PINN-based submarine pipeline scour and vibration coupling prediction method and system

Abstract

The invention discloses a PINN-based submarine pipeline scour and vibration coupling prediction method and system. Comprising the steps of generating training data through high-fidelity CFD and structural dynamics coupling simulation, simulating a turbulent flow field by adopting a partial averaging method, simulating pipeline vibration by adopting a vector finite element method, and realizing flow-pipe-sediment coupling solution through an immersed boundary method; a multi-task neural network fusing hydromechanics, structural dynamics and sediment transport equation constraints is constructed, space coordinates and time are used as input, flow velocity, pressure, pipeline displacement and scouring depth are output, a loss function comprises a data error term and a physical equation residual term, and the flow velocity, the pressure, the pipeline displacement and the scouring depth are calculated. Through residual errors of an automatic differential calculation equation, a pipeline vibration equation and a sediment transportation equation, a network is trained through an optimization algorithm, physical constraints are met, and meanwhile data are fitted; and applying the trained model to long-duration scouring and vibration coupling prediction of the pipeline, and rapidly outputting a spatio-temporal evolution result of vibration and scouring according to a given ocean current condition and an initial state of the pipeline.

Description

Technical Field

[0001] This invention relates to the field of subsea pipelines and physical information neural networks, and more specifically, to a method and system for predicting the coupling of scour and vibration in subsea pipelines based on PINN. Background Technology

[0002] Currently, traditional analysis methods for subsea pipeline vibration and scour problems mainly rely on finite element method (FEM) and Computational Fluid Dynamics (CFD) simulations. Existing pipeline vibration analyses often use finite element software (such as ANSYS and Abaqus) to calculate the pipeline's natural frequencies and vibration modes, but they cannot obtain information about the flow field around the pipeline, nor do they consider the effects of sediment. This method can only provide the structural response, neglecting fluid-structure interaction, and is difficult to address the coupled problem of vortex-induced vibration and sediment scour induced by actual ocean currents.

[0003] Existing numerical simulations of pipeline scour typically employ CFD software (such as Flow-3D and OpenFOAM) to establish three-dimensional models, using turbulence models like RNG k-ε and sediment transport models to calculate the flow field and scour morphology. Flow-3D's built-in scour module predicts sediment transport through mass conservation and convection-diffusion equations, and calculates bed-loaded sediment transport rates using the Meyer-Peter–Müller formula. However, these methods have significant limitations when simulating subsea pipeline scour problems. First, most studies treat local pipeline scour and vortex-induced vibration separately, lacking a simultaneous solution for both. For example, existing models primarily focus on the multiphase fluid-structure interaction scour problem involving the pipeline, sediment, and flow field, without simultaneously considering the dynamic vibration response of the pipeline structure. Second, this method is computationally intensive, usually requiring high-resolution meshes and long transient calculations to obtain stable results, resulting in extremely long computation times and very high computational resource requirements. Furthermore, the accuracy of the simulation results heavily depends on the grid quality, the precise selection of parameters for the turbulence model and sediment model. These factors together limit the applicability of this method in engineering practice, making it difficult to achieve rapid prediction and multi-scheme comparison.

[0004] In terms of algorithm module design, existing CFD simulation systems are also quite complex. For example, the three-dimensional scour simulation system disclosed in CN113947042A (Three-dimensional numerical simulation method and system for fluid scour of subsea pipelines) consists of multiple functional modules: the first setting module is responsible for establishing the simulation task in the preset CFD software and setting parameters such as calculation time and fluid compressibility; the second setting module sets physical models and parameters such as turbulence model and sediment model; the establishment module is used to create a three-dimensional scour model including pipeline components, sand bed components and fluid components; the third setting module performs computational domain mesh generation and sets initial and boundary conditions; the fourth setting module specifies the output time step and data type; the calculation module runs the solution and obtains flow field information and the scour pit expansion process. This series of modules constitutes a large and complex process, requiring users to perform a lot of preprocessing and iterative solutions, and it is limited to numerical simulation, lacking artificial intelligence assistance, and unable to quickly respond to real-time changes in the ocean current environment in marine engineering.

[0005] In summary, the existing finite element / CFD analysis process has the following drawbacks: high computational cost, long solution time, inability to simultaneously and efficiently predict the coupled evolution of flow-structure-sediment, lack of intelligent algorithms that integrate prior physical knowledge, and difficulty in meeting the needs of rapid safety assessment throughout the entire life cycle. Summary of the Invention

[0006] This invention overcomes the shortcomings of the prior art and proposes a method and system for predicting the coupling of scour and vibration in subsea pipelines based on PINN.

[0007] The first aspect of this invention provides a PINN-based method for predicting the coupled scour and vibration of subsea pipelines, comprising: Data generation module: PINN training data is generated using high-fidelity CFD and structural dynamics simulation. Specifically, the partially averaged Navier-Stokes method is used to simulate the fluid turbulence field, the vector finite element method is used to simulate the vibration response of the elastic pipe, and the coupled solution of flow-pipe-sediment is achieved through the iterative immersed boundary method. The system sets preset prior conditions, runs short-duration CFD-FSI simulations, collects multi-physical quantity data of the pipe under different conditions, and forms the training set and validation set of PINN. The PINN model construction and training module constructs a multi-task neural network that integrates constraints from fluid mechanics, structural dynamics, and sediment transport equations. The network input includes spatial location (x, y, z) and time, and the output consists of physical quantities such as fluid velocity field, pressure field, pipe vibration displacement, and bed scour depth. A preset loss function is set. The neural network is trained using Adam and L-BFGS optimization algorithms to fit the training data while satisfying physical constraints. This PINN model incorporates the turbulence control equations, pipe kinematics equations, and bed scour evolution equations, guided by physical priors. Prediction and Application Module: The trained PINN model is used for long-term prediction. In the prediction stage, given different ocean current conditions and the initial state of the pipeline, the PINN model makes predictions and outputs the spatiotemporal distribution of the pipeline vibration-scour coupling process. The system generates the pipeline vibration displacement change curve over time, the evolution curve of scour depth and span length in real time through the spatiotemporal distribution, and extracts the scour morphology prediction formula in reverse based on the network output.

[0008] In this solution, the setting of the preset loss function specifically refers to: The preset loss function consists of several parts: first, a data error term, which is used to ensure that the network output matches the simulation or experimental data; second, a physical residual term, which is calculated by automatic differentiation of the residuals of the Navier-Stokes equations and continuity equations, pipeline elastic vibration equations, and sediment transport equations, and used as a loss analysis.

[0009] In this scheme, the preset prior conditions include: The system sets prior conditions for pipeline geometry, seabed characteristics, and fluid parameters.

[0010] In this solution, the multi-physical quantity data specifically includes: Data on multiple physical quantities of the pipeline, including flow field velocity, pressure field, pipeline displacement, and bed sediment distribution, were collected under different vibration frequencies, amplitudes, flow velocities, and burial depths.

[0011] In this scheme, the training data includes: The multi-physical quantity data is divided according to a preset ratio to generate training and validation sets; Both the training and validation sets include physical quantity data under different conditions.

[0012] In this scheme, the evolution curves of the pipeline vibration displacement over time, scour depth, and overhang length also include: Multiple sets of first and second vectors are set according to different ocean current conditions and initial pipe states; Each pair of first and second vectors corresponds to a type of ocean current condition and pipeline initial state. The first vector includes ocean current condition parameters, and the second vector includes the initial state of the pipeline. The Manhattan distance is introduced to calculate the deviation of two sets of vectors under two different ocean current conditions, and the first deviation rate is obtained. The pipeline vibration displacement variation curves under two ocean current conditions and the pipeline in its initial state were obtained, and linear regression prediction was performed on the two curves. The prediction period was set to short duration, and two sets of predicted displacements were obtained. Calculate the deviation rate between the two sets of predicted displacements to obtain the second deviation rate; The predictive effectiveness of the variation curve is evaluated based on the difference between the first and second deviation rates.

[0013] A second aspect of the present invention also provides a PINN-based subsea pipeline scour and vibration coupling prediction system. The system includes a memory, a processor, and a data interface. The memory includes a PINN-based subsea pipeline scour and vibration coupling prediction program. When executed by the processor, the PINN-based subsea pipeline scour and vibration coupling prediction program performs the following steps: Data generation module: PINN training data is generated using high-fidelity CFD and structural dynamics simulation. Specifically, the partially averaged Navier-Stokes method is used to simulate the fluid turbulence field, the vector finite element method is used to simulate the vibration response of the elastic pipe, and the coupled solution of flow-pipe-sediment is achieved through the iterative immersed boundary method. The system sets preset prior conditions, runs short-duration CFD-FSI simulations, collects multi-physical quantity data of the pipe under different conditions, and forms the training set and validation set of PINN. The PINN model construction and training module constructs a multi-task neural network that integrates constraints from fluid mechanics, structural dynamics, and sediment transport equations. The network input includes spatial location (x, y, z) and time, and the output consists of physical quantities such as fluid velocity field, pressure field, pipe vibration displacement, and bed scour depth. A preset loss function is set. The neural network is trained using Adam and L-BFGS optimization algorithms to fit the training data while satisfying physical constraints. This PINN model incorporates the turbulence control equations, pipe kinematics equations, and bed scour evolution equations, guided by physical priors. Prediction and Application Module: The trained PINN model is used for long-term prediction. In the prediction stage, given different ocean current conditions and the initial state of the pipeline, the PINN model makes predictions and outputs the spatiotemporal distribution of the pipeline vibration-scour coupling process. The system generates the pipeline vibration displacement change curve over time, the evolution curve of scour depth and span length in real time through the spatiotemporal distribution, and extracts the scour morphology prediction formula in reverse based on the network output.

[0014] A third aspect of the present invention also provides a computer-readable storage medium comprising a PINN-based subsea pipeline scour and vibration coupling prediction program, wherein when the PINN-based subsea pipeline scour and vibration coupling prediction program is executed by a processor, it implements the steps of the PINN-based subsea pipeline scour and vibration coupling prediction method as described in any of the preceding claims.

[0015] The following technical effects can be achieved through this invention: This invention introduces physical prior knowledge through PINN, significantly improving prediction accuracy and reducing data requirements. Compared to purely data-driven methods, this method utilizes physical residuals to guide network learning, reducing fitting errors for unknown flow fields and scour evolution, achieving results close to high-fidelity simulations, while reducing the number of simulation samples required by several times.

[0016] Compared with existing technologies, the present invention significantly improves computational efficiency. In practical applications, CFD-based coupled simulations often take several hours to several days, while the trained PINN prediction can complete the long-term prediction of the same process within seconds to minutes, realizing rapid presentation and corresponding field evolution.

[0017] This invention integrates flow field, structural vibration, and sediment scour information, providing more comprehensive engineering decision support. The scour pattern prediction formula based on PINN inversion is physically interpretable and can be used to guide pipeline design and maintenance; simultaneously, the vibration prediction results help assess the risk of vortex-induced resonance.

[0018] In summary, this method outperforms existing technologies in terms of prediction speed, computational cost, model stability, and ease of use. By combining artificial intelligence with engineering physics principles, this system can provide efficient and accurate prediction and decision support for the safe operation of pipelines throughout their entire lifecycle. Attached Figure Description

[0019] Figure 1 A flowchart of a PINN-based method for predicting the coupling of scour and vibration in subsea pipelines is shown. Figure 2 The flowchart of the PINN model construction process of the present invention is shown; Figure 3 A block diagram of a PINN-based subsea pipeline scour and vibration coupling prediction system of the present invention is shown. Detailed Implementation

[0020] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0022] Figure 1 A flowchart of a PINN-based method for predicting the coupling of scour and vibration in subsea pipelines is shown.

[0023] like Figure 1 As shown, the first aspect of the present invention provides a method for predicting the coupled scour and vibration of subsea pipelines based on PINN, comprising: Data generation module: PINN training data is generated using high-fidelity CFD and structural dynamics simulation. Specifically, the partially averaged Navier-Stokes method is used to simulate the fluid turbulence field, the vector finite element method is used to simulate the vibration response of the elastic pipe, and the coupled solution of flow-pipe-sediment is achieved through the iterative immersed boundary method. The system sets preset prior conditions, runs short-duration CFD-FSI simulations, collects multi-physical quantity data of the pipe under different conditions, and forms the training set and validation set of PINN. The PINN model construction and training module constructs a multi-task neural network that integrates constraints from fluid mechanics, structural dynamics, and sediment transport equations. The network input includes spatial location (x, y, z) and time, and the output consists of physical quantities such as fluid velocity field, pressure field, pipe vibration displacement, and bed scour depth. A preset loss function is set. The neural network is trained using Adam and L-BFGS optimization algorithms to fit the training data while satisfying physical constraints. This PINN model incorporates the turbulence control equations, pipe kinematics equations, and bed scour evolution equations, guided by physical priors. Prediction and Application Module: The trained PINN model is used for long-term prediction. In the prediction stage, given different ocean current conditions and the initial state of the pipeline, the PINN model makes predictions and outputs the spatiotemporal distribution of the pipeline vibration-scour coupling process. The system generates the pipeline vibration displacement change curve over time, the evolution curve of scour depth and span length in real time through the spatiotemporal distribution, and extracts the scour morphology prediction formula in reverse based on the network output.

[0024] It should be noted that by outputting the spatiotemporal distribution of the pipeline vibration-scour coupling process, there is no need to perform a time-consuming CFD simulation again. PINN is a physical information neural network.

[0025] This invention, through an integrated prediction program, combines short-duration CFD with long-duration PINN calculations, significantly improving prediction efficiency. The partially averaged Navier-Stokes (PANS) method is used to simulate and calculate various fluid responses.

[0026] Here, the Vector Finite Element Method (VFIFE), also known as the Immersed Boundary Method (IBM), is used for coupled analysis. The short execution time is typically from a few seconds to a few minutes.

[0027] Compared with the prior art, the present invention can achieve the following technical effects: Embedding of physical laws: The PINN architecture is adopted, and the governing equations of fluid, structure and sediment dynamics are directly embedded into the loss function of the neural network as physical constraints, so that the model training not only depends on data, but must also follow the underlying physical laws.

[0028] Training data generation and use: Limited, high-quality initial data are generated through short-term, high-fidelity CFD simulations and provided to PINN as “samples” of physical laws, thereby significantly reducing the reliance on large amounts of experimental or full-cycle simulation data.

[0029] The essential difference in the prediction process: In the prediction phase, the trained PINN model can directly and quickly output the physical quantities of the entire field at any time, realizing the rapid deduction of long-term evolution and avoiding the stepwise numerical integration required by traditional CFD methods, which is extremely costly.

[0030] Modeling capability for complex working conditions: This method naturally incorporates actual working conditions such as hydrodynamics and sediment disturbance through physical equations at the algorithm level, rather than relying on empirical assumptions, thereby enhancing the model's generalization ability and applicability in complex ocean current environments.

[0031] Integrated Solution: This invention constructs an end-to-end coupling prediction process, from input of pipeline and environmental parameters to training data generation, PINN network training, and finally directly outputs the coupling prediction results of vibration and scour and interpretable physical formulas, realizing a complete closed loop from problem to solution.

[0032] Figure 2 A flowchart of the PINN model construction process of the present invention is shown.

[0033] According to an embodiment of the present invention, the setting of the preset loss function specifically includes: The preset loss function consists of several parts: first, a data error term, which is used to ensure that the network output matches the simulation or experimental data; second, a physical residual term, which is calculated by automatic differentiation of the residuals of the Navier-Stokes equations and continuity equations, pipeline elastic vibration equations, and sediment transport equations, and used as a loss analysis.

[0034] It should be noted that the PINN model utilizes physical priors to guide learning, enabling it to achieve high-precision models even with limited training data.

[0035] According to an embodiment of the present invention, the preset prior conditions include: The system sets prior conditions for pipeline geometry, seabed characteristics, and fluid parameters.

[0036] According to an embodiment of the present invention, the multi-physical quantity data specifically includes: Data on multiple physical quantities of the pipeline, including flow field velocity, pressure field, pipeline displacement, and bed sediment distribution, were collected under different vibration frequencies, amplitudes, flow velocities, and burial depths.

[0037] According to an embodiment of the present invention, the training data includes: The multi-physical quantity data is divided according to a preset ratio to generate training and validation sets; Both the training and validation sets include physical quantity data under different conditions.

[0038] It should be noted that the preset ratio can be 80% (training set) and 20% (validation set).

[0039] According to an embodiment of the present invention, the evolution curves of the pipeline vibration displacement over time, the scour depth, and the span length further include: Multiple sets of first and second vectors are set according to different ocean current conditions and initial pipe states; Each pair of first and second vectors corresponds to a type of ocean current condition and pipeline initial state. The first vector includes ocean current condition parameters, and the second vector includes the initial state of the pipeline. The Manhattan distance is introduced to calculate the deviation of two sets of vectors under two different ocean current conditions, and the first deviation rate is obtained. The pipeline vibration displacement variation curves under two ocean current conditions and the pipeline in its initial state were obtained, and linear regression prediction was performed on the two curves. The prediction period was set to short duration, and two sets of predicted displacements were obtained. Calculate the deviation rate between the two sets of predicted displacements to obtain the second deviation rate; The predictive effectiveness of the variation curve is evaluated based on the difference between the first and second deviation rates.

[0040] It should be noted that the pipeline vibration displacement variation curve refers to the pipeline vibration displacement variation curve over time. Given different ocean current conditions and pipeline initial states, there are often multiple conditions and states. In multiple sets of first and second vectors, each ocean current condition and pipeline initial state corresponds to a set of first and second vectors, used to simulate and predict multiple scenarios for multi-dimensional risk prediction of the actual pipeline condition. Each dimension of the first vector corresponds to a parameter of each ocean current condition, and each dimension of the second vector corresponds to the pipeline initial state. Ocean current condition parameters include vibration frequency, amplitude, flow velocity, and burial depth, while pipeline initial states include flow field velocity, pressure field, pipeline displacement, and bed sediment distribution.

[0041] It is worth mentioning that in the long-term pipeline prediction process using the PINN model, multiple pipeline parameters require coupled process analysis and spatiotemporal distribution prediction, and corresponding curves need to be generated in real time, such as pipeline vibration displacement change curves and evolution curves of scour depth and overhang length. However, the effectiveness and accuracy of the change pattern analysis and model prediction of the evolution curves cannot be simply inferred from the model loss. Furthermore, existing technologies lack a process for evaluating the predictive effectiveness of the multiple prediction parameters of the PINN model. Based on this, this invention introduces a multi-parameter vector mode to vectorize and differentiate the input parameters under different ocean current conditions. After the curve evolution is generated, linear regression prediction is performed on the relevant prediction curves to evaluate the consistency of the deviation rate before and after. If the consistency is low, i.e., the difference is high, it indicates low prediction effectiveness. Correspondingly, the training mode and loss function of the PINN model can be adjusted to improve the model's accuracy. Generally speaking, a low difference can be used to evaluate the model's declining prediction accuracy in repeated training and input / output under different ocean current conditions and pipeline models, and to reasonably adjust and update the model parameters. Linear regression can be used to fit and predict using the linear equation y(x)=kx+b.

[0042] According to an embodiment of the present invention, the data generation module further includes: The input and output of the PINN model are vectorized. The input includes spatial position (x,y,z) and time, and the output is the physical quantities of fluid velocity field, pressure field, pipe vibration displacement and bed scour depth, so as to obtain the input vector and output vector. In the process of multiple pipeline simulation predictions, the input vector and output vector of each prediction process are collected, and the input vector and output vector are concatenated to obtain the prediction vector; Construct a generative model based on GAN, which includes a generator G1 and a discriminator G2; The predicted vectors corresponding to multiple pipeline simulation prediction processes are used as real data to be imported into G1 for feature learning, and the same simulation vectors of the same dimension are repeatedly generated. The simulated vectors are imported into G2 for discrimination. The discrimination process incorporates high-fidelity CFD for data verification. Before verification, the pipeline parameters of the simulated vectors are analyzed, and the generated model is adjusted based on the verification results. Repeatedly train the generator G1 and discriminator G2 in adversarial mode until the generator model reaches Nash equilibrium. A simulated dataset is generated by the generation module, and the parameters are parsed based on the simulated dataset to obtain the first training dataset.

[0043] It should be noted that each pipeline simulation prediction process includes a prediction vector. In the coupled data analysis of pipeline scour and vibration, with limited training data, overfitting of the PINN model or other existing pipeline simulation prediction models can easily occur, and prediction accuracy is affected to some extent by the training dataset. Therefore, this embodiment introduces a GAN model to effectively solve the above problems. By vectorizing the pipeline parameter inputs and outputs of multiple prediction processes, the GAN model can facilitate feature learning and improve feature extraction efficiency. Multiple adversarial training sessions generate relatively realistic simulation data, which is then used as new training data for the secondary training of the PINN model. High-fidelity CFD is introduced into the adversarial training for data validation of the discriminator; this data validation is used to determine whether the parameters conform to the actual situation.

[0044] Example 1: Example 1 uses a "submarine natural gas pipeline project" as an example: Data Generation: Training data was first obtained using CFD + structural coupling simulation. Three-dimensional models of the pipeline and seabed were built in the computational domain. Turbulence was simulated using the RANS / PANS model for the fluid, and the pipeline structure was simulated using VFIFE elastic beam elements. The flow-pipe interface was coupled using IBM. Different flow velocities and pipeline vibration forces (simulated by an excitation device to simulate eddy currents) were applied, and short-time simulations (time history T=100s) were run to record data such as the vibration displacement w(t) of the pipeline at different locations, the velocity field {u}(x,y,z,t) around the pipe, and the scour depth S(x,z,t) on the seabed. The resulting dataset contains over one hundred samples, covering typical scour evolution processes. t represents time, and x, y, and z represent the velocity components of the flow field in the corresponding directions.

[0045] PINN Model Training: A PINN network was constructed with input tensors (x, y, z, t) and outputs (u, v, w, p, S), representing the velocity field around the pipe, velocity components, vibration displacement, pressure, pipe displacement, and bed scour depth, respectively. A multi-loss function was designed, including mean squared error loss from CFD simulation data and physical loss. The physical loss term consists of the residuals from the Navier-Stokes momentum equation, continuity equation, pipe dynamics equation, and sediment transport equation. Deep learning frameworks such as TensorFlow were used for training. Adam optimization was first applied, followed by L-BFGS fine-tuning iterations to ensure the network output simultaneously satisfies both known data and physical constraints. Training results show that the network can accurately reproduce the velocity and scour evolution during short-time simulations and exhibits good consistency.

[0046] Coupled Prediction: A trained PINN model is used to make predictions over a longer time period. For example, with a fixed input flow velocity of 1.2 m / s, a simulation is run for t=10,000 s. The network quickly outputs the pipe amplitude variation curve over time and the evolution of the scour pit along the pipe axis. The results show that the length of the pipe overhang gradually increases and tends to reach equilibrium, while the scour depth increases exponentially and stabilizes at approximately 0.45 m. Based on the scour evolution data predicted by PINN, new empirical formulas can be fitted, such as formulas for predicting the equilibrium scour depth, which have higher applicability under current conditions compared to existing empirical formulas. Furthermore, PINN predicts that the pipe amplitude under the dominant vibration frequency is approximately 10% of the pipe diameter, a result close to the CFD simulation experimental value. This embodiment also applies the PINN model to safety assessments under different flow velocities and spacing conditions, finding that it can quickly obtain the coupled evolution process of pipe vibration and scour, assisting the project team in optimizing the pipe design scheme.

[0047] Performance Verification and Iteration: The PINN prediction results were compared with CFD simulations and measured data to verify the accuracy and efficiency of the method. The results show that this method maintains similar accuracy to high-fidelity simulations while reducing computation time by more than 90%, significantly saving engineering computational costs. Based on actual project feedback, the network structure and loss weights were iteratively adjusted, further improving the model's predictive ability under complex flow field conditions.

[0048] As can be seen from the above embodiments, this method and system can efficiently and stably predict the local scour-vortex-induced vibration coupling process in actual submarine pipeline projects, providing strong technical support for the safe operation of pipelines.

[0049] Figure 3 A block diagram of a PINN-based subsea pipeline scour and vibration coupling prediction system of the present invention is shown.

[0050] A second aspect of the present invention also provides a PINN-based subsea pipeline scour and vibration coupling prediction system. The system includes a memory, a processor, and a data interface. The memory includes a PINN-based subsea pipeline scour and vibration coupling prediction program. When executed by the processor, the PINN-based subsea pipeline scour and vibration coupling prediction program performs the following steps: Data generation module: PINN training data is generated using high-fidelity CFD and structural dynamics simulation. Specifically, the partially averaged Navier-Stokes method is used to simulate the fluid turbulence field, the vector finite element method is used to simulate the vibration response of the elastic pipe, and the coupled solution of flow-pipe-sediment is achieved through the iterative immersed boundary method. The system sets preset prior conditions, runs short-duration CFD-FSI simulations, collects multi-physical quantity data of the pipe under different conditions, and forms the training set and validation set of PINN. The PINN model construction and training module constructs a multi-task neural network that integrates constraints from fluid mechanics, structural dynamics, and sediment transport equations. The network input includes spatial location (x, y, z) and time, and the output consists of physical quantities such as fluid velocity field, pressure field, pipe vibration displacement, and bed scour depth. A preset loss function is set. The neural network is trained using Adam and L-BFGS optimization algorithms to fit the training data while satisfying physical constraints. This PINN model incorporates the turbulence control equations, pipe kinematics equations, and bed scour evolution equations, guided by physical priors. Prediction and Application Module: The trained PINN model is used for long-term prediction. In the prediction stage, given different ocean current conditions and the initial state of the pipeline, the PINN model makes predictions and outputs the spatiotemporal distribution of the pipeline vibration-scour coupling process. The system generates the pipeline vibration displacement change curve over time, the evolution curve of scour depth and span length in real time through the spatiotemporal distribution, and extracts the scour morphology prediction formula in reverse based on the network output.

[0051] The data interface is used to store parameter data, PINN model data, and prediction data collected and generated by the data generation module.

[0052] When the system is running, it can perform one or more steps of the above-described PINN-based method for predicting the coupling of scour and vibration in subsea pipelines.

[0053] A third aspect of the present invention also provides a computer-readable storage medium comprising a PINN-based subsea pipeline scour and vibration coupling prediction program, wherein when the PINN-based subsea pipeline scour and vibration coupling prediction program is executed by a processor, it implements the steps of the PINN-based subsea pipeline scour and vibration coupling prediction method as described in any of the preceding claims.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0055] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0056] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0057] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0058] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0059] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the coupled scour and vibration of subsea pipelines based on PINN, characterized in that, include: Data generation module: PINN training data is generated using high-fidelity CFD and structural dynamics simulation. Specifically, the partially averaged Navier-Stokes method is used to simulate the fluid turbulence field, the vector finite element method is used to simulate the vibration response of the elastic pipe, and the coupled solution of flow-pipe-sediment is achieved through the iterative immersed boundary method. The system sets preset prior conditions, runs short-duration CFD-FSI simulations, collects multi-physical quantity data of the pipe under different conditions, and forms the training set and validation set of PINN. The PINN model construction and training module constructs a multi-task neural network that integrates constraints from fluid mechanics, structural dynamics, and sediment transport equations. The network input includes spatial location (x, y, z) and time, and the output consists of physical quantities such as fluid velocity field, pressure field, pipe vibration displacement, and bed scour depth. A preset loss function is set. The neural network is trained using Adam and L-BFGS optimization algorithms to fit the training data while satisfying physical constraints. This PINN model incorporates the turbulence control equations, pipe kinematics equations, and bed scour evolution equations, guided by physical priors. Prediction and Application Module: The trained PINN model is used for long-term prediction. In the prediction stage, given different ocean current conditions and the initial state of the pipeline, the PINN model makes predictions and outputs the spatiotemporal distribution of the pipeline vibration-scour coupling process. The system generates the pipeline vibration displacement change curve over time, the evolution curve of scour depth and span length in real time through the spatiotemporal distribution, and extracts the scour morphology prediction formula in reverse based on the network output.

2. The method for predicting the coupled scour and vibration of subsea pipelines based on PINN according to claim 1, characterized in that, The preset loss function is specifically defined as follows: The preset loss function consists of several parts: first, a data error term, which is used to ensure that the network output matches the simulation or experimental data; second, a physical residual term, which is calculated by automatic differentiation of the residuals of the Navier-Stokes equations and continuity equations, pipeline elastic vibration equations, and sediment transport equations, and used as a loss analysis.

3. The method for predicting the coupled scour and vibration of subsea pipelines based on PINN according to claim 1, characterized in that, The preset prior conditions include: The system sets prior conditions for pipeline geometry, seabed characteristics, and fluid parameters.

4. The method for predicting the coupled scour and vibration of subsea pipelines based on PINN according to claim 1, characterized in that, The multi-physical quantity data specifically includes: Data on multiple physical quantities of the pipeline, including flow field velocity, pressure field, pipeline displacement, and bed sediment distribution, were collected under different vibration frequencies, amplitudes, flow velocities, and burial depths.

5. The method for predicting the coupled scour and vibration of subsea pipelines based on PINN according to claim 1, characterized in that, The training data includes: The multi-physical quantity data is divided according to a preset ratio to generate training and validation sets; Both the training and validation sets include physical quantity data under different conditions.

6. The method for predicting the coupled scour and vibration of subsea pipelines based on PINN according to claim 1, characterized in that, The pipeline vibration displacement over time curve, the scour depth, and the span length evolution curve also include: Multiple sets of first and second vectors are set according to different ocean current conditions and initial pipe states; Each pair of first and second vectors corresponds to a type of ocean current condition and pipeline initial state. The first vector includes ocean current condition parameters, and the second vector includes the initial state of the pipeline. The Manhattan distance is introduced to calculate the deviation of two sets of vectors under two different ocean current conditions, and the first deviation rate is obtained. The pipeline vibration displacement variation curves under two ocean current conditions and the pipeline in its initial state were obtained, and linear regression prediction was performed on the two curves. The prediction period was set to short duration, and two sets of predicted displacements were obtained. Calculate the deviation rate between the two sets of predicted displacements to obtain the second deviation rate; The predictive effectiveness of the variation curve is evaluated based on the difference between the first and second deviation rates.

7. A PINN-based system for predicting the coupled scour and vibration of subsea pipelines, characterized in that, The system includes: a memory, a processor, and a data interface. The memory includes a PINN-based subsea pipeline scour and vibration coupling prediction program. When the processor executes the PINN-based subsea pipeline scour and vibration coupling prediction program, it performs the following steps: Data generation module: PINN training data is generated using high-fidelity CFD and structural dynamics simulation. Specifically, the partially averaged Navier-Stokes method is used to simulate the fluid turbulence field, the vector finite element method is used to simulate the vibration response of the elastic pipe, and the coupled solution of flow-pipe-sediment is achieved through the iterative immersed boundary method. The system sets preset prior conditions, runs short-duration CFD-FSI simulations, collects multi-physical quantity data of the pipe under different conditions, and forms the training set and validation set of PINN. The PINN model construction and training module constructs a multi-task neural network that integrates constraints from fluid mechanics, structural dynamics, and sediment transport equations. The network input includes spatial location (x, y, z) and time, and the output consists of physical quantities such as fluid velocity field, pressure field, pipe vibration displacement, and bed scour depth. A preset loss function is set. The neural network is trained using Adam and L-BFGS optimization algorithms to fit the training data while satisfying physical constraints. This PINN model incorporates the turbulence control equations, pipe kinematics equations, and bed scour evolution equations, guided by physical priors. Prediction and Application Module: The trained PINN model is used for long-term prediction. In the prediction stage, given different ocean current conditions and the initial state of the pipeline, the PINN model makes predictions and outputs the spatiotemporal distribution of the pipeline vibration-scour coupling process. The system generates the pipeline vibration displacement change curve over time, the evolution curve of scour depth and span length in real time through the spatiotemporal distribution, and extracts the scour morphology prediction formula in reverse based on the network output.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a PINN-based subsea pipeline scour and vibration coupling prediction program. When the PINN-based subsea pipeline scour and vibration coupling prediction program is executed by a processor, it implements the steps of the PINN-based subsea pipeline scour and vibration coupling prediction method as described in any one of claims 1 to 6.

Images