I think I started thinking of a way to improve gas mileage about 5 years ago. I have been tinkering with it on-and-off since then. One of the features of the Buick RoadMaster that Ali and I bought was the throttle-body fuel injection because I thought it would be easier to modify than the direct injection of newer engines. Despite learning lots about how I would tackle the problem, I don't think I'm going to worry about ever actually implementing it; instead, I'll focus more of my energy on human-powered vehicles like custom bicycles and such.
I read an article that included a graph of engine efficiency for a Volkswagen (I think) that plotted efficiency (horsepower per gallon of fuel) as a color against throttle position and engine RPM. In this particular engine, efficiency varied between about 5% and about 30%. One way to think of it is that for any given engine speed (i.e. 2500 RPM), the efficiency the engine converts fuel to mechanical power varies with throttle position. A rudimentary observation is that the engine is more efficient at 50% throttle than at either 5% or 95%. Also, there is a "sweet spot" — a throttle position that is the most efficient (or a range that's pretty close) — for any given engine speed.
Automotive designers have not done much with this information as far as I can tell. They try to make the "sweet spot" bigger for efficient cars, they try to set the top gear in the transmission so average highway speeds are in the sweet spot, and in a few cars, they switch off half the cylinders sometimes to try and change the sweet spot.
My idea is to change the fuel system more radically. First, switch to "throttle by wire" — make the accelerator pedal more akin to a "torque selector" than a "gas pedal". The actual engine throttle would be computer-controlled to try and maintain the most efficient engine output for its current output RPM. To control the amount of power the engine actually produces, the computer would disable fuel to suppress firing of certain cylinders at a ratio that approximates the desired power output requested at the accelerator pedal.
For example, if you're driving up a grade on the highway and need the engine to deliver 40 horsepower, the car might be running at 1800 RPM and you'd have the throttle at 40%. Let's say this gets you an engine efficiency of 12%, but at 1800 RPM, the "sweet spot" 60% throttle you can get 20% efficiency. At 60% throttle, though, the engine delivers 80 horsepower and you'd be accelerating. So the computer would turn off the cylinders half the time so the effective engine output would be 40 horsepower, but the engine efficiency would nearly double — and so would your gas mileage.
The trouble is, it's quite a time-consuming, complicated project. The first step is to measure the engine data — and that starts with building sensors and recording equipment to get a good set of efficiency and power output data for a spectrum of both throttle positions and RPM. Then it's a matter of analyzing that data to get the target throttle positions and ratio calculations to match the existing performance of the accelerator pedal. But then it gets complicated: you need a throttle actuator, an electronic accelerator pedal, and a way to send the engine computer corrected data from the exhaust oxygen sensor (i.e. turning off half the cylinders increases exhaust oxygen a lot) — probably more sensors too, and a computer to process all that information real-time. Of course you need to make it safe, and be able to record data so you can present it truthfully.
And if everything goes perfectly, it's a gamble as to how much improvement you'd actually get. The thing that kept me interested in the project was the prospect of doubling the mileage — going from 20 MPG to 40MPG. I think it's more realistic to consider a 10%-20% improvement. But without the big step of collecting data, I don't even know at all.
I promised myself this year that I'd put forth extra effort and really try to make it work. I made this promise for 3 years now, and I still have no system. So I'm relieving myself of pretending to get it done. At the end of March, I was supposed to have a working Controller Area Network (CAN) to reliably communicate data between the various microcontrollers in my system. It's now the middle of May and I have no such system. So it's time to zip up the body bag and pack this one away. I learned a lot in doing research, and I'm glad I did, but I just don't think I'm going to bother finishing it.