[WIP] Jet Stirred Reactors

[skrsna]

• Ready for review

• Code is functional and successfully runs JSRs given a ChemKED JSR file (Jet Stirred Reactor (JSR) PyKED#126)

Todo:

• Multiprocessing
• Efficently save .h5 files
• Refactoring
  • eval_model.py
  • simulation.py
• Error functions
  • Difference between area under the curves using numpy.trapz
• Multiple datasets

[rwest]

I would suggest integrate the difference between the curves, rather than find the difference of the integrals.

[kyleniemeyer]

I would suggest integrate the difference between the curves, rather than find the difference of the integrals.

I'm trying to visualize the difference here... in practice the answer should be the same, no? Though it's easier/more efficient to just integrate once.

[rwest]

I was imagining these two curves

They have the same area under them, so the difference in area under curve is zero, suggesting they are identical and there is no error in the model.

If you find the difference between the curves, then the area between that and the y=0 axis (ok, so perhaps I should have said integrate the absolute value of the difference between the curves), the error is significant.

[kyleniemeyer]

@rwest got it, that makes sense!

[rwest]

It's by no means a perfect measure of fit. For example, these two pairs of curves would have the same "error".

But it's a start, to get the code working.

[kyleniemeyer]

Agreed.

FYI, when I was implementing the error evaluation for perfectly stirred reactor simulations in my older MARS, which marched along the upper branch, I used three points: near extinction turning point, near equilibrium (e.g., a residence time of around 0.1 s), and the logarithmic midpoint between those two points.

The overall error was the the max of the error in residence time at the extinction turning point and the error in temperature at the other two points.

[rwest]

JSR simulations needn't save time history.
Probably this whole method should be in the AutoIgnitionSimulation class, not the Simulation super-class

[rwest]

For our JSR simulations it's maybe worth reading 2E08.pdf and keeping an eye on Cantera/enhancements#31 and Cantera/cantera#1021

[sevyharris]

I'm resurrecting this PR because I've got a flame speed addition in the works that depends on this.

Here's what I've done:

• Finished adding multiprocessing
• Added handling for multiple datasets
• Made a bunch of PEP8/style changes because I can't help myself
• Refactored the simulation files back into a single simulation.py file

Here's the remaining work:

• Get existing tests back up and running
• Add JSR tests
• Refactor jsr_eval_model into eval_model

Here's what I don't plan to do for this PR:

• Improve the error calculations. I made a curve_matching branch that can (in theory) handle this, but it deserves its own PR
• Plotting

[rwest]

Incomplete review; going to wait until refactoring finished. Not clear to me yet how multiple different simulation types are going to coexist and reuse appropriate code/classes.

[rwest]

In general I'm not sure I like the idea of importing something under a different name - could likely confuse people down the line. Is there a good reason for it?

[rwest]

Perhaps I should wait to review until refactoring finished, but this file/module/class seems a little confused right now about if it's specific to Autoignition, or if it's generic for all simulations.

So I'm going to pause reviewing now. Let me know if/when I should take another look.

[sevyharris]

Okay, @rwest I think this one is ready for review.