Throughput analysis of junction designs

[gustaphe]

I just passed a course in college level experimental design, which apparently means my report was statistically valid.
Might not be a very interesting read for everyone, but someone might like it.
If nothing else, it could be an inspiration for bringing Factorio into even more parts of your life.

Here's an excerpt:

4.2 TL;DR

Use Selfdefeating’s celtic knot intersection [8] if you want to look cool while going fast,
and Drizznarte’s four-way crossing [4] if you want fast left turns.
Nothing conclusive can be said on roundabouts vs. crossings.

[Koub]

You should definitely have a discussion with aaaargh

[Ohz]

Is there an interest to do a double sized junction? Junction so large that a train can fit between any crossed rails?

[featherwinglove]

I heard of a guy on Reddit who failed his Master's thesis because he was playing Factorio too much to defend his dissertation. I'm glad to see someone using Factorio in his post-sec homework instead!

[aaargha]

Oooh nice, more train science. Using statistical analysis to ensure the results are actually valid? What sorcery is this?

You should definitely have a discussion with aaaargh

Naaahh

Ahem, anyway...

I've been doing something similar over in another thread and, if you want to keep exploring this beyond your report, I'd be really grateful if you took a look and, if you're still interested, perhaps discuss improvements to the setup/measuring? I think my current setup does/could do most of the things in the outlook, though I'm not sure if my methods are good enough.

Now on to your work: Your approach seems much more precise than what I'm using, I also like that you can measure things for individual paths. Looking at this I really should make some improvements to automating my tests more.

In order to eliminate risk of blockage, each destination needed at least three stations,to allow all three trains to visit simultaneously

While this is sound reasoning won't using terminus stations pretty much make this moot? Trains leaving/entering the stations still have to fight for the same blocks, thus blocking each other, would not a RoRo design be preferable in this case?

As for the bias for/against N, could this possibly be explained by trains having different priorities when trying to enter shared blocks? My current theory (only supported by really basic testing) is that it's based on train ID, basically build order, where the lowest ID has the highest priority as it is processed first. If nothing else this could possibly skew the results when comparing left/right turns to going straight.

I also think it's interesting that the Celtic knot is not the clear winner when it comes to left turn performance as it is the only intersection of those tested that allows two opposing trains to turn left at the same time.

I'd take a closer look a the signalling of the intersections but the pictures a bit too small for details, from what I can see though is seems like there are some differences in how optimized the signals are on each intersection. Also, just for curiosity, do you use "properly" sized output blocks or are you going for straight throughput?

I'd guess that the reason you're having trouble differentiating A, B and E is probably that they're all good enough to handle the relatively small amount of traffic you're using. But it's pretty hard to compare as I'm measuring the amount of trains passing the intersection in 15 minutes and you're using station to station time. (I was about to do some serious errors trying to compare throughput with this )

Anyway, it was a very interesting read, good job and thank you for sharing.

Is there an interest to do a double sized junction? Junction so large that a train can fit between any crossed rails?

If you post a BP string in the thread above I'd be happy to run it through my setup, at least as long as you keep it shorter than about 1.5k tiles

[gustaphe]

Honestly, I'm not sure if I even tried to see if someone had done this before me. My priority was passing the course, helping the community was secondary a bonus.

The reason I went with stations like this was simplicity. Not to compare it to RoRo, but the most important part for me was that trains could clear the testing area, because I figured any competition at the choke point would cancel out in the statistics.

I'm thinking the N bias might otherwise be caused by whatever did the inserter weirdness. When that was fixed, this report was long ago submitted. Might redo the tests to see, if I find the time.
Now that you can get train ID from a station to the circuit network (again, implemented too late for me to use ) your theory should be quite easilly testable. I don't think I ran enough trains for them to ever arrive at the junction simultaneously, or for enough of them to do so to have an effect.

The signalling is one of my insecurities about the experiment, I can't say for sure that they are optimized, or even equally unoptimized. I tried to be consistent, but I'm not certain I managed.

But yes, all in all I'm satisfied. I'll take a look at your stuff soon and see if I can add anything, but at a glance your setup seems randomized enough for statistical analysis to be unneccessary.

[gustaphe]

I heard of a guy on Reddit who failed his Master's thesis because he was playing Factorio too much to defend his dissertation. I'm glad to see someone using Factorio in his post-sec homework instead!

You have to pass a lot of courses to get an engineering degree, so I can't guarantee I won't be there with him. So far I don't think it has cost me any grades that I wasn't going to fail anyway.