Method for predicting strength of k-type pipe joint of jacket based on error back propagation

Abstract

This invention provides a method for predicting the strength of K-type pipe nodes in jacket structures based on error backpropagation. The method mainly includes the following steps: (1) establishing a finite element model of the K-type pipe node; (2) writing a command stream to achieve parameterized batch modeling of the pipe node using finite element methods; (3) preparing and preprocessing the data; (4) establishing a pipe node strength prediction model based on a neural network; and (5) implementing the pipe node strength prediction model using programming. This invention is logically conceived and enables direct prediction of the strength range using the numerical values ​​of the pipe node design parameters, thereby avoiding extensive finite element analysis.

Description

Technical Field

[0001] This invention relates to the field of marine platform pipe joint manufacturing technology, specifically to a method for predicting the strength of K-type pipe joints of jackets based on error backpropagation. Background Technology

[0002] The construction and innovative development of offshore platform engineering are crucial foundations for the marine industry. Many offshore platforms in my country have been in service for years, and pipe structures are widely used. Among these, pipe joints are a key technical focus and challenge in platform structural design and construction. Pipe joint structures experience significantly higher stresses than other parts, and the weld seams are concentrated. Load transfer between components must occur through pipe joints, making them the most vulnerable areas for cracking and failure. Their design must meet certain strength and stiffness requirements while considering various parameters: for example, for K-type pipe joints, the cross-sectional diameters and wall thicknesses of chords and struts, the spacing between struts, and the included angles, etc. Since pipe joint optimization involves multiple parameters, finding the optimal parameter combination through experiments is impractical. Therefore, rapidly and accurately establishing a pipe joint structural model is the foundation and prerequisite for pipe joint analysis and research.

[0003] In the analysis of stress at pipe joints on offshore platforms, the finite element method (FEM) is currently the most widely used method. It can, to some extent, replace experiments, allowing researchers and designers to effectively complete optimization work at a lower cost. While the FEM can predict pipe joint strength during optimization, each combination of design parameters requires modeling, analysis, and result extraction. Clearly, optimization based on the FEM requires multiple numerical simulations and result comparisons, consuming a significant amount of time to find a relatively good combination. This drawback is particularly pronounced when performing three-dimensional model analysis.

[0004] In conclusion, it is necessary to further innovate existing technologies. Summary of the Invention

[0005] To address the technical problems existing in the background art mentioned above, this invention proposes a pipe node strength prediction method based on error backpropagation. Its concept is reasonable and realizes the direct prediction of the strength range using the numerical values ​​of the design parameters of the jacket structure pipe node, thereby avoiding a large amount of finite element analysis.

[0006] To address the aforementioned technical problems, this invention provides a method for predicting the strength of K-type pipe nodes in jacket structures based on error backpropagation, which mainly includes the following steps:

[0007] (1) Establish the finite element model of the K-type pipe node;

[0008] (2) Write command flows to implement finite element parameterized batch modeling of pipe nodes;

[0009] (3) Perform data preparation and preprocessing;

[0010] (4) Establish a pipe node strength prediction model based on neural networks;

[0011] (5) Implement the pipe node strength prediction model using programming.

[0012] The aforementioned method for predicting the strength of K-type pipe nodes in jacket platforms based on backpropagation of errors, wherein the specific process of step (1) is as follows: First, based on the commonly used K-type pipe nodes in jacket platforms, the geometric model of the pipe node is constructed using finite element software, and chord members and struts are established, with the angle between the chord members and struts being... Then, based on the completion of the geometric model construction, the finite element model mesh is generated. According to the actual situation and calculation requirements, the regular quadrilateral mesh is manually generated for analysis to obtain the model mesh generation results. Finally, the material properties, element types, boundary conditions and load application conditions are set in the finite element software.

[0013] The method for predicting the strength of K-type pipe nodes of the jacket based on backpropagation of error, wherein the specific process of step (2) is as follows: use a set of parameters to define the geometric dimensions of the pipe node model, constrain the relationship between the geometric dimensions, realize the topological relationship between the geometric dimensions, determine the relevant parameters of the geometric model that need to be changed, and display the stress cloud map after the finite element model is established and Solve is completed.

[0014] The aforementioned method for predicting the strength of K-type tube nodes in jacket structures based on error backpropagation, wherein the geometric dimensions of the tube node model can be defined according to dimensionless parameters:

[0015] (1);

[0016] (2);

[0017] (3);

[0018] (4);

[0019] In the above formulas (1)-(4), L represents the length of the chord tube; D represents the outer diameter of the chord tube; d represents the outer diameter of the strut tube; T represents the wall thickness of the chord tube; t represents the wall thickness of the strut tube; α is the ratio of the length of the chord tube to its outer diameter, reflecting the flexibility of the chord tube; β is the ratio of the outer diameter of the chord tube to the diameter of the strut tube, reflecting the load transfer and stress distribution; γ is the ratio of the outer diameter of the chord tube to the wall thickness of the chord tube, reflecting the radial stiffness of the chord tube; τ is the ratio of the wall thickness of the strut tube to the wall thickness of the chord tube, reflecting the relative bending stiffness of the chord tube and the strut tube.

[0020] The method for predicting the strength of K-type pipe nodes of the jacket structure based on backpropagation of error, wherein the specific process of step (3) is as follows: after running the example model in batches, extract the finite element results calculated by the finite element analysis of the model, namely the combination of different design parameters of the pipe node and the maximum stress value obtained, and organize the extracted finite element results; normalize the finite element results and assign a small initial value to avoid the data being 0.

[0021] The aforementioned method for predicting the strength of K-type tube nodes in a jacketed structure based on backpropagation of errors includes a maximum stress value categorized into three types: "safe," "uncertain," and "dangerous," with 0, 1, and 2 representing "safe," "uncertain," and "dangerous," respectively.

[0022] The method for predicting the strength of K-type pipe nodes of the jacket structure based on backpropagation of error, wherein: step (4) is to predict the structural strength of the pipe node based on the finite element results, and then use the backpropagation algorithm to establish a pipe node strength prediction model, and the architecture of the pipe node strength prediction model includes an input layer, a hidden layer and an output layer.

[0023] The input layer combines the structural characteristics of pipe nodes in actual engineering practice, selects pipe node structural parameters, and uses the structural parameters as input nodes.

[0024] The hidden layer uses a single-layer hidden layer error backpropagation algorithm to predict the safety of the pipe nodes; the number of hidden layer nodes is determined by combining multiple experiments and referencing empirical formulas; the referenced empirical formulas are as follows:

[0025] (5);

[0026] (6);

[0027] (7);

[0028] In equations (5)-(7) above, m represents the number of hidden layer nodes, n represents the number of input layer nodes, l represents the number of output layer nodes, and the constant a = 1~10;

[0029] The output layer selects three safety states as output: "safe", "uncertain", and "dangerous".

[0030] The method for predicting the strength of K-type pipe nodes in a jacketed structure based on backpropagation of errors, wherein: the pipe node strength prediction model in step (5) can be divided into six parts: initializing the network, training the network, querying the network, assigning variable values, importing data, and calculating the accuracy.

[0031] The specific process of step (5) is as follows: First, the neural network model is trained. During the training process of the neural network model, the network weights are continuously adjusted based on the learned sample data and the result error is backpropagated to complete the training. Finally, the trained model is verified and tested.

[0032] The initialization of the network requires setting the number of input layer nodes, hidden layer nodes, and output layer nodes of the neural network model. The number of these nodes defines the shape and size of the neural network model. In addition, the initialization of the network also requires setting the learning rate, initializing the connection weights, and defining the activation function. The initial connection weights are randomly generated, and the random numbers follow a normal distribution with a mean of 0 and a standard deviation of -0.5 times the number of hidden layer nodes (i.e., the initial weights between the input and hidden layers) and -0.5 times the number of output layer nodes (i.e., the initial weights between the hidden and output layers).

[0033] The activation function is the sigmoid function, and the formula is as follows:

[0034] (8);

[0035] The training network consists of two parts. The first part calculates the output for a given training sample. The second part compares the calculated output with the expected output, optimizes and improves the network weights based on the difference, and calculates the loss function. Regarding each weight parameter and bias The gradient is used to update each weight parameter. and bias To minimize the loss;

[0036] For each layer First, calculate the loss function relative to the activation value. Partial derivatives:

[0037] (9);

[0038] Next, the output relative to the linear transformation is calculated. Partial derivatives:

[0039] (10);

[0040] In the above formula (10) The derivative of the activation function;

[0041] Then, update the weights and biases of each layer:

[0042] (11);

[0043] (12);

[0044] The weights and biases are updated using gradient descent.

[0045] (13);

[0046] (14);

[0047] In the above formula (13) The learning rate;

[0048] The query network needs to transmit input signals from the input layer nodes. All inputs are combined with all correct link weights to generate a combined, regulated signal matrix, which is then transmitted to each hidden layer node. The hidden layer uses link weights to adjust the signal before transmitting it to the output layer. The output layer uses link weights to gradually update the signal according to the backpropagation algorithm to minimize the prediction error, and finally outputs the result. When data from the input layer is fed to a given hidden layer node or output layer node, the sigmoid activation function is applied to suppress signals from these nodes while adjusting the data from the input layer using link weights.

[0049] The variables are assigned values ​​according to the backpropagation network structure. The neural network model has 5 input layer nodes, 4 hidden layer nodes, 3 output layer nodes, and a learning rate of 0.1.

[0050] The imported data requires the creation of a neural network object, importing the training dataset to train the network and importing the test dataset to test the network. If the training data is insufficient, the method of iterating through all sample data multiple times can be adopted, that is, increasing the number of training rounds in the neural network model to train the network and improve the training effect.

[0051] The output of the pipe node strength prediction model is the safety status of the pipe node under different design parameters. The maximum value represents the predicted status. Since the output is three numbers in a one-dimensional array between 0 and 1, some code is added to intuitively represent the training effect by means of accuracy measurement.

[0052] The method for predicting the strength of K-type tube nodes in a ductwork system based on backpropagation of errors includes the following steps for training the neural network model: First, initialize the weights and biases of the neurons in the network and select an input sample. Next, perform forward propagation to calculate the outputs of the hidden and output layers, and then calculate the error based on the difference between the expected and actual outputs. Then, propagate the error to each hidden layer through backpropagation, calculate the error of the hidden layer, and adjust the weights and biases. Finally, check if the training of the current sample is complete. If not, continue adjusting; otherwise, increase the number of iterations and select the next sample, repeating the process. Finally, the entire training process ends when all samples have been trained.

[0053] By adopting the above technical solution, the present invention has the following beneficial effects:

[0054] This invention presents a reasonable method for predicting pipe node strength based on backpropagation of errors. It achieves safe prediction of pipe node strength and can be used in the design stage of pipe nodes. It simulates multiple parameters, reduces the number of experiments, and provides a faster assessment of the structural strength of pipe nodes. It can effectively solve the problem of cumbersome finite element analysis in the current design stage of pipe nodes. At the same time, it can effectively improve efficiency, reduce costs, and accelerate the design and development of pipe nodes.

[0055] This invention addresses the need to consider numerous design parameters during pipe node design by employing a backpropagation (BP) neural network to establish a pipe node strength prediction model. This model is suitable for handling information processing problems involving complex and fuzzy conditions that require simultaneous consideration of multiple factors. The model learns from a small set of representative samples and continuously adjusts its weights. The numerous adjustable parameters within the network offer high flexibility and strong fault tolerance, enabling it to analyze and handle relatively complex nonlinear systems, such as prediction and control problems. Attached Figure Description

[0056] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0057] Figure 1 This is a flowchart illustrating the pipe node strength prediction method based on error backpropagation in an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram of a K-type pipe node model in an embodiment of the present invention;

[0059] Figure 3 This illustrates the impact of the number of hidden layer nodes on prediction accuracy in this embodiment of the invention.

[0060] Figure 4 This is a schematic diagram of the BP network training process in an embodiment of the present invention. Detailed Implementation

[0061] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The present invention will be further explained below with reference to specific embodiments.

[0063] like Figure 1 As shown in the figure, this embodiment provides a pipe node strength prediction method based on error backpropagation, which includes the following steps:

[0064] S100. Establish the finite element model of the K-type pipe node.

[0065] Based on the commonly used K-type pipe nodes in jacket platforms, a geometric model of the pipe node is constructed using finite element software, establishing chords and struts. The angle between the chord and strut is... ,like Figure 2 As shown. The pipe node model established in this embodiment is as follows. Figure 2 As shown, the length is 600 mm extending outward from the intersection of the chord and the strut axis, and the angle between the chord and the strut is θ.

[0066] The K-type node symmetrical structure established in this embodiment has the same values ​​for parameters such as the angle of the two struts and the wall thickness of the strut tube. The geometric parameters are set as shown in Table 1-1.

[0067]

[0068] Then, based on the completion of the geometric model construction, the finite element model mesh is generated. According to the actual situation and calculation requirements, the regular quadrilateral mesh is manually generated for analysis. While ensuring the calculation accuracy, the calculation efficiency is effectively improved, and the model mesh generation result is obtained.

[0069] Finally, in the finite element software, material properties (including yield strength, elastic modulus, Poisson's ratio, stress-strain relationship, material yield criterion, etc., without considering the influence of residual stress in the weld in the joint area on the pipe joint), element type (selecting SOLID185), boundary conditions, and load application are set to achieve state simulation of the pipe joint. In this embodiment, the material selected for the finite element model of the pipe joint is CCSD36, and the yield strength is... An ideal elastoplastic model was chosen for the stress-strain relationship. The elastic modulus and Poisson's ratio were defined, and the material yield criterion followed the Von Mise yield criterion and related flow laws. The influence of residual weld stress in the nodal area on the pipe joint was not considered. SOLID185 was selected for analysis. This element is defined by 8 nodes, each with 3 degrees of freedom for translation along the x, y, and z directions. The specific boundary conditions are as follows: the left end of the chord is a fixed hinge support, and the right end is a sliding hinge support, allowing sliding along the chord axis; one end of the strut is welded to the chord, and the other end is a cantilever. Under load application, axial force and bending moment are applied to the two cantilever ends respectively, thus simulating the state of the pipe joint. A node was established at the center of the strut end face using MASS21, rigidly coupled to all nodes on the strut end face. The load transferred from the upper part to the pipe joint was simulated by applying a load to this point.

[0070] S200. Write a command flow to implement finite element parametric batch modeling of pipe nodes.

[0071] A set of parameters is used to define the geometric dimensions of the pipe node model and constrain the relationships between the dimensions to realize the topological relationships between the geometric dimensions. The relevant parameters of the geometric model that need to be changed are determined, and the stress cloud diagram is displayed after the finite element model is established and Solve is completed.

[0072] The geometric dimensions of the pipe node model can be defined based on dimensionless parameters:

[0073] (1);

[0074] (2);

[0075] (3);

[0076] (4);

[0077] In the above formulas (1)-(4), L represents the length of the chord tube in mm; D represents the outer diameter of the chord tube in mm; d represents the outer diameter of the strut tube in mm; T represents the wall thickness of the chord tube in mm; t represents the wall thickness of the strut tube in mm; α is the ratio of the length of the chord tube to its outer diameter, reflecting the flexibility of the chord tube; β is the ratio of the outer diameter of the chord tube to the diameter of the strut tube, reflecting the load transfer and stress distribution; γ is the ratio of the outer diameter of the chord tube to the wall thickness of the chord tube, reflecting the radial stiffness of the chord tube; τ is the ratio of the wall thickness of the strut tube to the wall thickness of the chord tube, reflecting the relative bending stiffness of the chord tube and the strut tube.

[0078] Referring to the Design Specification for Offshore Steel Structures published by Det Norske Veritas (DNV) in 2010, the effective range of geometric dimensions for pipe nodes on offshore platforms is: 4≤α≤40; 4≤β≤40; 4≤γ≤40; 4≤τ≤40.

[0079] SOLID185 elements were selected for modeling. A top-down approach was used to directly create the volume, and then various Boolean operations were used to generate the geometric model of the pipe nodes. After appropriate meshing, the finite element model was obtained.

[0080] The specific parameterization process is as follows:

[0081] S201. Determine the relevant parameters of the geometric model that needs to be changed, and assign values ​​to the variables;

[0082] S202. Set the analysis environment, define the element type as SOLID185, real constants, and material properties, etc.

[0083] S203. Establish the chords and struts, and use Boolean operations to obtain the geometric model of the pipe node;

[0084] S204, Mesh generation;

[0085] S205. Apply constraints to both ends of the chord, apply full constraints to the left end and X and Y direction constraints to the right end, and apply loads after rigid coupling at the cantilever end face of the strut.

[0086] S206. After completing Solve, display the stress cloud diagram.

[0087] Furthermore, after extracting the complete command flow, the command flow is parameterized, and all command flow files that need to be calculated are stored in the folder in the form of inp. Then, the corresponding batch modeling command flow is written, and the batch modeling calculation is realized by running it in the command window of the finite element software. The establishment and analysis calculation of all models are completed by changing the path command flow.

[0088] S300, Perform data preparation and preprocessing.

[0089] After running the example models in batches, the different combinations of design parameters for pipe nodes and the obtained maximum stress values ​​from the finite element analysis were extracted. The finite element results and data were then organized, totaling 110 sets. Specifically, the strength influencing factors considered included chord diameter, chord wall thickness, strut diameter, strut wall thickness, and strut angle. The maximum stress values ​​were categorized into three types: "safe," "uncertain," and "dangerous." Based on a reduction factor of 7%, this invention classifies maximum stresses less than 330 MPa as "safe," between 330 MPa and 355 MPa as "uncertain," and greater than 355 MPa as "dangerous," using 0, 1, and 2 to represent "safe," "uncertain," and "dangerous," respectively, facilitating data reading for the neural network model. The extracted data was normalized and assigned small initial values ​​to avoid zero values ​​causing the neural network to have weights that do not function effectively. The original data was processed into comparable standard sequences as much as possible, significantly improving training performance and prediction accuracy compared to the original method.

[0090] S400, Establishing a pipe node strength prediction model based on neural networks

[0091] Design a neural network model, the architecture of which includes:

[0092] In the input layer, based on the structural characteristics of pipe nodes in actual engineering, structural parameters of the pipe nodes are selected and used as input nodes. Specifically, based on the symmetrical structural characteristics of the K-type pipe node in this embodiment, several parameters that have a significant impact on the safety of the pipe node are identified: chord diameter, chord wall thickness, strut diameter, strut wall thickness, and strut angle. These five parameters are used as input nodes.

[0093] In theory, a three-layer backpropagation algorithm model can obtain any m-dimensional to n-dimensional mapping, and its computational results are superior in practical applications. This invention uses a single-layer hidden layer backpropagation algorithm to predict the safety of management nodes. The number of hidden layer nodes is generally determined by combining multiple experiments and referencing empirical formulas. The range of the number of hidden layer nodes in this invention is 3~12. The reference empirical formula is as follows:

[0094] (5);

[0095] (6);

[0096] (7);

[0097] In equations (5)-(7) above, m represents the number of hidden layer nodes, n represents the number of input layer nodes, l represents the number of output layer nodes, and the constant a = 1~10.

[0098] Specifically, an experimental method was used to determine the optimal number of hidden layer nodes. As the number of hidden layer nodes increased, the training time also increased, and the accuracy peaked at 4. Figure 3 As shown. Therefore, under the current circumstances, the optimal number of hidden layer nodes is 4.

[0099] For the output layer nodes, according to the invention method, three safety states, namely "safe", "uncertain", and "dangerous", are selected as the output.

[0100] The pipe node strength prediction model of this invention adopts the backpropagation algorithm (BP network) with the sigmoid function as the activation function, as shown in the following formula:

[0101] (8).

[0102] S500. Implement the pipe node strength prediction model using programming. The pipe node strength prediction model can be divided into six parts: initializing the network, training the network, querying the network, assigning variable values, importing data, and calculating the accuracy. The specific process is as follows:

[0103] First, train the neural network model, such as Figure 4 As shown, the state and activation value of each layer are calculated first, up to the last layer (i.e., the signal propagates forward); the error of each layer is calculated, and the error calculation process proceeds from the last layer forward (i.e., the error propagates backward); the gradient of the connection weights of each neuron is calculated; the parameters are updated according to the gradient descent rule (the goal is to reduce the error, and the above steps are iterated until the stopping criterion is met); during the training of the neural network model, the network weights are continuously adjusted based on the learned sample data through backpropagation of the error, thereby completing the training; finally, the trained model is verified and tested.

[0104] Specifically, initializing the network requires setting the number of input layer nodes, hidden layer nodes, and output layer nodes in the neural network model. The number of nodes defines the shape and size of the neural network model. In addition, initializing the network also requires setting the learning rate, initializing the link weights, and defining the activation function (using the sigmoid function). The most important part of the network is the link weights, which are used to calculate the feedforward signal and backpropagate error, and are continuously optimized using gradient descent when attempting to improve the network. In this invention, the initial link weights are randomly generated, and the random numbers follow a normal distribution with a mean of 0 and standard deviations of -0.5 times the number of hidden layer nodes (initial weights between the input and hidden layers) and -0.5 times the number of output layer nodes (initial weights between the hidden and output layers), respectively.

[0105] Specifically, training the network consists of two parts. The first part is to calculate the output for a given training sample. That is, let the input sample be X, the model parameters be W, and the target output be Y. The calculation process of the neural network is as follows:

[0106] For each layer :

[0107] ;

[0108] In the above formula The initial input is the activation value of the previous layer. ; For the first The weight matrix of the layer, It is the bias vector; For the first The linear transformation output of the layer;

[0109] The second part compares the calculated output with the expected output and optimizes the network weights based on the difference; that is, backpropagation is the process of optimizing network weights based on gradient descent. The goal of backpropagation is to calculate the loss function... Regarding each weight parameter and bias The gradient is used to update each weight parameter. and bias To minimize losses.

[0110] For each layer First, calculate the loss function relative to the activation value. Partial derivatives:

[0111] (9);

[0112] Next, the output relative to the linear transformation is calculated. Partial derivatives:

[0113] (10);

[0114] in, This is the derivative of the activation function.

[0115] Then, update the weights and biases of each layer:

[0116] (11);

[0117] (12);

[0118] The weights and biases are updated using gradient descent.

[0119] (13);

[0120] (14);

[0121] in, This refers to the learning rate.

[0122] Specifically, the query network needs to transmit input signals from the input layer nodes. All inputs are combined with all correct link weights to generate a combined, regulated signal matrix, which is transmitted to each hidden layer node. The hidden layer uses link weights to regulate the signal before transmitting it to the output layer. The output layer uses link weights to gradually update the signal according to the backpropagation algorithm to minimize the prediction error. Each link weight is gradually updated according to the backpropagation algorithm to minimize the prediction error, and finally, the signal is output from the output layer. When data from the input layer is fed to a given hidden layer node or output layer node, the sigmoid activation function is applied to suppress signals from these nodes while regulating the data from the input layer using link weights.

[0123] Specifically, the variables are assigned values ​​according to the error backpropagation network structure. The neural network model has 5 nodes in the input layer, 4 nodes in the hidden layer, and 3 nodes in the output layer, with a learning rate of 0.1.

[0124] Specifically, importing data requires creating a neural network object, importing the training dataset to train the network, and importing the test dataset to test the network. When research conditions are limited and training data is insufficient, an iterative approach can be adopted, which involves increasing the number of training epochs in the neural network model to improve training effectiveness.

[0125] Specifically, the output of the pipe node strength prediction model is the safety status of the pipe node under different design parameters, with the maximum value representing the predicted state. Since the output is a one-dimensional array of three numbers in the range of 0 to 1, which is not convenient for researchers to study and analyze, some code is added to adopt an accuracy measurement method to intuitively represent the training effect.

[0126] like Figure 4As shown, the process of training the neural network model described above is as follows: First, initialize the weights and biases of the neurons in the network, and select an input sample; then, perform forward propagation to calculate the outputs of the hidden layers and the output layer, and then calculate the error based on the difference between the expected output and the actual output; propagate the error to each hidden layer through backpropagation, calculate the error of the hidden layer, and adjust the weights and biases; then check whether the training of the current sample is complete. If not, continue adjusting; otherwise, increase the number of iterations and select the next sample, repeating the above process; finally, when all samples have been trained, the entire training process ends.

[0127] This embodiment establishes an artificial neural network for strength analysis and safety prediction based on different design parameters of K-type pipe nodes. Using 110 randomly ordered sample data as training data and 30 test data, the artificial neural network is analyzed and optimized. After training with 5 input nodes, 3 output layer nodes, 4 hidden layer nodes, and a learning rate of 0.1, the results show that the safety prediction accuracy for K-type pipe nodes can reach up to 92.86%, meeting the engineering error requirement of 10%. In practical engineering, using a BP neural network model to predict safety instead of finite element analysis using finite element software can significantly improve the efficiency of pipe node design and construction.

[0128] This invention has a reasonable concept and realizes the direct prediction of the strength range using the numerical values ​​of the design parameters of the K-type pipe node of the jacket, thereby avoiding the need for multiple numerical simulations and result comparison analyses based on the finite element method, which consumes a lot of time to find a relatively good combination. It can effectively improve efficiency, reduce costs, and accelerate the design and development of K-type pipe nodes.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the strength of K-type pipe nodes in a jacketed structure based on error backpropagation, characterized in that, The main steps include: (1) Establish the finite element model of the K-type pipe node; the specific process is as follows: first, based on the commonly used K-type pipe node of the jacket platform, use finite element software to construct the geometric model of the pipe node, establish the chord and strut, and the angle between the chord and strut is θ; then, based on the completion of the geometric model construction, perform finite element model mesh generation, manually generate regular quadrilateral mesh according to the actual situation and calculation requirements, and analyze to obtain the model mesh generation result; finally, set the material properties, element type, boundary conditions and load application in the finite element software; (2) Write a command flow to implement finite element parametric batch modeling of pipe nodes; The specific process is as follows: use a set of parameters to define the geometric dimensions of the pipe node model, constrain the relationship between the geometric dimensions, realize the topological relationship between the geometric dimensions, determine the relevant parameters of the geometric model that need to be changed, and display the stress cloud diagram after the finite element model is established and Solve is completed. (3) Data preparation and preprocessing; the specific process is as follows: after running the example model in batches, extract the finite element results of the model finite element analysis calculation, namely the different design parameter combinations of the pipe node and the maximum stress value obtained, organize the extracted finite element results; normalize the finite element results and assign a small initial value to avoid the data being 0; (4) Establish a pipe node strength prediction model based on neural networks; Specifically, the strength of the pipe node structure is predicted based on the finite element results, and then the pipe node strength prediction model is established using the backpropagation algorithm. The architecture of the pipe node strength prediction model includes an input layer, a hidden layer, and an output layer. The input layer combines the structural characteristics of pipe nodes in actual engineering practice, selects pipe node structural parameters, and uses the structural parameters as input nodes. The hidden layer uses a single-layer hidden layer error backpropagation algorithm to predict the safety of the pipe nodes; the number of hidden layer nodes is determined by combining multiple experiments and referencing empirical formulas; the referenced empirical formulas are as follows: m < n-1 (6); m = log₂n (7); In equations (5)-(7) above, m represents the number of hidden layer nodes, n represents the number of input layer nodes, l represents the number of output layer nodes, and the constant a = 1 to 10; The output layer selects three safety states as output: "safe", "uncertain", and "dangerous". (5) Implement the pipe node strength prediction model using programming.

2. The method for predicting the strength of K-type pipe nodes in jacket structures based on error backpropagation as described in claim 1, characterized in that, The geometric dimensions of the pipe node model can be defined based on dimensionless parameters: In equations (1)-(4) above, L represents the length of the chord tube; D represents the outer diameter of the chord tube; d represents the outer diameter of the strut tube; T represents the wall thickness of the chord tube; t represents the wall thickness of the strut tube; α is the ratio of the length of the chord tube to its outer diameter, reflecting the flexibility of the chord tube; β is the ratio of the outer diameter of the chord tube to the diameter of the strut tube, reflecting the load transfer and stress distribution; γ is the ratio of the outer diameter of the chord tube to the wall thickness of the chord tube, reflecting the radial stiffness of the chord tube; τ is the ratio of the wall thickness of the strut tube to the wall thickness of the chord tube, reflecting the relative bending stiffness of the chord tube and the strut tube.

3. The method for predicting the strength of K-type pipe nodes in jacketed structures based on error backpropagation as described in claim 1, characterized in that: The maximum stress value is divided into three categories: "safe", "uncertain", and "dangerous", with 0, 1, and 2 representing "safe", "uncertain", and "dangerous" respectively.

4. The method for predicting the strength of K-type pipe nodes in a jacket structure based on error backpropagation as described in claim 1, characterized in that: The pipe node strength prediction model in step (5) can be divided into six parts: initializing the network, training the network, querying the network, assigning variable values, importing data, and calculating the accuracy. The specific process of step (5) is as follows: First, the neural network model is trained. During the training process of the neural network model, the network weights are continuously adjusted based on the learned sample data and the back propagation of the result error to complete the training. Finally, the trained model is verified and tested. The initialization of the network requires setting the number of input layer nodes, hidden layer nodes, and output layer nodes of the neural network model. The number of these nodes defines the shape and size of the neural network model. In addition, the initialization of the network also requires setting the learning rate, initializing the connection weights, and defining the activation function. The initial connection weights are randomly generated, and the random numbers follow a normal distribution with a mean of 0 and a standard deviation of -0.5 times the number of hidden layer nodes (i.e., the initial weights between the input and hidden layers) and -0.5 times the number of output layer nodes (i.e., the initial weights between the hidden and output layers). The activation function is the sigmoid function, and the formula is as follows: The training network consists of two parts. The first part calculates the output for a given training sample. The second part compares the calculated output with the expected output, optimizes and improves the network weights based on the difference, and calculates the loss function L with respect to each weight parameter W. (l) and bias b (l) The gradient is used to update each weight parameter W. (l) and bias b (l) To minimize the loss; For each layer l, first calculate the loss function relative to the activation value A. (l) Partial derivatives: Next, calculate the output Z relative to the linear transformation. (l) Partial derivatives: d (l) =d (l+1) ·g′(Z (l) ) (10); In the above formula (10), g′(Z) (l) () is the derivative of the activation function; Then, update the weights and biases of each layer: The weights and biases are updated using gradient descent. In equation (13) above, η is the learning rate; The query network needs to transmit input signals from the input layer nodes. All inputs are combined with all correct link weights to generate a combined, regulated signal matrix, which is then transmitted to each hidden layer node. The hidden layer uses link weights to adjust the signal before transmitting it to the output layer. The output layer uses link weights to gradually update the signal according to the backpropagation algorithm to minimize the prediction error, and finally outputs the result. When data from the input layer is fed to a given hidden layer node or output layer node, the sigmoid activation function is applied to suppress signals from these nodes while adjusting the data from the input layer using link weights. The variables are assigned values ​​according to the backpropagation network structure. The neural network model has 5 input layer nodes, 4 hidden layer nodes, 3 output layer nodes, and a learning rate of 0.

1. The imported data requires the creation of a neural network object, importing the training dataset to train the network and importing the test dataset to test the network. If the training data is insufficient, the method of iterating through all sample data multiple times can be adopted, that is, increasing the number of training rounds in the neural network model to train the network and improve the training effect. The output of the pipe node strength prediction model is the safety status of the pipe node under different design parameters. The maximum value represents the predicted status. Since the output is three numbers in a one-dimensional array between 0 and 1, some code is added to intuitively represent the training effect by means of accuracy measurement.

5. The method for predicting the strength of K-type pipe nodes in a jacket structure based on error backpropagation as described in claim 4, characterized in that, The process of training the neural network model is as follows: First, initialize the network weights and neuron biases, and select an input sample; Next, forward propagation is performed to calculate the outputs of the hidden and output layers, and then the error is calculated based on the difference between the expected and actual outputs. The error is propagated to each hidden layer through backpropagation, the error of the hidden layer is calculated, and the weights and biases are adjusted. Then check if the training of the current sample has ended. If not, continue to adjust; otherwise, increase the number of iterations and select the next sample, repeating the above process. Finally, once all samples have been trained, the entire training process is complete.

Images