How mflaschel/gsmn works
GSMN is a PyTorch framework for learning and simulating generalized standard material behavior. The primary workflow generates benchmark biaxial histories, persists them, reloads them for training, evaluates neural Helmholtz and dissipation potentials through the GSMN material solver, and saves the trained model and diagnostics. Controls drive time-strain simulations; training configuration, optimization, plotting, and serialization support that workflow. Some orchestration is inferred from documented scripts and unsampled portions of those files.
Components 16
Simulation Control
- Benchmark Generator
main_biaxial_test_3D.py - Load Control
gsmn/Control_time_strain_tension_unloading.py - Control Adapter
gsmn/Control.py
Material Potentials
- Benchmark Potentials
gsmn/Potential_benchmark.py - Free Energy Network
gsmn/Potential_neural_network.py - Dissipation Network
gsmn/Potential_neural_network.py
Material Solver
- GSMN Material
gsmn/GSMN.py - Newton Solver
gsmn/GSMN.py - Material State
gsmn/GSMN.py
Training Runtime
- Training Script
main_training_biaxial_multiple.py - Training Config
gsmn/Config_training.py - Loss Optimization
main_training_biaxial_multiple.py
Persistence Output
- Test Data Store
gsmn/Inout.py - Result Store
gsmn/Inout.py - Training Plots
gsmn/Plot.py
Other components
- Engineer
Connections 23
- Engineer → Benchmark Generator: runs generator
- Benchmark Generator → Load Control: creates paths
- Benchmark Generator → Benchmark Potentials: builds potentials
- Benchmark Generator → GSMN Material: builds material
- Benchmark Generator → Test Data Store: saves histories
- Control Adapter → GSMN Material: dispatches loads
- Load Control → Control Adapter: provides increments
- GSMN Material → Free Energy Network: evaluates energy
- GSMN Material → Dissipation Network: evaluates dissipation
- GSMN Material → Newton Solver: solves internals
- Newton Solver → Material State: updates state
- Engineer → Training Script: runs training
- Training Config → Training Script: configures training
- Training Script → Test Data Store: loads histories
- Training Script → Control Adapter: reconstructs controls
- Training Script → Free Energy Network: constructs network
- Training Script → Dissipation Network: constructs network
- Training Script → GSMN Material: trains model
- Training Script → Loss Optimization: runs updates
- Loss Optimization → Free Energy Network: updates weights
- Loss Optimization → Dissipation Network: updates weights
- Training Script → Result Store: saves results
- Result Store → Training Plots: plots diagnostics