling are consistent with the physical liquid storage tank. Finally, OpenFOAM is used for modeling.

[0052] In this embodiment, the hydrodynamic performance calculation and prediction in step S2 is a combination of numerical calculation using the OpenFOAM open-source computational fluid dynamics code, physical model experimental methods, and machine learning prediction methods.

[0053] In this embodiment, the numerical calculation steps based on the OpenFOAM open-source computational fluid dynamics code are as follows: the initial values ​​of the model, the fluid computational domain grid, boundary conditions, numerical algorithm, and environmental conditions are determined sequentially using the numerical simulation process; the calculation process can be stopped at any time, and then the pressure, flow rate, and liquid level data results in the tank at the stopped time node are obtained. The above results are compared with the measured results. If data updates are needed, the measured results are used as the initial conditions for the next calculation of the numerical calculation model as input quantities to continue the calculation.

[0054] In this embodiment, the numerical simulation process uses the finite volume method to decompose the computational domain and generate a mesh. Within each computational cell, the Reynolds-averaged Navier-Stokes equations are solved to obtain the pressure-velocity coupling relationship at each point within the tank. Initial values ​​are obtained from data collected on a physical storage tank, ensuring that the boundary conditions maintain the same properties as the physical tank. If the gradient of the impermeable flow along the wall is zero, the tank moves under external environmental loads, ensuring that the wall position remains consistent with the movement of the physical tank. The experimental platform for the numerical simulation process needs to meet the environmental factors faced by the tank. Model similarity uses the Frode criterion with gravity as the primary factor to scale down the force, characteristic length, velocity, mass, and time parameters. Simultaneously, detection sensors are set at corresponding locations for model testing. Different loads are simulated on the experimental platform for different problems, and relevant hydrodynamic performance calculations and analyses are performed.

[0055] In this embodiment, the calculation steps of the neural network model based on machine learning in step S2 are as follows: First, the results obtained from the physical object, experimental model, and numerical model should be classified and organized, and the total sample size should be divided into two parts of 50% and 50%. One part is used as the sample data for training the neural network model, and the other part is used as the verification data for verifying the reliability and accuracy of the model. The neural network model is divided into two parts: feedforward and feedback. By repeatedly calculating the training sample data, the accuracy of the neural network model is improved, and finally, reliable feature quantity output is achieved.

[0056] In this embodiment, the step S3, which involves using data acquisition results from different types of sensors as measured data to interact with hydrodynamic performance calculation results, is as follows:

[0057] S3-1. Collect the four main parameters of pressure, displacement, liquid level change, and temperature from the output of physical objects, experimental models, experimental model data, and machine learning model data.

[0058] S3-2. Organize and store the data based on the differences in sampling frequencies of different models;

[0059] S3-3. The data interaction process is divided into four stages: data mapping, verification, assimilation, and feedback. First, the collected and categorized data are mapped simultaneously, and data from different models at the same time are compared. The sampling frequency, from highest to lowest, is: numerical model, experimental model, physical model, and machine learning model; the data accuracy weight, from highest to lowest, is: physical model, experimental model, numerical model, and machine learning model. In the verification process, physical data is used to correct the experimental model data, the corrected experimental model data is used to correct the numerical model data, and the corrected numerical model data is used to train the machine learning model and verify it against the physical data. Finally... Data assimilation between different models is achieved, with a verification error requirement of ≤5%. After completing the above interaction process, data feedback is provided, redundant and inaccurate data is cleaned up, and feature extraction is completed. Simultaneously, during the data acquisition process, measured data and numerical calculation results can be collected at the same time, and the two results are compared. If there are errors in the pressure, velocity, and liquid level results, and the error is greater than 5% of the measured results, the error is calculated by the formula (measured data - calculated data) / measured data > 5%. If the numerical calculation data input conditions are updated and replaced, the above data interaction process is automatically judged by a pre-written program, realizing unattended automatic operation.

[0060] The above structure utilizes different types of sensors for data acquisition and multi-model hydrodynamic calculations. The calculation results are obtained from the open-source fluid dynamics software OpenFOAM and hydrodynamic experimental calculations, and interact with the above-mentioned measured data (the experimental data is obtained after data acquisition by different types of sensors, including data collected during the test process and on-site data collected during the production process of the storage tank. The types and quantities of sensors here should be understood as being set according to requirements, that is, any type and quantity are included in this patent, such as pressure, liquid level, and temperature).

[0061] In step S3, the performance evaluation and diagnostic prediction module mainly plots and visualizes performance curves for liquid surface state, sloshing amplitude, flow field state, and wall pressure data. This allows for the analysis of the variation patterns of the hydrodynamic performance of the storage tank under different environmental loads. Using these patterns, the vibration modes and stress concentration locations of the storage tank are determined, providing data support for subsequent diagnostic prediction. Furthermore, the diagnostic prediction module mainly performs an overall evaluation of the storage tank's performance and predicts the risks and accidents that may be caused by changes in loads such as wind, earthquakes, and foundation. The diagnosis module retrospectively analyzes faults that have occurred due to the passage of time and external loads to clarify the causes of the faults.

[0062] Furthermore, the performance evaluation mainly involves plotting and visualizing performance curves based on data such as liquid surface condition, sloshing amplitude, flow field condition, and wall pressure, thereby analyzing the variation patterns of the hydrodynamic performance of the storage tank under different environmental loads. Further, based on these patterns, the vibration modes and stress concentration locations of the storage tank are derived, providing data support for subsequent diagnostic predictions.

[0063] In this embodiment, the performance evaluation in step S4 mainly assesses the various parameter data that affect the safety performance of the storage tank, and performs performance curve plotting and visualization output to analyze the variation law of the hydrodynamic performance of the storage tank under different environmental loads.

[0064] The diagnostic prediction process primarily involves a comprehensive assessment of the storage tank's performance and prediction of potential risks and accidents caused by changes in loads such as wind, earthquakes, and foundation conditions. Diagnosis, on the other hand, involves retrospectively analyzing past failures caused by the passage of time and external loads to clarify the causes of those failures. By integrating these processes, decision-making recommendations can be provided for the design, maintenance, structural optimization, and performance improvement of the storage tank.

[0065] like Figure 5 As shown, the internal logic and data interaction process of the twin model construction are as follows: The OpenFOAM calculation process in the figure consists of the following steps in sequence: structural modeling, mesh drawing, initialization of physical field processing, parameter setting, algorithm setting, solution, and post-processing. The processing of each step uses conventional techniques and is therefore not described in detail. The following section describes a virtual tank and a physical tank constructed based on the basic data of the storage tank, with corresponding parameters. The virtual tank consists of a machine learning model and a numerical analysis model, while the physical tank consists of an experimental model and a real tank. Its operation is comprised of the following two aspects:

[0066] First, the results of the numerical analysis model are verified by the experimental data such as liquid level changes, temperature, displacement and pressure collected in the experimental model. If the numerical analysis results do not match the experimental results, the numerical analysis model is optimized to obtain more accurate numerical analysis results. If the numerical analysis results match the experimental results, the results are input into the machine learning model so that the machine learning model can output a reliable learning result after obtaining a large amount of data.

[0067] Secondly, a series of abnormal data are generated when a real storage tank is subjected to external stimuli. The learning results obtained by a machine learning model are used to predict the risk of this series of abnormal data. At the same time, the prediction results are fed back into the numerical analysis model to realize the process of abnormality and the source analysis of the problem.

[0068] like Figure 6The diagram shows the specific setup and execution process of the OpenFOAM numerical model in the hydrodynamic performance calculation module 16. The specific description is as follows: The interaction process between the numerical model and the parameters collected by experimental and field-measured sensors is described; the numerical analysis model is built using the open-source OpenFOAM code, and its workflow is as follows: building the numerical model, mesh generation, setting analysis parameters, selecting the solver, and computation and post-processing; the setting of the model parameters, analysis parameters, and initial conditions of the numerical analysis model is mainly completed by the contents of the 0.orig, constant, and system files.

[0069] The 0.orig file contains the initial conditions required for the model, the constant file mainly contains the elements required for model operation, such as motion, and the system file mainly contains the model construction, mesh generation, analysis parameters, and solver settings. These three files enable the establishment and operation of the numerical analysis model. The experimental model uses five types of sensors (e.g., wall thickness sensor, machine vision, water level gauge, pressure sensor, and temperature sensor) to collect data on wall thickness, geometric dimensions, water level, pressure, and temperature. Changes in wall thickness and geometry affect the model's structure; depending on the degree of change, mesh re-generation is necessary, requiring updates to the shape and mesh in the numerical analysis model. Water level, pressure, and temperature affect the relevant hydrodynamic conditions of the numerical model, thus requiring updates to the numerical solver and boundary conditions.

[0070] In this embodiment, the vertical cylindrical storage tank model can be replaced by a liquid storage container of any shape, and the sensor installation position is determined accordingly based on the container characteristics. The numerical software solves the Navier-Stokes equations based on the finite volume method to calculate characteristic results such as pressure, liquid level and flow velocity. Other numerical solution methods can also be used to solve the pressure-velocity coupling equation of fluid motion, but it needs to be able to stop at any time, replace the input parameters at a specific moment, and start the calculation again.

[0071] Compared with existing technologies, the technical advantages of the digital twin evaluation system and method for hydrodynamic performance of liquid storage tanks obtained by this invention are as follows: This method fully leverages the advantages of traditional theoretical analysis / numerical simulation methods combined with the latest digital twin and machine learning methods to achieve accurate and rapid analysis of the hydrodynamic performance of liquid storage tanks in actual engineering projects, as well as risk prediction and failure process inversion, providing a full life-cycle assessment of tank performance. It replaces the initial parameters of the original numerical model with measured data from the actual storage tank as the input parameters of the numerical model, avoiding inaccuracies caused by changes in the structural parameters of the actual storage tank over time and improving the accuracy of the numerical model. Simultaneously, it replaces the process of training an artificial intelligence model using machine learning in the digital twin method with a numerical model that considers the actual physical processes, improving model prediction efficiency while providing more comprehensive and complete data on the fluid movement process within the storage tank. This enables hydrodynamic performance analysis, risk warning, and accident cause tracing.

[0072] This invention can effectively improve the accuracy of calculating the hydrodynamic performance of storage tanks using numerical calculations and machine learning; it also increases the speed and flexibility of obtaining results through actual measurements and physical model measurements, allowing for the consideration of more environmental variables and structural geometric features; by combining these two methods, comprehensive and real-time fluid motion data within the storage tank can be obtained, which can be used for storage tank risk prediction and accident problem tracing. This improves the safety of storage tanks and guides their design.

Claims

1. A digital twin evaluation system for the hydrodynamic performance of a liquid storage tank, characterized in that, Includes the following modules: A digital twin model module (15) is composed of a physical data acquisition module (1-1), a similar physical model testing module (1-2), and a liquid storage tank numerical simulation module (1-3). The physical data acquisition module (1-1) is used to acquire the changes of four parameters—pressure, displacement, liquid level change, and temperature—of the actual service-ready engineering liquid storage container under the influence of the external environment. The similar physical model testing module (1-2) is used to perform similarity processing based on the geometric dimensions of the physical storage tank and using the Froude gravity similarity criterion. The liquid storage tank numerical simulation module (1-3) uses the open-source computational fluid dynamics software OpenFOAM for modeling and ensures that all parameters and conditions in the modeling are consistent with those of the physical liquid storage tank. The numerical simulation calculation method, physical model experimental method and machine learning prediction method based on OpenFOAM open source computational fluid dynamics code are used together to calculate and analyze. The calculation results can be stopped at any time, and the pressure, flow velocity and liquid level data of the tank at the stop time point are obtained. The above results are compared with the measured results. If the data needs to be updated, the measured results are used as the initial conditions for the next calculation of the numerical calculation model as input to continue to calculate the hydrodynamic performance of the pressure, flow velocity and liquid level data of the tank (16). The four parameters of pressure, displacement, liquid level change and temperature output by physical objects, numerical models, experimental models and machine learning models are collected and classified. The data are sorted and stored by integrating the output results of each model through the hydrodynamic performance calculation module (16). Then, the collected and classified data are interactively mapped at the same time. The data of different models at the same time are compared and corrected. After completing the above interactive process, the data is fed back and redundant inaccurate data is cleaned up before the data collection and interaction module (17) for feature extraction is completed. The performance curves of various parameters affecting the safety performance of the storage tank are plotted and visualized. The variation law of hydrodynamic performance of the storage tank under different environmental loads is further analyzed, the vibration mode and stress concentration location of the storage tank are obtained, and then the risks and accidents that may be caused by the performance of the storage tank under load changes are predicted. The failures that have occurred due to the passage of time and external loads are retrospectively analyzed, thereby clarifying the cause of the failure. The performance evaluation and diagnosis prediction module (18) is used, in which the digital twin model module (15), hydrodynamic performance calculation module (16), data acquisition and interaction module (17) and performance evaluation and diagnosis prediction module (18) are linked in sequence, and the data acquisition and interaction module (17) is also linked to the digital twin model module (15).

2. A digital twin evaluation method for the hydrodynamic performance of a liquid storage tank, comprising using the digital twin evaluation system for the hydrodynamic performance of a liquid storage tank as described in claim 1, characterized in that, Specifically, the following steps are included: S1. Twin model construction steps, wherein the twin model construction includes data acquisition steps for the physical liquid storage tank, similar physical model construction steps, and numerical modeling steps for the liquid storage tank; S2. Hydrodynamic performance calculation steps, which include numerical calculation steps based on OpenFOAM open-source computational fluid dynamics code, physical model test analysis steps based on hydrodynamic experiments, and neural network model calculation steps based on machine learning. S3. The step of using data acquisition results from different types of sensors as measured data to interact with hydrodynamic performance calculation results; S4. Input the data collected in step S3 into the internal pre-built program. The relevant analysis algorithm performs post-processing and displays the relevant data in the form of charts in a window interface after statistical analysis. Time domain or frequency domain analysis is used as the analysis content. The data after interactive comparison of relevant parameters is analyzed to analyze the possible differences. Then, the obtained results are used for performance evaluation, risk prediction and fault diagnosis.

3. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 2, characterized in that, In step S1, the data acquisition steps for the physical storage tank are as follows: by collecting the changes in four parameters of the engineering structure in actual service under the influence of the external environment, the sampling frequency and total sampling amount should have a sufficient time span. Generally, the sample data period should not be less than 6 months, the sampling frequency should not exceed 30 minutes, and the basic materials of the physical storage tank need to be collected once before the collection process.

4. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 2, characterized in that, In step S1, the similar physical model construction step is to perform similarity processing based on the geometric dimensions of the physical storage tank. The steps are as follows: using the Froude gravity similarity criterion, the relevant parameters in the model are scaled using the Froude number ratio. At the same time, in the modeling process, the similarity of the model material and the liquid properties inside the storage tank is taken as a premise to ensure that the parameters in the numerical model modeling are consistent with the physical liquid storage tank. Finally, OpenFOAM is used for modeling.

5. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 2, characterized in that, In step S2, the hydrodynamic performance calculation and prediction is performed by combining multiple methods, including numerical calculation using the OpenFOAM open-source computational fluid dynamics code, physical model experimental methods, and machine learning prediction methods.

6. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 5, characterized in that, The numerical calculation steps based on the OpenFOAM open-source computational fluid dynamics code are as follows: The initial values ​​of the model, the computational domain mesh, boundary conditions, numerical algorithm, and environmental conditions are determined sequentially using the numerical simulation process; the calculation process can be stopped at any time, and then the pressure, flow velocity, and liquid level data in the tank at the stopped time point are obtained. The above results are compared with the measured results. If data updates are needed, the measured results are used as the initial conditions for the next calculation of the numerical calculation model as input quantities to continue the calculation.

7. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 6, characterized in that, The numerical simulation process employs the finite volume method for computational domain decomposition and mesh generation. Within each computational cell, the Reynolds-averaged Navier-Stokes equations are solved to obtain the pressure-velocity coupling relationship at each point within the tank. Initial values ​​are obtained from data collected on a physical storage tank, ensuring that the boundary conditions maintain the same properties as the physical tank. If the impermeable flow rate along the wall has a zero gradient, the tank moves under external environmental loads, ensuring that the wall position remains consistent with the movement of the physical tank. The experimental platform for the numerical simulation process must meet the environmental factors faced by the tank. Model similarity is achieved using the Frode criterion with gravity as the primary factor, scaling down the force, characteristic length, velocity, mass, and time parameters. Simultaneously, detection sensors are installed at corresponding locations for model testing. Different loads are simulated on the experimental platform for different problems, and relevant hydrodynamic performance calculations and analyses are performed.

8. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 2, characterized in that, The calculation steps of the neural network model based on machine learning in step S2 are as follows: First, the results obtained from the physical object, experimental model, and numerical model should be classified and organized, and the total sample size should be divided into two parts of 50% and 50%. One part is used as the sample data for training the neural network model, and the other part is used as the verification data for verifying the reliability and accuracy of the model. The neural network model is divided into two parts: feedforward and feedback. By repeatedly calculating the training sample data, the accuracy of the neural network model is improved, and finally, reliable feature quantity output is achieved.

9. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 2, characterized in that, The steps in step S3 for interacting the data acquisition results from different types of sensors as measured data with the hydrodynamic performance calculation results are as follows: S3-1. Collect the four parameters of pressure, displacement, liquid level, and temperature output from physical objects, experimental models, experimental model data, and machine learning model data by category. S3-2. Organize and store the data based on the differences in sampling frequencies of different models; S3-3. The data interaction process is divided into four stages: data mapping, verification, assimilation, and feedback. First, the collected and categorized data are mapped simultaneously, and data from different models at the same time are compared. The sampling frequency, from highest to lowest, is: numerical model, experimental model, physical model, machine learning model; the data accuracy weight, from highest to lowest, is: physical model, experimental model, numerical model, machine learning model; and the hydrodynamic performance calculation speed, from highest to lowest, is: machine learning model, numerical model, experimental model, physical model. In the verification process, physical data is used to correct experimental model data, the corrected experimental model data is used to correct numerical model data, and the corrected data model data is used to train the machine learning model and verify it against the physical data. Ultimately, different models are compared. The data assimilation process is as follows: verification error is required to be ≤5%. After completing the above interaction process, data feedback is performed, redundant and inaccurate data is cleaned up, and feature extraction is completed. At the same time, during the data acquisition process, measured data and numerical calculation results are collected simultaneously and compared. If there are errors in the pressure, velocity, and liquid level results, when the error is greater than 5% of the measured result, the error calculation formula is (measured data - calculated data) / measured data > 5%. Then, the input conditions of the numerical calculation data are updated and replaced. At this time, the numerical calculation is stopped and the measured data is used to replace the numerical calculation result at the corresponding time as the initial value for the next calculation step. The above data interaction process is automatically judged by a pre-written program to achieve unattended automatic operation.

10. The digital twin evaluation method for the hydrodynamic performance of a liquid storage tank according to claim 2, characterized in that, The performance evaluation in step S4 involves assessing the various parameters affecting the safety performance of the storage tank, plotting performance curves, and visualizing the output to analyze the variation of the hydrodynamic performance of the storage tank under different environmental loads.

Images