Fundamental Limits in Combine distance between the header tip and ground.

By raising or lower-
ing the header with an actuator, usually hydraulic, the header
Harvester Header Height Control height can be adjusted. If the header height is too large, there is a
reduction in harvest yield since much of the viable crop will be
left unharvested. Conversely, if the header height is maintained at
Yangmin Xie too low a level, equipment damage or operator fatigue will result.
Mem. ASME The primary solution approach taken to date is the feedback
e-mail: xie3@illinois.edu system depicted in Fig. 2. Some look-ahead feedforward
approaches have been attempted in industry using laser, ultra-
Andrew G. Alleyne sonic, and radar sensors. However, each has been sufficiently
challenging as to preclude introduction in practice. The accepted
e-mail: alleyne@illinois.edu
feedback sensor is usually a “feeler” that drags along the ground.
Rotation of the feeler relative to the header mount is measured
Mechanical Science and Engineering Department,
and translated into header height as shown in Fig. 1. The goal is to
University of Illinois at Urbana-Champaign, design the controller CðsÞ in Fig. 2 so as to regulate the header
Urbana, IL 61801 height to a prescribed setpoint in the face of unknown ground
disturbances. Physically, the ground disturbances are induced by
the interaction between the ground and the flexible tires; when the
Ashley Greer combine is driving over irregular terrain, the changing ground
e-mail: GreerAshleyE@johndeere.com height works as disturbance signal to excite the vibration of the
combine body. Therefore, the header height control can be seen as
Dustin Deneault the ability to have the header track the changing profile of the ter-
e-mail: deneaultdustind@johndeere.com rain at the header tip, and at the same time reject the disturbances
under the tires. The higher the closed loop bandwidth (frequency
John Deere Company, of disturbance that can be rejected or frequency of reference that
can be tracked), the more rapid a change in terrain can be accom-
Champaign, IL 61820-7484
modated and the faster the vehicle can traverse the field. These all
lead to greater efficiency and productivity. Header height control
has been a challenging issue in industry for decades, and hence
This paper investigates fundamental performance limitations limited harvesting speeds have occurred as a result.
in the control of a combine harvester’s header height control While relevant, this control problem has received relatively lit-
system. There are two primary subsystem characteristics that tle attention from the research community. Early approaches of
influence the achievable bandwidth by affecting the open feedback control were proportional-type controllers with an input
loop transfer function. The first subsystem is the mechanical con- dead zone operating around the set-point [3]. One of the few
figuration of the combine and header while the second subsystem recent investigations to utilize modern control techniques intro-
is the electrohydraulic actuation for the header. The mechanical duced a linear quadratic Gaussian controller to automatically track
combine þ header subsystem results in an input–output represen- changing terrain shapes [4]. Another reduced order state feedback
tation that is underactuated and has a noncollocated sensor/actu- controller was proposed by using a sky hook damper to simplify
ator pair. The electrohydraulic subsystem introduces a significant an optimal full state feedback controller and reject the output dis-
time delay. In combination, they each reinforce the effect of turbance [5]. The feedback control in Refs. [4] and [5] works well
the other thereby exacerbating the overall system limitation of in simulation at relatively low frequencies: below 1 Hz. Field tests
the closed loop bandwidth. Experimental results are provided to illustrate that the achievable bandwidth of a header height control
validate the model and existence of the closed loop bandwidth system is usually much lower in practice [6]. However, to increase
limitations that stem from specific system design configurations. the working efficiency and obtain desired header height control
[DOI: 10.1115/1.4023209] performance at the same time, a closed loop bandwidth well above
1 Hz is demanded by modern machines. This closed loop band-
1 Introduction width is desired to accommodate terrain variations resulting from
With the world population increasing over the next several dec- combine forward motion as depicted in Fig. 1. For a desired
ades, agriculture will be called upon to provide greater yields in vehicle speed of approximately 7 miles per hour, which is at the
food production with relatively little increase in land usage. upper limit of current harvesting speeds, the desired closed loop
Therefore, it is imperative that efficiencies associated with auto- bandwidth specification is 3 Hz or better.
mation become part of the overall solution. A key aspect is the In this article, the authors explore and explain the fundamental
machinery used to perform the agricultural tasks; one example of causes of the bandwidth limitations in the feedback control of
this, the combine harvester system, is discussed in this article. The the header height system. The rest of paper is organized as
combine harvester is used to extract the crops from the field and
during this harvesting process seed loss is a critical problem [1]. It
has been estimated that approximately 75% of the crop losses
occur at the header [2] and a significant portion of the header loss
is caused by improper setting of the header height. Therefore, the
header height control problem under study is motivated by the
interest in improving the efficiency and productivity of the har-
vesting process, specifically to increase the harvest yield and
decrease the total harvest time.
Figure 1 shows a schematic of a combine harvester system
operating in the vertical plane. The header height is defined as the

Contributed by the Dynamic Systems Division of ASME for publication in the

JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROL. Manuscript received
May 10, 2011; final manuscript received July 20, 2012; published online March 28,
2013. Assoc. Editor: Nariman Sepehri. Fig. 1 Combine system

Journal of Dynamic Systems, Measurement, and Control MAY 2013, Vol. 135 / 034503-1
C 2013 by ASME
Copyright V

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 2 Schematic of feedback header height control

follows. Section 2 introduces the models for the combine system The combine system discussed in this paper is such a typical
shown in Figs. 1 and 2. The two subsystems that are most rele- underactuated system, which can be simplified as the planar multi-
vant to the control limitations are presented: (i) the mechanical body system shown in Figs. 3 and 4. As such, it contains two rigid
subsystem and (ii) the hydraulic actuation subsystem. Section 3 bodies: the combine body and the header. There are three DOFS
utilizes the models of Sec. 2 and presents an analysis explaining with one actuator amounted between the header and the combine
the performance limitation. Section 4 verifies the model and body. The active DOF is the header rotation around the attach-
validates the limitation analysis. A conclusion provides a sum- ment point A with respect to the combine body, and the corre-
mary and offers insight as to possible remedies that could be sponding generalized coordinate is c. The two passive DOFs are
undertaken. the combine body rotation and vertical translation relative to its
center of gravity, and the corresponding generalized coordinates
are h and t, respectively. The output sensor is installed on the
2 System Modeling header tip to measure the header height with respect to the ground.
Therefore, the sensor is noncollocated with the actuator yet its
2.1 Mechanical Subsystem Modeling. Underactuated sys- measurement is influenced by all three DOFs. The mathematical
tems are those that possess fewer numbers of actuators than the model for this underactuated and noncollocated mechanical sys-
number of degrees of freedom (DOFs). Assume an underactuated tem is established as follows.
manipulator has n independent DOFs, m of which are actuated, Figures 3 and 4 illustrate both the rigid body dynamic analysis
and the remaining l ¼ n  m DOFs are termed passive. As illus- and the internally generated forces (FAx ; FAz ; Fl ) for the combine
trated in Ref. [7], the corresponding
  n generalized coordinates body and header, respectively. In this combine system, flow
can be written as qT ¼ qT1 ; qT2 , where q1 2 Rl and q2 2 Rm control valves are used to lift and lower the header. Assuming
correspond to the passive DOFs and active DOFs, respectively. the flow compressibility and the cylinder leakage are relatively
The dynamic equations of the n DOF system can be written as small, the control input to the mechanical system can reasonably
follows [8]: be assumed to be the velocity of the hydraulic cylinder l_c . Equa-
tions (3)–(12) present geometric relationships between the system
m11 q€1 þ m12 q€2 þ h1 þ /1 ¼ 0 (1) variables defined in Figs. 3 and 4. Equations (13)–(15) represent
force balances by which the three primary dynamic equations can
m21 q€1 þ m22 q€2 þ h2 þ /2 ¼ s (2) be represented. Equations (16)–(19) represent relationships among
forces, motion of bodies, and external disturbances caused by ver-
_ 2 Rl and h2 ðq; qÞ
where the vector functions h1 ðq; qÞ _ 2 Rm con- tical displacement of the ground. Nomenclature for the variables
tain Coriolis and centrifugal terms (likely small in the current presented in Eqs. (3)–(19) along with values representative of an
application), the vector functions /1 ðqÞ 2 Rl and /2 ðqÞ 2 Rm actual combine is shown. Exact manufacturer values could not be
contain gravitational terms, and s 2 Rm represents the input gener- made available, but the values in Nomenclature are sufficiently
alized force. accurate to make subsequent analysis valid.

Fig. 3 Force analysis for combine body Fig. 4 Force analysis for header

034503-2 / Vol. 135, MAY 2013 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 8 l2c ¼ l2ins þ l2f  2lins lf cos c (3) two passive DOFs discussed above, and q2 ¼ ðDcÞ corresponds to
>
>
>
> the displacement of the active DOF. The coefficient matrices are
>
>
>
>
> l2ins ¼ l2c þ l2f  2lc lf cos q (4) also given based on the linearized system. The deviation of header
>
> height output can be expressed as a linear combination of the dis-
>
>
>
> a þ c þ fh þ b ¼ p=2 (5) placements of the three DOFs as in Eq. (23)
>
>
>
>
>
> M1 q€ þ H1 q_ þ U1 q ¼ 0 (21)
>
> fh þ b  q ¼ u (6)
>
>
>
>
> M2 q€ þ H2 q_ þ U2 q ¼ s (22)
Kinematics

< xA ¼ lt1 cosðut1  hÞ (7)

 
>
> zA ¼ v  lt1 sinðut1  hÞ (8) q
>
> y ¼ ½ C1 C2  1 (23)
>
> q2
>
>
>
> xcgh ¼ xA þ lcgh cosðb þ fh  fcgh Þ (9)
>
>  
>
> 1 0
>
> where M1 ¼ ½ M11 M12 , M11 ¼ m11 ¼ , M12 ¼ m12
>
> zcgh ¼ zA  lcgh sinðb þ fh  fcgh Þ (10)
>
>
>    0 1 
>
> 0:22314 1:2764 0:23168
>
> a þ h ¼ a0 (11) ¼ , H1 ¼ ½ H11 H12 , H11 ¼ ,
>
>
>
:
0:12265
  1:5577 2:814 
1 h ¼ lh sin b þ zA þ h0 (12) 0 77:762 4:0543
H12 ¼ , U1 ¼ ½ U11 U12 , U11 ¼ ,
8 0  33:721 155:5
>
> Ih b€ ¼ mh glcgh cosðb þ fh  fcgh Þ  Fl lf sinðqÞ (13) 1:0936
>
> U12 ¼ , M2 ¼ ½ M21 M22 , M21 ¼ m21 ¼ 0, M22
>
> 12:539
<
Dynamics

mcom v€ ¼ FAz þ Fl sin u  mcom g þ Ff þ Fr (14) ¼ m22 ¼ 1, H2 ¼ ½ H21 H22 , H21 ¼ 0, H22 ¼ 0, U2 ¼ ½ U21 U22 ,
>
> U21 ¼ 0, U22 ¼ 0, s ¼ ks l€c ¼ 1:7016l€c , C1 ¼ ½ 7:3596 1 ,
>
> Icom h€ ¼ Ff a  Fr b  Fl lt2 sinðu þ ut2  hÞ
>
> C2 ¼ 4:5989.
:
1 þ FAx lt1 sinðut1  hÞ  FAz lt1 cosðut1  hÞ (15)
8 2.2 Hydraulic Subsystem Modeling. As mentioned in
>
> FAx ¼ mh x€cgh þ Fl cos u (16) Sec. 2.1, an electrohydraulic actuator is used to control the angle
>
>
>
> between the header and the combine body. The dynamics in the elec-
>
< FAz ¼ mh y€cgh þ Fl sin u þ mh g (17)
trohydraulic system come primarily from the valve. Assuming the
Forces

>
> fluid is incompressible, the steady state valve flow is proportional to
>
> Ff ¼ kf ðah  zf þ vÞ  bf ðah_  z_f þ vÞ
_ (18) the current command Iin as given in Eq. (24). The flow dynamics are
>
>
>
: thereby dominated by the second order characteristics between the
1 Fr ¼ kr ðbh  zr þ vÞ  br ð  bh_  z_r þ vÞ
_ (19) current command and the actual valve displacement given in Eq. (25)
pffiffiffiffiffiffi
To maintain a desired header height, the absolute header height h Q1;static ¼ K DpðIin Þ (24)
in Fig. 4 should track the time-varying ground profile zh by control-
ling the cylinder velocity l_c . To obtain the open loop transfer func- Q1; static
Q1;dynamic ¼ (25)
tion from zh (the tracking reference) to h (absolute header height), s2 þ 2fv xv s þ x2v
Eqs. (3)–(19) are linearized about an equilibrium point using Eq.
(20). The kinematic relationships in Eqs. (3)–(12) are linearized where K is a flow coefficient, Dp is the pressure difference across
using small angle approximations where appropriate. The equilib- the valve, fv and xv are the damping ratio and natural frequency
rium point considered is a header height of 0.15 m with the vehicle of the valve. For the experimental system, the valve bandwidth
on level ground with corresponding values given in Table 1. was validated as 10 Hz.
 If one assumes the pressure difference upstream and downstream
@f ðxÞ
*
*
 Dx ¼ 0 (20) of the flow control valve is constant and the valve dynamics are suf-
@x *
*
x ss
ficiently high bandwidth, the electrohydraulic system can be further
  simplified to be a cylinder velocity l_c proportional to the current
*
where
* x ¼ a b h t q u c Fl FAx FAz Ff Fr xA zA xcgh zcgh , command Iin with a time delay. As will be seen, the delay incorpo-
* *
f x represents Eqs. (3)–(19), xss are the steady state value of x at rates frictional effects [9] in the cylinder seals and linkage bearings.
* *
the equilibrium point, and Dx are the deviations of x from the When the system is operating at different header positions, the time
equilibrium point. delay varies due to the kinematic dependency of the nonlinear fric-
With the data from Nomenclature and Table 1, we can obtain tion characteristics in the mechanical system. For clarity of exposi-
17 linear equations from Eqs. (3)–(19). Since the system is a three tion, the delay is assumed to be constant. The hydraulic system
DOF system, choose the variables ðDh Dt DcÞ as the inde- dynamics can then be considered as given in following equation,
*
pendent variables, and others in Dx as dependent variables. The where khydr is the corresponding coefficient (see Nomenclature).
resulting dynamics can be represented by Eqs. (21) and (22),
where q1 ¼ ðDh DtÞT corresponds to the displacements of the l_c ðtÞ ¼ khydr Iin ðt  TÞ (26)

3 Fundamental Limitations to Combine

Table 1 Variable value at the equilibrium point
Header Height Control
Symbol Value Symbol Value Symbol Value 3.1 Underactuated and Noncollocated Systems. In the fol-
lowing analysis, we will show how the characteristics of the
ass 0.124 rad qss 0.489 rad css 1.124 rad
underactuation and noncollocation influence the position of the
uss 0.167 rad lc,ss 1.535 m bss 0.0227 rad
hss 0.011 rad vss 0.0596 m xA,ss 2.76 m open loop zeros and poles for the linearized system, and how such
zA,ss 0.948 m FAx,ss 113,277 N FAz,ss 29,881 N zeros and poles would induce limitations on feedback control.
Fl,ss 114,878 N Ff,ss 137,267 N Fr,ss 58,733 N First, we examine the mechanical subsystem to illustrate how zero
xcgh,ss 4.711 m zcgh,ss 1.39 m dynamics, or open loop zeros, result from plant dynamics. To
make the analysis relevant to the case of the header height

Journal of Dynamic Systems, Measurement, and Control MAY 2013, Vol. 135 / 034503-3

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 problem, only a single input single output system is considered.  1  
 q1 ðsÞ ¼ M11 s2 þ H11 s þ U11 M12 s2  H12 s  U12 q2 ðsÞ
Im 0
Defining a coordinate transform matrix T ¼ , the coor- (36)
C1 C2 
dinates can be transformed as q ¼ Tq, where qT ¼ qT1 ; qT2 . Since
Substituting Eqs. (34) and (36) into Eq. (23), the transfer function
jC2 j 6¼ 0, the transform matrix T is nonsingular, and this from input u to output y then can be expressed as
coordinate transformation is valid.The inverse of the matrix T can
I 0 C1 ðM12 s2  H12 s  U12 Þ þ C2 ðM11 s2 þ H11 s þ U11 Þ
be calculated as T ¼ m  , where C1 ¼ ð1=C2 ÞC1 and
1 yðsÞ ¼
C1 C2 M11 s2 þ H11 s þ U11
C2 ¼ 1=C2 . Applying the coordinate transformation above to the ks s
system in (21) and (22) and substitute generalized force s with  uðsÞ (37)
ðM22 s2 þ H22 s þ U22 Þ
s ¼ ks u_ (u is the output of hydraulic cylinder velocity l_c ), the
new system can be expressed as Eqs. (27)–(29). In the new coordi- which clearly illustrates how the open loop zeros for the underac-
nate system, the output y is not correlated to the passive DOFs q1 tuated and noncollocated system are dependent on the passive
anymore. DOFs and the poles are determined by both the active and the pas-
 1q ¼ 0
 1 q€ þ H1 q_ þ U
M (27) sive DOFs. Particularly in this combine system, the matrices of
the active DOF dynamics H22 , U22 are null, so all the nontrivial
 2 q ¼ ks u_
 2 q€ þ H2 q_ þ U
M (28) poles are determined by the passive DOFs, which can be calcu-
lated by the eigenvalues of the matrix P in Eq. (38). Additionally,
y ¼ q2 (29) the derivative on the numerator cancels one integrator and induces
a fifth order system instead of sixth.
 1 ¼ M1 T 1 , H1 ¼ H1 T 1 , U
where M  1 ¼ U1 T 1 , M
 2 ¼ M2 T 1 ,    
q€1 q_ 1
 1  1
H2 ¼ H2 T , and U2 ¼ U2 T . Define a feedback controller as ¼P
  q_ 1 q1
u ¼ M 1
~ 22 ~ 2 q
ks u_  H~2 q_  U (30)  1 1  (38)
M11 H11 M11 U11
P¼
I 0
where M ~ 22 ¼ M  22  M 21 M  12 , H~2 ¼ H2  M
 1 M  21 M
 1 H1 , and
11 11
~2 ¼ U
U 2  M  21 M 1 
 11 U1 .
The previous analysis can be demonstrated numerically with
Substituting Eq. (30) into Eqs. (27) and (28) results in the sys- the linearized combine system. Using the linearized system values
tem given in Eqs. (31) and (32). By inspection, the zero dynamics from Table 1, and the analysis from (34)–(37), gives
of the system are represented by Eq. (31) since the dynamics of q1
do not affect the output y. Note that M  11 deviate from
 11 , H11 , U y 4:822ðs þ 1:49 6 12:2iÞðs þ 0:88 6 11:3iÞ
the original coefficient matrices of the passive DOFs (M11 , H11 , Pcom ðsÞ ¼ L ¼
l_c sðs þ 1:48 6 12:4iÞðs þ 0:57 6 8:7iÞ
U11 ) with terms (C1 M12 , C1 H12 , C1 U12 ).
(39)
M  11 q1 ¼ M
 11 q€1 þ H11 q_ 1 þ U  12 q2
 12 €q2  H12 q_ 2 þ U (31)
The four nontrivial poles have undamped natural frequencies of
€q2 ¼ u (32) 2.1 Hz and 1.4 Hz with corresponding damping ratios of 0.104
and 0.069, respectively. The undamped natural frequencies of the
where M  11 ¼ M11 þ C1 M12 , H11 ¼ H11 þ C1 H12 , U  11 ¼ U11 zeros are 2.07 Hz and 1.87 Hz with the damping ratios as 0.114
þ C1 U12 , M
 12 ¼ M12 C2 , H12 ¼ H12 C2 , and U 12 ¼ U12 C2 . and 0.0894, respectively. Figure 5 shows the open loop poles and
The zero dynamics then can be further represented as Eq. (33). zeros for the system demonstrating an interlacing of poles and
For the underactuated and noncollocated system, expressed as zeros close to the jx axis. This type of phenomenon is also present
Eqs. (21)–(23), the zeros of the system are the eigenvalues of the and well recognized in other underactuated control systems, such
matrix N. as flexible structures [10].
   
q€1 q_ 1
¼N
q_ 1 q1
  1   1 U  (33)
M11 H11 M 11 11
N¼
I 0

The open loop poles also have close and explicit relationship
with the dynamics of the DOFs. For an underactuated system, the
dynamics of the active DOFs q2 are directly determined by the
input u without coupling to the passive DOFs q1. These coefficient
matrices M12 , H12 , U12 are always null. Applying a Laplace trans-
formation on Eq. (22) gives the transfer function from u to q2 as

 1
q2 ðsÞ ¼ M22 s2 þ H22 s þ U22 ks suðsÞ (34)

Additionally, the response of the passive DOFs q1 is determined

by the behavior of active DOF q2. Then, Eq. (21) can be rewritten
as
M11 q€1 þ H11 q_ 1 þ U11 q1 ¼ M12 q€2  H12 q_ 2 þ U12 q2 (35)

The transfer function from q2 to q1 can be obtained by again Fig. 5 Open loop poles and zeros of linearized combine me-
using Laplace transforms chanical system

034503-4 / Vol. 135, MAY 2013 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 6 Closed loop sensitivity plot of linearized combine me-
chanical system
Fig. 7 Closed loop sensitivity plot of linearized combine sys-
It is well known that input/output pairs with low frequency and tem with actuator delay
lightly damped zeros and poles can limit the effectiveness of any
feedback control approach [11,12]. Due to the proximity between the desired closed loop system bandwidth of 3 Hz. The delay can
open loop poles and zeros as shown in Fig. 5, there is difficulty in vary somewhat with the hardware configuration but will exist in
moving closed loop poles very far from their open loop location. some form for all header height actuations systems due to cost
Clearly, this will be a fundamental limitation regardless of any and manufacturing constraints of these agricultural systems. It
feedback controller C(s). cannot be eliminated by means of feedback. According to
In the following, a frequency domain analysis of the system Ref. [12], the bandwidth limitation caused by a time delay can be
limitations is given. Figure 6 gives the closed loop sensitivity expressed by Eq. (41). Therefore, in the present combine system,
function associated with the system of Eq. (39) using a propor- the time delay from the actuator will limit the closed loop band-
tional feedback gain tuned by experience to be 0.3. It clearly width to below 0.53 Hz.
shows the effects on the magnitude and phase plots caused by the
zeros and poles. From the Fig. 6, the closed loop bandwidth is lim- 1
xc < (41)
ited to below approximately 1.43 Hz. To obtain a higher closed T
loop bandwidth, a more sophisticated controller C(s) could be
introduced to decrease the sensitivity function over a broader where xc is the achievable bandwidth, and T is the delay. This
range of low frequencies. However, according to the Bode sensi- limitation deteriorates the already low bandwidth induced by
tivity integral (Eq. (40)), any reduction in the sensitivity function the mechanical structure characteristics discussed in Sec. 3.1.
at lower frequencies would result in an increase in higher fre- Figure 7 shows a sensitivity function of the closed loop system
quency [13]. from Fig. 6 but including a delay of 0.3 s. Clearly, the situation
ð1 has deteriorated by the extra phase decrease, thereby making the
p available bandwidth further below the desired target.
lnjsðjxÞjdx ¼  lim sCðsÞPðsÞ (40)
0 2 s!1
4 Model Validation
The effort of this C(s) to improve the sensitivity function in fre-
quency under 1.43 Hz will cause a “piling up” of the sensitivity The modeling and analysis of Secs. 2 and 3 posit the existence
function at and above 1.43 Hz. This will make the system lose of fundamental limitations to feedback control for the header
robustness at these higher frequencies possibly leading to instabil- height control problem in combines. This section describes the
ity. Fundamentally, the performance is limited by the position of experimental procedures and results of the models created in
open loop zeros and poles, which is due to the noncollocated and Sec. 2 to validate the analysis on a real-world system. First, the
underactuated nature of this system as shown above. Figure 6 hydraulic system model given in Sec. 2.1 is validated in the time
illustrates the challenge faced by any feedback controller in domain. Subsequently, the mechanical system model is validated
achieving a closed loop bandwidth on the order of the desired in the frequency and time domains. The results of this section
3 Hz value. indicate the validity of the plant models used in the analysis of
This system-level behavior is not unique to the header height Sec. 3.
control system on a combine. Any underactuated system with a
noncollocated sensing and actuation and lightly damped, low nat- 4.1 Hydraulic Subsystem Model Validation. To validate
ural frequency passive DOFs will introduce similar pole and zero the simplified electrohydraulic system from Sec. 2.2, the hydraulic
pairs (or even worse, unstable zeros or poles) in the open loop actuation had to be separated from the combine. Figure 8 illus-
transfer function thereby fundamentally limiting the bandwidth trates a test stand designed to perform this task. The foundation
achievable by any controller [14,15]. Below, in Sec. 3.2, we illus- pile acts as reaction wall for the actuator to push on with no pitch
trate that the situation is even more challenging when the actua- or heave dynamics as would be found on the actual combine. A
tion subsystem contains delays. header is also attached to provide a realistic inertial load for the
actuator to move. The valve and pump systems are replicated
from a production combine system and a height sensor is installed
3.2 Time Delay Systems. It is well known that time delays on the header tip to measure the header height. Additionally, sev-
in feedback systems reduce available bandwidth in order to main- eral pressure sensors are installed throughout the hydraulic system
tain closed loop stability [12,16]. This is true irrespective of the for diagnostic purposes.
feedback approach taken. Due to the subsystem design, the actua- To measure the time domain response of the hydraulic system,
tor delay present in the combine header height system is up to a step command is applied to the control valve and the available
0.3 s as will be illustrated in Sec. 4.1. This time is large relative to signals are monitored as shown in Fig. 9. The sampling rate for

Journal of Dynamic Systems, Measurement, and Control MAY 2013, Vol. 135 / 034503-5

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 8 Hydraulic system test bed

Fig. 10 Closed loop step response of the test system

Fig. 9 Open loop step response of hydraulic system for input

current

the sensors is 20 Hz. There is a combination of responses which

include the time to open valves (t1), to stroke the load sensing
pump (t2), to build up pressure of the pump output (t3), and to
overcome system friction (t4). Summing these effects results in a
total time delay T ¼ 0.3 s. This delay constant can vary with dif-
ferent combines and different operation positions. However, for
this particular investigation, it is treated as constant for clarity of
exposition.
To verify that a pure time delay (Eq. (26)) represents the major
characteristic of the hydraulic system, the closed loop step
responses from simulation and experiment are compared in
Fig. 10. A proportional controller is used to make the header fol-
low a step reference. The results of Fig. 10 indicate that the simu- Fig. 11 Experimental combine used for field test results
lations results match the response of the actual system sufficiently
well so as to have confidence in the simplified model in Eq. (26).
For this validation, it was not possible to introduce a perfect actu-
ator and thereby separate the electrohydraulic subsystem from the
4.2 Mechanical Subsystem Model Validation. The mechanical subsystem. However, given the validity of the electro-
mechanical subsystem experiments were performed on an experi- hydraulic subsystem demonstrated in Sec. 4.1, the effect of actua-
mental John Deere combine þ header system shown in Fig. 11. tion model error affecting the validation of the mechanical

034503-6 / Vol. 135, MAY 2013 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 suspensions can be added between the tires and the combine body
to increase the number of actuators and therefore eliminate the
underactuation in the system. Other possibilities include redesign-
ing key parameters in the system, such as suspension elements, to
improve the low natural frequency and lightly damped character-
istics of the passive DOFs. As can be seen, there are multiple
ways to address the mechanical system problem. However, all of
these must be considered in light of realistic cost and design con-
straints. As for the actuator delay problem arising from the elec-
trohydraulic system, it may be possible to reduce or eliminate the
delay with very high performance servo hydraulics. As with the
mechanical redesign, these types of system changes would have to
be performed under realistic cost and design constraints. High
performance servo-actuators may not be appropriate for an all-
weather all-terrain agricultural vehicle and may not meet market
price points. The work presented here illustrates a practically rele-
vant problem; the search for an optimal solution remains an open
control engineering question.
Fig. 12 Closed loop frequency response comparisons
between experimental results and simulation
Acknowledgment
The authors appreciate the support of Deere & Company for
subsystem model is minimized. The mechanical subsystem model this paper. Dustin Denault’s assistance for the experimental test-
from Sec. 2 was validated by both frequency domain and time do- ing performed on the combine and the test stand was essential and
main responses. Since the open loop system in (24) is type 1, a greatly appreciated.
sine sweep frequency response must be performed in a closed
loop manner. For the frequency domain responses, a simple con-
troller was utilized to generate a closed loop transfer function
both in simulation and experimentally. Nomenclature
A series of sinusoidal height references, constant amplitude a; b ¼ the distance in x direction between front/rear wheel
with varying frequency, were fed to both simulated and experi- axis and gravity center of combine body
mental closed loop systems for a stationary combine on level (2 m;1.3 m)1
ground. The output heights from the header height sensor were bf; br ¼ the damping constant of front and rear tires
collected and compared with the reference signals. Performing a (22,400 kg/s; 26,300 kg/s)
“frequency by frequency” analysis of magnitude and phase differ- h0 ¼ the original height of the A point (1.2 m)
ences allowed for the construction of a frequency response plot a0 ¼ the original value of angle a (0.113 m)
in Fig. 12. The result from experiment is compared with the Icom ; Ih ¼ the inertias of combine body and header with respect
simulation for the nonlinear system given by Eqs. (3)–(19). The to the gravity center and point A separately
nonlinear model fits the physical system well in the magnitude (66,000 kg m2; 22,000 kg m2)
plot. The phase plots have some differences, primarily due to the lt1 ; lt2 ; lcgh ¼ structural length (refer to Fig. 3) (2.9 m, 3 m, 2 m,
assumption of a constant friction level, and hence a constant 0.8 m)
delay, in the hydraulic subsystem. In the experimental system, the lins ; lh ; lf ¼ structural length (refer to Fig. 4) (4.6 m, 1.7 m)
friction varies with a change in relative orientation between the mcom ; mh ¼ the masses of the combine body and the header
combine and the header which is the primary cause of the phase (15,000 kg; 5000 kg)
differences. While the nonlinearity of the system would lead to kf ; kr ¼ the spring constant of front and rear tires
amplitude dependent responses, the nature of Fig. 12 clearly illus- (1,303,720 N/m; 1,673,600 N/m)
trates the matching of the major dynamic modes. By comparing khydr ¼ coefficient from valve current to the velocity of the
the linearized system with the nonlinear system, the linearization cylinder (0.032 m/s/A)
preserves the main model information and can describe the system fh ; fcgh ¼ structural angle (refer to Fig. 3) (0.3 rad, 0.1 rad)
behavior accurately enough. For further validation in the time do- ut1 ; ut2 ¼ structural angle (refer to Fig. 4) (0.3 rad, 0.5 rad)
main, see Ref. [6]. FAx; FAz ¼ the forces at the point A in x and z directions
(variable)
5 Conclusion and Discussion Ff; Fr ¼ the forces on the combine body at the front and rear
tires (variable)
This paper used system dynamics and common analytical tools
Fl ¼ the force from the hydraulic cylinder on the header
to gain insights into the fundamental limits for a combine har-
and combine body (variable)
vester header height control system. The modeling and analysis of
h ¼ the absolute header height (variable)
Secs. 2 and 3 clearly demonstrate the challenges present in header
lc ¼ the cylinder length (variable)
height control and the reasons for the relative bottleneck in
xA ; zA ¼ the distances between the combine body gravity
increasing vehicle speeds. The underactuation and noncollocation
center and the point A in x and z directions
properties of the mechanical system determine the position of the
(variable)
open loop poles and zeros, which results in a system bandwidth
xcgh; zcgh ¼ the distances between the header gravity center and
upper limit. A time-delay limitation from the hydraulic actuator
the point A in x and z directions (variable)
further limits the achievable closed loop performance.
a,q,u ¼ refer to Figs. 3 and 4 for the geometric meaning of
Clearly, a drastic improvement in the overall system perform-
these terms (variable)
ance cannot be achieved solely by feedback control design. To
b ¼ the header angle with respect to x axis (variable)
eliminate or decrease the undesirable mechanical characteristics,
h ¼ the pitch displacement of combine body (variable)
the mechanical system needs to be redesigned. For example, by
replacing the tires with tracks, the DOFs for the system reduce to
1
one and the system becomes fully actuated. Alternatively, active Values are given inside parentheses.

Journal of Dynamic Systems, Measurement, and Control MAY 2013, Vol. 135 / 034503-7

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 v ¼ the vertical displacement of combine body [7] Arai, H., and Tachi, S., 1991, “Position Control of a Manipulator With Passive
Joints Using Dynamic Coupling,” IEEE Trans. Rob. Autom., 7(4), pp.
(variable) 528–534.
c ¼ the angle between header and combine body [8] Spong, M., Seth, H., and Vidyasagar, M., 2006, Robot Modeling and Control,
(variable) John Wiley & Sons, New York.
[9] Olsson, H., Aström, K. J., Canudas de Wit, C., Gäfvert, M., and Lischinsky, P.,
1998, “Friction Models and Friction Compensation,” Eur. J. Control, 4, pp.
176–795.
[10] Lindner, D. K., Reichard, K. M., and Tarkenton, L. M., 1993, “Zeros of Modal
References Models of Flexible Structures,” IEEE Trans. Autom. Control, 38(9), pp.
[1] Hobson, R. N., and Bruce, D. M., 2002, “PM—Power and Machinery: Seed 1384–1388.
Loss When Cutting a Standing Crop of Oilseed Rape With Two Types of Com- [11] Aström, K. J., 2000, “Limitations on Control System Performance,” Eur. J.
bine Harvester Header,” Biosyst. Eng., 81(3), pp. 281–286. Control, 6(1), pp. 2–20.
[2] Glancey, J. L., 1997, “Analysis of Header Loss From Pod Stripper Combines in [12] Skogestad, S., and Postlethwaite, I., 2005, Multivariable Feedback Control
Green Peas,” J. Agric. Eng. Res., 68(1), pp. 1–10. Analysis and Design, John Wiley & Sons Ltd, West Sussex.
[3] Kaminski, T. L., and Zoerb, G. C., 1965, “Automatic Header-Height Control [13] Stein, G., 2003, “Respect the Unstable,” IEEE Control Syst. Mag., 23(4), pp.
for Grain Crops,” Trans. Am. Soc. Agric. Eng., 8(2), pp. 284–287. 12–25.
[4] Lopes, G. T., Magalhães, P. S. G., and Nóbrega, E. G. O., 2002, “AE—Automa- [14] Balas, G. J., and Doyle, J. C., 1994, “Robustness and Performance Trade-Offs
tion and Engineering Technologies: Optimal Header Height Control System for in Control Design for Flexible Structures,” IEEE Trans. Control Syst. Technol.,
Combine Harvesters,” Biosyst. Eng., 81(3), pp. 261–272. 2(4), pp. 352–361.
[5] Xie, Y., Alleyne, A., Greer, A., and Deneault, D., 2010, “Header Height Control [15] Thibeault, N. M., and Smith, R., 2001, “Fundamental Limits in Robustness and
of a Combine Harvester System,” Proceeding of ASME Dynamic Systems and Performance for Unstable, Underactuated Systems,” IEEE Trans. Autom. Con-
Control Conference (DSCC), Cambridge, MA. trol, 46(8), pp. 1265–1278.
[6] Xie, Y., Alleyne, A. G., Greer, A., and Deneault, D., 2011, “Fundamental Lim- [16] Magyar, B., Ho†s, C., and Stépán, G., 2010, “Influence of Control Valve Delay
its in Combine Harvester Header Height Control,” Proceedings of the 2011 and Dead Zone on the Stability of a Simple Hydraulic Positioning System,”
American Control Conference, San Francisco, CA. Math. Probl. Eng., 2010, p. 349489.

034503-8 / Vol. 135, MAY 2013 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use