noahbagz / ShipGen

Generate Parametric Ship Hull with a Guided Tabular Diffusion Model

ShipGen

Generate Parametric Ship Hulls with a Guided Tabular Diffusion Model. The following sections detail training a guided diffusion model to generate parametric ship hull designs

Data Set

This project uses the ShipD Dataset. The ShipD Dataset is a dataset of 30,000 parametric ship hull designs. Included in the dataset are performance metrics, meshes, and images of each hull.

Sample code for the dataset is found here:

https://github.com/noahbagz/ShipD

The dataset, code, and documentation:

https://www.dropbox.com/sh/jg98r425v7ly89l/AAA49uMr7_mhaVmRDrPq0NU_a?dl=0

Running the Code

After you download the ShipD Dataset, run the jupyter notebooks in this order:

1. "SetUp_Diffusion_TrainingData.ipynb" This will format the hull data and the performance data to prepare it for Machine Learning.
2. "Generate_negative_samples.ipynb" This creates a set of hulls that violate at least one constraint for classification purposes.
3. "Run_guidedDiffusion_Notebook.ipynb" This notebook trains the diffusion model, constraint classifier, and regression models. It also generates samples.

Other Resources

Check out MIT's DeCoDE Lab for more machine learning tools and datasets for engineering design:

https://decode.mit.edu

The ShipGen project page is found at:

https://decode.mit.edu/projects/ShipGen/