MPG Baseline for Insight 1 Nailed-Long!
 

OK, the final data is in. After some false starts, I have completed the testing and arrived at a fuel economy(FE) baseline that I can use for aero testing. I now have high confidence data that I believe will stand up to independent scrutiny.

As you may recall, I have been concerned about establishing a reliable and repeatable Insight 1 FE baseline against which to measure FE improvements. The immediate aero modification goal is an underbelly plus boat-tail prototype. The discussion which follows is geared primarily to the Insight 1, though some of the methods are appropriate for other cars. The Insight 1 is a difficult animal to drive consistently for a variety of reasons. Complications include: a. no cruise control, b. lean burn mode, c. low power, and d. high consumption air conditioner.

The steps as I saw them were:
1. Find a course. It is necessary to find a course which allows for a near-constant speed round trip between two fixed points. Optimally, the course starts and ends a short distance from one’s home or base of operation. The course should be as level as is available to you in your geographic area. Ideally, there would be no speed variation for hills, but this is a virtually impossible constraint on the Insight 1 in most areas of the country. It is necessary that the course be two way so that wind variations are cancelled or minimized. The course should be such that the Insight can maintain lean burn at the test speed. The turn around should be easy and uncomplicated, such as a cloverleaf without lights. The course should be of some significant length (statistically undefined) such that minor errors in driving technique or conditions will not produce significant distortions of the data. My course length is 48.2 miles.
2. Use a consistent driving technique. It is necessary to have repeatable data, therefore it is necessary to employ a consistent driving technique. Though there are hypermiling techniques which will improve FE, i.e. move the baseline upward, some of the these techniques are inconsistent in their application and results. Two examples are forced auto stop(fas) and pulse and glide(p&g). The advanced hypermiling techniques should not be used for baseline testing. A nice smooth accelerator application, anticipation of holdups and lane changes, and adherence to a speed band criteria are necessary. If a stop occurs anywhere along the course, the data run must be abandoned.
3. Maintain lean burn. The Insight 1 produces dramatically improved FE when the engine is in lean burn mode. As has been observed, the lean burn condition is primarily dependent on engine load. Therefore, the Insight will click out of lean burn if the cruising speed is too high, or if a constant speed is demanded on hill climbing. This situation leaves the test driver with two options: a. he may attempt to maintain lean burn at all times by allowing minor speed variations on hills, or b. he may set the test speed high enough that the car is never in lean burn during the test run. I chose the first option. The second option is fraught with problems which will produce inconsistent FE results, the most obvious being that the car will go into lean burn on downhill sections to speeds as high as 75 MPH, an illegal and impractical speed in my area. If the car clicks into and out of lean burn frequently, then the data set contains an unknown mix of engine conditions and will be unrepeatable.
4. Set a speed. Aerodynamic drag rises with the square of the vehicle speed. It is therefore necessary to have a repeatable speed criteria. Since small variations in speed are expected, and experienced along most practical test courses, it is necessary to establish some speed constraint other than constant speed. I chose to constrain the driving of the course to a constant average speed for the course. By using a ScanGauge, one can measure the trip average speed frequently and make small adjustments to the throttle. For aero testing the test speed needs to be as high as practical for the conditions, while allowing the engine to maintain lean burn virtually at all times. Through trial and error on my selected course I found this speed to be 60 MPH average, with an observed 2 MPH allowance for the turnaround. This resulted in a ScanGauge displayed average trip speed of 58 MPH.
5. Gather multiple data. In a rigorous statistical sense, it is probably impractical to gather enough data on a course of realistic length to meet standards of statistical significance. Having rusty statistical skills, I don’t actually know how to approach the problem of statistical significance. It is noted that others have felt that 3 runs were adequate for an FE baseline. I chose to make 6 runs, just to be a bit more assured of consistency.
6. Minimize the variables. As others have found and commented, it is absolutely necessary to minimize the number of variables affecting the data set. Common error sources are wind, temperature, barometric pressure, traffic conditions, and driving style. It goes without saying that the best data is produced when there is little wind and when the temperature is within some reasonable bound. Consistent driving style is a given.

Having laid down this elaborate but necessary criteria, the course was driven 6 times (actually 7 times, but one test was hopelessly compromised by driver error). As one would expect the data appears to be a normal distribution. It has a small standard deviation, indicating that the data set is of “high quality.” The results of the individual runs follow:

Run 1:Temp. 81F, wind 0, 87.8MPG
Run 2 Temp. 75F, wind nw@2, 90 MPG
Run 3 Temp. 90F, wind nw@2, 89.5 MPG
Run 4 Temp. 90F, wind sw@3 91.8 MPG (MPG adjusted for speed by 1 MPH by speed squared method)
Run 5 Temp. 91F, wind o, 90 MPG
Run 6, temp 93F, wind w8, 89.5 MPG
The data has the strong appearance of a well-defined normal distribution, with 4 data points tightly clumped and two outliers. The mean of the data is 89.76 MPG, and the standard deviation is 1.285 MPG, very tightly clumped high quality data.

So the bottom line, after all this thrashing about, is:

My Insight will do 90 MPG at 60 MPH on the Richmond area interstates, in the summer. Now to the mods!

Nice job in attempting to get reliable and real-world data for gas consumption! As MetroMPG has pointed out in the past, this is not trivial. My coast-down numbers also have made me take a cautious approach.

I don't know if this will help much, but at work we use MiniTab for statistical analysis on occasion, and a sampling of 3 or even 6, at least statistically speaking is not much to work with for a sample size.

In the past, some of our company statisticians have promoted the use of at least 30 samples just to get a rough idea on how well the manufacturing process is working, and how much standard deviation the data has.

I try to keep a gage history of at least 20 samples and perform either 2 or 3 sigma standard deviations on the data set to see how a particular gage is performing historically.

When doing coast-down testing last fall/winter, I accumulated around 40 or 50 runs before getting enough data to really start making predictions on how the barometric pressure and temperature affect the coast down times. I still have a ways to go on this, and having the tail extension will change the aero enough to setup the chance for more runs this fall/winter.

Hope this helps.

Jim.