Julia framework for automated simulations in COM3

This framework was developed for the purpose of easily simulating compression tests of PVA-ECC-based 2D metamaterials using COM3.
It contains all the necessary tools to generate truss-like geometries, mesh them, write COM3 files, run instances of COM3 and read from COM3 output files, as well as post-processing of the data such as exporting results to .csv and producing plots. It also contains a Reinforcement Learning environment built upon Reinforce.jl which allows for easy implementation of reinforcement learning algorithms using COM3 simulations to calculate rewards.
The whole framework is made thread-safe so that many simulations can be run in parallel on CPUs with many cores, improving performance on big simulation batches.

Starting up with the framework

This framework is a module coded using the Julia programming language, a high performance language for scientific purposes.
You can either use it in a Linux environment or in Windows. However, Linux is a more appropriate environment and installing a Linux distro such as Ubuntu inside Windows using the Windows Subsystem for Linux (WSL) (https://docs.microsoft.com/en-us/windows/wsl/install-win10) will result in a better experience and control.
You will need to first install Julia (https://julialang.org/) and Git for Windows (https://gitforwindows.org/). Once that is done, use the git GUI to clone this github repository.
Go to the windows command line, ( by typing cmd in the search bar for instance) and type in :

setx PATH "%PATH%;C:\Users\[Your-User-Name]\AppData\Local\Programs\Julia-1.6.1\bin"

after replacing with your correct username.
Then launch git bash (or the command line in Linux) and run the following in the Metamaterials folder (use cd to change directories inside the bash, pwd to display the current directory) :

julia setup.jl

This script will install relevant packages for this framework.
Once this is done, you can run a simple batch of simulations by writing in the git bash terminal, making sure you are in the Metamaterials directory :

julia batchSimulator.jl -batchsize N

where N is the number of simulations you want to run.
For an overview of the supported arguments by the framework, type :

julia batchSimulator.jl -help

If you want to run simulations simultameously across several threads, type :

julia -t (Your desired number of threads) batchSimulator.jl (Further options...)

Note that this is generally faster as the simulations are quite small scale, thus not taking advantage of many processing threads. Running instances in parallel, however, allows for faster computation.

Structure of the Julia module

The module defined in MetamatModule.jl is comprised of several source files, each defining functions for a specific part of the framework :

• structures.jl defines the data structures used in the module (Point,Element,Skeleton,Model,Material,Simulation) along with various constructors and methods for them.
• geometryTools.jl contains the geometry generator as well as plotting functions to this end.
• simulationTools.jl holds the tools which build and run a simulation.
• COM3RWTools.jl gives the different methods for writing to and reading from COM3 files.
• MetamatRLEnv.jl defines a Reinforcement Learning environment according to the Reinforce.jl package, where the agent can change one geometric feature at each step.
• MetamatRLEnv2.jl defines a similar Reinforcement Learning environment, where geometric features or material properties can be altered at each step.

Main scripts

• batchSimulator.jl is the main tool which allows for high volume of simulations of random metaaterials to be performed with parallel computing.
• DQN.jl holds a script for training of a Deep Q-Network on geometric features only.
• materialsDQN.jl holds a script for a Deep Q-Network based on the second Reinforcement Learning environment, inluding both geometric and material properties variations.

Other tools in the Resources folder

• plotDeformed.jl is a tool to plot the deformations and internal stress of a structure if executed inside the results folder of a simulation.
• annotateMetamatPlot.jl creates interactive plots from results of a batch of simulations, where hovering over the points reveals the geometries.
• points.csv and elements.csv define the base mesh that is reused and translated to mesh the metamaterials automatically.