A fatigue life prediction system and method for hybrid-driven marine riser systems

Abstract

This invention provides a hybrid-driven fatigue life prediction system and method for marine riser systems, belonging to the technical field of marine riser systems. The system includes a fluid-structure interaction (FSI) solution module, a physical monitoring module, a data transmission and processing module, a state prediction module, a state correction module, and a platform monitoring module. The FSI solution module establishes a high-fidelity FSI numerical simulation model; the physical monitoring module collects core state data of the riser system; the data transmission and processing module processes and transmits the monitoring data in real time; the state prediction module establishes a state prediction proxy model; the state correction module compares and corrects the measured sensor data with the predicted data; and the platform monitoring module converts the corrected global vibration response into fatigue life assessment results. This addresses the problem that high-fidelity mechanism models cannot operate in real time, and that pure data model extrapolation is unreliable, leading to a trade-off between accuracy and efficiency in deep-water riser fatigue life prediction.

Description

Technical Field

[0001] This invention belongs to the field of marine riser system technology, and more specifically, relates to a hybrid-driven marine riser system fatigue life prediction system and method. Background Technology

[0002] If a deep-water riser fails, the platform must be shut down immediately or even evacuated. The economic and safety consequences have prompted the industry to upgrade "life prediction" from post-design assessment to real-time decision-making during service. Over the past decade, the "mechanism-data hybrid-driven" framework has been considered a feasible route to balance accuracy and efficiency: first, establish physical common sense with high-fidelity fluid-structure interaction (FSI) simulations, and then use deep learning to map local monitoring snapshots to the overall response, thereby replacing dense arrays with a small number of sensors. However, most existing hybrid frameworks remain in the "offline demonstration" stage—the mechanism and data modules only run sequentially in laboratory-level simulations or short-term sea trials. Once they enter long-term, variable-condition service scenarios, their inherent defects are rapidly amplified. First, the mechanism and data are still "two separate entities": the physical equations only provide samples in the early training stage, and the online phase relies entirely on a black-box network. When the velocity profile changes abruptly or the tension at the top of the riser changes due to platform drift, the network extrapolation becomes unstable, and the error amplifies exponentially. Secondly, the update mechanism is lacking: most solutions simplify "calibration" to periodic retraining, requiring the raw data to be transmitted back to the shore-based supercomputer. The training window is measured in weeks, making it impossible to capture the transient damage accumulation during extreme events such as typhoons. Thirdly, the hardware topology follows the "end-to-end floating" approach, with all sensors directly connected to the platform via long-distance analog cables. The number of cable cores increases linearly with the channel, leading to increased weight and changes in bending stiffness of the riser, which in turn interferes with the vortex-induced vibration modes being measured. Fourthly, the early warning logic is simplistic, relying solely on a binary judgment of "whether the cumulative damage exceeds the threshold," lacking continuous interpretation of the damage rate. On-site personnel still need to manually interpret the warnings after receiving them, missing the optimal intervention window. Summary of the Invention

[0003] In view of this, the present invention provides a hybrid-driven fatigue life prediction system and method for marine riser systems, which solves the problem that the accuracy and efficiency of deep-water riser fatigue life prediction cannot be simultaneously achieved due to the inability of high-fidelity mechanism models to run in real time and the unreliability of pure data model extrapolation.

[0004] This invention provides a hybrid-driven fatigue life prediction system for marine riser systems, comprising:

[0005] The fluid-structure interaction solution module is configured to establish a fluid-structure interaction mechanical model of the riser based on computational fluid dynamics and structural dynamics theory, and generate a high-fidelity numerical simulation dataset using a separate solution strategy and dynamic mesh technology.

[0006] The physical monitoring module, which includes a flow profiler and acceleration or strain sensors, is arranged in an array along the axis of the riser to acquire flow field and structural response information in real time.

[0007] The data transmission and processing module integrates wired and wireless transmission networks to compress, filter, and add timestamps to the raw signals output by the physical monitoring module in real time, forming a continuous monitoring data stream.

[0008] The state prediction module embeds a deep learning framework that uses the fluid-structure interaction mechanics model as a physical constraint. It is jointly trained using the numerical simulation dataset and historical monitoring data to output the predicted distribution of the global vortex-induced vibration response of the riser.

[0009] The state correction module receives the measured data for correction and compares the error with the predicted distribution. When the error exceeds a set threshold, it triggers an online learning algorithm to dynamically update the model parameters of the state prediction module and sends back the corrected global response.

[0010] The platform monitoring module, deployed on the offshore operation platform, performs fatigue damage calculations on the corrected global response and generates life assessment results, while also providing visualization, storage, and multi-level early warning output.

[0011] The technical effects of the hybrid-driven marine riser system fatigue life prediction system provided by this invention are as follows: By coupling a high-fidelity fluid-structure interaction mechanism model with a deep learning framework, the system embeds physical conservation laws into the data-driven channel, enabling global eddy-induced vibration prediction to maintain physical consistency even in the absence of full-scale measurement points; the online correction and life assessment closed-loop operation can instantly detect fatigue hotspots and trigger early warnings at the platform, significantly reducing the risk of misjudgment caused by model drift or extreme sea conditions, and realizing real-time and proactive riser life management.

[0012] Based on the above technical solution, the fatigue life prediction system for a hybrid-driven marine riser system of the present invention can be further improved as follows:

[0013] This includes both hardware and software subsystems:

[0014] The hardware subsystem includes an online platform host, an underwater offline host, and a sensor array for the physical monitoring module. The online platform host is responsible for data display and early warning, while the underwater offline host completes data preprocessing and wireless transmission.

[0015] The software subsystem encapsulates the underlying code and interactive interface of the fluid-structure interaction solution module, state prediction module, and state correction module, and supports physical model parameter configuration, training task scheduling, result export, and historical query.

[0016] The beneficial effects of adopting the above-mentioned improvement scheme are as follows: The hardware-software layered architecture moves the computationally intensive modeling and training tasks to on-site maintainable servers, while the underwater offline host responsible only for data collection and compression is moved to the seabed, which not only shortens the wireless transmission bandwidth requirements, but also facilitates remote iteration of algorithm versions on the platform; The unified software interface enables model updates, data export and querying to be completed in the same interactive interface, reducing cross-disciplinary collaboration costs and improving the efficiency of continuous system upgrades.

[0017] Furthermore, the fluid-structure interaction solution module further includes:

[0018] The flow field solution submodule is based on secondary development of an open-source fluid solution platform and completes the calculation of unsteady flow fields.

[0019] The structural solution submodule uses a self-developed structural dynamics solution kernel to realize the assembly of riser unit matrix and displacement solution.

[0020] The fluid-structure interaction submodule enables bidirectional coupling and transfer of flow field loads and structural displacements through a script interface, ensuring energy conservation at the coupling interface.

[0021] The beneficial effects of adopting the above-mentioned improvement scheme are as follows: By using the "loose coupling" method between the open-source fluid solver and the self-developed structural solver, bidirectional load-displacement iteration can be achieved by using script-level data exchange without deep intrusion into the original solution kernel. This retains the advantage of the open-source community continuously improving the flow field algorithm, and can quickly replace the structural constitutive or element type, so that the coupled framework has plug-and-play scalability for risers with different diameters, wall thicknesses or top tensions.

[0022] Furthermore, the velocity profiler in the physical monitoring module is vertically deployed along the key elevation of the riser to acquire the velocity-depth profile in real time; the acceleration or strain sensors are arranged in groups with multiple redundancy, of which at least one group is used to provide the true value to the state correction module, and the remaining groups are used to provide input to the state prediction module.

[0023] The beneficial effects of adopting the above-mentioned improved scheme are as follows: adopting dual-mode monitoring of "flow profiler + grouped sensors": the profiler provides real-time inflow conditions and provides instant calibration for the model boundary; the redundant grouping strategy dedicates some sensors to calibration, avoiding overfitting caused by using the same set of data for both training and validation, so as to maintain the reliability of the prediction-calibration chain even when the sensors gradually age or fail locally.

[0024] Furthermore, the wired transmission network of the data transmission and processing module adopts distributed composite optical fiber, which is laid in parallel with the riser; the wireless transmission network adopts a complementary link of underwater acoustic communication and radio frequency communication to realize cross-media data relay from the underwater offline host to the platform online host.

[0025] The beneficial effects of adopting the above-mentioned improved scheme are as follows: the fiber-optic-underwater acoustic-radio frequency hybrid link aggregates the high-sampling raw data nearby through the fiber optic cable and then uploads it wirelessly via frequency hopping, which reduces the attenuation of analog signals over long distances and avoids the deployment difficulties brought about by pure fiber optic through-platforms; the complementarity of underwater acoustic and radio frequency can automatically switch in scenarios with high wave height obstruction or radio frequency blind spots, ensuring that the monitoring flow is not interrupted under harsh sea conditions and providing continuous data raw materials for subsequent fatigue assessment.

[0026] This invention provides a method for predicting the fatigue life of a hybrid-driven marine riser system, comprising:

[0027] A fluid-structure interaction mechanical model for risers was established, and a separate solution strategy and dynamic mesh technology were adopted to generate high-fidelity numerical simulation datasets in batches.

[0028] Velocity profilers and arrays of acceleration or strain sensors are deployed at key locations on the riser to collect flow field and structural response data in real time, and the data stream is transmitted to the underwater offline host for preprocessing via a wired and wireless converged network.

[0029] The preprocessed data and the numerical simulation dataset are input into a deep learning framework with embedded physical constraints to train a surrogate prediction model for the global vortex-induced vibration response of the riser.

[0030] The measured correction data is compared with the prediction results to identify the error. When the error exceeds the standard, the online learning algorithm is activated to incrementally update the parameters of the surrogate model until the prediction accuracy meets the requirements.

[0031] The corrected global response is converted into fatigue damage indicators and the remaining life is further assessed. At the same time, the life curve is visualized, stored and graded early warning is realized on the work platform.

[0032] The technical advantages of the hybrid-driven fatigue life prediction method for marine riser systems provided by this invention are as follows: The method integrates "batch mechanism simulation - proxy training - online correction - life assessment" into an automated pipeline: In the simulation stage, virtual samples covering multiple sea conditions are generated at once, reducing the need for on-site measurements; in the training stage, physical equations are used to constrain network weights, ensuring that the extrapolation results conform to energy conservation; in the correction stage, closed-loop correction is performed using measured snapshots to form a self-evolving model; in the assessment stage, life curves are directly output, realizing a one-click operation of "data in, life out," significantly reducing human intervention and reliance on expert experience.

[0033] Based on the above technical solution, the fatigue life prediction method for a hybrid-driven marine riser system of the present invention can be further improved as follows:

[0034] Furthermore, the online learning algorithm employs a recursive least squares incremental update strategy, modifying only the weight parameters with the highest contribution in the proxy model to maintain existing knowledge and reduce computational load.

[0035] The beneficial effects of adopting the above-mentioned improved scheme are as follows: recursive least squares incremental update is used, and only the high-sensitivity parameters are fine-tuned. There is no need to retrain the entire deep network, and the update can be completed in seconds at the edge computing node of the platform. It not only maintains the existing knowledge memory, but also avoids the computing power consumption caused by full parameter reconvergence, so that transient disturbances such as ship swaying and short-term flow change can be quickly absorbed, and the continuous stability of the prediction output can be maintained.

[0036] Furthermore, the fatigue damage index is obtained by extracting the equivalent number of cycles using the rainflow counting method, and the life consumption rate of the riser under the current sea conditions is calculated by combining it with the linear cumulative damage rule.

[0037] The beneficial effects of adopting the above-mentioned improved scheme are as follows: by combining rainflow counting and linear cumulative damage, the complex irregular stress history is transformed into an equivalent cycle, which can be directly connected to the SN curve in the engineering design specifications without additional calibration; the method is not sensitive to the cycle order, and can still give a conservative and traceable damage estimate when there are intermittent or missing segments in the monitoring signal, ensuring that the life assessment results have standard consistency and auditability.

[0038] Furthermore, the tiered early warning system is divided into multiple threshold levels based on the lifespan consumption rate. When the consumption rate exceeds any threshold, an early warning message of the corresponding level is automatically generated and pushed to the designated terminal.

[0039] The beneficial effects of adopting the above-mentioned improvement scheme are as follows: the multi-level early warning maps the continuous life consumption rate to segmented thresholds. Once the threshold is crossed, different colors or sounds will be automatically triggered, and the information will be pushed to the duty room and the shore management center simultaneously. Since the threshold corresponds to the consumption rate rather than the absolute life, even if the riser design life varies, it can provide a unified and intuitive decision reference, making it convenient for the operator to arrange inspections or window replacements in advance and avoid sudden shutdowns.

[0040] Furthermore, after each online update, the platform records the parameter change trajectory and performance indicators of the agent prediction model, automatically generates a model update log, and supports backtracking queries.

[0041] The beneficial effects of adopting the above-mentioned improvement scheme are as follows: by recording the parameter trajectory and performance indicators before and after each update, a model family tree with timestamps is formed; when prediction anomalies occur in the future, it can be quickly rolled back to the previous reliable version and the mutation node can be located. At the same time, it provides a complete parameter tuning log for offline in-depth analysis, realizes the closed-loop improvement of "running and reviewing at the same time", extends the technical life cycle of the entire prediction system and reduces the risk of loss of operation and maintenance knowledge.

[0042] Compared with the prior art, the present invention has at least the following technical effects:

[0043] This solution uses "mechanism embedding in the network, correction embedding in the hardware, and early warning embedding in the process" as the overall innovation fulcrum, and redesigns a long-term system that can survive in the existing machine room of the drilling platform without the need for onshore supercomputing. Its core lies in allowing physical laws to run through the data lifecycle: During the training phase, each snapshot generated by the fluid-structure interaction solver comes with residuals of mass, momentum, and energy conservation, which are synchronously backpropagated as part of the loss function, so that the network weights satisfy the set of physically feasible solutions in addition to the mathematically optimal solution; During the online phase, the platform monitoring module writes the difference between the latest measured strain segment and the network output as a recursive least squares objective function, and only performs rank-one correction on the weights of high-sensitivity sub-layers. The update process is completed in seconds and does not require the return of the original data, thereby breaking the chain of "extreme events → model drift → damage misjudgment". In terms of topology, the system adopts a two-tier architecture of "underwater offline host - platform online host": the underwater host completes edge compression and timestamp alignment, and splits the data stream into two paths, "for prediction" and "for correction," through a complementary underwater acoustic-radio frequency link. The platform side only needs to maintain a bandwidth of hundreds of megabits, and the stiffness of the riser body is no longer affected by the increased weight of the transmission cable. On the software side, physical modeling, agent training, life assessment, and early warning push are encapsulated into orchestratable services. The operator can trigger "one-click model update" or "rollback to the previous trusted version" on the web page, reducing the operation threshold to the level of an ordinary instrument worker. The multi-level early warning engine no longer only looks at cumulative damage, but maps the damage rate, remaining life, and environmental severity to the same decision plane, automatically generating gradient suggestions of "continue observation - planned detection - immediate shutdown," and interfacing with the platform's SMS system to achieve a closed loop from data to command without human intervention. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the fatigue life prediction system for marine riser systems of the present invention;

[0045] Figure 2 This is a flowchart of the fatigue life prediction method for marine riser systems of the present invention;

[0046] Figure 3 This is the overall hardware layout diagram of the marine riser system fatigue life prediction system of the present invention;

[0047] Figure 4 This is a flowchart of the fluid-structure interaction solution for the marine riser system of this invention;

[0048] Figure 5 This is a flowchart of the marine riser system state prediction algorithm of the present invention;

[0049] Figure 6 This is a flowchart of the marine riser system state correction algorithm of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0051] like Figure 1 The diagram shown is a flowchart of a mechanism-data hybrid driven fatigue life prediction system for marine riser systems provided by this invention. It includes a fluid-structure interaction solution module, a physical monitoring module, a data transmission and processing module, a state prediction module, a state correction module, and a platform monitoring module, specifically comprising the following:

[0052] The fluid-structure interaction solution module includes a flow field solution submodule, a structure solution submodule, and a fluid-structure interaction submodule. The three submodules are described below:

[0053] The flow field solution submodule is a secondary development based on the OpenFOAM open source platform. It supports two methods of mesh generation: creating or importing external meshes. It can automatically select the turbulence model according to different Reynolds numbers and uses the mesh motion solver "displacementLaplacian". After setting up the example file, it is solved by the PimpleFoam incompressible Newtonian fluid transient solver integrated within OpenFOAM. Under the condition that the initial cylinder has no motion, the fully developed initial flow field is calculated, which lays the foundation for fluid-structure interaction solution.

[0054] The structural submodule is independently developed based on C++ Eigen and the Intel MKL library, enabling preprocessing, solving, and post-processing of structural fields. Preprocessing includes structural element meshing, section property setting, assembly of system element mass and stiffness matrices, and assembly of tensioner and soil stiffness matrices. The solver primarily uses statics based on the Newton-Raphson algorithm and Newmark-... β Dynamic solution; post-processing includes structural visualization and plotting of mechanical property curves, etc.

[0055] The interaction submodule is the core module of the entire solver, responsible for real-time communication. Leveraging OpenFOAM's ability to compile, load, and execute C++ code at runtime, it creates an interaction interface between the flow field and structure solutions. The interactive solution implementation scheme between multiple flow field computational domains is as follows: a Linux shell script is used to concatenate the interaction interfaces across multiple examples, enabling the flow field of each slice to be solved serially in time and in parallel in space. To ensure solution efficiency, typically one flow field slice corresponds to several structural elements to guarantee the continuity of the structural response.

[0056] The physical monitoring module collects real-time flow field environmental data and structural response data around the riser system using flow profilers and a monitoring sensor network array deployed at key locations on the riser. The monitoring sensors are divided into prediction sensors and correction sensors. Prediction sensors are deployed at the vortex-induced vibration mode nodes of the riser for model input feature extraction; correction sensors are deployed at the inverse nodes of the vortex-induced vibration modes of the riser for model prediction accuracy verification and correction.

[0057] The data transmission and processing module includes a wired transmission submodule and a wireless transmission submodule. The wired transmission submodule establishes a transmission link through distributed composite optical fibers to transmit monitoring data to the underwater offline host in real time; the wireless transmission submodule uses a combination of underwater acoustic communication and radio frequency communication to establish a transmission link between the underwater offline host and the platform's online host, realizing the transmission of monitoring data, prediction results, and model parameters.

[0058] The state prediction module integrates a fluid-structure interaction (FSI) mechanism model with a deep learning algorithm to construct a proxy model for predicting the fatigue life of marine riser systems. Through a physical information neural network architecture, the FSI control equations are embedded into the deep learning network as physical constraints, enabling accurate reconstruction from local monitoring data to the global vortex-induced vibration response of the riser system.

[0059] The state correction module compares the measured data from the correction sensor with the corresponding position state prediction data. When the error exceeds a preset threshold, it initiates an online incremental learning algorithm based on recursive least squares to dynamically adjust the surrogate model parameters in the state prediction module. The fine-tuned parameters are transmitted to the underwater offline host via a wireless transmission link for model updates, achieving online self-evolution of the prediction model.

[0060] The platform monitoring module is deployed on a cloud server cluster. It receives corrected global vortex-induced vibration response data of the marine riser system and performs fatigue life assessment using the rainflow counting method and Miner's linear cumulative damage rule. It enables data storage, analysis, visualization, and early warning management, and provides real-time monitoring and risk warnings for the riser system's operational status based on a multi-level early warning mechanism.

[0061] like Figure 2 The diagram shown is a flowchart of a mechanism-data hybrid-driven method for predicting the fatigue life of a marine riser system provided by this invention, which includes the following steps:

[0062] Step 1: Based on computational fluid dynamics and structural dynamics theories, a fluid-structure interaction (FSI) mechanical model of the marine riser system is established. A high-fidelity FSI numerical simulation model is constructed using a separate solution strategy and dynamic mesh technology. A large-scale numerical simulation dataset containing eddy-induced vibration response under various sea conditions is generated for subsequent model training and validation. The calculation process is as follows:

[0063] (1) The initial flow field is fully developed by applying the flow field solver, which lays the foundation for fluid-structure interaction solution;

[0064] (2) Static solution based on Newton-Raphson algorithm and Newmark- β The method solves the structural dynamics of the marine riser system and obtains the structural response;

[0065] (3) Select different sections along the axial direction of the marine riser and define the computational domains of the flow field and structure respectively. Complete the interactive solution between the flow field and structure in each computational domain to obtain the three-dimensional fluid-structure interaction solution of the marine riser system. Finally, carry out batch calculations to obtain a high-fidelity numerical simulation dataset.

[0066] Step 2: Deploy flow profilers and acceleration or strain sensor arrays at key locations on the riser to collect flow field and structural response data in real time, and transmit the data stream to the underwater offline host for preprocessing via a wired and wireless converged network.

[0067] Step 3: Input the preprocessed data and the numerical simulation dataset into a deep learning framework that embeds physical constraints, and train to obtain a surrogate prediction model for the global vortex-induced vibration response of the riser.

[0068] Step 4: Compare the actual calibration data with the prediction results to check the error. When the error exceeds the standard, start the online learning algorithm to incrementally update the surrogate model parameters until the prediction accuracy meets the requirements.

[0069] Step 5: Convert the corrected global response into fatigue damage indicators and further assess the remaining lifespan. Simultaneously, implement real-time visualization, storage, and tiered early warning of the lifespan curve at the operating platform. The fatigue damage calculation method for the marine riser system is as follows:

[0070] Post-processing was performed on the state response data of the deep-water drilling riser system, including displacement, rotation angle, bending moment, and stress. The Palmger-Miner theory was then used to calculate the cumulative fatigue damage of the riser system. If the damage accumulated to a value of 1, the structure was considered to have failed. The calculation method is as follows:

[0071] When any load is applied to the structure, the damage generated inside the structure is as follows:

[0072] ;

[0073] In the formula, D represents the damage generated inside the structure, and N represents the number of cycles that cause the structure to fail when the applied external force is S.

[0074] When the external force is cyclicated n times, the damage value produced on the structure is:

[0075] ;

[0076] external force The number of cycles required to cause structural failure can generally be determined from the SN curve.

[0077] The structural failure is determined according to Miner's linear cumulative damage rule. Without considering a safety factor, the structure is considered to fail when the damage accumulates to 1. Therefore, the failure criterion for the riser can be considered as follows:

[0078] ;

[0079] Please see Figure 3 The overall hardware layout of the marine riser system fatigue life prediction system of the present invention is introduced.

[0080] The fatigue life prediction system for marine riser systems includes hardware such as an online platform host, an underwater offline host, and monitoring sensors. The online platform host primarily houses the platform monitoring module, responsible for receiving monitoring signals transmitted from the underwater offline host and performing data processing, analysis, and display. The underwater offline host primarily houses the data transmission and processing module, responsible for converting the monitoring data from the physical monitoring module into wireless monitoring signals and sending them to the online platform host. Specifically:

[0081] The specific layout of the physical monitoring module is as follows: the flow profiler adopts an acoustic Doppler velocity profiler (ADCP), which is vertically arranged at 5-10 times the diameter upstream of the riser, and the measurement range covers the entire length of the riser; the monitoring sensor group is arranged at the vortex-induced vibration mode node position of the riser, including no less than two sets of acceleration or strain sensors, which are used for model input feature extraction and model prediction accuracy verification and correction.

[0082] The specific layout of the data transmission and processing module is as follows: the wired transmission network adopts a distributed composite optical fiber network, arranged along the riser axis, and transmits multiple sensor data simultaneously to ensure the real-time performance and integrity of the monitoring data; the wireless transmission network adopts a hybrid scheme combining underwater acoustic communication and radio frequency communication, consisting of three parts: an encoder, a modulator, and a transmitter. Among them, the encoder adopts LDPC encoding to improve the anti-interference capability of underwater transmission; the modulator adopts adaptive OFDM modulation, which dynamically adjusts the modulation order according to the channel conditions; the transmitter is based on MIMO technology, and achieves spatial diversity through a multi-element acoustic wave transmitter to establish a transmission link between the platform's online host and the underwater offline host.

[0083] Please see Figure 4 The fluid-structure interaction solution process for the marine riser system of this invention is introduced.

[0084] The fluid-structure interaction (FSI) solution employs a separate solution strategy. The flow field solution is based on the PimpleFoam solver on the OpenFOAM platform, while the structural solution is based on a self-developed C++ program. The FSI interaction is implemented through an interface built on Linux Shell. Within each time step, the flow field solver calculates the fluid forces acting on the riser surface and transmits them to the structural solver. The structural solver calculates the riser deformation and updates the mesh, iterating until convergence. The solution is performed sequentially, first fluid, then solid. Data exchange occurs between the flow field and structure within each time step. A sequential solution combined with relaxation iteration is used to achieve the required computational accuracy. The solution steps are as follows:

[0085] (1) Set the flow field and structural boundary conditions respectively, solve the initial flow field, and obtain the fluid physical quantities at time n. With structural physical quantities The fluid force applied to the structural boundary by the flow field is The structural displacement is At the fluid-solid interface, for variables such as load, temperature, heat flux, and displacement, the fluid domain and the solid domain should satisfy the basic principle of variable conservation, that is, they should satisfy the following coupling conditions:

[0086] ;

[0087] In the formula, This represents the fluid stress at the fluid-solid interface; Indicates the fluid normal at the fluid-solid interface; This represents the solid stress at the fluid-solid interface; Indicates the solid normal at the fluid-solid interface; This indicates the fluid displacement at the fluid-solid interface; This represents the solid displacement at the fluid-solid interface; This indicates the fluid heat flow rate at the fluid-solid interface; This represents the solid heat flow at the fluid-solid interface; Indicates the fluid temperature at the fluid-solid interface; This indicates the temperature of the solid at the fluid-solid interface.

[0088] (2) Based on the fluid force at time n Solve the governing equations of the marine riser system to obtain the displacement of the structural boundary center. The mesh movement velocity at the boundary is obtained by interpolating the moving mesh. The governing equation for the dynamic mesh of the flow field is the displacement Laplace diffusion equation, and the calculation method is as follows:

[0089] ;

[0090] In the formula, Represents the Laplace operator; The diffusion coefficient is denoted as . This represents the displacement of the grid nodes.

[0091] (3) Update the flow field mesh and adjust it according to the obtained boundary motion velocity. Based on the pressure-velocity coupling algorithm, a relaxation iteration factor is set to solve for the velocity of the flow field. and pressure Furthermore, the force exerted by the fluid on the structural boundary at time n+1 is obtained. .

[0092] (4) The force exerted on the structure by the fluid at time n+1 Import the structural mechanics governing equations, solve them repeatedly, and obtain the displacement of the structural boundary center at time n+1. .

[0093] (5) Solve sequentially to the final time step, integrate the flow field and structural physical quantities, and output the vortex-induced vibration response data of the marine riser system.

[0094] Please see Figure 5 The present invention introduces the marine riser system state prediction algorithm.

[0095] A response prediction model for the full-state coupled system of a drilling platform and deepwater drilling riser is established based on a gated recurrent neural network (GRU). The prediction model takes monitoring data (strain, displacement, rotation angle, etc.) from sensors as input and outputs displacement, rotation angle, bending moment, Mises stress, and force of the marine riser system, thus forming a state prediction algorithm for the marine riser system. Furthermore, the fatigue life of the marine riser system is calculated based on the rainflow counting method. The calculation process includes:

[0096] (1) A large number of simulation conditions were set up in three dimensions: ocean current, water depth and different system configurations. Sufficient calculations were carried out based on the fluid-structure interaction solution module to establish a simulation database of state response of marine riser system.

[0097] (2) Normalize the data in the system state response database to generate high-quality data that can be used to train the prediction model of the gated recurrent neural network.

[0098] Normalization aims to avoid the impact of varying data sizes on model accuracy, thereby improving the training speed and accuracy of the state prediction model. Min-Max Scaling is used to normalize the system state response database, linearly mapping the original data and transforming it to a specific range. The calculation formula is as follows:

[0099] ;

[0100] In the formula, Represents the original data. For the normalized data, This represents the minimum value of the original data. This represents the maximum value of the original data.

[0101] (3) Integrating the fluid-structure interaction mechanism model, a state prediction model for the marine riser system based on a gated recurrent neural network (GRU) is established, and the predicted state response parameters of the fully coupled drilling platform-deepwater drilling riser system are output. The update formulas for each gate structure in the GRU neural network at the current moment are:

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] In the formula, This represents the input information at time t. This represents the hidden state at time t-1. The hidden state contains data information from historical time points. This represents the hidden state passed to the next time step. Indicates the candidate hidden layer state. The weight matrix represents the updated gate state. The weight matrix represents the reset gate state. The weight matrix representing the hidden layer state. This represents the bias vector for updating the gate state. The bias vector represents the state of the reset gate. These represent the bias vectors of the hidden layer states. During the backpropagation of the loss value in the model training process, the weights and thresholds are updated using two different formulas; therefore, the weight matrix... , , Separately write as , , , , , Different matrices, This represents the weight matrix from the hidden layer state to the update gate and reset gate. This represents the weight matrix between the update gate and the input to the current time step. Other weight matrices are similar and will not be described in detail here. The activation function is the sigmoid function. For the hidden layer activation function, , , These represent the activation functions for the update gate, reset gate, and output gate, respectively.

[0108] The forward propagation steps during the training of the GRU neural network model are as follows:

[0109] Initialize the output of the hidden layer Initialize GRU model weights , , and bias terms , , The initial value;

[0110] Gated recurrent neural networks use the state values ​​of the hidden layers from the previous time step. Input information at the current moment Update the initial weights of the gates. and initial bias term An activation result is calculated to determine the amount of information to retain;

[0111] Reset the door based on the previous hidden state value. Input information at time t Reset the initial weight of the door and initial bias term Calculate the alternative values ​​that could be added to the hidden state at the current moment;

[0112] The hidden layer state at the current moment is determined by updating and resetting the gate state values ​​at the current moment, and the output value of the gated recurrent neural network at the current moment is determined based on the output results.

[0113] Please see Figure 6 The present invention introduces the state correction algorithm for marine riser systems.

[0114] The state correction algorithm employs an online incremental learning mechanism, using recursive least squares (RLS) to dynamically optimize model parameters. The specific calculation process includes:

[0115] (1) Real-time acquisition of monitoring data from calibration sensors, comparison with the output of the state prediction module, calculation of prediction error and determination of whether it exceeds the tolerance threshold. The error calculation adopts the relative error criterion:

[0116] ;

[0117] ;

[0118] In the formula, To calibrate the sensor The measured value at time [time]. Output values ​​for the state prediction model. and These are the upper and lower limits of the measurement range. The threshold is set to 5%-10% of the measured value, depending on the sensor accuracy and engineering requirements. When the error exceeds the threshold, the parameter update process is triggered.

[0119] (2) The recursive least squares method with a forgetting factor is used to adjust the weight parameters of the state prediction model online, thereby realizing the recursive least squares parameter update. The calculation method is as follows:

[0120] ;

[0121] ;

[0122] ;

[0123] In the formula, for k The model parameter estimation vector at time step; for k- The estimated vector of model parameters at time 1; for k The actual measured value of the sensor at any given time; The input feature vector contains monitoring data such as strain, tilt angle, and flow velocity. Forgetting factor (value range 0.95–0.99); for k Gain matrix at time step; It is the covariance matrix; for k- Covariance matrix at time 1; superscript T This represents the transpose of a vector.

[0124] (3) Rerun the state prediction model based on the corrected parameters, and calculate the matching degree between the corrected prediction results and the measured data. The calculation method is as follows:

[0125] ;

[0126] ;

[0127] In the formula, MAE is the mean absolute error. When MAE < 0.05 and If the value is greater than 0.95, the model correction is considered to have converged; otherwise, the parameter update is retried. N The number of samples; For the first i Measured values ​​of a sample; For the first i The model prediction value for each sample; It is the arithmetic mean of all measured values; The coefficient of determination.

[0128] (4) The updated model parameters are sent to the underwater offline host via a wireless transmission link. Differential coding technology is used to transmit only the parameter changes to reduce the amount of data transmission. After receiving the parameters, the underwater offline host performs a hot update of the model.

[0129] In summary, this invention, through the deep integration of mechanisms and data, the synergistic cooperation of hardware and software, and the organic combination of prediction and correction, has constructed a complete and reliable fatigue life prediction technology system for marine riser systems, providing important technical support for the safe development of deep-water oil and gas resources.

[0130] To better understand and implement this invention, the following is an example of a specific application scenario: A mechanism-data hybrid driven riser fatigue life prediction system is deployed on a deep-water semi-submersible platform at a depth of approximately 1,500 meters in a certain sea area. The riser is 3,000 meters long, with its top end connected to the platform's moon pool via a tensioner, and its bottom end connected to the subsea production tree. The system hardware includes three underwater offline hosts, one distributed composite optical fiber, twenty-four fiber Bragg grating strain sensors, one acoustic Doppler current profiler, one platform online host, and a set of underwater acoustic-radio frequency communication floats. The number and location of the sensors can be adjusted proportionally to the length of the riser's flexible section, satisfying the principle of "reducing specific numerical limitations."

[0131] During installation, the fiber optic strain sensors are arranged in a dual-redundancy configuration of "prediction group - correction group": four sensors are evenly distributed circumferentially at the third, seventh, and tenth flexible nodes of the riser, where vibration locking is most likely to occur, with two belonging to the prediction group and two to the correction group; the velocity profiler is vertically fixed inside the flow guide shroud on the outside of the riser, with a measurement range covering water depths from zero to 300 meters, outputting real-time velocity-direction profiles. The composite optical fiber is bundled parallel to the riser's outer armor, providing both power and transmitting the original spectral signal; the underwater offline host is installed ten meters above the riser's anti-bend device, running an embedded Linux system to perform spectral demodulation, strain calculation, timestamp alignment, and data compression, generating one JSON-formatted monitoring frame per second, with a buffer capacity that can cover 72 hours. The underwater acoustic-radio frequency (ORF) float maintains its relative position to the platform through ultra-short baseline positioning, with adaptive bandwidth switching to ensure that the link packet loss rate remains below an acceptable threshold even in sea state 8.

[0132] During the offline preparation phase, engineers at the shore-based workstation used an open-source fluid solver and a self-developed structural solver to generate high-fidelity snapshots covering common combinations of internal flow rates, top tensions, and velocity. This process took a total of one week. The output files were encrypted and then imported into the platform's online host via a supply ship. The platform's online host ran a physical information neural network training service, using the strain distribution of each snapshot as a label. Simultaneously, the Navier-Stokes equation residuals were written as differentiable loss terms and embedded into the network. After two hundred iterations, a lightweight proxy model was obtained, with a model size of less than 100 MB, enabling millisecond-level forward inference on a single GPU.

[0133] During online operation, the underwater offline host splits the latest monitoring frame into a "prediction stream" and a "correction stream" every second: the prediction stream is sent to the platform in real time via the underwater acoustic link, input into the surrogate model, and outputs the full-field strain envelope; the correction stream is temporarily stored locally and uploaded via RF burst every ten minutes. After receiving the correction packet, the platform monitoring module extracts the measured strain with the same degree of freedom as the prediction field and calculates the error vector. If the root mean square error exceeds the set threshold, the state correction module immediately starts recursive least squares, performing rank-one correction only on the last two layers of weights in the surrogate model. The update process is completed within ten seconds, with the new weights replacing the old model and recording the version number. The updated full-field strain is sent to the fatigue calculation engine, which uses rainflow counting to extract the equivalent loop, combines it with Miner's linear cumulative damage rule to output the current hourly damage increment, and adds it to the historical cumulative damage to obtain the remaining lifespan percentage. When the remaining lifespan is below the lower limit of the green zone, the platform SMS automatically pushes a "planned inspection" suggestion to the duty room; if it is below the orange zone, a "prepare for replacement" warning is triggered. At the same time, the latest model version, environmental parameters and damage curves are packaged into a PDF report and sent back to the shore-based management center via satellite link for engineers to review.

[0134] This embodiment has been operating continuously in this sea area for six months, experiencing two typhoons. During the typhoons, the current profiler recorded abrupt changes in flow direction, and the surrogate model immediately provided an increased strain envelope. The correction module then fine-tuned the peak region to match the measured values ​​using actual data, all without manual intervention. Based on the early warning, the platform completed the tensioner adjustment three days in advance, avoiding a potential fatigue over-limit maintenance. Through the combination of "few sensors + lightweight surrogate + online correction," this embodiment maintains prediction accuracy while compressing the traditional "weekly" assessment cycle requiring shore-based supercomputing support to "minutes," fully verifying the feasibility and engineering adaptability of the system and method.

[0135] 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 hybrid driven offshore riser system fatigue life prediction system, characterized by, Comprise: A fluid-structure coupling solving module configured to establish a riser fluid-structure coupling mechanical model based on computational fluid dynamics and structural dynamics theory, and generate a high-fidelity numerical simulation dataset based on a separation solving strategy and dynamic mesh technology, the fluid-structure coupling solving module further comprising: A flow field solving submodule for completing unsteady flow field calculation based on secondary development of an open-source fluid solving platform; A structure solving submodule for realizing riser unit matrix assembly and displacement solving by using a self-written structural dynamics solving kernel; A fluid-structure interaction submodule for realizing bidirectional coupling transmission of flow field load and structure displacement through a script interface, and ensuring energy conservation of the coupling interface; The fluid-structure coupling solving adopts a separation solving strategy, the flow field solving is based on a PimpleFoam solver of an OpenFOAM platform, the structure solving is based on a self-developed C++ program, and the fluid-structure interaction is realized through an interaction interface based on a Linux Shell, in each time step, the flow field solver is responsible for calculating fluid force acting on the riser surface and transmitting it to the structure solver, the structure solver is responsible for calculating the riser deformation and updating the mesh, and iterative calculation is performed until convergence, the calculation is sequentially solved in the order of fluid first and then solid, the flow field and the structure exchange data in each time step, and a sequential solving + relaxation iteration scheme is adopted to obtain the required calculation precision; A physical monitoring module comprising flow velocity profilers and acceleration or strain sensors arranged in an array along the riser axis for real-time acquisition of flow field and structure response information, the flow velocity profilers in the physical monitoring module are vertically arranged along the riser at key elevations to real-time acquire flow velocity-depth profiles; the acceleration or strain sensors are arranged in multiple redundant groups, at least one group of which is used to provide true values to the state correction module, and the remaining groups are used to provide inputs to the state prediction module; A data transmission and processing module integrated with wired and wireless transmission networks, which performs real-time compression, filtering and time stamping on the original signals output by the physical monitoring module to form a continuous monitoring data stream, the wired transmission network of the data transmission and processing module adopts a distributed composite optical fiber laid in parallel with the riser; the wireless transmission network adopts a complementary link of underwater acoustic communication and radio frequency communication to realize cross-medium data relay from an offline host underwater to an online host on the platform; A state prediction module embedded with a deep learning framework taking the fluid-structure coupling mechanical model as a physical constraint, which is trained jointly with the numerical simulation dataset and historical monitoring data to output a predicted distribution of the global vortex-induced vibration response of the riser; A drilling platform-deepwater riser full-state coupling system response prediction model is established based on a gated recurrent neural network, the prediction model takes monitoring sensor monitoring data as input, the monitoring sensor monitoring data includes strain, displacement and rotation angle, and takes riser system displacement, rotation angle, bending moment, Mises stress and force as output, thereby forming a riser system state prediction algorithm, and calculating the fatigue life of the riser system based on a rainflow counting method; The calculation process comprises: A large number of simulation conditions are set in three dimensions of current, water depth and system configuration, and sufficient calculation is carried out based on the fluid-solid coupling solving module to establish a simulation database of the state response of the marine riser system; The data in the system state response database is normalized to generate high-quality data that can be used to train the gated recurrent neural network prediction model; The fluid-solid coupling mechanism model is fused to establish a marine riser system state prediction model based on the gated recurrent neural network, and the prediction results of the state response parameters of the drilling platform-deep water drilling riser fully coupled system are output; The state correction module receives measured data for correction and compares the error with the prediction distribution. When the error exceeds the set threshold, the online learning algorithm is triggered to dynamically update the model parameters of the state prediction module, and the corrected global response is returned. The state correction algorithm adopts an online incremental learning mechanism and realizes dynamic optimization of model parameters based on the recursive least squares (RLS) method. The platform monitoring module is deployed on the offshore operation platform to perform fatigue damage calculation on the corrected global response and generate life assessment results, while providing visualization, storage and multi-level warning output.

2. A hybrid driven offshore riser system fatigue life prediction system according to claim 1, wherein, It also includes a hardware subsystem and a software subsystem: The hardware subsystem includes a platform online host, an underwater offline host and a sensor array of the physical monitoring module. The platform online host is responsible for data display and warning, and the underwater offline host completes data preprocessing and wireless transmission; The software subsystem encapsulates the underlying code and interactive interface of the fluid-solid coupling solving module, state prediction module and state correction module, and supports physical model parameter configuration, training task scheduling, result export and historical query.

3. A method of fatigue life prediction for a mixed drive offshore riser system, characterized by, A method for predicting the fatigue life of a hybrid-driven marine riser system as claimed in any one of claims 1-2, the method comprising the following steps: A riser fluid-solid coupling mechanics model is established and a separation solving strategy and dynamic mesh technology are used to generate a high-fidelity numerical simulation data set; Flow velocity profilers and acceleration or strain sensor arrays are arranged at key positions of the riser to collect real-time flow field and structural response data, and the data stream is transmitted to the underwater offline host for preprocessing through a wired and wireless fusion network; The preprocessed data and the numerical simulation data set are input into a deep learning framework embedded with physical constraints to train a proxy prediction model of the riser global vortex-induced vibration response; The measured correction data and the prediction results are compared for error, and when the error exceeds the standard, an online learning algorithm is started to incrementally update the proxy model parameters until the prediction accuracy meets the requirements. The corrected global response is converted into a fatigue damage indicator and further evaluated for remaining life, while real-time visualization, storage and hierarchical warning of the life curve are realized on the operation platform.

4. The method of predicting the fatigue life of a hybrid driven marine riser system according to claim 3, wherein, The online learning algorithm adopts a recursive least squares incremental update strategy, which only modifies the weight parameters with the highest contribution in the proxy model to maintain existing knowledge and reduce computational load.

5. A method of predicting the fatigue life of a hybrid driven marine riser system according to claim 4, wherein, The fatigue damage indicator extracts the equivalent cycle number through the rainflow counting method and calculates the life consumption rate of the riser under the current sea conditions by combining the linear cumulative damage rule.

6. A method of predicting the fatigue life of a hybrid driven marine riser system according to claim 5, wherein, The hierarchical early warning divides multiple threshold values according to the life consumption rate, and when the consumption rate crosses any threshold value, the early warning information of the corresponding level is automatically generated and pushed to the specified terminal.

7. The method of predicting the fatigue life of a hybrid driven marine riser system according to claim 6, wherein, The agent prediction model records the parameter change trajectory and performance indicators after each online update by the platform end, automatically generates a model update log and supports backtracking queries.

Images