Computers and Electronics in Agriculture 121 (2016) 412–419

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Original papers

Application of fibre sensor in grain drill to estimate seed flow under field
operational conditions
A.A. Al-Mallahi ⇑, T. Kataoka
Research Faculty of Agriculture, Hokkaido University, Kita-9, Nishi-9, Kita-ku, Sapporo, Hokkaido 060-8589, Japan

a r t i c l e i n f o a b s t r a c t

Article history: A new sensing system to estimate the mass flow of seeds in the grain drill was developed and evaluated
Received 30 July 2015 in this paper. This was a manipulation of an estimation model developed indoors using fibre sensors; by
Received in revised form 19 December 2015 considering the operational conditions of the machine and environment. Therefore, an original method to
Accepted 5 January 2016
install the sensor within the grain drill was tried, and modifications on the estimation model and algo-
rithm were performed.
The experiment to test and evaluate the sensing system consisted of running a grain drill-propelling
Keywords:
tractor in the farm of Hokkaido University several times to sow rye seeds, where each run was approx-
Mass flow
Sowing
imately 160 m. Two fibre sensors were installed at two different metering units, that use axial flute roll-
Fibre sensor ers, inside the grain drill. Also, the seeds being discharged during the experiment were collected and
Estimation rate weighed at the end of each run. The results of 52 trials to estimate the mass flow of seeds indicated that
Sensing system in approximately 98% of the trials the estimation rate exceeded 90%. Moreover, the overall estimation
rate was approximately 95%. These results seemed not to be biased by the change of the sensor set
nor the sensing location.
The modifications on the estimation model were successful in eliminating the overestimation or under-
estimation bias that resulted by changing the speed in the indoor experiments. In addition, the modifi-
cations on the estimation algorithm could eliminate error in the sensor output values that may result
from dust, vibration, or variation in the sensor internal resistance. These results showed that the sensing
system can be used practically to monitor the seed flow in the grain drill which would have many appli-
cations in precision agriculture practices.
Ó 2016 Elsevier B.V. All rights reserved.

1. Introduction The performance of the grain drill is evaluated according to the

The grain drill is a tractor propelled vehicle used primarily to
sow grain seeds. Sowing differs from seeding in that; while seeding
aims to place seeds one by one at a certain spatial interval, sowing
aims to cover the farm by a uniform distribution of seeds. There-
fore, unlike seeders, which have multiple small hoppers placing
seeds in relatively far distances, grain drills have one big hopper
with many outlet pipes close to each other so as to ensure the max-
imum possible coverage of the ground.
Power in the grain drill is required to rotate internal shafts. One
of these shafts carries sowing metering flute rollers which deliver
seeds to the outlet pipes. In order to optimize the application of
the grain drill, the width of the rollers may be adjusted manually.
This is one of the measures that allow using the drill to sow differ-
ent variety of seeds classified according to their sizes.
⇑ Corresponding author.
E-mail address: ahmad@bpe.agr.hokudai.ac.jp (A.A. Al-Mallahi).
uniformity in distributing seeds throughout the sowing operation.
Since all shafts are connected to the tyre of the grain drill, the
increase of the forward speed of the grain drill will cause a faster
rate of sowing by the rollers. This will ensure a theoretical unifor-
mity of sowing throughout the land regardless of the forward
speed. Nevertheless, Maleki et al. (2006) reported that seed unifor-
mity using fluted-rollers is still impaired due to sudden releases of
seed batches and suggested modifications on the design of the roll-
ers to improve uniformity.
Therefore, the recent years have witnessed some new
approaches to develop more controlled sowing metering systems.
One of these approaches is based on using pneumatic device to
improve the precision of the metering system in releasing grains,
as designed and tested by Yasir et al. (2012). Another approach is
based on a more radical modification on the conventional metering
system. It suggests replacing the chain mechanism that transmits
power from the rotation of the tyres to the flute roller shaft, by
an electronic system that synchronizes the rotational speeds of

http://dx.doi.org/10.1016/j.compag.2016.01.006
0168-1699/Ó 2016 Elsevier B.V. All rights reserved.
 A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419 413

Nomenclature

N natural number {1, 2, 3, . . .} d density of clump during one period of time

L length of clump during one period of time m mass of clumps during one period of time
V value of voltage output during the experiment bi coefficient of the i-th variable
a initial value of voltage output

the tyres and the shaft, as developed and tested by Zhai et al. its governing algorithm to the outdoor operational conditions
(2014), as well as Kamgar et al. (2015). The latter approach aims including, machine vibration, dust, and different tractor running
mainly to overcome the problem of slippage that occur while speeds. Accordingly, a grain drill was set up, firstly, to run an
sowing. experiment of sowing rye seeds by installing two sets of optical
Despite these advancements in trying to improve the perfor- fibre sensors. Next, the ability of the sensing system to estimate
mance of the grain drill, a full system that provides a feedback from the mass flow was assessed. This was achieved by comparing
the seeds about their flow is yet to be tackled. Currently, a rough actual mass of seeds collected during the experiment with the
evaluation of the flow may be drawn based on yield at the end of mass estimated after considering all the operational impacts on
the season; by analyzing the amount and distribution of the yield the estimation model and algorithm.
in comparison with past seasons. Nevertheless, there are several
factors that might affect the yield during the growing season,
2. Materials and methods
beside seed distribution, and decisions about sowing can hardly
be taken without further considerations, as shown by De Bruin
2.1. Grain drill
and Pedersen’s study on soybeans (2008), as well as Weisz and
Heiniger’s study on small grain seeds (2012).
The grain drill used to conduct the outdoor experiment was
Additionally, in the current sowing practices using grain drills,
Tume (KL, 2500). Fig. 1 shows a back view of the grain drill. It is
the operator must periodically monitor the amount of seeds in
designed to apply fertilizers simultaneously while sowing. There-
the hopper so as to avoid running out of seeds while the grain drill
fore, the hopper is divided, in fact, into a front and back hoppers
is operating. Since there does not exist any way to neither accu-
that contain fertilizers and seeds, respectively. The width of the
rately estimate the input of seeds to the hopper nor having infor-
drill is 2.5 m, through which the seeds and fertilizers are dis-
mation about the sowing operation itself, the operator will be
charged downwards using an axial flute roller system. There are
forced to fill the hopper more frequently than he or she should.
20 and 12 metering units in the seed and fertilizer hoppers,
This means that the efficiency of the operation would become
respectively.
much lower than it could be; if the operator were to have a feed-
Each metering unit consists of a shutter, the flute roller, and a
back about the sowing operation and the seed distribution.
spring loaded bottom valve as described by the sketch of Fig. 2,
In order to equip the grain drill with a feedback mechanism of
which is a simplification of the metering unit description in the
its operation, it is necessary to develop a sensing system which
datasheet of the drill manufacturer (Tume-agri, n.d). The shutter
can estimate the flow rate of grain seeds under the operational
may separate the metering unit from the hopper, while the inclina-
conditions. Traditionally, most of the research efforts to develop
tion of the bottom valve determines the amount of seeds that can
seed sensing systems were concentrated on seeding mechanisms
be pushed off by the flutes of the roller. The pushed seeds form
rather than sowing because of the high cost of bigger seeds, such
clumps that flow through a funnel towards outlet pipes that lead
as beans, and the vital effect of distance between seeds on the
these clumps towards the designated locations in the soil. The
yield. Some of these researches used opto-electronic sensors
frame, on which all outlet pipes are arranged, is hooked to the
(Kocher et al., 1998; Lan et al., 1999), high speed cameras
frame of the hopper by a trough; whereas the clearance between
(Karayel et al., 2006; Zhan et al., 2010), or infrared sensor equipped
the two frames is 12 mm.
with an embedded system (Okopnik and Falate, 2014). On the
The flow rate per unit area is assigned manually by a hand
other hand, trials to develop sensing system for small seeds in
wheel which can be rotated to adjust the width of the flute roller
the grain drill have been scarce until recently.
exposed to the bottom valve. However, the feeding rate is usually
Earlier, the authors developed a model to estimate the mass
calibrated for each seed variety only prior to the sowing process
flow of small seeds using a fibre sensor (Al-Mallahi and Kataoka,
using calibration curves provided by the manufacturer for each
2013). Similarly, Raheman and Kumar (2015) used an infrared sen-
seed variety; and should not be adjusted while operating. In Japa-
sor equipped with and embedded system to detect the seed flow.
nese farms, it is common to run the grain drill at a fixed speed
The difficulty in developing a sensing system of the flow of small
which does not drop below 5 km/h, but does not exceed 9 km/h.
seeds in a grain drill has been that the seeds flow in clumps. While
These figures are similar to the standard speeds recommended
the latter research work was able to detect the passage of the
by drill manufacturers such John Deere and Massey Fergueson
clumps at different flow rates, the former work was able to go a
(Campbell, 2002), who suggest a speed that should not exceed 6
further step of estimating the mass of each clump by manipulating
mph (approximately 9.6 km/h). Also they go in line with a recom-
the sensor data statistically and developing a mass estimation
mendation by the agricultural decision maker at Iowa State
model. However, all trials to develop such systems have been
University (Edwards, 2007) who recommends 5 mph (approxi-
developed based on indoor experiments. Therefore, there has been
mately 8 km/h) as the operational speed of the grain drill.
a need to test these systems outdoors, so as to adapt any sensing
system developed indoors to the outdoor operational conditions
where environmental effects may be uncontrolled. 2.2. Digital fibre sensor
Consequently, the objective of this research work was to adapt
the sensing system developed by the authors to outdoor opera- The sensor used to detect the flow of seeds was an off-the-shelf
tional conditions. This could be achieved by installing the sensors digital fibre sensor (Keyence, FS-N10). It consists of a light trans-
on the grain drill, and modifying the mass estimation model and mitter and receiver as well as an amplifier connected by fibre
414 A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419

Tractor
Laptop housing

Grain drill Trough

Plastic bag

Fig. 1. Grain drill propelled by tractor in one of the runs during the outdoor experiment, where the outlet pipes were replaced by plastic bags to collect actual mass flow.

Seed hopper

Shutter

Discharge unit

Trough
Loading spring

Flute roller
Clearance
Bottom valve
Outlet pipe frame

Outlet pipe
Funnel

Fig. 2. Simplified sketch of the seed discharge unit and the seed outlet arrangement underneath the hopper.

cables. The amplifier functions also as a light source, sent towards
the emitter using one fibre cable towards the receiver using
another cable. A beam of Light, of wavelength 620 nm, travels from
the emitter towards the receiver in the shape of a beam of
40
3 mm. When the beam of light is interrupted by obstacles,
their shade will fall on the surface of the receiver causing differ-
ence in intensity, which is translated into an analogous voltage
output by the amplifier.
2.3. Installation of sensor in grain drill
Two sets of the sensor were installed in the grain drill at the
outlet pipe frame, so as to monitor the mass flow at the 5th and
15th metering units counting from the right side of the grain drill.
Since each sensor consists of a light emitter and receiver, between
which, a beam of light should cover the area through which seeds
flow, it was necessary to ensure that the two parts of the sensor are
installed in parallel. Therefore, a Pi-shaped metal plate (P) was
designed as the frame on which the sensor elements were fastened
to ensure parallelism.
Since the sensor height was 5 mm, the frame was chosen to be a
3-mm thick so as to ensure its installation within the clearance
between the frames. In order to reduce the effects of dust that
Moreover, the 2-mm thickness of the Pi-shaped frame could serve
as a vertical clearance for the sensor. Since the light beam of the
sensor has the shape of a rectangle of 40
3 mm, the overall setup
was arranged so that all seeds discharged towards the outlet pipes
can be detected by the sensor; with the least possible effect of dust
during sowing. A sketch of the top and front views of the sensors
setup is displayed in Fig. 3.
2.4. Mass flow calculation
2.4.1. Automatic correction of sensor output
The mass flow estimation model was developed based on the
observation that grain seeds flow in clumps. Besides, it was
assumed that; the changes of the fibre sensor light intensity; and
the time period during which one clump passes in front of the sen-
sor, correspond to the thickness (d) and length (l) of the clump
respectively (Al-Mallahi and Kataoka, 2013). Accordingly, the fol-
lowing equations were created to calculate the mass,
m ¼ b1 d þ b2 l ð1Þ
where d and l are calculated using,
X
8n

might result by the flow of seeds while sowing, a horizontal clear- l¼ bk ð2Þ
k¼1
ance of 3 mm was left between the Pi-shaped frame and the funnel.
 A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419 415

Pi-shaped frame

Sensor emitter Sensor receiver

(a)

Vertical clearance
Funnel
Horizontal clearance

Plastic bag

(b)

Fig. 3. (a) Top view of the fibre sensor installing arrangement on the outlet pipe frame including the Pi-shaped sensor frame. (b) Front view of the arrangement with the
outlet pipe replaced by plastic bags for seed collection to measure actual mass.

while, assumption was verified by setting a short period of time within

 which the sensor will detect seeds. This period was chosen to be
1; v<a
b¼ ð3Þ 0.8 s because it was the minimum resolution of the mass digital
0; vPa scale used by the authors (Al-Mallahi and Kataoka, 2013). They
and, found that even in a time period as short as 0.8 s there were
moments that the sensor, whose detection frequency was 8 times
X
8n
faster than the scale, did not detect any clumps. Consequently, this
d ¼ 8an  vk ð4Þ was the period chosen (after trying other longer periods) to make
k¼1
the initial voltage output correction within. Also, all the seeds
The model was created and the indoor experiment was con- which were falling in this period of time were considered as one
ducted by fixing the value of the initial voltage output of the sensor clump when the assumptions of the model were designated. The
at 4.91 V, which was the sensor analogue output value when no initial output correction was made by continuously setting the ini-
clumps were crossing the sensor beam of light. However, it was tial output to 0 V within this time interval; as expressed in the
noticed during the experiment that the value of the initial voltage flowchart of Fig. 4, which shows the iteration and conditions to
might fluctuate below 4.91 mainly due to the internal resistance of determine the initial voltage output before the mass estimation
the sensor itself when it is on for extended time. While this draw- takes place.
back in the sensor performance was tolerable under the supervised
experiment, it needed to be addressed when using the sensor in
2.4.2. Considerations of flute roller size and speed
the field.
The results of testing the estimation model created by the
Moreover, the considerations taken while installing the sensor
authors (Al-Mallahi and Kataoka, 2013) at three different roller
in the grain drill to avoid dust could reduce but not eliminate the
rotational speeds, indicated that the model tends to overestimate
dust effects on the sensor performance. As dust was generated
and underestimate the mass of seeds at low and high speeds,
not only from the sowing process but also from the soil disturbed
respectively. Moreover, since the width of the roller to be exposed
by the interaction with the tyres of the machines, the initial voltage
to the seeds to perform the sowing rate of 100 kg/ha is different
output would gradually drop by dust accumulation on the light
from the width used in the indoor tests, the change of the geome-
emitter and receiver.
try of the roller according to the type of seed and the desired flow
In order to overcome this problem, a modification was made on
rate should be considered. Accordingly, the estimation model
the mass estimation algorithm so that the initial output could be
would be modified to consider different rollers and rotational
corrected automatically rather than setting it manually. The algo-
speeds as expressed in Eq. (5),
rithm, to continuously correct the initial voltage, was based on
the assumption during the indoor experiment that there should m ¼ b0 ðb1 d þ b2 lÞ ð5Þ
be a time lapse between any two consecutive clumps; and that
one clump takes fractions of a second to cross the sensor beam where b0 can be considered as the coefficient that corrects the mass
of light, even at the highest flow rate of seeds discharge. The flow estimation at different roller speeds. Therefore, one of the goals
416 A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419
Start
T = 0, Vi = 0
T = T + 0.1s
Measure V
No
V > Vi?
Yes
Vi = V Vi = Vi
Yes
T < 0.8s
No
Estimate mass
Fig. 4. Algorithm to auto-correct sensor data output and estimate mass of clumps
in period of 0.8 s.
of the outdoor experiment was to determine the values of b0 at the
high and low speeds.
2.5. Experimental setup and data collection
The outdoor experiment, conducted in the experimental farm of
Hokkaido University, consisted of running the grain drill back and
forth for a designated distance so as to demonstrate sowing rye
seeds. The length of the row was approximately 160 m and the tar-
get flow rate per hectare unit was 100 kg, which is the recom-
mended rate by farmers in Hokkaido. Therefore, In order to
determine the required width of the roller, a manual calibration
test was conducted using rye seeds prior to sowing. The procedure,
which included raising the grain drill, turning the tyre 15 times by
hand, and collecting the seeds in troughs before being measured,
indicated that two-thirds of the full width of roller should be
exposed to the seeds to achieve the desired rate.
The installed sensors were connected to a laptop to log the raw
data obtained by the sensors and to estimate the mass that flows in
each run using Eq. (1). On the other hand, in order to measure the
actual mass in each run, the outlet pipes were replaced with plastic
bags to collect the seeds. This arrangement can be seen in the pic-
ture of Fig. 1, which was taken during the experiment. The red
housing at the right side of the grain drill contained the laptop
used to collect and process data during the experiment.
The details of the experiment including the operational forward
speed and actual mass sown in each run are illustrated in Table 1.
Since the data were collected at two different outlets in each run,
the total number of mass estimation trials using the sensors was
52. In addition to these 26 runs, there were three additional runs
at each speed, during which the shutters of the outlet chambers
were locked to disallow flow of seeds. The purpose of these 9 runs
was to examine the effect of vibration due to the motion of the
tractor and the drill on the sensor data. Finally, in order to correct
the underestimation and overestimation of the mass at different
speeds, the data at each speed were divided into two sets so that
one set was used as a training data to calculate the speed coeffi-
cient (b0) and the other set of data was considered as the testing
data to validate the results.
3. Results and discussions
3.1. Evaluation of sensor installation and algorithm functionality
In order to evaluate the sensor installation and the automatic
correction of the sensor output, the data collected from the sensors
at the 9 runs without sowing were observed. Although drops in the
initial voltage of the sensor output did happen, they were gradual
and tended to fluctuate which indicates that they were caused by
the changes of the internal resistance of the amplifier of the sensor.
In addition, the maximum drop which was observed during all the
runs did not exceed 2% of the initial voltage at that run. These
observations indicate that it was unlikely that the drops were
caused by external factor such as dust, because drops in the initial
voltage caused by blocking the beam of light would tend to be sud-
den, big, and continuous. There was no indication of any significant
effect of vibration on the results. Prior to the outdoor experiment,
it was observed that there were fluctuations in the sensor output
while moving the fibre cables of the sensor which was, neverthe-
less, solved by fastening the cables on the outlet pipe frame.
Applying the estimation model with the automatic estimation
algorithm, the mass flow was estimated as zero throughout the 9
runs, which meant that the drops in the initial voltage were suc-
cessfully overlooked by the sensing system. Therefore, it could be
concluded that the sensor installation and the estimation algo-
rithm were suitable to overcome the factors that may affect the
estimation model while working in the outdoor environmental
conditions.
3.2. Optimization of mass estimation model
As a first step towards optimizing the estimation model, Eq. (1)
was applied on the sensor data; which resulted in the values of the
estimated mass shown in Table 1. Plotting these values on the
graphs of Fig. 5, showed explicitly an overestimation of the mass
values and that this overestimation tends to increase proportion-
ally with the decrease of the forward speed. Overestimations and
underestimations are caused by the assumption on which the
model was constructed. That is, the mass of one clump is deter-
mined by its intensity and length where the former is estimated
from the drops in the fibre sensor voltage and the latter is esti-
mated from the number of times the drops occur.
As discussed by the authors (Al-Mallahi and Kataoka, 2013),
they observed that clumps caused by the rotational motion of
the flute roller may be light or heavy, and that light clumps
appeared more frequently at the low speeds. They found that the
individual light clumps are overestimated which result in a general
overestimation at the low speeds. On the other hand, there is a ten-
dency to underestimate the heavy clumps because they consist of a
big number of seeds passing instantaneously in front of the sensor.
Since the sensor is able to detect the seeds from one side only,
some seeds become occluded, causing the underestimation.
Consequently, the roller speed coefficient (b0) was introduced
as one of the possible solutions to improve the estimations. It
was calculated by, simply, dividing the summation of the esti-
mated mass of the 1st, 4th, and 7th runs of both sensors at each
speed, on the summation of the actual mass of the same runs.
Hence, one-third of the data were used to calculate the coefficients
which are shown in Table 2. Next, Eq. (5) was applied again on the
Table 1
Actual, estimated, and corrected masses by two sensors at 3 different speeds.

Operation speed 5 km/h

A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419
Trial number (–) 1 2 3 4 5 6 7 8
Sensor A B A B A B A B A B A B A B A B
Actual mass (g) 250.9 242.1 250.8 243.3 251.8 248.2 250.2 247.5 276.0 256.8 278.4 263.6 277.2 253.0 280.8 258.2 – –
Estimated mass (g) 303.3 268.5 303.4 309.7 328.3 321.6 295.4 299.7 351.9 295.5 344.1 310.7 321.3 323.1 333.5 270.1 – –
Corrected mass (g) 254.7 225.4 254.8 260.1 275.7 270.0 248.0 251.7 295.5 248.1 289.0 260.9 269.8 271.3 280.0 226.8 – –
Estimation rate (%) 101.5 93.1 101.6 106.9 109.5 108.8 99.1 101.7 107.1 96.6 103.8 99.0 97.3 107.2 99.7 87.8 – –
Error (%) 1.5 6.9 1.6 6.9 9.5 8.8 0.9 1.7 7.1 3.4 3.8 1.0 2.7 7.2 0.3 12.2 – –
Operation speed 7 km/h
Trial Number (–) 9 10 11 12 13 14 15 16 17
Sensor A B A B A B A B A B A B A B A B A B
Actual mass (g) 266.0 258.0 257.0 250.6 264.8 264.8 258.0 258.6 258.0 258.0 91.4 252.2 192.7 183.2 256.4 242.9 251.6 247.6
Estimated mass (g) 311.7 309.3 271.6 264.2 305.3 298.7 307.9 307.0 291.4 290.7 97.3 296.1 207.3 212.6 305.8 309.8 290.2 269.1
Corrected mass (g) 266.6 264.6 232.3 226.0 261.2 255.5 263.4 262.7 249.3 248.7 83.3 253.3 177.3 181.9 261.6 265.0 248.2 230.2
Estimation rate (%) 100.2 102.6 90.4 90.2 98.6 96.5 102.1 101.6 96.6 96.4 91.1 100.4 92.0 99.3 102.0 109.1 98.7 93.0
Error (%) 0.2 2.6 9.6 9.8 1.4 3.5 2.1 1.6 3.4 3.6 8.9 0.4 8.0 0.7 2.0 9.1 1.3 7.0
Operation speed 9 km/h
Trial number (–) 18 19 20 21 22 23 24 25 26
Sensor A B A B A B A B A B A B A B A B A B
Actual mass (g) 247.4 239.6 261.6 249.0 248.8 249.0 254.8 244.6 134.9 251.2 251.7 251.0 262.0 238.0 278.2 246.6 270.5 245.2
Estimated mass (g) 276.2 270.5 300.2 282.7 253.7 270.1 259.3 269.4 147.0 276.0 252.9 245.5 252.9 245.5 280.3 283.8 260.6 258.3
Corrected mass (g) 260.8 255.4 283.4 266.9 239.5 255.0 244.9 254.4 138.8 260.6 238.8 231.8 238.8 231.8 264.6 268.0 246.1 243.9
Estimation rate (%) 105.4 106.6 108.3 107.2 96.3 102.4 96.1 104.0 102.9 103.7 94.9 92.4 91.2 97.4 95.1 108.7 91.0 99.5
Error (%) 5.4 6.6 8.3 7.2 3.7 2.4 3.9 4.0 2.9 3.7 5.1 7.6 8.8 2.6 4.9 8.7 9.0 0.5

417
418 A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419

400.0
350.0
$FWXDOPDVV
300.0

Mass (g)
250.0
(VWLPDWHGPDVV
200.0
150.0 &RUUHFWHGPDVV
100.0
50.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Run number
(a) Sensor A

400.0
350.0
$FWXDOPDVV
300.0
Mass (g)

250.0
(VWLPDWHGPDVV
200.0
150.0 &RUUHFWHGPDVV
100.0
50.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Run number
(b) Sensor B

Fig. 5. Actual, estimated, and corrected masses for the 26 trials at each discharge unit by the two different sets of fibre sensor.

Table 2
Values of speed coefficient (b0) and results of statistical validation tests at each speed.

Grain Speed Actual Corrected Actual Corrected F-test (-) F-critical (–) p-value at t-statistic (–) t-critical (–) p-value
drill coefficient mass Mass variance (–) variance (–) F-test (–) at t-test (–)
speed (b0) (–) average (g) average (g)
(km/h)
5 1.191 258.05 261.35 170.88 383.45 2.244 2.403 0.064 0.560 2.056 0.580
7 1.169 239.54 235.06 1888.81 2137.73 1.132 2.272 0.400 0.300 2.032 0.766
9 1.059 245.78 245.77 869.21 905.54 1.042 2.272 0.467 0.001 2.032 0.999
All speeds – – 247.40 246.85 1027.88 1242.65 1.209 1.592 0.250 0.082 1.984 0.935

sensor raw data to obtain the corrected mass values which are number of samples when the data of all speeds were treated as
shown in Table 1, and plotted in Fig. 5. These results indicated that one data set. At first, F-tests were applied to verify the equality
the corrected masses were hovering around the actual masses. It is of the variances between the data sets. The results illustrated in
worth to indicate that in trials 13a, 14a, and 15b, the actual mass Table 2, showed that the F-test values were smaller than F-
flow was significantly less than all the other trials because the hop- critical at each speed. Consequently, the t-test, assuming equal
per was left to become empty while sowing. variances, was performed on the data sets of each speed as well
In order to validate the robustness of the model, a t-test was as the data sets at all speeds, obtained by combining the actual
performed on the data sets of the actual and corrected masses of and corrected values measured and calculated respectively at each
each speed independently, as well as on the data sets of the total speeds. Based on the t-test results the null hypothesis of assuming
equal means of the actual and corrected data sets (i.e. assuming
that the corrected values are biased by the estimation model)
300.0 was not rejected as the t-statistic values were smaller than t-
critical for all the data sets. These results were supported by the
p-values which were much bigger than 0.05.
280.0 In order to understand the relationship between the actual and
Corrected Mass (g)

corrected masses by both sensors and at all speeds, Fig. 6 was gen-
erated on which the actual mass was plotted against the corrected
mass in 49 trials out of the total of 52 trials. Only the 3 trials, men-
260.0 At 5 km/h
tioned in the previous paragraph were left out so as to visually
At 7 km/h make the graph more relevant. The correlation coefficient was
At 9 km/h 0.635, 0.968, and 0.868 at 5, 7, and 9 km/h respectively. It was
240.0
observed that only in 1 trial out of the 52, the estimation rate
was less than 90%. Also, the average corrected mass estimation
was 95.27%, which did not change significantly by the sensor
220.0 (95.53% at Sensor A and 95.01% at Sensor B). Rather than observing
220.0 240.0 260.0 280.0 300.0
overestimations and underestimations at low and high speeds
Actual Mass (g) respectively, the distribution of the estimation error was similar
in magnitude regardless of the speed. This indicated that the esti-
Fig. 6. Actual vs corrected masses of 49 trials at the 3 different tractor speeds.
 A.A. Al-Mallahi, T. Kataoka / Computers and Electronics in Agriculture 121 (2016) 412–419
mation model could be optimized to any speed within the drill
operating recommended speeds by calculating the roller speed
coefficient at that particular speed.
3.3. Evaluation of mass flow estimation system
Despite applying the sensor and the estimation model to uncon-
trolled environmental conditions such as dust and vibration, the
results seemed to be more robust compared to the ones indoor.
Firstly, two sensors were used to do the estimation simultane-
ously; and their corrected mass estimation values were very simi-
lar. This indicates that the installation was well designed. Secondly,
the number of data sets in the outdoor experiment was big,
whereas the deviation of the results was small when compared
to the indoor experiment.
Because of the compactness of the fibre sensor and the simplic-
ity of the frame, it was possible to install the sensor at the very nar-
row location on the top of the outlet frame. As the sensors wiring
was also fastened on the side of the same frame, the sensors could
be installed without any modification on the grain drill frame or
body, which will reduce the costs to the minimum if decisions will
be made to adapt the mass flow sensing system to grain drills by
manufacturers in the future.
4. Conclusion
The main achievement in this research work was the success in
applying the indoor seed mass flow estimation methodology on
the grain drill during actual sowing. The compactness of the fibre
sensor and the design of the Pi-shaped sensor frame ensured the
possibility of installing the sensor in a space as narrow as 12 mm
and the parallelism of the sensor components, which is a crucial
factor in obtaining accurate estimation of the flow.
Modifications were made on the estimation algorithm and the
estimation model to cope with the factors which might affect the
sensing accuracy outdoors. The modification on the algorithm
was successful in eliminating dust, vibration, as well as sensor
internal resistance; through introducing an automatic corrected
initial sensor output. Moreover, while the model, indoors, tended
to overestimate and underestimate the flow of low and high speeds
respectively, this tendency was overcome in the outdoor applica-
tion; providing better estimation results. This could be achieved
by taking the shape and the speed of the seed sowing axial flute
roller into consideration, within the estimation model in the shape
of a roller speed coefficient.
In 98% of the trials, the estimation rate was more than 90%,
while the average estimation rate was approximately 95%. The
big number of trials (52 runs by the grain drill) as well as getting
approximately the same results from two different sensors at
two different metering units, enhanced the robustness of the
results. Since the modified estimation method could not only elim-
inate the negative possible impacts working resulting from operat-
419
ing outdoors, but also avoid overestimations and
underestimations, the results were considered satisfactory for the
objectives of this research work. In addition, the simplicity in
installing the system without any need to modify the body of the
grain drill means that this estimation methodology is ready to be
adopted so as to develop a complete seed monitoring and estimat-
ing system in the grain drill for applications in precision
agriculture.
Acknowledgement
A part of this research was supported by the commissioned pro-
ject of the Ministry of Agriculture, Forestry, and Fisheries entitled
by ‘‘Development of automatic and assistant agricultural system”.
References
Al-Mallahi, A.A., Kataoka, T., 2013. Estimation of mass flow of seeds using fibre
sensor and multiple linear regression modelling. Comput. Electron. Agric. 99,
116–122.
Campbell, D. 2002. Grain Drills <http://www1.agric.gov.ab.ca/$department/
deptdocs.nsf/all/eng3079#42> (Retrieved December 14, 2015).
De Bruin, J.L., Pedersen, P., 2008. Effect of row spacing and seeding rate on soybean
yield. Agron. J. 100 (3), 704–710.
Edwards, W., 2007. Ag Decision Maker: An Agricultural Economics and Business
Web Site <http://www.extension.iastate.edu/agdm/crops/pdf/a3-26.pdf>
(Retrieved December 14, 2015).
Kamgar, S., Noei-Khodabadi, F., Shafaei, S.M., 2015. Design, development and field
assessment of a controlled seed metering unit to be used in grain drills for
direct seeding of wheat. Inform. Process. Agric. 2 (3–4), 169–176.
Karayel, D., Wiesehoff, M., Zmerzi, A.O., Muller, J., 2006. Laboratory measurement of
seed drill seed spacing and velocity of fall of seeds using high-speed camera
system. Comput. Electron. Agric. 50, 89–96.
Kocher, M.F., Lan, Y., Chen, C., Smith, J.A., 1998. Opto-electronic sensor system for
rapid evaluation of planter seed spacing uniformity. Trans. Am. Soc. Agric. Eng.
41 (1), 237–245.
Lan, Y., Kocher, M.F., Smith, A., 1999. Opto-electronic sensor system for laboratory
measurement of planter seed spacing with small seeds. J. Agric. Eng. Res. 72,
119–127.
Maleki, M.R., Jafari, J.F., Raufat, M.H., Mouazen, A.M., De Baerdemaeker, J., 2006.
Evaluation of seed distribution uniformity of a multi-flight auger as a grain drill
metering device. Biosyst. Eng. 94 (4), 535–543.
Okopnik, D.L., Falate, R., 2014. Usage of the DFRobot RB-DFR-49 infrared sensor to
detect maize seed passage on a conveyor belt. Comput. Electron. Agric. 102,
106–111.
Raheman, H., Kumar, R., 2015. An embedded system for detecting seed flow in the
delivery tube of a seed drill. In: Proceeding of International Conference on
Advances in Chemical, Biological and Environmental Engineering, Singapore,
pp. 236–241.
Tume-Agri Ltd. n.d. KL 2500: User manual (In Finnish). P.15 <http://www.tumeagri.
fi/site/wp-content/uploads/HKL-2500-3000-4000-1995-2000-k%C3%A4ytt%
C3%B6ohje.pdf> (Last accessed: July 28, 2015.
Weisz, R., Heiniger, R., 2012. Small grain seeding rates in North Carolina. Small grain
production guide 2011–12. In: Weisz, R., (Ed). North Carolina Cooperative
Extension Service, North Carolina, pp. 21–26.
Yasir, S.H., Liao, Q.X., Yu, J.J., He, D.L., 2012. Design and test of a pneumatic precision
metering device for wheat. Agric. Eng Int: CIGR J. 14 (1), 16–25.
Zhai, J.B., Xia, J.F., Zhou, Y., Zhang, S., 2014. Design and experimental study of the
control system for precision seed-metering device. Int. J. Agric. Biol. Eng. 7 (3),
13–18.
Zhan, Z., Li, Y., Chen, J., Xu, L., 2010. Numerical analysis and laboratory testing of
seed spacing uniformity performance for vacuum cylinder precision seeder.
Biosyst. Eng. 106 (4), 344–351.