Deep learning-based topology optimization with a minimum compliance loss function.
Finite element meshing and analysis is handled in the FEA module, while the loss function and analytical gradients are handled in the spalg module.
To run the model type python main.py 'SAVE_PATH' 'MODEL_NAME' 'DATA_PATH'. The following additional arguments can also be parsed:
parser.add_argument('--pretrained',default="") # save path for pretrained weights
parser.add_argument('--batch_size', default=16, type=int)
parser.add_argument('--epochs', default=100, type=int)
parser.add_argument('--lr', default=2e-4, type=float) # learning rate
parser.add_argument('--clip', default=1, type=float) # gradient clipping value
parser.add_argument('--dens_penal', default=2, type=float) # SIMP penalization
parser.add_argument('--vol_penal', default=1e1, type=float) # volume penalization
Send a mail to maoel@mek.dtu.dk if you want to obtain the datasets needed to train the model.