A deep-sea riser fatigue life prediction method based on physical information driving

Abstract

The application provides a deep-sea riser fatigue life prediction method based on physical information driving, which comprises the following steps: 1) collecting original data of riser fatigue test; 2) calculating physical information parameters of the riser; 3) taking material performance, riser geometric parameters, crack geometric parameters, stress and strain, load parameters and crack fatigue physical information parameters as model inputs, taking crack propagation rate as model output, carrying out normalization and dimensionless treatment, and constituting a deep-sea riser characteristic data set; 4) carrying out data segmentation and recombination based on time sequence characteristics of data, and building a long short-term memory deep learning model containing physical constraints based on characteristics of crack propagation; 5) obtaining an optimal physical information driven deep learning model; 6) carrying out fatigue life prediction through the optimal physical information driven model, and carrying out uncertainty evaluation. The application can realize probability prediction and uncertainty quantification of the riser fatigue life, and can provide a reference for reliability evaluation of the in-service riser structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ocean engineering, and particularly relates to a deep-sea riser fatigue life prediction method driven by physical information. BACKGROUND

[0002] Deep sea is an important way to transport oil and gas in current deepwater oil and gas engineering. Due to the complexity of deep sea environment, the riser is subjected to cyclic stress and produces fatigue damage, and the girth weld of the pipeline is prone to cracking. The current common fatigue life prediction methods include experimental method, numerical simulation method and data-driven method. The experimental method includes standard part experiment and full-size experiment, and the core is to fit the fatigue damage model based on physical quantities such as strain, stress or energy. The numerical simulation method focuses on simulating the physical state of the pipeline through numerical methods and boundary conditions, and connecting S-N curve to predict the fatigue life. Although the above methods have been successfully applied to the research of riser fatigue, they still have some defects and deficiencies, mainly including: (1) The theoretical method needs a large number of physical assumptions, which greatly reduces the accuracy of reflecting the actual physical phenomenon. (2) The numerical simulation method cannot fully reflect the real changes of the model due to its own limitations. In actual application, the interaction between factors such as size, material, manufacturing process and operating environment makes the structure safety uncertain.

[0003] Unlike experimental and numerical methods that focus on causal reasoning, data-driven methods focus on revealing the correlation within the data. However, it needs to be noted that since data-driven models are essentially black-box systems, their prediction performance is heavily dependent on the amount and quality of available data. The high cost of fatigue testing and the scarcity of data points, combined with the limited finite element data that has not yet reached high-fidelity accuracy, pose challenges for training powerful data-driven models. Chinese patent CN115081321B discloses a corrosion fatigue life prediction method for marine welded structures. The patent realizes the life prediction of marine welded structures by combining extreme gradient boosting algorithm and deep convolutional neural network through edge oversampling method for data enhancement. However, due to the lack of crack description in the input features of the data, and the lack of consideration of the influence of parameter uncertainty, there is a great limitation.

[0004] To solve the problems existing in the data-driven method, physical information and physical constraints need to be introduced into the model to improve the accuracy and explainability of the model and reduce the dependence of the model on high-cost data. Chinese Patent CN117993304B discloses a physical information driven metal material notch fatigue life machine learning prediction method, which adds notch geometric parameters and physical information parameters in the input features of the data, effectively building a fatigue life machine learning model with strong generalization ability. However, all the data selected by the model come from standard part experiments, and there is a certain deviation between standard part experiments and full-size experiments, especially under high stress. Therefore, the model has certain limitations.

[0005] Machine learning has been widely used in the field of deep-sea riser fatigue life prediction, but there are few reports on the fatigue life prediction of girth welds under complex stress conditions. On the other hand, a unified life model for multiple sizes and wide life ranges has not been established, and the physical information driven machine learning model provides a new idea for constructing a unified life prediction model. SUMMARY

[0006] In view of the technical problems in the above background art, the present application provides a deep-sea riser fatigue life prediction method based on physical information driving, which is reasonable in concept, can evaluate the crack propagation rate of girth welds under multiple sizes and multiple stress conditions, can realize the probability prediction and uncertainty quantification of riser fatigue life, and can provide a reference for the reliability evaluation of in-service riser structures.

[0007] To solve the above technical problems, the present application provides a deep-sea riser fatigue life prediction method based on physical information driving, which mainly includes the following steps:

[0008] 1) Collecting original data of riser fatigue test, including material performance, riser geometric parameters, crack geometric parameters, stress and strain, load parameters and crack propagation rate;

[0009] 2) Calculating the physical information parameters of the riser

[0010] The maximum circumferential stress and generalized strain at the crack of the riser are obtained by analytical method, and the damage parameter at the crack of the riser is further obtained;

[0011] 3) Taking the material performance, riser geometric parameters, crack geometric parameters, stress and strain, load parameters and crack fatigue physical information parameters as model inputs, and taking the crack propagation rate as model output, the original data of the riser fatigue test in the above step 1) and the physical information parameters of the riser in the above step 2) are normalized and dimensionless processed, to form a deep-sea riser feature data set;

[0012] 4) Based on the temporal characteristics of riser fatigue test data, the data is segmented and reorganized. The riser fatigue test dataset is divided into training set and test set in a 7:3 ratio. A long short memory deep learning model with physical constraints is built based on the characteristics of crack propagation.

[0013] 5) Train the long short-term memory deep learning model using the training set in the riser fatigue test dataset, obtain the optimal parameters of the long short-term memory deep learning model through the Bayesian optimization method, and compare different models to obtain the optimal physical information-driven deep learning model.

[0014] 6) Fatigue life prediction is performed by driving a deep learning model with optimal physical information, and uncertainty assessment is conducted.

[0015] The physical information-driven fatigue life prediction method for deep-sea risers, wherein the material properties in step 1) include elastic modulus E, Poisson's ratio v, and yield strength σ. limit The geometric parameters include two parts: riser geometric parameters and crack geometric parameters. The riser geometric parameters include inner diameter d, outer diameter D, wall thickness t, and pipe length L. The crack geometric parameters include crack depth a and crack width c. The stress and strain parameters include the maximum principal stress σ. max hydrostatic pressure σ m Von Mises stress σ eq Strain range Δε t And stress intensity factor K, load parameters including stress ratio R σ and frequency f.

[0016] The aforementioned physical information-driven method for predicting the fatigue life of deep-sea risers, in step 2), assuming the riser crack propagates radially, then the circumferential stress σ at the riser crack... rmax Represented by stress intensity factor:

[0017]

[0018] In equation (1) above, K is the force intensity factor and r is the crack length;

[0019] Generalized strain ε at the riser crack f Using the stress triaxiality of materials and critical strain ε o To indicate:

[0020]

[0021] In equation (2) above, For hydrostatic pressure; σ eq The stress is Von Mises; σ1, σ2, and σ3 are the three components of the principal stress, respectively, and ε oThe critical strain of the initial crack of the material.

[0022] The deep-sea riser fatigue life prediction method based on physical information driving, wherein in the step 2), the damage parameter at the riser crack is obtained by the following formula (3):

[0023] SWT=σ rmax Δε t / 2 (3);

[0024] In the above formula (3), Δε t is the total strain range around the girth weld.

[0025] The deep-sea riser fatigue life prediction method based on physical information driving, wherein in the step 3), the material performance is dimensionless and normalized by using the material performance parameter of the material at room temperature:

[0026]

[0027] In the above formula (4), M is the performance parameter of the material before dimensionless and normalization; M d is the performance parameter of the material after dimensionless and normalization; M s is the material performance parameter at room temperature.

[0028] In the step 3), the riser geometric parameter is normalized by using the size of the standard pipeline, and the crack geometric parameter is normalized by using the standard threshold value:

[0029]

[0030] In the above formula (5), G and G d are the geometric parameters before and after dimensionless and normalization, G th contains the riser geometric parameter and the crack geometric parameter, and G d represents the threshold value for dimensionless and normalization, which is selected according to the size of the standard pipeline and the measured maximum value of the crack fatigue test.

[0031] In the step 3), the stress-strain parameter is normalized by using the standard threshold value:

[0032]

[0033] In the above formula (6), φ is the stress-strain parameter before dimensionless and normalization; φ d is the stress-strain parameter after dimensionless and normalization; φ th represents the threshold value for dimensionless and normalization, which is selected according to the measured maximum value of the crack fatigue test.

[0034] In the step 3), the load parameter is normalized by using the following formula (7):

[0035]

[0036] R in the above formula (7) σ represents a stress ratio, represents a dimensionless and normalized stress ratio, f represents a load frequency, and f d represents a dimensionless and normalized frequency, f th is a threshold value specified by a riser fatigue test standard;

[0037] The physical information parameter in the step (3) is normalized by the following formula (8):

[0038]

[0039] σ in the above formula (8) rmax represents a maximum hoop stress, represents a dimensionless and normalized maximum hoop stress, ε f represents a generalized strain, ε d represents a dimensionless and normalized strain, SWT represents a damage parameter at a crack, and SWT d represents a dimensionless and normalized damage parameter; σ th , ε th , and SWT th is selected according to the maximum value in the analytical calculation result;

[0040] The crack propagation rate in the step 3) is normalized by the following formula (9):

[0041]

[0042] CGR in the above formula (9) represents a crack propagation rate, CGR d represents a dimensionless and normalized crack propagation rate, CGR th is selected according to the maximum value of the experimental data.

[0043] The deep-sea riser fatigue life prediction method based on physical information driving, wherein the data segmentation and reorganization based on the time sequence characteristics of the riser fatigue test data in the step 4) is to first divide each subset into a training set and a test set by using a stratified sampling method; then, a physical constraint is introduced into a loss function of a long short-term memory deep learning model to improve the interpretability of the model, and the loss function L(θ) is composed of two parts, the first part is the deviation between the predicted value and the true value, and the second part is the deviation of the predicted result under the physical constraint, which can be specifically represented as:

[0044] L(θ)=L d (θ)+λL p(θ) (10) ;

[0045] L in the above formula (10) d (θ) represents the deviation of the predicted value from the true value; L p (θ) represents the deviation of the predicted value from the physical law; λ is a penalty factor for balancing the influence of the two parts on the loss function; L d (θ) is calculated by RMSE, L p (θ) is determined by calculating the derivative of the crack propagation rate;

[0046] The crack propagation process is divided into three stages: initiation stage, stable stage and rapid expansion stage; L p (θ) expression is:

[0047]

[0048] In the above formula (11), N represents the number of cycles, a represents the crack depth, and I(A>B) is a state indicator function, which is 1 when A is greater than B, indicating that the physical constraint is violated, otherwise it is equal to 0, indicating that it is consistent with the physical constraint.

[0049] The deep-sea riser fatigue life prediction method based on physical information driving, wherein the specific process of step 5) is as follows:

[0050] 5.1) Select 6 cross-validation R 2 The average value As the optimization target function, determine the optimization range of the number of layers N layer , the number of units N units and the learning rate η of the long and short memory deep learning model, combine Bayesian optimization with K-fold cross-validation to obtain the best parameters of the long and short memory deep learning model, and the selected target function in the optimization process is the average value of K cross-validation R 2

[0051]

[0052] In the above formula (12), R i 2 is the determination coefficient of the i-th cross-validation; is the average value of K cross-validation.

[0053] 5.2) Select three key hyperparameters from the long and short memory deep learning model, i.e. the number of layers N layer , the number of units N units and the learning rate η, and use the Bayesian optimization method to obtain the final hyperparameter optimization result;

[0054] ​5.3) Compare the prediction results of the optimal physical information driven deep learning model with those of the traditional model to obtain the optimal model.

[0055] The deep-sea riser fatigue life prediction method based on physical information driving, wherein the specific process of step 6) is as follows: the trained physical information driven deep learning model is used for fatigue life prediction, the prediction results of the deep learning model without introducing physical parameters and the traditional physical fatigue life model are compared, and the life uncertainty is quantitatively evaluated.

[0056] The deep-sea riser fatigue life prediction method based on physical information driving, wherein the specific steps of quantitatively evaluating the life uncertainty are as follows:

[0057] 6.1) Sample the crack depth a, crack width c, maximum principal stress σ max , and strain range Δε t in the deep-sea riser feature data set by Latin hypercube sampling to obtain sufficient samples; the sampling range is limited to the first 80% of the deep-sea riser feature data set; it is assumed that the first 80% of the data in the deep-sea riser feature data set is subject to normal distribution, wherein the experimental measurement value represents the mean value, and the standard deviation is determined by the measurement error;

[0058] 6.2) The data samples in the first 80% of the deep-sea riser feature data set are segmented to meet the input form of the physical information driven deep learning model, and then the trained physical information driven deep learning model is used for prediction to obtain corresponding crack propagation data;

[0059] 6.3) The riser crack depth is calculated according to the crack propagation data, and the calculation formula of the crack depth is as follows:

[0060]

[0061] In the above formula (13), a i represents the crack depth at time i, a i+1 represents the crack depth at time i+1, represents the crack propagation rate at time i, and dN represents the cycle number;

[0062] 6.4) The cycle number N when the crack depth reaches the crack depth measured by the experiment is calculated according to the failure crack depth as the criterion, and this N is the fatigue life;

[0063] 6.5) The P-S-N curve is drawn according to the fatigue life result to realize quantitative evaluation of the uncertainty.

[0064] By adopting the above technical scheme, the present application has the following beneficial effects:

[0065] The physical information driven deep sea riser fatigue life prediction method has reasonable conception, the physical information and physical constraint are introduced to improve the accuracy and interpretability of the model, the problems of poor interpretability and poor robustness of the pure data driven deep learning model are solved, the well trained physical information driven deep learning model can evaluate the crack propagation rate of the girth weld under multiple size and multiple stress conditions, has good extrapolation ability and robustness, the LHS and the physical information driven deep learning model are connected, the probability prediction and uncertainty quantification of the riser fatigue life can be realized, and the reliability evaluation of the in-service riser structure can be provided. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0067] Figure 1 The flow chart of the physical information driven deep sea riser fatigue life prediction method of the present application;

[0068] Figure 2 The prediction result graph of the data driven model used in the physical information driven deep sea riser fatigue life prediction method of the present application;

[0069] Figure 3 The prediction result graph of the physical information driven model used in the physical information driven deep sea riser fatigue life prediction method of the present application;

[0070] Figure 4 The SCR riser P-S-N curve graph drawn according to the physical information driven model result in the physical information driven deep sea riser fatigue life prediction method of the present application;

[0071] Figure 5 The uncertainty quantification graph drawn according to the physical information driven model result in the physical information driven deep sea riser fatigue life prediction method of the present application;

[0072] Figure 6 The S-N curve graph drawn according to the physical information driven model result and the industry standard in the physical information driven deep sea riser fatigue life prediction method of the present application. DETAILED DESCRIPTION

[0073] The technical solutions of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0074] The present application will be further explained and described below in conjunction with specific embodiments.

[0075] As shown in the drawings, the present embodiment provides a deep-sea riser fatigue life prediction method based on physical information driving, taking API X65 SCR fatigue life test as an example to evaluate the SCR (crack propagation rate) fatigue life, including the following steps: Figure 1

[0076] S100, collect the original data of SCR riser full-size, small-size and standard fatigue test, including material performance, riser geometric parameters, crack geometric parameters, stress and strain, load parameters and crack propagation rate; wherein the original data are collected from the related experiments of "Deep-sea No. 1" sub-project "deepwater riser test test system construction" and the published literature and material manual at home and abroad. The aforementioned material performance includes elastic modulus E, Poisson's ratio v, yield strength limit σ limit , geometric parameters include riser geometric parameters and crack geometric parameters, riser geometric parameters include inner diameter d, outer diameter D, wall thickness t and pipe length L, crack geometric parameters include crack depth a and crack width c, stress and strain include maximum principal stress σ max , hydrostatic pressure σ m , Von Mises stress σ eq , strain range Δε t and stress intensity factor K, load parameters include stress ratio R σ and frequency f.

[0077] S200, calculate the physical information parameters of the SCR riser

[0078] The maximum circumferential stress and generalized strain at the crack are obtained by analytical method, and the damage parameter at the crack is further obtained. Wherein, it is assumed that the crack of the riser expands along the radial direction, then the circumferential stress σ rmax at the crack of the riser is represented by the stress intensity factor:

[0079]

[0080] In the above formula (1), K is the force intensity factor, and r is the crack length;

[0081] The generalized strain ε f at the crack of the riser is represented by the stress triaxiality of the material: ​and critical strain ε o to represent:

[0082]

[0083] In the above formula (2), σ is the hydrostatic pressure, σ eq is the Von Mises stress, σ1, σ2 and σ3 are the principal stresses in three directions, ε o is the critical strain of the initial crack of the material.

[0084] The damage parameter SWT at the crack of the riser is obtained by the following formula (3):

[0085] SWT = σ rmax Δε t / 2 (3).

[0086] In the above formula (3), ΔE t is the total strain range around the girth weld.

[0087] S300, the material properties, riser geometric parameters, crack geometric parameters, stress-strain, load parameters and crack fatigue physical information parameters are taken as model inputs, and the crack propagation rate is taken as model output, the original data of the riser fatigue test in the above step S100 and the physical information parameters of the riser in the above step S200 are normalized and dimensionless, and a deep sea riser feature data set is constructed.

[0088] Wherein, the material properties are dimensionless and normalized by using the material properties of the SCR material at room temperature, as shown in the following formula (4):

[0089]

[0090] In the above formula (4), M and M d are the material properties before and after normalization, M s is the material property at room temperature.

[0091] The riser geometric parameters are normalized by using the size of the SCR standard pipeline, and the crack geometric parameters are normalized by using the standard threshold value obtained by the SCR experiment, as shown in the following formula (5):

[0092]

[0093] In the above formula (5), G and G d are the geometric parameters before and after normalization; G contains the riser geometric parameters and the crack geometric parameters; G th represents the threshold value for dimensionless and normalization, which is selected according to the size of the standard pipeline and the maximum value measured in the crack fatigue test.

[0094] The stress-strain parameter is normalized by using the standard threshold value obtained from the SCR test, as shown in the following equation (6):

[0095]

[0096] In the above equation (6), φ is a dimensionless stress-strain parameter before normalization; φ d is a dimensionless stress-strain parameter after normalization; φ th represents the threshold value for normalization, which is selected based on the maximum value measured in the crack fatigue test.

[0097] The load parameter is normalized by using the following equation (7):

[0098]

[0099] In the above equation (7), R σ represents the stress ratio, represents a dimensionless stress ratio after normalization, f represents the load frequency, and f d represents a dimensionless frequency after normalization, f th is the threshold value specified in the riser fatigue test standard.

[0100] The physical information parameter is normalized by using the following equation (8):

[0101]

[0102] In the above equation (8), σ rmax represents the maximum hoop stress, represents a dimensionless maximum hoop stress after normalization, ε f represents the generalized strain, ε d represents a dimensionless strain after normalization, SWT represents the damage parameter at the crack, and SWT d represents a dimensionless damage parameter after normalization; σ th , ε th , and SWT th are selected based on the maximum value in the analytical calculation result.

[0103] The crack propagation rate is normalized by using the following equation (9):

[0104]

[0105] In the above equation (9), CGR represents the crack propagation rate, CGR d represents a dimensionless crack propagation rate after normalization, CGR thThe maximum value of the data obtained by the experiment is selected.

[0106] The specific input and output of the data set are shown in Table 1.

[0107] Table 1: Data set input characteristics

[0108]

[0109]

[0110] S400, data segmentation and reorganization based on the time sequence characteristics of the riser fatigue test data, the riser fatigue test data set is divided into training set and test set according to the proportion of 7:3, and a long short memory deep learning model containing physical constraints is built based on the characteristics of crack propagation.

[0111] In the embodiment, the data contains crack propagation data under different stress levels, and the crack data distribution under high, medium and low stress levels is 10%, 30% and 60% respectively. The data subsets are divided according to different stress levels; according to the principle of 20 sample points per group, 60000 data points are redivided. The division process ensures that the sample points come from the same data subset and the sequence is unchanged, and finally 300 new data groups are obtained;

[0112] The above data segmentation and reorganization based on the time sequence characteristics of the riser fatigue test data is to use stratified sampling method to stratify sample each subset according to training: test = 7:3, and obtain the training set and test set of the deep learning model for training and testing.

[0113] Then the physical constraint is introduced into the loss function of the long short memory deep learning model to improve the interpretability of the model. The composition of the loss function L(θ) contains two parts, the first part is the deviation of the predicted value and the true value, and the second part is the deviation of the predicted result under the physical constraint, which can be specifically represented as:

[0114] L(θ)=L d (θ)+λL p (θ) (10);

[0115] In the above formula (10), L d (θ) represents the deviation of the predicted value and the true value, L p (θ) represents the deviation of the predicted value and the physical law, and λ is a penalty factor for balancing the influence of the two parts on the loss function; L d (θ) is calculated by RMSE, and L p (θ) is determined by calculating the derivative of the crack propagation rate;

[0116] The crack propagation process is divided into three stages: initiation stage, stable stage, and rapid propagation stage; the expression for the loss term is:

[0117]

[0118] In the above formula (11), N represents the number of cycles, a represents the crack depth, and I(A>B) is a state indicator function. When A is greater than B, its value is 1, indicating that the physical constraints have been violated; otherwise, it is equal to 0, indicating that it is consistent with the physical constraints.

[0119] S500: Train a Long Short-Term Memory (LSTM) deep learning model using the training set in the riser fatigue test dataset. Obtain the optimal parameters of the LSTM deep learning model through Bayesian optimization. Compare different models to obtain the optimal physical information-driven deep learning model. The specific process is as follows:

[0120] S501, Select 6-time cross-validation R 2 average The number of layers N in the Long Short-Term Memory (LSTM) deep learning model is determined as the objective function for optimization. layer Number of units per layer N units The optimization range of the learning rate η is determined by combining Bayesian optimization and K-fold cross-validation to obtain the optimal parameters for the long short-term memory deep learning model. The objective function selected in the optimization process is the N-fold cross-validation R. 2 average

[0121] In equation (12) above, R i 2 Let be the coefficient of determination for the i-th cross-validation; This is the average value of K cross-validations.

[0122] S502. Select three key hyperparameters from the Long Short-Term Memory (LSTM) deep learning model for optimization. These are: the number of layers N in the LSTM model. layer Number of units per layer N units Using the learning rate η, the final hyperparameter optimization results are obtained by Bayesian optimization method; Table 2 lists the optimization range of these hyperparameters in detail.

[0123] Table 2 Hyperparameter Optimization Range

[0124]

[0125] The final hyperparameter optimization results are shown in Table 3; it should be noted that these results are the converged results of the hyperparameter optimization, and the combination of hyperparameters... The value is the maximum, indicating that the prediction ability of the model under this combination of hyperparameters is the most stable. The number of optimizations needs to be determined according to the size of the database. In this embodiment, 1000 times are selected.

[0126] Table 3 Hyperparameter optimization results

[0127]

[0128]

[0129] S503, compare the prediction results of the optimal physical information driven deep learning model and the traditional model, and obtain the optimal model.

[0130] In this embodiment, the performance of the data-driven model without introducing physical information and physical constraints is compared with that of the physical information driven model. For the data-driven model, this embodiment also performs hyperparameter optimization, and the prediction results of the two models are shown in Figure 2 , Figure 3 It is not difficult to see that for the LSTM model, only 75% of the predictions are within ±1.5 error range, and 63% of the predictions are within ±2 error range. Figure 2 Some abnormal values of prediction difference marked with circles are shown. This indicates that LSTM has potential improvement in prediction stability and generalization. Figure 3 Detailed prediction results using the PI-LSTM model are provided. More than 95% of the PI-LSTM model predictions are within ±1.5 error range, and all predictions are within ±2 error range. Compared with the LSTM model, the prediction accuracy of the PI-LSTM model is significantly improved, effectively alleviating Figure 2 the abnormal values shown. The results show that incorporating physical information as prior knowledge significantly improves the accuracy and stability of the LSTM model.

[0131] S600, use the trained physical information driven deep learning model for fatigue life prediction, compare the prediction results with the deep learning model without introducing physical parameters and the traditional physical fatigue life model, and quantitatively evaluate the life uncertainty; specifically including the following steps:

[0132] S601, sample the crack depth a, crack width c, maximum principal stress σ max , strain range Δε t in the deep sea riser feature data set by Latin hypercube sampling (LHS) to obtain enough samples. The sampling range is limited to the first 80% of the deep sea riser feature data set, and the data here basically meets the Paris theorem. Assuming that the first 80% of the data in the deep sea riser feature data set follows a normal distribution, the experimental measurement value represents the mean value, the standard deviation of the crack depth is 0.05 mm, and the standard deviation of the crack width is 0.1 mm, the maximum principal stress σmax Standard deviation is 0.5 MPa, strain range Δε t Standard deviation is 0.1, and the total sample number is 500.

[0133] S602, the first 80% of the data samples of the deep-sea riser feature data set are segmented according to 20 sample points to meet the input form of the physical information driven deep learning model, and then the trained physical information driven deep learning model is used for prediction to obtain corresponding crack propagation data.

[0134] S603, the crack depth of the riser is calculated according to the crack propagation data, and the calculation formula of the crack depth is as follows:

[0135]

[0136] In the above formula (13), a i represents the crack depth at i moment, a i+1 represents the crack depth at i+1 moment, represents the crack propagation rate at i moment, and dN represents the cycle number;

[0137] S604, taking the failure crack depth measured by the test as the criterion, the cycle number N when reaching the crack depth in the above formula (13) is calculated, and the N is the fatigue life.

[0138] S605, according to the fatigue life result, a P-S-N curve is drawn to realize quantitative evaluation of uncertainty. Figure 4 The P-S-N curve of the riser is given, Figure 4 It is shown that under the condition of 50% survival probability, there is a strong correlation between the experimental life and the predicted fatigue life. In addition, the experimental life is between 99% and 1% survival probability, and the change of the girth weld fatigue life is mainly due to the geometric inconsistency of the weld. Figure 5 It is shown that the uncertainty range of the predicted life is within ±1.5 error range, which shows that the uncertainty quantization result of the PI-LSTM model is in good agreement with the experimental life data. In order to further verify the physical consistency of the PI-LSTM model, Figure 6 The S-N curve of the pipeline girth weld under the BS7608 standard is given, and it is not difficult to see that the PI-LSTM model is very close to the BS7608 D curve.

[0139] The application has reasonable concept, can evaluate the crack propagation rate of the girth weld under the conditions of multiple sizes and multiple stresses, can realize probability prediction and uncertainty quantization of the riser fatigue life, and can provide reference for reliability evaluation of the in-service riser structure.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting the fatigue life of deep-sea risers based on physical information, characterized in that, Includes the following steps: 1) Collect raw data from riser fatigue tests, including material properties, riser geometry, crack geometry, stress and strain, load parameters, and crack propagation rate; 2) Calculate the physical information parameters of the riser. The maximum circumferential stress and generalized strain at the crack in the riser were obtained by analytical method, and the damage parameters at the crack in the riser were further obtained. 3) The five types of parameters, namely material properties, riser geometry parameters, crack geometry parameters, stress and strain, load parameters and crack fatigue physical information parameters, are used as model inputs, and crack propagation rate is used as model output. The raw data of riser fatigue test in step 1) and the physical information parameters of riser in step 2) are normalized and dimensionless respectively to form a deep-sea riser feature dataset. 4) Based on the temporal characteristics of riser fatigue test data, the data is segmented and reorganized. The riser fatigue test dataset is divided into training set and test set in a 7:3 ratio. A long short memory deep learning model with physical constraints is built based on the characteristics of crack propagation. 5) Train the long short-term memory deep learning model using the training set in the riser fatigue test dataset, obtain the optimal parameters of the long short-term memory deep learning model through the Bayesian optimization method, and compare different models to obtain the optimal physical information-driven deep learning model. 6) Fatigue life prediction is performed by driving a deep learning model with optimal physical information, and uncertainty assessment is conducted; In step 4), data segmentation and reorganization based on the temporal characteristics of riser fatigue test data involves first dividing each subset into a training set and a test set using stratified sampling. Then, physical constraints are introduced into the loss function of the long short-term memory deep learning model to improve its interpretability. The loss function L(θ) consists of two parts: the first part is the deviation between the predicted value and the true value, and the second part is the deviation of the prediction result under physical constraints. Specifically, it is expressed as follows: L(θ)=L d (θ)+λ·L p (i) (10); In the above formula (10), L d (θ) represents the deviation between the predicted value and the actual value; L p (θ) represents the deviation between the predicted value and the physical law; λ is the penalty factor, which is used to balance the influence of the two parts on the loss function; L d (θ) is calculated using RMSE, L p (θ) is determined by calculating the derivative of the crack propagation rate; The crack propagation process is divided into three stages: initiation stage, stable stage, and rapid propagation stage; L p The expression for (θ) is: In the above formula (11), N represents the number of cycles, a represents the crack depth, and I(A>B) is a state indicator function. When A is greater than B, its value is 1, indicating that the physical constraints have been violated; otherwise, it is equal to 0, indicating that it is consistent with the physical constraints.

2. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 1, characterized in that: The material properties in step 1) include elastic modulus E, Poisson's ratio v, and yield strength σ. limit The geometric parameters include two parts: riser geometric parameters and crack geometric parameters. The riser geometric parameters include inner diameter d, outer diameter D, wall thickness t, and pipe length L. The crack geometric parameters include crack depth a and crack width c. The stress and strain parameters include the maximum principal stress σ. max hydrostatic pressure σ m Von Mises stress σ eq Strain range Δε t And stress intensity factor K, load parameters including stress ratio R σ and frequency f.

3. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 1, characterized in that, In step 2), assuming the crack in the riser extends radially, the circumferential stress σ at the crack in the riser... rmax Represented by stress intensity factor: In equation (1) above, K is the force intensity factor and r is the crack length; Generalized strain ε at the riser crack f Using the stress triaxiality of materials and critical strain ε o To indicate: In equation (2) above, For hydrostatic pressure; σ eq The stress is Von Mises; σ1, σ2, and σ3 are the three components of the principal stress, respectively, and ε o This represents the critical strain for the initial crack formation in the material.

4. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 1, characterized in that, In step 2), the damage parameters at the riser crack are obtained by the following formula (3): SWT=s rmax No t / 2 (3); In equation (3) above, Δε t σ represents the total strain range around the circumferential weld. rmax This refers to the circumferential stress at the crack in the riser.

5. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 1, characterized in that, The material properties in step 3) are obtained by dimensionlessly and normally processing the material property parameters at room temperature: In equation (4) above, M represents the dimensionless and unnormalized performance parameters of the material; M d M represents the dimensionless and normalized performance parameters of the material; s These are the material property parameters at room temperature; In step 3), the riser geometric parameters are normalized using the dimensions of standard pipes, and the crack geometric parameters are normalized using standard thresholds. In equation (5) above, G and G d G represents the dimensionless geometric parameters before and after normalization, and includes both riser and crack geometric parameters. th The threshold value used for dimensionless and normalized values ​​is selected based on the size of the standard pipe and the maximum value measured in the crack fatigue test. The stress-strain parameters in step 3) are normalized using standard thresholds. In equation (6) above, φ represents the dimensionless stress-strain parameter before normalization; φ d φ represents the dimensionless and normalized stress-strain parameters. th This represents the threshold value used for dimensionless and normalized values, which is selected based on the maximum value measured in the crack fatigue test. The load parameters in step 3) are normalized using the following formula (7): In equation (7) above, R σ Indicates the stress ratio. The stress ratio is expressed as the ratio of the dimensionless stress to the normalized stress, where f represents the load frequency. d f represents the dimensionless and normalized frequencies. th The threshold value specified in the riser fatigue test standard; The physical information parameters in step 3) are normalized using the following formula (8): In equation (8) above, σ rmax Indicates the maximum circumferential stress. ε represents the dimensionless and normalized maximum circumferential stress. f Represents generalized strain, ε d SWT represents the dimensionless and normalized strain, and SWT represents the damage parameter at the crack. d σ represents the dimensionless and normalized damage parameters; th ε th and SWT th The maximum value is selected based on the analytical calculation result. The crack propagation rate in step 3) is normalized using the following formula (9): In equation (9) above, CGR represents the crack propagation rate. d CGR represents the dimensionless and normalized crack propagation rates. th The selection is based on the maximum value of the data obtained from the experiment.

6. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 1, characterized in that, The specific process of step 5) is as follows: 5.1) Select 6-time cross-validation R 2 average The number of layers N in the long short-term memory deep learning model is determined as the objective function for optimization. layer Number of units per layer N units The optimization range of the learning rate η is determined by combining Bayesian optimization with K-fold cross-validation to obtain the optimal parameters for the long short-term memory deep learning model. The objective function selected in the optimization process is the K-fold cross-validation R. 2 average In equation (12) above, R i 2 Let be the coefficient of determination for the i-th cross-validation; The average of K cross-validations; 5.2) Select the three most critical hyperparameters from the Long Short-Term Memory deep learning model, namely the number of layers N. layer Number of units per layer N units Given the learning rate η, the final hyperparameter optimization result is obtained using the Bayesian optimization method; 5.3) Compare the prediction results of the optimal physical information-driven deep learning model with those of the traditional model to obtain the optimal model.

7. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 1, characterized in that, The specific process of step 6) is as follows: use the trained physical information to drive the deep learning model to predict fatigue life, compare the prediction results with those of the deep learning model without physical parameters and the traditional physical fatigue life model, and quantitatively evaluate the life uncertainty.

8. The method for predicting the fatigue life of deep-sea risers based on physical information as described in claim 7, characterized in that, The specific steps for quantifying and assessing the aforementioned lifetime uncertainty are as follows: 6.1) Crack depth a, crack width c, and maximum principal stress σ were centrally analyzed in the deep-sea riser feature data using Latin hypercube sampling. max Strain range Δε t Sampling is performed to obtain a sufficient number of samples; the sampling range is limited to the first 80% of the deep-sea riser feature dataset; it is assumed that the first 80% of the deep-sea riser feature dataset follows a normal distribution, where the experimental measurements represent the mean and the standard deviation is determined by the measurement error. 6.2) The first 80% of the data samples in the deep-sea riser feature dataset will be segmented to meet the input format of the physical information driven deep learning model. The trained physical information driven deep learning model will then be used to make predictions to obtain the corresponding crack propagation data. 6.3) Calculate the crack depth of the riser based on the crack propagation data. The formula for calculating the crack depth is as follows: In the above formula (13), a i a represents the crack depth at time i. i+1 This represents the crack depth at time i+1. dN represents the crack propagation rate at time i, and dN represents the number of cycles. 6.4) Using the experimentally measured failure crack depth as the criterion, calculate the number of cycles N in the above formula (13) when this crack depth is reached. This N is the fatigue life. 6.5) PSN curves are plotted based on fatigue life results to achieve quantitative assessment of uncertainty.

Images