University of Bielefeld -  Faculty of technology
Networks and distributed Systems
Research group of Prof. Peter B. Ladkin, Ph.D.
Back to Abstracts of References and Incidents Back to Root

Engineering Analysis of the Landing Sequence

DLH2904 Airbus A320 accident, Warsaw, 14 September 1993

Clive Leyman

Visiting Professor in Mechanical Engineering and Aeronautics, City University, London
and Pucklechurch Consultants

23 January 1996
Because of the difficulty in obtaining numerical data, much Internet discussion on aircraft accidents is qualitative in nature. However, without quantitative information it is not possible to judge whether the effects being discussed are really important. This is particularly true when, as is often the case, the behaviour of a particular computer or software is concerned. In order to shed more light on the circumstances surrounding the crash of the Airbus A320 Flight DLH2904 at Warsaw, the official report has been carefully studied, and with the addition of some aerodynamic and performance parameters "guesstimated" using the author's experience, the accident has been modelled in a manner that allows `what if' calculations to be made. It is hoped that by publishing these data, a more informed judgement can be made by laypersons.


It should be stressed that this work IN NO WAY questions or casts doubt on the findings of the official Polish report on the accident - it merely seeks to quantify the effects already identified in that report. In addition, it has to be said that despite the author's previous connections with Airbus, these results have been obtained by a completely independent route, and that they do not represent the Airbus position in any way.

Sequence of Events

For the benefit of readers who have not seen the official report, the sequence of events was as follows (page numbers refer to the official report):-

Approaching Warsaw to land on R/W 11, the crew were told that an earlier landing aircraft had experienced severe windshear on final approach. They were given an airport wind of 25 km/hr (13kts) at 1600, i.e 10kt headwind/8kt crosswind (p9). The crew elected to make a manual approach, without autothrottle (p28) and `autobrake' landing mode (p42). Following company practice in windshear conditions they chose a speed about 15kts above the normal reference speed (145kts instead of 130kts) From the CVR, it is evident that they were monitoring wind speed carefully throughout the approach. Analysis of the data shows that at between 300ft and 200ft altitude the wind was a tailwind of around 25kts, gusting +/- 5kts. [Aircraft Flight Manual limit was 10kts] At about 150ft the wind changed to a 15kt tailwind, gusting +/- 5kts. This was commented on by the crew, but they made no throttle adjustment. The airspeed increased to 155kts +/- 5kts, and the aircraft rose slightly above the ILS flight path.

Here are graphs of altitude and windspeed and airspeed on approach, for comparison.

At 50ft the airspeed was 159kts, and the tailwind about 15kts. The landing flare was longer than usual, and the aircraft touched down on one wheel, apparently conforming with company practice for a crosswind landing. De-rotation was normal, but the aircraft was still held with one wing high, so that for some time, only the right mainwheel and the nose wheel were in ground contact. The runway was flooded, with standing water to a depth of 3mm. The logic to trigger deployment of reverse thrust and lift dumpers required either both mainwheels to be on the ground [signalled by position switches on the landing gear torque links], or one mainwheel on the ground and a rotational wheelspeed equivalent to a ground speed of at least 72kts. Brake application depended on mainwheels spinning up to 80% of a computed reference speed, which is normally the rotational equivalent of ground speed, with special treatment during the transitional touchdown "spin up" phase. Because of the excess airspeed, the aerodynamic lift was much greater than normal, so that the ground reaction on the landing gear was very low. Taken in conjunction with the flooded runway, this provoked aquaplaning of the tyres in contact with the runway, so that they never got up to the required rotational speed, and brake application and the application of spoilers/ reverse thrust was delayed. The aircraft slithered along the runway for a further 9 seconds, until aerodynamic and aquaplaning drag had reduced the speed sufficiently to compress both main gears [the other wheel having by now been put onto the runway]. Once the automatic sequence had been triggered, the aircraft started to decelerate, but the deceleration was significantly lower than normal, even for a wet runway. When it became clear that they were not going to stop inside the runway, the crew decided to veer off to the right, but because of the low runway friction although the aircraft heading changed, the trajectory continued more or less along the line of the runway until the aircraft collided with an embankment, with tragic results.

`What if' results

In a complex situation such as that under consideration, the calculated magnitude of any effect will depend on the values assumed for the parameters being held constant. Since some of these effects are sequential and some act in parallel it is really not practical to give precise rates of exchange for each. To establish a realistic range for these rates of exchange, the following procedure was used. A set of parameters was identified, each of which plausibly had an effect upon the landing distance. These effects were `removed', cumulatively, in the reverse order of their occurrence in the landing sequence. That yields the (continually reducing) numbers in the column `Reverse Sequence', in the table below. These effects were also evaluated as if each were the only variant parameter, and these distances are included below in the column `Individual Effect'. It should be clear that the `Reverse Effect' column underestimates the magnitude of the earlier effects, and the summation of the `Individual Effects' overestimates the potential distance reduction, since many of these factors operate in parallel, and not sequentially (as the arithmetic assumes).

Effect Reverse Sequence Individual Effect Remarks
Distance m Distance m
a) Calculated to complete stop 3160 3610 Datum for exchange rates
b) No aquaplaning 2910 2910 With CAA `Reference Wet' Friction
c) No Brake/spoiler delay 2510 2745 Trigger: one MLG +NLG
d) Manual Braking 2210 3145 Removes 0.25g decel'n limit
e) No long flare float 2105 3035 DLH FCOM value
f) Remove excess altitude 1985 3055 Aircraft altitude = 50ft at glideslope = 50ft
g) Flight Manual tailwind 1825 2955 Reduce tailwind value to 10kts
h) Reduce excess airspeed 1660 2700 Airspeed at 50ft = Vref + 20kts

Calculated distance from threshold until 70kts speed reached
under actual prevailing conditions = 2828m

Available Field Length 2830m

Certification Field Length 1553m (Flight Manual Wet Field Length, 0 Wind)

All distances are given from the runway threshold. The actual distance from the threshold to 70 kts on the day was 2830m, which compares well with 2828m as a starting point. Although not shown, the calculated `certification' distances also compare very well with the values given in the FCOM. We may therefore be reasonably confident in the exchange rates. One other piece of information is required for full understanding of the situation - with full flap deflection there is a powerful `ground cushion' effect. This is common to all civil aircraft with a low wing mounting. As a result, considerable aerodynamic lift can be generated, even with the aircraft running at essentially zero incidence as is the case with all wheels on the ground. This lift is normally 'killed' by deflection of the lift dumpers so as to maximise the ground reaction and increase braking force. With the excess airspeed present in the Warsaw incident, the aerodynamic lift before deployment of spoilers was considerable - it is estimated to be :-
Airspeed kts EAS 152 149 143 136 130 120
Lift tonnes 51.5 49.5 45.6 41.2 37.7 32.1
Ground Reaction tonnes 7.3 9.3 13.2 17.6 21.1 26.7


One could use these figures to justify almost any scenario that one wished. Most observers have latched on to the delay of 9 secs between nosewheel contact and start of the brake sequence as a primary cause of the accident, and on the face of things this is a reasonable reaction. From the value of aerodynamic lift, one can see the reasons for this delay - at the speed existing at the time of nose wheel contact (149kts), the ground reaction was only just sufficient to break out the rather high stiction (6.3 tonnes) on this particular design of landing gear. Less obvious, and not mentioned in the official report, is that the low ground reaction at this time, when combined with worn tyres and a flooded runway provide ideal conditions for aquaplaning, which made it impossible for the second condition of spoiler/reverse thrust triggering to be fulfilled, or for the brakes to be effective (see above). [One may note also that even if the landing had been with wings level, the brake sequence would still have been delayed by about 4 secs until the speed had fallen to about 143kts and the ground reaction increased to 12.5 tonnes]

The table of `what if' effects shows that if the brake logic had been different - say one MLG plus the NLG as on the A330/340, then the immediate triggering of the braking sequence would have shortened the distance to 2510m, and the accident could have been avoided.To this extent therefore the observers are correct, but are they correct to suggest this to be the prime cause of the accident?. Equally valid questions are whether the accident could have been avoided if the air and ground speeds had been more tightly controlled, or if the runway had not been flooded and the tyres worn.

The Lufthansa FCOM wording clearly allows pilots to increase airspeed by up to a maximum of 20kts if they believe there to be a risk of wind shear, and gives landing distances up to a maximum of 10kts tailwind. It is reasonable therefore to expect the aircraft to function correctly at these speeds.[Note however that the AI FCOM gives a limit of 15kts]

Calculations consistent with the values given above show that purely due to the reduction in ground speed, the landing distance at the DLH FCOM limits would have been 2690m, but in fact the reduced airspeed would have increased the ground reaction sufficiently to have compressed both oleos (wings level) and the delay in braking would have been avoided, with a consequent reduction of distance to 2225m. It is therefore equally true to say that the control of air and ground speeds was a primary cause of the accident. In fact, one might argue that since the high speeds existed as input conditions to the brake sequence, they are relatively more likely to have been a cause of the accident.

Similarly, if aquaplaning was absent (and the evidence is that the whole of the ground roll was subject to aquaplaning), then the delay in initiating braking would have been avoided, and the deceleration increased such that the landing distance would have been reduced to around 2350m, and there would have been no over-run.

Here are graphs of flare and derotation, ground deceleration and runway friction.

This is not to say that the braking system logic is immune from criticism.

Although the logic will `work' at airspeeds up to 23kts above normal approach speeds, i.e. 3 kts above the FCOM limit, there remains the possibility that the tendency of the aircraft to float on the runway at these levels of speed could, in other circumstances, produce problems. In addition, if the FCOM allows approach at up to 20kts above the normal speed, it would be necessary to ensure that the system had adequate operating margins to cater for gusts etc. In this case, a margin of 3kts in winds that might be as much as 30kts seems not sufficient.[But with the AI FCOM limit of Vref + 15kts, the margin is a much more reasonable 8kts ] One can either change the braking logic, which would react to the particular circumstances of Warsaw, or try to reduce the floating tendency at high approach speeds, which would address the fundamental problem.

[It is understood from informal contacts that Lufthansa made a temporary `fix' by recommending use of less than full flap in extreme weather conditions, and that Airbus have since introduced new logic to partially reduce lift immediately on touchdown, whilst retaining enough performance to allow a baulked landing to be safely executed.]

However, the crew cannot escape criticism either. Given their reaction to the weather conditions, one must ask what advice they were given in the FCOM as to how to treat windshear. In fact they are given quite a lot of advice, of which perhaps the best is "AVOID, AVOID, AVOID".

The FCOM contains the following advice/instructions (not given in the order they appear) Bold characterisation is the author's.

In the light of this advice, one must question some of the crew's decisions :-

Why did they disconnect the autothrottle? 25kts +/- 5kts is hardly enough to produce a level of turbulence sufficient to provoke poor performance, and this decision seems to have been made very early in the approach. Use of the autothrottle would probably have prevented that final, critical, speed increase to 29kts above Vref.

If they were flying in a 25kt tailwind into a airfield with 13kts headwind, why did they persist with an increased approach speed?. At this point, either they were not believing the airfield wind, in which case they were about to make a landing at 2.5 times the Flight Manual limit, or they should have been expecting to encounter a headwind (performance increasing) windshear, in which case they did not need the extra airspeed margin.

There is no mention on the CVR of any consideration of the effects of their chosen speeds on the landing distance required, or indeed of any reference to the FCOM for guidance on this point, although it must in fairness be added that the guidance that they would have received would have been inadequate to cover the extreme combination of conditions that existed.

Despite the FCOM advice on not allowing the aircraft to float during landing flare, this is exactly what happened, although here again this was very probably a result of the high airspeed at the threshold.

[Put flare and derotation graph in here -(warsaw2)]

In the author's opinion, the official report underestimated the effect of aquaplaning as a cause of the accident. It is apparent from the records that the deceleration actually got less as the ground speed reduced, which at variance with normal expectations that available runway friction increases at low speeds. More detailed analysis shows that the braking friction coefficient was more or less constant throughout the ground roll at about 0.1. This is consistent with an aquaplaning situation, far below the friction to be expected on a normally wet runway. The report suggests aquaplaning only on the concrete at the end of the runway, but there is another explanation for the measured reduction in longitudinal deceleration, namely that if the tyres are aquaplaning, then the total resultant tyre force is constant, and the reduction in deceleration along the aircraft X axis is simply due to the change in aircraft heading and the resolution of the constant tyre force along the X and Y axes. As mentioned above, if aquaplaning could have been avoided, then the wheel spin-up would have been as expected, the spoiler/reverse sequence started earlier, and the overall deceleration increased.


If we take the USAF definition that a `cause' is "an act, omission, condition, or circumstance which either starts or sustains a mishap sequence. It may be an element of human or mechanical performance. A given act, omission, condition, or circumstance is a "cause" if correcting, eliminating, or avoiding it would prevent the mishap or mitigate damages or injuries", then the crew decisions, aircraft system design and airfield/environmental conditions are all potential `causes' in this accident, and it is not profitable to try to isolate a single item.

Every effort has been made to ensure this text is syntactically identical to the original source.
Inquiries to
Peter Ladkin gratefully acknowledges Clive Leyman for permission for this use of his material.

Back to `Incidents and Accidents'.

Peter B. Ladkin, 1999-02-08
Last modification on 1999-06-15
by Michael Blume