Pipeline damage prediction method under combined action of impact load and corrosion defect

Abstract

The invention discloses a pipeline damage prediction method under the combined action of impact load and corrosion defects, and relates to the technical field of submarine pipeline systems. The pipeline damage prediction method under the impact load-corrosion defect combined action mainly comprises the steps that a nonlinear finite element analysis model is constructed according to corrosion defect size information, pipeline geometric size information, seabed soil geometric size and falling object geometric characteristics; constructing a sample space set by using the nonlinear finite element analysis model according to key factors; and constructing a pipeline damage prediction model, training the pipeline damage prediction model by using the sample space set to obtain a trained pipeline damage prediction model, and performing prediction to obtain a prediction result. By implementing the method for predicting the damage of the pipeline under the combined action of the impact load and the corrosion defect, the damage of the pipeline structure can be efficiently and reasonably predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of submarine pipeline systems, and more particularly to a pipeline damage prediction method under the combined action of impact load and corrosion defects. BACKGROUND

[0002] As the main medium for transporting oil and gas resources, submarine pipelines have the advantages of high efficiency and economy, and are widely used in the marine oil and gas development industry, so the structural safety is of great importance.

[0003] With the frequent human marine activities and the increasing service life of submarine pipelines, the combined action of impact load and corrosion defects has become a major factor threatening the structural safety of submarine pipelines.

[0004] Statistical data shows that impact load and corrosion caused by third-party activities are the main reasons for pipeline structural failure, and the risk of pipeline structural failure under the combined action of the two further increases, becoming a major factor threatening the safety of pipeline service.

[0005] However, in the current specifications and existing research methods, only the bearing capacity, structural failure mechanism and damage prediction of pipelines under the action of single impact load or corrosion defects have been studied, and there is no effective analysis method for the bearing capacity and damage assessment of pipeline structures under the combined action of impact load and corrosion defects, so the structural safety state cannot be accurately judged, which introduces potential risks for the marine oil and gas development industry.

[0006] In summary, with the increasing frequency of human marine activities and the increasing service life of pipelines, the combined action of impact load and corrosion defects has gradually become a key factor considered in pipeline integrity management, so it is necessary to develop a corresponding damage assessment method to accurately judge the structural safety state and develop maintenance strategies.

[0007] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0008] The purpose of the present application is to provide a pipeline damage prediction method under the combined action of impact load and corrosion defects, which can realize efficient and reasonable prediction of pipeline structural damage.

[0009] The present application provides a pipeline damage prediction method under the combined action of impact load and corrosion defects, comprising the following steps: S1: According to the corrosion defect size information, the pipeline geometric size information, the seabed soil geometric size and the falling object geometric characteristics, a nonlinear finite element analysis model is constructed; S2: According to the key factors, a sample space set is constructed by using the nonlinear finite element analysis model; S3: Constructing a pipeline damage prediction model, using the sample space set to train the pipeline damage prediction model to obtain a trained pipeline damage prediction model, and using the trained pipeline damage prediction model to perform prediction to obtain a prediction result.

[0010] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for predicting pipeline damage under the combined action of impact load and corrosion defects.

[0011] The implementation of the pipeline damage prediction method under the combined action of impact load and corrosion defects provided by the present invention has the following beneficial effects: Based on theoretically sound nonlinear finite element analysis, the present invention constructs an impact load-soil-pipeline interaction mechanical analysis system to ensure an accurate description of the structural response; a nonlinear finite element analysis model is constructed to achieve an accurate description of the damage to pipelines with corrosion defects under impact loads. Specifically, based on the dynamic finite element theory, the explicit central difference algorithm and the field variable iteration algorithm are used to realize the theoretical construction, numerical discretization and solution of the control equations of the impact falling object-pipeline with corrosion defects-soil interaction mechanical analysis system, capturing the material strain rate effect during impact loading and the effects of stress concentration and bearing capacity reduction caused by corrosion defects on the structural response under impact loads, thereby achieving a reasonable description of pipeline impact damage.

[0012] The present invention maps the failure behavior characteristics of the above-mentioned mechanical analysis system architecture to the digital feature space through deep learning theory, constructs a pipeline damage prediction model based on nonlinear finite element analysis and deep learning theory, deconstructs the construction and solution process of the mechanical system control equations with matrix algebra operations, establishes a nonlinear quantitative relationship between the structural response characteristics and its influencing factors, ensures the efficiency and accuracy of structural damage prediction, and realizes efficient and reasonable prediction of pipeline structural damage; for given impact energy and corrosion defect conditions, this method can perform real-time evaluation of the structural safety status, providing theoretical and technical support for pipeline safety design and operation and maintenance strategy formulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 This is a flow chart of the pipeline damage prediction method under the combined action of impact load and corrosion defects provided by the present invention; Figure 2 This is a schematic diagram of the arrangement of a non-defective pipeline impact test under flexible seabed soil conditions provided by the present invention; Figure 3 This is a schematic diagram comparing the axial strain time history curves of the pipeline specimen obtained by the numerical analysis and the experiment provided by the present invention; Figure 4Schematic diagram of the arrangement of the impact test on a pipeline with corrosion defects under rigid seabed conditions provided by the present invention; Figure 5 This is a schematic diagram comparing the axial and hoop strain time history curves of the pipeline specimen obtained through numerical analysis and experiment provided by the present invention; Figure 6 This is a schematic diagram of a linear regression diagram of the predicted results and the actual results provided by the present invention; Figure 7 It is a schematic diagram of an implementation flow chart provided by the present invention. DETAILED DESCRIPTION

[0014] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0015] Figure 1 A schematic diagram of a pipeline damage prediction method under the combined effects of impact load and corrosion defects in this embodiment is shown. In this embodiment, the pipeline damage prediction method under the combined effects of impact load and corrosion defects includes the following steps: S1: Based on the corrosion defect size information, pipeline geometry information, seabed soil geometry and the geometric characteristics of the falling object, a mesh model of the impact falling object-corrosion defect pipeline-soil and a nonlinear finite element analysis model are constructed; In an exemplary embodiment, step S1 specifically includes: S11: Construct a mesh model of impact falling object-corrosion defect-pipeline-soil based on the corrosion defect size information, pipeline geometry information, seabed soil geometry and falling object geometry characteristics; In an exemplary embodiment, step S11 specifically includes: S111: Constructing a grid model of the pipeline containing corrosion defects based on the corrosion defect size information and pipeline geometric size information; In an exemplary embodiment, step S111 specifically includes: S1111: Construct a corrosion defect geometric model based on the corrosion defect size information; In an exemplary embodiment, the corrosion defect size information includes corrosion depth, corrosion length, and corrosion width; S1112: Constructing a complete pipeline geometric model based on pipeline geometric dimension information; In an exemplary embodiment, the pipeline geometric dimension information includes outer diameter, wall thickness, and length; S1113: performing assembly and Boolean operations on the corrosion defect geometric model and the complete pipeline geometric model to obtain a pipeline geometric model containing corrosion defects; S1115: Using the Lagrangian algorithm to perform grid discretization on the geometric model of the pipeline containing corrosion defects, a grid model of the pipeline containing corrosion defects is obtained. ; S112: Construct a seabed soil mesh model based on the seabed soil geometry; In an exemplary embodiment, step S112 specifically includes: S1121: Construct a seabed soil geometric model based on the seabed soil geometric dimensions; In an exemplary embodiment, the geometric dimensions of the seabed soil body include length, width, and height; S1122: Use the Lagrangian algorithm to discretize the seabed soil geometry model and obtain the seabed soil mesh model ; S113: Constructing a mesh model of the falling object according to the geometric characteristics of the falling object; In an exemplary embodiment, step S113 specifically includes: S1131: Constructing a geometric model of the falling object based on its geometric features; In an exemplary embodiment, the geometric characteristics of the falling object include shape 、 size; S1132: Use the Lagrangian algorithm to discretize the geometric model of the falling object and obtain the falling object mesh model ; S114: Assembling the grid model of the pipeline with corrosion defects, the grid model of the seabed soil, and the grid model of the falling object to obtain a grid model of the impact falling object-pipeline with corrosion defects-soil; S12: constructing a nonlinear finite element analysis model based on the impact falling object-corrosion defect pipeline-soil mesh model, and solving it to obtain a solution result; In an exemplary embodiment, step S12 specifically includes: S121: Constructing a nonlinear finite element analysis model based on the impact falling object-corrosion defect pipeline-soil mesh model, wherein the nonlinear finite element analysis model includes a control equation; In an exemplary embodiment, the control equation is as follows: , in, 、 、 They are the unit node mass matrix, stiffness matrix and external load matrix in the nonlinear finite element analysis model; 、 are the unit node acceleration matrix and displacement matrix respectively; S122: Solving the control equation using a field variable iteration algorithm and an explicit central difference algorithm to obtain a solution result; In an exemplary embodiment, the field variable iteration algorithm is as follows: , , , , , , in, express Moment Undrained shear strength at each soil unit integration point; represents the function that characterizes the undrained shear strength with the field variable value as the independent variable, and Indicates the The field variable values ​​at the start and end of the current calculation time step at each soil unit integration point, , They represent the end time of the current calculation time step. The maximum principal strain of each soil element and its increment, and Indicates the starting time of the current calculation time step. The maximum and minimum principal strains of each soil element, , They represent the end time of the current calculation time step. The minimum principal strain of each soil element and its increment, and and Respectively represent the last calculation time step The variables in the solution correlation variables of each soil unit are the maximum and minimum principal strains and the cumulative plastic strain; and and Respectively represent the current calculation time step The variables in the solution correlation variables of each soil unit are the maximum and minimum principal strains and the cumulative plastic strain; and Indicates the start and end time of the current calculation time step The accumulated plastic strain of each soil element is Indicates the end time of the current calculation time step. Maximum shear rate per soil unit; Indicates the time increment of the current calculation time step; is the soil strength factor value at the soil element integration point at the end of the current calculation time step; is the ratio of the shear strength of the soil in the fully remolded state to its initial value; is the ratio of the shear strength of the soil in the fully remolded state to its initial value; is the cumulative plastic strain corresponding to a 95% reduction in soil strength due to remolding; is the strain rate factor; represents the reference shear rate; In an exemplary embodiment, the explicit central difference algorithm is as follows: , , , , wherein, , , , respectively represent the element node displacement, velocity, acceleration and external load matrix of the nonlinear finite element analysis model, and the subscript , , respectively represent the midpoint time of the previous calculation time step, the midpoint time of the current calculation time step and the end time of the current calculation time step, and respectively represent the time increment of the previous and current calculation time steps; , , , respectively represent the element node stiffness, mass, geometric relationship and physical relationship matrix of the nonlinear finite element analysis model; S2: constructing a sample space set according to the key factors by using the nonlinear finite element analysis model; In an exemplary embodiment, step S2 specifically comprises: S21: constructing an input data set according to the key factors In an exemplary embodiment, the key factors include the corrosion defect depth 、 the corrosion defect length 、 the corrosion defect width 、 the impact energy; S22: obtaining structural response data by using the nonlinear finite element analysis model according to the input data set, and constructing an output data set according to the structural response data; S23: constructing a sample space set according to the input data set and the output data set; S3: Constructing a pipeline damage prediction model, using the sample space set to train the pipeline damage prediction model to obtain a trained pipeline damage prediction model, and using the trained pipeline damage prediction model to perform prediction to obtain a prediction result; In an exemplary embodiment, step S3 specifically includes: S31: Constructing a pipeline damage prediction model based on a multilayer perceptron framework; In an exemplary embodiment, the pipeline damage prediction model includes an input layer, a hidden layer, and an output layer. The input layer includes four perceptrons for inputting corrosion defect depth, corrosion defect length, corrosion defect width, and impact energy, respectively. The hidden layer includes 20 perceptrons for coordinating with the perceptrons in the previous and next layers to construct a nonlinear mapping system to achieve deconstruction and mapping of the pipeline structure damage mechanics analysis system. The output layer includes a perceptron for outputting the ellipticity of the pipeline mid-span cross-section. S32: Divide the sample space set according to a preset ratio and normalize it to obtain a training set and a test set; In an exemplary embodiment, the preset ratio is 0.8:0.2; S33: training the pipeline damage prediction model using a gradient descent algorithm based on the training set to obtain a trained pipeline damage prediction model; S34: using the trained pipeline damage prediction model to predict the test set to obtain a prediction result; In an exemplary embodiment, the above pipeline damage prediction method under the combined action of impact load and corrosion defect further includes: performing prediction performance evaluation based on the prediction results; As an exemplary embodiment, each set of input data in the test set constructed in step S2 is Input pipeline damage prediction model and get its corresponding prediction output , and with Output data The prediction output of the pipeline damage prediction model is calculated as follows: and Output data Coefficient of determination R 2 ), Correlation coefficient, R ), mean absolute percentage error, MAPE ), thereby evaluating the prediction performance of the pipeline damage prediction model for unknown data.

[0016] , , , Where: is the coefficient of determination; R is the correlation coefficient; MAPE is the mean absolute percentage error; is the predicted value of the target object obtained by the prediction model, is the true value of the target object in the sample space dataset; , Represents the mean and standard deviation of the true value and predicted value corresponding to the target object; is the sample size of the sample space dataset.

[0017] In some embodiments, the above-mentioned pipeline damage prediction method under the combined action of impact load and corrosion defects can also be implemented in the following manner.

[0018] In this embodiment, the pipeline damage prediction method under the combined action of impact load and corrosion defect includes the following steps: (1) The dynamic finite element analysis theory, explicit central difference algorithm and field variable iteration algorithm are coupled to construct a nonlinear finite element analysis (NFEA) model of the interaction between impact falling objects, pipelines with corrosion defects and soil. The theoretical construction, numerical discretization and solution of the control equations of the mechanical analysis system are completed. The effects of the strain softening and strain rate (SSSR) effect of the soil, the stress concentration caused by corrosion defects and the reduction of bearing capacity on the pipeline structure response under impact loads are effectively captured, and a reasonable description of the structural failure behavior and damage degree is achieved. (2) Based on the uniform sampling method, the NFEA model is driven to perform sampling simulation calculations and construct a sample space set , providing basic data support for the development of damage prediction models; (3) A pipeline damage prediction model is established based on the multilayer perceptron (MLP) framework in deep learning theory. DS By optimizing the parameter calibration of the pipeline damage prediction model, the mechanical analysis system described by the NFEA model can be mapped and deconstructed in the digital feature space to quantify the nonlinear relationship between the structural failure behavior characteristics and related influencing factors, and accurately and efficiently predict the degree of structural damage.

[0019] In this embodiment, the pipeline damage prediction method under the combined action of impact load and corrosion defect includes the following steps: 1 Construction of nonlinear finite element analysis model 1.1 Establishment of the impact falling object-corrosion defect pipeline-soil mesh model According to the geometric characteristics of the falling object, pipeline and soil, the corresponding geometric model and mesh model are established to determine the numerical solution domain of the mechanical analysis system described by the NFEA model. .

[0020] 1.1.1 Construction of a Pipeline Grid Model with Corrosion Defects (1) Based on the corrosion defect size information, including corrosion depth d d , corrosion length l d , corrosion width w d Establishing a geometric model of corrosion defects ; (2) Based on the pipe geometry information, including the outer diameter D p , wall thickness t p ,length l p Establish complete pipeline geometry model ; (3) and Assembled into the same space and The center of the circle corresponding to the central section is used as the origin to establish a cylindrical coordinate system Based on this coordinate system, adjust Position: Let the model center coordinates be ( i = 0, r = (0.5D p – 0.5d d ), z =0). The corrosion defect to be generated is located on the outer surface of the center of the pipeline; (4) Yes and Perform Boolean operations to obtain the geometric model of the pipeline with corrosion defects M pd,geo ; (5) Using Lagrangian algorithm M pd,geo Perform grid discretization to obtain the corresponding pipeline grid model with corrosion defects M pd,ms ( X , Y , Z ).

[0021] 1.1.2 Construction of seabed soil grid model (1) Construct a geometric model of the seabed soil based on its geometric dimensions, including length, width, and height M s,geo ( L s , H s , W s ); (2) Using Lagrangian algorithm M s,geo ( L s , H s , W s ) to discretize the grid and obtain the corresponding seabed soil grid model M s,ms ( X , Y , Z ).

[0022] 1.1.3 Construction of falling object grid model (1) Based on the geometric characteristics of the falling object, including its shape MP foj ,size D foj , construct the geometric model of the falling object M foj,geo ( MP foj , D foj ); (2) Using Lagrangian algorithm M foj,geo ( MP foj , D foj ) to discretize the grid and obtain the corresponding falling object grid model M foj,ms ( X , Y , Z ).

[0023] 1.1.4 Mesh Model Assembly (1) M foj,ms , M s,ms , M pd,ms Assemble into the same space to M s,msThe center point of the upper surface is used as the origin to construct the three-dimensional Cartesian coordinate system CSYS car ( X , Y , Z ); (2) Based on CSYS car ( X , Y , Z ), adjust the relative position of the grid model to ensure that the pipeline is located at the center of the seabed surface, and the impact load acts on the mid-span position of the pipeline: M pd,ms The coordinates of the center of the central cross section are ( x = 0, y =0.5 D p , z = 0), let the coordinates of the center of the falling object be ( x = 0, y = ( D + h t + f z ( MP foj , D foj )), z = 0), where f z ( MP foj , D foj ) represents the vertical coordinate of the centroid of the falling object, h t Indicates the initial longitudinal distance between the falling object and the outer surface of the pipe.

[0024] 1.2 Establishment and solution of model control equations 1.2.1 Establishment of NFEA model control equations Based on the mesh models and their corresponding solution domains established in step 1.1, the material property characteristics of each mesh model are defined, and the mass and stiffness matrices in the NFEA model control equations are determined. On this basis, the initial conditions, boundary conditions, and contact effects are defined for the system composed of each mesh model, completing the theoretical establishment and numerical discretization of the NFEA model control equations shown in Equation (1).

[0025] (1) Where: M , K , F ext They are the unit node mass matrix, stiffness matrix and external load matrix in the NFEA model respectively; , U They are the unit node acceleration matrix and displacement matrix respectively (1) Definition of material property characteristics of the mesh model. The material property characteristics of the falling object, soil and pipeline are assigned to the mesh model, including their mass information and the constitutive relationship model determined by the strength parameter information: 1) For the falling object mesh model, the ideal linear elastic model is used as its constitutive model; 2) For the soil mesh model, the Mohr-Coulomb model is used as its constitutive model, and its strength parameter is the undrained shear strength. Su Characterization, elastic modulus is E s = 500 Su , Poisson's ratio is v s = 0.49; 3) For the pipeline network model, the Cower-Symons model is used as its constitutive model.

[0026] (2) Definition of boundary conditions. 1) For the falling object mesh model, its structural deformation during the impact process is ignored, and rigid body constraints are applied between its mesh nodes and centroid to simplify it into a rigid body; 2) For the soil mesh model, fixed end constraints are applied to its bottom surface and directional displacement constraints are applied to its side boundaries; 3) For the pipe mesh model, fixed end constraints are applied to its two end boundaries.

[0027] (3) Definition of initial conditions. 1) For the falling object mesh model, apply the impact velocity v imp As an initial condition, the quality of m foj The initial impact energy of the mechanical analysis system constructed by the input NFEA model can be obtained E imp =0.5 m foj v 2 imp ; 2) For the soil grid model, an initial geostress field is applied to it and balanced with gravity, so that the soil unit is in a convergent stress state at the initial moment of calculation, where the stress of each unit is maintained within its initial yield surface, ensuring that the soil unit is in an ideal state with initial stress but no initial strain, so as to approximate the mechanical state of soil that has undergone a long geological evolution process under real conditions.

[0028] (4) Definition of contact action. A general contact algorithm is used to describe the contact action between the falling object mesh model, the soil mesh model, and the pipeline mesh model. The normal contact property is hard contact, and the tangential contact property is Coulomb contact.

[0029] 1.2.2 NFEA model solution The governing equations constructed in step 1.2.1 are solved using an explicit central difference algorithm. Furthermore, the strain softening and strain rate effects of the soil are incorporated into the NFEA model through a field variable iteration algorithm to accurately describe the effects of soil strength changes caused by shear intrusion during impact on the pipe-soil interaction and the pipeline structural response.

[0030] (1) Construction of field variable iteration algorithm. 1) Modify the constitutive model of the soil mesh model and define its undrained shear strength as a function with field variable (FV) as the independent variable according to formula (2), so that the field variable (FV) can be used in the calculation and analysis process. FV Associated with SRSS effect to achieve intensity update; 2) In the current calculation time step t Read the plastic strain information at the soil unit integration point, including the end time of the current calculation time step ( t + Δ t j )'s maximum plastic principal strain , minimum plastic principal strain , and the end time of the previous calculation time step ( t j-1 + Δ t j-1 = t )'s maximum plastic principal strain , minimum plastic principal strain and the accumulated plastic strain , and calculate the end time of the current calculation time step according to formula (3) – (4) The cumulative plasticity and maximum shear rate of the soil are further calculated according to the theoretical empirical formula shown in formula (5) to characterize the SSSR effect. α SRSS ; 3) As shown in formula (6), the soil strength factor value is assigned to the field variable at the unit integration point, and the soil strength value at the unit integration point is updated according to formula (2) to achieve real-time capture of the strength change during the shear flow process of the soil; 5) At the end of the calculation step at the current time ( ), the maximum plastic strain, minimum plastic strain and cumulative plastic strain are calculated according to formula (7). x Stored in the solution dependent variable (SDV) for reading at the next calculation time step.

[0031] (2) (3) (4) (5) (6) (7) Where: Su t k express t Moment k Undrained shear strength at each soil unit integration point; FV Indicates the value of the field variable at the unit integration point, and its subscript k Indicates the k Soil unit, superscript t , Respectively represent the start and end time of the current calculation time step; e , Δ e Represent the principal strain and its increment respectively, and its subscripts (1, k ),(3, k ) indicates the k The maximum and minimum principal strains of each soil element, superscript t , Respectively represent the start and end time of the current calculation time step; x represents the accumulated plastic strain, and its subscript k Indicates the k Soil unit, superscript t , Respectively represent the start and end time of the current calculation time step; SDV denotes the decorrelated variables, and the subscripts ε1, ε3, and ξ denote SDV The variables stored in are the maximum and minimum principal strains and the accumulated plastic strains. , t Indicates the previous and current calculation time steps; Indicates the end time of the current calculation time step. k The maximum shear rate of a soil unit; Δ t j Indicates the time increment of the current calculation time step; represents the reference shear rate; is the cumulative plastic strain corresponding to a 95% decrease in soil strength due to remodeling; d rem is the ratio of the shear strength of the soil in the fully reshaped state to its initial value; m is the strain rate factor; The end time of the current calculation time step k The soil strength factor value at each soil unit integration point.

[0032] (2) Explicit central difference algorithm is used to solve the problem. 1) At each time step t In the process, the control equations are solved according to the explicit central difference algorithm according to equations (8)-(10). During the solution process, the constitutive model of the soil is modified according to the field variable iteration algorithm described in step (1), and the soil strength is updated in real time; 2) According to equation (11), the solved unit node displacement matrix is ​​entered into the geometric relationship and physical relationship equations to obtain the corresponding node stress and strain matrix, thereby accurately describing the structural response; 3) Let , j = j + 1, repeat steps 1) – 2) to solve the control equation for the next calculation time step until the termination time is reached t ≥ t ter , completing the solution of the entire impact process.

[0033] (8) (9) (10) (11) Where: , , , F ext Respectively represent the unit node displacement, velocity, acceleration and external load matrix of the NFEA model, and their subscripts , , Respectively represent the midpoint of the previous calculation time step, the midpoint of the current calculation time step and the end time, 、 Represents the time increment of the previous and current calculation time steps respectively; K , M , B , D They represent the unit node stiffness, mass, geometric relationship and physical relationship matrix of the NFEA model respectively.

[0034] 2 Construction of sample space dataset Sample space dataset DS = { x i , e i ova ( x i ); i = 1, 2, 3, …, N USTo provide basic data support for establishing a pipeline damage prediction model based on deep learning theory. According to the nonlinear mapping relationship between the input data and the output data of the pipeline damage prediction model to be established, the DS is composed of two parts of data sets, including the input data set DS in = { x i ; i = 1,2, 3, …, N US} and the output data set DS in = { e i ova ( x i ); i = 1, 2, 3, …, N US}, which represent the key factors affecting the structural failure behavior characteristics and the structural response, respectively.

[0035] 2.1 Construction of input data set For the key factors affecting the pipeline structural failure behavior x , the uniform sampling method is used to randomly sample according to the value range of the key factors to construct the input data set, so as to fully cover the potential working conditions under various value combinations of the key factors and improve the robustness of the pipeline prediction model. According to the previous research, the key factors considered in this embodiment include: corrosion defect depth x , corrosion defect length d d , corrosion defect width l d , impact energy w d . With the above factors as input features, the input data set can be represented as E imp = { DS in = ( x i = ( d d,i , l d,i , w d,i , E imp , i ); i = 1, 2, 3, …, N US}.

[0036] 2.2 Construction of output data set The output data set is constructed by randomly sampling the pipeline failure mode DSini Each set of sample data in x i = ( d d,i , l d,i , w d,i , E imp , i ) is input into the NFEA model constructed in step 1, and the mechanical analysis system constructed under the conditions of this set of sampling data is solved to obtain the structural response data. e i ova ( x i ) is the characteristic of the structural response, so the output data set obtained by the calculation and analysis of the input data set can be expressed as DS out = { e i ova ( x i ), i = 1, 2,3, …, N US}.

[0037] 2.3 Dataset Assembly The one constructed in step 2.1 DS in Chinese elements x i Constructed in step 2.2 DS out Chinese elements e i ova ( x i )According to the one-to-one relationship{ x i , e i ova ( x i )} to assemble and construct the sample space set DS = { DS in , DS out} = { x i , e i ova ( x i );i = 1,2,3, …, N US}.

[0038] 3 Pipeline damage prediction model construction 3.1 Model topology design Damage prediction model based on the multi-layer perceptron MLP framework in deep learning theory MLP d Topology design. This embodiment adopts a three-layer structure: (1) Input layer: consists of 4 perceptrons, corresponding to the input data d d , l d , w d , E imp (2) Hidden layer: It consists of 20 perceptrons, which are used to coordinate the perceptrons between the previous and next layers to build a nonlinear mapping system, and realize the deconstruction and mapping of the pipeline structure damage mechanics analysis system; (3) Output layer: It consists of 1 perceptron, which corresponds to the output data, that is, the ellipticity of the pipeline mid-span cross section. e ova .

[0039] 3.2 Sample Space Dataset Division and Preprocessing (1) Create the DS The training set was randomly divided into two groups in a ratio of 0.8:0.2. DS tra and test set DS te ; DS tra Parameter optimization for pipeline damage prediction models, DS te Used to test the prediction performance of pipeline damage prediction model.

[0040] (2) Yes DS tra and DS te Normalization is performed according to formula (12) to uniformly scale the data of different scales and dimensions in the dataset to the interval [0, 1], eliminate the dimensional differences between features, reduce the weight deviation of different features in the pipeline damage prediction model, and ensure the consistency of the contribution of each feature to the model, so as to improve the convergence and stability of the model.

[0041] (12) Where: x The target data in the dataset; x min ,x max for x The minimum and maximum values ​​of x nor for x The normalized value of .

[0042] 3.3 Model Parameter Optimization Based on the training dataset in step 3.2 DS tra , using gradient descent algorithm to predict pipeline damage model MLP d Optimize the trainable parameters in the model, that is, the bias of each perceptron in the model b n l The bias matrix b and the connection weights between perceptrons w l k,n The weight matrix W (1) In each iteration cycle, the input data is forwarded layer by layer from the input layer to the output layer according to the model topology structure according to formula (13); (2) The cost function is constructed according to formula (14) with the prediction error of the output layer C ( W , b ), whose independent variables are the model trainable parameters, namely the model weight matrix W and bias matrix b (3) According to the back propagation algorithm, the output layer error can be expressed as the cost function with respect to the perceptron output gradient and estimated according to formula (15). On this basis, the output layer prediction error is propagated backward layer by layer according to the formula, and the gradient of the cost function with respect to the trainable parameters is estimated according to formula (16) through the chain rule. ; (4) According to According to formula (17)-(18), the trainable parameters ( W, b ) to make corrections; (5) Repeat steps (1) – (4) to conduct the next round of iterative training until the total number of iterative cycles reaches the predetermined number n e , and get the optimized trainable parameters ( W opt , b opt ), and then build an optimized pipeline damage prediction model MLP d,opt ( x ; W opt , b opt ).

[0043] (13) (14) (15) (16) (17) (18) (19) (20) In the formula: a n l represents the activation value of the i-th perceptron in the j-th layer; l n represents the activation value of the i-th perceptron in the j-th layer; represents the activation value of the i-th perceptron in the j-th layer; k w k,n l is the weight between the i-th perceptron in the j-th layer and the i-th perceptron in the k-th layer; l n k b n l represents the bias of the i-th perceptron in the j-th layer; l n n l is the output value of the i-th perceptron in the j-th layer; l n f (·) represents an activation function; C ( W , b ) is a cost function, W , b respectively represent the weight matrix and the bias matrix of the model; y exp is an expected output vector, representing the true value of the target object; a out is an activation value vector of the model in the output layer, representing the predicted value of the target object; d out represents the output error vector of the model in the output layer; z out is an output value vector of the model in the output layer; d l is an error vector of the model in the j-th layer; l W l+1 is the i-th perceptron in the j-th layer in the model​​​​​​​​​​​​ Layer and l The weight matrix between layers; d l+1 For the model The error vector of the layer; z l For the model l The output vector of the layer; The cost function is about l Tier n Perceptrons and Tier k The partial derivatives of the weights between the perceptrons; The cost function is about l Tier n The partial derivatives of the biases of the perceptrons; or is the learning rate.

[0044] 3.4 Model prediction performance evaluation The one constructed in step 2 DS te Each set of input data x i Input pipeline damage prediction model and get its corresponding prediction output e i ova,pre ( x i ), and with DS te Output data e i ova ( x i ) for comparison. According to formulas (21)-(23), the prediction output of the pipeline damage prediction model is calculated { e i ova,pre ( x i ); i = 1, 2, 3, …, N te}and DS te Output data { e i ova ( x i ); i = 1, 2, 3, …, N te Coefficient of determination, R 2), correlation coefficient (Correlation coefficient, R ), mean absolute percentage error (Mean absolute percentage error, MAPE ), so as to evaluate the prediction performance of the pipeline damage prediction model for unknown data.

[0045] (21) (22) (23) In the formula: R 2 R2is the determination coefficient; R R is the correlation coefficient; MAPE MAPE is the mean absolute percentage error; op i Y is the target object prediction value obtained by the prediction model, tp i Y is the real value of the target object in the sample space data set; m tp s tp m op s op , and (Y, Y) represent the mean and standard deviation corresponding to the real value and the prediction value of the target object; N DS N is the sample size of the sample space data set.

[0046] The model is verified as follows: the pipeline model damage prediction method proposed in this embodiment is constructed based on nonlinear finite element analysis. Therefore, the accuracy of the NFEA constructed in step 1 needs to be verified first. The pipeline impact experiment published in an international journal with detailed experimental process records is used as the research object, and the NFEA model constructed in this embodiment is used to reconstruct the numerical value. Two experiments are selected for verification, including: (1) an impact experiment of a defect-free pipeline installed on a flexible soil seabed, which focuses on testing the performance of the model in describing the interaction between the pipeline and the soil during the impact process; (2) an impact experiment of a pipeline with corrosion defects installed on a rigid seabed, which focuses on testing the performance of the model in describing the effect of corrosion defects on the structural response under impact load.

[0047] 1.1 Impact experiment of defect-free pipeline on flexible soil seabed The experimental arrangement is shown in Figure 2 The pipeline sample is installed on the upper surface of the clay seabed, and the two ends are fixed and constrained. The outer diameter of the pipeline sample is D p ​​​= 60 mm, wall thickness of t p = 5 mm, length of l p = 1000 mm, yield strength of steel material of s y = 234 MPa, ultimate tensile strength of s uts = 451 MPa. The cohesion of seabed soil is c = 4.9 kPa. The impact load is applied at the mid-span of the pipeline by a wedge-shaped falling object with a mass of m = 300 kg and a velocity of v imp = 4.429 m·s -1 . A series of strain gauges (SG1-SG3) are arranged at the mid-span region of the pipeline to record the axial strain time history curve of the pipeline specimen during the impact process.

[0048] According to the above experimental arrangement and related parameters, the NFEA model established in this embodiment is used to numerically reconstruct the experiment. Figure 3 The comparison of the axial strain time history curve obtained by NFEA calculation and analysis and the measured results is given in the following table. It can be seen that the comparison results of the strain time history curve at strain gauge SG2 are relatively ideal, and the time history curves obtained by the two analysis methods are highly consistent in trend and value, and the relative error of the maximum strain value is 7.7%. In comparison, the time history curves obtained by the two methods at strain gauge SG1 have some deviation, although they are consistent in trend, but in value, the strain result of the NFEA model is higher than the experimental result, and the maximum strain relative error is 18.2%. Since strain gauge SG1 is located at the impact center, the structural response strength at this position is high, and it involves complex pipe-soil interaction and local buckling of the pipeline, so the strain rate effect of steel and soil materials has a great influence on the structural failure behavior during the impact process. The parameters in the NFEA model describing the above strain rate effect are difficult to calibrate due to lack of data, which ultimately leads to this error. In comparison, strain gauge SG2 is relatively far from the impact center, the structural response strength at this position is low, and the effect of strain rate effect on structural failure behavior gradually slows down, so the numerical analysis result gradually approaches the true experimental result. The above comparison and analysis show that the NFEA model constructed in this embodiment can reasonably describe the influence mechanism of pipe-soil interaction on the structural response under impact load, and accurately predict the failure behavior of the pipeline under the condition of flexible seabed soil.

[0049] 1.2 Impact experiment of pipeline with corrosion defects under rigid seabed condition The experimental arrangement is shown in Figure 4As shown in Figure 2, the pipe sample is placed at the center of the rigid seabed surface and is fixed at both ends. The outer diameter, wall thickness and length of the pipe sample are D p = 60 mm, t p = 5 mm, l p = 500 mm. The elastic modulus, yield strength and ultimate tensile strength of pipeline steel materials are respectively E p = 199.81 GPa, s y = 411 MPa, s uts = 443 MPa. The corrosion defect is located at the center of the outer surface of the top of the pipeline mid-span. Its shape is rectangular, and its depth, length and width are d d = 3 mm, l d = 30mm, w d = 12 mm. The impact load is given by m = 300 kg, and the impact velocity is v imp = 4.429 m·s -1 A wedge-shaped drop is applied to the mid-span of the pipeline in a free-fall manner. A series of strain gauges (SG1(1 ’ ) – SG4(4 ’ )) is used to record the strain history curve of the pipe sample during the impact process, where SG1(1 ’ ) and SG3(3 ’ ) is used to measure axial strain, SG2(2 ’ ) and SG4(4 ’ ) is used to measure hoop strain. To eliminate accidental errors, three impact tests were conducted under the same experimental configuration, and three sets of strain history data were recorded.

[0050] According to the above experimental arrangement and related parameters, the NFEA model constructed in this embodiment is used to numerically reconstruct the experiment. Figure 5Comparisons between the strain time histories calculated by the NFEA model and measured results are presented for different strain gauge locations (axial strain at SG1 and SG3; hoop strain at SG2 and SG4). Overall, the results obtained by the two methods agree well. At strain gauges SG1, SG2, and SG4, the strain time histories calculated by the NFEA model show consistent trends with the experimental results, and the strain values ​​fall within the range of the three sets of measured strain data. For strain gauge SG3, the strain time histories calculated by the NFEA model agree well with the measured results, but the strain values ​​are slightly larger than the experimental results, exceeding the range of the three sets of measured strain data. Due to the sharp shape of the wedge-shaped drop, the impact load exerted by the drop causes significant localized buckling of the pipeline, resulting in significant unit deformation and a significant increase in the compressive strain at the impact center. In this case, to meet the structural deformation coordination condition, the axial strain in the region adjacent to the impact center is enhanced through stress redistribution to balance the compressive strain. Therefore, the axial strain in the adjacent region SG3, as analyzed by the numerical model, is larger. From the above comparison and analysis, it can be seen that the NFEA model constructed in this embodiment can effectively capture the structural failure behavior characteristics under the action of corrosion defects and reasonably describe the impact damage degree of the pipeline.

[0051] The following is an example of pipeline damage prediction application. A typical pipeline with corrosion defects in an actual submarine pipeline project is used as the research object. The damage prediction method proposed in this embodiment is used to predict its damage under impact load to verify the practicality of this method. The geometric dimensions of the submarine pipeline and the mechanical parameters of the steel material are listed in Table 1. The corrosion defect is located at the center of the outer surface of the top of the pipeline mid-span. Its shape is rectangular, and the range of geometric dimensions is listed in Table 2. The impact load is caused by a certain impact velocity. v imp And the quality is m foj = A 3000 kg spherical drop is applied to the middle of the pipe span, v imp and its corresponding impact energy E imp The value range of is shown in Table 3.

[0052] Based on the above parameters, the pipeline damage prediction method proposed in this embodiment is used to predict structural damage. The degree of structural damage is characterized by the ellipticity of the pipeline mid-span section. Figure 6 The predicted ellipticity value obtained by the prediction model is given e ova,pre The actual ellipticity value calculated by the NFEA model in the dataset e ovalinear regression graphs (Fig. (a) corresponds to the training data set; Fig. (b) corresponds to the test data set). It can be seen that for both the training data set and the test data set, the data points are closely distributed on the ideal straight line e ova,pre = e ova around, and the fitted straight line of the data points is nearly coincident with the ideal straight line, indicating that the prediction result is in good agreement with the true value, and the correlation is strong. More specifically, as can be seen from the error statistical indicators listed in Table 4, the determination coefficient, the correlation coefficient, and the mean absolute percentage error of the prediction model under the conditions of the two data sets are maintained in the ranges of R 2 = [0.987, 0.996], R = [0.994, 0.998], MAPE = [1.82%, 2.82%], indicating that the robustness and generalization ability of the model are high, and the prediction performance does not fluctuate significantly due to changes in the data set and the corresponding prediction domain. The above example application and performance analysis show that the damage prediction model constructed in this embodiment has high fitting ability for known data and high prediction ability for unknown data, can realize efficient and reasonable estimation of pipeline damage under the conditions of given impact energy and corrosion defect size, and meets the engineering design requirements, thus proving the feasibility and practical value of the method of the present application.

[0053] Table 1 Geometric dimensions of target submarine pipeline and mechanical parameters of steel material

[0054] Table 2 Geometric dimensions of corrosion defects

[0055] Note: (a) D p , t p The values are consistent with those in Table 1. Table 3 Impact velocity and impact energy

[0056] Table 4 Error indicators of pipeline damage prediction model

[0057] It should be noted that Figure 7 is the implementation flowchart of this embodiment.

[0058] This embodiment provides a computer program product, comprising a computer program which, when executed by a processor, implements the steps of the pipeline damage prediction method under the combined action of impact load and corrosion defects described above.

[0059] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A pipeline damage prediction method under the combined action of impact load and corrosion defect, characterized in that: The following steps are involved: S1: Construct a nonlinear finite element analysis model based on the corrosion defect size information, pipeline geometry information, seabed soil geometry and falling object geometry characteristics; S2: constructing a sample space set based on key factors using the nonlinear finite element analysis model; S3: Constructing a pipeline damage prediction model, using the sample space set to train the pipeline damage prediction model to obtain a trained pipeline damage prediction model, and using the trained pipeline damage prediction model to perform prediction to obtain a prediction result.

2. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 1 is characterized in that: Step S1 specifically includes: S11: Construct a mesh model of impact falling object-corrosion defect-pipeline-soil based on the corrosion defect size information, pipeline geometry information, seabed soil geometry and falling object geometry characteristics; S12: Based on the impact falling object-corrosion defect pipeline-soil mesh model, a nonlinear finite element analysis model is constructed and solved to obtain a solution result.

3. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 2 is characterized in that: Step S12 specifically includes: S121: Constructing a nonlinear finite element analysis model based on the impact falling object-corrosion defect pipeline-soil mesh model, wherein the nonlinear finite element analysis model includes a control equation; S122: Solving the control equation using a field variable iteration algorithm and an explicit central difference algorithm to obtain a solution result.

4. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 3 is characterized in that: The control equation is as follows: ， in, 、 、 They are the unit node mass matrix, stiffness matrix and external load matrix in the nonlinear finite element analysis model; 、 are the unit node acceleration matrix and displacement matrix respectively.

5. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 3 is characterized in that: The field variable iteration algorithm is as follows: ， ， ， ， ， ， in, express Moment Undrained shear strength at each soil unit integration point; represents the function that characterizes the undrained shear strength with the field variable value as the independent variable, and Indicates the The field variable values ​​at the start and end of the current calculation time step at each soil unit integration point, , They represent the end time of the current calculation time step. The maximum principal strain of each soil element and its increment, and Indicates the starting time of the current calculation time step. The maximum and minimum principal strains of each soil element, , They represent the end time of the current calculation time step. The minimum principal strain of each soil element and its increment, and and Respectively represent the last calculation time step The variables in the solution correlation variables of each soil unit are the maximum and minimum principal strains and the cumulative plastic strain; and and Respectively represent the current calculation time step The variables in the solution correlation variables of each soil unit are the maximum and minimum principal strains and the cumulative plastic strain; and Indicates the start and end time of the current calculation time step The accumulated plastic strain of each soil element is Indicates the end time of the current calculation time step. Maximum shear rate per soil unit; Indicates the time increment of the current calculation time step; The end time of the current calculation time step Soil strength factor value at each soil unit integration point; is the ratio of the shear strength of the soil in the fully reshaped state to its initial value; is the cumulative plastic strain corresponding to a 95% decrease in soil strength due to remodeling; is the strain rate factor; represents the reference shear rate.

6. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 3 is characterized in that: The explicit central difference algorithm is as follows: ， ， ， ， in, 、 、 、 Respectively represent the unit node displacement, velocity, acceleration and external load matrix of the nonlinear finite element analysis model, and their subscripts , , Respectively represent the midpoint of the previous calculation time step, the midpoint of the current calculation time step and the end time, and Represent the time increment of the previous and current calculation time steps respectively; , , , They represent the unit node stiffness, mass, geometric relationship and physical relationship matrix of the nonlinear finite element analysis model respectively.

7. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 1 is characterized in that: Step S2 specifically includes: S21: Constructing input dataset based on key factors S22: obtaining structural response data using a nonlinear finite element analysis model according to the input data set, and constructing an output data set according to the structural response data; S23: Construct a sample space set according to the input data set and the output data set.

8. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 1 is characterized in that: Step S3 specifically includes: S31: Constructing a pipeline damage prediction model based on a multilayer perceptron framework; S32: Divide the sample space set according to a preset ratio and normalize it to obtain a training set and a test set; S33: training the pipeline damage prediction model using a gradient descent algorithm based on the training set to obtain a trained pipeline damage prediction model; S34: Using the trained pipeline damage prediction model to predict the test set, and obtain a prediction result.

9. The pipeline damage prediction method under the combined action of impact load and corrosion defect according to claim 8 is characterized in that: The pipeline damage prediction model includes an input layer, a hidden layer, and an output layer. The input layer includes four sensors, which are used to input corrosion defect depth, corrosion defect length, corrosion defect width, and impact energy respectively. The hidden layer includes 20 sensors, which are used to coordinate the sensors between the previous and next layers to build a nonlinear mapping system to achieve deconstruction and mapping of the pipeline structure damage mechanics analysis system. The output layer includes sensors for outputting the ellipticity of the pipeline mid-span cross-section.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the pipeline damage prediction method under the combined action of impact load and corrosion defect are implemented.

Images