Professional Documents
Culture Documents
net/publication/303543353
CITATIONS READS
2 237
3 authors:
Dimitri N. Mavris
Georgia Institute of Technology
1,305 PUBLICATIONS 8,662 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Alek Gavrilovski on 22 August 2018.
Dimitri N. Mavris
S.P. Langley NIA Distinguished Regents Professor
Director of the Aerospace Systems Design Laboratory, ASDL
Georgia Institute of Technology
Atlanta, GA, United States of America
ABSTRACT
Helicopter Flight Operations Quality Assurance (HFOQA) systems promise safety improvements in flight operations
through the use of on-board data from regular flights. HFOQA systems can provide data pertaining to many types of
accidents where human factors have been implicated because they track the manner in which the vehicles are operated.
For helicopters, most implementations of such systems on helicopters rely on experts to determine pre-set limits on
combinations of flight parameters. These limits are also known as “safety events”. A common practical problem that
arises in HFOQA systems is the need to have sufficient knowledge of a condition before events can be defined and used
in a proactive manner. There has been recent interest in using alternative approaches to detecting faults and unsafe
events in aviation and to solve this inherent limitation of HFOQA. In this work, a model-based approach is taken in
an effort to extend the capabilities of traditional HFOQA analysis, particularly in terms of definition and detection
of monitored conditions. For localized conditions, the use of simple models in place of traditional safety events
is investigated and demonstrated. The detection based on the model evaluation shows a good correspondence to the
flight condition it was designed to capture, and at the same time incurring a minimal amount of additional computation.
The model-based boundaries typically account for changes in vehicle parameters and operating conditions, whereas
traditional safety events have to be modified through an iterative process. For a more general approach, a dynamic
model was considered. By evaluating the vehicle’s response to a set of control inputs spanning a range about the trim
state, it was possible to determine the boundaries of safe input. The inputs falling outside this boundary were also
evaluated for the risk associated with them based on the time required to reach a critical state. In both cases, the results
offer improvements over the current state-of-practice, and can be deployed in a data-monitoring system directly or
following intermediate post-processing steps.
INTRODUCTION AND BACKGROUND ational experiences have demonstrated the potential for safety
improvement through HFOQA, as well as some additional op-
Increasing helicopter safety is an on-going concern and de- erational,maintenance and regulatory benefits (Refs. 1–5) The
spite recent successes, further efforts are needed to sustain the present work is concerned with the application of rotorcraft
improvements and continue the positive trend. Helicopters simulation to enable improved detection and analysis capa-
have traditionally been associated with greater risks,especially bilities in HFOQA by analyzing critical scenarios in a virtual
due to a high number of critical dynamic components whose environment. The result of the simulation studies are poten-
failure would likely result in an accident. In recent surveys tial improvements in the conditions that can be tracked, earlier
of rotorcraft accidents, the largest category of accidents was detection and more straightforward development of condition
found to be related to pilot actions and decision making, as indicators (safety events) that comprise most of the analytic
shown in Table 1. This fact makes Helicopter Flight Opera- capability in HFOQA.
tions Quality Assurance(HFOQA) particularly useful in im-
HFOQA is known under many names, depending on the
proving rotorcraft safety. Several studies and reports of oper-
purpose of the system and the local regulatory climate where
Presented at the AHS 72nd Annual Forum, West Palm Beach, it is implemented. It is often referred to as Helicopter Flight
Florida, May 17–19, 2016. Copyright
c 2016 by the Ameri- Data Monitoring (HFDM) as well. The purpose of such a sys-
can Helicopter Society International, Inc. All rights reserved. tem is to record and store flight data, and to enable post-flight
1
tion systems usually pertains to the kinematics, dynamics and
Table 1. Rate of crew-related accidents performance of the monitored vehicle or fleet. This creates
number of % pilot some overlap with the next type of improvement to monitor-
Source Years
accidents action ing systems. Model-based approaches typically make use of
Iseler & De some characteristic of the monitored system which can be ac-
1990-1996 756a 70
Maio6 curately and adequately described using mathematical formu-
Harris, Franklin 1963-1998 lations, which yield insights beyond what is already contained
7618a 64
& Kasper7 in the data. Marzat (Ref. 18) makes a great case for the ben-
EHEST8 2000-2005 311 70 efits of model-based system monitoring, as well as provides
2001-2002, an extensive overview of different approaches for a variety of
US JHSAT9 2006
523 84 aerospace systems.
a Number reflects single-engine helicopters Figure 1 shows an overall concept of the operation of a
HFOQA system. The vehicles generate data during regular
analysis that would identify potential unsafe trends. For flight
operations, which is typically analyzed for the presence of
critical components, the industry has specified a rate of 10−9
safety events through the use of expert-defined thresholds (top
failures per flight hour. Monitoring of flight critical compo-
left of figure). In practice, these safety events are most often
nents using Health and Usage Monitoring Systems (HUMS)
combinations of values on a subset of the recorded parame-
has contributed to the low accident rate due to system fail-
ters, which limits the analysis only to the available parameters
ures (Ref. 10). In comparison, pilot-related accidents in ro-
and the pre-defined events. Models can be used alongside the
torcraft operations occur at a rate of 10−5 to 10−6 per flight
existing data to provide derived parameters which are much
hour (Ref. 11). For this reason, HFOQA systems can pro-
more closely related to the hazard that is being monitored.
vide benefits by gathering meaningful information pertaining
Other benefits include the possibility to derive additional pa-
to many types of accidents where human factors have been im-
rameters which are not contained in the available data set,
plicated because they track the manner in which the vehicles
but may be related to the recorded data through some phys-
are operated. HFOQA systems promise safety improvements
ical principles. For example, in civil aviation researchers have
in flight operations through the use of on-board data from reg-
considered the risk of runway overrun by combining the land-
ular flights.
ing point data with a braking model of the aircraft, which al-
Most implementations of such systems on helicopters rely lows for a comparison between the remaining runway length
on expertise or prior experience to determine pre-set limits and the braking ability of the plane (Ref. 19). In this con-
on combinations of flight parameters. These limits are also ceptual framework, the models can contribute through crew
known as “safety events”. HFOQA operates primarily by feedback as well as direct implementation for the purpose of
comparing these thresholds to incoming data, and alerting the detecting hazards in the flight data.
operator when a pre-defined level of a certain parameter is
breached. In the present state of practice, variability is in-
troduced due to the subjective margin of safety which is as- TECHNICAL APPROACH
sumed in each safety event definition. Another concern is Safety conditions of interest
the tradeoff between false alerts and missed detection, which
operators address through periodic fine tuning of the safety In this work, we consider several conditions which have con-
events which are tracked. Because of the subjectivity associ- tributed to many helicopter accidents, particularly loss of con-
ated with such a process and the fact that some conditions are trol accidents. Studies of helicopter accidents (Refs. 1,9) have
simply too rare to accurately determine where the detection revealed the main categories of accidents to be related to the
level should be set, it is quite often the case that these systems pilot’s operation of the vehicle. The goal is to address the
must go through many iterations before acceptable levels of more significant areas of concern first, and the measure of the
performance are achieved (Refs. 4, 12). significance of an accident type has historically been the num-
Two general directions of research for improving monitor- ber of such occurrences over a period of time. As discussed
ing systems are toward data-based and model-based methods in (Ref. 20), the accident rate as a metric is not ideal but pro-
to extract additional knowledge from flight data, beyond the vides a direction for the focusing of development efforts on
traditional metrics. more proactive metrics. The industry is in general agreement
Data-based methods encompass a rapidly expanding field about the types of safety concerns that helicopters are most
of machine learning, data mining, and other statistical tech- commonly exposed to, and these known hazards provide the
niques which make use of the vast amount of data gath- starting point for this investigation.
ered (Ref. 13). The ability to analyze a large set of flight Of particular interest is the management of performance
data records and extract anomalous or otherwise non-standard during normal operation and during emergency procedures
events has seen much attention in recent years (Refs. 14–17), such as autorotation, dynamic rollover, loss of trail rotor ef-
with very positive results. In most cases, the application of fectiveness and entry into the vortex ring state, among others.
a data-driven algorithm requires some degree of familiarity These hazards are frequently discussed in rotorcraft literature,
with problem-domain knowledge, which in the case of avia- such as the Helicopter Flying Handbook (Ref. 21). Below
2
Expert based event Model based event definition discovery & detection
definition & detection
Simple physics based Reduced event models
model events from off-line simulation
Crew
Feedback
are brief summaries of the conditions considered in this work, ing gear provides a pivot point when it is prevented from mov-
and for more details the reader is referred to the handbook and ing. A static rollover is primarily experienced on helicopters
similar literature. with a high-mounted tail rotor, where the increased moment
The Vortex Ring State (VRS) is a well-known aerody- about the ground contact point can cause a tipover. In either
namic condition that is encountered at moderate forward ve- case the recovery requires pilot inputs that may be counter-
locities and descent rates that are roughly 0.5 to 1.5 times the intuitive, as cyclic control alone is usually not sufficient to
value of the hover induced velocity through the rotor disk. prevent a rollover. Further, during normal flight an increase in
In VRS, the flow enters a recirculating pattern, which has collective input also increases the control moments possible
the effect of reducing thrust and causing an increase in de- through cyclic input. While in contact with the ground, the in-
scent rate. This condition is especially hazardous close to the crease in the moment about the pivot point far outweighs any
ground while operating at high altitudes above sea level. An- gains in control forces and moments at the rotor, so this typ-
other concern with VRS is that the typical response of the ical pilot response can be counterproductive in rollover pre-
helicopter is reversed, such that collective application may in- vention.
crease the descent rate. The primary recovery method is to fly
away from the recirculating flow. Flight condition representation
Dynamic rollover is another hazard of interest, where the
helicopter rolls over while on the ground due to the combined Accurately representing a flight condition requires the under-
action of the main and tail rotors with a pivot point established lying physics to be properly modeled. The modeling choices
by the landing gear. While dynamic rollover is a well-known reflect this fact, as well as the need to balance the fidelity with
hazard, its name hides the fact that there are several possible the intended application in the context of a HFOQA system.
ways to encounter a rollover, both dynamic and static. In dy- While the requirements are not narrowly defined and com-
namic, the vehicle is translating or lifting off, while the land- putational power can be reasonably expected to continue in-
3
creasing, parsimonious choices are preferred. In addition to well within the operational flight envelope. Where our model
modeling effort and computational time, consideration was deviates from convention is in the fact that ground based ma-
also given to the availability of model parameter values and neuvering is also considered, and the inflow model must be
models elements were chosen so that parameters were either valid in and around the vortex ring state.
readily available or could be estimated.
Below we provide an overview of the model elements, with
To analyze the flight conditions of interest, two model- a particular focus on the unique aspects of the present model.
ing approaches were taken. The first approach aims to use These elements are of greatest importance to the conditions
a simple physics-based equation to detect the event from pa- which were considered in this investigation. For example,
rameters available in the data. A simple mathematical model fuselage aerodynamics were neglected in the dynamic model
is used to derive a physics-based boundary to help identify a due to the intended use in dynamic rollover analysis. Thus
particular condition. In prior work (Refs. 20,22) it was shown it is important to ensure the relevant elements are present if
that physics-based models can aid the typical HFOQA sys- the analysis is carried out in other parts of the operating en-
tem, and the approach was demonstrated in a case of power velope. The approach using the dynamic model is general
management during autorotation practice. The model allows enough to be carried out with a different model from the one
operational changes to be taken into account, creates an op- presented, which would not alter the manner in which the re-
portunity to adapt the detection to other vehicles, and helps sults are post-processed or used in a monitoring setting.
integrate other sources of information, such as wind, through
kinematic relationships. This approach is again used in the
present work, using a dedicated model to define the VRS Fuselage equations of motion
boundary and detect flight conditions which are within the
VRS region. The use of this type of simplified model allows The fuselage is represented as a rigid body with mass and iner-
rapid execution times and is ideal for direct implementation tia properties, upon which all forces and moments developed
in a data monitoring system (Fig.1), with the obvious need to by the rotors, other aerodynamic surfaces and any additional
develop models for every safety condition. However, for well components of the helicopter are applied to, expressed in the
known hazards this effort is warranted since the models im- body frame and applied about the CG.
prove the detection capability by providing a more localized A commonly used earth frame in flight dynamics is the
condition boundary while simplifying the extension of moni- North East Down (NED) frame. The NED frame follows the
toring capabilities to other monitored vehicles. vehicle cg and has the x-axis pointing north,the y-axis is ori-
A more general approach is to use a single model with suf- ented toward east of the vehicle, and the z-axis is pointed
ficient fidelity to simulate the vehicle’s response at many con- down toward the center of the Earth. A common and valid
ditions of interest, with the potential to automate the discov- approximation is to assume a flat earth model, which means
ery and definition of hazard boundaries. Similar approaches the curvature and rotation of the earth is neglected. Local vari-
have been taken by other researchers in the field, primar- ations in terrain can still be included, however.
ily with regard to maneuver evaluation and envelope protec-
It is convenient and common to express the forces and mo-
tion (Refs. 23–29). In this manner, a single approach which is
ments acting on the fuselage in the body-frame, which has
more general and aims to discover unsafe conditions through
its origin at the vehicle CG and is aligned with the vehicle’s
pertubations of control inputs and the monitoring of a metric
primary dimensions. The x-axis points toward the front, the
defined to indicate when limits are reached. In this work, a
y-axis is oriented toward the right, and the z-axis is perpendic-
flight dynamics model is implemented in Matlab based on es-
ular to both and pointed downwards. The body translational
tablished theory from 30, 31 and used to study the dynamic
and rotational rates are expressed in this frame. Equations 1
rollover condition. This type of approach is more suitable
and 2 are the expressions for the translational and rotational
to explore the extremes of the operational envelope and help
velocities of the vehicle.
identify conditions which are not well defined, or are rare
events (Ref. 32) that have not been adequately observed in
U
practice. V ∗body = V (1)
W
Model components
Based on the types of conditions typically monitored by p
HFOQA and the fact that most such conditions pertain to ω ∗body = q (2)
pilot-related actions, the model must have the capability to r
relate vehicle motion and performance parameters with pilot
In the above equations, the * superscript denotes quantities
inputs, as well as other vehicle parameters and environmen-
expressed in the body-frame. Fuselage rotational rates are the
tal parameters. Since most of the accidents whose causes are
pitch rate p, roll rate q, and yaw rate r.
tracked by HFOQA are not mechanical but rather pertain to
the pilots, a flight dynamic simulation similar to that described To evaluate the transient motion of the helicopter as it re-
in Padfield (Ref. 31) as “Level 1” is sufficient for most flight acts to pilot inputs, the derivatives of the translational and an-
4
gular rates are needed, which yield the translational and angu- X is then performed using an appropriate
Integration of Ẋ
lar accelerations as follows: method (commonly a 4th order Runge-Kutta integration
scheme) which results in the following rigid body state vector:
U̇
∗
V body = V̇ = −ω
V̇ ω ∗body ×V
V ∗body + (F
F ∗g + F ∗cg )/m (3) X = [X Y Z U V W φ θ ψ p q r]T (9)
Ẇ
Rotors
ṗ The main rotor represents the most significant source of forces
ω ∗body = q̇ = I −1 (−ω
ω̇ ω ∗body × Iω
ω ∗body + M ∗cg ) (4) and moments acting on the helicopter in flight. In this work,
ṙ the rotor is modeled using individual rigid blades having flap
and pitch degrees of freedom, with the airloads calculated us-
where m is the mass of the helicopter and I is the moment of ing blade element theory (Refs. 30, 31, 33). A full 360 degree
inertia tensor expressed in the body frame. F ∗g is the force lookup table for a sample airfoil (Ref. 30) is used to obtain
of gravity resolved in the body frame. F ∗cg and M ∗cg are the the lift and drag coefficients at all angles of attack. The tail
forces and moments generated through the rotor, aerodynamic rotor is modeled in a similar fashion, with the distinction that
surfaces, fuselage drag and any other component, such as the flap degrees of motion for each blade are neglected. In order
landing gear. to generate control moments, the rotor blades flap to reorient
To obtain the vehicle position in the earth frame, with re- the thrust vector according to pilot input, so the calculation of
spect to some origin, another transformation is required to this motion is very important for a flight dynamics simulation.
bring the body-frame values to the earth NED frame. The The blade motion is calculated for each blade using a flapping
transformation of quantities between the earth, body, and any equation of the form:
other reference frame used by the model is facilitated through
the use of a rotation tensor. The rotation tensor can be con- MA = Bβ̈ + Ω2 (B cos β + mb exg R2 ) sin β (10)
structed using a series of rotations, R = Rφ Rθ Rψ about the where MA is the applied aerodynamic moment in the positive
pitch, roll and yaw axes of the vehicle (3-2-1 sequence): flapping direction, and B is the component of the mass mo-
The rotation matrix is orthonormal and its inverse and trans- ment of inertia matrix Ib corresponding to rotations about the
pose are equivalent. The transpose is used to express the y-axis. This equation is a simplified version neglecting fuse-
body-frame velocities in the earth frame, yielding the position lage motions. Using the same treatment as the rigid body de-
derivatives: velopment for the fuselage, the motions of each blade can be
Ẋ
calculated in response to fuselage rates. The present model
Pe = Ẏ = RT V ∗body
Ṗ (5)
assumes the acceleration of the fuselage are much smaller
Ż
than the accelerations of the blade, which reduces the stiff-
Calculating of the fuselage orientation in the earth frame is ness of the problem and improves the computation time. Ad-
done by transforming the body rotational rates into the earth ditionally, the purpose of this model is to examine large scale
frame and integrating. The following Euler-angle transforma- motions of the overall helicopter, so higher-harmonic terms
tion is used for this purpose, which must be inverted before are neglected since their primary effect is on the vibration re-
pre-multiplying the body rotational rates: sponse of the vehicle.
1 0 − sin(θ )
Inflow
Leb = 0 cos(φ ) cos(θ ) sin(φ ) (6)
0 − sin(φ ) cos(φ ) cos(θ ) One of the most important aspects of rotor modeling is cal-
culating the airloads, which depend primarily on the inflow
The resulting rate of change of the earth-referenced bank, el-
through the rotor disk, or else the airflow experienced by the
evation and heading angles is
rotor. There are many choices available in the helicopter lit-
φ̇
erature (Refs. 34, 35), with varying degree of fidelity and ease
−1 ∗
α e = Leb ω body = θ̇ (7) of implementation. For the purpose of this effort, the in-
ψ̇ flow model should consider flight into the VRS regime, since
this condition is a known safety hazard. Several model have
The body-frame and earth-frame quantities describing the mo- been proposed that consider VRS in particular, such as the
tion of the vehicle can be assembled in a vector of state deriva- Wolkovitch (Ref. 36), Peters (Ref. 37), ONERA (Ref. 38)
tives for the fuselage: and Johnson (Ref. 39) models. Most of these models pro-
vide an expression for the boundary of the VRS region based
Pe
Ṗ on a physically relevant criterion, such as vortex convection
∗
V body
V̇ T speeds, stability, Vz drop and similar. This type of boundary is
X =
Ẋ α e = Ẋ Ẏ Ż U̇ V̇ Ẇ φ̇ θ̇ ψ̇ ṗ q̇ ṙ
(8)
ideal for immediate online application for monitoring or limit
∗ protection (Ref. 38). Several of these models also provide an
ω body
ω̇
5
extended inflow model that is valid up to and within the VRS with the exception that the a term is retained. Equation 12 is
region. Further, Johnson shows a good agreement for dynamic valid for a constant forward velocity, which is the reason it
simulation of entry into VRS, and for this reason, this model only contains Vz terms. To include the fit in the flight dynam-
was implemented in the current simulation and used in both ics simulation and in the VRS boundary, a non-dimensional
as a standalone simplified condition boundary and within the lookup table is created, with the values multiplied by the hover
dynamic model. induced velocity to obtain dimensional quantities. A further
development shown in (Ref. 39) is the extension of the VRS
νh2 model to time-domain simulation by casting the result as a
ν=p (11)
Vx2 + (Vz + ν)2 first order differential equation witht the time constant defined
in equation 13.
0.7 0.7
3 τrev = =q (13)
momentum theory
λh CT
2
curve fit
2.5 VRS model Since the same approach is taken for both main rotor and tail
rotor inflow, two additional states are added to the simulation
2 and integrated in time.
8/8h
ing systems. The results from the model evaluations are either
implemented directly in a monitoring system as boundaries in 2000
Flight data
Vortex Ring State Model use in HFOQA -3000 VRS boundary
VRS encounter
In HFOQA implementations using simplified models for con- -4000
dition detection, the model is used in the same manner as reg- 0 20 40 60 80 100 120 140 160
Horizontal speed [knots]
ular safety events, with the exception that the monitoring lev-
els (condition indicators) are physically meaningful quantities
output by the model instead of pre-defined thresholds estab- Fig. 5. Detection of VRS using boundary from Johnson
lished by experts and fine-tuned over a period of time. In prior (Ref. 39).
work (Ref. 22) it was seen how a simple performance model
of a helicopter can help distinguish unsafe flight segments dur- encountered during the flight. The trajectory is a clear indi-
ing an autorotation practice flight. The results are positive and cation of the rapid and nearly vertical descent during vortex
indicate that the application of physics-based models can lead ring state encounter. Though this example uses a simulated
to reduced false alerts by virtue of boundaries that adapt to the set of flight data, the analysis is nearly identical in the real-
vehicle and flight condition based on parameters available in world case. The primary difference would be related to issues
the flight data and external or estimated data. with noise and other artifacts in the data, though most of the
The same principle was applied to the detection of flight hardware used for this purpose is very reliable and performs
near the Vortex Ring State regime in the present work. The sufficient conditioning of the data before the data upload is
VRS boundary is usually expressed in terms of the horizon- completed.
tal and vertical velocities Vx and Vz (Ref. 39), and many re- Because the detected condition is related to the properties
searchers are in general agreement on the location of this re- of the flow through the rotor, the same principles can be ap-
gion. Figure 5 shows this boundary for a simulated flight data plied to any rotorcraft, conceivably allowing the same model
record, defined based the inflow model described above. The to be applied across a range of vehicles with minor modifi-
boundary is defined as the point of vertical descent rate where cations. This is a direct benefit for HFOQA operation, since
the inflow gradient changes, seen as a “bump” in fig.2. The the manual safety event definition can be aided by physically
helicopter was intentionally flown into the VRS region several
times, and at each time step the flight state of the vehicle was
checked against the boundary valid at that flight state. In this 4000
flight data
case the detection is based on a logical test against the VRS 3500 VRS encounter
boundary, which is identical to the approach broadly used in ground elevation
HFOQA. The primary difference is that the VRS boundary 3000
shown here is scaled with the hover induced velocity, so ve-
Altitude [ftmsl ]
-40
0 2 4 6 8 10 12
time [sec]
Yaw rate [deg/sec]
40
20
0
-20
-40
0 2 4 6 8 10 12
time [sec]
10
Roll rate [deg/sec]
-10
-20
0 2 4 6 8 10 12
time [sec]
4
Velocity [m/sec]
Fig. 12. Time to maximum roll angle vs. cyclic and tail
rotor control combination.
2
gradual taxi turn
rollover
0 In addition to identifying the regions where negative out-
0 2 4 6 8 10 12 comes are possible, it is be beneficial to quantify the reaction
time [sec]
time available as a measure of the risk associated with each
point of the envelope. In figure 12 the combinations of in-
put which do not result in a crash within the simulation time-
window are assigned a much longer time value than found in
Fig. 10. Comparison of a typical turn and a rollover case. the rest of the simulated trajectories. In this figure, the green
regions are regarded as safe control inputs, with more reaction
time available than the rest of the input space. Outside of this
region, the time to reach a critical condition is progressively
shorter, and is calculated for every combination of input using
the first hitting time metric. The type of mapping of Fig. 12
is one of the main results of this work that improves over the
standard HFOQA events based on static logical thresholds.
First, the result is more a continuum than a crisp thresholds,
though an level can be easily specified. This continuum can
be representative of the proximity to a certain condition, rather
than just having an indication of whether a condition has oc-
curred. Another benefit of the representation of figure 12 is
the ability to identify additional hazards or interesting behav-
ior around what would otherwise be considered the extent of
the operational envelope.
Comparing the output from the dynamic model to the
boundary from the operator and even the one defined using
the simplified 2-D model (Ref. 22), we can see that there are
more complex regions visible in the output of the dynamic
Fig. 11. Maximum roll angle achieved vs. cyclic and tail model used in this investigation. Cornering moments are not
rotor control combination. considered in the 2-D case, and they tend to exacerbate the
rolling motion induced by the tail rotor, but oppose the rolling
10
as “safe” in Fig. 12 when the calculation is based purely on
roll angle, are shown to have excessive yaw rates in figure 14.
Without considering yaw rate, an optimistic boundary may
be adopted, leading to potential problems in the presence of
obstacles or terrain. Therefore additional limits on the yaw
rate and translation velocity are specified which limit the re-
gion. The alternative is that some other manner of detecting a
critical condition, such as collision detection, must be imple-
mented for these vehicle states.
At present, both the unsafe outcomes and these undesir-
able and risky alternative outcomes will be considered to be
beyond the limit of acceptable operation. Considering mul-
tiple critical conditions simultaneously reduces the safe input
region to a smaller portion of the input space. Figure 15 shows
the combined results of control input sweeps at three different
collective (thrust) settings.
The results are contour plots defined for equal values of
Fig. 13. Roll rates vs. lateral cyclic and tail rotor input. time-to-condition for each collective setting. The 5 and 10
second levels were chosen for illustration purposes, but the
entire gradient up to the maximum time evaluated in the sim-
ulation run is available. An operator may select different time
thresholds to use as training material, or in a data monitoring
system where the contour can be implemented as a boundary
in a manner similar to the simple models described above, as
well as existing HFOQA events. In addition to providing a
result which can be used to define a detection boundary, the
result in Fig. 15 contains information about how the set of
acceptable inputs is altered by changes in collective. One ob-
vious but very important aspect is the drastic reduction in the
size of the region. As the vehicle nears the moment of takeoff,
the region of lateral cyclic and tail rotor input gets very small
around the trim point.
A defining characteristic of helicopter piloting technique is
the need to constantly compensate for cross-coupling effects
between the different controls. In Fig. 15 this is reflected in
the fact that the “center” of the contour moves with the trim
Fig. 14. Yaw rates vs. lateral cyclic and tail rotor input. point, which changes depending on the thrust setting. From
the figure it is clear that at higher thrust settings, much higher
values of tail rotor input are required.
motion when the turn is initiated using the main rotor. It seems
reasonable to expect different cases of rollover due to cyclic The type of mappings shown in Fig. 15 are immedi-
or tail rotor input, but the simulations show that there are addi- ately useful for communicating hazards in crew training ma-
tional possible outcomes, that may not be initially regarded as terial, as well as for implementation in a database as detection
unsafe, but are certainly undesirable. These are cases where thresholds, operating on real data. The simpler models can be
the helicopter begins a rapid yawing motion due to excessive applied directly in a data-based monitoring system, whereas
tail rotor input, while the cyclic is applied in a way that op- the results of the dynamic simulation studies must be com-
poses the rolling moments enough to keep the vehicle in this puted offline.
condition for an extended period of time. Another example is Another benefit of the present approach is that regardless
the case where the vehicle is taxiing at high main rotor thrust of the parameters used to define the critical condition bound-
levels - excessive cyclic and high roll angles on the ground ary (such as representing VRS in terms of Vx and Vy ), the
can cause the tires to break free and the vehicle to enter a temporal information provides a gradient that points to the
slide. These two cases are both undesirable, but may well be region with higher time-to-critical condition, which is associ-
recoverable, which will be the topic of further research. For ated with a lower risk. In the rollover case, the critical bound-
example, consider figures 13 and 14, where the maximum roll aries are established based on roll angle, yaw rate, and roll
rate and maximum yaw rate are shown. The maximum roll rate, but the result is presented in terms of time to reach any
rate occurs roughly at the same location in the input space as of those conditions. The introduction of another critical con-
the maximum roll angle. However, parts of the region marked dition would not affect the presentation of the results, which
11
2. The use of a dynamic simulation was also shown for the
purpose of establishing bounds on safe operation in ad-
vance of any specialized event definition. This approach
required the analysis to be performed ahead of time.