Wellhead rapid fatigue analysis method and system based on physical information neural network, medium and equipment

Abstract

The invention relates to a wellhead rapid fatigue analysis method and system based on a physical information neural network, a medium and equipment. The method comprises the following steps: S1, inputting material characteristics, and predicting a temperature field of a shaft based on a first physical information neural network model; and S2, inputting a temperature field, and predicting a stress field based on the second physical information neural network model. And S3, collecting a stress field, and calculating based on the third physical information neural network model to obtain an accumulated fatigue damage value of the wellhead. And S4, calculating the remaining life of the wellhead according to the accumulated fatigue damage value and the service time of the wellhead. Based on the physical information neural network, the method enables the calculation efficiency of the fatigue life of the underwater wellhead to be high and accurate, can still maintain the high precision of the model under the condition of limited data volume, does not need to depend on a large amount of data to guarantee the accuracy of the model, considers the temperature effect, and enables the prediction result of the fatigue life of the underwater wellhead to be more accurate and closer to the reality.

Description

Technical Field

[0001] The present invention relates to the technical field of oil drilling and production engineering, and in particular to a wellhead rapid fatigue analysis method, system, medium and equipment based on physical information neural network. Background Art

[0002] Currently, deepwater oil and gas exploration has expanded from traditional deepwater areas to ultra-deepwater areas. As the core of deepwater energy development systems, the subsea wellhead assembly (SWE) consists of core components such as the casing head, wellhead carrier, casing hanger, and guide base. It performs multiple functions, including wellhead sealing, pressure control, and drilling equipment support, fulfilling the dual missions of pressure protection and fluid management. Its engineering reliability directly impacts the stable operation of the entire production system. In the marine environment, the SWE is dynamically connected to the floating platform via a riser. During drilling operations, it is subjected to cyclic dynamic loads caused by waves, ocean currents, platform drift, and the sway of the blowout preventer stack. Over long-term service (typically ≤20 years), these cyclic loads can easily induce cumulative fatigue damage in the wellhead system, leading to seal failure, structural fracture, and even major safety incidents such as blowouts. Current research focuses primarily on static performance, such as casing load capacity, wellhead stability, and contact surface sealing, while insufficient attention has been paid to fatigue life prediction. Moreover, most of the predictions for wellhead fatigue life focus on the use of SN curves, which require a large number of tests to obtain. When there is no available SN curve for the material, the difficulty of obtaining the fatigue life of the wellhead increases significantly. The relevant technologies have the following disadvantages:

[0003] 1. The calculation speed is slow. When calculating the fatigue life of underwater wellhead systems, traditional methods are often inefficient due to the limitations of calculation strategies. First, the time-domain dynamic analysis based on finite elements requires implicit integration of each load cycle, and a single cycle calculation takes a long time. Secondly, contact nonlinear problems (such as the bite surface of high and low pressure wellheads) need to be solved iteratively, and contact state determination and convergence control significantly increase the single-step time. In addition, in order to ensure the accuracy of local stress concentration areas such as welding defects, millimeter-level meshing is required. What's more serious is that traditional methods often require multiple rounds of parameter sensitivity studies (such as ocean current spectrum parameters, material damage thresholds, etc.), and repeated calculations further increase the time cost, making it difficult to meet the needs of rapid evaluation and optimization design in engineering practice.

[0004] 2. Requires a large amount of monitoring data. Traditional life prediction methods typically rely on extensive monitoring data to ensure model accuracy. First, methods based on statistical regression or physical models require data from multiple operating conditions throughout the equipment's lifecycle, including parameters such as temperature, pressure, and vibration, to construct a failure probability distribution or stress-life curve (SN curve). For example, the Weibull distribution requires historical failure data from similar equipment, while mechanical fatigue analysis requires high-precision strain history data. Second, the diverse nature of complex operating conditions requires diverse data. For example, marine equipment must cover varying flow rates, salinities, and load spectra, forcing long data acquisition cycles and high dimensionality. Furthermore, traditional methods often adopt a "data-driven" approach, relying on raw signals (such as vibration waveforms with thousands of points per second) rather than feature extraction, resulting in a significant increase in data storage and computational burdens. In the face of sudden failures or low-probability events, capturing abnormal features requires super-threshold monitoring, further increasing data volume. These factors create significant bottlenecks in data acquisition, storage, and processing for traditional methods, limiting their feasibility in terms of real-time performance and cost-effectiveness.

[0005] 3. A large number of experiments are required, and the economic efficiency is relatively poor. The fatigue life of the material requires destructive testing at multiple stress levels, which requires a long experimental cycle and a large amount of sample consumption. Secondly, complex equipment needs to consider the coupling effects of multiple factors, such as the interaction of temperature, load, corrosion, etc. In order to cover all working condition combinations, the experimental design often grows exponentially, greatly increasing time and economic costs. In addition, the simulation of extreme working conditions is limited. For example, the ultra-high temperature test of aircraft engines is difficult to fully reproduce the real scene, resulting in insufficient model extrapolation capabilities. The high threshold for obtaining experimental data restricts the universality of the method, especially for new materials or customized equipment. The high experimental cost forces a contradiction between prediction accuracy and feasibility.

[0006] 4. Temperature effects are rarely considered. Traditional wellhead life prediction methods typically fail to account for the impact of dynamic temperature changes, resulting in predictions that deviate from reality. First, high-temperature environments accelerate material degradation: metals are susceptible to creep, intergranular corrosion, and strength loss under long-term high temperatures. Traditional models often use room-temperature parameters, which underestimates the cumulative damage caused by thermal stress and thermal fatigue. For example, when a wellhead flange undergoes temperature cycling (such as temperature fluctuations caused by frequent injection and production), microcracks develop due to differences in thermal expansion coefficients. However, traditional methods only consider static pressure and cannot capture these dynamic failure mechanisms. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method, system, medium and equipment for rapid wellhead fatigue analysis based on physical information neural network in response to at least one of the above-mentioned defects.

[0008] The technical solution adopted by the present invention to solve the technical problem is: a fast wellhead fatigue analysis method based on physical information neural network, comprising the following steps:

[0009] S1. Input material properties and predict the temperature field of the wellbore based on the first physical information neural network model;

[0010] S2, inputting the temperature field and predicting the stress field based on the second physical information neural network model;

[0011] S3, collecting the stress field, and calculating the cumulative fatigue damage value of the wellhead based on the third physical information neural network model;

[0012] S4. Calculate the remaining life of the wellhead based on the accumulated fatigue damage value and the service time of the wellhead.

[0013] In some embodiments, the physical equations embedded in the loss functions of the first physical information neural network model, the second physical information neural network model, and the third physical information neural network model are different.

[0014] In some embodiments, step S1 includes:

[0015] Conduct mathematical modeling and discretization on the time and space solution domain of the temperature field;

[0016] Discretize the space-time solution domain into computational units and randomly extract internal points to generate a training coordinate set;

[0017] Set physical constraints;

[0018] configuring a first physical information neural network model;

[0019] Embed the heat transfer control equation into the loss function and construct the convergence criterion;

[0020] Adaptively adjust the weight coefficient according to the convergence state;

[0021] Perform hyperparameter tuning;

[0022] According to the constructed convergence criterion, determining whether the first physical information neural network model is less than a preset loss;

[0023] If so, the first physical information neural network model is deployed to the entire domain, and the material properties are input into the first physical information neural network model for prediction to output the temperature field of the wellbore.

[0024] In some embodiments, the step of embedding the heat transfer control equation into the loss function and constructing a convergence criterion includes:

[0025] The heat transfer control equation between the fluid in the drill string and the wellbore annulus is reconstructed as a residual term, and combined with the temperature gradient conservation and energy dissipation inequality to suppress the excessive sensitivity of the neural network to noisy data. Moreover, the residual term, initial condition loss and boundary condition loss are constructed into the convergence criterion.

[0026] In some embodiments, the residual term is:

[0027]

[0028] in,

[0029] Subscript p represents the drill pipe, subscript a represents the annulus, v is the velocity of the fluid at the wellhead, t is the time node, m; z is the well depth node, m; f p —Temperature residual in the drill pipe; f a —Temperature residual in the annulus; v p —Fluid velocity in drill pipe, m / s; T p —Drill pipe temperature, °C; T a —annulus temperature, °C; N b —spatial node at the drill bit; S b —Heat source term of drill bit, J; c l —Specific heat capacity of drilling fluid, J / (kg·℃); ρ1—Drilling fluid density, kg / m 3 ; A p —Drill pipe cross-sectional area, m 2 ; A a —Annulus cross-sectional area, m 2 ;v a —Flow rate of fluid in annulus, m / s; T e — initial formation temperature, °C; r pi —Inner diameter of drill pipe, m; r ci —Inner diameter of the annulus, m; U p —Drill pipe comprehensive heat transfer coefficient, W / (m 2 ℃); U a —Annular space comprehensive heat transfer coefficient, W / (m 2 ·℃).

[0030] In some embodiments, the initial condition loss is:

[0031]

[0032] Among them, L IC —Initial condition loss; N IC —Number of initial condition coordinate points; T' p (z i ,0)—the initial drill pipe temperature inferred by the model, ℃; T' a (zi ,0)—the annulus temperature at the initial moment inferred by the model, ℃.

[0033] In some embodiments, the boundary condition loss is:

[0034]

[0035] Among them, L BC —Boundary condition loss; N BC —Number of boundary coordinate points; T' p (0, t i )—wellhead drill pipe temperature at different times inferred by the model, ℃; T' p (h b , t i )—bottom hole drill pipe temperature at different times inferred by the model, ℃; T' a (h b , t i )—bottomhole annulus temperature at different times inferred by the model, ℃; T0—drilling fluid injection temperature, ℃; h b —Bottomhole node.

[0036] In some embodiments, step S2 includes:

[0037] Input the temperature field, which includes the temperature value of each point of the structure at each moment;

[0038] Initializing the second physical information neural network model;

[0039] Embed the elastic-plastic constitutive relation into the loss function, and decompose the loss function into a mechanical strain equation and a temperature strain equation, wherein the mechanical strain is constrained by a structural dynamics equation;

[0040] Performing a fully coupled solution on the mechanical strain equation and the temperature strain equation, and judging whether the fully coupled solution result converges according to a stress equilibrium condition and a no-slip boundary condition;

[0041] If so, output the stress field time series data.

[0042] In some embodiments, the stress balance condition is:

[0043] σ f n=-σ s ·n

[0044] Among them, σ f represents the stress tensor of the fluid at the interface; σ s represents the stress tensor of the solid at the interface; n is the unit normal vector of the interface;

[0045] and / or

[0046] The no-slip boundary condition is that the velocity of the fluid at the solid wall is zero.

[0047] In some embodiments, step S3 includes:

[0048] collecting the stress field and performing time-space grid alignment;

[0049] The actual fatigue parameters of the material are standardized to construct a spatiotemporal training sample set;

[0050] A third physical information neural network model is established, and a bidirectional LSTM network layer is used to extract the time series characteristics of the load sequence;

[0051] The damage evolution law equation is embedded in the loss function to simultaneously optimize the theoretical deviation between the data matching error and the Miner damage accumulation criterion through a customized loss function. The loss function contains a mean square error term and a physical regularization term based on the SN curve.

[0052] A hybrid data training strategy is adopted to fuse measured data with synthetic data generated by the damage evolution law equation in a preset ratio, and the network parameters are iteratively updated through an adaptive learning rate optimization algorithm;

[0053] The accumulated fatigue damage value of the wellhead is outputted through the trained third physical information neural network model.

[0054] In some embodiments, the damage evolution law equation is:

[0055]

[0056] Where σ is the stress tensor, ∈ is the strain tensor, ∈˙ is the strain rate, and T is the temperature. The specific form of the function f depends on the material constitutive model and damage mechanism.

[0057] In some embodiments, the damage evolution equation under uniaxial stress is:

[0058]

[0059] Where ω is the damage variable, M, χ, and x are material constants, and σ is the stress.

[0060] In some embodiments, the material properties include the specific heat capacity of the wellhead material (cement sheath), the heat transfer coefficient of the drilling fluid, the surface temperature, the geothermal gradient, the well depth, the wellhead diameter, and the diameter of the drill pipe.

[0061] In addition, the present invention also provides a wellhead rapid fatigue analysis system based on physical information neural network, comprising:

[0062] a temperature field prediction module for inputting material properties and predicting the temperature field of the wellbore based on the first physical information neural network model;

[0063] A stress field prediction module, configured to input a temperature field and predict a stress field based on a second physical information neural network model;

[0064] Fatigue prediction module, used to collect stress fields and calculate the cumulative fatigue damage value of the wellhead based on the third physical information neural network model;

[0065] The life assessment module is used to calculate the remaining life of the wellhead based on the accumulated fatigue damage value and the service time of the wellhead.

[0066] In addition, the present invention also provides a computer-readable storage medium storing a computer program, which is suitable for loading by a processor to execute the steps of the above-mentioned physical information neural network-based wellhead rapid fatigue analysis method.

[0067] In addition, the present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the steps of the above-mentioned physical information neural network-based wellhead rapid fatigue analysis method by calling the computer program stored in the memory.

[0068] The implementation of the physical information neural network-based wellhead rapid fatigue analysis method, system, medium and equipment of the present invention has the following beneficial effects: the present invention is based on the physical information neural network, which makes the calculation of underwater wellhead fatigue life efficient and accurate, and can maintain high precision of the model even when the amount of data is limited, without relying on a large amount of data to ensure the accuracy of the model. Taking into account the temperature effect, the underwater wellhead fatigue life prediction results are more accurate and closer to reality. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0070] Figure 1 1 is a flow chart of a method for rapid wellhead fatigue analysis based on a physical information neural network provided by an embodiment of the present invention;

[0071] Figure 2 1. It is a flow chart of the temperature field prediction step in the wellhead rapid fatigue analysis method based on physical information neural network in some embodiments of the present invention;

[0072] Figure 3 is a flow chart of the stress field prediction step in a fast wellhead fatigue analysis method based on a physical information neural network in some embodiments of the present invention;

[0073] Figure 4is a flow chart of the fatigue prediction step in a fast wellhead fatigue analysis method based on a physical information neural network in some embodiments of the present invention;

[0074] Figure 5 It is a structural diagram of a wellhead rapid fatigue analysis system based on physical information neural network provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0075] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, a detailed description of the specific embodiments of the present invention is now provided with reference to the accompanying drawings. It should be noted that, unless otherwise expressly specified or limited, terms such as "mounted," "connected," "connected," "fixed," and "disposed" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediary; internal connections between two elements, or interactions between two elements. When an element is referred to as being "on" or "under" another element, the element can be "directly" or "indirectly" located above the other element, or one or more intervening elements may be present. The terms "first," "second," and "third," etc., are used solely to facilitate the description of the present technical solution and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features indicated. Therefore, features designated as "first," "second," and "third," etc., may explicitly or implicitly include one or more of such features. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0076] In the following description, specific details such as particular system structures and techniques are provided for purposes of illustration, not limitation, to facilitate a thorough understanding of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.

[0077] refer to Figure 1 In a preferred embodiment, the physical information neural network-based wellhead rapid fatigue analysis method of this embodiment includes the following steps:

[0078] S1. Input material properties and predict the temperature field of the wellbore based on the first physical information neural network model. It is understood that material properties include but are not limited to the specific heat capacity of the wellhead material (i.e., cement sheath), the heat transfer coefficient of the drilling fluid, the surface temperature, the geothermal gradient, the well depth, the wellhead diameter, and the diameter of the drill pipe. In some embodiments, material properties may also include the cross-sectional area of ​​the gas production string, the specific heat capacity of the gas, the equivalent diameter of the pipe, the convective heat transfer coefficient of the inner wall of the drill pipe, the convective heat transfer coefficient of the outer wall of the drill pipe, the convective heat transfer coefficient of the inner wall of the casing, the convective heat transfer coefficient of the inner wall of the wellbore, the depth of the casing shoe, the well depth, the thermal conductivity of the drill pipe and casing, the thermal conductivity of the cement sheath, the rotation speed, the outer diameter of the annulus, the wellbore diameter, the Joule coefficient, the gas production string temperature, the drill bit torque, the fluid flow rate in the gas production string, the mechanical penetration rate, etc.

[0079] S2. Input the temperature field and predict the stress field based on the second physical information neural network model.

[0080] S3. Collect the stress field and calculate the cumulative fatigue damage value of the wellhead based on the third physical information neural network model.

[0081] S4. Calculate the remaining life of the wellhead based on the accumulated fatigue damage value and the service time of the wellhead.

[0082] It can be understood that the first physical information neural network model, the second physical information neural network model and the third physical information neural network model use the same neural network framework, but the physical equations embedded in the loss function are different.

[0083] In some embodiments, such as Figure 2 As shown, step S1 includes:

[0084] Mathematical modeling and discretization are performed on the time and space solution domain of the temperature field.

[0085] The space-time solution domain is discretized into computing units, and internal points are randomly selected to generate a training coordinate set.

[0086] Set physical constraints.

[0087] The first physical information neural network model is configured.

[0088] The heat transfer governing equations are embedded into the loss function, and convergence criteria are constructed.

[0089] The weight coefficient is adaptively adjusted according to the convergence state.

[0090] Perform hyperparameter tuning. This step adjusts deep learning hyperparameters, such as activation functions and learning rates. Specifically, hyperparameter tuning involves a first-stage parameter optimization (using the L-BFGS-B algorithm for global parameter optimization) and a second-stage parameter fine-tuning (using the Adam optimizer to accelerate local convergence). Hyperparameter tuning can help the model converge more effectively.

[0091] According to the constructed convergence criterion, it is determined whether the first physical information neural network model is less than a preset loss.

[0092] If so, the first physical information neural network model is deployed to the entire domain, and the material properties are input into the first physical information neural network model for prediction to output the temperature field of the wellbore.

[0093] Specifically, the steps of embedding the heat transfer control equation into the loss function and constructing the convergence criterion include:

[0094] The heat transfer governing equations between the fluid in the drill string and the wellbore annulus are reconstructed as residual terms, and combined with the temperature gradient conservation and energy dissipation inequalities to suppress the neural network's oversensitivity to noisy data. Moreover, the residual terms, initial condition loss, and boundary condition loss are constructed as convergence criteria.

[0095] It can be understood that regarding temperature field prediction, in the physics-driven neural network framework of the positive solution problem, the loss function serves as the key coupling interface between the control equation and the deep learning model. The convergence criterion is constructed by including three core error terms: the residual term of the control differential equation, the initial condition loss (initial state matching term), and the boundary condition loss (geometric boundary constraint term). For multi-region temperature field modeling of drilling conditions, the heat transfer control equations of the fluid in the drill string and the wellbore annulus are specifically reconstructed as residual terms. Combined with prior constraints such as temperature gradient conservation and energy dissipation inequality, the neural network's excessive sensitivity to noisy data, as well as boundary condition loss and initial condition loss, is suppressed, and adaptive optimization is performed to ultimately obtain the wellbore temperature field.

[0096] Specifically, the residual term of this embodiment is:

[0097]

[0098] in,

[0099] It can be understood that the subscript p represents the drill pipe, the subscript a represents the annulus, v is the velocity of the fluid at the wellhead, t is the time node, m, z is the well depth node, m. f p —Temperature residual in the drill pipe. f a —Temperature residual in the annulus. v p —Fluid velocity in drill pipe, m / s. T p —Drill pipe temperature, °C. Ta —Annulus temperature, °C. N b —Spatial node at the drill bit. S b —Heat source term of drill bit, J. c l —Specific heat capacity of drilling fluid, J / (kg·℃). ρ1—Drilling fluid density, kg / m 3 . A p —Drill pipe cross-sectional area, m 2 . A a —Annulus cross-sectional area, m 2 .v a —Flow rate of fluid in annulus, m / s. T e —The initial formation temperature, ℃. pi —Inner diameter of drill pipe, m. ci —Inner diameter of the annulus, m. U p —Drill pipe comprehensive heat transfer coefficient, W / (m 2 ℃). U a —Annular space comprehensive heat transfer coefficient, W / (m 2 ·℃).

[0100] The initial condition loss is:

[0101]

[0102] Among them, L IC —Initial condition loss. N IC —Number of initial condition coordinate points. T' p (z i ,0)—the initial drill pipe temperature inferred by the model, ℃. T' a (z i ,0)—the annulus temperature at the initial moment inferred by the model, ℃.

[0103] The boundary condition loss is:

[0104]

[0105] Among them, L BC —Boundary condition loss. N BC —Number of boundary coordinate points. T' p (0, t i )—wellhead drill pipe temperature at different times inferred by the model, ℃. T' p (h b , t i )—Drill pipe temperature at the bottom of the well at different times inferred by the model, ℃. T' a (h b , t i )—the bottom hole annulus temperature at different times inferred by the model, °C. T0—the drilling fluid injection temperature, °C. h b —Bottomhole node.

[0106] In some embodiments, such as Figure 3 As shown, step S2 includes:

[0107] Input the temperature field, which includes the temperature values ​​of each point on the structure at each moment.

[0108] Initialize the second physical information neural network model.

[0109] The elastic-plastic constitutive relation is embedded in the loss function, and the loss function is decomposed into mechanical strain equation and temperature strain equation. The mechanical strain is constrained by the structural dynamics equation.

[0110] The mechanical strain equation and the temperature strain equation are fully coupled and solved, and whether the fully coupled solution results converge is determined based on the stress equilibrium condition and the no-slip boundary condition.

[0111] If so, output the stress field time series data.

[0112] Specifically, the stress balance condition of this embodiment is:

[0113] σ f n=-σ s ·n

[0114] Among them, σ f Represents the stress tensor of the fluid at the interface. σ s Represents the stress tensor of the solid at the interface. n is the unit normal vector of the interface.

[0115] It should be noted that the no-slip boundary condition of this embodiment is that the velocity of the fluid at the solid wall is zero.

[0116] It can be understood that stress field prediction is an important part of wellhead fatigue analysis, which is responsible for converting temperature field data into stress field inside the structure, providing input for subsequent fatigue damage assessment. This step is based on the deep integration of physical information neural network (PINN) and thermoelasticity theory to achieve efficient and high-precision thermal stress solution, especially suitable for complex temperature fields and material nonlinear problems. The following describes its design principles and implementation process in detail: input data T (i.e. temperature field), including the temperature value of each point of the structure at each moment. Thermal stress calculation is based on the constitutive relationship and equilibrium equation of linear elastic thermodynamics. Its control equation is decomposed into strain caused by mechanical factors and strain caused by temperature. Mechanical strain is constrained by structural dynamics equations, and then the final output stress field time series data is obtained according to stress equilibrium conditions and boundary conditions. This embodiment can avoid the error accumulation of the traditional "heat first, then force" step-by-step method by performing a fully coupled solution, which is particularly suitable for strong nonlinear problems.

[0117] In some embodiments, such as Figure 4As shown, step S3 includes:

[0118] Acquire the stress field and perform space-time grid alignment.

[0119] The actual fatigue parameters of the material are standardized to construct a spatiotemporal training sample set.

[0120] A third physical information neural network model is established, and a bidirectional LSTM network layer is used to extract the timing characteristics of the load sequence.

[0121] The damage evolution law equation is embedded in a loss function to simultaneously optimize the data matching error and the theoretical deviation from the Miner damage accumulation criterion through a customized loss function. The loss function includes a mean square error term and a physical regularization term based on the SN curve. It can be understood that the damage evolution law equation predicts the structural lifespan by quantifying the propagation pattern of internal defects in the material.

[0122] A hybrid data training strategy is adopted to fuse the measured data with the synthetic data generated by the damage evolution law equation in a preset ratio, and the network parameters are iteratively updated through an adaptive learning rate optimization algorithm.

[0123] The accumulated fatigue damage value of the wellhead is output through the trained third physical information neural network model.

[0124] It can be understood that this step quantifies the cumulative damage caused by cyclic loading based on the obtained stress time history and material fatigue properties. The corresponding fatigue prediction module can include a stress acquisition module, a physical feature extraction layer, a deep neural network architecture, a physical constraint calculation unit, and a damage accumulation output layer. First, the stress time history is preprocessed to convert the temperature field T and the stress field σ eq Resample to a unified spatiotemporal grid to ensure the spatiotemporal consistency of damage calculation. Then, standardize and pre-process the actual fatigue parameters of the material to construct a training sample set containing time dimension features. Subsequently, a deep physical information neural network model is established, and a bidirectional LSTM network layer is used to extract the temporal characteristics of the load sequence. A physical constraint module is embedded before the output layer, and a customized loss function is used to simultaneously optimize the data matching error and the theoretical deviation of the Miner damage accumulation criterion, where the loss function contains a mean square error term and a physical regularization term based on the SN curve. A hybrid data training strategy is adopted to fuse the measured data and the synthetic data generated by the physical equation at a ratio of 1:4, and the network parameters are iteratively updated through an adaptive learning rate optimization algorithm. Finally, the cumulative fatigue damage value is directly output by the trained model. The cumulative fatigue damage value is combined with the service time to calculate the remaining life. This embodiment can use a five-fold cross-validation to ensure the generalization of the model, and set a physical consistency verification module to enforce the monotonically increasing damage characteristic.

[0125] It can be understood that the physical constraint module of this embodiment can be implemented by changing the physical equations in the temperature solution process to embed the physical control equations, namely the structural dynamics equations and the NS equations, which are well-known equations.

[0126] In some embodiments, the damage evolution law equation is:

[0127]

[0128] Where σ is the stress tensor, ∈ is the strain tensor, ∈˙ is the strain rate, and T is the temperature. The specific form of the function f depends on the material constitutive model and damage mechanism.

[0129] Specifically, the damage evolution equation under uniaxial stress is:

[0130]

[0131] Where ω is the damage variable, M, χ, and x are material constants, and σ is the stress.

[0132] This embodiment is based on a physical information neural network, which makes the calculation of underwater wellhead fatigue life efficient and accurate. It can maintain high precision of the model even when the amount of data is limited, without relying on large amounts of data to ensure the accuracy of the model. Taking into account the temperature effect, the prediction results of underwater wellhead fatigue life are more accurate and closer to reality.

[0133] refer to Figure 5 In another preferred embodiment, the physical information neural network-based wellhead rapid fatigue analysis system of this embodiment includes:

[0134] The temperature field prediction module is used to input material properties and predict the temperature field of the wellbore based on the first physical information neural network model.

[0135] The stress field prediction module is used to input the temperature field and predict the stress field based on the second physical information neural network model.

[0136] The fatigue prediction module is used to collect the stress field and calculate the cumulative fatigue damage value of the wellhead based on the third physical information neural network model.

[0137] The life assessment module is used to calculate the remaining life of the wellhead based on the accumulated fatigue damage value and the service time of the wellhead.

[0138] It is understandable that traditional FEA (Finite Element Analysis) numerical analysis methods rely on mesh discretization and iterative solutions, which consume a lot of computing resources. Compared with related art fatigue prediction methods, this embodiment can achieve the following technical effects:

[0139] 1. Innovatively introduce Physics-Informed Neural Networks (PINN), which directly embeds thermodynamic control equations (such as heat conduction equations and elastic-plastic constitutive relations) into the neural network loss function to achieve rapid solutions under the constraints of physical laws.

[0140] 2. High accuracy in multi-physics field coupling. Traditional multi-physics field analysis requires sequentially solving single physical fields, such as temperature and stress fields, and coupling between fields through data mapping. This leads to two core flaws: cross-field error propagation and redundant computational resources. This paper builds a field-synchronized solution framework based on a physical information neural network (PINN), integrating the heat conduction equation, elastic-plastic constitutive relations, and damage evolution law equations into the neural network loss function, thereby improving prediction accuracy.

[0141] 3. The wellhead temperature gradient distribution under transient working conditions can be captured. The present invention introduces the non-steady-state heat conduction equation as a neural network constraint term to construct a fully coupled solution framework of temperature-stress-deformation. This can capture the nonlinear change of the material thermal expansion coefficient with temperature and predict fatigue damage under complex high-temperature and high-pressure environments.

[0142] 4. Data requirements are relatively low. Purely data-driven neural networks rely on massive amounts of labeled samples and are subject to the risk of extrapolation failure, making them difficult to meet the needs of working condition prediction. This invention encodes mechanical prior knowledge into the network architecture and strengthens the physical connection between stress intensity factors and temperature fields through a hybrid loss function. This can maintain prediction confidence even under conditions of sample scarcity, breaking through the traditional proxy model's over-reliance on experimental data.

[0143] 5. The requirements for monitoring data are low. High-quality wellhead stress time history data can be obtained through the physical constraints of the physical information neural network, which is very economical.

[0144] In another preferred embodiment, the computer-readable storage medium of this embodiment stores a computer program suitable for loading by a processor to execute the steps of the physical information neural network-based rapid wellhead fatigue analysis method described in the above embodiment. Based on the physical information neural network, this embodiment achieves efficient and accurate calculation of underwater wellhead fatigue life. It maintains high model accuracy even with limited data, eliminating the need to rely on large amounts of data to ensure model accuracy. It also considers temperature effects, resulting in more accurate and realistic predictions of underwater wellhead fatigue life.

[0145] In another preferred embodiment, the computer device of this embodiment includes a memory and a processor, wherein the memory stores a computer program. The processor executes the steps of the physical information neural network-based rapid wellhead fatigue analysis method of the above-described embodiment by calling the computer program stored in the memory. Based on the physical information neural network, this embodiment achieves efficient and accurate calculation of underwater wellhead fatigue life. It can maintain high model accuracy even with limited data, eliminating the need to rely on large amounts of data to ensure model accuracy. It also considers temperature effects, resulting in more accurate and realistic predictions of underwater wellhead fatigue life.

[0146] As you can see, the SN curve is an important tool for evaluating the fatigue performance of materials under cyclic loading. SN curves generated from experimental data can be used to predict material fatigue life, assisting in engineering design and material selection. In the petroleum industry, SN curves are widely used to evaluate the fatigue performance of various equipment and structures, ensuring their safety and reliability in actual use.

[0147] The elastoplastic constitutive relationship is a mathematical model that describes the stress-strain relationship of a material in the elastic deformation and plastic deformation stages. It includes two main parts: an elastic part: describes the behavior of the material in the elastic deformation stage. A plastic part: describes the behavior of the material in the plastic deformation stage. The focus of this embodiment is not on the construction of the elastoplastic constitutive relationship itself, but on the application of the elastoplastic constitutive relationship, that is, the specific process of establishing the elastoplastic constitutive relationship can refer to the existing technology, which will not be repeated here. It should be noted that, with regard to the relevant equations mentioned in the embodiments of the present invention, other equations that do not provide a specific form can refer to the existing formulas, which will not be repeated here.

[0148] The computer-readable storage medium of the present invention can be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0149] The processor of the present invention is used to provide computing and control capabilities to support the operation of the entire device. It should be understood that in the embodiments of the present application, the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0150] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0151] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0152] It is understandable that the above embodiments only express the preferred implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the patent scope of the present invention. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present invention, the above technical features can be freely combined, and several deformations and improvements can be made, all of which fall within the scope of protection of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should fall within the scope of coverage of the claims of the present invention.

Claims

1. A fast wellhead fatigue analysis method based on physical information neural network, characterized in that: The following steps are involved: S1. Input material properties and predict the temperature field of the wellbore based on the first physical information neural network model; S2, inputting the temperature field and predicting the stress field based on the second physical information neural network model; S3, collecting the stress field, and calculating the cumulative fatigue damage value of the wellhead based on the third physical information neural network model; S4. Calculate the remaining life of the wellhead based on the accumulated fatigue damage value and the service time of the wellhead.

2. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 1 is characterized in that: The physical equations embedded in the loss functions of the first physical information neural network model, the second physical information neural network model and the third physical information neural network model are different.

3. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 1 is characterized in that: Step S1 includes: Conduct mathematical modeling and discretization on the time and space solution domain of the temperature field; Discretize the space-time solution domain into computational units and randomly extract internal points to generate a training coordinate set; Set physical constraints; configuring a first physical information neural network model; Embed the heat transfer control equation into the loss function and construct the convergence criterion; Adaptively adjust the weight coefficient according to the convergence state; Perform hyperparameter tuning; According to the constructed convergence criterion, determining whether the first physical information neural network model is less than a preset loss; If so, the first physical information neural network model is deployed to the entire domain, and the material properties are input into the first physical information neural network model for prediction to output the temperature field of the wellbore.

4. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 3 is characterized in that: The step of embedding the heat transfer control equation into the loss function and constructing a convergence criterion includes: The heat transfer control equation between the fluid in the drill string and the wellbore annulus is reconstructed as a residual term, and combined with the temperature gradient conservation and energy dissipation inequality to suppress the excessive sensitivity of the neural network to noisy data. Moreover, the residual term, initial condition loss and boundary condition loss are constructed into the convergence criterion.

5. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 4 is characterized in that: The residual term is: in, Subscript p represents the drill pipe, subscript a represents the annulus, v is the velocity of the fluid at the wellhead, t is the time node, m; z is the well depth node, m; f p —Temperature residual in the drill pipe; f a —Temperature residual in the annulus; v p —Fluid velocity in drill pipe, m / s; T p —Drill pipe temperature, °C; T a —annulus temperature, °C; N b —spatial node at the drill bit; S b —Heat source term of drill bit, J; c l —Specific heat capacity of drilling fluid, J / (kg·℃); ρ1—Drilling fluid density, kg / m 3 ; A p —Drill pipe cross-sectional area, m 2 ; A a —Annulus cross-sectional area, m 2 ;v a —Flow rate of fluid in annulus, m / s; T e — initial formation temperature, °C; r pi —Inner diameter of drill pipe, m; r ci —Inner diameter of the annulus, m; U p —Drill pipe comprehensive heat transfer coefficient, W / (m 2 ℃); U a —Annular space comprehensive heat transfer coefficient, W / (m 2 ·℃).

6. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 5 is characterized in that: The initial condition loss is: Among them, L IC —Initial condition loss; N IC —Number of initial condition coordinate points; T' p (z i ,0)—the initial drill pipe temperature inferred by the model, ℃; T' a (z i ,0)—the annulus temperature at the initial moment inferred by the model, ℃.

7. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 6 is characterized in that: The boundary condition loss is: Among them, L BC —Boundary condition loss; N BC —Number of boundary coordinate points; T' p (0, t i )—wellhead drill pipe temperature at different times inferred by the model, ℃; T' p (h b , t i )—bottom hole drill pipe temperature at different times inferred by the model, ℃; T' a (h b , t i )—bottomhole annulus temperature at different times inferred by the model, ℃; T0—drilling fluid injection temperature, ℃; h b —Bottomhole node.

8. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 1 is characterized in that: Step S2 includes: Input the temperature field, which includes the temperature value of each point of the structure at each moment; Initializing the second physical information neural network model; Embed the elastic-plastic constitutive relation into the loss function, and decompose the loss function into a mechanical strain equation and a temperature strain equation, wherein the mechanical strain is constrained by a structural dynamics equation; Performing a fully coupled solution on the mechanical strain equation and the temperature strain equation, and judging whether the fully coupled solution result converges according to a stress equilibrium condition and a no-slip boundary condition; If so, output the stress field time series data.

9. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 8 is characterized in that: The stress equilibrium condition is: s f ·n=-σ s ·n Among them, σ f represents the stress tensor of the fluid at the interface; σ s represents the stress tensor of the solid at the interface; n is the unit normal vector of the interface; and / or The no-slip boundary condition is that the velocity of the fluid at the solid wall is zero.

10. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 1, characterized in that: Step S3 includes: collecting the stress field and performing time-space grid alignment; The actual fatigue parameters of the material are standardized to construct a spatiotemporal training sample set; A third physical information neural network model is established, and a bidirectional LSTM network layer is used to extract the time series characteristics of the load sequence; The damage evolution law equation is embedded in the loss function to simultaneously optimize the theoretical deviation between the data matching error and the Miner damage accumulation criterion through a customized loss function. The loss function contains a mean square error term and a physical regularization term based on the SN curve. A hybrid data training strategy is adopted to fuse measured data with synthetic data generated by the damage evolution law equation in a preset ratio, and the network parameters are iteratively updated through an adaptive learning rate optimization algorithm; The accumulated fatigue damage value of the wellhead is outputted through the trained third physical information neural network model.

11. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 10, characterized in that: The damage evolution law equation is: Where σ is the stress tensor, ∈ is the strain tensor, ∈˙ is the strain rate, and T is the temperature. The specific form of the function f depends on the material constitutive model and damage mechanism.

12. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 11, characterized in that: The damage evolution equation under uniaxial stress is: Where ω is the damage variable, M, χ, and x are material constants, and σ is the stress.

13. The method for rapid wellhead fatigue analysis based on physical information neural network according to claim 1, characterized in that: The material properties include the specific heat capacity of the wellhead material (cement sheath), the heat transfer coefficient of the drilling fluid, the surface temperature, the geothermal gradient, the well depth, the wellhead diameter and the diameter of the drill pipe.

14. A wellhead rapid fatigue analysis system based on physical information neural network, characterized in that: include: a temperature field prediction module for inputting material properties and predicting the temperature field of the wellbore based on the first physical information neural network model; A stress field prediction module, configured to input a temperature field and predict a stress field based on a second physical information neural network model; Fatigue prediction module, used to collect stress fields and calculate the cumulative fatigue damage value of the wellhead based on the third physical information neural network model; The life assessment module is used to calculate the remaining life of the wellhead based on the accumulated fatigue damage value and the service time of the wellhead.

15. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which is suitable for being loaded by a processor to execute the steps of the wellhead rapid fatigue analysis method based on physical information neural network according to any one of claims 1 to 13.

16. A computer device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the steps of the method for rapid wellhead fatigue analysis based on physical information neural network as described in any one of claims 1 to 13 by calling the computer program stored in the memory.

Images