DeepJEB++: 15,360 Simulated Jet-Engine Brackets for 2D-to-3D Augmentation

DeepJEB++ is a foundation-model-driven dataset for structural engineering design. It contains 15,360 simulation-labeled jet-engine brackets generated by adapting Stable Diffusion v1.5 and TRELLIS. Each design is paired with automatically computed linear-elastic finite-element labels under four load cases.

Use Cases

• Training generative models for mechanical part design based on the described 2D-to-3D augmentation pipeline.
• Benchmarking structural analysis models using the automatically computed finite-element labels.
• Developing data augmentation techniques for engineering datasets based on the foundation-model-driven generation method.
• Researching the intersection of generative AI and finite-element analysis for design automation.

Strengths

• Contains 15,360 individual design samples.
• Each sample includes automatically computed finite-element analysis labels for four distinct load cases.
• Generated using adapted versions of established models Stable Diffusion v1.5 and TRELLIS.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other scale metrics like file size and format are unknown.
• Data may reflect bias inherent to the generative models and simulation parameters used.

Provenance

Source

KAIST-SmartDesignLab

Collection Method

Generated by adapting Stable Diffusion v1.5 and TRELLIS models, with automated finite-element simulation labeling.

Freshness

Last updated 2026-06-10 12:58:56; freshness should be verified.