Академический Документы
Профессиональный Документы
Культура Документы
x
E
a cos u
2
y
E
a sin u
2
_
3
and with the positive direction of the Ox-axis it creates the following angle:
O
E
A
B
M
x
C
y
A
x
y
E
a
d
b
s
c
r
x
x
k
x
y
A E
p
D
Fig. 1. Geometry of a four-bar linkage.
R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246 237
h arctg
y
E
a sin u
x
E
a cos u
_ _
; p 6 h 6 p: 4
The coordinates of the point D are determined by the expressions
x
D
x
A
x
E
c cosh t c; 5
y
D
y
A
y
E
c sinh t c; 6
where t is the coecient whose value is
t = 1, for the case when the point D is above the line segment s (Fig. 1), and
t = 1, for the case when the point D is below the line segment s (crossed mechanism),
and where c, the angle created between the rocker and the line segment s, is dened by the expression
c arccos
s
2
c
2
b
2
2 s c
_ _
; 0 6 c 6 p: 7
The angle w created between the coupler BD and the positive part of the Ox-axis is determined by the
expression
w arctg
y
D
y
B
x
D
x
B
_ _
: 8
Finally, the position of the point M of the coupler, the point moving along the desired path, is given by the
following equations:
x
M
x
A
a cos u d cos w r cos
p
2
w
_ _
; 9
y
M
y
A
a sin u d sin w r sin
p
2
w
_ _
: 10
3. Optimum synthesis of a four-bar linkage and tools employed
3.1. Formulation of the optimisation problem
The paper considers the following optimisation problem:
Find optimum dimensions of the mechanism for the desired rectilinear path traced by the point M (Fig. 1)
of the coupler of the four-bar linkage so that the objective function has the minimum value.
Thus dened optimisation problem can be given the following general mathematical formulation:
minimise f X; 11
subject to g
j
X 6 0; j 1; . . . ; n
g
; 12
f(X) is the objective function, g
j
(X) 6 0 represent the constraints dened by the search space, n
g
is the total
number of constraints.
X = {x
1
, . . . , x
D
}
T
represents the design vector consisting of D design variables. The design variables are the
values which should be dened during the optimisation procedure. Each design variable is dened by its lower
and upper boundaries. For the case of the four-bar linkage (Fig. 1), the design vector is X = (a, b, c, d, r, x
A
,
y
A
, x
E
, y
E
, u
0
, u
1
)
T
.
By introducing the constraints in the objective function, Eqs. (11) and (12) can be transformed into the fol-
lowing form:
Minimise F X f X PX; k
j
; 13
where P(X, k
j
) are the penalty functions which can be presented in the following way:
238 R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246
PX; k
j
n
g
j1
k
j
max0; g
j
X
2
_ _
: 14
When the solution is found outside the region considered, then the current parameters of the solution are
squared and multiplied by large positive numbers (penalty factors) k
j
, and then added to the numerical values
of the objective function.
3.2. DE algorithm
The DE algorithm is briey described here, and the control parameters of the algorithm are also dealt with.
A detailed description of the DE algorithm can be seen in Refs. [712].
DE is a simple, but still strong evolutionary algorithm for realisation of the global minimum in numerous
real optimisation problems. The DE algorithm, similarly to GA, has the following control parameters: the
population size NP, the crossover constant CR and the mutation constant F. Coding of chromosomes with
real numbers, i.e. presentation of chromosomes as vectors of real values, is used in numerical applications
of DE in optimisation processes.
Generation of the initial population is performed stochastically. At that, the population size NP is
commonly ten times bigger than the number of design variables. At the beginning, each design variable is a
random value which is found within the dened upper and lower boundaries. While dening the bound-
aries, attention should be paid that the values of design variables are not out of the range which is really
acceptable.
The mutation constant in DE is a real parameter, which controls the increase of dierence between two
individuals in the search space. The dierence between two randomly chosen vectors denes the magnitude
and direction of mutation. When the dierence is added to a randomly chosen vector, it becomes a mutant
vector. The basic idea of DE is that mutation is self-adaptive in the search space and the current population.
At the beginning of the optimisation process, the magnitude of mutation is large because the vectors in the
population are far away from the search space. When the process starts to converge, the magnitude of muta-
tion starts to decrease. The self-adaptive mutation in DE leads the solution of the optimisation process toward
the global minimum.
For taking the best values for CR, there are certain basic rules dened in [12]. High values are eective for
all problems, but they are not always the fastest. The problems with heavy interaction between design vari-
ables generally require a high CR. But, if interaction between design variables is lower, a lower CR can be
used, which results in obtaining a satisfactory solution with a smaller number of iterations (faster solution).
The interaction between design variables in this paper is high so that a high CR, i.e. CR = 0.8 is used in all
examples. Also, on the basis of recommendations in [12], the values of other control parameters in all exam-
ples are: NP = 110 and F = 0.6.
3.3. Method of variable controlled deviations
For the purpose of achieving ideal operation of a mechanism within a device, it is necessary that at a certain
section of the path the observed point should move according to the precisely dened law. The desired law of
motion in a general case can be represented by the shape function and the position function.
In a large number of cases, the mechanism will perform its operation successfully even when the working
part of its path does not coincide with the ideal one. It is enough that the real path is in the controlled space
around the ideal path where the allowed deviations are prescribed in advance.
For the four-bar linkage in this paper, the motion of the coupler point on a segment is dened by expres-
sions (9) and (10), within which all design variables gure either directly or indirectly.
Solution of the problem is reduced to forming of the objective function in which there are also deviations
bigger than the allowed ones. Minimisation of the objective function results in bringing the mechanism path
within the allowed deviations around the given path. As the values of real deviations do not gure in the syn-
thesis process, it can be concluded that, if there is one solution, then there is an innite number of indenite
R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246 239
similar solutions. The following relations give the projections of dierences between actual and allowed
deviations:
n
i
jy
M
i y
d
ij dyi if jy
M
i y
d
ij dyi > 0;
0 otherwise;
_
15
g
i
jx
M
i x
d
ij dxi if jx
M
i x
d
ij dxi > 0;
0 otherwise:
_
16
i 1; 2; . . . ; N;
where N is the number of given points on the path, x
M
and y
M
are the real coordinates of the point M dened
by Eqs. (9) and (10), x
d
and y
d
are the coordinates of the point M, i.e. the coordinates lying on the given path,
dx and dy are the allowed deviations of the position of the point M from the given position.
Without reducing the generality of consideration, it is taken that the given path is the segment of the length
L on the straight line (l), which is dened by y
d
= c = const. The coordinate x
d
is dened by the expressions
x
d1
x
p
for u u
0
;
x
di
x
p
Li 1 x
x
=N 1 for u
0
< u < u
0
u
1
;
x
dN
x
k
x
p
L for u u
0
u
1
;
i 2; 3; . . . ; N 1 and x
x
2 0; 1:
17
By introducing the variable x
x
whose value during optimisation is randomly changed in the interval (0, 1), the
whole path is found within the required controlled space.
The controlled deviations dx and dy change during the optimisation process. At the beginning of the syn-
thesis process, the boundaries of allowed deviations are set considerably out of those that should be accom-
plished and they are marked as maximum values in Fig. 2. The synthesis process is activated and as soon as the
objective function reaches the zero value, i.e. when all points of the generated path are in the prescribed envi-
ronment of the given path, the allowed deviations decrease, and the synthesis process continues. Variable con-
trolled deviations both for the shape function and the position function are thus introduced in the
optimisation process, directly in the DE algorithm. The rule for decrease of allowed deviations during the opti-
misation process is described by the following equations:
dy dy c
1
dx dx
_
; when dy > dy
mid
and dx > dx
min
;
dy dy
dx dx c
2
_
; when dy 6 dy
mid
and dy Pdy
min
and dx > dx
min
;
dy dy c
3
dx dx
_
; when dy 6 dy
mid
and dy > dy
min
and dx 6 dx
min
:
18
M(x
d
(i), (i))
y
dx
min
M(x
M
(i),
M
(i))
y
dx
max
d
y
m
i
n
d
y
m
i
d
d
y
m
a
x
x
y
desired path
d
Fig. 2. Decrease of controlled deviations.
240 R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246
The values c
1
, c
2
and c
3
are positive numbers given at the beginning of the optimisation process. It is desirable
for c
3
to be signicantly smaller than the remaining two constants because in that case the path approaches the
given path more safely and the local minima are avoided. The value dy
mid
represents a mid-value of the con-
trolled deviation dy.
Fig. 3 presents the given functions in the eld of allowed deviations. The shaded parts represent undesired
deviations which should be minimised.
3.4. Objective function
On the basis of Eq. (13), the objective function is dened by the following expression:
f a; b; c; d; r; x
A
; y
A
; x
E
; y
E
; u
0
; u
1
k
N
i1
n
2
i
g
2
i
m
j1
k
1
g
2
1j
k
2
g
2
2j
k
3
g
2
3
; 19
where N is the number of given points representing the rectilinear portion, and k, k
1
, k
2
and k
3
are large num-
bers inuencing the importance of values multiplied by them.
The rst member in the given expression for the objective function represents the sum of squares of devi-
ations of the current path out of the controlled space around the given path, while the second and the third
members refer to the rest of the path and Grashof conditions, respectively.
The constraints are dened in the following manner:
(1) Scope of initial values of design variables. In the DE algorithm, the initial values of design variables are
dened by their upper and lower boundaries, i.e. a broader range of values which can be taken as the
initial solutions in the relevant area is given. Ref. [12] proposes the constraints directly in the DE algo-
rithm, which allow that the values of design variables remain within the mentioned boundaries during
the whole optimisation process. This has also been applied in this paper so that the constraints which
unable negative values of certain design variables have not been introduced in the objective function.
Namely, it is enough to take positive numbers for the lower and upper boundaries of those variables
whose values must not be negative. This refers to the lengths of the crank, the coupler and the rocker.
(2) The rest of the path. In the Eq. (19), m is the number of given points representing the rest of the path, and
g
1
and g
2
are the constraints which refer to the global shape of the rest of the path:
g
1j
xj x r xj x l for xj x r > 0 or xj x l < 0;
0 for xj x r < 0 and xj x l > 0;
_
20
g
2j
yj y up yj y low for yj y up > 0 or yj y low < 0;
0 for yj y up < 0 and yj y low > 0;
_
21
L
x
1
Fig. 3. (a) The shape function; (b) the position function.
R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246 241
where x_r is the right-hand boundary, x_l is the left-hand boundary, y_low is the lower boundary and y_up is
the upper boundary of the rectangular area, and x(j), y(j) are the real coordinates of position of a point of the
coupler, which lie outside the rectilinear part of the path.
It happens that the optimisation process results in stretching of the rest of the path or in its undesired shape.
That is the reason why a rectangular area around the rectilinear portion of the path in which the rest of the
path should lie is given. The constraint of the whole path by the rectangular area has shown to be more
favourable than the constraint of the working angle because dierent values of working angles as well as a
much more acceptable path can be obtained.
(3) Grashof conditions. The sum of lengths of the shortest and the longest member of a four-bar linkage must
be smaller than the sum of lengths of the remaining two members. The constraint referring to Grashof
conditions in Eq. (19) is g
3
:
g
3
l
max
l
min
l
mid1
l
mid2
for l
max
l
min
l
mid1
l
mid2
> 0;
0 otherwise;
_
22
where l
max
= max(a, b, c, f), l
min
= min(a, b, c, f), f
x
2
E
y
2
E
_
, l
mid1
and l
mid2
are the lengths of the remaining
two members.
4. Results
The described method has been applied for the synthesis of a four-bar linkage which accomplishes rectilin-
ear motion. The method has been used in 10 examples in which the desired path is a segment of the straight
line parallel to the Ox-axis. The beginning of the segment coincides with the coordinate beginning, and its
length is 400 mm. The number of precision points has been varied for dierent cases. The smallest number
of observed precision points is 16, and the largest one is 80. All of them are distributed along the given segment
of the straight line and in each iteration their positions are changed stochastically. Since the DE method has
been applied, and it requires the initial boundaries for design variables, the same values have been used for all
examples and they are presented in Table 1.
The constants used in the optimisation process are presented in Table 2. The maximum number of itera-
tions in all examples is 5000.
Tables 3 and 4 show the nal results of the optimisation processes. The presentation of 10 obtained solu-
tions (mechanisms) is given, as well as the number of precision points, the number of iterations, the minimum
deviations of the shape function and the position function. The values of deviations of the shape function, the
position function and the working angle are also given in percentage.
The use of a large number of precision points has aimed at as precise accomplishment of rectilinear portion
of the path as possible. Synthesis has also been performed for dierent values of minimum deviations of the
shape function and the position function.
Table 1
Initial values of design variables
a (mm) b (mm) c (mm) d (mm) r (mm) x
A
(mm) y
A
(mm) x
E
(mm) y
E
(mm) u
0
(rad) u
1
(rad)
Lower boundary 0 0 0 900 900 900 900 900 900 3 3
Upper boundary 900 900 900 900 900 900 900 900 900 3 3
Table 2
Constants used in the optimisation process
c
1
(mm) c
2
(mm) c
3
(mm) dy
max
(mm) dx
max
(mm) dy
mid
(mm) m e (error)
2 2 0.1 dy
min
+ 10 dx
min
+ 10 dy
min
+ 2 5 0
x_l (mm) x_r (mm) y_low (mm) y_up (mm) k k
1
k
2
k
3
270 670 300 300 10
6
10
8
10
8
10
6
242 R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246
According to Tables 3 and 4, it is obvious that in certain examples some values of the design variables are
out of the boundaries dened in Table 1. Retention of the design variables within the given initial boundaries,
for small values of allowed deviations of the shape function and the position function, has made the synthesis
process unrealisable. Therefore, constraints up to some values of allowed deviations of the shape function and
the position function have been used, all until the entrance of the mechanism into the zone of solution, and
then those constraints have been excluded and the optimisation process has continued. Exclusion of the con-
straints at the very beginning would result in an unfavourable solution, in the sense of disproportion of the
members of the mechanism or negative lengths of the members.
Figs. 47 present the nal mechanisms from Example 1 and Example 10, as well as their paths. The mech-
anism from the rst example has the biggest minimum deviation of the shape function 0.05 mm and the
Table 3
The obtained mechanisms with deviations of the shape function and the position function in (%), as well as the values of the working angle
also in (%)
Example 1 Example 2 Example 3 Example 4 Example 5
a (mm) 125.978916 96.705406 139.591303 135.497672 129.412911
b (mm) 411.399256 371.127732 604.312170 467.980994 468.645202
c (mm) 541.496506 328.692471 893.584293 480.876733 449.268731
d (mm) 592.409710 775.161111 681.718428 901.629932 978.557388
r (mm) 128.981898 281.463296 236.511665 83.625963 8.220348
x
A
(mm) 41.142091 4.050515 65.002319 112.894205 160.207318
y
A
(mm) 479.253399 727.970387 581.438747 767.149999 844.281041
x
E
(mm) 302.434924 177.225533 393.056215 336.678285 351.114253
y
E
(mm) 108.279994 124.917147 243.074691 14.002536 22.629577
u
0
(rad) 1.835903 1.535475 1.930653 2.039952 2.194976
u
1
(rad) 2.472382 2.199662 2.157540 1.898473 1.852450
N 16 32 48 64 80
Number of iterations 567 744 758 834 2080
dy
min
(mm) 0.05 0.04 0.03 0.02 0.01
dx
min
(mm) 7 3 3 3 3
dy (%) 0.025 0.02 0.015 0.01 0.005
dx (%) 3.5 1.5 1.5 1.5 1.5
u
1
(%) 39.34918 35.00871 34.33832 30.21514 29.48266
Table 4
The obtained mechanisms with deviations of the shape function and the position function in (%), as well as the values of the working angle
also in (%)
Example 6 Example 7 Example 8 Example 9 Example 10
a (mm) 176.819373 177.504353 199.385132 189.437426 203.611519
b (mm) 619.410129 625.399922 733.773373 683.792342 717.963041
c (mm) 747.520042 753.337923 1079.669236 766.947339 878.658085
d (mm) 1006.536306 1018.699233 994.863267 1207.147526 1174.481687
r (mm) 143.022337 142.953350 183.812366 118.031209 111.341095
x
A
(mm) 201.388994 200.485390 213.072039 265.279043 279.296622
y
A
(mm) 834.444226 845.523541 805.770289 1015.939001 966.625118
x
E
(mm) 460.200949 464.503441 557.097831 507.155936 553.408409
y
E
(mm) 102.977265 102.817834 336.628634 59.485926 156.426044
u
0
(rad) 2.325397 2.310155 2.365534 2.417108 2.464141
u
1
(rad) 1.591044 1.579201 1.521278 1.379626 1.380132
N 80 80 80 80 80
Number of iterations 1195 1241 1274 1286 1399
dy
min
(mm) 0.005 0.004 0.003 0.002 0.002
dx
min
(mm) 3 3 3 3 3
dy (%) 0.0025 0.002 0.0015 0.001 0.001
dx (%) 1.5 1.5 1.5 1.5 1.5
u
1
(%) 25.32251 25.13376 24.21189 21.95743 21.96548
R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246 243
position function 7 mm, with the smallest number of precision points 16. In contrast to this one, the mech-
anism from the tenth example has the deviation of the shape function of 0.002 mm, or, in percentage
0.001%, and the deviation of the position function whose value is 3 mm or 1.5%. The number of precision
points for this mechanism is 80. These are the two mechanisms representing all mechanisms shown in Tables
3 and 4.
600 400 200 0 200 400 600 800 1000
1000
800
600
400
200
0
200
400
Fig. 4. Mechanism from Example 1.
300 200 100 0 100 200 300 400 500 600 700
500
400
300
200
100
0
100
200
300
400
500
Fig. 5. Path of the mechanism from Example 1. The number of precision points is 16, and the maximum deviation from the straight line is
0.025%.
244 R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246
5. Conclusion
This paper described the process of optimal synthesis of a four-bar linkage by the method of variable con-
trolled deviations with the application of the DE algorithm. The method of variable controlled deviations is
suitable because deviation of motion of the working member point in relation to the projected path can be
followed at any moment. In the largest number of cases, when variable controlled deviations are not applied
600 400 200 0 200 400 600 800 1000
1000
800
600
400
200
0
200
400
Fig. 6. Mechanism from Example 10.
300 200 100 0 100 200 300 400 500 600 700
500
400
300
200
100
0
100
200
300
400
500
Fig. 7. Path of the mechanism from Example 10. The number of precision points is 80, and the maximum deviation from the straight line
is 0.001%.
R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246 245
to the synthesis process, and when the projected deviations are considerably small, realisation of the synthesis
of a four-bar linkage becomes practically impossible. Namely, if constant controlled deviations are taken in
the optimisation process, the value of the objective function stagnates and convergence is impossible. How-
ever, the application of variable controlled deviations for a relatively small number of iterations results in
the solutions in which the objective function decreases until the zero value. Deviations are controlled by pre-
scribing their allowed values within which motion of the coupler point is followed. The method is illustrated
by the example of synthesis of a four-bar linkage where the point on the working member moves along the
rectilinear segment the length of which is 400 mm. For the given example, this paper presents the results of
synthesis in which a very high accuracy was realised. Namely, a mechanism in which the deviation of the shape
function is 0.001%, and the deviation of the position function is 1.5%, was obtained. These values correspond
to the working angle of 79.07 or 1.380132 rad, i.e. 21.96548% of the coupler motion cycle. Further research in
this eld will be directed to elaboration of the ways of application of this method for the case of an arbitrary
shape of the path which should be realised by a point on the working member of the four-bar linkage during
its motion.
References
[1] R.L. Norton, In search of the perfect straight line and constant velocity too design charts for optimum straight-line crank-rocker
four bar linkages with near constant velocity, in: Proc. 6th Applied Mechanisms and Robotics Conference, Cincinnati, OH, 1999.
[2] R.L. Norton, Design of Machinery an Introduction to the Synthesis and Analysis of Mechanisms and Machines, third ed., McGraw-
Hill Inc., New York, 2004.
[3] R. Bunduwongse, K.L. Ting, Synthesis of straight line mechanism with fth order burmester points at inection pole and
transmission angle control, in: Second Applied Mechanisms and Robotics Conference, Cincinnati OH, II.A.3-1, 1991.
[4] S.R. Dordevic, A New Approach to Technological Requirements of Motion of Mechanisms in Their Synthesis, Doctoral
Dissertation, Faculty of Mechanical Engineering, University of Belgrade, Belgrade, 1988 (in Serbian).
[5] J.A. Cabrera, A. Simon, M. Prado, Optimal synthesis mechanisms with genetic algorithms, Mechanism and Machine Theory 37 (10)
(2002) 11651177.
[6] A. Kunjur, S. Krishanamury, Genetic algorithms in mechanical synthesis, Journal for Applied Mechanisms and Robotics 4 (2) (1997)
1824.
[7] R. Storn, K. Price, Dierential evolution a simple and ecient heuristic for global optimization over continuous spaces, Journal of
Global Optimization 11 (4) (1997) 341359.
[8] K. Price, R. Storn, A simple evolution strategy for fast optimization, DR. Dobbs Journal 264 (1997) 1824.
[9] R. Storn, On the usage dierential evolution for function optimization. <http://http.icsi.berkley.edu/~storm/litera.html>.
[10] J. Lampinen, A Bibliography of Dierential Evolution Algorithm, Lappeenranta University of Technology, Finland, 2001.
[11] S. Kukkonen, J. Lampinen, Comparison of generalized dierential evolution to other multi-objective evolutionary algorithms, 2004.
<http://www.imamod.ru/~serge/arc/conf/ECCOMAS_2004/ECCOMAS_V2/proceeding/pdf/716.pdf> (Online).
[12] K.V. Price, R.M. Storn, J.A. Lampinen, Dierential Evolution A Practical Approach to Global Optimization, Springer-Verlag,
Berlin Heidelberg, 2005.
[13] P.S. Shiakolas, D. Koladiya, J. Kebrle, On the optimum synthesis of six-bar linkages using dierential evolution and the geometric
centroid of precision positions technique, Mechanism and Machine Theory 40 (3) (2005) 319335.
246 R.R. Bulatovic , S.R. Dordevic / Mechanism and Machine Theory 44 (2009) 235246