EPA Notifies Fiat Chrysler of Clean Air Act Violations

The testing process is required to pass a gauge reproducibility and reliability study. See https://en.wikipedia.org/wiki/ANOVA_gauge_R%26R.

Essentially, the goal is to analyze the results when a given vehicle is tested in different conditions. In your scenario, each of three operators might take a dozen different makes and models of vehicles on a "normal drive" on each of three consecutive days, resulting in 3 operators x 3 measurements x 12 vehicles = 108 NOx numbers. They would then analyze the results statistically, checking the average value and standard deviation. This should prove that each of the three operators produces the same result ("reproducibility"), and that each day produces the same result ("reliability").

There will be some variation - the vehicle may change a little from one run to another, even if the control inputs are identical. And the inaccuracy of the probe will also produce some small random variation. Those can't be helped - though they should be known. But if you use different operators, lead-foot Larry needs to produce test results that are indistingushable from when slow Sam runs the test, and they both need to produce the same values on windy days and snowy days.

This is easy if the only function of the operator is to strap the vehicle to a dynamo, check the HVAC, tire pressure, and so on, plug the control cable into the OBDII port (or wherever it goes), and push the "start" button on the tester.

Perhaps the best solution is to collect data on hundreds of actual, reasonable, human drivers and feed those noisy, difficult to distinguish profiles into the test system. But that means expensive dynamos, and lots of hours of testing.

Isn't this like saying that shooting deaths were a much bigger problem in the Old West, so it's not fair to jail someone just for shooting a few people? The crucial fact is that you're breaking the law and killing people in the process, not that you would have killed even more people if the law hadn't been there.

> The natural state of all groups with the same interest is collusion.

Once you reach marginal gains in engineering, you don't need collusion to have all the engineers doing the same things.

A good example is power consumption in microprocessors. For a given technology node, you will burn a certain amount of power for a certain size of chip. Period. You can be much worse than everybody else, but it's really hard to be much better.

The law does give a pass for murder now. Just half the number it gave until a few years ago. So it's really hard to argue that a company in this situation is morally responsible for the murder of some number of people. As that would be equivalent to saying that the government has decided how many people you can murder, perfectly legally, every year. So unless you accept this, it's not murder.

As someone who manufactures automotive test equipment, please apply a bit of Occam's razor. A lot of the "shady" stuff they're doing is just reducing and simplifying the cost of testing.

It would always be nice to run a more accurate, more real-world test. Certain Ford specs, for example, require multi-axis robot arms with independently mobile manikin forms to apply loads to the seats. Old Ford specs, and some current competitor specs, just have a hemispherical indenter that moves through a sinusoidal profile on a straight line at a fixed speed by an electric motor.

Hardware for the "realistic" test costs hundreds of thousands of dollars, plus about a man-year of expert engineering time by the test lab or a systems integrator to turn the robot arm into standards-compliant, turn-key test equipment. We're talking high-speed data aquisition, 3D modeling, CNC machining, user studies, and all that fun stuff. Checking that it works correctly requires high-speed cameras, pressure mats, and a lot of work.

The old-style test requires bit of steel, a few bearings, an on/off switch, an electrical counter or timer (that even an ME can wire up, and that's saying something!) and an electric motor for cost a few thousand dollars, and can be built by anybody with a welder and a few wrenches, maybe a machinist lathe and old Bridgeport mill if they want to get fancy. You can verify the correct operation with a pair of calipers, a stopwatch, and a gentle kick to make sure the bearings are still tight.

Or there are multi-million-dollar sled systems that throw assembled prototype vehicles at barriers to see how they break. I don't even want to get into the complexities of those. And a lot of what you measure there can be tested by chaining sub-components to a hydraulic ram and slowly tearing them apart to see what maximum force they can take.

The two systems produce roughly the same test results. In rare cases, there's an effect that the simple system misses that would be caught by the complex one. In my realm, thankfully, it would be difficult and pointless to design it to pass only the simple test. In many cases, it's not about honesty, it's about pragmatism. In this case, the fault lies not with the technology but with the intentional effort that the manufacturer went through to cheat the test.