[siar]

Hi Ralf,
thank you for the details on the piston and membrane runs – that is exactly what I was missing. I’ll build it into the model when I find the time; at the moment I’m rather short of it, so no hurry on my side either.
Also thanks for posting the information on Discord. I had a look there but, to be honest, couldn’t quite find my way around – the channels and threads were a bit much for me, so your summary helps a lot.
Fingers crossed for the next steps with your engine, especially the pressurized runs. Looking forward to seeing how it goes.
Best regards,
Siar

[siar]

Hi Ralf, DeltaEC runs natively on Linux for me — no Wine needed. There is DeltaEC 7.0.2 for Linux on the DeltaEC page (it’s labelled “DeltaEC 702 Linux”, built on Ubuntu 22.04): https://www.losalamosthermoacoustics.org/deltaec.html
I’d simply let an AI coding agent (Claude Code, Codex CLI, …) with shell access do the setup: download, unpack, install any missing libraries, do all the troubleshooting. The same agent can then drive DeltaEC via the console mode — that’s what I do.  Best regards, Siar

[siar]

DeltaEC in batch mode on Linux, driven by an AI coding agent

Hi Ralf, hi everyone,

a workshop note rather than a result. I have been rebuilding your printed loop (compliance, hose, T-junction with membrane) in DeltaEC on Linux, and one thing turned out to be worth sharing: DeltaEC 7.0.2 has a console mode (DeltaEC -c model.out) that runs without any X server. It is a small text menu on stdin: r runs the model, w writes the result into the .out, y confirms the overwrite, e exits. With that, a shell script can run a model non-interactively, and an AI coding agent (I use Claude Code) can write dozens of variants of the .out, run them in a loop, and read guessz/targs back from the end of each file as the starting point for the next. Parameter sweeps over mesh, hose length or heater temperature then become a few minutes of unattended work instead of an evening of clicking. The agent also makes a reasonable DeltaEC assistant: it reads the manual, explains segment parameters and finds the reason when a run does not converge.

The limit is the usual one: the model is only as good as its inputs. For your loop I have the CAD parts, but two numbers would help more than anything else, if you happen to have them: the moving mass in grams for the runs where you measured 1943, 2039 and ~2600 rpm on the membrane, and the heater temperature at which the engine starts by itself. No hurry, and thanks for publishing the files and the test runs in such detail.

[Author]

Posts