Dataset for classification of forming tool types for aircraft parts based on neural network models using CAD

This study presents a data collection using CAD image-based deep learning and machine learning models to classify forming tool types for aircraft parts. Focusing on sheet metal components i.e. which constitute a substantial portion of aircraft structures and are often produced via fluid-cell hydroforming, this dataset uses visual information embedded in CAD images, particularly flange geometries, to classify tool types without requiring physical prototypes or specialized sensors. A dataset was built from publicly available CAD models (sourced from Zenodo) and additional models generated in CATIA, resulting in 12,432 images across three visualization modes (Normal, Hidden Line, and Wireframe) and multiple orientations. Images were categorized into six tool types based on the flange configuration. Several convolutional neural network architectures and machine learning models were evaluated to validate their classification performance on the dataset. ResNeXt achieved the highest accuracy of 96% on Normal View with Wireframe View. ResNeXt was evaluated with accuracy and F1-score to address class imbalance.. This scalable, cost-effective, and readily applicable CAD image-based deep learning and machine learning model for classifying aircraft parts will significantly help people who want to propose similar solutions for small-batch, multi-variety manufacturing environments.