After the 2014 ICCT report revealed that these light-duty passenger diesel vehicles were emitting too much NOx and US regulators confronted VW about the results, VW did some testing and proposed a voluntary software recall to recalibrate the various emissions control devices on the affected vehicles. From the California Air Resources Board's (CARB's) in-use compliance letter to Volkwagen AG, 2015-09-18, following that recall:
To have a more controlled evaluation of the high NOx observed over the
road, CARB developed a special dynamometer cycle which consisted of
driving the Phase 2 portion of the FTP repeatedly. This special cycle
revealed that VW's recall calibration did increase Diesel Exhaust
Fluid (DEF) dosing upon initial startup; however, dosing was not
sufficient to keep NOx emission levels from rising throughout the
cycle. This resulted in uncontrolled NOx emissions despite the SCR
reaching sufficient operating temperatures.
What's significant here is that CARB changed the test cycle for this additional testing. The EPA's standard test cycles are publicly available and consist of set patterns of acceleration and braking, each intended to simulate a real trip* characterized by a certain type of driving (urban highway, rural up- and downhill, and so on). Here's a visualization of Federal Test Procedure 75 (FTP-75):
The manufacturers typically know the parameters of the test cycles to which their vehicles will be subjected in a given market. They then design their vehicles to pass the test. It's the regulator's responsibility to make sure the testing provides a reasonably accurate picture of how the vehicle will perform during its lifetime. This transparency should benefit everyone involved by:
- giving the manufacturers a clear path to compliance and helping them to plan R&D and control costs;
- reassuring legislators and the public that the regulator is effectively enforcing the laws that protect public health and the environment;
- allowing the regulator to perform controlled testing on a limited number of vehicles and extrapolate that to the in-use fleet, rather than testing every vehicle (not economically feasible).
However, making the details of test cycles known also has the potential to allow manufacturers to circumvent testing in a way that gives them a business advantage without actually providing the mandated emissions reduction. This is sometimes called "cycle beating" and it's nothing new in the automotive industry.** What may be new here is how VW went about beating the cycle. Continued from the compliance letter, emphasis mine:
CARB shared its test results with VW on July 7, 2015. CARB also shared
its results with the EPA. Several technical meetings with VW followed
where VW disclosed that Gen1, Gen2 and the 2015 model-year improved
SCR vehicle (known as the Gen3) had a second calibration intended to
run only during certification testing. During a meeting on September
3, 2015, VW admitted to CARB and EPA staff that these vehicles were
designed and manufactured with a defeat device to bypass, defeat, or
render inoperative elements of the vehicles' emission control
That goes way beyond a failure of due diligence. You can imagine a situation where an automaker is "teaching to the test," i.e., designing their emissions system with a strict focus on what they have to do to pass testing, and just fails to realize or discover some flaw that will result in non-compliant levels of emissions on average. You might call that shoddy or incompetent work; you could even say the regulator's at fault there, for not having a better suite of testing.
In this case, VW designed the software that controls the vehicle's emissions system to cheat the test. It's very much like hiring someone to write an essay or sit an exam for you, while you take the credit; the vehicle system that EPA and CARB were testing was literally not the same system that consumers were purchasing and driving, making the test meaningless.
The International Business Times, referencing ICCT Senior Fellow John German as the source, reported earlier this week that:
It’s not yet known exactly what Volkswagen was measuring. German said
that it will probably be months before we find out what exactly the
company was doing. The EPA explained that the "device" can't be turned
off by the user, and Volkswagen will now be required to fix the issue
without the owner incurring charges.
While we don't know exactly how VW did it, it's clear that the software they wrote for these vehicles contained some sort of algorithm(s) to detect, based on data from the comprehensive suite of sensors included in modern vehicles' OBD systems and their knowledge of the parameters of US regulatory test cycles, when the vehicle was and was not being run through one of these test cycles. This Washington Post article is the best lay summary I've read so far.
Some reports suggest that all they did was detect whether the steering wheel was moving, because when hooked up in the lab there are no turns—but since the CARB letter suggests that a "special dynamometer cycle" was able to get the control unit to switch out of the special "test" mode, I expect the actual algorithm was a bit more sophisticated than that. The EPA's 2015-09-18 Notice of Violation to VW makes this specific allegation:
The "switch" senses whether the vehicle is being tested or not based
on various inputs including the position of the steering wheel,
vehicle speed, the duration of the engine's operation, and barometric
pressure. These inputs precisely track the parameters of the federal
test procedure used for emission testing for EPA certification
As for why the lab tests didn't detect the presence of this cheating program, they're just not designed to catch deliberate, sophisticated fraud. These programs are expensive, they're funded by taxpayer dollars and the regulations they enforce makes cars more expensive, meaning taxpayers who own vehicles get hit with costs from multiple directions. Auto manufacturers are large and have a lot to lose by cheating the system—the financial impact of this scandal on VW was already in the billions of dollars even before the news broke and sent their stock down 40% (tens of billions in market capitalization) and before any fines have been announced (also potentially in the tens of billions).
Costs are a huge, enormous, cannot-be-overstated issue in the world of environmental (and public health) regulation. Cost, not science, is nearly always what kills advancements in this area. So we trust that auto makers will make a reasonable effort to comply and we pay for testing that basically amounts to: Put the car on rollers; "drive" it through a test cycle with a bag around its tailpipe; see what's in the bag.
This sort of testing, as long as the test cycles are known in advance, is just not capable of detecting fraud within the control software. In fact, considering the level of sophistication possible in modern software algorithms, even hiding some or all of the details of the test cycle might not be sufficient to defeat this approach entirely, as long as the manufacturer is determined enough and the testing takes place in artificially controlled lab conditions.
What the ICCT group did was to use a PEMS*** to measure tailpipe emissions under real-world conditions, while driving on actual roads with actual traffic patterns that VW's software didn't recognize as corresponding to emissions testing. This allowed them to gather more accurate data about the real performance of the vehicle emissions systems—but at a cost of $50,000 for only three vehicles and with a very limited scope of what they were looking for.
* Research Note 96-11: Driving Patterns and Emissions: A New Testing Cycle (CARB 1996) provides a brief description of how a test cycle is developed.
** For example, see Cycle-Beating and the EU Test Cycle for Cars (Kågeson 1998).
*** Say it just like it sounds—"pems"—or call it a lab-in-a-box, but don't confuse it with PeMS, PEMS, PM CEMS...