Engineering Structures 158 (2018) 155–163

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Short communication

Engine to wing structural design under critical loads caused by a propeller T

blade loss
⁎
I. Armendáriza, , J. Olarreaa, J. García-Martínezb
a
E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid, Madrid, Spain
b
Materials and Structures Department, National Institute of Aerospace Techniques (INTA), Torrejón de Ardoz, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: To support the propeller blade loss loads in a turboprop aircraft the engine mounting system has to be carefully
Blade loss designed. The joints have to withstand the flight loads, however, during the blade loss event they have to fail
Pylon support design before the wing gets damaged. Using a FEM model and Monte Carlo simulation techniques the possible failure
Elastomeric failures sequences for a broad spectrum of flight conditions have been obtained, including different propeller fre-
Parameter uncertainties
quencies, blade loss sizes, angular positions where the blade is lost and also material properties. The structural
FEM
model includes non-linear behavior, damages, and several types of failure together with stochastic variables
Monte Carlo simulations
which can incorporate parameter uncertainties. Finally, the pylon to wing support is designed to guarantee, with
high level of confidence, no major hazard on the aircraft due to this dynamical phenomenon.

1. Introduction
Although structure performances have been greatly improved in the
last years, the huge increase in the demand of air traffic is responsible
for the fact that the rate of occurrence per airplane departure for pro-
pulsion system malfunction, or inappropriate crew response accidents,
has remained essentially constant for many years, as it is mentioned in
[1]. In aviation safety databases like [2–6] or [7] are reported more
than 644 engine occurrences since 1919 until July 2017 (powerloss,
fire, flame-out, fuel issues, propeller reverse pitch, simulated engine
failure, etc.). Specifically, they include the following occurrences: 64
uncontained engine failures, other 54 engine separations and at least 35
turboprop blade separations. One of the last accidents was on the 1st of
November of 2014 where a de Havilland DH-114 Heron aircraft made
an emergency landing following the in-flight separation of a turboprop
and the prop struck onto other engine causing substantial damage as it
is documented in [7]. Some months later, the 16th of April of 2015, a
Swearingen SA227-AC Metro III was substantially damaged after an
uncontained engine failure during the climb. A post-accident ex-
amination of the airplane (documented in [6]) revealed that a rotor
from the right engine had separated.
1.1. Turboprop blade loss phenomenon
Blade loss is a relatively common flight incident in turboprop air-
planes. As an example, the 25th of October of 2013 where a Fokker F-27
Friendship 500F was damaged, Rolls-Royce Dart 532-7 engine suffered
an uncontained failure. This resulted in the loss of the propeller and the
front part of the engine. Propeller blades sliced through the fuselage of
the airplane, exiting on the other side as it is documented in [6]. On one
hand, accidents in turboprops are less common due to the fact that the
engines have fewer moving parts than in turbofans. They offer greater
reliability, smoother operation and have longer time between over-
hauls. On the other hand, more hours before they have to stop for in-
spection implies to increase the probability of a blade loss due to un-
noticed crack growth. This type of cracks is deeply studied in [8] which
joined to the obvious safety implications. It confirms the importance of
thoroughly studying the blade loss event with all the parameters of
influence. The blade loss event probability is related to diverse factors
such as: blade material, blade structural design, time between over-
hauls, NDE techniques used…This parameter, of significant importance,
can be obtained by the manufacturer/operator of the aircraft.
Pylon to wing support for aircraft engines is designed to withstand
the loads produced during the normal flight operation without trans-
mitting excessive vibrations to the wing that could compromise the
structure. In normal conditions the largest loads occur in the thrust and
vertical directions. Nevertheless, some events of engine malfunction
may produce an imbalance and compromise the aircraft structure. In
most cases, the powerplant installation design makes that no single
failure or malfunction jeopardizes the safe operation of the airplane.
Each powerplant is isolated from the others and configured in order to
stop the rotation of any engine individually if necessary. An inoperative

⁎
Corresponding author.
E-mail address: armendarizbi@gmail.com (I. Armendáriz).

https://doi.org/10.1016/j.engstruct.2017.12.025
Received 12 August 2017; Accepted 12 December 2017
Available online 23 December 2017
0141-0296/ © 2017 Elsevier Ltd. All rights reserved.
I. Armendáriz et al.
engine does not constitute a safety issue since airplanes are designed to
fly under such circumstance. However, once the blade is lost a powerful
vibration is produced in the engine as it is reported in [9] and shown in
Fig. 8, but the consequences vary function of several parameters, being
the blade loss size one of major importance [10,11]. In order to bound
the dynamic loads transmitted from engine to the pylon and, finally, to
the wing the stiffness and strength of the elastomeric joints must be
thoroughly selected. They are designed so they fail to interrupt the load
path to the wing and, thus, avoiding the failure of any other part of the
airplane.
In order to study the phenomenon two important techniques must
be selected: an appropriate parametric analysis tool together with a
model order reduction technique. Among the wide variety and quantity
of these techniques, the Monte Carlo Technique (MCT) and the Craig-
Bampton reduction have been chosen regarding their compatibility,
versatile characteristics and simplicity [12–15].
This paper is structured as follows. After a brief description of the
structural model, the selection of the Monte Carlo Technique, the
parameters of influence (with their distribution functions) and the load
cases are presented. Afterwards, the results are analyzed and the key
parameters are identified. Finally, it is shown the procedure to obtain
with any desired confidence level the pylon to wing allowables. Along
these sections some clarifying examples are given.
2. Description of the model
The most severe failures occur when it is impossible to prevent se-
vere vibration transmission to the structure of the airplane. Most
common ones are propeller unbalance at assembly or a crack in the
propeller hub that can possibly result in propeller blade loss, to which
this paper is devoted. The engine mounting system (EMS) model must
absorb these vibrations and, if necessary, detach the engine from the
structure before fatal structural damages may occur. In order to simu-
late the dynamic behavior of the phenomena, two differentiated ele-
ments of the structure have been modeled: the engine and the engine
mounting system (also referred in the article as pylon). The hypothesis
of a rigid and fixed wing is used in the model for the boundary con-
ditions and limited mass and stiffness matrices are used, with special
attention to the joints.
The representative case that has been chosen is a heavy-duty mili-
tary transport turboprop. A draft of the structure is shown in Fig. 1 and
more detailed descriptions of the models can be found in [11,16],
where different hypotheses for the simulation are considered and
compared in detail.
Fig. 1. EMS schematic sketch: EMS to engine and EMS to wing joints.
156
Engineering Structures 158 (2018) 155–163
2.1. Engine
The selected model for the engine has two important characteristics:
it is reliable (it has been verified and checked, see [11,16]) and with a
reduced number of degrees of freedom.
A highly dynamic phenomenon requires a model with mass, inertia
and eigenmodes that matches the real structure. It is also necessary the
non-linear behavior of the six elastomeric devices, which has been
adequately modeled: linear stiffness in shear and non-linear in com-
pression/tension, and different failure criteria depending on the axis. A
draft of their location and position on the structure is shown in Fig. 1.
On the other hand, the Monte Carlo technique that has been used for
the analyzes requires low order models, otherwise it would have a
prohibitive computational demand. The Craig-Bampton reduction has
been chosen in order to fulfill this requirement [12].
2.2. Engine Mounting System (EMS)
This structural element connects the engine to the wing and,
therefore, supports and transmits the forces that reach the wing.
Additionally, when this structural element is not considered in the
blade loss simulation the failure results are different, as shown in [11].
The engine joint is designed through the elastomeric devices, while
the attachment to the wing (also referred as pylon to wing support)
consists of four points: two connected to the forward part of the wing in
X and Z directions, and other two to the rear part of the wing, of which
one support loads in Y and the other in Z (see Fig. 1). Therefore, this
attachment can be designed through the comparison of these loads with
the corresponding allowables. Fig. 2 shows the volume (inside red box)
in which no allowable (Fx1 and Fz1,Fx 2 and Fz 2,Fy3 and Fz4 ) has been
reached, and, therefore, there is no structural failure. It also shows an
example of the forces (blue line) and the required allowables (green
box) to support the load case.
2.3. FEM analysis considerations
Different schemes are considered for the integration of the dynamics
of the problem (Eq. (1)). In every case, the time step is a key parameter
which is obtained in relation with the highest eigenmode required and
Fig. 2. Pylon to wing force evolution (blue line), its maximum values (green box) and the
allowable limits (red box). (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
I. Armendáriz et al.
the stability of the scheme.
The β-Newmark schema has been chosen as the most appropriate. It
gathers three convenient characteristics: speed, stability and robust-
ness. A detailed comparison of the different schemes performance is
shown in [16].
My″ + Cy′ + Ky = F → My″ (t ) + Cy′ (t ) + Ky (t ) = Fnonlinear (y,t )
A proportional structural damping is applied to the model, being the
damping coefficient an input variable, as specified in Section 3. Al-
though the damping is often considered proportional it is not an easy
task. This methodology (MCT with damping as an input) has the ad-
vantages of determining the influence of this damping in the response.
The final results are more accurate and useful with the usual determi-
nistic procedures.
The blade loss unbalance is a load (summarized in Eq. (2)), which is
dependent of some input variables considered in Section 3: blade loss
size () and propeller rotational frequency (Fc and Moop ) and propeller
rotational frequency (ω).
Fy = −Fccos (ωt ) Fz = −Fcsin (ωt )
My = Moopsin (ωt ) Mz = −Moopcos (ωt )
Finally, the kinetic and potential energy levels and balances have
been controlled during the simulations in order to check the accuracy of
the selected integration schemes and to verify the failure sequence. The
blade loss event means an increase of energy in the engine, mostly ki-
netic. This kinetic energy is partly transformed to potential energy in
the elastomeric devices and partly transferred to the wing as kinetic
energy. As the potential energy is later returned to the structural
system, the properties of the elastomers are the key design parameters
to bound the loads that reach the wing. In this energy exchange process
two irreversible effects are simulated: the structural damping and the
elastomer failures. A simplified blade loss simulation is represented in
Fig. 8.
Therefore, when an element fails this exchangeable energy becomes
irreversible, and this energy path is no longer possible. Those events
make each load case different, the simulation progresses and separates
each one from the others. Finally, as all the elastomeric elements fail
the EMS and the engine separate from each other. The engine continues
vibrating violently due to the instability and the EMS vibrates with an
almost constant level of energy (exchanging kinetic and elastic energy).
This level of energy is always diminished smoothly when the structural
damping is considered.
3. Load cases analyzed
Several load cases are analyzed by modifying the initial conditions
(the default values in Table 1). Table 1 summarizes the range and
distributive functions that has been considered. These are characteristic
parameters involved in the blade loss phenomena. The angular position
of the blade (ANG), the propeller rotational speed (RPM), the blade loss
size ratio (SIZE) the elastomeric devices stiffness ratio (STIFF) and
strength ratio (STR), and finally, the structural damping (DAMP) are
the parameters varied in this work. A more detailed description of the
parameters can be found in [11,16].
Table 1
Properties of the input variables.
Variable Variation type Interval
Blade loss size (SIZE) Uniform (0, 1)
Elastomeric stiffness (STIFF) Normal N(0, 1)
Elastomeric strength (STR) Normal N(0, 1)
Sensitivity to rpm (RPM) Uniform (69, 88) [rad/s]
Structural damping (DAMP) Uniform (0%, 5%)
Blade loss angular position (ANG) Uniform (0°, 360°)
Engineering Structures 158 (2018) 155–163
3.1. MCT input
The Monte Carlo Technique has been chosen in order to obtain an
statistical approach to the blade loss process. It depends on several
parameters and some of them must be considered from an stochastic
point of view. They are subjected to non-deterministic fluctuations and
(1)
a deterministic approach would lead to incomplete and inaccurate re-
sults.
These inputs and their probability distribution are listed in Table 1.
The selected number of simulations is always between 1000 and
10,000. This technique allows versatile input parameter configuration,
so the influence of each individual parameter and their interaction can
be studied.
4. Results
MCT permits a versatile selection of the input parameters, which
allows the correlation study of individual parameters and the inter-
dependence of two or more parameters together. These results are
shown in detail in [11], while here it is examined the parameters re-
(2) lation with the failure sequence and the pylon to wing allowables. As a
summary, Table 2 and Fig. 4 present the results function of the input
variation. The analysis reveals that the impact of each parametric
variation over the results is different; the standard deviations in Table 2
show their order of influence. The evolution of the analysis when
changing the blade loss size can be much different than when changing
the elastomeric stiffness or the engine frequency. The elastomeric
strength, the structural damping and the angular position of the blade
when it is detached are in a medium level of parametrical influence.
Specifically, Table 2 presents the results of the next variables: the
first failure time (FFT), which is the time of the first elastomeric failure,
and the last failure time (LFT), in which the last remaining elastomer
breaks and, therefore, the engine is completely detached from the air-
craft. The difference between these times (referred as FLT in the fol-
lowing) gives the total time from the beginning of the process to the
engine detachment. It has also been considered: the first failure energy
(FFE), which is the energy of the system when the first elastomer
breaks, the last failure energy (LFE), which is the energy of the system
when the last elastomer breaks, the maximum wing X, Y & Z sum forces
(MWX, MWY & MWZ), which are the maximum X, Y & Z forces that
reach the wing during the simulation. Precisely, Table 2 shows means
and standard deviations of the most important variables when varying
the first column parameters. PAF is the blade loss angle in the propeller
plane when the first elastomer breaks, and PAL is the blade loss angle in
the propeller plane when the last elastomer breaks.
Additionally, the correlation of the failure probability with the
pylon to wing allowables is studied in the next sections.
4.1. Failure sequence results
These are the results of a deep and detailed study where all the
possible failure sequences between the engine and the pylon are con-
sidered (6 elastomeric joints). A six-number figure has been used to
refer the failure sequence. The first number (from 1 to 6) indicates the
order in which the TOP + attachment fails. The second number (from 1
to 6) designates the order in which the TOP- elastomer breaks.
Consequently, the third, fourth, fifth and sixth number (from 1 to 6)
indicate the order in which the SIDE+, SIDE-, REAR + and REAR-
elastomer breaks respectively. Therefore, if the sequence number is
Value
123456 means that the chronological failure sequence is TOP+, TOP-,
1 SIDE+, SIDE-, REAR + y REAR + and REAR-. Table 3 and Fig. 3 in-
1 dicate the failure sequence rate of occurrence when all input para-
1 meters are combined (see Table 1). In Fig. 3 only the occurrence rate of
ωdefault
the most probable failure sequences (above 3%) are shown.
0
0° An important result that can be extracted from this analysis is that
from 720 possible failure sequences only 46 occur. Next, the most
157
I. Armendáriz et al. Engineering Structures 158 (2018) 155–163

Table 2
Output parameters properties.

LOAD CASES FFT [s] LFT [s] FLT [s] FFE [J] LFE [J] MWX [N] MWY [N] MWZ [N] PAF [°] PAL [°]

SIZE μ 0.0442 0.1266 0.0824 7930 13,119 313,102 243,175 253,614 θini 170 0
(E + EMS ) σ 0.0228 0.1000 0.0797 3915 5582 109,562 101,409 154,733 θfin 180 10

RPM μ 0.0278 0.0669 0.0391 13,141 21,503 223,687 197,536 143,490 θini 340 0
(E + EMS ) σ 0.0001 0.0013 0.0012 252 2138 5383 6205 9767 θfin 170 360

ANG μ 0.0306 0.0757 0.0450 13,712 21,660 340,466 201,073 257,747 θini 170 150
(E + EMS ) σ 0.0015 0.0129 0.0128 922 2677 89,630 49,981 86,664 θfin 330 170

STIFF μ 0.0281 0.0658 0.0377 12,823 18,765 263,458 208,001 147,308 θini 160 180
(E + EMS ) σ 0.0007 0.0051 0.0050 218 1646 43,411 27,714 40,731 θfin 170 190

STR μ 0.0275 0.0742 0.0467 12,652 19,697 255,455 200,416 182,289 θini 170 0/160
(E + EMS ) σ 0.0008 0.0165 0.0162 316 3375 52,226 26,546 92,090 θfin 170 50/200

DAMP μ 0.0286 0.0945 0.0659 11,810 19,175 224,477 168,536 236,314 θini 170 10
(E + EMS ) σ 0.0010 0.0146 0.0142 661 1860 33,505 11,052 69,636 θfin 240 50

SIZE & ANG μ 0.0443 0.1317 0.0874 8428 14,706 357,928 242,325 313,238 θini 0 0
(E + EMS ) σ 0.0218 0.0946 0.0749 4404 6464 161,059 92,356 241,025 θfin 360 360

SIZE & RPM μ 0.0440 0.1220 0.0780 8209 14,510 289,338 235,145 233,283 θini 0 0
(E + EMS ) σ 0.0234 0.0899 0.0692 4096 6561 128,034 99,520 194,100 θfin 360 360

SIZE & STIFF μ 0.0447 0.1260 0.0813 8136 13,639 314,920 244,503 249,958 θini 170 350
(E + EMS ) σ 0.0244 0.0989 0.0766 4071 5820 120,768 100,411 163,437 θfin 180 30

SIZE & STR μ 0.0445 0.1248 0.0802 7989 13,721 328,886 252,132 265,176 θini 160 350
(E + EMS ) σ 0.0234 0.0934 0.0730 3983 6414 126,674 100,045 172,826 θfin 170 10

SIZE & DAMP μ 0.0462 0.1713 0.1251 7615 12,792 338,488 202,738 400,658 θini 0 0
(E + EMS ) σ 0.0249 0.1657 0.1439 3811 5883 260,376 80,312 471,957 θfin 360 30

ALL μ 0.0458 0.1600 0.1142 8130 14,053 364,870 198,296 414,076 θini 0 0
(E + EMS ) (∗Fig. 4) σ 0.0230 0.1406 0.1203 4266 6416 270,579 76,536 478,869 θfin 360 360

( μ = Mean, σ = Standarddeviation , [θini,θfin] = Angularintervalpronetofail , E = Engine , EMS = EngineMountingSystem and (∗number) = Figure number where the results are shown).

Table 3
Table of the failure sequence occurrence rate for ALL parameter variations.

SEQUENCE ALL SEQUENCE ALL

123456 1.06% 123465 0.45%

124356 1.28% 124365 0.35%
132456 0.30% 132465 0.21%
134256 0.25% 134265 0.17%
142356 0.04% 142365 0.01%
143256 0.35% 143265 0.13%
213456 3.03% 213465 1.19%
214356 6. 56% 214365 1.66%
231456 1.76% 231465 0.36%
234156 12. 54% 234165 4.38%
241356 4.65% 241365 0.93%
243156 14. 30% 243165 4.38%
312456 0.64% 312465 0.16%
314256 0.02% 321456 0.54%
321465 0.38% 324156 0.78% Fig. 3. Failure sequence rate of occurrence for ALL parameter variations.
324165 0.39% 341256 2.19%
341265 0.75% 342156 9. 38%
342165 3.69% 412356 0.08% probable failure sequences 214356 (6.6%), 234156 (12.5%), 243156
412365 0.00% 413256 0.12% (14.3%), 342156 (9.4%) and 432156 (11.8%) are named as groups 1, 2,
421356 0.12% 421365 0.10% 3, 4 and 5 respectively. The group 6 refers to all the other sequences
423156 1.16% 423165 0.31%
among the 46 possible.
431256 2.65% 431265 0.89%
432156 11. 82% 432165 3.57% The results in Figs. 5 and 6 show that almost in all the simulations,
no matter the input variation, there are certain patterns:

• SIZE variation. Sequences 4 and 5 are the most probable. Sequence 4

occurs for extreme values of the blade loss ratio >0.75 or <0.3

158
I. Armendáriz et al. Engineering Structures 158 (2018) 155–163

Fig. 4. Results comparing input vs. input, input vs. output and output vs. output variables for ALL parameter variations.

Fig. 5. Several results of different input parameter variations vs. a certain output variable, the failure sequence is also displayed.

(almost all and none of the blade). The same happens for sequence 2
which occurs mostly for ratios >0.5 (see Fig. 6).
• RPM variation. There are no 1, 2 and 3 sequences, and sequence 4
takes place only when the frequency is 75–80 [rad/s]. Sequence 5
appears just between 73–75 [rad/s] (see Fig. 5). Even though there
are lower and greater frequencies it is only when sequence 4 occurs
when the maximum Fy (MWY) reaches a maximum. It also happens
with maximum Fx force (MWX), however, in this case it is not ex-
clusively for that sequence.
• ANG variation. There are no clear patterns and it is difficult to reach
a conclusion. Sequence 2 produces the very last elastomer failure
(LFT) compare to the last failure of other sequences. The same
happens to FLT (see Fig. 5).
• STIFF variation. Apart from 6, sequences 4 and 5 are the most

159
I. Armendáriz et al. Engineering Structures 158 (2018) 155–163

Fig. 6. Failure sequences in different input parameters vs. SIZE parameter.

probable. In sequence 2 and 5 the first elastomer failure takes place An important result that must be emphasized is that the maximum
in the very beginning of the phenomenon. The opposite occurs in force in Y axis (MWY) exceeds the allowable limit in most of the si-
sequence 4 (see Fig. 5). mulations. A simulation case has been graphed in Fig. 2, where the
• STR variation. Aside from 6, sequences 4 and 5 are the only that take maximum force of each axis (green box) are the variables MWX, MWY
place. Precisely, sequence 4 occurs when STR values between 1–1.1 & MWZ. The allowable limits are graphed in Fig. 2 (red box), and it is
(see Fig. 5). Sequence 4 produces medium range of values (not ex- clearly showed that it is only in MWX and MWZ where the values do not
treme) of the OUTPUT parameters. exceed the limits. On the other hand, MWY, and therefore, Fy force
• DAMP variation. Sequences 4 and 5 are the most probable. Sequence exceeds the limits.
1 does not take place. Sequences 2 and 5 make earlier the failure of Designers are interested in lower failure probabilities and this
the last elastomer (lower LFT). The same happens with the time methodology is able to analyze this aspect. The failure probabilities in
between the first and last elastomer failure (FLT), it is shorter for each direction have been calculated for the different groups of input
sequence 2 and 5. Sequence 2 is the sequence which produces less parameter variations in order to study their impacts. Therefore, Table 4
maxima forces at the wing (MWX, MWY and MWZ). shows the failure probabilities for different combinations of allowable
values.
If the input parameters that are varied are more than one the results The failure probability of the pylon to wing attachments can be
can be, or not, very different. For example, SIZE&ANG (variation of calculated with this methodology for a certain allowable limit. For
SIZE and ANG parameters, the rest are fixed at their default values) do example, results for Fx = 600 [kN], Fy = 100 [kN] and Fz = 600 [kN],
not show extra information apart from the conclusions already men- which are characteristic values for these type of engine, are showed in
tioned. However, SIZE&RPM, SIZE&STIFF, SIZE&STR y SIZE&DAMP Table 4 (the first 3 columns).
show clear patterns as shown in Fig. 6. Each sequence takes place more If the allowables are Fx = 600 [kN], Fy = 400 [kN] and Fz = 800 [kN],
or less at the same range of SIZE ratio, which is the most important and
influential parameter. The differences depend on the other parametrical
variations which modify slightly the failure sequence areas (see Fig. 6, Table 4
Attachment failure probability for each group of load cases.
in which the different colors are in the same areas).
Fx [kN] Fy [kN] Fz [kN] Fx [kN] Fy [kN] Fz [kN]
600 100 600 600 400 800
4.2. Pylon to wing attachments results
SIZE 1.3% 100.0% 2.5% 1.3% 8.3% 1.0%
The maximum forces on the wing (MWX, MWY & MWZ) are easily RPM 0.0% 100.0% 0.0% 0.0% 0.0% 0.0%
ANG 0.7% 98.2% 0.0% 0.7% 0.1% 0.0%
determined due to the FEM model configuration. As it was underlined
STIFF 0.0% 100.0% 0.0% 0.0% 0.0% 0.0%
before, the introduction of the EMS in the simulation affects the results STR 0.0% 100.0% 0.0% 0.0% 0.0% 0.0%
in several ways: it delays the failure of elastomers in time, it shortens DAMP 0.0% 100.0% 0.0% 0.0% 0.0% 0.0%
the time lapse between the first and the last failure, the increment of SIZE & ANG 5.5% 99.0% 6.5% 5.5% 7.5% 3.7%
energy during the event is slower and the angles prone to fail can in- SIZE & RPM 2.1% 98.4% 2.7% 2.1% 7.5% 1.7%
SIZE & STIFF 2.4% 98.9% 2.4% 2.4% 8.9% 1.4%
cidentally change (PAF & PAL). Another result that can be extracted is
SIZE & DAMP 8.4% 92.2% 11.1% 8.4% 2.1% 9.4%
that the slope of the total energy of the model is lower when the pylon is ALL 9.8% 97.0% 12.9% 9.8% 2.2% 9.7%
introduced in the simulation.

160
I. Armendáriz et al.
Fig. 7. Level of confidence vs. force allowables.
the probability to reach them (failure probability) is lower than 10%,
no matter which parameter is varied. Precisely, when all variables are
varied (ALL) the probability of failure of each axis are 9.84%, 2.2% and
9.7% respectively. With this methodology the allowables can be opti-
mized to any desired level of confidence. For instance, with
Fx = 600 [kN], Fy = 300 [kN] and Fz = 800 [kN] as allowables the prob-
ability of failure still lower than 10% (9.84%, 9.7% and 9.7% respec-
tively).
4.2.1. Pylon to wing attachments design
When a wing attachment is being designed, the question that has to
be answered is which allowables are needed to make it safe enough to
withstand all the life-time forces with high level of confidence. As it has
been mentioned before the results show that in most of the simulations
the maximum force in Y axis (MWY) exceeds the allowable limit. How
much should that allowable be increased to withstand a blade loss? To
answer this question, Fig. 7 shows the probability of non-failure func-
tion of the allowable. Moreover, a structural modification in order to
achieve the required values of strength along Y axis would not imply a
noticeable mass increment because longerons can withstand this load
with minimal structural modifications (the current pylon to wing joints
design derives of the fact that this Y axis load is not relevant in most of
the usual load cases for the wing and engine but it is for the blade loss
event).
If only the angle varies (ANG) the value of the allowables to with-
stand the event must be higher than 100 kN for Fx , Fy and Fz . If the
values are 456 kN, 268 kN and 365 kN for Fx , Fy and Fz , respectively, in
90% of the events the wing will not suffer any damage. This percentage
can be increased to 95% or even 99% when the values Fx , Fy and Fz
161
Engineering Structures 158 (2018) 155–163
allowables are 572 kN, 329 kN and 519 kN respectively. In this case the
pylon to wing structure supports the event no matter which is the angle
when the blade is lost.
This analysis can be carried out for all the input parameters and
there are different evolutions of the curves. All of them start to increase
(probability of failure start to be less than 100%) for allowables around
100 kN in all directions. However, STR, RPM, DAMP and STIFF varia-
tions increase them to higher values. Normally, Z axis probability of
failure decreases first, which is a curve increase in Fig. 7 (at approxi-
mately Fz allowable of 100 kN). Secondly it happens to Fy (180 kN ap-
proximately) and lastly Fx (200 kN approximately). This means that
those are the values that start to make the wing safe from the blade loss
event. In the case of taking into consideration ANG and SIZE variations,
the decrement of the failure probability is smoother, and therefore, the
allowable values to reach 90%, 95% or 99% level of confidence are
considerably higher. Table 5 shows these allowable values.
Finally, if all the parameters are free (ALL) the evolution of the
curve are as it is shown in last chart of Fig. 7. The percentage increment
of the confidence level is even lower than the previous cases, however,
it is comparable with SIZE and SIZE & other parameter variation. In
case of SIZE&ANG, SIZE&DAMP and ALL to withstand the 99% of the
blade loss events the Fx and Fz allowables has to be higher than
1000 kN. Therefore and according to Table 5, if the allowable values are
590 kN, 299 kN and 787 kN for Fx , Fy and Fz , only 10% of flights which
lose a blade would suffer wing damages (which may produce a major
hazard).
I. Armendáriz et al. Engineering Structures 158 (2018) 155–163

Fig. 8. Schematic evolution of the blade loss event.

5. Conclusions the end of the phenomenon. It has been included non-linear behavior as
well as failure criteria to reproduce accurately the event. Even though
The proposed FEM model is suitable to simulate the blade-loss the normal modes of the structure change when a soft mount fails and
collapse sequence. Moreover, it is able to run without instabilities until so does the dynamic behavior of the system. The elastomeric model is

162
I. Armendáriz et al. Engineering Structures 158 (2018) 155–163

Table 5
Allowable design values for different safety criteria.

ALLOWABLE Fx [kN] Fy [kN] Fz [kN]

Realiabilityfactor 90% 95% 99% 90% 95% 99% 90% 95% 99%

ANG 456 501 572 268 279 329 365 425 519
RPM 229 230 231 206 206 207 157 158 160
SIZE 471 498 617 381 450 545 407 456 774
STIFF 338 361 362 242 257 273 194 210 422
STR 366 369 370 246 249 256 355 362 366
DAMP 278 304 309 184 193 199 302 339 344
SIZE & ANG 524 606 >1000 378 446 535 516 692 >1000
SIZE & RPM 410 482 769 380 435 534 357 449 998
SIZE & STIFF 465 525 705 393 447 525 402 461 903
SIZE & STR 219 267 553 160 202 307 249 332 731
SIZE & DAMP 509 931 >1000 312 349 442 689 >1000 >1000
ALL 590 920 >1000 299 344 463 787 >1000 >1000

able to capture the sequence of collapsing. online version, at http://dx.doi.org/10.1016/j.engstruct.2017.12.025.

The results confirm the expected and remarkably dependence on the
blade loss size, in a second order of influence on the elastomeric
strength, the structural damping and the angular position of the blade
loss, and almost irrelevantly on the elastomeric stiffness and the engine
frequency. The analyzes have been performed using distinct strategies,
distinct schemes and a statistical methodology (Monte Carlo tech-
nique). The model is able to study the range of each parameter where
the simulation behaves similarly. These results are easily presented
through colored figures. And not only the range of each parameter have
been identified but also the specific combinations of input parameters
that lead to certain result. Consequently, when there are only small
changes in the initial variables the research identifies which parameters
are responsible for significant differences in the development of the
simulation and why. Therefore, the most important parameters that can
change hazardously the behavior of the structure are identified.
Moreover, the results determine how long does it take to detach the
engine and which are the consequences to the wing during this interval.
Furthermore, the blade loss phenomenon probably damages the wing
along Y axis (with the initial allowable limits for the pylon to wing
attachments) due to the fact that they are not designed to support this
type of load. It has been suggested a structural modification in order to
achieve the required values of strength along Y axis. In addition, a
specific design method has been carried out in order to optimize al-
lowable values which withstand the blade loss event with a high level of
confidence. Thus, an optimal design of the pylon to wing attachment is
achieved with any desired confidence level.
Acknowledgements
The article was completed thanks to the National Institute of
Aerospace Techniques (INTA) and the Universidad Politécnica de
Madrid (UPM). Moreover, it is important to underline also the colla-
borative relations with Airbus Military.
References
[1] Sallee G, Gibbons DM. Propulsion system malfunction plus inappropriate crew re-
sponse (psm + icr). Tech rep. AIA/ AECMA; 1998.
[2] Accident and incident data. Tech rep. Federal Aviation Administration; Last up-
dated: July 27th 2017. < http://www.faa.gov/data_research/accident_incident/
> [accessed July 31st, 2017].
[3] Aviation accident database and synopses. Tech rep. National Transport Safety
Board; Last updated: July 27th 2017. < http://www.ntsb.gov/_layouts/ntsb.
aviation/index.aspx > [accessed July 31st, 2017].
[4] Bureau of aircraft accidents archives. Tech rep. Bureau of Aircraft Accidents
Archives; Last updated: July 13 th 2017. < http://www.baaa-acro.com/advanced-
search-result/ > [accessed July 31st, 2017].
[5] Incidents and news in aviation. Tech rep. The Aviation Herald; Last updated: July
29th 2017. < http://http://www.avherald.com/ > [accessed July 31st, 2017].
[6] Asn aviation safety database »airplane - engines. Tech rep. Flight Safety Foundation;
Last updated: July 30 th 2017. < http://aviation-safety.net/database/events/event.
php?code=AC > [accessed July 31st, 2017.
[7] Olguin JR. 2015 accidents and incidents. Tech rep. Instituto Boliviano de Historia
Aeronáutica, Latin American Aviation Historical Society and Air-Britain (Historians)
Ltd.; Last updated: 07 October 2015. < http://www.aviacionboliviana.net/civil/
accidentes/index.htm > [accessed August 18, 2016].
[8] In-flight separation of propeller blade results in uncontrolled descent and fatal ac-
cident to twin-turboprop commuter aircraft. Accident prevention 1997; 54(2).
[9] Sinha S. Rotordynamic analysis of asymmetric turbofan rotor due to fan blade-loss
event with contact-impact rub loads. Sound Vib 2013;332:2253–83.
[10] Kushan MC, Diltemiz SF, Sackesen I. Failure analysis of an aircraft propeller. Eng
Fail Anal 2007;14:1693–700.
[11] Armendáriz I, Olarrea J, García-Martínez J. Parametric analysis of a highly dyna-
mical phenomena caused by a propeller blade loss. Eng Fail Anal 2015;57:528–43.
http://dx.doi.org/10.1016/j.engfailanal.2015.08.011.
[12] Craig RR, Bampton MCC. Coupling of substructures for dynamic analyses. AIAA J
1968;6(7):1313–9.
[13] Rixen DJ. A dual craig-bampton method for dynamic substructuring. J Comput Appl
Math 2004;168:383–91.
[14] Nolde N. The analysis of extremes in multivariate models. Ph.D. thesis. M.Sc.
Statistics, Simon Fraser University. ETH ZURICH; 2010.
[15] Zu-Qing Qu. Model order reduction techniques with applications in finite element
analysis, vol. XVI; 2004.
[16] Armendáriz I, López J, Olarrea J, Oliver M, Climent H. Case study: analysis of the
response of an aircraft structure caused by a propeller blade loss. Eng Fail Anal
2014;37:12–28. http://dx.doi.org/10.1016/j.engfailanal.2013.11.0.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, in the

163