Arrhenius

Inspired by ReactionMechanismSimulator.jl, this project aims at developing a mini package for interpreting combustion chemical kinetic models and compute reaction source term. The name of Arrhenius.jl is reflecting the fact that the distinction between combustion and other chemical reacting flow are temperature-dependent kinetics and large activation energy.

Arrhenius.jl is designed with following in mind:

• Combustion software 2.0
• Differential programing
• Physics informed machine learning
• Combustion simulation education.

We are in an early-development. Expect some adventures and rough edges.

Installation

pkg> add https://github.com/DENG-MIT/Arrhenius.jl

Publication

• Machine Learning Approaches to Learn HyChem Models: demonstrate 1000 times faster than genetic algorithms using commercial software for optimizing complex kinetic models.
• Arrhenius.jl: A Differentiable Combustion Simulation Package: overview of Arrhenius.jl and applications in deep mechanism reduction, uncertainty quantification, mechanism tuning and model discovery.

Usage

You can start from the example of pyrolysis of JP10 (an aviation fuel power the flight) under the folder of example. It will guide you on how to implement the governing equations with a couple of lines of code. You will also learn how to use ForwardDiff.jl to differentiate the solver.

Currently, the package relies on Cantera and ReacTorch for interpreting the reaction mechanism. If you want to have a try, you don't need to install Cantera and ReacTorch, since there are already some pre-compiled reaction mechanisms under the folder of mechanism. Otherwise, you can install Cantera and ReacTorch to compile it using the python script interpreter.py under the folder of mechanism. You can also ask for help in the discussion forum and our developers can compile the model for you.

Examples

Note that some of the examples are in development and you can have early access by contacting Weiqi Ji

• Active Subspace of Reaction Mechanism
• Pyrolysis of JP10
• Perfect Stirred Reactor
• Auto-ignition:
• Compute Jacobian using AD
• Couple with CRNN and Neural ODEs
• Deep Reduction: Two-stages mechanism reduction with deep learning

Validation with Cantera

In the example of pyrolysis.jl, we compare the results with Cantera. The example involves solving the equations of mass fractions and temperature under constant pressure.