A generalized dual potential for inelastic Constitutive Artificial Neural Networks: A JAX implementation at finite strains (Source code and data)

This dataset contains

the source code

the data and examples

of the generalized inelastic Constitutive Artificial Neural Network (iCANN) enhanced by a novel network architecture.

The corresponding publication is:

Holthusen, H., Linka, K., Kuhl, E., Brepols, T.

A generalized dual potential for inelastic Constitutive Artificial Neural Networks: A JAX implementation at finite strains

JAX implementation

activation_functions: Contains the activation functions

BaseClasses: The `heart' of our code. Contains the general object-orientated classes to create neural networks and iCANNs

build_network: Add contribution of individual iCANNs (parallel connection)

config_training: Configuration file for current training options

energy: Feed-forward neural network for Helmholtz free energy

execute_model_jit: Execute discovered model with specific parameters (weights). Use Just-InTime-Compilation (JIT) for fast evaluation (EXECUTABLE)

execute_model: Execute discovered model with specific parameters (weights). No use of JIT (EXECUTABLE)

execute_training: Execute training for specific training data based on config_training. Training is specified within runs/ (EXECUTABLE)

helpers_material: Helper routines for continuum mechanics

helpers: Helper routines to read in data

iCANN_explicit: Explicit time integration scheme for iCANN

iCANN_impllicit: Implicit time integration scheme for iCANN

potential: Feed-forward neural network for dual potential

Directories

loadings: Contains training and testing files with data

results: Contains discovered weights and corresponding losses

runs: Specify the config_training file. Execute training for specific example (EXECUTABLE FILES)