Ieee 00954519

48 IEEE Control Systems Magazine October 2001
S
ustainable agriculture
aims at the production of
high-quality food and raw
materials in sufficient
quantity for a wide range
of consumers. Further
objectives are the rational use of natu-
ral resources and preservation of the
environment. For this reason, modern
field machinery and equipment should
be able to cope with complex agricul-
tural processes and to execute difficult
operations at high efficiencies and
without environmental pollution.
To control the performance of these
machines, much information has to be
captured by sensors and transmitted
to and stored in data logging systems
for further processing. Moreover, agri-
cultural production takes place in an open system that has
various relations to its surroundings. Therefore, when these
machines and processes are in operation, the state of the
surrounding systems, as well as the interactions between
the agricultural production process and its environment,
must be taken into account. Mass and energy flows must be
accompanied by information flows. These facts require the
introduction of an information-based agriculture, the
so-called “precision agriculture.”
Precision agriculture means that the production pro-
cesses must be strictly controlled according to the de-
mands of plants, soil, and environment in a site-specific way.
The area of these sites is much smaller than the area of
whole fields. For intensive cultures, such as vegetables, a
SPECIAL SECTIONSPECIAL SECTION
By Josse De Baerdemaeker, Axel Munack,
Herman Ramon, and Hermann Speckmann
De Baerdemaeker (josse.debaerdemaeker@agr.kuleuven.ac.be) and Ramon are with the Katholieke Universiteit Leuven, Laboratory for
Agro-Machinery and -Processing, Kasteelpark Arenberg 30, 3001 Leuven, Belgium. Munack and Speckmann are with the Institute for Tech-
nology and Biosystems Engineering of the Federal Agricultural Research Centre, 38116 Braunschweig, Germany.
©199121stCENTURYMEDIAAND©2000iSWOOPLTD.
0272-1708/01/$10.00©2001IEEE

site may even be only one plant. For animal production, this
means that each animal is treated individually. Such a
site-specific treatment requires the transmission of great
amounts of data, such as individual values for references,
states, and controlled variables, together with information
about weather conditions, date, time, and location. Addi-
tionally, technical equipment and production processes
should be upgraded with new knowledge, improvements,
and enhancements in a simple and compatible way. The
maintenance and service of modern machines and process
equipment should be handled according to their actual
wear, operation times, and circumstances. This necessi-
tates sampling, transmission, and processing of data in a
compatible way, since the data may be generated, transmit-
ted, and processed in different units. In summary, compati-
ble data transmission is a necessary condition for achieving
all the aims formulated above. Communication technology
thus serves as the backbone of precision agriculture. In the
following, we give three examples for advanced precision
agriculture components: a combine harvester, sprayer, and
fertilizer spreader. This will be followed by a description of
the “backbone” communication, which is organized in the
form of a specific agricultural bus system and protocol.
Spatial variability in soil conditions such as texture,
structure, soil moisture, and soil fertility give rise to local
variations in crop yield. Although the lack of spatial unifor-
mity of the factors that influence the growth of field crops,
and hence their productivity, has been known and appreci-
ated since early times, agricultural practice hardly takes
into account this spatial variability in traditional arable
crop production. The recent availability of reliable, inexpen-
sive, and precise systems for on-the-go acquisition of the
world position of soil tillage tools, machines for crop protec-
tion, fertilizers and harvesters during field operation (the
global positioning system, or GPS, supported by dead reck-
oning systems), and parallel advances in sensor technology,
precision mechanisms, and the information processing
power of computers, have led to adoption of the concepts of
precision agriculture, site-specific farming, or spatially vari-
able application. In site-specific agriculture, different field
operations are adapted to variations in soil conditions, crop
growth stage and yield, and the spread of weeds and disease
infestation within each individual field. Intrafield variations
are captured and the registered data are translated into nu-
merous field maps (e.g., weed, disease, yield, and fertilizer
and pesticide application maps) with high resolution. These
October 2001 IEEE Control Systems Magazine 49
Mapping External Information
Decision Support
Data Collection: Growing Season
Variable-Rate Fertilizing
Variable-Rate Spraying
Variable Treatment
Data Collection: Harvest
* Prices
* Weather
* Data Processing
* Data Analysis
* Advice
* Soil Analysis
* Weed Patches
* Diseases
* Crop Growth Stage
* Chlorophyll Content
* Yield
* Quality
* Moisture Content
Figure 1. Arable precision farming cycle [1].

maps are the core of site-specific crop management that
guarantees a more rational use of raw inputs such as seed
for sowing, fertilizer, pesticides, and fuel for mobile agricul-
tural machines.
In the near future, a modern farm could be managed as
shown schematically in Fig. 1. Based on historical data about
each field, such as crop rotation, crop yield, soil status, infes-
tation spread, and climatic conditions, decision models de-
termine the essential site-specific soil tillage, pretreatment of
the seedbed, and sowing density. During the growth season,
the modern farmer decides about site-specific application of
fertilizer, supported by crop growth models and field mea-
surements, the most important of which are soil coverage of
crops in the early growth stage and evolution of the chloro-
phyll content of green leaves. A spraying machine equipped
with optical sensors for the detection of diseases and weeds
is used for the treatment of local infestations. During harvest-
ing, sensors register the online bulk mass flow of harvested
raw produce in the harvesting machine, along with product
properties that are of important commercial value, such as
protein and moisture content of cereals and sugar content of
sugar beets. These data, related to the captured absolute po-
sition of the machine, are mapped in historical records to
support site-specific crop management in subsequent grow-
ing seasons.
Examples of Advanced Precision
Agriculture Components
Objectives
Site-specific agriculture requires the application of machin-
ery equipped with high-precision devices. Unfortunately,
most performance specifications that must be met by ma-
chines or machine parts for use in precision agriculture can
no longer be met through a traditional sequential design of
the mechanism, the controllers, and the information sys-
tems. Increasingly, the improvement or adaptation of agri-
cultural machines requires the application of a mechatronic
design methodology to meet the stringent performance re-
quirements that are essential for site-specific field opera-
tions. In a mechatronic design process, performance of the
mechanism can be improved considerably or even opti-
mized through the concurrent and integrated development
of precision mechanisms, modern controllers, and
advanced information systems.
In this respect, three recently developed mechatronic
systems for spatially variable application in arable crop
management will be discussed. The first mechatronic de-
sign is a high-precision mass flow sensor built into harvest-
ing machines for online measurement of crop yield during
harvesting. Next, the adaptation of a spraying machine for
selective spraying of those areas with significant infestation
is discussed. Finally, a flow rate control system, imple-
mented on a slurry tank spreader for variable-rate applica-
tion of liquid manure, is discussed.
Mass Flow Sensor for Combines
Sensor Requirements
During the past 15 years, research on yield sensors has fo-
cused mainly on the development of reliable grain flow sen-
sors on combine harvesters for measuring the grain yield
during harvesting. Although many sensors have been pro-
posed, only a few proved to be suitable for commercial ap-
plication [2] due to the severe performance criteria
imposed on the sensors, the most important of which are:
• The sensor should be able to measure the grain flow
with sufficient accuracy such that measurement er-
rors are less than 5%.
• Machine motion and vibration should not disturb the
accuracy of the sensor.
• Analysis of the measurement signal before it becomes
suitable for deriving yield maps should be simple and
straightforward.
• The accuracy of the sensor must remain independent
of variations in bulk properties.
• Requirements for recalibration and maintenance of
the sensor should be minimal.
• The sensor should have an appropriate design for
easy integration in combines.
The yield sensor developed by Strubbe [3] shows very
promising results, as it amply meets these performance re-
quirements.
Grain Yield Sensor
The grain flow sensor proposed by Strubbe [3] is mounted
at the outlet of the grain elevator, as shown in Fig. 2. The sen-
sor consists of a 90° curved plate or chute, supported at the
elevator housing by two pendulum rods that can rotate
around a pivot point as shown in Fig. 3. A beam spring keeps
the sensor in its initial position when the machine is at rest.
A counterweight is fixed to the opposite tips of both rods
such that the pivot point coincides with the center of gravity
of the whole assembly so as to render the sensor insensitive
to translational vibrations of the combine. In addition, this
suspension drastically reduces the influence of driving up-
hill or downhill on the zero reading of the sensor. Normally,
the threshed grain kernels are thrown by the pin parcels
into the storage tank. To lead the grain flow smoothly into
GPS Data Logger
Grain Flow Sensor
Radar Speed
RPM Elevator
Cutting Width
Figure 2. Curved-plate grain mass flow sensor with additional
instrumentation installed in a combine [1].

the sensor, a deflection plate and a rotor are in-
stalled at the head of the elevator.
The grain mass flow entering the sensor ex-
erts a force on the curved plate, causing the as-
sembly to start rotating around its pivot point
against the spring force. This final force is the re-
sult of the gravity force Fg, the centrifugal force
Fcf , and the friction force Ff between the grain
mass and the curved plate body [3] and thus is a
function of the total grain mass m on the plate.
Consequently, the registered instantaneous de-
flection of the beam spring by a linear inductive
distance sensor is a measure of the mass flow
variations in the curved plate. According to Fig.
4, forces acting on an elementary particle dm
moving in the chute are
dF gdmg = (1)
dF
v
R
dmcf =
2
(2)
dF Nf = µ (3)
with
N
v
R
g= − +






2
sin( )δ θ and dm
Q
v
Rd= θ
(4)
in which g is the gravitational force (m/s2
), v is the speed of
the particle in the chute (m/s),µ is the friction coefficient,Q
is the mass flow rate of kernels in the chute (kg/s), R is the
radius of curvature of the chute (m), δ is the inclination an-
gle of the curved plate (rad), andθ is the angle (rad) indicat-
ing the position of the elementary particle in the plate, as
indicated in Fig. 4.
Unfortunately, the friction coefficient in the friction force
is a function of kernel characteristics such as crop type and
moisture content. As a consequence, the sensor must be
recalibrated for different crops and varying harvesting con-
ditions, a very time-consuming and delicate operation. By
using theoretical models, Strubbe [3] proved that the influ-
ence of friction on the moment M acting on a chute depends
on the location of the pivot point A of the chute relative to
the center of curvature O( )0 0, of the plate. The position of A
with respect to O( )0 0, is determined by the polar coordi-
nates ( )α,r , as shown in Fig. 4. The parameter r is the dis-
tance between the center of curvature and A, and α is the
angle between r and the entrance of the chute (rad).
Strubbe concluded that for any chute construction, at
least one pivot point could be found where the moment of
the resulting forces acting on the chute becomes almost fric-
tion independent. Those pivot points are located on a
straight line whose position and orientation depend on the
entrance speedv0 of the kernels and the configuration of the
chute assembly.
The forces acting on the elementary particle in the chute
createamomentaroundpivotpoint A,whichcanbewrittenas
[ ]
dM
U
U
r r R Q gRd
=
− +



× − + − −
sin( )
sin( ) cos( )
δ θ
θ α µ θ α µ θ (5)
with the two dimensionless numbers
U
v
gR
= And U
v
gR
0
0
2
= .
(6)
Integration of (5) over the arc length of curvature of the plate
results in the moment M exerted by the total grain mass on
Force
Clamped Spring
Rotation Point
Counterweights
Support Arms
Curved Plate
Distance SensorRotor
Deflection Plate
Figure 3. Detail of grain elevator top with grain mass flow sensor [3].
O(0,0)
M
A r( , )α
r
N
µN
dm
gdm
α
ν0
R
θ
θe
δ
Figure 4. Forcesandmomentsactingonthemeasurementplate[1].

the chute in pivot point A. The sensitivity of (5) for variations
in µ relative to a standard friction coefficient, which equals
0.4 for most biological materials, is expressed by
( )
∆M
M M
M
=
−
−






( ) ( . )
( . )
.
.
µ
µ
04
04
100
10 04
(%).
(7)
Equation (7) can be embedded into a numerical optimi-
zation problem as an objective function with the follow-
ing design parameters: entrance speed of the kernels,
inclination angle of the assembly, arc length and radius
of curvature of the chute, distance between the center of
curvature and the pivot point, length of the pendulum
rods, and orientation of the rods with respect to the
plate. The combine construction and, in particular, the
shape of the grain elevator impose explicit constraints
on the design parameters. To avoid the mass flow losing
contact with the curved plate, the centrifugal force
should always be larger than the normal component of
the gravitational force pointing in the direction of the
center of curvature. This statement can be considered
an implicit constraint. By minimizing ∆M for several nu-
merical values of the friction coefficient µ, the straight
line on which the influence ofµ on the measured moment
M is minimal, the optimal entrance speed, and the config-
uration of the chute can be calculated. Different proto-
type chutes with optimized design parameters can be
found in Strubbe [3].
The optimized grain flow sensor has been tested over
several years under widely varying harvesting conditions
ranging from winter barley with a moisture content of 12% to
corn with a moisture content of 40%. The regression lines in
Fig. 5 show that the sensor is independent of crop type and
condition. Only the harvesting season influences the slope
of the regression line, indicating that the sensor should be
calibrated once a year at the start of the harvesting season.
The accuracy of the grain yield sensor was evaluated on har-
vested areas of various sizes ranging from 120 to 2000 m².
The registered yield error is due to inaccuracies in the mea-
surement procedure and sensor inaccuracies. The error in
percentage yield increases with decreasing harvested area.
For a harvested area of 400 m², matching the grid size of 20 m
× 20 m for soil sampling, the maximum error
was 5%. For an area of 2000 m2
, the maximum
error decreased to 3%. The error in estimating
the yield of a 6-ha field in The Netherlands was
less than 1.8%.
Grain Yield Maps
To transform the mass flow rate data from the
yield sensor into a yield map, additional infor-
mationiscollectedbythefollowingsensors[1]:
• A capacitive moisture sensor is mounted
in the grain elevator to convert the mass
flow rate measured at a certain moisture content into
a mass flow rate with a standard moisture content
(e.g., 14%).
• As a larger cutting width directly influences the mass
flow rate in the curved plate, an ultrasonic distance
sensor is installed on the header of the combine to
measure the cutting width of the knife, which influ-
ences the mass.
• A precise Doppler radar sensor to measure the travel
speed of the combine, in combination with the ultra-
sonic sensor outputs, is necessary to relate the actual
harvested surface to the measured grain mass flow in
the chute.
• To relate the grain yield to the correct location in the
field, the absolute position of the
combine is determined by a differ-
ential GPS (DGPS).
• The transportation time the grain
kernels need to reach the yield
sensor after the crop is cut by the
cutter bar and the smearing effect
of the return loop where unthresh-
ed ears are brought back into the
threshing process should be com-
pensated in the yield measure-
ments. To this end, Maertens et al.
[4], [5] developed an analytical
model of the grain flow process in
the New Holland TF78 combine.
FlowRate[t/h]
20
18
16
14
12
10
8
6
4
2
0
0.0 0.5 1.0 1.5 2.0
y x
R
= 10.263
= 0.9752
Wheat 14% (Boigneville)
Dry Peas 14% (Boigneville)
Corn 30% (Buken)
Corn 35% (Herent)
Regression Corn 30%
Grain Flow Sensor Signal [V]
Figure 5. Grain flow sensor output signal for different crops and harvest conditions as a
functionoftheflowrate.Theaveragemoisturecontentpercropisgivenin%wetbulbdensity[1].
Precision agriculture means that the
production processes must be
strictly controlled according to the
demands of plants, soil, and
environment in a site-specific way.

This model starts by representing
the biomass flow above the cutter
bar. Subsequently, it describes the
transport time of the biomass
through the feeding auger and the
transport time of the unthreshed
kernels in the threshing-sieving
mechanism. Once the grain has
fallen through the concaves of the
threshing drums, the kernel distri-
bution and transport time on the
grain pan and the sieves is mod-
eled. A similar model is provided
for the return loop. In a final step,
the residence time of the kernels in
the grain elevator before reaching
the yield sensor is modeled.
Site-Specific Spraying
Chemical Crop Protection
Agricultural production suffers from se-
vere losses due to insects, plant dis-
eases, and weeds. Owing to an exponen-
tially growing world population, crop
protection has become one of the most
important field operations for increas-
ing productivity and crop yield.
The most widely used practice in
weed control is spraying herbicides uniformly over the agri-
cultural fields at various times during the cultivation cycle
of arable crops. To guarantee their effectiveness, over-
application of pesticides is commonly advised; however, ex-
cessive use of pesticides raises the danger of toxic residue
levels on agricultural products. Because pesticides, and es-
pecially herbicides, are a major cost factor in the produc-
tion of field crops and have been identified as a major
contributor to ground water and surface water contamina-
tion, their use must be reduced dramatically.
Fortunately, most weed populations develop in patches
in the field, with large areas of the field remaining free of
weeds or having a very low weed density in the early stage
of infestation (Fig. 6). As a consequence, herbicides would
be used more efficiently if they were applied in the appropri-
ate dose, where they are needed, and not to areas with insig-
nificant weed densities. Thus, weeds have been suggested
as the primary target for spatially selective pest control.
To set up a local weed treatment, the weed populations
must be evaluated in the field. In this respect, two concepts
of site-specific weed control have been suggested [6]:
• Weed monitoring is carried out in separate operations
prior to the spraying operation (the mapping concept).
Weed distribution is represented in digitized weed
maps, which are later used during spraying operations
to activate the spraying system using the on-board com-
puter of the field sprayers. The instantaneous position
of the field vehicle is determined by a GPS receiver
mounted on the machine.
• Weed monitoring and spraying are carried out sequen-
tially in the same operation (the real-time concept). A
real-time weed detection system mounted on the field-
spraying machine detects “individual” weeds and
transmits that information to a control system that
controls the spraying equipment of the vehicle. This is
called weed-activated spraying (Fig. 7).
Necessary Equipment for
Real-Time Targeted Application
The modification or extension of current field-spraying ma-
chines involves three aspects:
1) The field sprayer should be equipped with a detection
system that can discriminate between weeds and
crop or soil. This implies the development of optical
sensors with appropriate classification software for
online discrimination between weeds and field crops.
2) Horizontal and vertical stabilization of a spray boom is
essential to ensure correct positioning of the spray
nozzles and detection system. To this end, a spray
boom suspension must be designed to absorb tractor
vibrations so that the spray boom behavior is stable.
Plants/0.25 m2
Distance[m]
60
40
20
0
0 20 40 60 80 100 120 140 160 180 200 220
Distance [m]
660
600
540
480
420
360
300
240
180
120
60
Figure 6. Measured weed density in an agricultural field [6].
Processing of Reflection Measurements
Setting Appropriate Spraying Action
Controller
Spraying ActionReflection Measurement
Detector
Weed
Crop
Valve
Spray Nozzle
Figure 7. Real-time concept in field spraying [6].

3) The response speed and accuracy of the spray equip-
ment must be improved to guarantee a minimal time
delay and difference in flow rate between the continu-
ously tuned dose and the actually sprayed dose. This
implies the application of appropriate pumps, pres-
sure control valves and flow rate controllers,
fast-locking valves to very quickly close sections in
the spray hoses, and special spray nozzles.
Optical Detection System
Solar radiation incident on green vegetation is partially re-
flected from, transmitted through, or absorbed by the vege-
tation. Light is selectively absorbed in the blue (about 400
nm) and red (about 650 nm) wavebands by the chlorophyll
of the plants. It is reflected in the green (about 600 nm) and
strongly reflected in the near-infrared (NIR) (between 750
and 1300 nm) wavebands by the complex internal structure
of the plant (Fig. 8).
Research performed by Vrindts [6] using a desktop
spectrophotometer showed that the difference in the spectra
of crop and weeds at various wavelengths can be used to
classify and subsequently discriminate between crop and
weeds. Under field conditions, the dis-
crimination becomes quite complicated
due to varying illumination conditions
and background reflection properties,
calling for advanced classification meth-
ods such as neural networks. In the work
performed by Moshou et al. [7], two eco-
nomically important crops, corn and
sugar beets, and various weed species
are discriminated from their reflection
ratio in the visual and NIR bands of the
spectrum. A variety of neural-net-
work-based methods have been used for
comparison with the proposed classifi-
cation method, local linear mappings
self-organizing map (LLM SOM). The
neural-network-based methods that
have been implemented include the
multilayer perceptron (MLP) trained
with backpropagation, learning vector
quantization (LVQ), and a variety of
methods based on the SOM. Probabilistic neural networks
(PNNs) have also been used. The study included corn (Zea
mais) (three to seven leaves), sugar beet (Beta vulgaris) (cot-
yledon stadium), buttercup (Ranunculus repens), Canada
thistle (Cirsium arvense), charlock (Sinapis arvensis), chick-
weed (Stellaria media), dandelion (Taraxacum officinale),
grass (Poa annua), redshank (Polygonum persicaria), stinging
nettle (Urtica dioica), wood sorrel (Oxalis europaea), and yel-
low trefoil (Medicago lupulina). Intensity variations in illumi-
nation were observed due to the spatial nonuniformity of the
emission pattern of the light source, to shadows, and to dif-
ferent degrees of specular reflection by leaves with different
orientations. Dividing each spectral band’s value with the
norm of the whole spectrum normalized the spectra.
Forthecorn/weedcase,17discriminatingwavelengthswere
selected, and for the sugar beet/weed case, 18 wavelengths.
The selected wavelengths appear in Table 1. The indicated
number of principal components was used as input to all the
classifiers. Because a very small number of samples were avail-
able from each weed class, this strategy was followed to exploit
theinformationcontentoftheavailabledatatothemaximum.
Results of the classification are shown in Table 2 [8]. The
proposed method proves superior compared to the other
classification methods. The classifier achieved a correct de-
tection rate of 97% for corn, indicating that only 3% of corn is
classifiedasweed.Althoughthedetectionrateofmostindivid-
ual weed species is much lower, the classifier led to a correct
detection rate of 92% for weed in corn. Similar results were ob-
tained for sugar beet, in which case a detection rate of 98%
could be achieved for the beets and 97% for the weeds [8].
Fully Active Horizontal Spray Boom Suspension
Laboratory Experiment
Unevenness in the spray distribution is caused primarily by
Reflectance(%) 60
50
40
30
20
10
0
200 400 600 800 1000 1200 1400 1600 1800 2000
Beet Lambsquarters Redshank Thistle Cockspur Soil
Wavelength [nm]
Figure 8. Examples of measured reflectances in sugar beet canopy [6].
Table 1. Wavelengths selected according to their class sepa-
ration ability together with principal components that con-
tain at least 95% of the original variance (PCA) [8].
Combination Wavelengths (nm) PCA
Corn/weeds 539, 540, 542, 545, 549, 557, 565,
578, 585, 596, 605, 639, 675, 687,
703, 814, 840
5
Sugar beets/
weeds
535, 542, 545, 554, 565, 578, 585,
595, 610, 628, 657, 666, 680, 690,
699, 720, 778, 804
8

malfunctioning of the hydraulic system, wind gusts, advection,
andbyhorizontalandrollingboomvibrations,bothinducedby
soil unevenness. The latter are caused by undesired rotational
motion of the tractor around a horizontal axis pointing in the
travel direction and can be attenuated quite easily by a passive
vertical suspension (e.g., pendulum and trapezoidal suspen-
sion) using gravity force to stabilize the boom.
Yawing and jolting tractor motions are responsible for
undesired horizontal spray boom motions. Yawing of the
tractor causes boom yawing, a rigid-body motion, and
asymmetric elastic deformations of the boom (Fig. 9),
whereas jolting of the tractor induces symmetric elastic
boom deformations.
Selective spraying is an additional argument for suppress-
ing horizontal boom vibrations. In online weed detection, in-
tegration of a defined number of the most recent horizontal
strips of pixels from the optical sensor provides information
for the decision algorithm that activates the spray system
controller. Excessive horizontal boom vibrations can lead to
errors in the decision algorithm and the spray controller. Un-
fortunately, it is extremely difficult to design a horizontal sus-
pension with satisfactory performance due to the absence of
an external force and a reference plane (e.g., gravity and the
soil for the vertical suspension). Until now, only very rudi-
mentary horizontal suspensions are commercially available,
consisting of rubber cushions fixed between the spray boom
and the suspension frame.
An active horizontal suspension has been developed in a
laboratory setup (Fig. 10). The experiment employs a plat-
form activated by two excitation actuators to reproduce
yawing and jolting tractor motions. A sledge is mounted on
top of this frame, with one translational degree of freedom
imposed by the two prismatic joints. The sledge represent-
ing the horizontal suspension bears a 12-m-long commercial
spray boom whose vertical joint preserves one rotational
degree of freedom with respect to the sledge. The complete
suspension provides a yawing and jolting degree of freedom
with regard to the platform.
The active suspension can be conceived either as a vibra-
tion compensator, where boom vibrations are attenuated by
the introduction of active damping into the structure, or as a
vibration isolator attempting to prevent transmission of vi-
brations from the tractor to the boom. In both cases, two
electrohydraulic actuators, installed between the platform
and the spray boom, and two accelerometers, mounted on
the boom and measuring boom accelerations, provide the
active suspension. For the isolator, the accelerometers
should be located at the transmission path of the tractor vi-
brations (i.e., as close as possible to the actuators). For the
compensator, the sensors should be fixed onto the boom
tips, where the displacements of the boom are largest and
easiest to observe.
In the case of the compensator, the complete electro-
hydraulic system, including the suspension and the boom
Table 2. Comparison of classification methods for the corn/weed case (percentages of correctly classified samples).
Probabilistic
Neural Network
Multilayer
Perceptron
Self-Organizing
Map
Learning Vector
Quantization
Local Linear
Mappings
Self-Organizing Map
Corn 93 96 89 92 97
R. repens 51 49 47 51 59
C. arvense 72 68 70 72 77
S. arvensis 70 64 91 70 81
S. media 72 68 66 72 71
T. officinale 66 47 58 66 72
P. annua 64 68 59 64 66
P. persicaria 66 77 58 66 78
U. dioica 46 52 44 44 52
O. europaea 96 99 88 96 99
M. lupulina 85 90 81 84 93
Figure 9. Spray liquid redistribution for yawing boom motion
(top view).

structure,arepartofthecontrolloop,whereasinthecaseofthe
isolator, the boom dynamics are not directly involved in the
control loop. In addition, due to the noncollocated configura-
tion of the compensator, nonminimum phase zeros slip into the
frequency band of interest, rendering control system design ex-
tremely difficult. Logically, isolator design is easier to accom-
plish, and its controller is more robust against changes in spray
boom dynamics.
The feedback control system of the vibration isolator
was developed using the following steps.
• Configuration of the control system and performance
specifications: In the active vibration isolator, the two
hydraulic actuators counteract tractor yawing and
jolting by moving the sledge in the opposite direction.
Tractor accelerations below 0.5 Hz are normally at-
tributed to maneuvers performed by the operator and
therefore must be fully transmitted to the boom. Since
only vibrational modes of the boom with correspond-
ing natural frequencies below 10 Hz contribute to an
uneven spray deposition pattern, the isolator should
attenuate boom accelerations between 0.5 and 10 Hz.
To avoid drift of the pistons in the hydraulic cylin-
ders, an internal proportional position control loop is
provided for each actuator. Linear variable
differential transformer (LVDT) sensors measure the
relative position of the piston rods with respect to the
housing of the corresponding cylinders. The gains of
the proportional controllers are tuned to force the ac-
tuators into a synchronized motion for identical input
signals. Accelerometers with a bandwidth of 150 Hz
measure the transmitted vibrations to the boom. Con-
sequently, the final control system is arranged in a
cascade configuration consisting of a slave loop posi-
tion controller for each hydraulic cylinder and a mas-
ter loop, providing the control actions for the
actuators based on the accelerometer measurements.
• Modeling of the system: Experiments on the setup re-
vealed that by steering the control actuators in phase or
antiphase, only translational or rotational modes of the
mechanical system could be excited. This implies that it
should be possible to derive two separate transfer func-
tions, G st( ) and G sr ( ), for the system, the former de-
scribing the translations and the latter describing the
rotations. Anthonis [9] has used analytical models to
show that the transfer function matrix from the physical
input coordinates (two actuators) to the output coordi-
nates (two accelerometers) is dyadic. This implies that
the transfer function matrix can be diagonalized by
changing the physical coordinates into coordinates ex-
pressing pure translations and rotations such that the
two-input, two-output system is transformed into two
single-input, single-output (SISO) systems.
Plan View
Spring and Hinge
Accelerometer
Prismatic Joint
Platform
Active Suspension Actuators
Sledge
Accelerometer
Spray Boom
Spring and Hinge
Excitation Actuators
Rotation Axis
Level 3
Level 4
Figure 10. Photo and sketch of the experimental arrangement [9].

Due to unsatisfactory accuracy of the
analytical models, a black-box model is
derived from the actuator inputs to the
accelerometer outputs. As the dyadic
structure still applies to the experimental
model, two SISO models are identified.
In the measurement setup, the digi-
tal excitation signal is converted into an
analog signal before applying it to the
hydraulic actuator. Similarly, the cap-
tured response signals from the acceler-
ometers are converted into a digital
output. The sampling frequency of the
excitation signal and the digital output
are selected to be ten times the maxi-
mum frequency of interest, a common
practice in system identification. As 20
Hz is a safe upper limit for the frequency
band of interest, a sampling frequency
of 200 Hz has been selected. Aliasing is
avoided by forwarding the output
through an eighth-order Butterworth
filter with a cut-off frequency of 20 Hz. To avoid inclusion of
the filter in the model, the input signal is filtered too.
The response to four different excitation signals is evaluated
[9]. Each excitation signal contains 4096 points and is applied
periodically to avoid leakage errors. After the system reaches
steady state, ten measurement periods are collected and aver-
aged to minimize the effect of noise in the frequency response
function(FRF).Thefirstandmostcommonlyappliedexcitation
signal in identification is a band-limited random sequence (0 to
20 Hz). Due to its random nature, not all frequency lines in the
frequency band of interest are equally excited, resulting in a
poor signal-to-noise ratio in certain bands. A second applied in-
put signal, the swept sine or chirp, provides better frequency
coverage [10]. To improve the signal-to-noise ratio, the ampli-
tude of the excitation signal could be raised, but in the test
setup the amplitude of excitation is limited by the available
power source for the actuators. Therefore, two special com-
pressed signals are designed, providing an optimal sig-
nal-to-noise ratio for a certain excitation level and minimizing
the measurement time for a given accuracy. A signal that can be
tailored to the needs of the experiment is the multisine. In this
study, a linearly spaced frequency grid is selected between 0.3
and 20 Hz. All selected frequencies receive the same amplitude.
By optimizing the phases, compressed signals are obtained, in-
troducing as much energy as possible into the structure at the
frequencies of interest for a given extreme amplitude of the in-
putsignal.Twodistinctoptimizationschemesareapplied,mini-
mizing the crest factor, but with a different selection of the
starting values of the phases, giving rise to two different excita-
tion signals. Note that the crest factor is defined as the peak
value divided by the effective root mean square (RMS) value of
the signal. For multisine 1, a time-frequency domain-swapping
algorithm [11] is employed and the initial values of the phases
are selected randomly. For multisine 2, the Chebyshev approxi-
mation method [12] is used for minimization and the Schroeder
phase coding [13] is used as a starting value for the optimiza-
tion process. After applying the four excitation signals to the
test setup, the FRF for each signal is calculated for the transla-
tions and the rotations as well (Fig. 11).
Instead of the more commonly used time domain identifi-
cation, a black-box frequency domain identification method
is applied, making it is easier to derive directly continuous
models that are more suitable for H ∞
controller design. Fre-
quency domain identification considers the transform do-
main description of systems and attempts to estimate the
parameters of transfer functions from an estimate of the sys-
tem’s FRF. The nonlinear least-squares estimator is used,
which tries to minimize the squared error between the mea-
sured FRF and the estimate of the FRF represented by a pro-
posed parametric transfer function
( ) ( )$ min $ ,Θ Θ
Θ
= −






=
∑arg FRF j P jk k
k
N
ω ω
2
1 (8)
in which
( )
( )
( )
( )
( ) ( )
$ ,
,
,
P j
B j
A j
b j
j a j
k
k
k
n i k
i
i
n
k
n
n i k
b
b
a
a
ω
ω
ω
ω
ω ω
Θ
Θ
Θ
= =
+
−
=
−
∑0
i
i
na
=
−
∑0
1
(9)
is a parametric estimate of the transfer function of the sys-
tem, evaluated on the imaginary axis at frequency point j kω .
The parametersbn ib − and an ia − are collected in the vectorΘ,
which has to be determined. Indices na and nb represent the
highest degree of the denominator and the numerator, re-
Amplitude[dB]
Amplitude[dB]
Phase[deg]
Phase[deg]
Translations
20 20
200 200
0 0
150
−20 −20
50
50
−40 −40
0 −50
0 0
0 0
5 5
5 5
10 10
10 10
15 15
15 15
20 20
20 20
Rotations
Frequency [Hz]
Frequency [Hz]
Frequency [Hz]
Frequency [Hz]
100
150
100
0
Figure 11. Frequency domain identification results (black line: model, grey line: average
FRF) [9].

spectively. ( )$ ,P j kω Θ stands for a model estimate ofG st( )or
G sr ( ).
An iterative scheme based on Gauss-Newton and
Levenberg-Marquardt algorithms searches the optimal pa-
rameter values of the selected transfer function in a
least-squares solution. The linear least-squares estimate
serves as an initial guess for the iteration.
The set of possible model structures for the parametric
transfer function can be reduced when some prior knowl-
edge is present. The position feedback on the control actu-
ators imposes a position on the central frame of the boom.
As this frame is rigid, and accelerometers measure its mo-
tion, a double differentiator should be incorporated in the
model structure. The presence of a double differentiator is
also visible in the FRFs of Fig. 11. For the translations, a
sixth-order model with numerator and de-
nominator of equal degree seems to provide
the best tradeoff between model complexity
and accuracy. In case of rotations, a
fourth-order model with numerator and de-
nominator of equal degree is selected (Table
3). The identification results are depicted in
Fig. 11.
Control System Design
The controller is designed using H ∞
control
theory in which the H ∞
or Chebyshev norm of
a certain cost function is minimized [9]. In this application,
the multiple-input, multiple-output (MIMO) design reduces
to a SISO design for translations and rotations. Because the
peak values of the control cost function are minimized in H ∞
controller design, it is intuitively understood that the optimal
control cost function is all pass, implying that the maximum
singular-value curve equals unity [14]. Therefore, the H ∞
de-
sign methodology is ideally suited for shaping transfer func-
tions.
A block diagram with all relevant input and output sig-
nals and transfer blocks is depicted in Fig. 12. The absolute
acceleration of the boom with respect to the soil y t( ) is
composed of the relative acceleration of the boom with re-
spect to the platform z t( ) induced by the control actuators
and the acceleration of the platform with respect to the soil
w t( ), representing tractor vibrations. Accelerometers mea-
sure y t( ) on which sensor noise d t( ) is added. It turns out
that the most important component in d t( )is low-frequency
accelerometer drift induced by the amplifiers. The trans-
mission path of vibrations from the tractor to the boom is
represented by the sensitivity function S s( ). This transfer
function should be shaped into a band-stop characteristic
such that vibrations causing large boom motions are fil-
tered on condition that the boom still follows uniform trac-
tor motions and accelerations imposed by the operator. The
decomposition of the system allows applying Fig. 12 sepa-
rately for the translations and the rotations as well.
In an H ∞
framework, shaping is accomplished by search-
ing for a controller such that
( ) ( )αW s S s1 1∞
< (10)
is fulfilled, where α is a tuning parameter. By raising α,
the steepness of the band-stop filter is increased until no
controller can be conceived anymore. At this point, the
optimal controller is found. To accomplish the desired
performance, S s( ) is augmented with a weighting func-
tion W s1( ), amplifying S s( )in the desired frequency band
(Fig. 13). W s1( ), displayed as (11), is constructed by cas-
cading two transfer functions. A tuning parameter α is
added to trade off between robustness and performance
of the controller:
W s
s s
s s
sn n n
d n n
n n
1
2
1
2
2
1 1 1
2
2
2 2
2
2
21 1
( ) =
+ ς +
+ ς +
×
+ ς
α
ω ω
ω ω
ω s
s s
n
d n n
+
+ ς +
ω
ω ω
2
2
2
2 2 2
2
2
.
(11)
Table 3. Poles and zeros of the identified transfer functions Gt
(s) and Gr
(s).
G st( ) G sr ( )
Zeros/2π Poles/2π Zeros/2π Poles/2π
0 −21.08 0 −14.50
0 −1.09210.75i 0 −0.5887±7.574i
−0.6659±11.27i −0.4714±8.026i −0.5963±7.880i −1.028
−0.3937±8.244i −0.8928
Accelerometer
Controller System
d(t)
+
+
−
u(t) z(t)
+
w(t)
H(s)
y(t)
Figure 12. Control problem design scheme.
Amplitude[dB]
Frequency [Hz]
50
45
40
35
30
25
20
15
10
5
0
0 2 4 6 8 10 12 14 16 18 20
Figure 13. AmplitudeplotofthesensitivitydesignweightW1(s)[9].

Parametersωn 1 andωn 2 determine the location and the width
of the band-stop characteristic. Tractor vibrations having a
frequency content close to the first natural frequency of the
boom must be penalized heavily. Therefore, to obtain satis-
factory vibration isolation, the first peak ofW s1( )(i.e., atωn 1)
is placed at the first natural frequency of the boom, which is
1.2 Hz for the translations and the rotations as well. As
high-frequencyvibrationsbeyond5Hzdonotsignificantlyin-
fluence the spray deposition pattern, the second peak of
W s1( ) (i.e., at ωn 2) is selected at 3 Hz. By changing the ratio
ς ςn d1 1/ and ς ςn d2 2/ in magnitude, the heights of the peaks at
ωn 1 and ωn 2 are modified. An additional performance crite-
rion is to avoid the propagation of accelerometer drift to the
output, which is accomplished by passing the accelerometer
signals through a high-pass filter( /( ))s s +1 before they enter
the controller. The final sensitivity func-
tions are shown in Fig. 14.
To guarantee a stable controller on
the real system, model imperfections
must be taken into account. During con-
troller design, a nonconservative multi-
plicative robustness test is performed in
an ad hoc manner [9]. During the H ∞
con-
trol synthesis, the problem formulation
proved to be ill-conditioned due to
jω-axis zeros introduced by the double
differentiators. This problem is solved by
applying the bilinear pole-shifting trans-
form technique [15]. Here the imaginary
axis is shifted 0.1 units to the right, which
seems to be sufficient to remove the
ill-conditionedness.
In addition, during controller design,
the high-pass filters applied to remove
the drift of the accelerometers were not
taken into account, leading to a small
amplification at low frequencies in the
sensitivity function (Fig. 14). Incorpo-
rating the high-pass filters in the controller design cancels
their effect, again resulting in actuator drift. Fortunately,
this small amplification could be reduced by lowering the
pole of the high-pass filter.
Experimental Validation of the Active Vibration Isolator
The controller is implemented on the laboratory setup. Its
performance is validated by measuring the boom tip dur-
ing excitation of the platform by means of a laser. When
slow or fast motions are imposed on the excitation table,
the control actuators do not react. In the midfrequency
range, boom vibrations are attenuated. An example of an
excitation of the boom with a stochastic tractor vibration,
with and without the controller, is depicted in Fig. 15. A re-
duction of the amplitude of the boom by a factor of more
than five is achieved.
Robustness is checked by adding mass to the sledge.
Even with a supplementary weight of 150 kg, which is ap-
proximately two times the weight of the sledge-boom as-
sembly, the controller remained stable and satisfactory per-
formance was achieved. A weight of 10 kg connected to the
boom tip, lowering the first natural frequency of the boom
from 1.2 to 0.7 Hz, could not destabilize the controller. In this
case, performance is lost.
Appropriate Spray Equipment
In selective spraying, the quality of the spray nozzles greatly
influences the dynamic properties of the spray equipment.
Their opening and closing times must be as short as possi-
ble to minimize their contribution to the dead time, time de-
lay, rise time, and peak time of the hydraulic system. It is
advisable that each nozzle operate independently to render
Translations
Amplitude[dB]
Amplitude[dB]Phase[dB]
Phase[dB]
10 10
0 0
−10 −10
−20 −20
−30 −30
10−2
10−2
10−210−2
100
100
100100
102
102
102102
Frequency [Hz] Frequency [Hz]
Frequency [Hz]Frequency [Hz]
Rotations
200
100
0
–100
–200
200
100
0
–100
–200
Figure 14. Designed performance of the controllers (sensitivity function) [9].
Displacement[m]
0.4
0.3
0.2
0.1
0
−0.1
−0.2
−0.3
0 5 10 15 20 25 30
Time [s]
Figure 15. Boom tip motions with (solid line) and without
(dashed line) controller [9].

the spray resolution as small as possible. The nozzles must
be safe to operate, implying a long life cycle and correct dos-
age. Their impact on the pressure in the hoses must be as
small as possible with a view to keeping the droplet spec-
trum stable. The ability to change the flow rate through the
nozzle without influencing the droplet spectrum would be
an advantage.
Solenoid and motor valves are mounted on a spray boom
to lock boom sections and cannot be used for operating indi-
vidual nozzles. During opening and closing, they create
pressure variations in the hydraulic equipment that are dif-
ficult to compensate. Their rise time is high and can in-
crease to 15 s for motor valves, pointing to unacceptably
slow dynamics. In addition, the operational safety of these
valves is questionable. Thus, solenoid and motor valves are
best avoided in selective crop protection.
In air-assisted spraying, the liquid pressure in an air jet
may be increased by a factor of three or four without consid-
erable variation in the droplet spectrum. The disadvantages
of air-assisted spraying are the need for a powerful (>10 kW)
and expensive (>2,500 euros) compressor and a double cir-
cuit, one for the spray liquid and one for the compressed air.
This restricts the application of air jets to the level of boom
sections while operating each individual air jet independ-
ently, making it unrealistically expensive. Air jets are unsuit-
able for patch spraying as well, due to the limited pressure
range in which the droplet spectrum remains constant. A
pressure variation with a factor of four results only in a dou-
bling of the flow rate.
In this respect, pulse-width-modulated
(PWM) nozzles offer new possibilities for selec-
tive spraying. Within a fixed time interval of 0.1
s, the cycle time (or with a cycle frequency of 0.1
Hz), the spray nozzle is switched on and off. To
open the nozzle, an electromagnet moves a pin
made of stainless steel upward against a spring
force. The ratio between the on position (i.e.,
duty cycle) and off position determines the flow
rate through the nozzle, which can vary by a fac-
tor of ten without changing the droplet spec-
trum significantly as long as the pressure in the
conduits remains stable during a variable flow
rate through the nozzle. As electrical conduits
are cheap and easy to install, each nozzle can easily be oper-
ated individually. In addition, Giles [16] showed that PWM
nozzles have very fast dynamics, so their transient behavior
after a new flow rate setting is negligible. However, theoreti-
cal studies supported by experiments [17] proved that with a
cycle frequency of 10 Hz, spray liquid is released in stripes,
especially when the duty cycle is small. To avoid this occur-
ring during spraying, the cycle frequency of PWM nozzles
should at least be doubled.
Fertilizer Spreader
The Problem of Spreading Liquid Manure
During the spreading of liquid manure, several factors may
cause an application that is not in agreement with the needs
of the plants and the capacity of the soil. Taking the actual
demand as determined by soil analyses and the previous
take-away by harvesting, there are some aspects that must
be observed during application.
At first, the manure may not be homogeneous if it was
stored in a manure tank for some time. This effect can be
eliminated by intensive mixing of the
manure within the stationary tank be-
fore filling the tank trailer.
The actual nitrogen content of the
manure must then be determined to
calculate how much (e.g., what vol-
ume) should be applied per hectare.
According to the principles of preci-
sion farming, the amount should be
calculated for small portions of the
field, since the demand may vary
greatly from area to area.
Therefore, the flow controller of the
tank trailer must react to set-point
changes quite rapidly. This also holds,
particularly in hilly regions, for vary-
Pump
Three-Way Valve
with Actuator
Slurry
Flow Rate
Measuring
Device
ControllerActual Speed (Measurement)
Slurry Volume per Hectare (Reference) Application Width
(Parameter)
Figure 16. Slurry flow rate control by flow branching.
Performance specifications for use
in precision agriculture can no
longer be met through a traditional
sequential design of the
mechanism, the controllers, and the
information systems.

ing tractor speed caused by wheel slip. Another situation in
which high-speed action of the flow controller is required is
during startup and stopping of the tractor when reaching
the boundaries of the field. The actual reference values,
measurements, and parameters are transmitted to the
spreader by the agricultural bus system (LBS; in German:
Landwirtschaftliches Bus-System [18]).
Various principles are known for the operation of slurry
tank spreaders. Here, flow control by branching is used: The
(more or less constant) flow of the pump is split into one
stream that is redirected into the tank and another stream
that is fed into the spreading device (Fig. 16). This principle
has the advantages that it does not require a volumetrically
operating pump and that the manure is continuously mixed
in the tank trailer, since a certain part of the pump flow is
refed into the tank.
Needless to say, we consider only the most advanced dis-
tribution systems, such as trailing foot equipment, which al-
low for a precise lateral distribution of the manure and an
outflow of the liquid very close to the soil. The development
reported here is not intended for use with traditional
spreaders such as splash plates, since their distribution
precision is not sufficient and unpredictable losses of nitro-
gen by ammonia emissions occur. Injection (trenching) be-
low the soil surface is also possible and greatly reduces
odor and ammonia emissions; however, nitrous oxide emis-
sions were reported to increase significantly (up to 230%),
making this kind of application inappropriate from a green-
house gas emissions standpoint [19].
Sensors and Actuators
According to Fig. 16, a sensor for determining the true speed
of the tractor must be available. This could consist of a
Doppler radar device [20] or a DGPS Doppler device, as
pointed out by Han [21]. In this way, the DGPS system may
serve for exact determination of position and speed. Thus,
two devices exist that provide sufficiently precise and reli-
able measurement data for the actual speed; therefore, the
problem of speed measurement is considered solved and
will not be discussed further here. Information about the ac-
tual speed is provided as a service by the LBS.
Several devices are available for sensing the manure flow
rate. Details of comparative tests were reported in [22] and
[23]. Precision and dynamic responses of magnetic induc-
tive devices (MIDs) proved to be the best of all sensors
tested, but even the dynamics of the MIDs proved to be too
slow for control purposes. For example, the comparison of
two MIDs from well-known manufacturers gave the follow-
ing results: MID1 exhibits a time delay of 1.2 s, followed by a
fast rise (lag time of approximately 0.25 s); MID2 shows a
shorter time delay of 0.5 s but is accompanied by a
first-order lag with lag time of more than 1 s. The latter may
be adjusted within narrow bounds. The implications for
control will be discussed later.
As for the actuators, the three-way valve may be manipu-
lated by hydraulic cylinders or an electric motor. Due to the
power requirements for fast motion of the three-way valve,
and the fact that a typical tractor provides greater hydraulic
than electrical power, the hydraulic motion is favored.
Control
Commercial Solutions
Flow controllers installed on commercially available slurry
tank trailers are usually equipped with three-point switches
as output devices. This is reasonable, since they reduce the
cost of the equipment substantially compared with a fully
analog power output. At the same time, the probability of
the valve becoming stuck is considerably reduced, since the
full hydraulic power can be imposed for every change in the
valve’s angular position. This is even more important for the
electric motor, since its force is much weaker, and the
valve’s position will probably not change if the full power of
the motor is not applied.
The time needed for full opening of the valve, starting
from the completely closed and ending at the completely
open position, which means a change in the angular posi-
tionαvalve from 0° to 50° in our case, was measured as 0.3 to
0.4 s for the hydraulic cylinder and 3.5 s for the electric mo-
tor. This means that closed-loop control with the electric ac-
tuator is achievable, whereas stable operation of the control
loop with the hydraulic actuator is not possible. Due to the
slow dynamics of the MID, the valve is completely
opened/closed before the measurement value has reported
any change in the flow rate.
A Kalman-Filter-Based Approach
The control problem addressed above can be solved since
there exists an (almost) nondynamical indirect measure-
ment that is related to the flow rate: the angular position of
the valve. Thus, a Kalman filter in conjunction with a Smith
predictor can be used to generate delay- and lag-free esti-
mates of the flow rate.
The relation between α, the angular position of the valve,
and FS , the true flow rate to the spreader, is a nonlinear one
that can also vary in time. The reason for the latter is that
the pump flow rate (even without the valve) varies depend-
ing on the viscosity of the slurry, the presence of obstacles
in the tube and hose system of the trailer, the pressure at the
entry of the pump (depending on the liquid level in the
tank), and the operating width of the application system
(part of the trailing feet may be switched off when reaching
the field boundary). The nonlinearity of the valve occurs
mainly in the fully open and fully closed positions, resulting
in an S-shaped pump/valve characteristic.
These effects lead to three consequences:
1) The pump/valve characteristic must be modeled as
nonlinear. Here, an approximation by two basis func-
tions was chosen, namely, a linear part with slope
Kvalve and a nonlinear part with sinusoidal shape and

gain factor Nvalve . This approximation is superior to a
more simple one that uses only one gain. That approx-
imation is applicable, too [24], but the gain must then
be adapted continuously. This indicates that such a
simple representation of the pump/valve characteris-
tic is only a formal approximation but does not pro-
vide sufficient prediction capabilities.
2) The factors Kvalve and Nvalve vary with time in an unpre-
dictable way. Therefore, they must be included in the
estimated state such that the model becomes nonlin-
ear and an extended Kalman filter must be used.
3) The linear dynamical part of the filter consists of the
model for the time lag of the MID only, which is a
first-order, time-lag system.
The (delay-free) model equations for design of the ex-
tended Kalman filter are as follows.
• Model for generation of the undisturbed MID output
signal (in first-order lag notation):
T
dF t
dt
F t
K t t N t
MID
MID
MID
valve valve
∗
∗
+
= − ⋅
( )
( )
( ) ( ) ( ) sα in
( )
( ).2
50
1π
α t
w t
°
+
(12)
• Model for the slope of the pump/valve system:
dK t
dt
w tvalve ( )
( )= +0 2 .
(13)
• Model for the nonlinear gain of the pump/valve sys-
tem:
dN t
dt
w tvalve ( )
( )= +0 3 .
(14)
• Model for generation of the real (noise-disturbed) MID
output signal:
F t F t v tMID MID( ) ( ) ( )= +∗
. (15)
Here wi and v are zero-mean, Gaussian white noise compo-
nents, and TMID =1 s. Since 0 50≤ ≤ °α and 0 500≤ ≤FS
L/min, a reasonable first estimate for the valve gain is
Kvalve =10 L/(min−deg). Nvalve is initially estimated as
Nvalve = 50 L/min. The zeros in (13) and (14) are included to
make clear that Kvalve and Nvalve are regarded as constant.
The system is linearized, and noise variances are defined
as follows:
{ } { } { } { }E w E w E w E v1
2
2
2
3
2 2
10 01 100 100= = = =, . , , .

50
α
Fs
Slurry Flow Rate
(Controlled Variable)
Time Lag
of MID
Delay
of MID
FMID
Plant
Valve and Pump
Actuator
Hydraulic Cylinder
Controller
Three-Point Switch
Fref
Fest
Angular
Position
Sensor
Model for SP
Direct
Input
F
L( )α
Delay for SP
Delay for EKF
−
−
−
−
α
Nonlinear Valve/Pump Model
Model for EKF
Direct
Input
F
K
N
Kalman Gain
K N sin 2• α − • π
Figure 17. Schematic diagram of the complete control loop; upper part: controller, actuator, plant, and measurement device; lower part:
extended Kalman filter with Smith predictor.

The output noise reflects the measurement accuracy of the
flow meter (2% of full scale), whereas the system noise com-
ponents refer mainly to the desired dynamics of variation in
the estimated parameters. With these assumptions, the
Kalman gain is computed for various set points α. A
first-order approximation of the gain factor curves results in
L( )
. .
. .
. .
α
α
α
α
=
⋅ +
⋅ +
⋅ −


00144 0426
00005 00057
00325 1092






.
(16)
The estimate of the flow rate is finally computed by
F t K t t N t
t
est valve valve( ) ( ) ( ) ( ) sin
( )
(= ⋅ − ⋅
°
< <α π
α
α2
50
0 50°).
(17)
Until now, the delay of the MID has not been considered
(τMID s= 05. ). For the continuous-time filter, this was included
by a Smith predictor (SP), which is built around the Kalman
gain. The Kalman feedback gain is considered as the regula-
tor of the EKF model control loop. This means that, at first, a
delay-free design of the control loop is possible, as noted
above, and subsequently the delay is taken care of via the
SP. The structure of the complete estimator is shown in the
lower part of Fig. 17. In addition to the filter, other compo-
nents of the control loop are shown: controller, actuator,
valve, MID. Within the filter, the upper loop is the Smith pre-
dictor loop, whereas the lower loop is the classical ex-
tended Kalman filter. The nonlinear block at the input of the
filter represents (17).
This combination of EKF and SP performs very well, as
demonstrated by simulation in [25]. Here, practical results
from the test stand at the Institute for Technology and
Biosystems Engineering of the Federal Agricultural Re-
search Centre (FAL) in Braunschweig, Germany, can be re-
ported. This test unit consists of the entire equipment of a
tank trailer, with a second tank to store the part of the liquid
manure flow that in practice is spread to the field.
The three-point switch was modified to provide better
dynamic performance. In each sample interval (sample time
= 1 ms), the difference between the estimate of the flow rate
and the reference value is computed, as well as the esti-
mated time for the hydraulically driven valve to attain a po-
sition corresponding to a flow that equals the reference. A
hydraulic pulse of corresponding length is then scheduled.
During the following samples, this estimated pulse length is
further adapted to currently measured and estimated val-
ues. This principle corresponds to a strategy of adaptive
predictive control with receding horizon and dead-beat.
In contrast to simulations, there is no chance to measure
or compute a real control deviation, since there exists no
measurement of the true and actual slurry flow rate. The
only available data for the actual flow rate are the estimates
of the filter and the MID measurements. The latter, however,
suffer from delay and time lag.
Fig. 18(a) shows part of a test run of about 70 s. The pre-
diction behavior of the filter is easily verified; in fact, the
solid red line representing the reference and the dotted
black line of the filter output can hardly be distinguished in
this scaling. At 24 s, the filter shows some motion, which is
due to a control action of the valve; the control algorithm re-
quires some fine-tuning. Fig. 18(b) demonstrates detail from
another run (4 s). The course starts with a control deviation
of approximately 10 L/min, which is less than 3% of the ac-
tual value of 370 L/min. The ramp in the reference causes
some controller action, which leads to a change in the valve
position after about 100 ms. Here, the capacity of the hy-
draulic components should be increased in the future to
supply hydraulic power more quickly and with more power.
The estimate of the Kalman filter with Smith predictor is
FlowRate[L/min]
FlowRate[L/min]
500
450
400
350
300
250
200
150
100
50
0
0 10 20 30 40 50 60 70
400
350
300
250
200
150
26 26.5 27 27.5 28 28.5 29 29.5 30
Time [s] Time [s]
Reference
MID Signal
Filter Output
Reference
MID Signal
Filter Output
(a) (b)
Figure 18. (a) Time course of one experiment in the test stand. (b) Time course of another experiment; detail.

considerably faster than the MID signal. The increase in the
MID signal, which is observed at the beginning of the tran-
sient phase, has no causal connection to the opening of the
valve; some nonstationary behavior can be observed from
time to time during the experiments.
Although the controller stops exactly at the point where
the estimated flow rate equals the reference value, the sub-
sequent estimates for the coefficients K and N lead to a cor-
rection for the estimate of the flow rate such that further
control action is required.
In summary, one can state that the delay and lag times of
the MID can be overcome by the designed filter and predic-
tor combination. Control action becomes much faster than
with the original commercial equipment. Additional work
on the test stand is necessary to clarify some observed ef-
fects and for fine-tuning.
Networks in Agriculture
The previous sections of this article described efforts to im-
prove agricultural machines, as well as the sensors and ac-
tuators used on them. To make use of the machines in an
efficient way and in accordance with various existing regula-
tions, further higher level information must be taken into ac-
count from different areas of the surroundings in the
broadest sense. This is true for both the examples listed
here and the complete range of agricultural production. The
application of this information for production planning and
the production process itself will increasingly be done with
the help of network systems. Fig. 19 schematically shows ag-
riculture embedded into its environment with various (mu-
tual) influences and effects. Arrows mark (main) influences.
The surrounding conditions are also interdependent, as in-
dicated. Note that the compilation in Fig. 19 is very incom-
plete; the number of influencing effects is much larger, and
the same is true for the number of interactions. In addition,
data transmission paths within the Internet are marked in
the picture that already exist or will be created in the near
future. An extensive network of agricultural institutions al-
ready exists (including agricultural software suppliers, ma-
chinery industry, administration, and public agricultural
information services such as the German DAINet). These
well-established services also make use of existing data net-
works such as the Internet.
For production management at the farm level, data must
be exchanged between production planning (mostly sta-
tionary) and production facilities (stationary and/or mo-
bile). Within the production facilities, data are also
transmitted for control purposes. Here, completely differ-
ent conditions must be met with regard to the amount and
time scale of data communication. These conditions cannot
Commerical
Networks (Internet)
Mode of Action
Communication
Lines
Soil
(Analysis)
(Agricultural)
Engineering
(Technical Data,
Service Areas)
Legislation,
Regulations
(Plant Protection,
Use of Fertilizer)
Biological Effects;
Genetic
Engineering
Research and
Development
(New Procedures and
Algorithms)
Climatic
Conditions
(Climate Data from
Weather Services)
Finance;
Commerce
(Banks, Trade,
Cooperatives)
Employment;
Staff
(Seasonal and
Temporary Workers)
Storage,
Conservation,
Processing
Temporal
Conditions
(Use of Services and
Machines)
Product Quality
and Quantity
Agriculture
Figure 19. The various influences that come to bear on agriculture and the relevant data communication channels.

be fulfilled by global networks such as the Internet. There-
fore, in this area, completely different data transmission
techniques must be used. Complex electronic control sys-
tems can only operate efficiently if their various compo-
nents are able to exchange data automatically. To ensure
compatible data exchange between different types of farm
equipment from different manufacturers, standardized data
communication systems need to be installed.
At present, the development and design of farm-specific
data networks have made greatest progress in the area of
plant production. Therefore, the following explanation con-
centrates on two networks and their standards (DIN 9684
and ISO 11783), which are designed for mobile agricultural
machinery. These networks mainly serve to exchange pro-
cess data, which are necessary for technical control, to in-
form the operator, and to exchange data with stationary
farm computers. It must be noted that the following text is
only a very concentrated summary of the comprehensive
standardization documents.
Network Realizations
in Plant Production
In plant production, some very special features exist. Pro-
duction processes are typically performed by mobile ma-
chinery, which often consists of combinations of several
working machines or agricultural implements. Modern ma-
chines and implements are controlled by electronic control
units (ECUs). These ECUs are coupled by a network as
shown in Fig. 20. This network additionally includes a hu-
man-machine interface (User Station) and a computer inter-
face between the mobile and stationary system areas (Task
Controller 1).
The German Agricultural Bus standard (LBS) [18] and the
Agricultural Bus standardized in ISO 11783 (Tractors, Ma-
chinery for Agriculture and Forestry—Serial Control and
Communication Network) provide open interconnection sys-
tems for on-board electronic systems [26], [27]. The main
purpose of the LBS is to standardize data transmission be-
tween different machines or parts of machines (tractor to im-
plement, tractor and implement to user station, tractor and
implement to farm computer, etc.), whereas the well-known
SAE J1939 standard is concerned with data exchange be-
tween various units belonging to one machine [28].
In designing such networks, several fundamental re-
quirements and preconditions must be considered.
• The network is anticipated as a basis for setting up and
running distributed process control systems (e.g., con-
trol of the distribution of fertilizer, application of pesti-
cides, irrigation). For these tasks, the network must
exchange data between technical components of the
agricultural machines with low time delay.
• Production processes are often performed by combi-
nations of machines and implements that are manu-
factured by different international companies. This
calls for a standardized network.
• In such combinations, implements are changed fre-
quently, which causes multiple connections and
disconnections at the physical bus line. Therefore,
Part 4 *)
Part 5 *)
LBS, DIN 9684
Physical Bus, Protocol; Part 2 *)
Part 3 *)Part 3 *)
User Interface
LBS Service
“User Station”
Mobile Stationary
Computer Interface
LBS Service
“Task Controller 1”
Diagnosis Interface
LBS Service
“Internal Diagnosis”
Front Implement 1
Electronic
Control Unit
Tractor
Electronic
Control Unit
Towed Implement 1
Electronic
Control Unit
Towed Implement n
Electronic
Control Unit
Plant Production (Mobile), Tractor-Implement Combination
*) Part of the Standard
with the Main Definitions
for the Specific Topic
Figure 20. Example of network in accordance with DIN 9684.

a serial bus with simple connectors and cables is
preferred.
• The changing of implements always alters network
configuration. To avoid additional workload for the
operator and the need for a special computer to do
network administration, the entire network must be
able to monitor, control, and reconfigure itself auto-
matically.
• The network must allow the operator to monitor and
control the machinery combination.
• For automated information-based farming, such as
precision farming using field operation maps with
position-specific set points, data must be available.
These data are prepared on the stationary farm com-
puter during production planning. Conversely, mea-
sured values such as soil parameters, yield data, and
the like, collected during field operation, have to be
transmitted to the farm computer as a basis for later
production planning. This can only be done by using
machine-readable data storage and exchange.
The five parts of DIN 9684 are derived from these precon-
ditions for the LBS.
Part 1: Point-to-Point Connection (Not Relevant Here)
The standard DIN 9684, part 1 (Agricultural Tractors and
Machinery—Interfaces for Signal Transfer—Point-to-Point
Connection) was published in 1989. Data already measured
on the tractor are made available by it for use on agricul-
tural implements. These data (ground speed, rotation
speed of power take-off (PTO), and the position of the
hitch) are transmitted in the form of pulses or as an analog
voltage signal. The standard was revised in 1995 and pub-
lished as international standard ISO 11786 (Agricultural
Tractors and Machinery—Tractor-Mounted Sensor Inter-
face Specifications).
Part 2: Serial Data Bus—Transport Protocol and Physical
Layer
Part 2 of the standard defines the data exchange protocol
and the physical bus of the LBS. For the LBS, the commercial
protocol CAN Version 2.0A [29] was selected. CAN (control-
ler area network) was designed by the German company
Bosch and was originally planned for use in automobiles.
CAN is conceptually a network for object-oriented data
transfer with random access and collision detection to the
bus (CSMA/CD) and with priority control. Object identifica-
tion and priority control are done by a CAN identifier (11-bit
length, 2048 objects), which must be unique and unambigu-
ous for all nodes. The CAN protocol, especially the use of the
CAN identifier (cf. Table 4), must be adapted for application
to the LBS because:
• The LBS is an open network for agricultural purposes
with frequently changing configurations of very differ-
ent combinations of field machinery. It needs a much
larger number of data objects.
• Additional identifier information needs to be placed
inside the data telegram to label the greater number of
data objects.
• The CAN identifier must be kept unambiguous. This is
realized by including dynamic addresses of the trans-
mitting nodes inside the CAN identifier.
• The dynamic addresses are only valid for the actual
network configuration and are defined during the ini-
tialization process.
• To retain priority control of messages, the CAN identi-
fier is subdivided into eight function groups, which
have different priorities.
This part also includes the definition of the physical net-
work layer:
• Data will be transmitted on a pair of twisted wires.
• The length of the bus line is limited to 40 m.
Table 4. Function groups of the LBS (use of the 11-bit CAN identifier).
Function Group Priority
(3 bit)
First Parameter Second Parameter
System control 0 Switch for logon/
system management (1 bit)
Implement (node) designator:
Type and position (7 bit)
Basis messages 1 Designator of basis message (4 bit) Transmitter address (4 bit)
Targeted messages 2 Receiver address (4 bit) Transmitter address (4 bit)
LBS services;
service ⇒ node
3 Receiver address (4 bit) Service designator
(transmitter address) (4 bit)
LBS services;
node ⇒ service
4 Service designator
(receiver address) (4 bit)
Transmitter address (4 bit)
Partner systems 5 Free (4 bit) Master address of partner system
(transmitter address) (4 bit)
Free 6 Free Free
Free 7 Free Free

• The transmission rate is 125 kbit/s (approximately
1,000 CAN frames/s).
• The number of physical nodes is limited to 20.
Part 3: System Functions, Identifier—Network Management
and Identifier Structure in the LBS
Part 3 of the standard first defines the network management.
The main parts of network management are the automatic
initialization, procedures to claim dynamic addresses, and a
management system to control nodes on the system level,
such as monitoring of active participants or release of inac-
tive nodes. These tasks have the highest priority of the eight
function groups (function group system control) in the LBS.
The function group “Basis Messages” with the next prior-
ity level is used to transmit so-called basis data. These data
are transmitted cyclically in a packed format and are pro-
vided to all active nodes on the bus (broadcast transmission
of measured values of ground speed, rotation speeds of the
engine, PTO, the position of the hitch, or time and calendar
information). Data for process control purposes, the LBS
process data, are also transmitted using this function group.
Process data are labeled with an extended data identifier in-
side the CAN data telegram. The identifier is structured;
thus, it is possible to recognize the meaning and relevance
of the data directly by filtering. The function group “Tar-
geted Messages” offers another way to transmit process
data with the possibility of sending the information directly
to an LBS participant (active node at the bus), which is spec-
ified by its dynamic address.
The network also offers in the number of CAN identifi-
ers a slot for OEM-defined data exchange to make it possi-
ble to use the protocol and data line for OEM-specific
purposes with the so-called LBS Partner Systems. Only a
few constraints from the LBS system management are pre-
scribed.
Two function groups are necessary for the use of LBS ser-
vices: one for the data direction from the service to the
nodes and one for the opposite direction. Part 3 only defines
the general handling of the LBS services. Because these ser-
vices are very different in functionality, each has its own de-
scription. Currently two services are defined.
The remaining two function groups are reserved for fu-
ture expansion.
Table 5. Comparison of parts of the ISO 11783 and DIN 9684 standards.
Number Title Status as of June 2000 Equivalence to LBS
ISO 11783 Tractors, Machinery for Agriculture
and Forestry—Serial Control and
Communication Data Network
DIN 9684 Agricultural Tractors
and Machinery—Interfaces for
Signal Transfer
ISO 11783, Part 1 General Standard for Agriculture
Mobile Data Communications
Working Draft (WD) DIN 9684, Part 2: Serial Data Bus
ISO 11783, Part 2 Physical Layer Final Draft Interna-
tional Standard (FDIS)
DIN 9684, Part 2
ISO 11783, Part 3 Data Link Layer Harmonized with SAE
J1939/21
DIN 9684, Part 3: System Func-
tions, Identifier
ISO 11783, Part 4 Network Layer for Agriculture Mo-
bile Data Communication
FDIS DIN 9684, Part 2 and Part 3
ISO 11783, Part 5 Network Management for Agricul-
ture Mobile Data Communication
FDIS DIN 9684, Part 3
ISO 11783, Part 6 Virtual Terminal Draft International
Standard (DIS)
DIN 9684, Part 4: User Station
ISO 11783, Part 7 Implement Messages Application
Layer for Agriculture
DIS DIN 9684, Part 3
ISO 11783, Part 8 Power Train Application Layer Harmonized with SAE
J1939/71
ISO 11783, Part 9 Tractor ECU Network Interconnec-
tion Unit
DIS DIN 9684, Part 3
ISO 11783, Part 10 Task Controller Application Layer
for Agriculture
WD DIN 9684, Part 5: Data Exchange
with the Management Informa-
tion System, Task Controller 1
ISO 11783, Part 11 Mobile Agriculture Data Element
Dictionary
WD

Part 4: User Station
Part 4 of the standard defines the human-machine interface,
the LBS service user station. This service includes elements
for display, data acquisition (alphanumeric keypad, soft
keys, joysticks, etc.), and direct control of machine func-
tions (function keys) for all active participants on the bus.
On the operator side, the user station has several differ-
ent graphic screens. A Data Display presents data relevant
to the working process. An Alarm Display informs the opera-
tor about alarm situations. A Soft Key Display shows the ac-
tual meaning of soft keys or menu items for menu control. A
Function Key Display indicates the meaning of function keys
for direct control of machine functions. The operator can al-
locate these screens and keys to a specific active LBS partic-
ipant. As the standard does not prescribe any physical
design of the user station, the manufacturer is allowed to in-
corporate its own ideas and has the freedom to install a
number of display and key groups for several implements,
such as front and rear hooked implements.
On the LBS participant side, the user station is a virtual
unit. This means the user station permanently simulates
availability for the participants, as well as for implements
that are not selected by the operator. The participants use
the user station with the help of predefined screen contents.
For this operating mode, each participant defines and loads
resources into the user station, for example, during the ini-
tialization process. Resources include all instructions to
generate display images or dialogue elements. Each partici-
pant can only use its own resources. Exchanging or updat-
ing of screen contents are initiated by a small number of LBS
data telegrams. The use of stored resources avoids a large
bus load during the working process of field machinery.
Part 5: Data Exchange with the Management Information Sys-
tem—LBS Service “Task Controller 1"
Part 5 of the standard defines the LBS service Task Control-
ler 1, the mobile-stationary interface. This service includes
three main parts. The first part is a nonstandard communi-
cation medium for transport between the stationary and the
mobile areas.
The second part is a standardized interface between ar-
bitrary management programs inside the stationary farm
computer and the communication medium. This interface
consists of the definition of standardized transfer files that
contain task control data, measurement data, and machine
data. On one side, the management programs have to gener-
ate or receive these transfer files, and on the opposite side, a
driver program has to handle the data exchange between
transfer files and the communication medium.
The third part is a standardized interface between the
medium and the ECU in the implement, which is placed in
the mobile area. It acts as a process controller, using incom-
ing process control data, machine data, and data about the
actual position in the field. The results are sent to the ECUs
via the CAN bus according to the definitions of LBS process
data. Measured data are collected by the service during the
working process. After field work is completed, these data
are stored at the communication medium.
Future LBS Expansion
The standards of the LBS can be adjusted to the expanding
demands of agriculture in appropriate periods. Therefore, it
must be possible to make adjustments by updating the rele-
vant software. The LBS already incorporates placeholders
for future tasks such as transparent data transmission,
printer capabilities, and different diagnoses [30].
ISO 11783: The International Standard
for an Agricultural Bus System
Agricultural machinery is produced by international indus-
try, so only international standardization can guarantee un-
impeded data transfer between agricultural systems.
Nienhaus [31] reports that in 1988, the establishment of a
subcommittee for electronics was discussed in Technical
Committee 23 (TC 23) of ISO. Subsequently, an independent
subcommittee, SC 19, with working group 1, the WG1 Mobile
Machinery group, was established, which is responsible for
the standardization of the agricultural bus according to ISO
11783.
Concluding Remarks
Site-specific agriculture requires the application of field ma-
chinery capable of precise, repeatable operations based on
models of systems processes. Such equipment requires a
host of high-precision sensors and actuators. Unfortu-
nately, most performance specifications for machinery sys-
tems used in precision agriculture can no longer be met
through the traditional sequential design of the mechanism,
the controllers, and the information systems. In the
mechatronic design process outlined in this article, the effi-
ciency of the design process and the performance of the
mechanisms can be improved considerably or even be opti-
mized through concurrent, integrated development of the
mechanisms, control systems, and advanced information
systems. Such advanced sensing systems with modern feed-
back controllers can generate significant demands for data
processing and require substantial communications band-
width. Standardized agricultural bus systems form the
backbone for the high-variability and high-bandwidth data
streams.
In this article, three example mechatronic designs of mo-
bile agricultural machinery have been discussed, and the
requisite communication system for these machines has
been presented.
In the first example, the problem of mechanical grain
yield sensors placed on combines was described. Such yield
sensors work by registering forces exerted by the harvested
mass grain as it flows onto the sensor’s chute and could be
very susceptible to variations in friction properties of the
grain kernels. Thus, the sensor must be recalibrated at regu-
lar intervals. The friction force is a function of a large num-
ber of parameters, including: entrance speed of grain,

inclination angle of the assembly, arc length and radius of
curvature of the chute, distance between the center of cur-
vature and the pivot point, the length of the pendulum rods,
and the orientation of the rods with respect to the plate. By
proper choice of these design parameters, the contribution
of friction in the measured force was reduced to an insignifi-
cant level. For the optimized sensor, the total number of cali-
brations can be limited to once per harvesting season,
independent of the condition and type of crop harvested.
However, sensor outputs can become contaminated by
smearing effects of the grain flow. An algorithm that de-
scribes the dynamics of grain flow eliminated these effects
by adapting the sensor placement, allowing more accurate
data for yield map construction.
Targeted spraying requires an integrated adaptation of
field-spraying machines on three levels: the equipping of a
spray boom with optical sensors for weed detection, the sta-
bilization of the spray boom to ensure correct location of the
spray nozzles on the target after optical detection, and the
improvement of the dynamics of the spray equipment hy-
draulics for fast and correct release of the prescribed dose.
Visual classifiers discriminate between field crops and weeds
based on a minimal number of spectral lines in the near-infra-
red band as registered by the optical sensors. For the
corn/weeds case, the proposed neural-network-based
method achieved a correct classification of 97% for corn and
92% for the weeds. In the sugar beet/weed case, it led to 98%
correct classification for sugar beets and 97% for the weeds.
Horizontal boom vibrations can create a mismatch between
spraynozzlesandtheweeddetectionsystem.Aspassivesus-
pensions are inadequate for suppression of boom vibrations,
a full active boom suspension was developed in which two
electrohydraulic actuators isolated the boom from tractor vi-
brations. At the first natural mode of the boom, amplitudes of
the vibrations were reduced by a factor of more than five. Op-
erating spray nozzles by pulse-width modulation of the sup-
ply can considerably improve the dynamic behavior of
hydraulic spray equipment, if the cycle frequency of the noz-
zles can be increased to at least 20 Hz.
Uniform, accurate spreading of liquid manure has high
relevance for crops and the environment. To avoid nitrogen
losses to the air (ammonia volatilization), the operation
should be carried out very close to the soil (e.g., by trailing
feet). Furthermore, the valuable fertilizer must be applied to
the plants according to actual demand, which varies with
the status of the plant, weather conditions, soil composi-
tion, nutrient content, humidity, and many other parame-
ters. This implies spatially variable dosing for the
application flow rate, which significantly increases the me-
chanical and control complexity of the machine. Actuators,
sensors, and control equipment for manure flow rate must
be considered. An advantage is that agricultural mobile ma-
chines have powerful hydraulic systems, enabling rapid
valve action, which must be accompanied by rapidly react-
ing sensors. Here we are faced with the problem that the
best-suited sensor with respect to robustness and precision
has built-in signal preprocessing with delay and time lag.
This problem could be overcome by implementing an ex-
tended Kalman filter with Smith predictor. The controller
consists of a three-point switch with an adaptive predictive
control strategy. Promising results from practical runs on a
test stand were demonstrated in this study.
Communications networking of production units has be-
come an important feature of agricultural production pro-
cesses and can be expected to continue to grow. Farm
operations can communicate with weather services, trad-
ers, contractors, suppliers, biological services, consultants,
and many other organizations. In these applications, the
Internet already plays a key role. For on-farm communica-
tion, which is mainly used for online or inline applications
on or among tractors and implements, a specific communi-
cation system, the agricultural bus system, has been devel-
oped. This standardized communication system serves as
the backbone for precision agriculture, as demonstrated by
the examples in this study.
References
[1] B. Missotten, “Measurement systems for the mapping and the evaluation
of crop production performance,” Ph.D. dissertation, K.U. Leuven, 1998.
[2] P. Reyns, B. Missotten, H. Ramon, and J. De Baerdemaeker, “A review of
combine sensors for precision farming,” Precision Agriculture J., to be
published.
[3] G. Strubbe, “Mechanics of friction compensation in mass flow measure-
ment of bulk solids,” Ph.D. dissertation, K.U. Leuven, 1997.
[4] K. Maertens, J. De Baerdemaeker, H. Ramon, and R. De Keyser, “An analyti-
cal grain flow model of a combine harvester: Part 1, design of the model,” J.
Agricultural Eng. Res., vol. 79, no. 1, pp. 55-63, 2001.
[5] K. Maertens, J. De Baerdemaeker, H. Ramon, and R. De Keyser, “An analyti-
cal grain flow model of a combine harvester: Part 2, analysis and application
of the model,” J. Agricultural Eng. Res., vol. 79, no. 2, pp. 187-193, 2001.
[6] E. Vrindts, “Automatic recognition of weeds with optical techniques as a
basis for site-specific spraying,” Ph.D. dissertation, K.U. Leuven, 2000.
[7] D. Moshou, H. Ramon, and J. De Baerdemaeker, “A weed species spectral
detector based on neural networks,” Precision Agriculture J., to be published.
[8] D. Moshou, E. Vrindts, B. De Ketelaere, J. De Baerdemaeker, and H. Ramon,
“A neural network based plant classifier,” J. Computers Electronics in Agricul-
ture (COMPAG), vol. 31, no. 1, pp. 5-16, 2001.
[9] J. Anthonis, “Design and development of an active horizontal suspension
for agricultural spray booms,” Ph.D. dissertation, K.U. Leuven, 2000.
[10] L. Clijmans, “Model-based approach to assess sprayer’s quality,” Ph.D.
dissertation, K.U. Leuven, 1999.
[11] E. Van Der Ouderaa, J. Schoukens, and J. Renneboog, “Peak factor
minimization using a time-frequency domain swapping algorithm,” IEEE
Trans. Instrumen. Measure., vol. 37, no. 1, pp. 145-147, 1988.
[12] P. Guillaume, J. Schoukens, R. Pintelon, and I. Kollár, “Crest-factor
minimization using nonlinear Chebyshev approximation methods,” IEEE
Trans. Instrumen. Measure., vol. 40, no. 6, pp. 982-989, 1991.

[13] M.R. Schroeder, “Synthesis of low peak factor signals and binary se-
quences with low autocorrelation,” IEEE Trans. Inform. Theory, vol. IT-16, pp.
85-89, 1970.
[14] R.Y. Chiang and M.G. Safonov, Robust Control Toolbox User’s Guide.
Natick, MA: MathWorks, 1992.
[15] R.Y. Chiang and M.G. Safonov, “H∞ synthesis using a bilinear pole shifting
transform,” J. Guidance, Contr. Dynam., vol. 15, no. 5, pp. 1111-1117, 1992.
[16] D.K. Giles, “Distributed network system for control of droplet size and
application rate for precision chemical application,” in Proc. Precision
Agriculture’99, Odense, Denmark, 1999, pp. 857-866.
[17] R. Delen, “Modelling of the behavior of a pulse width modulated spray
nozzle,” M.Sc. thesis, K.U. Leuven (in Dutch), 2000.
[18] DIN 9684: Landmaschinen und Traktoren—Schnittstellen zur
Signalübertragung. (DIN 9684: Agricultural implements and tractors—Inter-
face for signal transmission). Teil 1, Punkt-zu-Punkt-Verbindung (point to
point connection ), 1989, revised 1995. Teil 2, Serieller Daten-BUS (serial data
BUS), Jan. 1998. Teil 3, Systemfunktionen, Identifier (system functions, identi-
fier), Juli 1997. Teil 4, Benutzerstation (user station), Gelbdruck, April 1997.
Teil 5, Datenübertragung zum Management-Information-System,
Auftragsbearbeitung 1 (data exchange with the management information sys-
tem, task controller 1), Gelbdruck, April 1997. Berlin: Beuth Verlag GmbH,
1989-1998.
[19] E. Schürer and H. Kutzbach, “Emissions of nitrous oxide and methane af-
ter slurry application in grassland,” in Proc. AgEng98, Oslo, paper 98-E-012,
1998.
[20] Technical Data of Dickey-John Corporation, Auburn, IL, U.S.A., 1993.
[21] Y. Han, Personal communication. AERODATA, Braunschweig, 1994.
[22] E. Buning, “Ein Beitrag zur Optimierung der Längsverteilung von
Flüssigmist,” Dr.-Ing. dissertation, Tech. Univ. Braunschweig, Germany, 1997.
[23] E. Buning, A. Munack, and H. Speckmann, “Components and control sys-
tems design for high performance spreading of liquid manure,” in Proc. 13th
Int. CIGR Congress Agricultural Engineering, Rabat, vol. 2, pp. 333-338, 1998.
[24] A. Munack, E. Buning, and H. Speckmann, “A high performance control
system for spreading of liquid manure,” in Preprints 1999 IFAC Congress,
Beijing, Paper 4a-01-2, 1999.
[25] A. Munack, “High precision control system for spreading of liquid ma-
nure,” in Proc. ASAE/CSAE-SCGR Annual Inter. Meeting, Toronto, ASAE paper
no. 99 1104, 1999.
[26] H. Auernhammer and J. Frisch, Eds., “Landwirtschaftliches BUS-System
LBS. Mobile Agricultural BUS-System—LBS,” KTBL-Arbeitspapier 196,
Landwirtschaftsverlag, Münster, 1993.
[27] H. Speckmann and G. Jahns, “Development and application of an agricul-
tural BUS for data transfer,” Comput. Elect. Agric., vol. 23, pp. 219-237, 1999.
[28] SAE J1939: Recommended practice for truck and bus control and communi-
cations network. SAE, Warrendale, PA, 2000.
[29] Bosch, CAN Specification, Version 2.0, Robert Bosch, Stuttgart, 1991.
[30] H. Speckmann, “Providing measured position data for agricultural ma-
chinery,” Comput. Elect. Agric., vol. 25, pp. 87-106, 2000.
[31] C. Nienhaus, “Stand der internationalen Normungsarbeit. Progress Re-
port on International Standardization Procedures,” in Landwirtschaftliches
BUS-System LBS - Mobile Agricultural BUS System - LBS, KTBL-Arbeitspapier
196, pp. 182-189, Landwirtschaftsverlag, Münster, 1993.
Josse De Baerdemaeker graduated as an agricultural engi-
neer from the Katholieke Universiteit Leuven, Belgium. In
1975, he obtained an M.Sc. and a Ph.D. in agricultural engi-
neering from Michigan State University and later did post-
doctoral research at Cornell University and the University
of California, Davis. Currently, he is a Professor at the
Katholieke Universiteit Leuven. His teaching and research
areas focus on the interaction between physical processes
and biological products for the design and control of novel
technologies for the cultivation, harvest, handling and stor-
age of crops. He is the author or co-author of some 150 pa-
pers. He is active in international organizations related to
engineering and process control for biological systems and
served as President of the European Society of Agricultural
Engineers from 1996-1998.
Axel Munack received the Dr.-Ing. degree from the faculty of
electrical and mechanical engineering at the University of
Hannover, Germany, in 1980. From 1985 to 1988, he was Pro-
fessor for Simulation Techniques at the Technical Univer-
sity of Hamburg-Harburg, and since 1988, he has been
Director of the Institute of Technology and Biosystems Engi-
neering at the Federal Agricultural Research Centre (FAL),
Braunschweig, Germany. His areas of interest comprise ap-
plications of information technology in agricultural produc-
tion processes, use of plant oil as a substitute for diesel fuel,
and modeling and control of biotechnical processes. He is
author or co-author of more than 160 publications. In
1996-1997, he served as President of the FAL. He is Vice-Pres-
ident of the European Agricultural Engineering Society,
EurAgEng, and is Incoming President of the International
Commission of Agricultural Engineering, CIGR.
Herman Ramon graduated as an agricultural engineer from
Gent University. In 1993 he obtained a Ph.D. in applied biolog-
ical sciences at the Katholieke Universiteit Leuven. He is cur-
rently Professor at the Faculty of Agricultural and Applied
Biological Sciences of the Katholieke Universiteit Leuven, lec-
turing on agricultural machinery and mechatronic systems
for agricultural machinery. He has a strong research interest
in precision technologies and advanced mechatronic sys-
tems for processes involved in the production chain of food
and nonfood materials, from the field to the end user. He is au-
thor or co-author of more than 40 papers.
Hermann Speckmann received his Dipl.-Ing. degree from
the faculty of electrical and mechanical engineering at the
Technical University of Braunschweig, Germany, in 1972.
Since 1973, he has been a research engineer at the Federal Ag-
ricultural Research Centre (FAL) in Braunschweig. His work
deals essentially with automation and control of agricultural
machinery for both field and in-house operation, as well as
with data communication techniques for mobile machines
and tractor-implement combinations. During this work, he
has significantly contributed to the DIN 9684 standard.

Ieee 00954519

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (20)

Similar to Ieee 00954519

Similar to Ieee 00954519 (20)

Recently uploaded

Recently uploaded (20)

Ieee 00954519