Truth Space
Truth space
TDC
Kevin Cameron
How CAN YOU DISCOVER THE TRUTH? No, I’m not talking about the kind philosophers seek, just the ordinary useful kind that makes more power or reduces fuel consumption-physical stuff.
Easy? Just get on the dyno and start thrashing? Not so fast. According to California legend, a couple of decades ago a wealthy racing enthusiast decided to get to the bottom of the vexing two-stroke power mystery once and for all. He instructed his team to order 40 TZ250 cylinders and a hundred pistons from Yamaha, and to modify and test one variable at a time. Step-by-step testing was sure to uncover The Truth.
It didn’t. Instead, every modification produced less power than stock. He therefore concluded that the cylinders were as good as could possibly be made.
We know that’s not true-250cc roadrace engines were making 45 bhp then, but now they more than double that. How could The Method fail? I found out when I built a 250 for the 1977 Daytona race. I widened the exhaust port 3mm, raised it 1.5mm, took 20mm out of the header pipe and jetted-in a pair of carburetors 2mm bigger than stock. Rider Richard Schlachter took those quickie mods to third place. The wider, taller exhaust port would have been useless without running the engine at higher revs, so the pipe had to be cut. There wasn’t much point in letting exhaust out faster if intake wasn't increased to match. If we’d used the step-by-step method, we’d have gotten nothing, and finished back in the pack, where we’d come from originally.
My point here is that answers-Truth, if you want to call it that-can be complex, and finding answers may require the changing of more than one variable at one time.
Experienced Superbike tuners tell a parallel story about chassis setup. It's not easy to find a workable chassis/suspension setup, and the search can consume many a rider-year. They compare looking for a good setup with playing golf in a completely dark warehouse, on a perfectly smooth, featureless floor. Somewhere in that floor is the cup. You can play a long time and never find it. We can call the floor of our warehouse “setup space.” Its two dimensions are a mental analogy to the many more variables that actually affect motorcycle setup-ride height, damping curves, mass properties and so on. Somewhere in the “space” defined by those variables (the math people call them “dimensions”) lies the multi-dimensional analogy of the golf hole-the answer, The Truth, the correct setup.
In many kinds of research, we expect to find that the floor slopes down slightly when you get near the cup, so you can tell you’re “getting warm.” Yes, says your rider, that last rideheight change was better. Let’s try more. But what do you do when your problem best fits the “golf-in-a-darkwarehouse” model?
Scientists call their method for doing this “combinatorial analysis.” This simply means that you write up all the possible combinations and permutations of what you think are the important variables, and you try them as fast as you know how, hoping for a score. It’s basically the same as buying lottery tickets by the dozen or feeding silver to the dollar slot machine. At the racetrack, it means testing, testing, testing endless combinations, starting with (you hope) reasonable assumptions about approximately what the variables should be.
What if your assumptions are wrong? What if you’ve left out variables altogether? You fail-as many have.
A recent issue of the New York Times describes some fascinating diesel-engine modeling work now going on at the University of Wisconsin under Prof. Rolf D. Reiz. This work jumps well beyond shotgun research and finds a way to detect and use even an imperceptible slope in the floor of our darkened warehouse.
Their method requires that the motion of engine parts, the flow fields of gas in ports and cylinder, and the interaction of the fuel spray with the turbulent gas motion above the piston all be modeled in detail, using the best physical-mathematical descriptions available. This model runs on a modest supercomputer to produce a result, in terms of predicted power, fuel consumption, heat loss, production of emissions and so on. This much has already been done by others in many places.
What the Wisconsin researchers have added is an ability to perform self-directing computations that mimic biological evolution-what is called a genetic algorithm. The supercomputer calculates the results for a given set of values for the variables. Then it changes them and recomputes. If the recomputation improves some aspect of the result-even slightly-it explores farther in that direction. If there is no improvement or if the results are worse than before, the next computation goes a different way. Because the computer, with its multiple processors, can make these computations fairly quickly, it can explore and map a lot of “space” in a short time. It can find promising avenues of actual physical research in much the same way that nature worked out that horse-like animals survived better with longer legs, more acute vision, and a tendency to shy and bolt at anything unusual. Most valuable of all, it doesn’t take 55 million years to complete the experiment!
This genetic engine modeling allows exploration of aspects of engine design that have been too complex and timeconsuming for trial-and-error dyno methods. This includes the detailed shapes of cylinder heads and piston crowns, which determine how turbulence develops and how flame spreads. Engineers know a few big crude things about these questions, but little in the way of useful details. This is part of why new racing motorcycle engine designs so often take a couple of years or more to mature-it takes that long to figure out what’s really going on.
I know a lot of people are tired of computers, and tired of their Os and Is getting into everything. Sorry, there’s only one other method proven to work with certain kinds of complex problems, and that one takes the full 55 million years. We can’t wait that long. □
