25

Flight data recorders store massive amounts of data per flight and despite improvements year on year with data storage as with Moore's law, carrying larger and larger amounts of data can become expensive, through larger hard drives and the extra weight.

So, is a flight data recorder erased after every flight? If so, after how long and by whom, is it automatic on landing, engine shutdown or it's one of the pilots procedures on landing?

David Richerby
  • 11,875
  • 4
  • 46
  • 86
securitydude5
  • 8,529
  • 12
  • 57
  • 111
  • 15
    "larger and larger amounts of data can become expensive, through larger hard drives and the extra weight." Patently false. Standard 128GB microSD flash cards are cheap, have massive storage, and weigh nothing. – abelenky Apr 09 '19 at 11:54
  • Aparat from the mandatory crash-proof FDR, there are WiFi data recorders (WQAR) that are just downloaded at the airport, data continue to live on the ground. Other just use optical media to store Tb of data (related). – mins Apr 09 '19 at 12:18
  • 8
    I don’t feel Moore‘s law is applicable here, as cost would be prohibitive when attempting to certify aviation hardware (which has completely different reliability, environmental and robustness requirements) at the same rate as consumer hardware is developed. – Cpt Reynolds Apr 09 '19 at 18:01
  • @abelenky Standard mSD cards do not survive the washing machine let alone 4km under the sea in salt water. Nor do they fare well to fire. – Aron Apr 09 '19 at 18:54
  • 6
    @Aron I think the implication is that if we can fit 128GB on a thumbnail sized chip, we don't need to worry about "size and weight" of additional storage. Just harden your solid state storage and you're done. You're not adding additional drives or making a bigger tape these days. – JPhi1618 Apr 09 '19 at 19:52
  • 1
    @CptReynolds exponential divided by constant factor is still exponential. – Acccumulation Apr 09 '19 at 20:53
  • 1
    Don't forget Wirth's law when mentioning Moore's law: the amount of data that should be stored scales with the place available to store data. – Bergi Apr 09 '19 at 20:56
  • 2
    @Aron Nor does any hard drive, so that's a silly thing to say. The reason these things survive such extreme conditions is entirely down to the casing. And in point of fact, SSDs are way more robust to impact and temperatures than any hard drive. – Graham Apr 09 '19 at 23:19
  • Is there a related question here about what stops a flight recorder from running on, if it happens to still be connected to battery power for 25 hours after a crash ? – Criggie Apr 10 '19 at 19:58
  • My point is that mSD cards aren't archival quality storage devices. Can you ensure that the quantum tunneled charge on 9nm process floating gates can maintain their charge over a period of 2+ years in hostile environments? The data integrity of flash is terrible. – Aron Apr 11 '19 at 05:40
  • @abelenky On a typical flight a Dreamliner will record about a half a terabyte of data: If this data is not erased over the life of the aircraft, the aircraft will only have space to fly with the last 128 GB SD card. Moores law into perspective replacing the 128gb SD card with larger and larger SD cards of the same size is is cheaper more expensive than simply deleting useless data. Rather replace it every 4 or 10 years., perhapls in 20 years time one would store 2048 TB of data on an sd card the same size and weight. – securitydude5 Apr 14 '19 at 10:23

2 Answers2

41

Flight data recorders have changed over the years and evolved to what they are today. Modern recorders are recording up to 88 parameters. The storage of the recorder is dependent on the range of the aircraft. Early recorders were using a circular tape that ran through the recorder. E.g. loop recording all the data and automatically overwriting the oldest data on the tape.

The recorder of a Boeing 777 is based around modern solid state storage, capable of recording up to 25 hours continuously. Modern flight recorders are still recording continuously, so after the initial 25 hours, the storage is full. It will then start overwriting the oldest datablock stored. This ensures that you will always have the last 25 hours of data before a crash available for research.

The following quote comes from AERO Magazine Issue 02 - Spring 1998:

Flight data recorders were first introduced in the 1950s. Many first-generation FDRs used metal foil as the recording medium, with each single strip of foil capable of recording 200 to 400 hr of data. This metal foil was housed in a crash- survivable box installed in the aft end of an airplane. Beginning in 1965, FDRs (commonly known as "black boxes") were required to be painted bright orange or bright yellow, making them easier to locate at a crash site.

Second-generation FDRs were introduced in the 1970s as the requirement to record more data increased, but they were unable to process the larger amounts of incoming sensor data. The solution was development of the flight data acquisition unit (FDAU).

As shown in figure 2, the FDAU processes sensor data, then digitizes and formats it for transmission to the FDR. The second-generation digital FDR (DFDR) uses tape similar to audio recording tape. The tape is 300 to 500 ft long and can record up to 25 hr of data. It is stored in a cassette device mounted in a crash-protected enclosure.

FAA rule changes in the late 1980s required the first-generation FDRs to be replaced with digital recorders. Many of the older FDRs were replaced with second-generation magnetic tape recorders that can process incoming data without an FDAU. Most of these DFDRs can process up to 18 input parameters (signals). This requirement was based upon an airplane with four engines and a requirement to record 11 operational parameters for up to 25 hours (see "Parameters Explained" below).

Another FAA rule change that took effect October 11, 1991, led to the installation of digital FDAUs (DFDAUs) and DFDRs with solid-state memory on all Boeing airplanes before delivery. This FDR system was required to record a minimum of 34 parameter groups. The DFDAU processes approximately 100 different sensor signals per second for transmission to the DFDR, which uses electronics to accommodate data for a 25-hr period.

Today all Boeing current-production models use DFDR systems, which will store 64 12-bit words per second (wps) over a 25-hr period in electronic memory. At the end of the 25 hours, the DFDR will begin recording the most recent data over the oldest data. No tape removal is required with these systems. Each of these systems on every Boeing model (except the 777) have at least two data frames that are transmitted from the DFDAU to the DFDR (see "What Is a Data Frame?" below).

These separate data frames accommodate the different regulatory agency requirements. A 128-wps DFDR was available for the Boeing 777 and MD-90, allowing the development of one data frame that incorporated all regulatory agency requirements and that required operators to develop only one data frame decode algorithm. "How a FAA Rule Is Changed", below, explains the basis on which the FAA may propose rule changes.

Source

Community
  • 1
Brilsmurfffje
  • 5,512
  • 3
  • 26
  • 44
  • 10
    Makes me wonder what kind of ancient memory technology they are using. 64 x 12 bit words per second for 25 hours is not even 10MB. – rghome Apr 09 '19 at 13:14
  • 31
    @rghome Remember that reliability is of extreme concern, and that includes reliability after being lost in the ocean for a few years. There aren't many situations in which you'd need more than 25hrs worth of data, and 64 parameters once a second is enough to store most of what's needed. – David Richerby Apr 09 '19 at 13:27
  • This is interesting, but what's the answer to the question? The question is "Is flight data recorder erased after every flight?", not "How much data can an FDR hold?" The answer describes a loop, but doesn't say that the data is kept until overwritten. – T.J. Crowder Apr 09 '19 at 15:02
  • 1
    @T.J.Crowder The first paragraph says, "After these 25 hours it will start overwriting the oldest datapoint." That tells us that it's not totally erased after every flight, but rather partially erased on every time interval it records a new datapoint, to always have 25 hours of recorded data. – JoL Apr 09 '19 at 15:49
  • 8
    @JoL - No, it doesn't tell me that. It tells me that if it's not erased, it will start overwriting after 25 hours. It doesn't tell me that it's not erased. (This isn't pedantry. I genuinely couldn't tell from the answer whether the common practice was to erase them or not. David Richerby's answer is quite clear on that point.) – T.J. Crowder Apr 09 '19 at 15:56
  • 8
    @T.J.Crowder I see. I was looking at this question from the angle that the data needed to be erased to free up storage space, and so how is it erased? after every flight? Since "after these 25 hours it will start overwriting the oldest datapoint" solves that, I didn't see any reason why there would be any other type of erasure, and took that sentence as saying there wouldn't be. If there is any other motivation for erasing the data, I suppose you're right that this answer isn't explicit on that point. – JoL Apr 09 '19 at 16:51
  • @JoL - Fascinating, and I can totally see how you got there that way. :-) – T.J. Crowder Apr 09 '19 at 16:53
  • @JoL, I clarified my answer based on your comments, I hope it is more clear now :) – Brilsmurfffje Apr 09 '19 at 17:44
  • 1
    @Brilsmurfffje Well, I think it was already clear to me from the beginning. :) I think T.J. Crowder just wanted a "No." at the beginning of the answer. The comments were about whether or not "after these 25 hours it will start overwriting the oldest datapoint" implied that "No". It's still implicit with the edit, but personally I think it's clear enough. – JoL Apr 09 '19 at 17:55
  • 1
    Some recorders compress data in a fascinating way that records a full data set only if significant changes occur, and deltas in between. The capacity is there to hold the mandatory 25hrs for „significant changes“ every second, but since often time is spent in cruise with little happening, you can get several hundred hours of actual data on these recorder types. – Cpt Reynolds Apr 09 '19 at 18:00
  • 1
    @rghome, not ancient, durable. I bet your phone's memory card won't survive hitting a mountainside at 500 km/h, followed by sitting in a fire for several hours. – Mark Apr 09 '19 at 21:36
  • To that point, after the Lion Air MAX crash, they analyzed the data from the previous flight as well as the crash flight and found a similar problem with MCAS--and those pilots were still alive to explain how they dealt with it. – StephenS Apr 09 '19 at 23:20
  • @rghome The quoted document is ancient; it dates back (unchanged!) 20 years, to mid-1999 (maybe even 1998 if you go by the copyright footer). 10 MB of solid state memory in 1999 was quite a lot! Remember, this predates USB flash drives, and even they only started at around 128 MB a year or two later. Possibly they are still "good enough" and still in use (because certifying new formats is expensive). Or we might already have something newer (apart from QARs). – Bob Apr 10 '19 at 04:34
  • It's probably relevant to add that the 88 paramenters on a per second basis are what is required by FAA regulations. In actual fact there is a lot more data stored, but that data is confidential and varies by manufacturer and model. The A388neo for example is reported to store half a terabyte per flight, which is quite a lot more than even 25 hrs of FAA regs would produce, and probably at a much greater accuracy than per second. An impact at 400kph may take far less than a second to disconnect a black box from it's sensors so a higher polling rate would do wonders for analysis. – user1901982 Apr 10 '19 at 13:06
  • @rghome A standard 90 minute audio cassette holds about 2MB per side when encoded with a reasonable amount of error correction. That's essentially the same technology used up until the development of solid state storage. Rotating drives are far too sensitive to vibration and G forces to use in such an application. (I hope nobody was silly enough to try it anyway.) – Perkins Apr 10 '19 at 17:42
  • @JoL I would genuinely expect a manual erasure at the beginning of a flight, to aid later forensics if nothing else (no point in them potentially getting confused by data recovered from a previous flight). That being said, data erasure can have a lifetime cost with some technologies. Anyway, David seems to have shown that my expectation is not reality. – Lightness Races in Orbit Apr 11 '19 at 02:22
  • 1
    @LightnessRacesinOrbit I think it would hinder forensics more than it would aid them. The bar would have to be really, really low for a forensics team to get confused by that, and the state of the plane in previous flights seems like very important contextual data to me. Some event in a previous flight could explain the state of the current flight. As to the lifetime cost of data erasure, I think in contrast to erasing at the beginning of each flight, overwriting the oldest datapoint to preserve constant time is a much simpler mechanic that makes lifetime prediction much more accurate. – JoL Apr 11 '19 at 16:04
  • @JoL: Fair enough! – Lightness Races in Orbit Apr 11 '19 at 16:57
  • 1
    I’ll have to check when I get back to work, but I’m pretty sure the 777 is up to 50 hours now. Also, as an aside, I believe ALPA has some say in how long the recording can be. – Frank Apr 11 '19 at 22:57
28

So, is a flight data recorder erased after every flight? If so, after how long and by whom, is it automatic on landing, engine shutdown or it's one of the pilots procedures on landing?

Absolutely not! Recorded data isn't just needed for flights that crash: there are plenty of cases where something bad happens in mid-air but the plane can still make a controlled landing, in one piece. Automatic erasure would mean you'd never have data from those incidents.

Rather, recordings are made on a loop. Originally, this would have been a looped tape; now it's digital memory. Once the memory is full, each new chunk of data overwrites the oldest stored data, so there is always a record of the last X amount of time.

There isn't really ever a reason to erase data from the recorder, except by overwriting it with new data. Having data from previous flights on the recorder is only a very minor inconvenience for investigators: the data is timestamped, so it's easy to extract the data they need. Not erasing the old data means that it's still there even if you only realise you want to see it after the plane has set off on its next flight. It also allows investigators to look a little farther into a plane's history and see if problems on an incident flight had also occurred on previous flights, without major consequences.

David Richerby
  • 11,875
  • 4
  • 46
  • 86
  • 6
    "plenty of cases": a count which probably exceeds actual crashes by at least an order of magnitude. Given that a crash usually requires a whole series of things to go wrong (and each of those "things" usually doesn't go wrong), "we had a problem but were able to recover" is going to be much more common. – Martin Bonner supports Monica Apr 10 '19 at 10:19
  • This doesn't say anything about the possibility of erasure between flights more generally, say at the beginning of a flight (and it need not be automatic). I'm a little surprised that this isn't indicated, for I would have thought that if a crash does occur you don't really want the dataset peppered with old, irrelevant information when you're trying to do forensics on a potentially damaged recorder. Then again, as I've commented elsewhere, full erasure is likely to have a lifetime cost for the device. – Lightness Races in Orbit Apr 11 '19 at 02:24
  • 1
    @LightnessRacesinOrbit Lifetime cost is far from the biggest issue. Having data from previous flights on the recorder is at worst a very minor inconvenience for investigators: the data is timestamped. And it has the (huge) advantage of being able to obtain data from a recent flight if you decide you want it after the plane has already begun its next flight. Storing data for the last 24hrs means that you might even have several days to decide "Hey, we need the data from that flight on Tuesday now that three more pilots have reported that slightly weird thing." – David Richerby Apr 11 '19 at 08:45
  • @DavidRicherby Okay, that seems reasonable. :) (Would you consider merging some of this to the answer? You have some great points with value here.) – Lightness Races in Orbit Apr 11 '19 at 10:47
  • @LightnessRacesinOrbit Done! – David Richerby Apr 11 '19 at 12:51
  • Sweet, nice one. – Lightness Races in Orbit Apr 11 '19 at 13:19
  • A good example of using previous data is the Gulfstream G-IV crash at Hanscom Field, where investigators used the data to determine that "the pilots had neglected to perform complete flight control checks before 98% of their previous 175 takeoffs in the airplane, indicating that this oversight was habitual and not an anomaly." – Robie Basak Apr 15 '19 at 13:58