CNNs_for_mechanical_property_prediction_of_composites_dataset

Dataset for the ECAI 2025 Paper - Explainable Prediction of the Mechanical Properties of Composites with CNNs Filename and contents: composite_images.zip : collection of 2D black-and-white 500 x 500 pixel images of the composite microstructures (as .tif files) composite_image_coordinates.zip : collection of csv files containing the coordinates for the microstructure images model_generation_and_FE_simulations.py : python code to be inputted into Abaqus FE Software to create FE models (.inp files) and subsequently FE simulation job files (.odb files) to perform FE analysis extract_force_displacement_from_models.py : python code to be inputted into Abaqus FE software to extract force-displacement data from the .odb simulation files extract_E_and_YS_from_force_displacement.py : python code to be run on computer to conver force-displacement data into Youngs Modulus and Yield Strength composite_mechanical_properties.csv : csv file containing the Youngs Modulus and Yield Strength for all models, obtained from FE simulations. This is used for AI models load_images_function.py : python function to help load the .tif image files into desired python notebook for AI models. This is a supplementary file How to reproduce data (Mechanical properties generation): unzip 'composite_images.zip' and 'composite_image_coordinates.zip' into a folder locally open 'model_generation_and_FE_simulations.py' and add the location of the images and coordinates directory (from step 1) into the code. The comments on the very top of this python file point to the necessary lines to be edited to specify the path name Run the modified 'model_generation_and_FE_simulations.py' inside Abaqus FE software open 'extract_force_displacement_from_models.py' and add the location of the '.odb' files into the code. The comments on the very top of this python file point to the necessary lines to be edited to specify the path name Run the modified 'extract_force_displacement_from_models.py' inside Abaqus FE software open 'extract_E_and_YS_from_force_displacement.py' and add the location of the .csv files containing the force-displacement data outputted by Abaqus from step 5. Also add the path name for the .csv file that you want to store the mechanical properties in. The comments on the very top of this python file point to the necessary lines to be edited to specify the path names Run 'extract_E_and_YS_from_force_displacement.py ' as a python script. The final mechanical properties will be outputted. The attached 'composite_mechanical_properties.csv' can be used for sanity check How to reproduce CNNs and XAI results: Split the data from 'composite_mechanical_properties.csv' into a train-test-validation split. For the paper, an 85-10-5 split was used. if needed, use the 'load_images_function.py' to load the images into the file (e.g. jupyter notebook) you want to perform the AI analysis on for the customised CNNs (10 epochs) - follow the architecture shown in the paper to recreate the CNN models on TensorFlow 2.13.0 to predict the Youngs Modulus and Yield Strength本数据集对应ECAI 2025收录论文——基于卷积神经网络（Convolutional Neural Networks, CNNs）的复合材料力学性能可解释预测研究。 ### 文件名与内容 1. composite_images.zip：包含复合材料微观结构的2D黑白500×500像素TIFF格式图像集合 2. composite_image_coordinates.zip：包含各微观结构图像坐标信息的CSV文件集合 3. model_generation_and_FE_simulations.py：可导入Abaqus有限元（Finite Element, FE）软件运行的Python代码，用于生成有限元模型（.inp格式文件）及后续有限元分析作业文件（.odb格式文件）以开展有限元分析 4. extract_force_displacement_from_models.py：可导入Abaqus有限元软件运行的Python代码，用于从.odb格式仿真文件中提取力-位移数据 5. extract_E_and_YS_from_force_displacement.py：可在本地计算机运行的Python代码，用于将力-位移数据转换为杨氏模量与屈服强度 6. composite_mechanical_properties.csv：包含所有有限元仿真得到的复合材料力学性能数据的CSV文件，涵盖杨氏模量与屈服强度，可用于人工智能模型训练 7. load_images_function.py：辅助将.TIF格式图像加载至人工智能模型开发所需Python脚本（如Jupyter Notebook）的Python函数，为配套辅助文件 ### 力学性能数据复现流程 1. 将composite_images.zip与composite_image_coordinates.zip解压至本地同一文件夹 2. 打开model_generation_and_FE_simulations.py，按照文件顶部注释指引，修改代码中指定的路径行，填入步骤1中图像与坐标文件所在目录的路径 3. 在Abaqus有限元软件中运行修改后的model_generation_and_FE_simulations.py脚本 4. 打开extract_force_displacement_from_models.py，按照文件顶部注释指引，修改代码中指定的路径行，填入步骤3生成的.odb格式文件所在目录的路径 5. 在Abaqus有限元软件中运行修改后的extract_force_displacement_from_models.py脚本 6. 打开extract_E_and_YS_from_force_displacement.py，按照文件顶部注释指引，修改两处路径：一是填入步骤5中Abaqus输出的力-位移数据CSV文件所在路径，二是指定存储最终力学性能数据的CSV文件路径 7. 运行extract_E_and_YS_from_force_displacement.py脚本，即可得到最终力学性能数据。可通过附带的composite_mechanical_properties.csv文件进行结果校验 ### 卷积神经网络与可解释人工智能（Explainable AI, XAI）结果复现流程 1. 将composite_mechanical_properties.csv中的数据集划分为训练集、测试集与验证集，本论文采用85-10-5的划分比例 2. 若有需要，可使用load_images_function.py将图像加载至开展人工智能分析的开发环境（如Jupyter Notebook）中 3. 针对自定义卷积神经网络模型（训练轮次为10），按照论文中给出的架构，在TensorFlow 2.13.0环境下复现CNN模型，以实现杨氏模量与屈服强度的预测