So I go the Shocktek modified to field serviceable Bilstein route early last year with the intention of learning something about the art of testing and development, and I document the initial difficulties only implicitly (you can read about it in the H-T archive that you can't access), in the end anouncing that I've bought a set of Mugen N1's to get me to the Expo, and I drive the rest of the summer on Scott Zellners recommended starting settings and focus on my development as a driver - which I find a challenging enough task.

Now, with the end of the season and time to reflect and plan, I'm back to thinking about development. As it happens I made some skeptical remarks in a thread recently about the first two installments of Racecar Engineerings new series "Design of Experiment" that offers a description of a superior (in terms of the economy of testing time) alternative to the dominant rule "change one thing at a time". Paul Van Valkenburgh had confessed to a conversion of sorts along these lines a few issues earlier. I was shortsighted and unimaginitive in my remarks - I suggested that if what I had read to that point had appeared in Road & Tracks April issue it would read about the same. I brought up the subject at my first race car lunch in more than a month this last Tuesday, and in the process of asking my friends what they thought of the series I suggested the possibility of the methodology using a matrix of sorts. Nobody wanted to think about it too hard, so that was that. A couple of days later I find the new issue with part 3 and yup, that's what it is.

The author lays out a reasonably instructive example and provides some interpretive guidance. It is not a specific blueprint. It only suggests the method. The treatment of the subject so far has only briefly touched on the experimental and analytical hazards. I'm now eagerly awaiting the next installments without major scepticism, though I am concerned about interactions between factors being a significant problem and the specific selection of factors and their bounds being challenging as well - an experiment's success depending greatly on it's design. Hence the relevance of the name of the series.

If you're not a regular reader of Racecar Engineering, and this sort of thing sounds interesting and or like something you can see yourself carrying out reasonably rigorously, then you should be a regular reader.

Scott, who wishes his local newstands had RaceTech as well...but they don't.....so who is (John) Hill anyway?

 

The testing matrix was quite interesting. I too am interested in the next installment. I need to read a bit more in depth to see exactly how I could apply this and properly interpret the data, but it's something to think about. PS, my local Barnes & Noble HAD RaceTech last month, and I purchased a copy. I went there last weekend, no RaceTech. I think I'm going to have to throw a temper tantrum there, or hope the new shipment is on its way.

 

I had a class in college about statistical experiments and the sort. Cool class. (I had a class in college about it, that makes me an expert, right? ) The big thing with a test matrix and reducing its dimensions is that you have to assume second order effects are negligible. In other words, you kinda have to assume that things don't work together better than they work separately. I don't have the level of experience to say whether that's a good assumption for chassis setup or not. I'm guessing this guy seems to have the experience and think it is.

Another problem is that I can't see how you can possibly build up a statistically valid number of data points. It would seem nearly impossible to perform the statistical manipulations to gage the quality of your data and your assumptions.

 

I haven't read the article but I assume they are using Taguchi techniques to reduce the number of tests. We're doing the same thing in class right now. I have to build a catapult to shoot a golf ball, and we use the DOE to characterize the process and predict the settings to reach a target.

There are so many variables associated with vehicle performance, and the definition of "better" is so vague that I think without a good daq system you'd be chasing your tail to try to use this methodology.

I guess its time for a trip to B&N. If I wasn't a broke college student and the magazine wasn't so damned expensive I'd probably have a subscription.

 

The apparent scheme is quite simple - at least the author suggests that the "maths" need not be messy (those British). And better is simply the lap time

I've read it, I've thought about it, I've awakened thinking about it. I'm going to need help understanding how we arrive at the individual effects graph. So far all I can think is that the factor setting with the highest correlation to the lowest lap times would be the most responsive, and so on. Sure there is a risk of factors interfering with each other, but I'm thinking that the goal of this methodology is to speed the initial approach to what I called the absolute maximum in a previous thread - it seems reasonable to think that that absolute maximum might be proximate in the topography to the indicated value of the most responsive factor (assuming alot about the selection of factors again of course). This especially seems relevant to our particular cars inasmuch as there might be some uncertainty as to just what that most responsive factor might be (rear spring rate?).

I'm pretty sure that I won't be putting together a nice example of such a test program for the reason that I'm probably not up to the task as a driver, and I'm probably not up to the task as a wallet. But it suggests that in pursuing greater performance thru a program of simultaneous multiple changes (well documented to be sure) that one might realize useful improvement and less risk of ridicule (and that's really important to me .

Scott, who thinks he is currently underdamped in the front and in the process of remedying this he's going to have to change other stuff (more front downforce please)....aww, how bout some more rear wing too....yeah, that's better....

 

Heh.. it's too bad there isn't a computer game whose physics model you trusted well enough and was detailed enough to run some of the permutations out in that area. plus validating the results against reality would probably be close to as time consuming as the tests you would be trying to avoid.

I'm enjoying playing with Sports Car GT right now.. with the vehicle data files editors you can do some really nifty things.. and the amount of physics involved is actually rather impressive for a game. I enjoy switching a VW rabbit to RWD and trying to drive it without changing and spring/damping rates, or swaybars or camber or anything.. ok, now I've digressed..

 

With SCGT (Sports Car GT) I've taken an MR2 .veh file and made it handle as much as an integra as possible (understeering, bodyroll, power & acceleration)...Got it going really good on the Fort Worth Auto-X course too. pretty fun while I was running it, too bad my computer is down. I'm really missing SCGT. The best thing I have right now is Super 1 Karting simulation on my fathers PC...which is really sim BTW.

Applied Physics, Trial, and Error...my method for testing on my integra.