SICE Annual Conference 2007

Sept. 17-20, 2007, Kagawa University, Japan

Frequency Analysis Method using the Adaptive Algorithm and Application to Dynamic
Measurement of Mass and Weight
Yuuki SASAMOTO1 and Toshitaka UMEMOTO1
and Motoyuki Adachi2 and Yoichiro Kagawa2
1
Osaka Prefectural College of Technology, 26-12 Saiwai-cho Neyagawa,Osaka,Japan,572-8572
(Tel : +81-72-821-6401; E-mail: f06007@ipc.osaka-pct.ac.jp )
2
Yamato Scale Co., Ltd.,5-22 Saenba-cho, Akashi, Hyogo, Japan 673-8688

Abstract: With increased automation in the manufacturing industry, the demand for a quick and accurate mass measurement
system is growing. Currently, an active belt conveyer is in demand by many enterprises. However, there is a problem that the act
ive the active belt conveyer causes mechanical noises. The main noises are three kinds, which are the natural oscillation o
f measurement conveyer, oscillation by rotation of motor and belt pulley. They cause observable signals within the active belt
environment. Therefore, we have to remove those noises to acquire a greater accuracy. In the current mass measurement system, the
moving method is objected to remove those noises. Unfortunately, it is inadequate only in the current system method for more quick
and accurate measurement.
In this paper, we propose a frequency analysis method using the LMS algorithm. Moreover, we develop the system which appli
ed this method and moving average method. Finally, we use device actually marketed in order to verify the utility of the
system by comparative experiment between current system and proposed system.

Keywords: Dynamic Mass Measurement, Spectra analysis, LMS Algorithm

1. INTRODUCTION owever, increasing number of users now demands t

In order to measure the mass with accuracy, the m
easurement needs long time and has to be done un
der the condition of completely free from any
noises. However, it is difficult to make such ideal
surrounding s in most c a s e s o f industrial mass
measurement s , s u c h a s the automatic system for
weighing and sorting objects[1] [2] [3] [4]. The system is
generally called t h e Checkweighter , w h i c h i s
developed for examining and sorting the weight of all
produces on the production line. Checkweighter has to
carry products sent from the previous conveyer to the
next conveyer and also t o exclude some defective
products, lightweight or overweight, from the
production line. Both tasks must be done at the same
time. In other words, it is necessary to analyze not
only the measurement method but also the conveyance
method in order to develop the Checkweighers. The
required conditions that the conveyance method must
archive are three of the following.
1) A stable and reliable weighing result
2) A high conveying speed
3) A simple and flexible operation mechanism
Currently, as the conveyance method which meets abo
ve three requirements, the active belt conveyer is widel
y adopted in industr y. However, since the mass
detector of active belt conveyer is composed of the
built-in drive, it is impossible to avoid mechanical
noises caused by motion of the conveyer. These noises
must be removed in order to make the measurement
accurate enough to meet the needs of users. In the
current system, these noises are removed by the
simple moving average method. This method can rem
ove the noises completely if interval cycle of the
noises synchronizes with the moving average time. H
he machine which can quickly measure more variet
y of shapes of objects. In order to meet these dem
ands, the machine needs to measure the bigger
objects with higher conveying speed. However, it
causes reduction of time that objects are on
measurement conveyer. Consequently, the machine is
unable to obtain enough measurement time to matc
hes cycle of the noises with the moving average
time. As a result, all noises are not removed. It
means even if the specification of the system
makes remarkable progress, the accuracy of measuri
ng the object is not improved because of the nature
of simple moving average method.
Therefore, in this paper, we would like to propose
adaptive notch filter after we generalize our resent
research of the frequency analysis method of serial
processing type by using LMS algorism[5] [6] [7]. Mor
eover, we also propose the system that combine ad
aptive notch filter with the simple moving average
method. Finally, we use device actually marketed in
order to verify the utility of the system by comparati
ve experiment between current system and proposed
system. Samples used in comparative experiment ar
e either in standard specification or outside standard
specification.
2. PROBLEM SETTING
2.1 Outline of measurement system
The Checkweighter stated in this paper is the mass
measurement system of CSH22L type made in Yamato
Scale Co., Ltd, which is shown in Fig.1. Also, its model
is shown in Fig.2, and its specification is shown in
Table 1. As Fig.2 shows, the equipment is composed of
three conveyers; two conveyers for carrying and the

- 37 -
PR0001/07/0000-0037 ¥400 © 2007 SICE
Authorized licensed use limited to: Chulalongkorn University. Downloaded on September 9, 2009 at 22:54 from IEEE Xplore. Restrictions apply.
 Conveyer for Measurement

Feed-In Conveyer Feed-Out Conveyer

Belt Pulley

＋ ＋ ＋

＋
＋ ＋

Load Cell Motor

Fig.1 Checkwighter
Fig.2 Model of Equipment
Table 1 Specification of Equipment
Name Spec Table 2 Size of Samples
Number of Weight
Scope of Measurement 20～2200 [g] Size [mm]
Sample [g]
Max Speed of Conveyer 220 [piece/min]
A L300×W100×H15 160.8
Sampling Period 2 [msec]
B L400×W250×H15 1003.0
Length of Conveyer for Measurement 435 [mm]

shown in Table 2. The condition of experiment is

other conveyer for measurement, which is called the
measurement conveyer. In the measurement conveyer,
Load Cell is used as mass measurement sensor. The
objects are carried from the other two conveyers to the
measurement conveyer, and then Load Cell is distorted
under the weight of the object. Its strain value is
outputted as voltage value. The mass of object is
calculated from its voltage value.
2.2 Noisy elements caused in the equipment
As shown in chapter 1, the problem is that the active
belt conveyer is caused by mechanical noises. The main
noises are three kinds, which are the natural oscillation
of measurement conveyer, oscillations by rotation of
motor, and belt pulley. Two noises, oscillations by
rotation of motor and belt pulley, are calculated from
conveyer speed. If the conveyer speed is set at vc , the
rotation of belt pulley is shown to the following.
vc
fp =
πd p
Where, d p is the diameter of belt pulley. Also, the
rotation of motor, f m , is calculated by the proportion
of the number of motor’s teeth, M m , to the number of
pulley’s teeth, M p . Its equation is shown to the
following.
Mm
fm = fp
Mp
Therefore, the elements of frequencies are determin
ed by using these equations when users set conveye
r speed.
Then, the natural oscillation cause by a shape phase
determines how frequent the noise occurs. In this pa
per, we use the equipment which noise frequency i
s 28.75 Hz.
In this paper, we use two kinds of samples, either
over or under the standard specification size. It is
shown in Table 3. Noisy elements calculated by eq
uation (1) and equation (2) are shown in Table 4.
Table 4 shows that the noisy elements are lower fr
equencies.
2.3 Measurement time
The time when objects are carried on the measure
ment conveyer is called Measurement time. The ma
ss of object have to be measured while its time. M
easurement time, τ , is shown to the following.
L −L (3)
τ= 0
vc
Where, L0 is the length of measurement conveyer,
and L is the length of sample. From this equation,
measurement time is concerned with the conveyer s
peed and the length of sample. Therefore, increasin
g the conveyer speed or the length of sample cause
s the reduction of measurement time. However, sinc
(1) e the noisy elements are low frequencies, in the cu
rrent method, simple moving average method, all n
oises are not removed as showed in the preceding
chapter.
3. ADAPTIVE NOTCH FILTER
3.1 Frequency analysis method
In this paper, we propose a new adaptive notch filter,
(2) using the extended adaptive discrete Fourier transform
[7]
. First, the frequency analysis system is shown in
Fig.3. In this figure, G (k ) and X (k ) show Fourier
coefficient and input signal respectively in the time kT .
Here, T is sampling period, and is abbreviated as from
next. If the analysis frequency is set F , the input signal
is the complex exponential function, as follow:
X (k ) = exp( j 2πFkT ) (4)
The error signal ε (k ) required for adaptation is

- 38 -
Authorized licensed use limited to: Chulalongkorn University. Downloaded on September 9, 2009 at 22:54 from IEEE Xplore. Restrictions apply.
 Table 3 Condition for Experiment Table 4 Analysis Frequency for Experiment
Weighing Conveyer Measureme Analysis Frequency [Hz]
Number of Number Number of
Speed Speed nt Time Natural
Experiment of Sample Experiment Motor Belt Pulley
[piece/min] [m/min] [sec] Oscillation
1 80 35 0.24 1 13.75 6.88
A
2 120 52 0.14 2 20.43 10.22
3 8.25 4.13 28.75
3 49 21 0.1
4 10.61 5.31
4 B 62 27 0.08
5 12.58 6.29
5 74 32 0.06

Fig.4 Block Diagram for the Noise rejection

Fig.3 Spectra Analysis System with the LMS Algorithm

defined as the difference between the desired response

d (k ) and the output y (k ) , as follows:
ε (k ) = d ( k ) − y( k )
= d ( k ) − {G (k ) X ( k ) + G ( k ) X (k )} (5)
Where, G (k ) and X (k ) are conjugate complex
number of G (k ) and X (k ) respectively. In this paper,
we use the LMS algorithm[8] proposed by B.Widrow. Fig.5 Frequency characteristics of adaptive notch filter
The LMS algorithm minimizes the mean-square error
ε 2 (k ) by recursively altering the Fourier coefficient
G (k ) at each sampling instant according to the G (k + 1) X (k + 1)
expression = G(k + 1) X ( k + 1) + 4µ i ε (k ) X ( k ) X (k + 1)
G (k + 1) = G (k ) + 4µε (k ) X ( k ) (6) = G (k ) X (k ) X (1) + 4 µε (k ) X (1) (8)
Where, µ is a convergence factor controlling Where, we can define w(k ) as w(k ) = G (k ) X (k ) .
stability and adaptation. Equation (8) is shown to the following.
w(k + 1) = w(k ) X (1) + 4 µε (k ) X (1) (9)
3.2 Properties of adaptive notch filter
The transfer function of this equation is shown to the
Our research purpose is to derive an adaptive algorith following.
m to remove noise. Then, we think as a system in Fig.4. W ( z ) 4 µX (1)
Inthis figure, we think about the desired response d (k ) H R (z) = = (10)
E ( z ) z − X (1)
and the error signal ε (k ) as input signal and output sig
To similar, the transfer function of H J (z ) is shown
nal respectively. The transfer function of Fig.4 is shown
to the following. to the following.
E ( z) 1 (7) W ( z ) 4 µ X (1) (11)
H (z) = = H R (z) = =
D( z) 1 + H R ( z ) + H j ( z ) E ( z ) z − X (1)
When we use that input signal X (k ) is a complex Substituting equation and equation(11) into equation
exponential function, equation (6) is shown to the foll (7), we can obtain the following the transfer function
owing. H (z ) .

- 39 -
Authorized licensed use limited to: Chulalongkorn University. Downloaded on September 9, 2009 at 22:54 from IEEE Xplore. Restrictions apply.
 Table 5 Step Size Parameter for Experiment Table 6 Result of Comparative Experiment
Number of Step Size Parameter 3 standard deviation [g] progress
Experiment Number of
µ1 µ2 µ3 rate
Experiment Current Proposed
System System [%]
1 0.05 0.009 0.009
2 0.01 0.002 0.009 1 0.77 0.67 12.6
3 0.1 0.008 0.002 2 1.06 0.86 19.6
4 0.1 0.001 0.008 3 5.01 2.82 43.7
5 0.1 0.001 0.008 4 10.79 4.15 61.5
5 25.26 7.27 71.2
350000
342500
340000
Current System
Proposed System
Count Value

330000 342000

320000

Count Value
341500
310000

300000 341000

290000
340500
280000
340000
270000
0.0 0.5 1.0 1.5 2.0

Time [sec] 339500

0 5 10 15 20 25 30

Fig.6 Example of signal obtained from the system
z 2 − ( X (1) + X (1)) z + 1
H ( z) =
z 2 − (1 − 4µ )( X (1) + X (1)) z + 1 − 8µ
The frequency characteristics of adaptive notch filt
er for removing the natural oscillation is shown inF
ig.4. In this graph, µ is 0.09, and T is 2msec. Fr
om this property, the Gain near the analysis freque
ncy is greatly reduced. Hence, in the system propos
ed, only the specific frequency elements are remove
d.
4. DYNAMIC MASS MEASUREMENT EXP
ERIMENT
4.1 Outline of experiment
In this paper, we propose the system composed of
adaptive notch filter and simple moving average method.
Moreover we verify the utility of the system by
comparative experiment between current system and
proposed system. In the experiment, samples are either
in standard specification or outside standard
specification. As showed in chapter 2, the conditions of
samples and conveyer speed are shown in Table2 and
Table3.
4.2 Experiment with the current system
First of all, we examined the current system, whi
ch is using simple average method. In Checkweight
er used with this paper, a size of filter automaticall
y obtained from sample size and conveyer speed. T
hen, the simple moving average is applied to signal
s obtained from the road cell. In this experiment,
Number of Data
Fig.7 Example of result
(12)
these processing results are taken into a computer.
They are called count value C . An example of co
unt value is shown in Fig.6. The mass of objects i
s calculated from the following equation.
M
M = (C − C I ) S (13)
CS
Where, C I is standard count value. M S is standa
rd weight. C S is count value at standard weight.
4.3 Experiment with the proposed system
Next, the proposed system, which is using adaptiv
e notch filter and simple moving average method, i
s examined by using the signal obtained at the pre
vious experiment. In this experiment, in order to re
move three kinds of noises explained in chapter 2,
three kinds of adaptive notch filter for each noise a
re connected in series. The noisy frequencies shown
in Table 4 are used as analysis frequencies. As Ste
p Size Parameter, we use the value which gives th
e best accuracy. Its value is obtained by repeating
experiment at the same condition, and is shown in
Table 5. In this table, µ 1 is the value for natural o
scillation. µ 2 is the value for oscillation of motor.
µ 3 is the value for oscillation of belt pulley.
In Table 3, the minimum measurement time obtai
n at the previous experiment is 0.06sec. Because th
e sampling time is 2msec, the number of data is 3
0 points. So, we set to process only 30 points by t

- 40 -
Authorized licensed use limited to: Chulalongkorn University. Downloaded on September 9, 2009 at 22:54 from IEEE Xplore. Restrictions apply.
he adaptive notch filter. function and frequency characteristics. Next, we pro
posed the system, which applied adaptive notch filt
4.4 Result of experiment er and simple moving average method, and we test
The accuracy is compared between the current syst ed to verify the utility of the system by comparative
em and proposed system by repeating the experime experiment between current system and proposed sy
nt 60 times under the same condition. In the equip stem with using device actually marketed. In the ex
ment, the timing to output a signal is determined fr periment, two kinds of samples under and over the
om length of sample and speed of conveyer. Its ti standard specification size were used, and five kind
ming is called measurement timing point. In the ex s of conditions were set. As a result, we found tha
periment, mass of object is calculated from the sign t the proposed system can decrease the measuremen
al at measurement timing point, and the accuracy is t error to one third of the existing system makes. I
evaluated by three times of standard deviation. The nother words, the proposed system have the great p
result of experiment is shown in Table 6, and an e ossibility to improve the current specification.
xample of results is shown in Fig.7. As a result, w Therefore, the system proposed in this paper is
hen enough accuracy is obtained by using only the effective for Checkweighter to measure an object more
simple moving average method, the accuracy of pro accurate and is expected to make progress in its
posed system is not improved very well. However, performance.
when inadequate accuracy is obtained by using only
the simple moving average method, the accuracy of REFERENCES
proposed system is greatly improved. The progress
rate is 70 percent at the maximum. From Table 3, [1] Ono, T., Basic points of dynamic mass measurement,
the minimum measurement time obtain from these e Proceedings of SICE Annual Conference 1999,
xperiments is 0.06sec. This time corresponds to 30 (1999.8) pp.43-44
points as a number of data. From this reason, we s [2] Ono, T., Dynamic Mass Measurement, Journal of
et to process only 30 points by the adaptive notch Japan Society for Design Engineering, Vol.38, No.10
filter. In a word, the proposed system has the great (2003.10) pp. 489-495
possibility to improve the current specification even [3] Kurosu, S. et al., Continuous Mass Measurement on
in case of less measurement time. Conveyor Belt, Transactions of the Institute of
Electrical Engineers of Japan C,Vol.126,No.2,
(2006.2) pp.264-269,20060201
5. CONCLUSION [4] Kurosu, S. et al., Improved Continuous Weighing by
At present, Checkweighter is widely used in the Multi-stage Conveyor Belt Scale, Transactions of the
field of industrial measurement. However, when we Society of Instrument and Control Engineers,
Vol.40-12 (2004) pp. 1205-1210
use Checkweighter, it is impossible to avoid mecha
[5] Umemoto T. and Aoshima H., The Adaptive Spectrum
nical noises because of its structure. To remove the
Analysis for Transcription, Transactions of the
se noises, simple moving average method can be a Society of Instrument and Control Engineers, Vol.28
pplied. However, in order to remove noises, simple No.5 (1992.5)pp.619-625
moving average method needs to gather many num [6] Umemoto T. and Aoshima H., Selection of Step Size
bers of dates, and also it can only be applied in th Parameter of the Adaptive Filter for Spectrum
e noisy elements with low frequency. Accordingly, Analysis, Transactions of the Society of Instrument
even if the performance of the equipment is improv and Control Engineers, Vol.28 No.10
ed, the progress in accuracy is not able to be expe (1992.10)pp.1257-1262
cted. Therefore, in this paper, we focus the frequen [7] Umemoto T. and Aoshima H., Adaptive DFT and its
cy analysis method by using LMS algorism, and co Application, Transactions of the Society of Instrument
nsider the system in which an input is d (k ) and and Control Engineers, Vol.30 No.8
(1994.8)pp.959-965
an output is ε (k ) . First, we showed that adaptive n [8] B. Widrow and S. D. Stearns. Adaptive Signal
otch filter is composed by inspecting its transfer Processing. New Jersey: Prentice-Hall, Inc.(1985)

- 41 -
Authorized licensed use limited to: Chulalongkorn University. Downloaded on September 9, 2009 at 22:54 from IEEE Xplore. Restrictions apply.