Care to dispute it?

Surely an off switch is the cost effective way to overide all other systems.

Mind you there have been deaths associated with that too, people in the past turning their cars off to find they have no brakes on the second or third application after the power booster has returned to atmo pressure.

And no breakage or malfunction also means no information to go on. But the risk is still there. In fact there is ONLY risk until something actually happens, and the risk only changes as more information is obtained to identify the likelihood of a failure. Information is key here.


This is the contrary perspective of the same misunderstanding as above. When something fails, that information trumps all the statistics. IOW nobody ignores something new that is learned because the statistics tell him it's not a big deal.

And there you have it, the driver has absolutely no way to know whether his car has the problem or not, he can only rely on the statistics and with the (lack of) information he has, his risk is greater than 0. The statistics take precedence over concrete information because the concrete information doesn't yet exist.

Again you're comparing an informed perspective to an ignorant statistical one. But the statistics don't ignore information, they change to accept new information as it is learned. Eventually enough may be known about the incidence of the failure to be able to pinpoint exactly which cars will and won't fail, in which case the statistics will indicate 0 or 100% probability. Of course at that point nobody looks at it as a statistical concern any more.

Nope, generally parts fail because they have a defect. I suppose that with failure rates for some components so low that they amount to system noise, it could be perceived as a "random distribution," but that's due only to a lack of understanding of the specific failure. For example, in examining warranty repair frequency rates, clear patterns invariably emerge for specific models: alternator, wiper motor, mirror adjuster, brake squeal. On another model with no parts commonality in those areas, totally different components and rates. Obviously these are not random distributions.

Field service engineers see this all the time. Visit one dealer and inspect the parts in the scrap room; visit another dealer in the next state and see the same parts. Groundhog day. With a nice match-up of the nature of the defect to the operating conditions (temperature is a good one) it's not unheard of for failure rates to approximate 100 percent. The variability is in the conditions; mass-produced parts are consistent at least. When parts fail in something approaching recognizable frequency, it is in effect because they were designed and built to fail. It just wasn't appreciated in-process.

BTW, this is why you want a TV in the service waiting room. Let them talk sports or politics. When the subject comes around to cars, folks discover shared experiences.

Of course. Here is the first thing I could find on Google that shows what I know to be true. http://www-fars.nhtsa.dot.gov/Main/index.aspx

Take a look at the trend of US fatalities per 100 million vehicle-miles from 1994 to 2008. It goes from 1.73 to 1.27. In fact, if someone can find a graph with a greater range of years, you'd see pretty much continuous decline. Maybe the advances in car technology over the years have not been such an unmitigated disaster for our safety after all.

Okay, you seem to labor under a strawman of a definition of randomness. When I'm asking if the part failure has a randomness associated with it, I'm not asking whether the engine is as likely to fail as a radio knob. Believe it or not, most people who work with statistics are able to realize that cars have different parts, and that they might fail with different frequencies.

I'm asking an entirely different question. Let's say you have a bunch of parts, all of them coming from the same batch. Those parts are known to occasionally fail. Is there a random process involved in determining which parts from that batch would fail?

Not so much. If they were all exposed to the same loadings, then they'd all fail as a cluster. But you have three layers or more of noise - usage,environment, the other components in the vehicle.

That's not very logical. As I stated, frequency rates vary not just in different components on the same models, but in the same component on different models.



No. A randomized system can estimate how many parts in a batch may fail, but it can't identify which ones.

You are vaguely suggesting a process not completely unlike olden times. Pick an acceptable minimum success rate in production and run the part. Trouble is a car has (depending how you count) 20,000 components. If each part has a recognizable fail rate, by the time the car reaches the end of the assembly line it has dozens of defects built into it on purpose. And no joke: at the end of the line, some percentage of cars will not even run or move or will have some hideous deformity. So there was a repair line next to the end of the main line so brand new cars could be pushed away and rebuilt.

Needless to say that doesn't cut it anymore. Thus the move to zero defect manufacturing and 100 percent testing down to the nuts and bolts. This not only improves quality, it increases production and efficiency, because bad parts slow or stop the manufacturing process at every step. Or consider costs: If x percent of the parts shipped from plant A to plant B are defective, x percent of the transportation cost is lost and these wastes multiply throughout the system. There are still repair lines but they do totally different work. They are fixing paint flaws and trim alignments, not replacing engines or trying to find a door that will fit.

As a non engineer my knowledge of failure mode analysis is limited to what I have read and engineers have told me , often long ago.

I seem to recall that Weibull analysis is meant to characterise failure modes into infant mortality, wear out , random and terminal ( i.e sudden end failure. I think it was used to characterise failures after a lot of testing with the aim of identifying those that could be possibly prevented e.g PWB burn in for infant mortality, reduced by spending money eg wear out and the rest where you then took a judgement on acceptabilty. The latter would lead to unscheduled failure which was nuisoance if ti was say production machinery but rather dangerous if it was an aircaft part ( or say a steering linkage).

Is this still used as a tool,or have better things ovetaken it?

Seems to me it happens all the time and is happening here: according to some posters, any fear of SUA with these vehicles is irrational as the statistical probability is too low. Which is essentially rubbish. For some owners the risk is nonexistent while for others the risk is far greater than the overall frequency suggests. Which are which? Unknown. Ask Again Later, says one sophisticated modeling tool I keep close to hand.

Returning to the tedious air travel parallel... when pilots are caught drinking on the job, the airline's PR rapid response team will invariably haul out the statistics showing that commercial air travel is safer than napping in a hammock, operating a salad fork, etc. Great. Glad to hear it. Now please run the numbers for aircraft flown by pilots who were fall-down drunk. The selection group for this air safety issue is excessively broad; i.e., it also includes sober pilots and is skewing the hell out of your results. I will stick to my original theory that drunken pilots are in general bad for air safety.

Beyond that, you are not so much disputing as restating the issue in your own terms. Meh. I would just as soon argue about English literature.

Ok, here we go, now we're onto something. If you can estimate, but can't identify, you're talking probabilities.

I'm actually not suggesting anything at all about the manufacturing standards, so save the impassioned speech. I'm not saying it's a good or a bad thing that cars fail, or whether it's cost-effective for some of them to come out as junk from the assembly line. I'm just concerned with probability of failure of affected Toyotas, given that failures exist. That's how this all started.

What I am saying is that should parts fail, there is an essentially random process associated with them failing. Since cars fail when parts fail, then the failure of cars is a function of the failure of the parts, and thus there is likewise an essentially random process associated with cars failing. Thus, if I'm a perfectly rational Toyota owner, and I'm evaluating the likelihood of my Toyota flying off the highway at 120 mph, I'm justified in applying a probability theory in making my decision about continuing to drive the car.

Oh sure, it's a staple of statistical reliability modeling, many of which tools are based on Wiebull distributions and adaptations and refinements thereof. Variations on a theme. Like all predictive models, is most effectively used and abused on large, simple, mechanistic systems: IC chip component, ball bearings, ant colonies. When your departure gate is changed twice in 20 minutes, that is Ernst Waloddi Weibull smiling down upon you.

I swear I am not making this up: In the early days of the Ford Motor Co when the company was growing at an exponential rate, a running estimate of accounts payable was kept by placing the invoices in stacks and weighing them at regular intervals. Say, every Thursday. This is not quite as absurd as it sounds...okay, it is. But anyway, if you presume that various suppliers tend to use the same sort of raw invoice form and each tends to submit them on some sort of regular calendar basis, the system almost seems workable. But needless to say, the methodology was no threat to IBM and is nowhere close as accurate as simply adding the sums. It worked mainly because FoMoCo operated on a large cash surplus. Henry Ford hated banks (although his son Edsel owned one) as he distrusted their pencil-pushing ways -- apparently he preferred a grain scale. One of the neatest places in the Detroit area is the old EEE building/former Ford administration building on Oakwood Boulevard, which was recently closed -- it was last occupied by powertrain engineering. Henry had his office there -- and just down and across the hall, his safe.

So, basically, what you're saying is that there is uncertainty in statistics. Insightful observation.

And, actually, "ask again later" is your approach. Statistics is all about giving the best estimate of the true answer given the information available, knowing that the "true" and possible unknowable answer may diverge from the estimate. Your approach is, "I don't know which one is a lemon and which one isn't, I don't know what to do", which is what you're spoofing.

Again, you're knocking down a straw man that you yourself built. Nobody is saying that Toyotas are safe to drive because cars in general crash rarely. That's what the car analogy of your air travel analogy is.

We're saying that the fear of SUA is irrational because the affected Toyota models themselves experience them exceedingly rarely, all things considered. A defect that kills 10 people is really not that exceedingly dangerous for an activity that kills tens of thousands. In airline travel terms, if the statistically significant safety record shows that planes piloted by fall-down drunk pilots crash 1.05 times as often as planes piloted by sober drivers, then the airline would be justified in using statistics to claim that safety was not excessively compromised by the drunk pilots.

That's what I was getting it. Noise may technically be deterministic, but since it's a result of processes that are impossible to perfectly measure, it is practically a random process.

That's where I was also going with the shuffled deck example. Shuffling the deck, rolling the dice, flipping the coin, none of the outcomes of that are truly random. However, when properly done, the outcome is the result of a process that is extremely sensitive to initial conditions that cannot be perfectly measured, so all we can do is treat it as random.

I already stated my position on the matter: in its current condition the rough statistical data bruited abroad is useless either way in allowing the consumer to make an accurate assessment. Aside from supplying him/her with a vague awareness that some number of cars are capable of some failure, it says nothing at all. You say the stats indicate a low possibility approaching impossibility. I say the that depends entirely which owners you are talking to. So once again my advice to all owners is: do not be worried but mindful, and most importantly, know how to handle an SUA when it occurs. Sound advice for anyone no matter what make or model of car they drive, one would think.

You could look at it this way: When the flight attendants go to the exit rows, they don't lecture the passengers at length on the remote statistical possibility of an air catastrophe on this particular flight. They make sure the passengers know how to use the exit doors. This seems entirely sensible to me.

It depends on the cost. Giving instructions is practically free, so for that price any reduction in risk is worth it.

Where is the 'reduction in risk' in lecturing the passengers on the chances of a catastrophe on that very flight. Might be a few bruises in the rush to get off the plane...

Electric power steering on Toyota Corolla:

http://www.autonews.com/apps/pbcs.dll/arti.../100209863/1290

I was talking about reduction in risk from emergency exit instructions.

So we have been arguing about whether consumers can make drastic decisions (like whether to drive the Toyota parked in their driveway) based on statistical data?

Duh - I'm sorry I bought into this argument now.

I'm not sure that was the point.

Lecturing passengers on how to operate emergency exits doesn't reduce the risk of an aircraft mishap. The instructions are provided in order to (hopefully) help the passengers know what to do in the event that the risk of mishap is realized.

It sounds like you are referring to Environmental Stress Screening of electronics. The purpose of which is to precipitate latent manufacturing defects in electronics and electronic assemblies through the use of applied random vibration and thermal stresses.

I deal with this on a fairly regular basis.

"Normal" ESS profiles can take many hours or days to complete, depending on the size of the part tested. The profiles were developed based on years of research starting in the Apollo moon rocket days and culminating in a plethora of reports and standards developed and "perfected" in the late 80s early 90s. In the arena of vibration testing this led to the famous "Willoughby Curve" for random vibe testing. In an effort to reduce these testing cycle times electronics manufacturers have been following two paths. Throwing it out altogether, or trying to find harsher tests to make the infant mortality happen faster. In commercial electronics the approach of throwing ESS out isn't necessarily a bad one in that the obsolescence cycle is about 2 years, additionally consumer electronics are relatively cheap and buying a replacement widget won't break the bank. Harsher testing led to the development of Highly Accelerated Life Testing (HALT) and Highly Accelerated Stress Screening (HASS) by Gregg Hobbs.

On the non-commercial electronics side there are a multitude of possibilities. Some companies try to use smoke and mirrors to make it look like they've done a real ESS screen when in fact they've done a halfassed test. Some guys actually do ESS. And some guys who are looking for better, faster, cheaper, have adopted HALT and HASS as a way to "do ESS more better."

If you travel in the academic circles, there are a bunch of guys who will argue that "HASS isn't really HASS" and will debate the nature and basis of all screening methods. I haven't spent much time with those guys to understand their arguments.

Risk of injury, not risk of accident. I thought I didn't need to point it out.

ESS is actual, physical testing. A Weibull distribution is a tool in statistical analysis.

Um Weibull isn't even that. it is a way of fitting observed distributions to mathematical ones. Useful? yes. Predictive? Hmm.

It is very handy if you have tested 10 components to failure. Or more likely 7 fail and 3 don't by the time you give up.

Correct.

What I didn't get into is that Weibull methods are used as part of determining HALT and HASS profiles.

Toyota have hired Exponent, an engineering and science consultantancy firm to try find out if the electronic control can cause SUA. Their inital report says no, they have not been able to induce SUA. In all cases, the engine have either behaved like normal, shut down or shifted to fail safe mode with reduced power output.



Mercedes have been using a brake by wire system called Sensotronic without a direct hydraulic link between the pedal and the brakes (although it have a hydraulic back up). But they dropped the system after several problems, with exception of some small volume models like the Mercedes McLaren. The problems were probably fixed though, the system was said to be no less reliable than conventional brake systems but bad PR and high costs eventually ended its use.

Normal ABS/ESP systems can only control braking though the normal hydraulic system. Cut the power to the system and the eight servovalves will return to closed (four of them) and open postion (four of them), and the system will operate mechanically.



It isn't exactly surprising that ESP systems lower insurance costs, they can after all almost eliminate some common types of accidents and contribute to a much increased road safety (unless the drivers compensate by driving faster - which indeed many do). Cars and their system are also the cause of very few accidents, single digit percentage numbers usually.

What failed Mercedes demonstration? I specifically recall one failed Mercedes demonstration, but that wasnät related to brake by wire, but their radar based adaptive cruise control that also could brake the car to a complete stop if something blocks the road. Turns out you have to remember to turn the system on if it's going to work!



It doesn't have to be that complex. A torque limitation when the brake pedal is pressed can be enough. Sure, the engine will still fight the brakes, but the torque limitation can be chosen so that the brakes can slow down the car anyway. Systems like that have been in production for over a decade at least. There are also electronic throttles with mechanical throttle wires as backup, although I suspect that no car manufacturer no longer have such a system in production.

Some aircraft also have a mechanical back up system for the engine, with a wire from the throttle lever to the fuel system if there should be a problem with the electronic control. But today I would think most would prefer to use two separate FADEC units instead.

Aircraft engines also tend to use alternating current for many sensors, that way most interference can be filtered out since the system only look at certain frequencies.

A car is a system alright, but it is a system which includes such wildly unpredictable variables as drivers and passengers, which are much harder to program.

I don't mind electronic systems, but I distrust the all-in-one-basket systems whereby if there's a fault in the computer it can affect everything. I'm no belt-and-braces man, but my ISP, landline phone and mobile phone are all through different suppliers so if I every fall into dispute with one and they cut me off, I've still got the other two fully independent. Hasn't happened. But it pays to have independent backup.

The automatic breaking etc stuff is fine, but can you imagine what it would be like in Bangkok rush hour with cars coming at you from everywhere, erratic bicycles, pedestrians dashing through, not to mention and thousands of Honda 100 bikes? You'd never be able to move an inch. No doubt they'll get there one day, but is a way off I suspect.

What's the alternative to braking, running over them? Besides, isn't automatic braking already out on the market, as part of advanced cruise control systems?

My worry is that the systems can't cope with these traffic problems as well as the driver can.

No computer programmer or system can programme for every possible situation and combination of same, and there is a point wherre all you are doing is transferring control of the car from the driver, who is there, to the programmers and engineers, who are not.

The alternative isn't "running over them". That's a straw man argument.

It wasn't a straw man argument, or any other kind of argument for that matter. I just fail to see the problem you're describing. If someone darts out in front of you, I fail to see an option other than braking, so I don't see how a computer would make a suboptimal decision. Either doing nothing or swerving seems like a reckless thing to do.

Surely the point is that Terry was visualising a computerised system in Bangkok rush hour. My guess is that it would be overloaded, dither and stall, where a human driver would cope.

Thanks, Tony, that's what I was trying to convey.

In frantic traffic there's just so much going on that the sensors can't detect, and can't deal with. Too much input which the driver can see and the sensors cannot, where the computer wants to do one thing and the driver must of necessity NOT do it; or vice versa.

We westerners driving on our wide modern roads in formal traffic patterns which can be programmed for is one thing; but real traffic in much of the world is chaotic, not just to look at but in the mathematical sense too.

That would depend if the person who darted in front of you intends to keep going or not. Humans can detect intent and computers can't. Sure we get it wrong some times and the result can be deadly but at the same time our traffic flow highly is dependent on these constant judgement calls. Even more so in developing countries. Electronic brake activation culd bring traffic to completely safe standstill.

Automatic brake systems are already in production, Volvo offer such a system and to my knowledge there is nothing that prevents it from being used in Bangkok rush hour traffic. This kind of system tracks other cars and pedestrians (so far the system can't handle wild animals) using systems like radar and infrared cameras and offer warnings to the driver. If the system detects that an unavoidable collision is about to happen it can brake to avoid the collision or reduce the speed at impact.

Radar based system also can 'see' in fog, mostly aimed to look after other vehicles and larger objects, while infrared cameras are good at identifying humans.

But there are also simpler systems which can identify an emergency brake situation and then make sure the maximum brake performance is used during the deceleration. Many drivers unfortunatly don't use their brakes fully in such a situation and since a 10% decrease in speed before an impact decrease the risk for fatality by 40% that can have a large effect on road safety.



Well, humans are not so good at it either I'm afraid. In about half of all accidents involving a motorist hitting a pedestrian the motorist saw the pedestrian but didn't stop. It's in this kind of situation an automatic brake system can make a difference. The expected result is that four out of ten pedestrians can be saved, and 15% of the collisions can be avoided.

Err...

Well, I wrote 'is about to happen'.

If the system detects that say, full braking is required to avoid a collision it automatically starts to brake just before the collision becomes unavoidable (where the braking distance is longer than the distance to impact).

So long as I can turn all that shit off. But as with all new software products, manufacturers never accept that idea at first.

Down the track you won't be able to. If you crash and kill someone and the systems are switched off you'll be going to have more sex than you ever dreamed of. With big sweaty men.

Current Bosch recommendation is that the defeat switch for ABS TC and ESC is more of a hint to the system than a true on off button. It effecively watches what you are doing and decides that you really are screwing up then the sytems reactivate.

Absolutely, the overall rate is declining...

And there's no doubt that cars are getting safer. Mandatory belt wearing is playing a major part in all of this... do you have random breath testing there too?

In Australia, the mandating of the belt-wearing laws peeled about 25% off the death numbers in the first year (and it continued to drop) despite the fact that fewer than 50% of people were wearing belts properly. Then, ten years later, random breath testing came in across the country and saw a further 33% decline in deaths. And people are still caught drinking and driving.

Throughout it all, however, the governments have pushed a 'speed kills' line and heavily policed speed limits, increasing dramatically the areas where speed limits apply and reducing speed limits as well. So the public of today think mainly in terms of speed as a killer (yes, they also recognise drink driving) and don't recognise the importance of learning to drive properly. That, of course, is only needed if you're going to drive fast, isn't it?

Driving standards have dropped, have no fear. But it's masked by improved cars and further safety built into them.

Exactly, for the effort and expense that the Government puts in the death toll should be around 10% of what it was - but it isnt. We both know what it was like driving around in the 60's and 70's.

Notice a good read of those stats reveal a lot more than the surface quote above and motorcycle deaths, that mostly result from poor standards of car driving, has dramatically increased as there has been no major advances in safety for a bike parallel to the car.

It also doenst list amounts of actual accidents and serious injuries (injuries that would have been deaths if not for belts, bags, safety dashes and crumple zones back then).

You don't think that would be a liability too? Say you don't want to brake for a kangaroo because he's just on the other side of the tracks and a train is coming. System brakes for you, 'roo hops away safely and you are left with a few less members of your family. then a few years later the CEO of Bosch gets lots of sex.

Comes back to the net benefit to society argument. if they can demonstrate that on average they save more lives than they kill and it is cost effective then the system goes in. eg airbags, sure once in a blue moon they go off and cause more harm than good, but nobody is dragged off to jail. That is EXACTLY the argument the FAA uses for airliner safety improvements.

Hang on we are now differentiating between unknown accident situations and a real scenario that could/does/known to arise. So you go across a cross road and some dick decides to pull out of a parking spot or a pedestrian walks across on the other side and your car stops not completing the intersection and you get sided. Imagine the result of someone being killed and they knew the scenario could happen and it was on a white paper.

Thats is a real and known situation that can be avoided (could be as simple as a horn blast stopping the other idiots further moving into your path) or at least you have a choice to go for a gap or hit the car or person if you see the impending danger bearing on you from the side.

As I see it you are allowing the idiots to still get into their stupid situations but not allowing them to have a chance to get out of them in some cases - making the driver even less responsible again.

I don't think the FAA counts because thats generally a one chance scenario, what circumstance is there that some die and some don't in airplane crashes?

i6; wouldn't happen, the Kangaroo would have already jumped straight in front of the train ('Roo's are the stupidest animals on the planet) or the wife would have already splattered it with the .303 hanging out the window from 400 meters (while breast feeding the twins of course, we build 'em tuff here).

Keep em coming Cheapie.

Who was it originally pushing for airbags, anyone remember?

There was a bit of a exposé in Road & Track at the time. Insurance companies were backing the installation of airbags. Why?

That was the big part of the story. So there would be fewer deaths to pay out on? Not on your nellie!

They explained that basically payouts and premiums always balanced each other (all calculated by actuaries, of course). The profit of insurance companies coming from the invetment of the premium before it has to become part of a payout.

Airbags added a significant slice to the price of the car, hence they added to the premiums. Hence they added to insurance company profits in a much bigger way than belts did.

Interesting theory. Idiotic and nonsensical, but interesting.

Well, to take the other tack

We shouldn't add cost to a car by introducing safety systems because some people don't need them in some circumstances.


Any takers?

I never saw any retraction printed by R&T about this article, nor any letters to the editor taking them to task over it. So I've always assumed it to be somewhat factual...

If, however, you have reason to believe otherwise, can you please explain why you do?