Quote:
Originally Posted by arghx7
So here's what most likely happened. Somebody at Subaru or Toyota ran an engine dyno at a bunch of steady-state speed & load points. So at 4000rpm 1.00 g/rev they ran the engine for maybe a minute. All the spark timing, AVCS, etc were all adjusted at each point. The dyno measured the engine torque. It was all spit out into a big Excel-readable file by the dyno control system.
Then they loaded the data from the file either directly into the maps for the ECU, or into some Matlab tool or Excel macro which processed it and spit out the values. There wasn't necessarily a ton of thought put into the numbers in the table. Some guy followed some procedure, that's all. The torque curve of the engine was most likely established on the dyno at each individual point before those maps were ever made.
The FA20 doesn't use an actual model of engine torque, not from what's been seen so far. It's just those look-up tables, which as I said came right from an engine dyno into the maps with maybe some tweaks on top of that.
|
This guy knows what's up. Listen to him.
Tuning starts in the dyno lab. Steady-state testing is great for developing the base tune and working the bugs out of new functions/technology/sensors/you name it.
Once the software design has been validated conceptually, you move on to a production-level validation. Most of the manufacturers will send out fleets of company vehicles, driven by underpaid interns, co-ops, and engineers. These cars usually have a few non-production sensors and fancy data logging systems. Everybody drives like normal, and they get an accurate distribution of people's driving habits... and the car's response them. Send them into all sorts of weather, location, driving styles, etc.
Due to the adaptive variables and correction factors in modern software, it's pretty likely that the car will settle into a semi-usable tune without alot of fiddling. Inefficient and unresponsive, though. The fun is analyzing "how wrong" your open-loop controls were, and determining the root cause. Seeing how far you can push the system before it acts up in altitude, snow, etc.
Alot of modern datalogging systems allow you to update the engine variables in real-time. Most of the legwork is done via test-n-flash method for now.
Side Note: Model-based tuning is the way of the future... but it's a tough sell to management. Even with unlimited computing power, they'll still want to use "streamlined" versions that barely resemble the underlying physics. It's trickling into most code, but I'm not too satisfied with the implementations I've seen thusfar.