35th European Rotorcraft Forum 2009

DocumentID: 101191

HELICOPTER SLING LOAD POSITIONING

H. Brenner
([email protected])
Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR)
Institute of Flight Systems
Braunschweig, Germany

Abstract.1 The paper at hand addresses the precise positioning control of helicopter sling loads. Due
to a significant decrease in handling qualities caused by the external load, the manual load positioning can be
hindered. By means of linear quadratic optimal output regulation, the helicopter position as well as the load
sway are controlled in an integrated approach. The position control is part of an existing AFCS, which guaran-
tees for the basic helicopter stability. The controllers are developed and tested on the basis of a comprehensive
system simulation.

Notations C container
CG center of gravity
Symbols cmd command
A [−] system matrix ctvr control variable
a [m/s2 ] acceleration (u̇, v̇, ẇ)T cur current
B [−] control matrix g geodetic
C [−] observer matrix H helicopter
cS [N/m] sling spring constant i i-th sling
D [−] feedthrough matrix L load
dS [N s/m] sling damping constant LH load hook
I [kg m2 ] inertia tensor P pendulum
J [−] performance index R rod with swivel joint
Ky [−] output feedback matrix S sling
l, w, h [m] length, width, height trgt target value
m [kg] mass
R, S, Q [−] weighting matrices Abbreviations
TM [−] transformation matrix AFCS automatic flight control system
U [−] unit matrix OM optical marker
u [−] system control vector
V [m/s] velocity (u, v, w)T
x [−] system state vector
y [−] system output vector 1 Introduction

T
(Φ, Θ, Ψ ) [rad] Euler-angles he capability to transport sling loads of differ-
T
ω [rad/s] angular rates (p, q, r) ent goods to remote locations with poor ac-
δa [%] cyclic lateral control (A1s [◦ ]) cessibility may characterize the versatile uti-
δb [%] cyclic longit. control (B1s [◦ ]) lizability of helicopters in particular. However, due to
δc [%] collectiv control (ΘM R [◦ ]) the presence of the sling load, the dynamic behavior of
δp [%] pedal control (ΘT R [◦ ]) the helicopter is influenced in a way that the handling
qualities are degraded. Hence, the pilot workload is
Indices increased due to the task of controlling the sling load,
A, a aerodynamic which implies the damping of load pendulum motions
AP attachment point as well as the precise positioning of the attached load.

1 Paper presented at the 35th European Rotorcraft Forum, Hamburg, Germany, Sept. 22-25, 2009

©DGLR 2009 1
 35th European Rotorcraft Forum 2009

(a) approach (b) placing

Figure 1: Precise load positioning [1]

Flight tests with a CH-53G heavy-lift helicopter car- bodies and a rod in which the slings are fitted (fig.
rying mobile pioneer bridges as sling loads revealed 14). The rod is connected to the helicopter’s single
certain challenges in the precise load positioning (fig. load hook and features a swivel joint in order to allow
1) [1]: The central issue in precise load positioning is the load to turn without twisting the slings, which
given by the fact that the pilot cannot see the load could elsewise exceed the load limits of the slings.
in general. Hence, the detection of offsets in the de- The slings are modeled as flexible cables: Hence, the
sired load position is done by onboard crew members cable-forces represent the constraining forces within
or ground personnel giving instructions to the pilot. the two-body system. Different load aerodynamics
This procedure implies time delays until a proper cor- can be considered. The rigid body dynamics of the
rection of the position error can be executed by the helicopter, the load and the rod is discussed in the
pilot, which by then may not be effective anymore. appendix.
The longer the positioning maneuver takes, the more
the pilot workload increases, which in turn leads to
false control inputs and further position errors. The Aerodynamics
dynamic influence of the sling load on the helicopter
The aerodynamic forces and moments of the he-
increases with the weight of the load.
H,b , M H,b ) are nonlinear functions of the
licopter (F A A
In order to support the development of controllers helicopter motion and the atmosphere, which include
for the precise load positioning, the overall system was the relevant multi-dimensional effects sufficiently. For
modelled [2]. Both subsystems – helicopter and load the present work, linear aerodynamics of a CH-53D
– are considered as rigid bodies. The flexible slings cargo helicopter are implemented, leading to a quasi-
build a multiple-strap-harness. Based on the modeling nonlinear description of the helicopter dynamic in (58)
of the overall system, a comprehensive simulation was and (59). The derivatives are obtained from [3]: They
designed featuring a quasi-nonlinear CH-53 helicopter were derived by linearization of a generic nonlinear
model and a nonlinear load system. The sling load simulation code and cover a speed range from hover
model comprises different aerodynamic features. The up to 140 kts at a helicopter gross weight of 16 tons.
simulation tool provides trim calculation, linearization
and flight simulation. Hence, a basis for system anal- For the following investigations of the performance
yses and the controller development is available. and robustness of the positioning control, a cubic con-
tainer is considered as external load, featuring an edge
length of 2.4 m. The aerodynamic coefficients have
been derived in wind tunnel tests. The polars of the
static aerodynamic coefficients – which lead to a tan-
2 System Modeling and Simula- gential, and two normal forces, and a moment – were
tion identified in [4]. However, unsteady aerodynamic in-
fluences on the load surface are not taken into account
The modeling and simulation of the overall system due to the rather low frequencies of the system’s mo-
helicopter-slings-load is supported by using Matlab tions. The resulting aerodynamic forces and moments
& Simulink® . The system is built up of the two rigid are derived by (65) and (73).

©DGLR 2009 2
 35th European Rotorcraft Forum 2009

Constraining forces cause then system complexity is kept to a minimum,

which in turn leads to higher reliability. For this rea-
The load carrying harness consists of one or more
son, a camera sensor featuring digital image process-
slings, whose dynamic properties are defined by spe-
ing for measuring the pendulum dynamic is chosen: A
cific spring and damping constants, depending on the
camera, which is mounted underneath the helicopter,
sling material. Due to the relative motion between he-
tracks an optical marker (OM), which is placed on the
licopter and load, the slings are elongated, resulting in
load or at the slings, and generates visual information
forces (eq. 77), which in turn generate moments (eq.
of the load position and velocity with reference to the
82, 83) due to the offset between the sling attachment
helicopter body system (fig. 2). The load pendulum
points and the respective center of gravity (fig. 14).
The sling forces and moments as well as the aerody- angles ϕ̄˙ P and ϑ̄˙ P in the helicopter body system are
namic forces and moments are added to the vectorial then determined. These data are further processed us-
forces and moments in the equations of motion (58) ing the helicopter attitude and body rates – measured
and (59). by the onboard IMU – in order to derive the controller
variables ϕ̇P and ϑ̇P , which describe the load pendu-
lum motion in the geodetic system.
Automatic flight control system
This kind of sensor was developed by iMAR GmbH
In order to stabilize the basic unstable helicopter, [5] and means the basis for the simulation setup. The
an AFCS is implemented according to [3] (fig. 3). It digital image processing provides short time delays as
superposes the pilot inputs by ±10% in cyclic and col- well as sample rates, which are sufficiently high con-
lective control, and by ±3% in pedal control; the ac- cerning the rather slow load pendulum dynamic. Fig-
tuator rates are limited to 100%/s. ure 2 shows the mounting position of the camera; its
In section 4, the pendulum controllers are deter- field of view covers an opening angle of 60◦ , which can
mined on the basis of the AFCS-controlled, and thus, be extended to 180◦ at the expense of process time.
stable helicopter. Besides the derivation of the system pendulum an-
gles and rates, the digital image processing includes
linkage MR [°] Kálmán-filtering for the simulated prediction of the
c [% ]
position of the optical marker. The load dynamics is
b [% ] B1s [°]
calculated by means of a simplified analytical pendu-
a [% ]
- servos A1s [°] lum model. This information is needed, in case the
p [% ]
- external load is not located within the camera field
TR [°]
- of view, due to large pendulum deflection angles, for
limiter instance.
F/FW -
(, p ) State space model
F/B
(,q ) The equations of motion in (58), (59) and (61) de-
F/B scribe the overall system helicopter and sling load, and
( ,r ) are used in the numerical simulation in their full non-
F/B linear formulation. For analyses of the flight dynam-
ics – for instance, stability and controllability – the
Figure 3: AFCS basic sketch system of equations must be simplified. For this, the
state variables are bound to a working point, in order
to enable a linearization: The theory of linear systems
Sensor selection and modeling considers stationary flight conditions at an operating
point x0 . For the analysis and synthesis of linear sys-
The states of the model description do not deliver di-
tems, a multiplicity of tools – time-domain based as
rect information about the load pendulum sway. Thus,
well as complex-variable-domain based – is available.
the pendulum motions, which are to be controlled,
must be measured and provided as output variables. The linearization of the overall system leads to the
Different types of sensors are considered: For instance, state space model:
an IMU that is mounted on the load can measure the
(1) ẋ = Ax + Bu (state equation), x(0) = x0
body rates and transmit the data to the helicopter for
further processing. Another approach is the installa-
tion of a sensing arm that follows the motions of the (2) y = Cx + Du (output equation)
rod as an indicator for the load dynamic.
In general, sensors that do not require to transmit The state vector is given with
the information of the load dynamic from the load sys-
T
tem to the helicopter system are advantageous, be- (3) x = [xH , xR , xL ]

©DGLR 2009 3
 35th European Rotorcraft Forum 2009

xH , b
HCG yH , g HCG
xH , g H
y H ,b H

LH / camera LH / camera

-P M P

z -P M P

z
x y
OM OM
y L ,b
LCG
LCG
camera fi xL , b z L ,b z L ,b
eld of vie
w

Figure 2: Sensor installation

including the helicopter/load states (Λ = H, L) For the given system the feedthrough matrix is a zero
matrix:
(4) xΛ = (u, v, w, p, q, r, Φ, Θ)Λ
(11) D = 0 (12x4 )
and the states of the rod
 
(5) xR = ϕ̇, ϑ̇
R 3 Analysis of the System Flight
and the control vector Dynamics
T
(6) u = [δc, δb, δa, δp] The flight dynamics describes the character of the mo-
tions of the overall linear system; one important re-
as well as the output vector sult is the stability analysis. Applied for a cubical
   T load without aerodynamics, figures 4 and 5 show the
(7) y = (u, v, w, p, q, r, Φ, Θ)H , ϕ, ϕ̇, ϑ, ϑ̇ eigenmodes of the helicopter and the load at 60 kts for-
P
ward level flight. The considered weight of the cube is
The system state matrix consists of the main matri- 3000 kg and the length of the single sling is 7 m.
ces of the partial systems and the respective coupling The incorporated degrees of freedom within the
matrices eigenmodes were analyzed using the corresponding
⎡ ⎤ eigenvectors. A characterization is given in table 1.
AH AR→H AL→H The eigenmodes I, II, V, VI, VII, VIII mainly de-
(8) A = ⎣ AH→R AR AL→R ⎦ scribe the helicopter dynamic, slightly coupled with
AH→L AR→L AL the load dynamic. As a coupled motion of the sys-
tems helicopter-rod-load, the pendulum oscillation is
where the submatrix AH contains the classical heli- described by the eigenmodes III (lateral) and IV (lon-
copter derivatives. The control matrix results in gitudinal). A vertical oscillation of high frequency is
⎡ ⎤ given by IX : The mode couples the vertical axes of the
BH
helicopter and the load by the flexible sling. The in-
(9) B = ⎣ B R ⎦
0 corporation of the dynamics of the two degrees of free-
dom of the rod are described in X and XI. They are
both of high frequencies, because of high constraining
with B L = 0. The observer matrix is given with
forces acting at the rather light rod of 50 kg. Besides
CH 0 (8x2 ) 0 (8x8 ) the pendulum motions, the single suspended cube ex-
(10) C =
C H→P C R→P C L→P ecutes pitch and roll, which finds its expression in the

©DGLR 2009 4
 35th European Rotorcraft Forum 2009

Table 1: Characterization of eigenmodes

eigenmode characteristic eigenmode characteristic

I helicopter roll VIII coupled helicopter

mode longitudinal motion
II helicopter dutch- IX coupled vertical
roll mode oscillation
III coupled lateral X coupled DmL
pendulum motion lateral oscillations
IV coupled longitudinal XI coupled DmL
pendulum motion longit. oscillations
V helicopter lateral- XII load pitch mode
and roll mode
VI helicopter roll XIII load roll mode
mode
VII coupled helicopter
vertical motion

1.5 0.94
load off 0.88 0.76 0.5
1.5 0.94
load off 0.88 0.76 0.5
load on IV load on 4000kg
0.968 0.968
I III IV
5m I
1 1
II
0.984 0.984 II III
Im {s}
Im {s}

V V
0.5 0.994 0.5 0.994 500kg
30m

0.999 0.999

VI VII VIII VI VII VIII

0 0

-5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 -5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
Re {s} Re {s}
(a) close-up range (a) close-up range

50 0.0013
load off 0.0009 0.00055 0.0003
50 0.0017
load off 0.00115 0.0007 0.00035
45 load on
45 load on
500kg
40 40
0.0023
X XI
35 0.0019 X XI 35
IX
30
30
30 30
Im {s}

25 0.0028 25 0.0034
Im {s}

5m 4000kg
20 20
15 0.0042 IX 15 0.0055
10 10
0.009 30m 0.012
5 5 XII XIII
XII u. XIII
0 0
-0.07 -0.05 -0.03 -0.01 0 0.01 -0.09 -0.07 -0.05 -0.03 -0.01 0 0.01
Re {s} Re {s}
(b) far-field range (b) far-field range

Figure 4: Pole-zero maps for a steady-state horizon- Figure 5: Pole-zero maps for a steady-state horizontal
tal flight at 60 kts with a 3000 kg sling load, and a flight at 60 kts with a 7 m sling, and a variation of load
variation of sling length weight

©DGLR 2009 5
 35th European Rotorcraft Forum 2009

10
eigenmodes XII and XIII. Due to the lack of aerody- 0

amplitude [dB]
namic stabilization, the load’s yaw motion has neutral -10
-20
stability, caused by the yaw hinge of the rod – it is load pole III
-30
located in the point of origin, and is not shown for -40
clarity reasons. -50
Depending on system parameters like the cable 45
length and the load mass, the eigenmodes of the over- 0
-45

phase [°]
all system vary – particularly the pendulum motions
-90
in III and IV. The overall system shows a tendency to
-135
developing marginally stable pendulum motions (fig. -180 load off
4 and 5). When increasing the sling length, the fre- -225
load on
10-1 100 101
quencies of the eigenmodes III and IV decrease, as
 [rads-1]
well as of the vertical motion in IX, since the sling
spring constant is a function of its length (fig. 4b). By (a) Gδa ẏH (s)
increasing the load weight, the pendulum frequencies 10
rise (fig. 5a). This effect can be explained by a simpli- 0

amplitude [dB]
fied two-point dumb-bell model with the pendulum fre- -10
-20
quency given by (q.v. [6]): -30 load pole IV
-40
 -50
g mL
(12) ωP = · 1+ 45
lS mH 0
-45
phase [°]

Since the helicopter roll rate in I and II is coupled with -90

the load weight, the change in the pendulum frequency -135
affects these eigenmodes, too – the damping declines -180 load off
load on
-225
with an increase in load weight. Furthermore, heli- 10-1 100 101
copter body-rates are controlled by the AFCS, which  [rads-1]
further amplifies the reciprocal effect of decreasing (b) Gδb ẋH (s)
system damping as result of increasing system weight.

Figure 6: Frequency responses of the translational he-

licopter velocities at V = 0 kts, mL = 4000 kg, and
4 Position Controllers lS = 7 m
The solid lines indicate the helicopter without the
4.1 Problem Analysis external load whereas the dotted lines illustrate the
The positioning of helicopter sling loads is a twofold transfer behaviour of the translational displacement
task. On the one hand, the load must be held on its with the load attached. Considering the latter case,
position under the influence of disturbances. On the it can be seen that the poles of the pendulum oscilla-
other hand, the load must be placed precisely from one tions – pole III for the lateral and pole IV for the lon-
position to another with minimal load sway. gitudinal pendulum motion – considerably affect the
For the development of positioning controllers, a ref- dynamics of the translational helicopter displacement:
erence flight case is considered featuring the key vari- The amplitude margins as well as the gain margins are
ables V = 0 kts defining hover state, mL = 4000 kg lowered.
as maximum load weight in addition to the helicopter Hence, when positioning the helicopter in order to
weight, and lS = 7 m for a short sling leading to quick place the load, the load pendulum motions must be
pendulum motions. considered in the control strategy, too. Otherwise, the
Hoh et al. analyzed the dynamic impact of sling helicopter will be moved out of its target position con-
loads on the handling qualities of a CH-47D Chinook stantly due to the coupling influence of the swinging
helicopter [7]. One essential output was that the con- load. Thus, the need for an integrated control of the
trollability of the translational displacement of the he- translational displacement of the helicopter together
licopter is degraded due to the presence of a sling load. with the pendulum dynamic is determined.
This effect intensifies with an increase in load weight. This integrated control task of positioning a load
The dynamic influence of a sling load on the lateral and damping its oscillation simultaneously is well
and longitudinal helicopter speed above ground due to known from the gantry-crane control. Significant con-
cyclic control inputs – expressed by the transfer func- tributions were made by Iordanou in [8] and [9], and
tions Gδa ẏH (s) and Gδb ẋH (s) – of the CH-53D refer- by Corriga in [10], and by Wang and Surgenor
ence flight case is shown in figure 6. in [11] and [12], for instance. They realized optimal

©DGLR 2009 6
 35th European Rotorcraft Forum 2009

feedback control by using linear quadratic state regula- in order to test the achievable load-position accuracy
tion (LQR) in order to position the load time-optimal under the influence of a turbulent atmosphere. Addi-
with minimal load sway. This approach was adapted tional control inputs apply:
for the helicopter sling load positioning task at hand.
The goal is to position the helicopter, and simultane- T
(13) (uturb )H = [δc, δb, δa, δp]turb,H
ously damp the load pendulum motions. Hence, by
the time of arrival at the target position, the load’s
position is perpendicular and stable beneath the heli- Furthermore, the turbulences affect the load aero-
copter. dynamics, leading to additional inflow:

T
(14) (xturb )L = [u, v, w]turb,L

4.2 Controller selection aspects

In case of a load displacement from a current po-
The LQR-control is well suited for the interacting con- sition P 0 to a desired target position P trgt , the heli-
trol of multi-variable systems. By means of a per- copter velocity in (u, v, w)H and the covered distance
formance criterion, which is specially tailored to the ΨH =0
(Δx, Δy)H,g are determined by means of the integra-
objective of the control task, control parameters are tion of the measured body accelerations and the atti-
determined, which deliver optimal results for the pur- tude of the helicopter in ΦH and ΘH . Here, ΨH = 0
pose of a weighting of the state and control variables. denotes that the geodetic system is turned by the az-
The controller is derived as solution of an optimization imuth ΨH (fig. 8). The target position as well as the
problem. For the correct application of LQR-control, starting time t0 for positioning could be commanded
all state variables must be measurable. by the pilot, for instance. The control variables for
However, since the load pendulum motions are de- positioning the helicopter are given with the position
tected by means of a camera and digital image pro- offset and its derivatives in equation (15) according
cessing, the states of the load-body in xL cannot be figure 8.
measured and thus, the full state vector cannot be
determined. The consideration of a deterministic or
stochastic observer may ease the problem. Another ΨH =0 ΨH =0 ΨH =0
Δx x x
method of optimal control is given with the linear (15) = −
Δy y y
quadratic optimal output control: The weighted op- ctvr

trgt,H,g
 
cur,H,g
timal control is accomplished by feedback of the posi- = P trgt
tion offset and the pendulum deflections. Compared
to the observer design, this approach is advantageous
with
because of a minor implementation complexity. The
structure of the control loop is illustrated in figure 7. ΨH =0 ΨH =0 ΨH =0
A turbulence model according to Lusardi et al. in [13] x Δx x
(16) = +
is implemented and adapted to the CH-53 helicopter y trgt,H,g
Δy cmd
y H,0,g
  
=P0 =0
uturb H
xturb L

+ x 1 x + C y
B IMU +
+ + s
north

Image Proc.
 0
yctvr
H

A
H
full system  0
xctvr
H
Ptrgt

+
control
authority P0
upilot
uAFCS
AFCS
+
upos,opt control variables - y pos,cmd
K y,opt east

Figure 7: Output regulation Figure 8: Position control signals

©DGLR 2009 7
 35th European Rotorcraft Forum 2009

and 4.3 Linear Quadratic Output Regula-

 t⎛ ⎞ tion
ΨH =0 u
x Ψ =0 ⎝ v ⎠ The condition for optimality of the output regula-
= (T M )gbH · dt
y tion corresponds to the demand that a pre-defined
cur,H,g
t0 w H,b performance index J must be minimized during the
(17) ΨH =0
x changeover of the system from the initial state x0 to
+ the final state xe in the time intervall t ∈ [0, te ]. Gen-
y 0,H,g
   erally, the performance criterion is quadratic, weight-
=P0 =0
ing system states as well as control deflections. Hence,
the optimal controller is not derived against the back-
The control variable of the position velocity is given
ground of fulfilling demands on the performance of the
with
dynamic of the control loop. Instead, it is determined
ΨH =0 ΨH =0 ΨH =0 by the evaluation of the progress of the state and con-
Δẋ ẋ ẋ
(18) = − trol variables within a considered time period. For the
Δẏ ctvr
ẏ trgt,H,g
ẏ cur,H,g
   linear and time-invariant system of the equations (1)
=0 and (2), and the output feedback

and (24) u(t) = − K y y(t)

the time-variant performance index is formed in de-
(19)
 ⎛ ⎞ pendence on [14]:
ΨH =0 u̇t
ẋ
=
Ψ =0
(T M )gbH · ⎝ v̇ ⎠ dt J (x0 , u(t)) = y  (te )Sy(te )
ẏ  te
cur,H,g t0 ẇ
⎛
H,b
⎞ΨH =0 
(25) + x (t)C  Qy Cx(t)
0
u
+⎝ v ⎠ + u (t)Ru (t) dt
w H,b The achieved outputs in y(te ) – as deviations from
the desired output y = 0 – are evaluated by means
considering of S. High control deflections are unfavorable and are
therefore penalized by R. The progress in the out-
(20) put variables y is rated by the evaluation matrix Qy .

Ψ =0 cos Θ sin Θ sin Φ sin Θ cos Φ The matrices S, R and Qy are symmetric and positive
(T M )gbH =
0 cos Φ − sin Φ H
definite. In the following, the formulas of the linear
quadratic output regulation are explained for the sake
Besides the position offset of the helicopter and its of completeness.
derivation, the pendulum angles and rates are needed The aim in LQ is a stationary feedback gain ma-
for feedback control: trix K y by means of the minimization of the perfor-
   mance index J of equation (25). The feedback con-
Δϕ ϕ ϕ trol must be independent on the initial state x0 and
(21) = −
Δϑ P,H,b ϑ cmd,P ϑ cur,P time-invariant. By neutralizing the time limitation,
   equation (25) changes to the infinite time quadratic
=0
performance criterion. Hence, with

   (26) y(t → ∞) = 0
Δϕ̇ ϕ̇ ϕ̇
(22) = − the output evaluation by S can be omitted. In order to
Δϑ̇ P,H,b
ϑ̇ cmd,P
ϑ̇ cur,P
   use the performance criterion to find an optimal feed-
=0 back gain matrix that is suitable for different flight
states the dependence on x0 must be eliminated. For
Hence, the complete output vector for the linear
this, the performance obtained for a linear indepen-
quadratic optimal output control is given with:
dent set of initial states is averaged. This corresponds
to the assumption of the initial states in x0 being a
(23) random variable uniformly distributed on the surface
  T
of the n-dimensional unit sphere. Considering the ex-
Ψ =0
y = (Δx, Δẋ, Δy, Δẏ)H,g
H
, Δϕ, Δϕ̇, Δϑ, Δϑ̇ pected value E of the n-initial states
P,H,b
1
(27) E {x0 x0 } = U
n

©DGLR 2009 8
 35th European Rotorcraft Forum 2009

the optimization problem is given with: For the search of the minimum of the modified cri-
 ∞ terion J, the gradient of the trace of P is determined
min E {J} = min E x (t)C  Qy Cx(t) with respect to the elements in K y . The gradient fea-
Ky Ky 0 tures the characteristic to always point towards the
(28) 
  direction of the steepest descent. The approximation
+u (t)R u (t) dt of the extreme values follows this direction, and it is
therefore inevitably arriving at a local minimum. The
The solution of (28) as a result of the minimization gradient matrix is given by the following equation ac-
of the expected value E by the adaptation of K y leads cording to [14], [15], [16]
to the optimal time-invariant feedback control K y,opt ,
which is effective independent of the initial states. ∂J  
(37) = 2 R K y C − B P L C 
The overall system from (1) is rewritten to ∂K y

(29) ẋ(t) = A − BK y C x(t) , x(0) = x0 and P as the solution of the Riccati-equation in (34)
   with
A

(A − BK y C) P + P (A − BK y C)
In addition, x(t) is given with (38)  
+ C  Qy C + C  K y RK y C = 0
(30) x (t) = x0 eA t and L as the solution of the Ljapunow-equation:

Substituting (30) into the performance criterion leads (39) (A − BK y C) L + L (A − BK y C) + U = 0
to
 ∞ The minimum is arrived when the norm of the gra-
 dient approaches the pre-defined break value :
(31) J = x0 eA t Q eA t x0 dt = x0 P x0
0  
 ∂J 
(40)   <
with ∂K y 
 ∞
(32) P =

eA t Q eA t dt For the search of the minimum of the performance
0 criterion by the variation of the elements in K y , the
value of J is calculated along the gradient. A starting
and point K y,0 of the approximation method for the func-
  tion J ∈ Rn must be determined for that the closed
(33) Q = C  Qy C + C  K y RK y C
loop is stable. This first feedback gain matrix can
By means of partial integration the matrix P is either be derived by an iterative trial-and-error ap-
transformed to the linear Riccati-equation: proach or by some formalized approach (e.g. [17]).
The optimal solution is then approximated within i
(34) A P + P A = −Q iterative steps starting from the initial feedback gain
matrix. For the approximation a stepsize a is intro-
For the symmetric matrix P – as the solution of the duced, which is regulated by common techniques of
Riccati-equation – it applies: unidimensional search: For the paper at hand, the
Armijo condition was chosen. According to this pro-
min E {J} = min E {x0 P x0 } cedure, the step size is controlled and adapted by the
Ky Ky
(35) inequation (41). Hence, any overshooting beyond the
= min E {trace (P x0 x0 )} minimum as a result of large step sizes is avoided as
Ky
well as the conduction of too many steps of iteration
According to (28) and the initial state x0 – which due to very small step sizes. The performance index
is located on the unit sphere of (27) – and according of the potentially more optimal controller K y,i+1 is
to the fact that the expected value of the trace of P compared to the current controller K y,i :
equals the trace of P itself, hence, the optimization
of the expected value leads to the final form of the (41)
optimization problem:  
∂J ai ∂J −∂J
J K y,i − ai −J (K y,i ) < ,
1 1 ∂K y 2 ∂K y,i ∂K y,i
(36) min E {J} = min trace (P ) = min J
Ky Ky n Ky n If the inequation (41) is satisfied, a new and more
optimal controller is found:
The minimum of the function J is derived by means
of nonlinear optimization: A steepest descent algo- ∂J
rithm with Armijo-stepsizes was chosen. (42) K y,i+1 = K y,i − ai ·
∂K y

©DGLR 2009 9
 35th European Rotorcraft Forum 2009

Consequently, the step size is increased for the next An important task in the LQ optimal regulation is
step of iteration: the agreement of the weighting matrices Qy and R.
The ratio between the elements qij ∈ Qy and rkl ∈ R
(43) ai+1 = ζ · ai ,ζ ≥ 1 . determines the control rate: Small values of rkl lead
to an increase in rate but require greater actuator am-
If (41) is not satisfied, the step size a must be too plitudes and vice versa.
large, leading to The ratio among the elements qij determines, how
fast the states are run to zero. Hence, the choice of
(44) K y,i+1 = K y,i
Qy and R is essential since they mean the basis for
and the lowering of a according to: the evaluation of the progressions of the outputs and
control variables.
(45) ai+1 = ξ · ai ,ξ < 1 . A common approach to determine the weighting ma-
trices is given with the trial-and-error method. The
The step size is decreased until (41) is satisfied and elements are varied without considering any intercon-
a new controller is found. By means of the feedback nections nor dependencies between them. According
gain K y,i+1 the Riccati-equation in (38), and the to the number of system states, this procedure rapidly
Ljapunow-equation in (39), and the gradient accord- loses clearness and practicality due to a multliplicity
ing to (37) are determined. In a next step, the break of variations. In the paper at hand, the weighting ele-
condition in (40) is evaluated by its arrival the optimal ments are derived in general accordance to the method
controller is found: of Wang and Surgenor [11]. The maximum devia-
tions of the output variables qij as well as the max-
(46) K y,opt = K y,i+1 imum control deflections rkl are defined as guideline
values for subsequent parameter variations. The off
The process chain of the derivation of the optimal diagonal elements of Qy and R are assumed zero, and
output control is once again illustrated in figure 9. the q11 and q33 elements are set to one, which is the
weighting of the helicopter position. This procedure
determination of K y,i=0 for that A  B K y,i=0 C is stable (29) seems adequate, since the weighting characteristic is
determined by the ratio of the elements instead by
variation of qij ‰ Qy (51)
their absolute values. Hence, there are four tunable
variation of rkl ‰ R (53), (54) parameters left: q55 and q77 for the weighting of the
load angles, and r22 and r33 for the penalization of
steepest descent approximation with a and 
cyclic control deflections. The longitudinal and lateral
variation of i motions are weighted separately so that the output
determiantion of Pi und Li (38), (39) weighting matrix is given with
sJ (37), (40) ⎛ ⎞
1 0 0 0 0 0 0 0
sK y,i
 b ⎜ 0 0 0 0 0 0 0 ⎟
⎜ ⎟
⎜ 1 0 0 0 0 0 ⎟
⎜ ⎟
calculation of the new feedback gain (68): break ⎜ 0 0 0 0 0 ⎟
⎜ ⎟
⎜ (yH )2
max ⎟
K y,i+1  K y,i  a ¸
sJ (47) Qy = ⎜
⎜ (ϕP )2
max
0 0 0 ⎟
⎟
sK y,i ⎜ symm. 0 0 0 ⎟
⎜ ⎟
⎜ (xH )2 ⎟
adaptation of the stepsize a according to ⎜ max
0 ⎟
⎜ (ϑP )2
max ⎟
the rule of ARMIJO (67): ⎝ 0 ⎠

while
a sJ sJ
J K y,i+1  J K y,i p ¸ , and the weighting matrix of the control inputs with
2 sK y,i sK y,i
⎛ ⎞
0 0 0 0
do
⎜ (xH )2
0 ⎟
a   ¸ a ,  1 (44) max
⎜ (δb)2
0 ⎟
⎜ max ⎟
a  ¸a ,  1 (48) R = ⎜
⎜ (yH )2
max
⎟
0 ⎟
⎜ (δa)2max ⎟
i  i+1 ⎝ symm. 0 ⎠
K y,i  K y,i+1

K y,opt  K y,i For an assumed intervall of maximum position devia-

q ij ,rkl
tions of
Q xR
y
K y,opt
(49) (xH , yH )max ∈ [5 m , 20 m]

Figure 9: Algorithm for the determination of linear and an intervall of maximum pendulum angles of
quadratic optimal output feedback regulation (50) (ϕP , ϑP )max ∈ [5◦ , 20◦ ]

©DGLR 2009 10
 35th European Rotorcraft Forum 2009

the following variation range for q55 and q77 applies, Figures 10 and 11 show the achievable minimal per-
relative to the remaining elements in Qy and R: formance index J ∗ for the longitudinal and lateral po-
sitioning of the reference flight case due to the varia-
(51) (q55 , q77 ) ∈ [0.06 , 16] tion of the elements in Qy and R. The test case de-
scribes a positioning of the helicopter by Δx = 20 m
The maximum allowed control deflections – the in the longitudinal and by Δy = 20 m in the lateral
AFCS limits the cyclic control by ±10% – account for: direction, respectively: The position range was chosen
arbitrary since it does not influence the resulting opti-
mal weighting. The goal is to determine the weighting
(δb)max ≈ 3 cm
(52) parameters q55 , q77 , r22 , and r33 , which lead to the
(δa)max ≈ 2 cm optimal positioning of the overall system according to
the performance criterion of equation (56).
Thus, the variation ranges for r22 and r33 are:
The far field in figure 10a shows a decrease of J ∗
towards small values of q77 and r22 . For the deter-
(53) r22 ∈ [2.6 , 41.5]
mination of the performance index of the longitudinal
positioning, the weighting parameters associated with
(54) r33 ∈ [4.9 , 8.0] the lateral positioning are set to one and thus, neu-
trally rated. Both optimization processes are there-
The optimal feedback gains for various flight cases fore regarded as decoupled. In doing so, the processing
are determined within the intervals of Qy and R by time can be reduced. However, tests regarding a cou-
optimizing the performance criterion. pled variation of the weighting elements led to nearly
congruent results. The minimal performance index is
attained at q77 = 0.35 and r22 = 1.6 (q.v. fig. 10b). In
case of the lateral positioning the weighting elements
5 Performance Analysis q77 and r22 are set to one, and the minimal perfor-
mance index J ∗ is accomplished by the variation of
For the overall system illustrated in figure 14 at V =
0 kts, mL = 4000 kg, and lS = 7 m the linear quadratic
optimal output regulation is adopted. In order to start ˜102 ˜102
34 27
the optimization algorithm the initial feedback gain
matrix is set up, which stabilizes the system in equa- 30 25

tion (29): J*
26 23
⎛ ⎞
0 0 0 0 0 0 0 0 22 21
1 ⎜ 2 0 0 0 0 −0.1 −0.1 ⎟ 40
4
6
(55) Ky,0 = ⎜
⎝ 0 0 1 2 0.1 0.1 0 0 ⎠
⎟ 0 4 8 12 16
20
r22
0 0.2 0.6 1.0 1.4
2
r22
q77 q77
0 0 0 0 0 0 0 0
(a) far-field range (b) close-up range
The optimal feedback controllers (K y,opt )qij ,rkl –
they have been determined in dependence on Qy and Figure 10: Determination of the minimal performance
R – are virtual-flight-tested on their ability to posi- index J ∗ of the longitudinal positioning by a variation
tion the overall system in order to eventually define of q77 and r22 at the reference flight case
(QxR)
the optimal controller K y,opt . The meaning of the
optimal positioning must be formulated by means of
another performance criterion. For this, the criterion
of equation (25) determining the system energy can ˜102 ˜102
40 28
be used. An alternative is given in the integration of
26
the absolute error of the position offsets and the load
J * 30 24
pendulum deflections leading to:
22
20 20
(56) 40
 ∞
5
10 20
0.5
1.0 2
4
6
q55 15 0 r33 q55 1.5 r33
J∗ = ΔxH  + ΔyH  + ΔϕP  + ΔϑP  dt
0 (a) far-field range (b) close-up range

Analyses of both criteria showed that the position-

ing of the overall system leads to different results. Figure 11: Determination of the minimal performance
Evaluating the integrated absolute errors yields more index J ∗ of the lateral positioning by a variation of q55
optimal results. and r33 at the reference flight case

©DGLR 2009 11
 35th European Rotorcraft Forum 2009

q55 and r33 . Again, a decrease of the index develops U0 σw intensity

and its minimum is achieved at q55 = 0.3 and r33 = 4.1
(q.v. fig. 11). [f t/s] [kts] [f t/s] [kts]
Eventually, the following optimal feedback control
gain is determined for the reference flight case: 3 1.78 0.1 0.06 weak
10 5.92 0.6 0.36 normal
(57) 18 10.66 3.6 2.13 medium
Ky,opt = 28 16.58 7.1 4.20 strong
⎛ ⎞
0 0 0 0 0 0 0 0
⎜ 0.32 1.44 0 0 0 0 0.07 −0.10 ⎟
⎜ ⎟ Table 2: Turbulence intensities
⎜ 0 0 0.27 1.09 0.05 0.13 0 0 ⎟
⎝ 0 0 0 0 0 0 0 0 ⎠

and listed in table 2, ranging from weak (< 2 kts) to

The optimal positioning of the overall system is il- strong interferences (> 15 kts) in dependence on [13].
lustrated in figure 12. The helicopter is displaced The results of the position hold are illustrated in
by 20 m longitudinally and laterally at the same figure 13 for four different turbulence intensities over
time. The integrated load pendulum sway regula- a period of 100 seconds. The different panes show the
tion is switched on in figure 12a and deactivated relative deviations of the helicopter position, the pen-
in dulum sway and the load position from the respective
 12b for comparison – hence, the  control elements
k(2,7) , k(2,8) , k(3,5) , k(3,6) ∈ K y,opt are set to zero. initial conditions. Furthermore, the longitudinal load
The upper panes show the relative cyclic control inputs deviation is plotted against its lateral displacement. In
from the AFCS superposed by the automatic position case of weak turbulences the load position can be held
control. The control limits of ±10% with reference to within a field of 20 cm x 20 cm edge length. This po-
the trim control are not exceeded. It can be seen in sition accuracy decreases to 40 cm x 40 cm for normal
figure 12a that the load sway is being damped during turbulences down to 4 m x 4 m in case of strong turbu-
the system displacement towards the target position. lences, rather due to the helicopter dynamic than the
By the time of the arrival at the final position the load pendulum oscillations.
pendulum motions are entirely damped and stabilized.
0.5 2
If the load sway is not considered in the feedback con- xH,g [m] 0 0
trol, the helicopter control will not be able to hold the -0.5 -2
0.5 2
position accurately due to a remaining load sway. yH,g [m] 0 0
In a next step, the ability to hold the load position -0.5 -2

under the influence of turbulences is briefly analyzed. 1 5

'M P,H ,b >q@ 0 0
The disturbances affect the helicopter and the con- -1 -5
tainer; different turbulence intensities are considered 1 5
'-P,H ,b >q@ 0 0
10 -1 -5
a [%] 0.5
5 2
b [%] 0 xL,pos [m] 0 0
-5 -0.5 -2

20 0.5 2
xH,g [m] yL,pos [m] 0 0
10 -0.5 -2
' x H ,g > m s @ 0 20 40 60 80 100 0 20 40 60 80 100
0
20 time [s] time [s]
yH,g [m]
10 0.5 3
' y H ,g > m s @ 0.3 2
0
1
0.1
'M P,H ,b >q@ 10
xL,pos [m] 0
0 -0.1
'M P,H ,b >q s @ -1
-10
-0.3 -2
'-P,H ,b >q@ 10 -0.5 -3
0 -0.5 -0.3 -0.1 0.1 0.3 0.5 -3 -2 -1 0 1 2 3
'-P,H ,b >q s @ yL,pos [m] yL,pos [m]
-10
0 10 20 30 40 50 0 10 20 30 40 50
weak medium
time [s] time [s]
normal strong
(a) integrated pendulum (b) integrated pendulum
damping active damping inactive (a) weak to normal (b) medium to strong

Figure 12: Optimal positioning of the overall system Figure 13: Position hold accuracy of a 4t-container in
by Δx = Δy = 20 m in the reference flight case turbulent atmosphere

©DGLR 2009 12
 35th European Rotorcraft Forum 2009

6 Conclusion and Outlook angular-moment-theory in equation (61). Therefore,

the dynamics of the overall system is described by the
Sling loads influence the helicopter dynamics. The states
handling qualities are degraded and the pilot workload xH = (u, v, w, p, q, r, Φ, Θ)H
is increased significantly due to the additional task to
control the load. Analyses revealed that the transla- xL = (u, v, w, p, q, r, Φ, Θ)L
 
tional dynamic of the helicopter is severely affected by xR = ϕ̇, ϑ̇
the sling load. The need for pilot support is therefore R

evident: The paper at hand focusses on the develop- The general nonlinear equation of the translational
ment of supplementary AFCS-modes, which generate motion is given by:

control inputs that combine the automatic helicopter dV 1 #
positioning with an integrated load sway control. (58) = · F Λ,b − ω Λ,b × V Λ,b
dt Λ,b mΛ
On the basis of a comprehensive system simula-
tion featuring trim calculation, linearization and vir- and of the rotation by:
tual flight testing, position controllers are developed  #
dω
by means of an automatic optimization algorithm us- = I −1
Λ,b · M CGΛ,b
dt Λ,b
ing linear quadratic optimal output regulation. Anal- (59)  
yses show that the sling load can be positioned opti- − ω Λ,b × I −1
Λ,b · ω Λ,b
mal in terms of the minimization of the time until the
target position is reached and of the load sway. The applying for the helicopter and the load (Λ = (H, L)).
LQ-controllers are effective within the operating range The rod is considered as additional body with two
of the AFCS-actuators regarding the limited rate and degrees of freedom:
saturation.   T
dω
In a next step, the controller algorithm and the dig- (60) = ϕ̈, ϑ̈
ital image processing system will be implemented into dt R,b R

the DLR system simulator and the Flying Helicopter The analytical modeling is based on angular-moment-
Simulator (FHS) in order to analyze the controller ef- theory for systems, whose reference point is neither its
fectiveness for different helicopter types and system center of gravity nor its fixed-point:
configurations, and to evaluate pilot acceptance.   −1 #
dω
= I LH
R,b · M LH
R,b
dt R,b
(61)
Appendix  
− mR · r LH→R × aabsLH,g
Rigid-body dynamics
The sum of the resulting moments with respect to the
In a first step, the helicopter and load are described load hook is derived by
separately as two independent six degree-of-freedom ⎡ ⎤
# sin ϕ cos ϑ
rigid bodies. The general equations of the nonlinear lR mR g ⎣
M LHR,b = − · cos ϕ sin ϑ ⎦
translational and rotatory motions are given by (58) 2
cos ϕ cos ϑ R
and (59).
The index Λ = (H, L) enunciates the compatibil- (62)
⎡ ⎤
ity of the equations for the helicopter and the load, 0
respectively. For the validity of the rigid body expres- + ⎣ 0 ⎦ × F sling
R
sion, following conditions apply: lR
• the earth is considered as initial frame The absolute acceleration of the helicopter load hook
in the geodetic system is obtained by
• the helicopter and the load are considered as
rigid bodies (63) aabs
LH,g = T M gb · aLH,b
H abs

and the absolute acceleration of the load hook in heli-

• the helicopter and the load are symmetric relat-
copter body-axes:
ing to the xz-plane, leading to Ixy = Iyz = 0
LH,b = aH,b + ω̇ H,b × r HCG→LH
aabs abs
• external forces are concentrated in resultant   
tangential acc.
forces acting in the respective center of grav-
ity + ω H,b × (ω H,b × r HCG→LH )
(64)   
centripetal acc.

The rod is considered as additional body with two + 2 · ω H,b × v rel rel
LH,b +aLH,b
  
degrees of freedom; its dynamics is determined by coriolis acc.

©DGLR 2009 13
 35th European Rotorcraft Forum 2009

u, p HCG
H ,b
w, r v, q H ,b
H ,b
LH

xg FHS ,g yg FHS ,g
zg zg
LH LH
- R
M R

R R
S S
FR,g FR,g
S S
FL,g FL,g
R R

S ,i S ,i
FL,g FL,g

APi APi
LCG LCG

u, p L ,b v, q L ,b
w, r L ,b
w, r L ,b

Figure 14: Overall system helicopter, rod and sling load

2
The distance between the helicopter center of grav- (70) Vxy = u2 + v 2
ity and the load hook remains constant. Hence, the
relative acceleration arel
LH,b as well as the coriolis accel-
2
(71) Vyz = v 2 + w2
eration, both become zero.
The external forces and moments in the equations v
(72) β = arcsin $
(58), (59), and (61) result from aerodynamics and the (u2 + v 2 + w2 )
sling forces.
Due to the symmetry of the container, CY = CZ
Aerodynamics applies for the aerodynamic polar.
The polar of the aerodynamic moment coefficient
The container load features an edge length of 2.4 m. CM against the angle of attack indicates that for small
Its aerodynamic coefficients have been derived in wind angles i = 0◦ = 90◦ and for i ≈ 15◦ , the moment be-
tunnel tests. The polars of the static aerodynamic co- comes zero [4]. For angles between 0◦ and 15◦ , the
efficients – which lead to a tangential, and two normal reaction moment is positive and hence stable; for all
forces, and a moment – were identified in [4]. The other angles, the moment is instable causing the con-
resulting aerodynamic forces are derived by tainer to spin around its yaw axis.
⎛ 2 2
⎞ The aerodynamic moments depend on the cross sec-
CX (αxz ) Vxz + CX (β) Vxy
⎝ CY (β) Vxy 2 2 ⎠ tion surface and the container length. They are given
(65) F AL,b = qC · + CY (αyz ) Vyz
2 2 with
CZ (αxz ) Vxz + CZ (αyz ) Vyz ⎛ ⎞
2
CM (αyz ) Vyz
considering (73) M A L,b = qC · lC ·
⎝ CM (αxz ) Vxz 2 ⎠
2
ρ CM (β) Vxy
(66) qC = · hC · wC
2
The moments of inertia of the container used in
(67) αxz = arctan (w/u) equation (59) are calculated with
mL  2 2

(74) IL,xx = h C + wC
(68) αyz = arctan (w/v) 12
mL  2 
2 (75) IL,yy = l + h2C
(69) Vxz = u2 + w2 12 C
mL  2 2

(76) IL,zz = l C + wC
12

©DGLR 2009 14
 35th European Rotorcraft Forum 2009

Constraining forces [5] iMAR-GmbH, Gesellschaft für inertiale Mess-,

Automatisierungs- und Regelsysteme. http://
The resultant vector of the sling forces in the respec-
www.imar-navigation.de
tive body system (Λ = H, R, L) is given by the trans-
formation of the vectors of the geodetic sling forces [6] Cicolani, L.S. ; McCoy, A.H. ; Tischler, M.B.
; Tucker, G.E.: Flight Time Identification of
# a UH-60A Helicopter and Slung Load, TM-1998-
(77) F SΛ,b = T M Λ
bg · F S,i
Λ,g
112231 / NASA Ames Research Center. 1998. –
i
Technical Report
derived by the vectorial description of the sling force
⎛ ⎞S,i [7] Hoh, R.H. ; Heffley, R.K. ; D.G.Mitchell:
% % sin ϑ · cos ϕ Development of Handling Qualities Criteria for
% %
(78) F S,i
Λ,g = %F
S,i
% · ⎝ cos ϑ · sin ϕ ⎠ Rotorcraft with Externally Slung Loads / NASA,
cos ϑ · cos ϕ g U.S. Army RDECOM. 2006. – Technical Report
which is determined for each sling due to its elongation [8] Iordanou, H.N. ; Surgenor, B.W.: Experi-
and elongation rate: mental Evaluation of the Robustness of Discrete
% %
% % Sliding Mode Control versus Linear Quadratic
(79) %F S,i % = cS,i ΔlS,i + dS,i l˙S,i
Control. In: IEEE Transactions on Control Sys-
The attitude of each sling is given by (q.v. fig.14): tems Technology 5 (1997), S. 254–260
S,i [9] Iordanou, H.N.: Continuous Versus Discrete
yAP − yR
(80) ϕS,i = − arctan Sliding Mode Control as Applied to a Pneumatic
g
zAP − zR g
xAP − xR
S,i Positioning System, Queen’s University, Canada,
(81) ϑg = − arctan
S,i
Diss., 1998
zAP − zR g
[10] Corriga, G. ; Guiua, A. ; Usai, G.: An Implicit
The moment vectors due to the sling forces in the
Gain-Scheduling Controller for Cranes. In: IEEE
helicopter and load body system are given by
⎡⎡ Transactions on Control Systems Technology 6,
⎤LH ⎡ ⎤HCG ⎤
x x No.1 (1998), S. 15–20
⎢ ⎥
(82) M SH,b = ⎣⎣ y ⎦ −⎣ y ⎦ ⎦ × F H,b
S
[11] Wang, Z. ; Surgenor, B.: Performance Evalu-
z H,b z H,b ation of the Optimal Control of a Gantry Crane.
and In: Proceedings of ASME International Mechani-
⎡⎡ ⎤AP,i ⎡ ⎤LCG ⎤ cal Engineering Congress, 2003
x x
⎢ ⎥
(83) M S,i
L,b = ⎣⎣ y ⎦ −⎣ y ⎦ S,i
⎦ × F L,b [12] Wang, Z. ; Surgenor, B.: A Problem with the
z L,b z L,b LQ Control of Overhead Cranes. In: Journal of
Dynamic Systems, Measurement and Control 128
(2006), S. 436–440
References [13] Lusardi, J. A.: Control Equivalent Turbulence
Input Model for the UH-60 Helicopter, University
[1] Brenner, H.: Flight Testing of Pioneer Bridges of California, Diss., 2004
as Helicopter Slung Loads Using a CH-53G. In:
[14] Lunze, J.: Regelungstechnik 2, Mehrgrößensys-
Proceedings of the 33rd European Rotorcraft An-
teme, Digitale Regelung. 4. Auflage. Springer-
nual Forum, 2007
Verlag, 2006
[2] Brenner, H.: Simulation Model for Helicopter
[15] Yan, W.Y. ; Teo, K.L.: A Gradient Flow Ap-
Cargo Handling Investigations. In: Proceedings
proach to Computing LQ Optimal Output Feed-
of the 7th ONERA-DLR Aerospace Symposium,
back Gains. In: Optimal Control Applications and
2006
Methods 15, No.1 (1994), S. 67–75
[3] Heffley, R.K. ; Jewell, W.F.: A Compila-
[16] Levine, W.S. ; Athans, M.: On the Determi-
tion and Analysis of Helicopter Handling Qual-
nation of the Optimal Constant Output Feedback
ities Data - Volume One: Data Compilation /
Gains for Linear Multivariable Systems. In: IEEE
NASA Contractor Report 3144. 1979. – Tech-
Transactions on Automatic Control AC-15, No.1
nical Report
(1970), S. 44–48
[4] Ronen, T.: Dynamics of a Helicopter With a
[17] Shapiro, E.Y. ; Fredricks, D.A.: Lecture Notes
Sling Load, Department of Aeronautics and As-
in Control and Information Sciences, paper: Ap-
tronautics of Stanford University, Diss., August
plication of Constrained Constant Optimal Out-
1985
put Feedback to Modern Flight Control Synthesis.
Springer-Verlag, 1982

©DGLR 2009 15