Method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower drum base

Abstract

The invention relates to the technical field of engineering structure health monitoring and artificial intelligence, and discloses an offshore wind turbine tower drum base three-dimensional stress distribution prediction and visualization method, which comprises the following steps: generating batch working conditions by presetting a bending moment range and an interval, and calling a finite element software automation script to execute static analysis; node stress values are extracted and integrated into a structured training data set; customizing neural network training: constructing a full-connection network, and training a model by adopting a standardized preprocessing and early stop mechanism; generating an interactive cloud picture: generating a 3D stress cloud picture HTML (Hypertext Markup Language) file supporting rotation / zooming / data query based on a spatial interpolation algorithm and a Plotly tool, and automatically marking the coordinate and the numerical value of a maximum stress point; and end-to-end system deployment: packaging as a Flask Web application, and inputting a bending moment value by a user to generate an interactive cloud picture in real time. The stress analysis period is greatly shortened, the engineering application threshold is reduced, and rapid evaluation and remote cooperation of the offshore wind turbine tower drum base can be efficiently supported.

Description

Technical Field

[0001] This invention relates to the fields of engineering structure health monitoring and artificial intelligence technology, specifically a method for predicting and visualizing the three-dimensional stress distribution of offshore wind turbine tower foundations. Background Technology

[0002] In the structural design and safety assessment of offshore wind turbines, stress distribution analysis of the tower base is a crucial step. Traditional methods rely on finite element analysis (FEA) software for simulation calculations. This method requires establishing accurate geometric and material models, generating fine meshes, and applying boundary conditions and loads. The calculation process is complex and time-consuming, with a single simulation typically taking several hours, making it difficult to meet the demands of rapid iteration and real-time response in engineering design.

[0003] Furthermore, FEA analysis heavily relies on manual operation by professional engineers, including parameter setting, task submission, result extraction, and post-processing, resulting in high labor costs and low automation. When analyzing multiple load conditions, a large number of simulation tasks need to be repeatedly executed, leading to inefficiency and hindering the construction of load-response databases.

[0004] Although artificial intelligence technology has begun to be applied to structural mechanics prediction, existing methods mostly focus on overall performance assessment or simplified model prediction, lacking the ability to realize the transformation from load input to three-dimensional stress cloud for complex structures such as offshore wind turbine tower foundations. Figure 1 The technical solution for key generation, especially in achieving a highly consistent visualization output with finite element results while ensuring prediction accuracy, remains a technical challenge.

[0005] Document CN 117892598 A discloses an enhanced inverse finite element shape sensing reconstruction system for offshore wind turbine towers. The system's discretization generation module discretizes the tower structure to obtain discretized elements and generates shape functions. The data input module receives real-time strain measurement data. The measurement calculation module establishes a theoretical strain model based on the discretized elements and shape functions, and establishes an experimental strain model based on the real-time strain measurement data. The coupled calculation module constructs an error function based on the theoretical and experimental strain models, solves it to obtain a high-precision reconstruction result of the tower structure deformation field, and outputs the result to complete the reconstruction. However, when encountering problems, it cannot quickly respond to the stress analysis cycle.

[0006] There is an urgent need for a method that can quickly, automatically, and accurately generate three-dimensional stress cloud maps of offshore wind turbine tower foundations to improve engineering analysis efficiency, reduce labor costs, and support intelligent design and operation and maintenance.

[0007] Therefore, a new technical solution is needed to solve the above-mentioned technical problems. Summary of the Invention

[0008] To address the aforementioned issues, this invention discloses a method for predicting and visualizing the three-dimensional stress distribution of offshore wind turbine tower foundations. This method encapsulates the complex AI prediction process into a lightweight web application, eliminating the need for professional software installation and allowing operation solely through a browser. This significantly lowers the barrier to entry and makes the system suitable for rapid on-site assessments, remote collaboration, and multi-user sharing scenarios, thereby enhancing its practicality and promotional value.

[0009] The technical solution of this invention is: a method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation, comprising the following steps:

[0010] I. Acquisition and Processing of Datasets for Automated Finite Element Analysis

[0011] Step 1: Construct an equivalent finite element model of the offshore wind turbine tower base in the finite element analysis platform, determine key nodes, and apply boundary conditions and loads;

[0012] Step 2: Develop an automated execution script to control the finite element analysis platform to perform multi-condition simulations in batches, and export the stress data and corresponding load parameters of key nodes under each condition to form the original dataset;

[0013] Step 3: Extract and organize the original dataset to generate a structured training dataset containing load parameters and corresponding nodal stress values;

[0014] II. Construction and Training of Deep Learning Models

[0015] Step 1: Preprocess the training dataset, using load parameters as input features and nodal stress values ​​as output labels, and divide it into training set, validation set and test set;

[0016] Step 2: Construct a fully connected feedforward neural network model, which includes an input layer, at least two hidden layers, and an output layer. The input layer receives load parameters, and the output layer outputs the stress prediction values ​​of each key node.

[0017] Step 3: Train the model using the training set, optimize the model parameters using the validation set, evaluate the model performance using the test set, and obtain the trained stress prediction model;

[0018] III. 3D Interactive Stress Cloud Map Generation and Web Application Integration

[0019] Step 1: Receive the target load parameters input by the user, and obtain the stress prediction values ​​of each key node through the trained stress prediction model;

[0020] Step 2: Match the predicted stress values ​​with the three-dimensional coordinates of key nodes, generate a continuous stress distribution through spatial interpolation, and construct a 3D interactive stress cloud map based on visualization tools. The cloud map supports rotation, scaling, translation, and real-time viewing of stress values.

[0021] Step 3: Integrate stress prediction and 3D cloud map generation functions into the web application to enable users to input load parameters through a browser and obtain interactive stress cloud maps online.

[0022] Preferably, in step 1 of acquiring and processing the automated finite element analysis dataset, an equivalent finite element model is constructed. The offshore wind turbine tower base is discretized using plate and shell elements and beam elements, the main load-bearing components are retained, key nodes are determined, fixed boundary conditions and top bending moment loads are applied, and the equivalent finite element model is discretized using plate and shell elements PLATE and beam elements BEAM. It contains 720 key nodes, the boundary condition is that the bottom of the model is completely fixed, and the load is the bending moment load applied to the control point at the top of the model.

[0023] By adopting the above technical solution and training based on real finite metadata, the model can accurately learn the nonlinear mapping relationship between load and stress, and the prediction results are highly consistent with the simulation results, thus possessing engineering applicability.

[0024] Preferably, in step 2 of acquiring and processing the automated finite element analysis dataset, an automated script is developed to define the range and interval of bending moment load parameters. The script then iteratively executes the modeling and solving of each working condition through the batch processing interface of the finite element software, extracts the von Mises stress values ​​of key nodes, and exports them as a structured file containing bending moment values ​​and stress data. The automated execution script is a .com file suitable for SESAM software. The range of bending moment load parameters for multi-working-condition simulation is from 150,000,000 N·m to 1,200,000,000 N·m, with loading intervals of 500,250 N·m, generating a total of 2000 discrete working conditions. The original dataset is stored in JavaScript format files, with the file names containing the corresponding bending moment values.

[0025] By adopting the above technical solution, automated scripts are used to drive finite element analysis in batches and automatically extract results, reducing manual intervention, lowering labor costs, and improving the reliability and consistency of data generation.

[0026] Preferably, in step 3 of the automated finite element analysis dataset acquisition and processing, a Python script is used to traverse the original dataset file, merge all working condition data, and generate a training dataset stored in CSV format. The dataset contains bending moment values ​​and corresponding stress values ​​of each key node. The bending moment values ​​and the von Mises stress values ​​of each node are extracted and merged to generate a training dataset in CSV format. The CSV file contains column headers "moment,stress_1,stress_2,...,stress_n", where n is the number of key nodes.

[0027] Preferably, in step 1 of the deep learning model construction and training, data preprocessing involves standardizing the CSV dataset containing bending moment values ​​and key node stress values, dividing it into training, validation, and test sets. Preprocessing includes standardizing the input features and output labels using the StandardScaler class, with the standardization formula as follows: Where μ is the sample mean and σ is the sample standard deviation; the ratio of the training set, validation set, and test set is 70%:20%:10%.

[0028] Preferably, in step 2 of the deep learning model construction and training, the model construction involves: using TensorFlow / Keras to build a fully connected neural network. The network includes an input layer to receive bending moment values, at least three hidden layers, a ReLU activation function, and an output layer to output the predicted stress values ​​for each node. The structure of the neural network model is as follows:

[0029] Input layer: 1 neuron, receiving 1-dimensional load parameters;

[0030] First hidden layer: 64 neurons, activation function is ReLU;

[0031] The second hidden layer has 128 neurons and the activation function is ReLU.

[0032] The third hidden layer has 256 neurons and the activation function is ReLU.

[0033] Output layer: n neurons (n ​​is the number of key nodes), linearly outputting the predicted stress value.

[0034] By adopting the above technical solution, the model can be reused after training. It can output stress distribution in real time by inputting any bending moment value, which meets the needs of online monitoring, rapid evaluation and multi-condition simulation.

[0035] Preferably, in step 3 of the deep learning model construction and training, the mean squared error is used as the loss function during model compilation, the Adam optimizer is used, and the initial learning rate is 1×10⁻⁶. -4During training, an EarlyStopping callback function is introduced, and training stops when the validation set loss does not decrease for 10 consecutive rounds.

[0036] Preferably, in step 1 of the 3D interactive stress cloud map generation and web application integration, stress prediction involves receiving the bending moment value input by the user and predicting the stress value of key nodes of the offshore wind turbine tower foundation through a trained deep learning model.

[0037] Preferably, in step 2 of the 3D interactive stress cloud map generation and web application integration, the visualization preparation involves matching the predicted stress values ​​with the three-dimensional coordinates of the nodes, filtering surface nodes and performing spatial interpolation, loading the three-dimensional coordinate files of key nodes, matching the predicted stress values ​​with the coordinates, filtering the outer surface nodes of the cylinder, performing spatial interpolation using the scipy.interpolate.griddata function, constructing a 3D surface map using the Plotly library, mapping stress values ​​using the Viridis color scale, marking the location of the maximum stress value, and saving it as an HTML format file.

[0038] Preferably, in step 3 of the integration of 3D interactive stress cloud map generation with the web application, the cloud map generation involves: generating a 3D surface stress cloud map using the Plotly library, supporting rotation, scaling, and real-time viewing of stress values, and saving it as an HTML file; the web application is built on the Flask framework, including a front-end HTML form for receiving load parameters input by the user; the back-end routing receives the parameters and calls the stress prediction model and cloud map generation module to generate a 3D cloud map in HTML format; the cloud map is embedded in the webpage and returned to the user, supporting interactive operations on the browser side.

[0039] By adopting the above technical solution, a browser-based interactive 3D stress cloud map is generated, which supports rotation, scaling and data point query, intuitively displaying stress distribution and facilitating engineers to quickly identify high-stress areas.

[0040] The advantages of this invention are as follows: 1. This invention is based on AI to replace traditional simulation. It applies deep learning models to the stress distribution prediction of offshore wind turbine tower base, constructs an end-to-end mapping relationship of "load-stress", and achieves second-level prediction, which completely changes the traditional finite element analysis technology mode that is time-consuming and inefficient.

[0041] 2. This invention features an automated data generation and processing closed loop. It innovatively drives the batch operation of finite element software through automated scripts and combines Python scripts to automatically extract, organize, and generate standardized training datasets. This solves the problems of difficulty and high cost in obtaining high-quality engineering data required for AI model training, forming an automated closed loop of "simulation → data → model".

[0042] 3. This invention is aimed at generating 3D interactive cloud maps for engineering visualization. Based on AI prediction, it integrates spatial interpolation and Web visualization technology to automatically generate 3D stress cloud maps that can be interactively operated in a browser. It supports rotation, scaling, data hover query, and highlights the maximum stress point, significantly improving the intuitiveness and engineering applicability of the results.

[0043] 4. The end-to-end systematic implementation of this invention: The entire process of data preparation, model training, stress prediction and 3D visualization is coded and modularized to form a reusable and deployable complete system. It supports users to input arbitrary bending moment values ​​and output structured data and interactive cloud maps with one click, realizing a paradigm shift from "manual simulation" to "intelligent prediction". Attached Figure Description

[0044] Figure 1 The overall flowchart of the method for generating 3D interactive stress cloud diagrams for offshore wind turbine tower foundations provided by the present invention;

[0045] Figure 2 This is a flowchart illustrating the process of generating, extracting, and organizing file data to create a training dataset for automated finite element analysis (FEA) data according to the present invention.

[0046] Figure 3 This is a comparison chart of the interactive 3D stress cloud map generated by this invention and the 3D stress cloud map of finite element software, used to evaluate the prediction accuracy of the model.

[0047] Figure 4 This is a screenshot of the interactive 3D stress cloud map generated by the present invention, showing the stress distribution on the cylindrical surface, color gradation legend, and the label of the maximum stress point;

[0048] Figure 5 Deploy an end-to-end system architecture diagram to describe the module integration relationships: User input layer: GUI interface receives bending moment values; Output layer: automatically saved CSV data files and generates interactive 3D stress cloud diagrams.

[0049] The accompanying drawings of this invention are shown in color. Detailed Implementation

[0050] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0051] like Figure 1-5 As shown, this embodiment takes the offshore wind turbine tower foundation as the analysis object, uses DNV's SESAM software as the finite element analysis platform, employs Python as the development language and TensorFlow / Keras as the deep learning framework, and achieves rapid generation of 3D interactive stress cloud maps. The entire process is divided into the following three core stages:

[0052] I. Fully Automated Finite Element Analysis Dataset Acquisition, Processing, and Merging

[0053] Step 1: Establish an equivalent finite element model: Construct a three-dimensional finite element model of the offshore wind turbine tower base in SESAM software. Discretize the model using plate shell elements (PLATE) and beam elements (BEAM), reasonably simplify the non-load-bearing structure, retain the main load-bearing components, determine the final mesh generation scheme after mesh convergence analysis, and generate a total of 720 key nodes. Apply completely fixed boundary conditions at the bottom of the model and apply bending moment loads at the top control points to form an "equivalent model" that can be used for batch simulation.

[0054] Step 2: Develop automated execution scripts: Write automated command scripts (.com files) for the SESAM software to control the WADAM and HydroD modules to complete structural response analysis. The specific implementation steps are as follows:

[0055] By running the .com script through SESAM's batch processing interface (such as Xtract or Superfile operation), the system will automatically perform the modeling, solving and result export operations for all working conditions in sequence. The entire process requires no manual intervention, achieving high-efficiency and high-consistency batch simulation.

[0056] Define the range of bending moment loading parameters: the initial value is 150,000,000 N·m, the ending value is 1,200,000,000 N·m, the loading interval is 500,250 N·m, and a total of 2000 discrete bending moment conditions are generated.

[0057] Set up a loop structure in the .com script to assign each bending moment value to the load variable MOMENT_VAL in turn;

[0058] Call the SESAM command LOAD to apply MOMENT_VAL to the top node of the model, in the direction of around the Y-axis;

[0059] The ANALYSIS command is invoked to start the structural statics solver for calculation.

[0060] After the solution is completed, the POST module is called to extract the von Mises stress values ​​of 720 nodes;

[0061] The EXPORT command is used to export the bending moment value and corresponding nodal stress data of the current working condition as a separate JavaScript file. The file is named as stress_result_moment_XXXXXXXXXX.js, where XXXXXXXXXX is the current bending moment value.

[0062] The script automatically enters the next loop to process the next work condition until all 2000 work conditions are completed;

[0063] Step 3: Automatic Generation of Result Files: After the analysis of each working condition is completed, the von Mises stress values ​​of 720 nodes are automatically extracted by the SESAM post-processing module. The bending moment values ​​and the stress data of each node are packaged into a single .js file by calling the JavaScript interface or using SESAM's built-in data export function. The file is named as stress_result_moment_XXX.js, where XXX represents the corresponding bending moment value. The file contains key-value pair data of the structure. All files are stored in a preset output directory, forming a structured raw dataset.

[0064] Step 4: Data Extraction and Organization: Write a Python script to traverse the results directory using the os module, and use the json library to read and parse each .js file (or use the re module to extract text data) to extract the moment and stresses fields. Merge all the working condition data row by row to generate a unified CSV file moment_stress_training_data_with_header.csv, with column headers moment,stress_1,stress_2,...,stress_720. Each row corresponds to the complete stress response data of a bending moment working condition, forming a high-quality, structured training dataset. This dataset can be directly used for the training and validation of subsequent AI models.

[0065] II. Construction and Training of Interactive Stress Cloud Map AI Model for Offshore Wind Turbine Tower Foundation

[0066] Step 1: Data Preprocessing. Import the moments_stress_training_data_with_header.csv file generated by automated simulation into the Python environment. Use the pandas library's read_csv() function to read the file and form a structured data table. Extract the moment column from the data table to form the input feature matrix X, with a shape of (N,1), where N is the total number of samples (2000 in this example); extract the 720 columns from stress_1 to stress_720 to form the output label matrix y, with a shape of (N,720). To eliminate the influence of unit differences on model training, use the StandardScaler class in the sklearn.preprocessing module to standardize X and y respectively. The standardization formula is:

[0067]

[0068] Where μ is the sample mean and σ is the sample standard deviation. After processing, both input and output data are converted to a standard normal distribution with a mean of 0 and a standard deviation of 1. The scaler_X_mean, scaler_X_scale, scaler_y_mean, and scaler_y_scale parameters calculated during the standardization process are saved to local files for subsequent destandardization operations in the prediction stage. Then, all data are randomly divided into training, validation, and test sets at a ratio of 70%, 20%, and 10%, respectively, for model training, hyperparameter tuning, and final performance evaluation.

[0069] Step 2: Model Building. A fully connected feedforward neural network model is built using the TensorFlow / Keras deep learning framework. The specific structure is as follows:

[0070] Input layer: Receives 1-dimensional input (i.e. bending moment value), with 1 neuron; First hidden layer: Contains 64 neurons, with ReLU (Rectified Linear Unit) activation function, expressed as f(x) = max(0, x);

[0071] The second hidden layer contains 128 neurons and uses ReLU as the activation function.

[0072] The third hidden layer contains 256 neurons and uses ReLU as the activation function.

[0073] Output layer: Contains 720 neurons, no activation function (linear output), corresponding to the stress prediction values ​​of 720 nodes. The model stacks layers sequentially, forming a network architecture of "input → 64 → 128 → 256 → output". When compiling the model, the loss function is specified as Mean Squared Error (MSE), the optimizer is Adam (Adaptive Moment Estimation), and the initial learning rate is set to 1×10⁻⁶. -4 This model structure is implemented through a separate build_model.py script and saved as the stress_prediction_model.keras file, serving as the initial model structure for subsequent training.

[0074] Step 3: Model Training and Validation. Load the model structure from `stress_prediction_model.keras` and train it using the training set data. Set the training parameters as follows: batch size (BATCH_SIZE) to 64, training epochs (EPOCHS) to 300. During training, introduce the `EarlyStopping` callback function to monitor the validation set loss (val_loss). If the loss does not decrease for 10 consecutive training epochs, training automatically stops, and the model weights at the point of lowest validation loss are restored to prevent overfitting. In each training epoch, the model updates the weight parameters using the backpropagation algorithm and records the training loss and validation loss. After training, save the final model as `trained_stress_prediction_model.keras`. Evaluate the model performance using an independent test set, calculate the mean squared error (MSE) and coefficient of determination (R²) between the predicted and true values, and plot the loss curve during training, with the horizontal axis representing the number of training epochs and the vertical axis representing the loss value, to visually demonstrate the model's convergence. If the R² value is lower than the preset threshold (e.g., 0.95), the network structure can be adjusted (e.g., by increasing the number of hidden layers or neurons) or the hyperparameters can be optimized before retraining until satisfactory accuracy is achieved.

[0075] III. 3D Interactive Stress Cloud Map Generation and Online Web Application Integration

[0076] Step 1: Stress Prediction In the Python script visualization.py, the trained model file trained_stress_prediction_model.keras, which has been trained through the previous steps, is first loaded, along with the standardized parameters (including scaler_X_mean, scaler_X_scale, scaler_y_mean, scaler_y_scale) saved during the data preprocessing stage. The user inputs any target bending moment value (e.g., 150,000,000 N·m) through the command line or interactive interface. The program automatically standardizes the input value using scaler_X to obtain a standardized input vector. Subsequently, this vector is fed into the pre-trained neural network model for forward inference. The model outputs standardized stress prediction values ​​with 720 nodes. Finally, the mean and standard deviation of scaler_y are used to de-standardize the output results, restoring them to physically meaningful actual stress values ​​(unit: Pa), thus completing the rapid prediction from load to stress distribution.

[0077] Step 2: Data Saving and Visualization Preparation. Load the pre-prepared node coordinate file nodes_coords.csv, which contains the 3D spatial coordinates (x, y, z) of 720 nodes, with each row corresponding to the coordinate data of one node. Merge the 720 predicted stress values ​​obtained in the previous step with the coordinates of each node in sequence to generate a structured data table containing four columns: "x, y, z, stress". Save this data table as a CSV file, named stress_data_moment_XXXXXX.XX.csv, where XXXXXX.XX represents the input bending moment value for easy traceability and reuse. At the same time, call the generate_3d_surface_plot function to start the 3D contour plot generation process. This function first filters the nodes located on the outer surface of the cylinder based on the geometric features of the tower base (center coordinates (5.0, 5.0), radius 5.0m, height range 0.0–10.0m) as the basic data for contour plot drawing.

[0078] Step 3: Generate an interactive cloud map. For the selected surface nodes, use the scipy.interpolate.griddata function for spatial interpolation. Using a preset regular cylindrical mesh as the target, the nearest or linear interpolation method is used to map the stress values ​​of discrete nodes onto continuous mesh points, forming a smooth stress distribution surface. Then, the go.Surface component of the Plotly library is called to construct a 3D surface map. The X, Y, and Z coordinates correspond to the spatial position of the cylindrical surface. The color map uses the Viridis color scale, and the stress values ​​are marked with color bars. The units are automatically converted from Pa to MPa for easy reading. A red marker and text label are added at the location of the maximum stress value to display its specific coordinates and stress value. Finally, the generated graph is saved as a separate HTML file. This file can be opened in any modern web browser and supports interactive operations such as free rotation, scaling, and translation with the mouse. Users can also view the precise coordinates and stress values ​​of any position by hovering.

[0079] Step 4: System Integration and Deployment. The data loading, model prediction, result saving, and 3D visualization modules described above are integrated into a unified Python workflow, forming an end-to-end automated system. Users only need to run the main script and input the target bending moment value; the system can automatically complete the entire process from prediction to generating an HTML contour map. This system is encapsulated as a Flask-based web application: the front end provides an HTML form for users to input bending moment values; the back end receives the request, calls the prediction and visualization modules, generates the result, and embeds the HTML contour map into the webpage, returning it to the user. This achieves an integrated online service model of "input-calculation-display," meeting the needs of rapid on-site assessment and remote collaboration in engineering projects. The specific implementation steps are as follows:

[0080] Step 1: Build the Flask application main program, create a Python file app.py, import the Flask library and initialize the application instance;

[0081] Step Two: Prepare the front-end interface. Create a folder named "templates" in the project directory, and create a new HTML file named "index.html". Use standard HTML syntax to build a user input form, including a numeric input box. <input type＝"number"> This is used to input bending moment values, and there is a submit button. <button type＝"submit">The page layout is clean and clear, and it displays correctly on both PCs and mobile devices.

[0082] Step 3: Define backend routes. Define two route functions in app.py:

[0083] • / Route: Responds to GET requests and returns the index.html page for the user to input bending moment values;

[0084] The / predict route responds to POST requests and receives the bending moment value submitted by the front end.

[0085] Step 4: Integrate the prediction and visualization modules. In the ` / predict` route processing function, call the core functionalities in `visualization.py`, including:

[0086] Load the trained AI model

[0087] trained_stress_prediction_model.keras;

[0088] • Load standardized parameters and node coordinate data;

[0089] • Standardize the bending moment values ​​input by the user;

[0090] • Perform model inference to obtain predicted stress values ​​for 720 nodes;

[0091] • Perform de-standardization to restore the actual stress;

[0092] • Call the generate_3d_surface_plot function to generate a 3D stress cloud plot and save it as an HTML file (such as output_plot.html);

[0093] Step 5: Return the visualization results. Embed the generated output_plot.html file content into the response data, or save it to the static directory and then...

[0094] render_template returns a new HTML page containing the embedded chart. After the user submits the bending moment value, the page automatically redirects to the results display page, allowing the user to view the interactive 3D cloud map without refreshing the page.

[0095] Step Six: Deployment and Operation. Configure the Flask application to listen on a local machine or a specified IP port (e.g., http: / / localhost:5000). Users access this address through a browser, input the target bending moment value, and submit. The system automatically completes the calculation and displays a rotatable and scalable 3D stress cloud map in real time on the webpage, realizing an integrated online service of "input-calculation-display".

[0096] By using the above method, this invention encapsulates the complex AI prediction process into a lightweight web application. No professional software needs to be installed; the operation can be completed simply through a browser, significantly reducing the barrier to entry. It is suitable for rapid on-site assessment, remote collaboration, and multi-user sharing scenarios, thereby enhancing the system's practicality and promotional value.

[0097] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention; the objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any modifications or variations of the embodiments of the present invention may be made without departing from the stated principles.< / button>

Claims

1. A method for predicting and visualizing three-dimensional stress distribution in offshore wind turbine tower foundations, characterized in that, Includes the following steps: I. Acquisition and Processing of Datasets for Automated Finite Element Analysis Step 1: Construct an equivalent finite element model of the offshore wind turbine tower base in the finite element analysis platform, determine key nodes, and apply boundary conditions and loads; Step 2: Develop an automated execution script to control the finite element analysis platform to perform multi-condition simulations in batches, and export the stress data and corresponding load parameters of key nodes under each condition to form the original dataset; Step 3: Extract and organize the original dataset to generate a structured training dataset containing load parameters and corresponding nodal stress values; II. Construction and Training of Deep Learning Models Step 1: Preprocess the training dataset, using load parameters as input features and nodal stress values ​​as output labels, and divide it into training set, validation set and test set; Step 2: Construct a fully connected feedforward neural network model, which includes an input layer, at least two hidden layers, and an output layer. The input layer receives load parameters, and the output layer outputs the stress prediction values ​​of each key node. Step 3: Train the model using the training set, optimize the model parameters using the validation set, evaluate the model performance using the test set, and obtain the trained stress prediction model; III. 3D Interactive Stress Cloud Map Generation and Web Application Integration Step 1: Receive the target load parameters input by the user, and obtain the stress prediction values ​​of each key node through the trained stress prediction model; Step 2: Match the predicted stress values ​​with the three-dimensional coordinates of key nodes, generate a continuous stress distribution through spatial interpolation, and construct a 3D interactive stress cloud map based on visualization tools. The cloud map supports rotation, scaling, translation, and real-time viewing of stress values. Step 3: Integrate stress prediction and 3D cloud map generation functions into the web application to enable users to input load parameters through a browser and obtain interactive stress cloud maps online.

2. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 1 of the acquisition and processing of the automated finite element analysis dataset, an equivalent finite element model is constructed. The offshore wind turbine tower base is discretized using plate and shell elements and beam elements, the main load-bearing components are retained, key nodes are determined, and fixed boundary conditions and top bending moment loads are applied. The equivalent finite element model is discretized using plate and shell elements (PLATE) and beam elements (BEAM), and contains 720 key nodes. The boundary condition is that the bottom of the model is completely fixed, and the load is the bending moment load applied to the control point at the top of the model.

3. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 2 of the automated finite element analysis dataset acquisition and processing, an automated script is developed to define the range and interval of bending moment load parameters. The script then iteratively executes modeling and solving for each working condition through the batch processing interface of the finite element software, extracts the von Mises stress values ​​of key nodes, and exports them as a structured file containing bending moment values ​​and stress data. The automated execution script is a .com file suitable for SESAM software. The bending moment load parameter range for the multi-working-condition simulation is from 150,000,000 N·m to 1,200,000,000 N·m, with loading intervals of 500,250 N·m, generating a total of 2000 discrete working conditions. The original dataset is stored as a JavaScript format file, with the file name containing the corresponding bending moment value.

4. The method for predicting and visualizing the three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 3 of the automated finite element analysis dataset acquisition and processing, a Python script is used to traverse the original dataset file, merge all working condition data, and generate a training dataset stored in CSV format. The dataset contains bending moment values ​​and corresponding stress values ​​of each key node. The bending moment values ​​and the von Mises stress values ​​of each node are extracted and merged to generate a training dataset in CSV format. The CSV file contains column headers "moment, stress_1, stress_2, ..., stress_n", where n is the number of key nodes.

5. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 1 of the construction and training of the deep learning model, data preprocessing involves standardizing the CSV dataset containing bending moment values ​​and key node stress values, and dividing it into training, validation, and test sets. This preprocessing includes standardizing the input features and output labels using the StandardScaler class. The standardization formula is as follows: in, The sample mean. The standard deviation is the sample value; the ratio of the training set, validation set, and test set is 70%:20%:10%.

6. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 2 of the deep learning model construction and training, the model construction involves using TensorFlow / Keras to build a fully connected neural network. This network includes an input layer to receive bending moment values, at least three hidden layers, a ReLU activation function, and an output layer to output predicted stress values ​​for each node. The structure of the neural network model is as follows: Input layer: 1 neuron, receiving 1-dimensional load parameters; First hidden layer: 64 neurons, activation function is ReLU; The second hidden layer has 128 neurons and the activation function is ReLU. The third hidden layer has 256 neurons and uses ReLU as the activation function. Output layer: n neurons (n ​​is the number of key nodes), linearly outputting the predicted stress value.

7. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 3 of the deep learning model construction and training process, the mean squared error is used as the loss function during model compilation, along with the Adam optimizer and an initial learning rate of [missing information]. During training, an EarlyStopping callback function is introduced, and training stops when the validation set loss does not decrease for 10 consecutive rounds.

8. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 1 of the 3D interactive stress cloud map generation and web application integration, stress prediction involves receiving the bending moment value input by the user and predicting the stress value of key nodes of the offshore wind turbine tower foundation through a trained deep learning model.

9. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 2 of the 3D interactive stress cloud map generation and web application integration, the visualization preparation involves matching the predicted stress values ​​with the 3D coordinates of the nodes, filtering surface nodes and performing spatial interpolation, loading the 3D coordinate files of key nodes, and matching the predicted stress values ​​with the coordinates; filtering the outer surface nodes of the cylinder and performing spatial interpolation using the scipy.interpolate.griddata function; constructing a 3D surface map using the Plotly library, mapping stress values ​​using the Viridis color scale, marking the location of the maximum stress value, and saving it as an HTML format file.

10. The method for predicting and visualizing three-dimensional stress distribution of offshore wind turbine tower foundation according to claim 1, characterized in that: In step 3 of the integration of the 3D interactive stress cloud map generation with the web application, the cloud map generation involves: generating a 3D surface stress cloud map using the Plotly library, supporting rotation, scaling, and real-time viewing of stress values, and saving it as an HTML file; the web application is built on the Flask framework, including a front-end HTML form for receiving load parameters input by the user; the back-end router receives the parameters and calls the stress prediction model and cloud map generation module to generate a 3D cloud map in HTML format; the cloud map is then embedded in the webpage and returned to the user, supporting interactive operations on the browser side.

Images