AF356 tailstrike in yyz

[fdr]

Originally Posted by CVividasku

I disagree. I looked up the 350 FCOM and it does very clearly state that the flare law is a direct law.
If you say that a flight control surface behavior makes sense "when taking all factors into account", okay, but that only applies if you consider that the law is not direct anymore.

I'm sorry to disagree again.
You're talking about sampling error, you also talk about a 0.25 or 0.5s sampling interval for pitch parameters.
Yes, but, if there is a two second discrepancy between two parameters that are supposed to be linked, it can't be attributed to sampling error.
Two seconds being much larger 0.5s.

Then, if you look closely on the curves, I count sixteen datapoints per each 2 seconds on the pitch SSC. As well as on the elevator deflection curve. Which means a 0.125s sampling interval. I don't have the A350 calibration files to decode the DFDR data, but I'm pretty sure that you will find 8 words per second for both these parameters.
The A350 being a modern airplane, there are many more parameters than on the older 320s and it's really not surprising to have such precision.

Basing all your reasoning on something, that, by chance, was not recorded, does not sound very serious. It can happen but it's very unlikely.
And even if what you said happened, that is, a jolt from the pilot flying at the tailstrike, it does not change the problem at all. It reduces a bit its seriousness, but the deed was already done at this time.
It does not explain why the elevators had been staying so much nose up even while the PF was, almost steadily, releasing his nose up input.

I don't know if you've flown airbuses, but I've never seen anyone making alternating inputs quicker than 4Hz...
Due to the internal damping of the sidestick, it would be very difficult to make significant oscillations at a significant rate. I'm pretty sure the 0.125s interval would capture any involuntary jolt, if it was significant, that is, if it was large enough to produce an elevator deflection visible on the curves. I'm completely sure if I try to pull a small jolt on the airbus sidestick, I can't make it quicker than 0.25s (even very small deflections like 10% of travel) and even less so quicker than 0.125.

If the report was as thorough as an accident report, it could have a paragraph somewhere stating that they did tests like so and concluded. I'm 99% sure it's impossible to make a jolt without being detected (shorter than 0.125s).

Lastly, I don't believe the artifact story. How come the sensor doesn't have any problem later on ? If there is a data point at some value, it means it was measured at this value. The shock did not impact directly the sensor nor any of its components, the acceleration alone was not very large (yes it's measured at the C of G for the one we have, but seeing the very steady increase in pitch, if we computed, we wouldn't find a very large acceleration at the tail, as computed by Nz + q'*length)

So, to sum up, I don't really buy your arguments, and even if they were to explain some things, they wouldn't explain the main problem, it being the 2s desynchronisation between elevators and SSC.

There is no 2s desynchronisation between elevators and SSC.

• The SSC is not rate limited
• The control surface is rate limited

The difference between control input and response always has an error unless the system is static.
The control surface actuator can be and is saturated in this condition, which is a normal system condition when a high amplitude input is made over a short time frame. The only time that doesn't occur is where the linkage between the two is rigid, as in aerobatic aircraft with push pull control rods with high rigidity, and zero pivot slack.
The FCC output in these cases is a proportional response, it is whatever the pilot is inputting by the SSC. When in normal law, the FCC will make control outputs as required to meet the demand, and those are determined by a programmed response table for the output, these tables are dependent on configuration.

The time delay from the saturation of the actuator is approximately 1.0s. For the ~36s immediately before the tail strike, the lag in the commanded position of the control and actual is minimal, and what is normally observed where there has not been a high amplitude, short time period input. Saturated control actuators are observed on various occasions, but are not the general manner by which the aircraft is flown, In this case, the controls have gone from near full nose up to reversed to full nose down in about 2s, and then in the next 2s have been reversed towards ~1/2 nose up again, then over the following ~2s to half nose down, before being nulled. With the gyrations of the high amplitude inputs which saturate the controls, position lag occurs. That is completely normal system function, the control inputs are less normal, they are unfortunately an over control of the required inputs by the pilot.

The sampling rates are required to be no less than 0.25s for both of these chanels, the rate that is recorded in the DFDR is 0.125s.

I don't know if you've flown airbuses, but I've never seen anyone making alternating inputs quicker than 4Hz...

Yes, I have, A320, 330, 340, and all of it as flight test. Of more relevance is I have conducted incident review on A320, A330 and various other Airbus products, as well as B737, B747, B777 types.

Spoiler

 

 

The elevator deflection has a maximum deflection rate that can be applied, just because the pilot can rapidly move the SSC or prong, doesn't meant the elevator position does at the same time. This is common with all types, and is one of the reasons that the FDR records both the pilot control input and the control surface position. They are not mechanically linked, the control can only move at ithe maximum rate of the actuators. Same if it was a B777, B747, A320.. etc. That the elevator is rate limited is evident on 6 occasions on the zoomed chart of my last post. That was provided to show the response that occurs on all of these systems.

There is no obvious defect, there is a sampling related artefact that is what is observed routinely on data analysis of the FDR. If anything, this is better than most prior types.

Just because a chart exists does not mean it is showing the complete behaviour of the system, sampling rate makes a difference, and often throws up artefacts as a consequence. The data sentence does not have a synchronous timing of all of the samples, they are for the most part serial, which permits momentary variations to occur between an input and an output. The output in this case is a hydromechanical system which has system lag and rate limits.

The data points that are measured here I do agree are around the 8Hz, looking at the identifiable short period changes. The higher the frequency of the sampling the better, but it is still within the range that artefacts will arise. There is a general rule when undertaking a spectral analysis to determine the minimum sampling rate to avoid artefacts, and as the input to the SSC is analog but the output is digital to the control actuators control servo it is difficult to avoid artefacts over short periods.

The "2 second" slip that is suggested to have occurred appears once in the data set, immediately prior to the apparent time period of the tail strike (this is still LHR that is being discussed). Around 1 sample period before the tailstrike (0.125s, 8Hz), the system is indicating a consistent response. After this point, CVividasku contends that there is a "2 second" error in the control system, once.

The elevators are rate limited in their deflections, that is observable in the data at multiple points, and therefore the variation between commanded deflection and actual deflection will always exist, it is only a matter of degree. For the one point however, there is a change in the direction of the control position vs the control command, and that only requires one additional short period input to have been applied. (flight controls are always rate limited, but in normal operation that is not observed by the pilot, they don't get to the condition where the system limited response becomes apparent, most inputs are of limited amplitude in normal operation, and at rates that are trained to be smooth).

The proximity of the questioned response to the reported time period of the tail strike which is based on the best estimate of the investigators, and which is subject to the sampling errors and dynamics of the structure in response to loads, is of interest. The potential for an input from the pilots arm as a response to the impact of the tail is not zero. Any imparted load to the pilot at that point would be transferred to the SSC as a momentary transient and for this type of control can be a high amplitude input. What would be interesting to know is if the signal to and out of the FCC is/was stored in any form, I would suspect that is not the case. The DFDR data should record the sampled signal out from the SSC, the response is recorded for the control surface, but the output and the rate of the actuator is probably only identifiable in testing in the control system model. That has been done before, including environmental testing of the actuator servo, which discovered the B737 rudder PCU issue. A momentary bit flip would also achieve the same sort of problem, however even with the QF072 event, that was only a hypothesis that was floated, and the extended investigation found 3 occasions of data spikes from the ADIRU involved in that case, in 19 million hours of operation. If a high amplitude, short duration input was made on the SSC and was between the sampling of the SSC, that doesn't mean it was not detected by the FCC and responded to as a signal to the control surface. The data that is presented indicates that the control surface has moved in the opposite direction of the sample commands for a single command sample point. That sets up a rate limit case thereafter, which lags the control input thereafter, with a 1 sec lag. The change between the SSC command position and the control position is a variable; The SSC can go full scale deflection as fast as the body can move the control grip, which is less than 0.25s, but the control surface takes around 3 seconds to sweep full scale deflection, so lag of the control itself is a variable. Under normal operations this is not an issue, it is when high amplitude control inputs are made at short periods.

In the Heathrow case, proximate to the tail strike, there is a single point where there is an apparent anomaly. That is explainable by a single unrecorded high amplitude SSC input, that has occurred outside of the sampling. An inadvertent input is possible from the physiology of the pilot; the upper body, shoulder, and arm is subject to movement under load, and that transfers to the SSC as an input, which is the reason for the arm rest on the Airbus, it reduces APC/PIO/APIO in response to the dynamics of the human body coupling to a control input. A control anomaly that results in a one off event is not impossible, it is extremely improbable, around 10e12 or thereabouts. For that to happen within 0.125s of a tail strike that has already been set up to occur, and occurs once, and not again, is around 10e18 by quick calculation of probability. Doesn't mean it didn't happen, it just means that it is rather remote in probability, and other known factors are far more likely.

​​​​​​​Yes, but, if there is a two second discrepancy between two parameters that are supposed to be linked, it can't be attributed to sampling error.
Two seconds being much larger 0.5s.