Peri-Net: Analysis of Crack Patterns Using Deep Neural Networks

TensorFlow code for the forward model of the Peri-Net neural network for modeling LAMMPS peridynamics simulations.

Dependencies

• LAMMPS - The molecular dynamics software used to generate data (must be compiled with the PKG_PERI=on flag).
• Python - The code is designed to work with Python 2.7 and Python 3.5 or greater. Python2 is required for reading/writing in the LAMMPS format; Python3 is required for the TensorFlow network implementation.
• TensorFlow - The software library used to implement the deep convolutional neural network in Python.
• NumPy - Arrays are processed in Numpy prior to being fed as tensors into TensorFlow.
• Pandas - Python module used to read raw LAMMPS dump files into Python.
• ParaView - Open-source data visualization software used to analyze damage patterns.

Generating Data

# Generate Data with LAMMPS:

$ mkdir Data
$ OMP_NUM_THREADS=8 lmp -in in.peri
OR
$ lammps -in in.peri

# Convert LAMMPS data to array format:
$ python Convert_Data.py

Parallelization

The data generation process can be sub-divided into separate LAMMPS instances by breaking up the i loop in the in.peri file. For example, the file can be split into three files in_0.peri, in_1.peri, in_2.peri corresponding to a partion of the full i loop and executed separately by running:

$ ./Run_0_LAMMPS.sh
$ ./Run_1_LAMMPS.sh
$ ./Run_2_LAMMPS.sh        

where the starting count cstart is specified in terms of the istart and imax values. These starting counts can be pre-computed using the Peri_Files/plot.py file.

The array files can then be created using the Python multiprocessing module via:

# Convert dump files into NumPy arrays:
$ python Make_Arrays.py

The generated training data is created in the ./Data/ and ./Arrays/ subdirectories. Once the arrays have been created, the large ./Data/dump_*.peri files can be deleted; only the ./Data/indenter_*.txt files are required for training.

Train Model

# Train TensorFlow model:
$ python Train_Model.py

View Results

# Freeze TensorFlow model:
$ python freeze.py

# Test model at specific point (x,y):
$ cd Frozen
$ python load.py --x_val x --y_val y

# Convert prediction to dump file:
$ python2 Convert_Predictions.py

# View prediction in Paraview:
$ paraview p_disk_0.case &

# Plot true solution with id ID:
$ python load_soln.py --ID ID
$ python2 Convert_Solutions.py
$ paraview s_disk_0.case &