Question about "black boxes" on motor vehicles

How do the “black boxes” on motor vehicles work?

Question was triggered by the horrible bus - semi crash here in Saskatchewan back in April, when a youth hockey team was wiped out by a crash.

Media reports are that there are computer recorders from both the bus and the semi and they’ve been taken to a specialty lab in California for decoding, in hopes of shedding light on the accident.

How do those work? do they record for a lengthy time backwards (half an hour or more), or it is it just a minute or two? Were they installed in vehicles for the specific purpose of crash reconstruction, or do they have some other purpose which has the incidental effect of keeping records that help with crash reconstruction in some cases? are they in durable cases, like on airplanes? what type of skills do you need to decode them?

I don’t know if they put anything extra into commercial vehicles like buses, but a typical engine computer from any modern car stores a lot of data, just because it uses that data to optimize its performance. In addition to keeping track of things like the air pressure (since this determines the optimal mixture of air and gasoline), it also tracks how you drive, like what speeds you are going and how hard you accelerate and brake.

Even if the device isn’t designed to be a record-keeping device like an airplane black box, there is still an awful lot of data. They will be able to tell how fast each vehicle was moving, whether the driver was accelerating or braking (and how hard for each), what gear they were in, etc. They’ll basically be able to reconstruct exactly what each vehicle was doing just from the data that the engine computer uses to make itself efficient.

Apparently the trucking companies do, to gather various sets of information about safety and performance beyond just those registered by the control module. The persistence of data would be longer than for an airplane since it may not be able to be downloaded live while the truck’s on a long haul run.

Cue question as to whether the new standard announced in Carpenter applies to this data…

IIRC if the restraints control module (RCM) decides one or more airbags need to be deployed, modules retain a “snapshot” of the state of various systems. e.g. at the time airbag deployment was commanded It records the throttle position, whether the driver was braking, any ABS/stability control data such as steering angle, and so on. I don’t know how much of a time window it captures, but it’s not much. But unlike airliners there’s no dedicated “black box” module in the car.

It varies depending on the year and make of the vehicle, and what country your in. I can only speak as to the US. I am involved in accident reconstruction, and work with the things from time to time so know a bit about them.

The ‘black box’, or airbag control module (ACM), is continually monitoring several data streams from the vehicle. After 2012, this is generally vehicle speed, % throttle, engine RPM, brake on/off, ABS on/off, driver/front passenger seatbelts on/off. Some makes may also report the amount of steering input, which is nice as it tells us if the driver tried to avoid the collision.

The ACM is located as close as possible to the center of mass of the vehicle. Usually it’s in the hump between the front seats around the cupholders. It looks like a silver box the size of two packs of cigarettes. They’re not indestructible – we’ve had them destroyed in collisions by impact and also fire. They may also not record data if power is cut during the collision sequence before the data can be recorded. In general, though, we can pull data from them.

In newer cars there are two accelerometers in the vehicle - one for lateral g’s, one for longitudinal g’s. If there’s a deceleration event sufficient to ‘wake up’ the ACM, it will lock in the last five seconds of data, usually in 0.5 second increments, along with the change in speed (delta V) for the lateral and longitudinal direction. A Ford ACM can get us data in 0.1 second increments, along with lat/long g’s on those same increments. Fords can also give us data sets up to 25 seconds prior to the collision through the power control module. We also get lat/long g data during the collision sequence (about 0.15 seconds) in a nice graph and table. If there’s more than one event (car hits car, then another car), we may also get both impacts recorded.

Interestingly, since Tesla uses a front-facing camera for auto-navigation, it records still images every second or so. That’s not part of its ACM data, but a separate telematics log that’s sent to Tesla. We can get that data with a search warrant and see the conditions leading to the collision.

The ACM won’t tell us when the collision happened, but it does record ignition cycles. So if we download the ACM and it reports a collision at cycle 2145, and the download at cycle 2147, we know the deployment is relevant to the crash under investigation. If the collision is at cycle 1932 and the download is at cycle 2147, then it’s likely from an earlier, separate collision event and the car was put back on the road.

IIRC, ACMs began as just the gizmo to deploy the airbag. Over time, in a quest to understand more about what happens in a collision, more data streams were recorded. However, what was recorded and for how long varied by manufacturer. In 2012 congress passed a law mandating certain items to be recorded and for how long, and so after that year what we get when we download an ACM is pretty consistent. Before that, there is a lot of variability in what is recorded, or if we can even download data from the unit. Europe has different laws, but European vehicles sold in the US must comply with the 2012 ACM requirements (may also be for all of North America – can’t recall).

I have no experience downloading from trucks (missed the class this month, dang it, and we’ve just not had those collisions in my office yet), but I’d expect much of the data to be the same.

Downloading the data requires the BOSCH CDR tool, a special gizmo you plug into the vehicle DLC port under the dash and a laptop which decodes the ACM data and prints it in a special report. If you can’t access through the DLC port due to crash damage, you need special cables to plug into the ACM directly. We have a rolling suitcase crammed with different cables because manufacturers can’t seem to decide on a standardized cable, even within their own damn company.

Thanks for that detailed info, Patch! Yet again, the breadth of expertise on the Dope is astounding. Of course we would have an expert in electronic accident reconstruction here…

I would assume that Canadian vehicles meet the same 2012 standards as US ones, given the integration of the auto-manufacturing industry in North America.

I believe Canada does follow the same guidance for what goes in the ACMs (I would have to look it up, but I believe it was part of a North America treaty thing because so many cars are made outside the US). You need to follow it to sell cars in the US, so manufacturers make their ACMs meet the standards. It’s simple enough to do now, and it benefits everyone.

Except the drunk driving 90 mph who kills two people and claims he was going 40 mph. He’s toast. :slight_smile:

I think that Treaty would be NAFTA. That’s why the auto industry is worried by Trump’s talk about tariffs on cars and breaking up NAFTA. NAFTA, and before that the Canada-US Autopack which provided much of the background framework for NAFTA, have created a fully integrated N American auto industry, with common standards and integrated supply chains. They’re worried they could lose, bigly, by tariffs that break up that unity.

Commenting beyond that would take us to GD. :wink:

My Isuzu dealer offers a free driving performance check at service time, a percentage breakdown of speed ,revs, braking and idling time, there is clearly quite a lot of data available.

Patch ftw. I can add that the intent of the “black box” was not originally meant to assist with accident investigation so much as to log conditions immediately before an airbag deployment. In the days before side airbags, this would mean T-bone collisions without a deployment wouldn’t have useful data in the device. Also, (and this may have changed since my training) the box retains three “cycles” of data. Each cycle begins and ends with starting/attempting to start the car. This is a somewhat vexatious quality for investigators because the relevant data can be cleared during the normal course of accident cleanup:

  • Accident is recorded in cycle 1
  • Engine is dead, occupant may attempt to start the car in order to clear the roadway; accident data is moved to cycle 2
  • Tow truck driver may attempt to start the car, or just turn it on in order to get the mileage off a digital odometer; accident data is moved to cycle 3
  • Someone somewhere else in the custody chain turns on the car; accident data is cleared completely

Thanks for that, Inigo. I was wondering how it would work if the vehicle got turned on and off.

That may have been an issue with older ACMs, but on newer vehicles (2012 and on) the data is retained in the form of a “deployment event” or a “non-deployment event” and it isn’t as easy to clear.

In a deployment event, the ACM kicks off the airbag and stores the data. That data is locked, and cannot be cleared or overwritten. I have seen vehicles with a deployment event, and hundreds of key cycles later, there’s another deployment event. The vehicle was in a collision, the airbags deployed, and they never replaced the ACM when they repaired the vehicle. Generally, an ACM will only retain one or two deployment events.

In a non-deployment event, the ACM received enough of a ‘kick’ to wake it up, but it’s insufficient to deploy a bag. However, it’s severe enough that the ACM records the event. A non-deployment event can be overwritten by a deployment event, or even a non-deployment event, but until it’s overwritten by another event it sticks around. Again, how many events can be stored varies by ACM.

How many deployment/non-deployment events get stored varies by ACM. I’ve seen up to 4 in a continuing event where a vehicle takes multiple whacks. The data in those cases will overlap, and you may start getting wonky data due to power spikes/failures, airborne tires, occupants slamming into pedals, etc.

They key thing to note is you’re not erasing data by key cycles anymore. I know of one case where a driver claimed his ACM showed he couldn’t have caused a collision, only the ACM data revealed the event happened thousands of key cycles prior to the download. The stuff lasts.

Yup, we’ve had that issue as well. Early ACMs only had a longitudinal delta-V, and didn’t even give us a speed. It’s made worse by the fact that unless a vehicle is struck directly head-on, the longitudinal delta-V won’t tell the whole story as some delta-V will be in a lateral direction and that wasn’t recorded. Then we get to break out the calculators and do some reconstruction work from scene evidence and crush damage to try and figure out how fast the vehicle was going. With luck, we can nab a speed off a nearby security camera and skip some of that.

I don’t know anything about these devices but, as a programmer by trade, I’ll suggest that they are likely centered around a magnetic drive and will probably use something called a “circular buffer”. Basically, you log your data into the hard drive, just starting at the beginning and continuing on until you’ve filled the whole thing up. At that point, you loop and start overwriting the oldest data in the system.

Separately from the circle buffer you’ll have something that keeps track of the position to write at as new entries come in. This might be stored as a separate field on the drive, outside of the buffer, or they might have special hardware that maintains it separately since it will be updating so often. (Likely they use the same drive but have a separate write head on the drive?)

When you want to read from the thing, you just start from the write position and either go forward or backwards in the drive depending on whether you want things in chronological or reverse chronological order.

Likely, each entry will have its own timestamp, but that could just be a partial timestamp like “millisecond since ignition” or whatever, rather than a full thing. (Car manufacturers are really cheap on hardware so they’ll likely try to cram as much stuff into a small space as they can, if the reader well be able to use logical deduction to figure out what day, and year it was, rather than use extra bytes to store a full timestamp.) But, so, you can also order things based on the timestamps if the write position is lost, though it might have some uncertainty in it depending on how fine the timestamps are and whether they include a full timestamp somewhere in the log.

I’d be curious to know what the average log history is? The preceding hour’s worth of drive time? More than that?

Data storage in an ACM is extremely limited – manufacturers aren’t trying to track the vehicle over time, but just grab the data preceding the collision and during the event itself that deploys the bags. Typically, then, all they capture is the five seconds prior to deployment, and 0.1 - 0.2 seconds after to collect the collision deceleration info. There is nothing like a platter hard drive in the unit – it’s all chips, and they throw in the least amount the can to get the job done. Heck, the accelerometers that save your ass are only worth about $5.

The furthest back I know any unit goes is a PCM of a Ford, and the Telematics Data from a Tesla.

A Ford’s PCM records the last 25 seconds of engine operation continuously. If the car loses power in a crash, we can capture that data. If some damn fool goes and turns the key, it will overwrite it and we can end up with, say, 10 seconds of data before the crash and 15 seconds of some tow driver fiddling with the car (makes for interesting graphs). But it’s only going back that 25 seconds.

Teslas gather a bunch of data that is sent back to Tesla every few hours. This data is different than ACM data, but does include things like speed and what auto-drive features are on. That data is stored on a SIM card in the dash. I know that data goes back at least several minutes. There is a separate ACM that detects the collision and deploys the airbags.

Because of the limited data stored, there are no timestamps, and so there is no way to correlate a key cycle with a date. All I know is the deployment event is at key cycle 2310, I downloaded at key cycle 2313, and so it’s highly likely that event correlates with the collision under investigation. If the deployment event was at key cycle 1910 (e.g. it’s a hit and run and we find the car later) I can’t say that it was or not.

This article shows the interior of an ACM.

Even cheaper than I expected then! :smiley:

Well maybe real blackboxes work that way. I guess I don’t know.

Black boxes in the old days used wire and foil. From there the moved on to magnetic tape. These days they use flash memory.

A magnetic hard drive would crash the heads in the high-vibration environment of an airplane. Many hard drives also rely on an air cushion to keep the head floating above the platter, and those don’t work very well at high altitudes and lower air pressures. You could make a sealed drive to get around that issue though.

A solid-state hard drive is basically built around flash memory, and since they already use flash memory, packaging it up into a hard drive format would just add unnecessary weight and complexity.

Many passenger planes these days also periodically “phone home” via satellite, and send back all kinds of data about their performance. This is mostly used for maintenance purposes. Sending data this way is very costly, so planes only send a fairly minimal amount of data, nowhere near as much data as what is collected by a modern black box.

So if I’ve got it correctly, the information that is stored is info about the movement of the vehicle, not the usage of the controls? So it wouldn’t be:

"5:15:30 - accelerator has been pressed down for half an hour

5:15:45 - accelerator released

5:15:47 - pressure applied to brake"

It would be more like:

"5:15:30 - speed at 70 mph

5:15:45 - speed at 68 mph

5:15:47 - speed at 60 mph

5:15:45 - speed at 55 mph

5:15:40 - speed at 45 mph"

And smart guys like Patch can then reasonably infer that the vehicle was proceeding normally, but at 5:15:45, the slight drop in speed means the foot was taken off the gas, and then the rapid drop in speed from 5:15:47 onwards shows a hard braking?

I did some poking around. They actually store quite a bit of data, including the accelerator and brake. There’s no need to determine those from other sources like the speed data.

You can see the list of data required here (warning PDF):

The data list starts on page 228 and goes forward for several pages.

This is relatively recent, so older cars obviously won’t store anywhere near the same amount of data.