Comment by thrwyoilarticle | Hacker News Reader

Comment by thrwyoilarticle | Hacker News Reader

thrwyoilarticle

I've never been surprised about Dieselgate. If an engineer is given a metric to beat, they'll beat the metric, not the spirit of the metric. The same attitude of improving the test still exists, it's why manual sports cars have tall gears - the test will happen at a more efficient rev range than the users will actually use. Variable valve timings, variable cam profiles, more aggressive sports modes: these are just hiding the real emissions at a rev range the emissions test won't reach. And wasn't it accepted wisdom that you shouldn't believe the MPG or L/100km figures?

So-called 'tech' is just as bad. A defeat device is so similar to the everyday silicon valley attitude it's hard to believe it didn't originate there. Move fast and break things? Building companies based on massive amounts of user data, then complaining you can't feasibly do manual content review? We can't build a society based on the assumption that engineers will act in the public's best interest.

The challenge is the test is flawed if it stipulates to narrow/limited of a test range. This is why in SW we do automated tests, unit tests, but also real-user testing.

For emissions the only solution is in-situ testing over real-world operating conditions over a period of time. In other words real-user-testing of the vehicle as it will be used.

This would stop all such games immediately.

I would be very interested to see which brands / models of cars would see significant MPG reporting changes.

How much would this exacerbate / accelerate the challenges car companies face in complying with governmental MPG requirements?

for me it was also hard to grasp what went wrong here. i always thought if this test can be gamed something is flawed with this test. they were given the task to pass the test and they did although that did not translate to the outcome the engineers developing this test envisioned for it i guess. It feels like they changed the requirements and then went after anyone not passing anymore. i do understand that they purposefully tried to detect the test to game the result "somehow" but i cant help it but to think if its possible to reduce emissions for the test then its a viable strategy on the road given the same usage pattern... although i don't really know how they actually did it...

so the problem is that vw purposefully lowered engine performance during the test. they could have done that on the road, and been within spec, but then their cars would have felt sluggish, so instead they just made their cars lie during the epa test

VW detected a test cycle and turned emissions treatment, Adblue and some other things, on. When on the road, these systems were tuned down or fully turned off.

From my limited understanding it wasn't just lowering engine performance they couldn't be simultaneously be efficient with fuel and low in emissions (IIRC burning at a higher temperature is good for fuel economy but bad for other pollutants). It's very well possible that no single configuration could have made them meet the requirements.

If memory serves well, they used steering, rev and so on as input to decide if the car was on a test bench. Then they turned on emissions treatment, or rather turned it up. Obviously that works on the road as well.

As a tangent, VW was part of group of German manufacturers colluding in limiting Adblue tank capacity (I'd have too look up details). in order to maximize revenue, they also extended the refill periods to match inspection intervals. Basically so the OEMs could sell dirt chepa Adblue to customers. Now the smaller tanks were insufficient to treat emissions on the road without being refilled between inspections. Instead of mounting bigger tanks, VW decided, apparently, to just turn the system of while on the road. Clearly illegal because test cycle recognition is illegal. As opposed to using technical loop holes when road operations are concerned.

It was sort of the opposite - there was a control module called "kundenspezifische Akustikbedingung" or "customer acoustics condition" in Bosch diesel control units which was _deactivated_ by conditions like steering input that indicated real driving. So the car started in "low emissions" mode by default (essentially, running richer, which increased consumption and CO2 emissions but reduced NOx), and then leaned out (decreasing consumption but increasing NOx) once the "acoustic model" was deactivated.

https://cseweb.ucsd.edu/~klevchen/diesel-sp17.pdf

The acoustic mode was used to recognize the test cycle and to properly dose Adblue. It didn't impact fuel mixture, that was something VW later did to make the engine generally compliant. The new, leaner mixture resulted in some power losses.

Under all other operating modes, the Adblue dosage was reduced to almost zero. That is clearly illegal. In addition to that, VW did some other stuff especially for the US market to meet California's emission requirements.

What everyone else did was playing with temperature windows, under all operating conditions to reduce Adblue and optimizing for other things. Not ullegla per-se, as these windows kinda used to be legal, but definitely violating the spirit of regulations. Some manufacturers exaggerated more than others. But only VW cheated as clearly, as bluntly, as long and at as large a scale. No amount of VW PR can change that.

With the help from Bosch:

"Supplier’s Role Shows Breadth of VW’s Deceit" https://www.nytimes.com/2017/02/01/business/bosch-vw-diesel-...

Yep. Bosch tried to protect their asses by sending VW a letter that this software was test purposes only, any road or production use would be illegal. They still got fined.

"In June 2008, Bosch wrote a letter demanding that Volkswagen agree to pay any penalties if they were discovered using a defeat device."

The use of that software as a defeat device surely was a surprise to everyone when it came out years later...

The EDC17 analysis in this paper does not support this: “ Figure 4 shows how the fuel injection quantity (additive) correction (qCor) is modified by the acoustic condition.”

Not that it makes the cheating any less illegal, but I’m interested in the source of your analysis and why it differs from the model in the FR and the disassembly performed by that research group.

Then my source was too old, they just talked about Adblue dosage. What you say does make sense so, since after the software update cars in Germany had less power then before. My take is that Adblue dosage was the main issue so.

Thanks! Nice paper, I didn't know they had access to the SW specifications

You can find these (leaked of course) easily by googling “Funktionsrahmen”. They are shared widely by tuners.

Thats unbelievable. So they purposefully fucked with the car design to sell some juice? Beancounters beware. I wonder who it was that pitched such an unconscionable asinine idea. It seems like there were some missing NO men as well.

If I remember well, the main offenders were German OEMs that colluded with VW on that. other European OEMs mainly toyed around with other stuff and edge cases, some activities were later found to be illegal. Most so were able to fix it by "just" burning more Adblue and getting rid of temp windows and such things.

VW was the only one I know that was forced to tweak engine performance. IMHO the goal was optimize weight and cost, and engine design. Only VW seemed to have gone far enough to design engines that were technically barely, if at all, able to comply with emission requirements. So they resorted to cheating. Once that worked, it escalated to the point 11 million cars were affected. Kind a stupid if you ask me.

Is it the engineers fault? Or is this the result of antisocial business decisions?

If you’re not going to trust engineers, who are you going to place your trust in? Business men? Lawyers? Bureaucrats? Politicians?

If we’re looking at engineers as multi variable function optimizers, You’ll find that the field will be automated away in short order. And that will be one less group to trust.

Fact is, at any level, everyone has an assumption of trust for society. There are of course consequences to violating that trust. But then smart people find ways to insulate themselves from this consequences.

Regardless, the problems you speak of exist regardless of what professions exist. The real problem is how do you align incentives to align for better outcomes, and that is a hard problem to solve.

You make it sound like engineers are failing their moral duty and then ignore the people in charge of them who actually make the important decisions in any company. I can almost guarantee it wasn't some lowly engineers decision to be a bad faith participant in the testing; they build what they're told to. Also, comparing car ECU engineers to software engineers generally is a stretch, I don't think there's much of a common culture there.

I think older Aston Martins had a tall first gear precisely to give them a good 0-60 figure.

Reply on news.ycombinator.com