HEADER HEIGHT CONTROL OF A COMBINE HARVESTER SYSTEM

Yangmin Xie Ashley Greer

Andrew Alleyne Dustin Deneault
Mechanical Science and Engineering Department John Deere
University of Illinois at Urbana-Champaign Champaign, IL 61820-7484
Urbana, IL 61821 [email protected],
[email protected], [email protected] [email protected]

ABSTRACT approximately 75% of the crop losses occur at the header [1][2]
There are appreciable losses in crop output at all the stages and a significant ratio of the header loss is caused by improper
in grain production like harvesting, transportation, storage, etc. To setting of the header height. Figure 1 shows a schematic of a
reduce grain loss during the harvesting process, it is desirable to combine harvester system. The header height is defined as the
effectively control the header height of the combine harvester. The distance between the header tip and ground. By raising or lowering
desired header height control is able to follow the shape of the the header with an actuator, usually hydraulic, the header height
terrain and reject disturbances from the ground. This paper can be adjusted. However there is a delicate balance to the process.
presents an optimal state feedback LQR controller to achieve the If the header height is too large, there is a reduction in harvest
control objectives. By properly defining the states and choosing the yield since much of the viable crop will be left un-harvested.
cost function, the full state feedback controller is shown to Conversely, if the header height is maintained at too low a level,
accomplish the control goals. Then a reduced states feedback equipment damage or operator fatigue will result. This problem is
controller, which uses a skyhook damper in the controller, is used exacerbated by the fact that combine systems typically have low
to simplify the full state feedback controller. Simulation results frequency dynamics due to the lightly damped tires acting as
show that with less feedback information, the reduced order state springs, shown in Figure 2. The vehicle‘s natural frequencies [3]
feedback controller preserves the ability to achieve good reference also change somewhat with speed and soil condition thereby
tracking and disturbance rejection. posing a further challenge to any purely mechanical mitigation
strategy. The ground profile in a field contains two types of
spectral content: a relatively low-magnitude high-frequency
ground disturbance superposed on a relatively high-magnitude
INTRODUCTION low-frequency terrain modulation. Therefore, it is desirable for
With a growing world population, agriculture continues to be header height to be able to adjust and follow the terrain shape
an important societal need requiring attention by the technological while simultaneously suppressing the effect of the ground
community. Gains in modern agriculture can be attributed to disturbances on the cutting head.
several factors, one of which is increased mechanization and
One possible solution to the so-called header height problem
automation of the entire agricultural process. A key facet of the
is a terrain-following automatic control system to maintain header
overall mechanization is the improved functionality of modern
height. This has been a goal of agricultural mechanization and
machines that work to plant, monitor, harvest, and condition the
automation engineers since the 1960‘s [5]. Early approaches used
fields of operation. The modern combine is an example of such a
simple header control systems to adjust the header height so as to
machine that creates an essential link in the overall food chain to
be proportional to crop height [4]. Typically, these simple
allow for rapid recovery of crops from the field. While it has
controllers are proportional-type controllers with an input dead
evolved significantly over time, there are still several areas in
zone operating around the set point. With relatively little
which improvements through automation could be made to the
sensitivity outside the dead zone, this corresponds to an equivalent
combine. One of these is the header height control problem.
hysteretic on-off control approach in practice. Although these
The header height control problem is a long standing issue in simple controllers may be sufficient for static tests or for operating
the combine harvester industry. It has been estimated that on smooth terrain, typical fields are sufficiently irregular as to

1
excite combined dynamics, thereby deteriorating the header motion header height against gravitational forces. Section 4 develops a
tracking performance. reduced order version of the original LQR approach by noting,
similarly to [8], that a small set of feedback terms comprise the
bulk of the information used in the control inputs. Additionally,
Section 4 illustrates the connection between the particular LQR
feedback terms and more traditional Proportional-Integral-
Derivative (PID) approaches. Section 5 shows simulation results of
the proposed controller, both full LQR and reduced states LQR to
demonstrate the potential of the presented controllers. Section 6
concludes the paper.

SYSTEM MODELING
Figure 1: Schematic of a combine harvester system In this section, the combine system is modeled in a way that is
convenient for a further LQR controller design. Figure 2 presents
Unfortunately, this highly relevant problem has received little
the simplified system structure. It has three degrees of freedom: the
attention from the research community. One of the few
vertical and rotational motion of the combine body and the rotation
investigations to utilize modern control techniques introduced an
of the header around the attached point A with respect to the
optimal header height control method to automatically track
combine body. The header is rotated by a torque applied by the
changing terrain shape[6]. A Linear Quadratic Gaussian (LQG)
hydraulic actuator. The changes in ground height under the tires, zr
controller was used to design a state feedback controller with a
and zf, are considered as unknown disturbances to the system, and
Kalman filter to estimate the states information. The results
the ground height zh at the header position is the reference for
obtained good tracking performance in simulation when tracking a
tracking. So the header height in this paper means the distance
ground profile with spectral content under 1 Hz. This method,
between header tip and the ground. The pitch angle  is assumed
however, requires an accurate system model to construct a good
state estimator and, additionally, it did not display the ability to to be small for the model linearization, and half of the combine
reject high frequency disturbances effectively. system is modeled here due to its symmetry.

Previous efforts in the header height control community

focused mainly on the terrain following with less effort towards
disturbance suppression. Fortunately, similar vibration isolation
control has been studied extensively in the passenger vehicle active
suspension control problem. To isolate a vehicle from a ground
vibration disturbance, researchers used active control schemes to
adjust the suspension characteristic or forces between the tire and
vehicle body. This has been repeatedly demonstrated to
significantly decrease the vibration of a vehicle subjected to a
broad frequency spectrum of disturbances. One way to perform
this isolation of vibration is to design an optimal controller which
includes the velocity of the isolated mass as a term in the cost
function [7]. The work in [8] demonstrated that simple absolute
velocity feedback of the isolated mass (termed skyhook damper) Figure 2: Schematic of the mathematical system model
can provide many of the benefits of full state feedback while
requiring fewer state measurements [8]. In fact, the skyhook The force analysis for the combine system is illustrated in
damper approach has found numerous applications in the semi- Figure 3. To simplify the analysis, the force from the hydraulic
active and active suspension controls literature [9]. By analogy to actuator is equivalently transformed as a torque T on the header,
the automotive active suspension problem, the header can be and forces FAx and FAy from the combine body through the
treated as the isolated mass, and it is desired to isolate the header attachment point A. The header can be lifted or lowered by
from the disturbance vibrations coming from the combine body. adjusting the torque; therefore the torque is considered as the
Therefore, a skyhook damper approach is an attractive option to control input for the header height control design. The dynamics of
perform the required disturbance rejection in the header height the combine body in the proceeding direction are ignored because
control case. This would then need to be paired with a terrain the combine moves with a prescribed velocity. The Eq. of motions
following approach to achieve an overall header height solution. are derived using Newton‘s second law and are given in Eq. (1)–
The rest of the paper is organized as follows. In Section 2, the (5).

 
combine harvester is abstracted as a three degree of freedom
system, and a dynamical model for the overall combine harvesting I A  k f (a  z f  v)  b f (a  z f  v) a
system is established, including ground disturbances. In Section 3,
the general LQR method is used to design a state feedback

 kr (b  zr  v)  br (b  zr  v) b  (1)

controller taking into account the constant force needed to maintain T  FAxlt1 sin t1   FAy lt1 cos t1 

2
 
mAv  k f (a  z f  v)  b f (a  z f  v)  (2)
x1   lt1 cos t1   l2 cos  ss  v  zh
x2   lt1 cos t1   l2 cos  ss  v
 
 kr (b  zr  v)  br (b  zr  v)  mA g  FAy
x3  a  z f  v (8)
 
mh  l2 sin   lt1 sin t1   FAx (3)
x4  b  zr  v

m  v  l  cos(
h t1 t1 
)   l2 cos   FAy  mh g (4) x5  
x6  v
I h   T  mh gl2 sin  (5)
where x1 is the header height difference with respect to the set
point.; x2 is the header vertical velocity; x3 is the front tire
deflection; x4 is the rear tire deflection; x5 is the combine body
vertical velocity; x6 is the combine body rotation velocity.
Substituting Eq. (8) into the original nonlinear Eq. (1)-(5), and
rearranging terms yields the linear system equations below.

I A x5   k f x3  b f x3  a   kr x4  br x4  b
(9)
T  FAxlt1 sin t1   FAy lt1 cos t1 

mA x6   k f x3  b f x3    kr x4  br x4   FAy (10)

Figure 3: Force analysis.  

mh  tan   ss   x5lt1 cos t1   x6  x2   lt1 x5 sin t1   FAx (11)

The header height is described in Eq. (6). By assuming that  

mh x6  lt1 x5 cos(t1 )   x5lt1 cos t1   x6  x2   FAy (12)
the set point of the header height is held at 0.2m and the ground is
flat, the steady state values for the system are given in Eq. (7). I h  x5lt1 cos t1   x6  x2  (13)
 T
l2 cos  ss
y  h0  lt1 sin t1     sin  l1  v  zh (6)
x1  x2  zh (14)
 ss  0.55929 vss  0.054368  ss  0.0023478
(7)
FAy , ss  49715 Tss  42.848 FAx , ss  0 The linearized system can be described using a state space
realization as shown in Eq. (15).
Six dynamic states are chosen to describe the combine system, x  Ax  Bu  Ed
as shown in Figure 4, and Eq. (8) presents the linearized (15)
relationship between each state and the position and velocity of the y  Cx
system.
Where
0 1 0 0 0 0 
0 0 194.94 10.788 3.4247 1.84 

0 0 0 0 1.5226 1 
A 
0 0 0 0 2.0048 1 
0 0 32.901 28.107 1.441 0.1192 
 
0 0 114.33 58.075 0.10574 2.1338

 0   1 0 0 
 0.00026129   0 2.0166 0.17666 
   
 0  0 1 0 
B  E  
 0  0 0 1 
Figure 4: Definition of states.  0.000080964   0 0.34036 0.46027 
   
 0.000079301  0 1.1827 0.95101 
C  1 0 0 0 0 0
(16)

3
 d   zh zr  ability of an output feedback controller to place the close loop
T
u  T zf (17)
poles. By root locus design, there are always close loop poles and
zeros that are near these open loop poles and zeros, but they will
x is the dynamic state vector, the torque T is the control input
not cancel each other when considering the ability of the real
u, d includes the derivative of reference and disturbances, and y is
actuator . So the output tracking bandwidth is bounded by the
the system output, which is equal to x1. The matrices A, B, C and E
natural frequency of the combine body (by the poles of (sI-A22)-1),
are obtained using parameters from a real combine harvester
which is around 1 Hz in this particular model. This limitation can
system.
also be intuitively understood. If the irregular ground profile
For this under-actuated system, the torque input can not contains a high frequency disturbance, e.g. a bump, the bump will
arbitrarily adjust the header height because the transfer function cause the vibration of the combine body around its natural
from the torque input to the header height output is influenced by frequency. The combine body will then pass this energy to the
the combine body‘s dynamics. To clearly illustrate this header and induce large header height error. So the output
relationship, the state space system in Eq. (15) is rewritten as Eq. feedback controller can generally track low frequency references,
(18). States z1 are related to the header height, and states z2 but does not have the ability to track high frequency references,
represent the combine body dynamics. and therefore lacks the ability to reject high frequency
disturbances. The proposed state feedback controller will address
 z1   A11 A12   z1   B1   E1  this problem.
      u   d
 z2   0 A22   z2   B2   E2  (18)
z  3. LQR CONTROLLER DESIGN
y  C1 0  1  The following section applies the linear quadratic regulator
 z2  (LQR) method to generate a state feedback controller. LQR is a
common optimal control approach in linear systems. The cost
where, function is developed with a quadratic function with respect to the
z1   x1 x2  z2   x3 x6  states and control inputs, and an optimal state feedback controller
T T
x4 x5
can be obtained through minimizing the cost function.
0 1   0  (19)
A11    A12    However, if the controller is designed just based on the
0 0   A122  linearized system, there will be a DC offset in the output because
B1   0 B12  C1  [1 0]
T
of the gravity induced steady state error between the real system
and the linearized system. An integral controller is used to
The open-loop transfer function from the torque input to the eliminate the steady state error. Therefore, an augmented system is
system output is derived based on Eq. (20), and given in Eq. (21). introduced to implement the LQR method.
y
u
1

 C1  sI  A11  A12  sI  A22  B2  B1
1
 (20) Define an extra state x7 by Eq. (23):

x7  t    x1  t  dt
t
(23)
0
y 1
  (Guh1  Guh 2 ) Then the system can be extended, to design the LQR
u s2
controller, the disturbance matrix E can be ignored
Guh1  B12 (21)
x  Ax  Bu  Ed
Guh 2  A122  sI  A22  B2
1
(24)s
y  Cx
y 1  s  1.6 + 17i  s  1.6 - 17i 
 
x   x x7 
T
u s2  s  1 + 13i  s  1 - 13i  (22)
Where


 s  0.97 + 10.8i  s  0.97 - 10.8i   0
 s  0.8 + 10i  s  0.8 - 10i   A66 0  B
A B  
  0
where Guh1 is the transfer function from the torque input  
directly to the vertical header height acceleration x2 , while Guh2 is 1 0 0
E 
the transfer function from the torque input to x2 through the E  C  C 0
dynamics of combine body. Eq. (21) shows how the combine 0
body dynamics influence the effect of the torque on controlling The cost function of the LQR is given in Eq. (25), and the
header motion. feedback control law is described in Eq. (26).
By further calculation (Eq. (22)), it can be shown that the 

poles and zeros of Guh1+Gu2h , which are also the poles and zeros J   [ x '  t  Qx  t   u '  t  Ru  t ]dt (25)
0
the of open loop system, are close to each other; this limits the

4
 u   Kx (26) K  1.09 106 1106 8.08 105 40070 -17437 1105  (33)
The matrices Q and R are the weighting matrices for the
Bode Diagram
dynamic states and the control input, respectively. Recall that the
dynamic state x1 is the tracking error between the reference and 0
measured height, and the state x2 is the vibration velocity of the

Magnitude(dB)
header. Therefore, for the control objective of the reference -50
tracking and disturbance rejection, the cost function should Gh
emphasize the minimization these two states. The other four states -100 Gf
are related to the combine body movement which do not influence Gr
the control objectives, so they are not weighted heavily in the cost -150
-2 -1 0 1 2
10 10 10 10 10
function. Eq. (27) presents the construction of matrices Q and R for
the state feedback controller design.
200

 q1 0 0 0 0 0 0
0 q2 0 0 0 0 0 
100

Phase(deg)

0 0 0.1 0 0 0 0 Rr (27) 0
 
Q  0 0 0 0.1 0 0 0 -100
0 0 0 0 0.1 0 0 
  -200
0 0 0 0 0 0.1 0  10
-2 -1
10 10
0 1
10
2
10
0 0 q7  frequecy(rad) (rad/sec)
 0 0 0 0
The balance between the reference tracking and disturbance
Figure 5: Closed-loop Bode plots of full states LQR controller with
rejection can be realized by choosing different q1 and q2 in the
q1 = 104, q2 = 104, q7 = 102, r=10-8
matrix Q. If the priority is to keep a more precise tracking to the
ground profile, q1 should be increased; if suppressing vibration of
the header is more critical, q2 should be increased. q7 can be chosen In the low frequency range (smaller than 1 rad / sec ), the
as a relatively small value if the DC drift is assumed to be small. magnitude of transfer function Gh is close to 0dB, and the phase is
Assuming that there is no actuation limitation for the actuator in close to 0 degrees, which means that the system can track the
the combine system, the gain r in the matrix R can be chosen as ground profile within this frequency range. When the frequency is
small value. higher than 1 rad/sec, the magnitude of Gh, decreases, signifying
that the header does not track the signal at high frequencies. This is
By rewriting the system state-space representation as shown in
desirable because we do not want the header to respond to the
Eq. (28), we obtain the transfer functions from the reference zh and
bumps and noise of the ground profile. The magnitudes of Gf,, and
the disturbances zf and zf to the output y (Eq.s (29) — (31),
Gr remain low (much smaller than 0dB) at all frequencies, which
respectively). The system performance with the designed state
means the controller can successfully isolate the combine body
feedback LQR controller can then be evaluated through frequency
vibration caused by ground disturbances. The ground height
analysis.
changes at the front and rear tires show little influence to the
x   A  BK  x  E1 zr  E2 z f  E3 zr header height.
(28)
y  Cx Also it can be shown that adjusting q1 and q2 influences the
trade-off between the two control objectives. By reducing q2 (Eq.
 (C  sI  ( A  BK )  E1  1)s
y 1 (34)), the LQR algorithm tends to put more control effort
Gh  (29)
towards tracking the ground profile, and the tracking bandwidth
zh
will increase while poorer disturbance rejection performance is
 C  sI  ( A  BK )  E2 s
y 1 observed (see Figure 6).
Gf  (30)
zf q1  1104 q2  1103 q7  1102 r  1108 (34)

 C  sI  ( A  BK )  E3 s
y 1
Gr  (31)
zr
By appropriately choosing the coefficients (see Eq. (32)) in
the weighting matrices Q and R, the resulting state feedback
controller in terms of the optimal control gain K is presented in Eq.
(33). The Bode plots of the transfer functions Gh,, Gf,, and Gr are
shown in Figure 5.

q1  1104 q2  1104 q7  1102 r  1108 (32)

5
 Bode Diagram feedback controller is simply by eliminating the other four states
feedback from the LQR controller shown in section 3. To justify
0
this reduction, a comparison of the Bode plots between the full
Magnitude(dB)

-50
states and reduced states feedback controller are shown in figure 7
and figure 8. Their tracking abilities are nearly identical except that
Gh
-100 the Bode plot of the reduced states feedback controller has a small
Gf
Gr
deviationaround the natural frequency of the combine body. The
-150 disturbance rejection ability of the reduced states feedback is
-2 -1 0 1 2
10 10 10 10 10 compromised slightly (Figure 8). However, it is acceptable because
the maximum value of the magnitude is still less than -15dB, which
200
means the largest gain from disturbance to output is less than 0.2.
100 Furthermore, by tuning q1 and q2, this value can be further
Phase(deg)

decreased.
0
Bode Diagram
-100 10

Magnitude(dB)
-200
-2 -1 0 1 2
10 10 10 10 10 -10
frequecy(rad) (rad/sec)
-20
Gh of full states feedback controller
-30
Figure 6: Closed-loop Bode plots of full states LQR controller with Gh of reduced states feedback controller

q1 = 104, q2 = 103, q7 = 102, r=10-8 -40

-2 -1 0 1 2
10 10 10 10 10

4. REDUCED STATES LQR CONTROLLER

0
As is shown above, a full state feedback controller designed
by LQR method can accomplish the two control objectives. Phase(deg) -50
However, in practice, measurement of all the states is not practical.
Therefore, this section presents a reduced states LQR controller, -100
which preserves the ability of a full state feedback control with
fewer requirements for the feedback information. -150
-2 -1 0 1 2
10 10 10 10 10
The reduction of the states is from the concept called skyhook frequecy(rad) (rad/sec)

damping, which is widely used in active suspension problem as a

way to isolate vibrations [9]. According to [8], a skyhook damper Figure 7: Comparison of Bode plots for Gh between full states and
can capture the vibration isolation property of the full state reduced states feedback controller with q1 = 104, q2 = 104, q7 = 102,
feedback controller. A skyhook damper is simply a feedback r=10-8
controller using the velocity of the isolated mass. In the header Bode Diagram
height control problem, we want to suppress the vibration of the 0

header, so the header works as the isolated mass and a skyhook -50
Magnitude(dB)

damper in this case should use header velocity as feedback

-100 Gf of full states feedback controller
information. By inspecting the full states LQR controller, the Gr of full states feedback controller
feedback of state x2 works exactly as the skyhook damper. In -150 Gf of reduced states feedback controller

addition, the controller should keep the ability to track terrain -200
Gr of reduced states feedback controller

-2 -1 0 1 2
shape and to eliminate steady state offset, so this skyhook damper 10 10 10 10 10

should combine a tracking controller and the integral controller 200

mentioned in section 3. And the feedback of states x1, x7 work
exactly for these two purposes. 100
Phase(deg)

0
This combination results in a reduced states feedback
controller including a PI output feedback controller and a skyhook -100

damper (Eq. (35)). The PI controller uses the tracking error as its -200
-2 -1 0 1 2
feedback information to track the ground height zh. The skyhook 10 10 10
frequecy(rad) (rad/sec)
10 10

damper uses the vertical velocity of the header x2 to isolate the

vibration caused by disturbances.
Figure 8: Comparison of Bode plots for Gf,, Gr between full states
u  K1 x1  t   K 7 x7  t   K 2 x2  t  and reduced states feedback controller with q1 = 104, q2 = 104, q7 =
(35)
102, r=10-8
 K1e  t   K 7  e  t dt  K 2 x2  t 
Compared to the general form of a PID controller (Eq. (36)),
the reduced states feedback controller appears similar. There are
By the states reduction, only two variables need to be
standard proportional, integral and derivative terms. A key
measured. The way to get the feedback gains for the reduced states

6
difference is that the derivative gain acts on the absolute header
1
velocity, not on the rate of change of header height error. This
difference mitigates the effect of combined induced vibration on 0.8

the feedback signal. 0.6

u  K p e  t   Ki  e  t dt  K d e  t 
0.4
(36)
0.2

Height (m)
0

SIMULATION RESULTS -0.2

Simulation results based on the state feedback control -0.4
algorithms discussed above are provided in this section. The terrain -0.6
shape that we want to track is assumed to have a frequency smaller
-0.8 Ground profile
than 0.2 Hz, so the controller is designed to track signals below Desired header height
-1 Header height by reduced states feedback controller
this frequency and reject disturbances higher than this frequency. Header height by full states feedback controller
Considering the frequency spectrum of the ground profile, the 0 5 10 15 20 25 30
overall ground disturbance is assumed to be a combination of three Time (s)
sinusoidal signals w1, w2, w3 (Eq. (37)). w1 represents the changes
in terrain shape (generally with large amplitude and low Figure 9: Header height control performance comparison between
frequency), and w2 and w3 are caused by bumps and measurement reduced and full states feedback controllers
noises of the ground (generally with small amplitude and high 0.8
frequency). The desired header height with respect to ground is set
to be 0.15m. 0.78

w1  0.6sin  0.05  2 t 
0.76

0.74

w2  0.05sin 1.5  2 t  (37)

Height (m) 0.72

w3  0.02sin  5  2 t  0.7

0.68
The Q and R matrices are configured as Eq. (34). By the
0.66
corresponding Bode plot of Gh,, Gf and Gr (Figure 7, Figure 8), the
reduced states feedback controller has a tracking bandwidth around 0.64
Desired header height
0.2 Hz, and the magnitude of Gf and Gr are low in high frequency 0.62 Header height by reduced states feedback controller
Header height by full states feedback controller
range, thus it is capable to track w1 and reject w2 and w3. 0.6
5 5.5 6 6.5 7
Figure 9 shows the performance of the full state feedback Time (s)
controller and the reduced states feedback controller.
Figure 10 show that the performance of the reduced states Figure 10: Detail plots for Figure 9
feedback controller closely approximates the performance of the
1
full state feedback controller.
0.8
The limitation of output feedback control in the combine
0.6
system was discussed in Section 2. To further compare the
performance between a reduced state feedback control and output 0.4

feedback control, the performance of a PID controller (Eq. (36)) is 0.2

Height (m)

shown in Figure 11. The header vibrates significantly, and such a 0

high frequency vibration in real operation will cause operator
-0.2
fatigue and even equipment damage.
-0.4

-0.6

-0.8 Header height by PID controller

Desired header height
-1 Ground profile

0 5 10 15 20 25 30
Time (s)

Figure 11: Header height control performance using PID controller

7
 To prove the necessity of the integral controller, the controller has very a similar control performance as a full state
performance of the reduced states feedback controller is compared feedback controller.
to the one without integral controller. Although it is still able to
reject the high frequency disturbance, there is a large deviation
ACKNOWLEDGMENTS
when it tries to follow the terrain shape (Figure 12). This is more
The authors appreciate the technical and financial support by
obvious when the combine is going up hill. It is because the header
the John Deere Company.
requires a larger steady torque Tss (see Eq. (7)) to maintain the
required header height when going up hill. Without the integral
controller to compensate this offset of the Tss, the actual header REFERENCES
height is smaller than desired value. [1] J . L . Glancey, ―Analysis of Header Loss from Pod Stripper
Combines in Green Peas,‖ J . agric . Engng Res .vol 68 ,pp. 1 –
10, 1997.
0.8
[2] G. R. Quick; W. F. Buchele. ―Reducing Combine Gathering
0.6 Losses in Soybeans,‖ ASAE No 72-625, 1974
[3] G. E. Rehkugler, ―Dynamic Analysis of Automatic Control of
0.4 Combine Header Height‖, Transactions of the American Society
of Agricultural Engineers, vol 13 (2), pp. 225-231, 1970
0.2
[4] G. S. Pask, J. N. Wilson, G. C. Zoerb. ―Automatic Header-
Height Control System for Windrowers,‖ Transactions of the
Height (m)

0
American Society of Agricultural Engineers, vol 17 (4), pp. 597-
-0.2 602, 1974
[5] T. L. Kaminski and G. C. Zoerb. ―Automatic Header-Height
-0.4 Control for Grain Crops,‖ Transactions of the American Society
-0.6
of Agricultural Engineers, pp. 284-287, 1965
[6] G. T. Lopes, P. S. G. Magalhaes, E. G. O. Nobrega. ―Optimal
Header height by reduced states controller without integral term
-0.8 Header height by reduced states controller Header Height Control System for Combine Harvesters,‖
Desired header height Biosystems Engineering, vol 81(3), pp 261-272, 2002.
-1 Ground profile [7] A. G. Thompson , ‗An Active Suspension with Optimal Linear
0 5 10 15 20 25 30 State Feedback‘. Vehicle System Dynamics. vol 5, pp. 187-
Time (s) 203, 1975.
[8] C. Yue, T. Butsuen, J.K. Hedrick, "Alternative control laws for
Figure 12: Header height control performance comparison between automotive active suspensions", ASME journal of Dyn. Sys., vol.
controllers with and without integral feedback term 111, pp. 288, 1989.
[9] Hrovat, D. ‗Survey of advanced suspension developments and
CONCLUSION related optimal control application. Automatica‘, vol 33(10), pp
This paper presented a full state feedback controller designed 1781–1817, 1997
by the LQR method for solving the header height control problem.
To achieve the two control goals - tracking terrain shape and
rejecting high frequency disturbances - the cost function in the
LQR controller design is defined using tracking error and header
vibration velocity. High tracking bandwidth usually leads to less
disturbance suppression. Therefore, by adjusting the parameters
q1 and q2 in the Q matrix, the trade off can be balanced according
to design requirements. A reduced states feedback controller is
then proposed to simplify the LQR controller. A LQR controller
requires full state information, and some states, like the deflection
of the tire, are hard to measure in real applications. The reduced
states feedback controller, on the other hand, needs only two states
measurements, the tracking error and the vertical velocity of the
header. Techniques for measuring tracking error are already mature
in current combine systems, and the vertical velocity of the header
can be obtained by applying an accelerometer. The proportional
term in the reduced states feedback controller works to track
terrain shape; the derivative controller (skyhook damper) works to
reject disturbances; and the integral controller works to eliminate
steady state offset. The frequency domain analysis and time
domain simulations show that with only tracking error and header
velocity as feedback information, the reduced states feedback