State of The Impending Sim – Part IV

Accounting for the Timeline

I have already covered these in detail in the previous State of the Sims, so I’ll give you a recap here:

•  Switch to Electron, React, and Babylon.js – Meco had a relatively simple custom 3D and UI game engine; however, it became clear that we wanted to focus on extensibility, intuitive UI, and rocket engine design/simulation features rather than game engine development. This move also simplifies writing addons for the game UI, as they can be written with JavaScript and HTML5.

•  Set up AI workflows to boost development productivity – As you would know by now, generative AI like ChatGPT and Stable Diffusion looked like they were about to disrupt many industries. So, I wanted to make sure that Meco was not caught out. The TLDR is that we now have a neat AI game art pipeline to give us much more varied characters and environments when Missions are implemented in v0.2. Also, as of now, I am glad to report that despite trying my best to guide ChatGPT 4, it cannot handle programming solutions to anything more than grad-level aerospace/rocket science problems. However, AI proves promising as an assistant with knowledge of fundamental physics, equations, and terminology, so MecoGPT in-game would be splendid.

Gas Simulation Overhaul

This is the latest addition to the timeline, proving to be the most tricky yet. More than two years ago, I decided to implement a simplified gas simulation to move on to develop other parts of Meco. It had limitations as friction and heat transfer were ignored, and the network topology was limited by each junction being only able to have one inbound branch. Unfortunately, I discovered in July that this simplified methodology also rendered gas valves defunct.

Thus, it became apparent that I could no longer avoid implementing a 1D-CFD finite element gas simulation. I will write in detail how this is accomplished, but for our purposes here, let’s say that the code has to solve a large multi-variate non-linear problem in an incremental manner using quasi-newton methods with gradient descent such as described by Levenberg–Marquardt and Broyden–Fletcher–Goldfarb–Shanno (the limited-memory and box constrained version) in nested iterations—all that while attempting a solution 100,000 times a second on a modest PC.

We now have a Proof-of-concept coded in Python that passes the test cases needed. The accuracy of the simulation has been verified with Fanno Flow equations (for friction effects), and I’m working on validating it with Raleigh flow equations (for energy transfer effects) now. Unlike Fanno and Raleigh flow equations, the simulation can handle simultaneous friction and energy transfer in a duct. After all that, we port the code to C++ and integrate it with the rest of the simulation.