An Analysis of Hard Landing Incidents

Based on Flight QAR Data

Lei Wang 1,2,3, Changxu Wu1, Ruishan Sun 3, and Zhenxin Cui 3
1
Institute of Psychology, Chinese Academy of Sciences, 16 Lincui Road,
Beijing, 100101, China
2
Graduate University of Chinese Academy of Sciences, 19 A, Yuquan Road,
Beijing, 100049, China
3
Civil Aviation University of China, 2898 Jinbei Road, Tianjin, 300300, China
wanglei0564@[Link],
[Link]@[Link]

Abstract. Hard landing is one kind of typical landing incidents that can cause
passenger discomfort, aircraft damage and even loss of life. This paper aimed to
find out flight performance and operation features of hard landing incidents by
using the methods of variance analysis, regression modeling and flare operation
analysis based on flight QAR data. Results showed that pilots need to control
the aircraft to an appropriate groundspeed and descent rate before descending to
the flare initial point. Then control column and throttle operation in flare ma-
neuver would affect landing performance conjointly. The logistic model
showed that the vertical load of touching ground was actually linked with
touchdown attitude and configuration closely, including three variables of pitch
angle, roll angle and flap degree. These findings were expected to be applied in
practice to prevent hard landing incidents and even landing accidents.

Keywords: Hard landing, QAR, flight safety, flare.

1 Introduction

Final approach and landing is the most important flight phase because a pilot needs to
deal with more operations, decision-making, and workloads than other phases [1-4].
Hard landing is one kind of typical landing incidents which is defined as the main
landing gear impacts the ground with a greater vertical speed and force than in a nor-
mal landing. Hard landings can vary in seriousness from simply causing mild passen-
ger discomfort to situations resulting in serious vehicle damage, structural failure, and
even loss of life [5-7]. When an aircraft has experienced a hard landing it has to be
checked for damage before its next flight. Statistics also showed that hard landings
happened frequently.
Though many studies regarding hard landing have been conducted, most of them
have been based on models or experiments rather than real flight data [8-10]. Quick
Access Recorder (QAR) is an airborne system which can record all kinds of position
parameters, movement parameters, operation and control parameters, and alarm

D. Harris (Ed.): EPCE 2014, LNAI 8532, pp. 398–406, 2014.

© Springer International Publishing Switzerland 2014
 An Analysis of Hard Landing Incidents Based on Flight QAR Data 399

information in the whole flight phase. The hard landing was judged by the parameter
of vertical acceleration. It is generally monitored by using QAR data in most com-
mercial air carriers, but these data are also confidential for them. Meanwhile, there are
few aviation administrators whom enforced their airlines to install QAR equipment on
every commercial transport jet [13]. Therefore, QAR data were difficult and rarely
utilized into research. This paper aims to find out flight performance and operation
features of hard landing incidents through analyzing QAR data and put forward the
prevention measures at the same time.

2 Methods

2.1 QAR Data Collection and Processing

The 119 cases of QAR data in this study were collected from three commercial air-
crafts (Boeing 737-800) of a local airlines company. The original data is a CSV
(Comma Separated Value) file with thousands of rows and columns. Therefore, VBA
(Visual Basic for Applications) programing functions in Microsoft Excel was applied.
In the final landing stage, aircrafts always fly within profile of a landing glide path;
their position changes in the lateral axis are quite limited. Therefore, we focused on
longitudinal and vertical parameters in this study. Finally, 19 columns of relevant
original QAR data of every file were refined. 21 flight parameter variables were then
selected and calculated as shown in the following table based on VBA programs.
These parameter variables covered all flight and operational parameters in the critical
visual and manual landing stages from the flare initial height to touchdown. Mean-
while due to flare maneuver would reduce the aircraft’s descent rate to acceptable
levels so that it settles gently on the main landing gear. It was seemed as one of the
most skilled operation in flight [10], the pilot operation below 200 feet, especially the
flare operation was selected as the main subject for analysis.
Among that the Flare height meant the height of initiating flare operation and Flare
time meant the total time of aircraft flying from flare initial point to touch down point.
It should be noted that the variable of flare time means the total time from flare initial
point to touch down point. In addition, the flare operation initial point in this study is
higher than the standard 30 feet in most flight manuals. This is because any slight
backwards pulling of the control column could be recorded by a Quick Access Re-
corder, causing that the time and height of flare is earlier than theoretical value. The
variable of Touchdown Distance and Vertical Acceleration Touchdown were two
parameters using to determine long landing and hard landing. The Vertical Accelera-
tion Touchdown meant the maximum value of vertical acceleration when the main
landing gears touch the ground [11]. Based on the common statistical results of QAR
data and monitoring criterion of aviation operators [12-14], the threshold of determin-
ing hard landing for this aircraft type was set as 1.4 g in this study.
400 L. Wang et al.

Table 1. Selection of parameters

Classification Name Parameter name in QAR Units

Kinematics & Performance Flare height RADIO HEIGHT Feet
Flare time / Second
Groundspeed GROUND SPEED Knot
Descent rate VERT SPD Feet/min
Airspeed AIR SPD Knot
Vertical acceleration VERT ACCEL g
Touchdown distance / Feet
Operational Parameter Throttle resolver angle SELTD TRA FILTERED Degree
Control column position CONTRL COLUMN POSN Degree
Control wheel position CONTRL WHEEL POSN Degree
Control column force CONTRL COLUMN FORCE LBS
Control wheel force CONTRL WHEEL FORCE LBS
Flap handle position FLAP HANDLE POSN Degree
Speed brake handle posi- SPD BRAKE HANDLE Degree
Rudd pedal position RUDD PEDAL POSN Degree
Configuration & Attitude Flap FLAP Degree
Aileron AILERON POSN Degree
Elevator ELEV POSN Degree
Rudder RUDD POSN Degree
Pitch angle CAP DISP PITCH ATT Degree
Roll angle CAP DISP ROLL ATT Degree

2.2 Statistical Analyzing and Modeling

119 QAR data samples were divided into two groups with 65 cases of normal landing
(Group 1) and the other one was 54 cases of hard landing (Group 2). QAR data of 65
normal landing events and 54 hard landings were regarded as two groups of indepen-
dent samples. Each flight parameter variable of these 119 flights was also calculated
by using VBA program.
First, the flight performance parameters of all flights were analyzed. For the aim of
observing dynamic change of flight performance parameter variables in final landing
phase and their differences between two groups, the altitude of 200 to 0 feet was di-
vided into four flight phases (200-150-100-50-0 feet) and selected flight parameter
was measured and compared in every phase. The multivariate analysis process of
general linear model was introduced to compare the differences in the two groups.
Second, for the aim of finding the operation features of hard landing incidents and
their correlations with landing performance, the statistical methods of variance analy-
sis was used to find the difference of flare operation between normal landing and hard
landing, including their parameter differences at flare initial point and in the whole
flare process. One way ANOVA was used to examine variables which were subjected
to normal distribution and non-parameter K-W test for other ones.
 An Analysis of Hard Landing Incidents Based on Flight QAR Data 401

Third, aiming to find key flight parameters causing hard landing incidents, the lo-
gistic regression model on hard landing incidents was developed. In this study, the
occurrence of hard landing was defined as a binary and dependent variable, where the
value is 1 if it happened and 0 if it did not happen. The hard landing was judged by
the parameter of vertical acceleration. We selected 17 flight parameters from Table 1
as original covariates in this logistical model, which including all operational parame-
ters, configuration & attitude parameters and 3 kinematics parameters of
groundspeed, airspeed and descent rate. Due to flare is a continuous operation from
flare initial point to touchdown, the parameter value both at flare initial point and
touchdown point were sampled in and there were 34 independent variables in total.
The name and definition of each flight parameter is as showing in Table 1. The for-
ward stepwise method was then performed. The likelihood ratio test ( χ difference)
2

testing the change in –2LL (log likelihood) between steps was utilized to determine
automatically which variables to add or drop from the model. The final predictor va-
riables and coefficients of the model were obtained in the stepwise process. Simulta-
neously, the effectiveness of the model was checked and discussed below.

3 Results

3.1 Flight Performance Analysis

The variable of vertical acceleration is essentially both subjected to normal distribu-
tion and the results of Anderson-Daling test also proved it ( p > 0.05 ). For the selected
119 samples, the mean and standard deviation of Vertical Acceleration Touchdown
respectively was 1.387±0.082. The differences between 18 variables from 200 feet to
touchdown were analyzed by using repeated measure and one-way ANOVA. Here
only several important results regarding parameters of groundspeed, descent rate,
control column and throttle are presented. Groundspeed and Descent rate are the two
most flight performance parameters in landing and their change trend is as showing in
Figure 1.

Fig. 1. Difference analysis of groundspeed and descent rate

402 L. Wang et al.

As shown in Figure 1, the groundspeed of normal group is slightly greater than

hard landing group. The difference of variable Groundspeed is not significant in the
whole stage of 200-0 feet ( F (1, 117) = 1.763, p < 0.183 ). The results of repeated
measure ANOVA showed that the group effect of variable Descent rate is not signifi-
cant ( F (1, 117) = 2.410, p = 0.123 ). The descent rate of hard landing is slightly larg-
er than the normal group before 50 feet, also the flare initial point, which changes a
lot past 50 feet.

3.2 Flight Operation Analysis

Then, there were 65 normal landing samples (Group 1) and 54 hard landing samples
(Group 2). The descriptive statistic on flare initial height and operation time of the
two groups is as follows.

Table 2. Statistics on flare height and time

Group N Flare Height (M ± SD, feet) Flare Time (M ± SD, s)

Normal Landing 65 52.169 ± 23.521 8.031 ± 2.076

Hard Landing 54 51.963 ± 20.175 7.722 ± 2.141

As seen in Table 2, there is no significant difference between the flare initial height
of two groups, which are both around 50 feet ( F (1, 117) = 0.006, p = 0.941 ). Mean-
while, the flare time of two groups also does not indicate significant difference (
F (1, 117) = 0.633, p =0.428 ). Flare operation is considered one of the most technically
demanding aspects of piloting. The results of the difference analysis on variables at
the flare initial point are as shown in Table 3.
 An Analysis of Hard Landing Incidents Based on Flight QAR Data 403

Table 3. Difference analysis on variables of flare initial point

p(AN
Parameter Categories Variable Names Group Mean±SD OVA/K-
W)
Throttle Resolver Normal 49.277±1.786 0.069
Angle Hard 49.922±2.044
Normal 1.066±0.882 0.342
Control Column
Hard 0.930±0.616
Normal 2.111±1.010 0.620
Column Force
Hard 2.024±0.860
Normal 0.052±10.179 0.586
Control Wheel
Hard 0.954±7.233
Operation Parameter
Normal 0.001±0.462 0.624
Wheel Force
Hard -0.038±0.375
Normal 30.462±2.115 0.000
Flap Handle Position
Hard 32.963±4.609
Speed Brake Posi- Normal 2.991±0.888 0.540
tion Hard 2.898±0.740
Normal 0.570±0.314 0.747
Rudder Pedal
Hard 0.555±0.142
Normal 2.423±1.088 0.376
Elevator
Hard 2.576±0.720
Normal 1.383±2.124 0.441
Aileron
Hard 1.650±1.500
Normal 30.462±2.115 0.000
Flap
Configuration and Atti- Hard 32.963±4.609
tude Normal -0.134±0.693 0.600
Rudder
Hard -0.192±0.484
Normal 1.600±0.603 0.012
Pitch Angle
Hard 1.301±0.677
Normal -0.245±1.437 0.327
Roll Angle
Hard -0.466±0.894
Normal 148.923±4.748 0.259
Air Speed
Hard 147.907±5.003
Normal 145.815±7.104 0.461
Groundspeed
Hard 146.833±7.883
Flight Performance
Normal -803.077±121.718 0.638
Descent Rate
Hard -813.481±117.407
Vertical Accelera- Normal 1.051±0.040
0.298
tion Hard 1.044±0.033

For the normal landing and hard landing groups, there are only three variables
representing the significant difference at the level of 0.05, which are Flap Handle
Position, Flap and Pitch Angle.
404 L. Wang et al.

Fig. 2. Difference analysis of control column and throttle resolver angle

In Figure 2, the control column and throttle change greatly after passing 50 feet
(flare operation initial point). There is no difference between the control column of
the two groups ( F (1, 117) = 0.000, p = 0.998 ). There is also no difference found for
throttle operation before 50 feet ( F (1, 117) = 3.349, p < 0.07 ). The difference is
reflected after a flare starting when the pilot begins to decrease thrust.

3.3 Logistic Regression Model on Hard Landing

Table 4 shows the estimated parameters of the logistic model in predicting landing
incident type (hard landing or normal landing). Three predictors were included in the
final logistic regression model. The overall predictive percentage of the model was
72.6%, the sensitivity was 0.697and the specificity was 0.786.
As shown in Table 4, the Wald criteria indicated that Flap Handle Touchdown,
Pitch Angle Touchdown and Roll Angle Touchdown significantly contributed to the
occurrence of hard landings ( p < 0.01 ). Nagelkerke’s R of 0.677 indicated a relative-
2

ly strong relationship between predicting variables and hard landing.

 An Analysis of Hard Landing Incidents Based on Flight QAR Data 405

Table 4. Logistic regression values of the predicting variables

Predicting variables Wald ( χ 2 ) Adjust OR a 95% [Link] OR b

Flap Handle Touchdown 11.107** 1.172 1.074-1.296

Pitch Angle Touchdown 18.613** 0.531 0.393-0.713

Roll Angle Touchdown 15.984** 2.229 1.489-3.281
Constant 3.469# 0.058

** p < .01, * p < .05,# .05 < p < .10 and otherwise p ≥ .10.
a
Adjust ORs (odds ratio) predicted hard landing.
b
Confidence interval.

4 Discussion and Conclusions

In aviation safety research, the focus has typically been more on aviation accidents
where their occurrence rate has been decreased to quite a low level in most regions of
the world. However, unsafe incidents have often been ignored due to the difficulty in
obtaining and analyzing them in detail. Basing on flight QAR data, this study pro-
vided a new way to analyze unsafe incidents in the landing phase by considering a
history of individual instances recorded during flight. The main findings in this study
were concluded as following.

1. The results of multivariate analysis indicated that most flight parameter variables
with differences appeared in the stage of 50 feet to touchdown. Theoretically
speaking, many flight operations, including flares, need to be finished by pilots in
just a few seconds. While aircraft in low speed flight is sensitive to wind and other
weather factors, any small configuration changes during this stage could easily
complicate the decision of the proper action to take at the decision point. There-
fore, this phase is the most important operation stage and pilots should check the
ratio of descent rate and groundspeed carefully at the point of 50 feet.
2. Flare would reduce the aircraft’s descent rate to acceptable levels so that it settles
gently on the main landing gear, it would greatly influence vertical acceleration
through the two key factors of flare time and final flare pitch angle. The control
column and throttle operation would affect landing performance conjointly. Pilots’
quick and steady pulling up columns and softer throttle reduction are helpful for a
better flare operation and better landing performance.
3. The logistic model showed that the vertical load of touching ground was actually
linked with touchdown attitude and configuration closely, including three variables
of pitch angle, roll angle and flap degree. Among these, the pitch angle of the air-
craft is correlated with control column operation directly and therefore is a main
external indication of flare. As a matter of fact, the correlations between pitch an-
gle and vertical acceleration were strong at every stage from 200 to 0 feet.
4. These findings would be the basis of developing a mathematical and quantitative
model for further revealing the relationships between pilot operation and landing
406 L. Wang et al.

performance, which can also be applied in practice to prevent hard landing inci-
dents and even landing accidents.

Acknowledgments. We appreciate the support of this work from the National Natural
Science Foundation of China (No. 61304207, No.U1333112), the National Basic
Research Program of China (No.2010CB734105) and the Fundamental Research
Funds for the Central Universities (No. ZXH2012D001).

References
1. Hawkins, F.H.: Human Factors in Flight, 2nd edn. Ashgate. Brookfield, VT (1993)
2. Wickens, C.D., Hollands, J.G.: Engineering Psychology and Human Performance, 3rd edn.
Prentice Hall Press, Upper Saddle River (2000)
3. Harris, D.: The influence of human factors on operational efficiency. Aircraft Engineering
and Aerospace Technology 78(1), 20–25 (2006)
4. Shappell, S., Detwiler, C., Holcomb, K., Hackworth, C., Boquet, A., Wiegmann, D.: Hu-
man error and commercial aviation accidents: an analysis using the human factors analysis
and classification system. Human Factors 49(2), 227–242 (2007)
5. Sartor, P., Bond, D.A., Staszewski, W.J., Schmidt, R.K.: Value of an overload indication
system assessed through analysis of aviation occurrences. Journal of Aircraft 46(5),
1692–1705 (2009)
6. Nie, L., Shu, P., Huang, S., Wang, X.: Intelligent Diagnosis for Hard Landing of Aircraft
Based on SVM. China Safety Science Journal 19(7), 149–153 (2009)
7. Wang, X., Shu, P., Cao, L.: Incremental AHP based Weight Assessment for Risk Factors
of Civil Aircraft Landing Process. China Safety Science Journal 20(9), 112–115 (2010)
8. Rosa, M.A.V., Fernando, G.C., Gordún, L.M., Nieto, F.J.S.: The development of probabil-
istic models to estimate accident risk (due to runway overrun and landing undershoot)
applicable to the design and construction of runway safety areas. Safety Science 49(5),
633–650 (2011)
9. Molesworth, B., Wiggins, M.W., O’Hare, D.: Improving pilots’ risk assessment skills in
low-flying operations: The role of feedback and experience. Accident Analysis and Pre-
vention 38(5), 954–960 (2006)
10. Benbassat, D., Abramson, C.I.: Landing flare accident reports and pilot perception analy-
sis. International Journal of Aviation Psychology 12(2), 137–152 (2002)
11. Civil Aviation Administration of China. Implementation and Management of Flight Opera-
tion Quality Assurance. Advisory Circular: 121/135-FS-2012-45, CAAC, Beijing, China
(2012)
12. Qi, M., Shao, X., Chi, H.: Flight operations risk diagnosis method on Quick Access
Record exceedance. Journal of Beijing University of Aeronautics and Astronautics 37(10),
1207–1211 (2011)
13. Wang, L., Wu, C., Sun, R.: Pilot operating characteristics analysis of long landing based on
flight QAR data. In: Harris, D. (ed.) EPCE 2013, Part II. LNCS, vol. 8020, pp. 157–166.
Springer, Heidelberg (2013)
14. Sun, R., Han, W.: Analysis on parameters characteristics of flight exceedance events based
on distinction test. China Journal of Safety Science and Technology 7(2), 22–27 (2011)