View Single Post
      12-08-2007, 04:12 PM   #192
Lieutenant Colonel

Drives: Legacy GT - 13.704@99.39
Join Date: Jul 2007
Location: Manheim, PA

iTrader: (0)

Originally Posted by swamp2 View Post
Well, Bruce, what continues to be seriously flawed is your interpretation and understanding of the basics of simulation, accuracy, and random vs. systematic errors. All of the numbers you have posted (your personal runs, etc.) are basically “noise” clouding the basic issue here.
I guess that's true. If you're satisfied with hideous, systematically incorrect results, then I suppose I should be, as well. My systematically documented results over the years are then obviously without meaning. Point to you.

Originally Posted by swamp2 View Post
By the way I’ll never let you forget that you are the one so foolish to claim here on this very form that you “contributed” to a similar simulation software tool capable of validation with actual tests to within HUNDREDTHS OF A SECOND. Let me quote you, just to remind,

"I helped design a quarter-mile simulation tool a number of years ago, called "ShiftMaster", so I have a little knowledge of the topic. That tool was accurate to within around a hundredth anywhere during the quarter mile. I have no knowledge of CarTest, but assume from your rantings that it's pretty good."
Here's another shining quote for you to remind me about in the future:

Not only was ShiftMaster capable of delivering results at that level of accuracy, it pretty much had to be that accurate. We bought a number of such packages available at the time (this was a number of years ago), and pretty much all of them could deliver results down to around that level. In other words, not only was it possible amongst a number of available packages, it was routine.

I should qualify this a bit, I think, since the last time you and I had a dust up (about time to speed having essentially nothing to do with time to distance), it was due to a misunderstanding.

What I'm talking about is taking actual drag strip results, documented along with weather conditions, starting line techniques, subjective traction levels and whether the car was able to hook or not off the line and during shifts, plus shift points and style, plus etc if I've forgotten something. I used to write notes on the back of the timeslips. (Anal? Who me? I was potty trained at gunpoint.)

At the time, it was pretty straightforward to get extremely good accuracy with packages we bought before ShiftMaster came into being, and that package eventually got accuracy as close as anything else we tried. We also bought a ton of SAE papers on the subject, especially those we could find on rotational inertia and the vagaries of traction coefficients at various levels of slip.

I should also mention that we only tested in depth on material we knew we could trust, which was essentially our timeslips with our intimate knowledge of what had gone on during those runs.

Still, the result was a package that was competitive with others we had hands-on knowledge of at the time - and to repeat, they were all accurate to a level that you say is impossible. All of them.

As a final sort of disclaimer, let me also say that I'm quite sure that we would've stumbled from time to time on a car that more or less refused to be accurately simulated, for whatever reasons. Hadn't happened up to the point I lost interest, however, and I'm also quite sure that that would be the exception rather than the rule.

Originally Posted by swamp2 View Post
I don’t want to get into a detailed case by case validation of CarTest. That really is not the way to validate it anyway. I should do a detailed statistical analysis with means and standard deviations of the test data (for many vehicles) along with convergence and Monte-Carlo analysis (parameter variation, basically) on the simulation side and then I could rigorously establish the accuracy of the tool and the accuracy would be metric dependent (i.e. a different accuracy depending on the test). Not wanting to get so formal (and waste so much time!) I have resorted to a process of informed inductive reasoning; simply comparing it to a variety of cars using a variety of metrics combined with a good understanding of the test measurement process, simulation in general, statistics, error analysis and physics to show me that it is “reasonable” (I’ll define reasonable soon…). It has shown quite reasonable results for key stats such as 0-60, 0-100, 0-150 (all mph) as well as 1/4 mi time and traps. I have also done some work on in gear speed to speed times and some rolling start tests allowing gear changes. Again results have compared reasonably well with tests. I have not performed validations nor claimed accuracy outside of this fairly narrow domain of metrics (for instance speed gained in last 1/8th mi of the 1/4 mi). I do not doubt that some predictions from the software are worse than others. That is the basics of simulation, numerics and physics, it is generally harder to predict the derivative (and since your not really a math guy I mean formal mathematical derivate or slope of something) rather than the something itself. [b]So back on point what do I mean by “reasonable”? I mean within a few tenths here or there in the lower speeds, maybe a second or so on the fastest ones and generally a few tenths and always less than 5 mph for the 1/4 trap speed (often much better than that on trap). Why are these numbers “reasonable”? Simply because they are typically within the average numbers reported from tests plus or minus the variation in the reported test numbers. This is the hallmark of an acceptable/reasonable simulation. Each test is attempted to be controlled but in fact is fairly substantially uncontrolled, major contributing differences include driver, technique, temperature, surface quality, tires (type and inflation) age of vehicle, natural differences vehicle to vehicle, etc. If you want to see this “in action” simply have a look at the thread here called “More performance figures”. It shows this real world variability for the E92 M3 as well as for a bunch of its competitors. So the result of all this is a fairly large variation in test, none for simulation. The simulation is deterministic; same inputs always gives the same outputs, exactly.
Other than the defensive hyperbole, what I get out of this is, on a car with reported trap speeds between 105 and 110 mph, you're OK with a 100 mph simulation.

Wow. Tell me I'm wrong.

As an aside, my guess is that CarTest really *is* an accurate tool, but not in your hands just yet.

Originally Posted by swamp2 View Post
This all comes right back to your infamous quote above. There is simply no way to make a simulation tool that matches a widely variable real world test “perfectly” (with my basic definition of perfect here being +/- 1/100th second – right from the horses mouth).
As mentioned previously, there are apparently a bunch of proles out here on the actual planet who have had better luck than you have had.

Originally Posted by swamp2 View Post
The whole new car vs. old car thing is so simple I don’t even feel it warrants a reply but here you go. A broken in car will typically produce more power than a new car. If you can measure that effect on an accurate ENGINE dyno and put the results into the simulation guess what – you could predict these differences, again, “reasonably”.
Not sure why you bring this up, but that aside, I don't have a problem with what you've said here.

Originally Posted by swamp2 View Post
I also disagree with your closing PS. A high redline and gear ratios absolutely have a large impact on vehicle performance both directly and indirectly. Sure, as I have stated and agreed with many on the point, time and time again, the key figure for either track or strip performance is simply power to weight. But gearing and redline matter as well. Gears provide torque multiplication and they do so in all gears at all times. Larger ratios either gear or final = more torque multiplication = more acceleration, period (with the caveat that indeed the wheels are not spinning). Of course there is a limit to this, there is no such thing as a free lunch, you can’t simply raise and raise the gear ratio endlessly, traction is a limiting factor at lower speeds as well is keeping the engine in its best operating rpm band where is produces high torque lastly there is the limit of having such large ratios that it requires too much shifting. A high redline allows one to stay in a gear longer and maximize time spent accelerating instead of shifting (slowing down) and shifting to the next gear where acceleration will be markedly less. Would you disagree that raising the redline on a car, all other things equal (and assuming it still produces reasonable torque in the increased band) it will make the car accelerate better through the gears? The indirect benefits of a high redline and numerically larger gears is more evident on the track than the strip. Vehicles designed as such are typically low torque, high power, high redline, and light/low intertia drivetrain. All of these add up to a car that can not only accelerate but one that can handle.
There's a chance we may have another misunderstanding here, but that said, gearing and redline essentially have no bearing on the matter, which in fact is only power and weight. In addition to the information I've posted on the topic in another string in this file, there are in fact only three caveats here that I know of. The first is that, having a redline a good deal higher than the power peak means that you can wind the engine out higher, and thus arrive in the next gear further up on the power slope. In that single case, a "higher" redline (in this case meaning higher than the power peak, rather than in absolute terms) will give you an acceleration advantage. The second is that, given an engine with 414 HP and 295 foot pounds with an 8400 rpm power peak and another making 590 foot pounds and the same 414 HP at 4200 (each at their redlines), each with the same approximate internal rotating inertia and the 8400 rpm engine geared twice as aggressively as the 4200 rpm version (so that each engine can get into its power band with roughly the same alacrity), I'd bet on the 4200 rpm combo to win in a drag race, simply because it has less overall rotating inertia to contend with. The third is that I'm assuming any mass production car is geared in such a way as to take advantage of its power band in everyday use. No fair installing gearing tall enough so that the engine can't get into its power band until you've hit 80 mph in first gear, as an example. Hence my comment in another string that, after the first 60 feet or so, gearing essentially doesn't matter.

As we've said before, horsepower is the great simplifier in this context. You can certainly go through the torque-at-the-drive-wheels calculations, but when you do, you'll see that you didn't have to.


Last edited by bruce.augenstein@comcast.; 12-08-2007 at 11:55 PM. Reason: Change 295 foot pounds to 590 foot pounds