Available online at www.sciencedirect.

com

Mechanism and Machine Theory 44 (2009) 4256

Mechanism
and
Machine Theory
www.elsevier.com/locate/mechmt

Kinematic and sensitivity analysis and optimization

of planar rack-and-pinion steering linkages
A. Rahmani Hanzaki a,*, P.V.M. Rao b, S.K. Saha b
a

Mechanical Engineering Department, Shahid Rajaee University, Lavizan, Tehran 16788, Iran
Mechanical Engineering Department, Indian Institute of Technology Delhi, Hauz Khas, New Delhi 110 016, India
Received 10 May 2007; received in revised form 19 February 2008; accepted 26 February 2008
Available online 15 August 2008

Abstract
In this paper, the combined kinematic and sensitivity optimization of a rack-and-pinion steering linkage is performed.
This steering linkage is the most common steering system used in passenger cars. Although, the steering linkage has
received a lot of attention for the minimization of the steering errors, no attempt has been made so far to investigate
the sensitivities of optimum dimensions relative to variation of link lengths. The kinematic optimization of the linkage
is carried out using three homogenous design parameters. The objective of the proposed optimization is to minimize maximum steering error during cornering. This is followed by a sensitivity analysis to predict how the steering error is aected
by manufacturing tolerances, assembly errors, and clearances resulting due to wear. Since the optimized kinematic error is
very sensitive to the variations of the linkage parameters, the kinematic and post-optimal sensitivity optimization of the
steering linkage is performed in an integrated manner. The methodology proposed in this work helps the designers of
rack-and-pinion steering linkage to choose the linkage parameters whose maximum steering error (MSE) and sensitivity
are minimum.
2008 Published by Elsevier Ltd.
Keywords: Rack-and-pinion steering linkage; Steering error; Kinematic optimization; Post-optimal sensitivity; Cognates

1. Introduction
Among the steering linkages, rack-and-pinion steering linkage is the most widely used in passenger cars. It
consists of two steering arms, two tie rods, and a rack. The linkage has two common congurations, namely,
central take-o and side take-o, as shown in Fig. 1. In central take-o (CTO) conguration, the tie rods and
the rack are connected at the middle of the rack as depicted in Fig. 1a, while in side take-o (STO) conguration, these connections are at the rack ends as shown in Fig. 1b. Each of the above congurations can be
either trailing or leading type, as shown in Fig. 2a and b, respectively [1].
*

Corresponding author. Tel.: +98 91 23252729; fax: +98 22970052.

E-mail addresses: a.rahmani@srttu.edu (A. Rahmani Hanzaki), pvmrao@mech.iitd.ac.in (P.V.M. Rao), saha@mech.iitd.ac.in
(S.K. Saha).
0094-114X/$ - see front matter 2008 Published by Elsevier Ltd.
doi:10.1016/j.mechmachtheory.2008.02.014

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

Steering arm

Kingpin

Tie rod

Pinion

43

Tie rod

Steering arm

Rack

Pinion Rack
Kingpin

(a) Central take-off (CTO)

(b) Side take-off (STO)

Wheel

Fig. 1. Rack-and-pinion steering linkage and its congurations.

Front motion

Front motion

(a) Trailing

(b) Leading

Fig. 2. Trailing and leading type of rack-and-pinion steering linkages.

Wb

Wt

Fig. 3. Ackermann condition for a vehicle when turning.

In order to provide pure rolling to the road wheels and to reduce wear and tear of the tires, a steering linkage must handle the vehicle so that it follows Ackermann principle (see Fig. 3). This principle states that during low speed cornering when free from lateral inertia forces, the verticals drawn from the centers of the wheels
should meet at the center of bend, i.e., point O of Fig. 3. For a two-wheel steering vehicle, this point must lie
on the common axis of the rear wheels [1]. Referring to Fig. 3, the relation between the inner wheel angle, hI,
and the outer wheel angle according to Ackermann principle, hOA, is given as
hOA hI tan1

1
1
1
W t tan
cot hI 1=wb
cot hI W b

where wb = Wb/Wt is the normalized expression of the wheel base, Wb, with respect to wheel track, Wt. In
reality, Eq. (1) is never satised for every radius of orientations. Hence, there are eorts to synthesize the linkage so that Ackermann principle is satised for any orientation of the wheels as closely as possible. In order to
do that, it is necessary to obtain the angle hO for a given value of hI. Hence, an appropriate kinematic model of
the steering linkage is essential. In addition, the kingpin inclination and caster angles that provide compliance
to the steering linkage with the suspension system have little inuence on the motion transmission of the steering linkage. As a result, the real rack-and-pinion steering linkage, which is spatial in nature, can be modeled as
a planar linkage for the investigation of Ackermann condition. Such a simplication of the steering system has
been also used by other researchers, e.g. [2,3].

44

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

Error optimization studies in steering linkages have been attempted by many researchers. Zarak and Townsend [3] optimized the STO conguration as shown in Fig. 1b, where they considered the distance between the
inner and outer wheel turning centers as the steering error and minimized it for dierent rack travels. During
optimization, they used four parameters that were normalized with respect to the wheel track. Felzien and
Cronin [4] investigated the minimization of steering errors. They considered an integrated McPherson suspension and steering linkage model, and minimized the weighted sum of the squares of steering errors. Simionescu
and Smith [5] discussed Watt II function generating cognates, and showed that STO conguration of the steering linkage has innite number of cognates, of which one is the CTO conguration. Simionescu and Smith [2]
used three parameters, namely, a normalized link length/a link length ratio and two angles, in the case of
CTO/STO conguration to optimize the steering errors of the linkages. The choice of the design parameters
used by them, some of which are in terms of angles, is not appropriate however for link-length sensitivity analysis. Hence, the kinematic optimization carried out in this paper uses three homogeneous design parameters,
namely, those which have the units of length. Moreover, manufacturing tolerances, assembly errors, and clearances resulting due to wear, which are inherent to any real steering linkage, may aect the objective function
value signicantly. Hence, it is important to perform a post-optimal sensitivity analysis, in addition to kinematic optimization. Furthermore, a method based on rack-and-pinion steering linkage cognates given in [5] is
used to generalize the steering optimization. Using this methodology, the steering linkage can be optimized
once and the results are extended to desired rack-and-pinion steering linkage whether CTO or STO with
any rack length. To the best of the authors knowledge, no work concerning the sensitivity of the error for
the optimized dimensions of a rack-and-pinion steering system has been reported so far. In addition, the sensitivity minimization, which is carried out in this paper as a part of optimization process, is also new in the
context of rack-and-pinion steering linkage design. Such minimization of both the objective function and
the sensitivity has been considered important in the literature of robust optimal design [6,7]. In summary,
the contributions of the paper are
 A simple generalized methodology for the optimization of a rack-and-pinion linkage for both CTO and
STO is proposed.
 Post-optimal sensitivity of the steering linkage is performed.
 The steering linkage is optimized for minimum steering error as well as for minimum sensitivity.
 A methodology for multi-objective optimization problem is also presented, where the number of design
parameters is reduced by one, thus increasing the eciency and speed of the optimization.
This paper is organized as follows: Section 2 shows that the optimized results for a CTO linkage can be used
to nd the optimized parameters of a corresponding STO linkage, consequence that any STO conguration
has a CTO cognate conguration. Section 3 presents a generalized kinematic modeling of a planar six-bar
rack-and-pinion steering linkage, which would then be used in Section 4 to optimize the steering linkage under
study. Post-optimal sensitivity analysis is then carried out in Section 5, which is further used in Section 6 to
optimize the steering linkage that has minimum sensitivity in addition to minimum steering error. The proposed methodologies are illustrated in Section 7, followed by the conclusions in Section 8.
2. Cognates
The side take-o (STO) conguration of the steering linkage, shown in Fig. 1b, is more common in passenger
cars. This linkage noted AB0 C1C2D0 E in Fig. 4, has innite number of cognates [5], meaning that the input-output link behaviors of all the cognate linkages are same. For example AB00 C01 C02 D00 E whose links are parallel to
the original linkage is one of them. Continuing in a similar manner, as shown in Fig. 4a, linkage ABCDE is also
a cognate, which is nothing but the central take-o (CTO) conguration of the steering linkage introduced in
Fig. 1a [5]. Note from Fig. 4b that another CTO cognate for the STO linkage, i.e., AB0 C1C2D0 E of Fig. 4b, is
A1B1C2D0 E, where A1E = AELr [5]. As a result, if the link lengths of the STO linkages shown in Fig. 4a are
divided by AELr, the link lengths of their CTO cognate, ABCDE of Fig. 4a, will be obtained. Since in function
generating cognates, the angular variables of similar links of cognates are equal, the results obtained for the
CTO conguration can be utilized for the analysis of STO conguration cognates as well. It also implies that

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

lt

la

C 2

C 1

C2

C1

Lt

La

45

Lt

Lr

Wt

(a) A set of the cognates of the steering linkage

D
Lt
La

B1

C2
Lr

Lt

r
La

A1
E

C1

Wt

(b) Different arranging of the cognates

Fig. 4. Cognates of rack-and-pinion steering linkage.

the STO linkages are as sensitive as their CTO cognate. Hence, to extend the results obtained for a CTO conguration to its STO cognates, the following simple method can be used.
The real linkage link lengths except rack length are divided by (WtLr), while the wheel base and wheel
track, i.e., Wb and Wt, are divided by Wt. This normalization is opposed to that in previous works by other
researchers like Simionescu and Smith [2] and Zarak and Townsend [3], where the normalization is done with
respect to wheel track, Wt. By this normalization, the CTO linkage, namely, ABCDE of Fig. 4a is obtained.
Moreover, the normalized parameters, La, Lt, H, Wb, and Wt, are denoted as la, lt, h, wb, and wt, respectively.
After optimization, the optimum la, lt, and h are denormalized1 to nd La, Lt, and H of the optimized STO
linkage for a certain rack length. It is obvious that a CTO conguration can be treated in the same way
due to the fact that it is same as a STO linkage except its rack length, Lr, is zero. Note that Lr is dened
as the distance of the two tie rod ends connected to the rack for the kinematic problem.
3. Kinematic modeling
As explained in Section 2, the CTO conguration shown in Fig. 1a is considered here for the purpose of
kinematic modeling and analysis. The CTO steering linkage is kinematically a six-bar linkage [8] as shown
in Fig. 5. In this gure, the links are denoted by #1, . . . , #6 and the joints by 1, . . . , 7. Kutzbach criterion
illustrates that the DOF of this linkage is one. Moreover, vectors, l1 and l5 denote the steering arms of length
la, whereas vectors, l2 and l4, represent the tie rods of length lt. Furthermore, at the end of every link, a twodimensional coordinate system is considered that is xed to its previous link.
From Fig. 5,
l1 l2 wt =2 bi hj

where wt is the wheel track, b is the rack displacement, h is the distance from the front wheel axis to the rack
axis, and i and j are the unit vectors along X1 and Y1, respectively. Eq. (2) can be written in terms of its scalar
components as
1

Denormalizing here means converting the normalized parameters to real link lengths by multiplying the normalizing factor.

46

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

X2

#2

l1

Y1

#1

O1

#3

3, 4

+b

l2

wt

X1

l4

#4
3

X3

5
4
X4

6 X5

#5
l5
O7

X6

Fig. 5. Six-bar planar rack-and-pinion steering linkage for the CTO conguration.

la ch1 lt ch12 wt =2 b

and

la sh1 lt sh12 h

3; 4

where c and s stand for cosine and sine functions, respectively, h12  h1 + h2, and the normalized wheel track,
wt, is unity. Referring to the derivations shown in Appendix A, the initial angle of the left steering arm denoted
as h10 is obtained from Eq. (A.9) by using Eqs. (A.7) and (A.8). This is also equal to h60. If the range of the
inner wheel rotation is considered as 40, which is quite practical for most of the passenger cars, the results can
be obtained for the range of hI = 040. Note that, for the trailing conguration, as shown in Fig. 2a,
h1 = h10  hI, and for the leading conguration which is shown in Fig. 2b, h1 = h10 + hI when turning left. Furthermore for straight ahead conguration, the initial value of h6, i.e., h60, is equal to h10. To evaluate angle h6
in an arbitrary position of the mechanism, the expression of the rack displacement is calculated rst as
p
5
b la ch1  wt =2  D
where
D  l2a ch21  k 1 2la hsh1

and

k 1  l2t l2a h2

In Eq. (5), + sign should be used. This is followed by the derivation of an expression similar to Eq. (2) using the
right side of the steering linkage, Fig. 5, which would provide the values of h6 given by Eqs. (A.10) and (A.11).
Pressure angle, which plays important role in force transmission eciency of a linkage, is dened as the
angle between the velocity vector of the driven link and the driver link direction. The pressure angle at joint
2 of the steering linkage of Fig. 5 is shown by b2 and is the angle between the perpendicular to the steering
arm, i.e., #1, and the tie rod, namely, #2. According to the above denition, the pressure angle at joint 3
is the same as h3. The pressure angles at joints 4 and 5 are found similar to those at joints 3 and 2, respectively.
4. Steering error optimization
In this section, optimization of the steering linkage with CTO conguration is carried out to minimize the
steering error, dhO, which is dened as
dhO jhO  hOA j

where hO is the actual angle made by the outer front wheel during steering maneuver and hOA is the correct
angle for the same wheel based on the Ackermann principle given by Eq. (1). Referring to Fig. 5, the angle hO
is the variation of h1 or h6 when turning left, depending on if the steering linkage is trailing or leading, respectively. The objective here is to minimize the maximum value of the steering error, abbreviated as MSE during
entire rack travel [2], where wt is unity. The following constraints are considered:
(I) The initial assemblage of the linkage must be feasible by the link lengths, which implies that D10 of Eq.
(A.8) should be non-negative, i.e., D10 P 0.
(II) The input link should be able to rotate at least 40. In other words, when hI-max = 40, D of Eq. (6)
should be non-negative, i.e., D P 0.
(III) The rack displacement must be unidirectional while satisfying condition II meaning that jbij > jbi1 j, for
i = 1, . . ., n, discrete positions of the rack, where bi is the solution of Eq. (5) corresponding to the input
angle, hIi, at the ith step.

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

47

(IV) The output link should also be able to rotate in correspondence with the rack travel, b. In other words,
D6 of Eq. (A.11) must satisfy D6 P 0.
The kinematic optimization problem is now formally posed as
w  max jhO hI  hOA hI j
D10 P 0; D P 0; D6 P 0;

Minimize MSE;
Subject to

8a
8bd

and

jbi j P jbi1 j i 1; 2; . . . ; n

8e

Variations of the objective function, Eq. (8a), with respect to la, lt, and h are shown in Fig. 6a and b, as a sample, where la, lt, and h represent the normalized steering arm length, tie rod length and the distance from the
front wheel axis to the rack axis as already pointed out. In these gures, there are two groups of points lying

20

MSE (deg)

15

10
0

5
0.2

0
0.3

0.4

lt

0.5

0.6

0.7

0.4

(a) MSE, , vs. h and l t (trailing, w b =1.4, l a =0.14)

20

MSE (deg)

15

10

5
0
0.1
0.2

la

0.3
0.4

0.3

0.4

0.6

0.5

lt

(b) MSE, , vs. l a and l t (trailing, w b =1.4, h =0.1)

Fig. 6. Variation of maximum steering error.

0.7

48

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

on the horizontal planes where correspond to 20 and 10. The points whose MSEs are 20 indicate the congurations with MSEs more than 20, which are not acceptable. Hence, their MSEs are replaced by 20. Similarly, the at surface corresponding to MSE = 10 is plotted to show the non-feasible or non-practical values
of linkage parameters. In other words, the link lengths associated to MSEs of 10 in Fig. 6a and b is not feasible, even if it is feasible, it is impractical as the input link cannot rotate by desired angle of hI-max = 40. After
studying the plots, the following observations can be made:
(i) MSEs along a curved line of Fig. 6a and b are so close to each other that choosing a set of link lengths as
optimum values is not acceptable in practice. Hence, additional constraints like space limitation, rack
travel, sensitivity, etc. can be considered during optimization.
(ii) The slope of the MSE function, w, around the optimum points is very high and dierent on their right
and left neighborhoods, which specify high sensitivity of the function, w. Since undesirable variations of
the design parameters of the steering linkage, such as manufacturing tolerances of the link lengths, clearances in the joints, body deformation, and adjustment of the toe angle of the front wheels. usually exist,
an appropriate sensitivity analysis with respect to the design parameters is important.
Keeping the above two observations in mind, optimization is carried out using the optimization toolbox of
MATLAB in this stage. The results for wb = 1.4 are shown in Figs. 7 and 8, and closely match the results of
Simionescu and Smith [2].
In order to verify the correctness of the results, a number of optimum points were analyzed and validated
by the methodology given in [8].
5. Post-optimal sensitivity analysis
For sensitivity study, two main denitions are found in the literature; Shar and Smith [9] dene it as the
variation of the output with respect to a small variation of the input, whereas in [10], it is dened as the ratio of
the variation of the objective function to a small variation of the design parameters at optimum point. The
latter is also referred to post-optimal sensitivity and adopted here. In this context, Knappe [11] used partial
derivatives to study the inuence of parameter deviations on the output. He obtained the total output deviation by adding the deviations due to individual parameters, then used these results to assign the practical tolerances. Shar and Smith [9] formulated tolerances and clearances corresponding to the allowable output
error using partial derivatives for a 10-bar linkage. Chakraborty [12] made a probabilistic model of mechanisms considering the tolerances on link lengths and joint clearances. Dhande and Chakraborty [13] treated

0.6

0.55

0.95

0.9

MSE (deg)

l*

0.5

0.45

4
0.4

4
0.85
0.8

2
0.75

0.35

0.7
0

0.1

0.2
h

0.3

(a) Optimum tie rod length

0.4

0.65

0.1

0.2
h

0.3

0.4

(b) Maximum steering error,

Fig. 7. Optimum tie rod length and corresponding maximum steering error for trailing CTO conguration, wb = 1.4, and la of (1)
la = 0.14, (2) la = 0.16, (3) la = 0.19, and (4) la = 0.24.

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

0.8

0.5

0.75

0.65

MSE (deg)

3
2

0.45

l*

0.7

1
2

0.4

3
0.35

0.6

0.55

49

0.1

0.2
h

0.3

0.4

(a) Optimum tie rod length

0.1

0.2
h

0.3

0.4

(b) Maximum steering error,

Fig. 8. Optimum tie rod length and corresponding maximum steering error for leading CTO conguration, wb = 1.4, and la of (1)
la = 0.14, (2) la = 0.16, (3) la = 0.19, and (4) la = 0.24.

the tolerances and clearances of a four-bar function generating linkage as optimization parameters, and
obtained the optimum values using the allowable output deviation as a constraint. Note that, for an optimization problem with continuous dierentiable objective function, the condition for optimality is that the rst
derivatives of the objective function are zero, as indicated in Fig. 9a. In such cases, the second derivatives leading to the Hessian matrix are used to analyze the post-optimal sensitivity of the output function with respect to
the variations of design variables [10]. In the case of an objective function whose gradient is not continuous
near the optimum point, as illustrated in Fig. 9b, the rst derivatives in the neighbourhood can be used to
study the sensitivity. As mentioned in item (ii) of Section 4, post-optimal sensitivity analysis is important
to investigate the inuence of the variation of the design parameters, namely, la, lt, and h, on the variation
of maximum steering error (MSE), w of Eq. (8a). Here, sensitivity is dened as the ratio of the variation of
the objective function, w, to the small variations in the design parameters la, lt, and h.
In the present case, the optimum point is a cusp, as evident from the plots of Fig. 6, i.e., the rst derivatives
of the objective function, w, with respect to the design parameters, la, lt, and h, at the optimum point are not
dened. Hence, the derivatives in the positive and negative neighborhoods of the optimal point are used
instead in the post-optimal sensitivity analysis.
For this purpose, the following six-dimensional sensitivity vector is introduced:

T
9
Dw  wla wla  wlt wlt  wh wh

F(x)

where wla  ow=ola for la  la > 0 and wla   ow=ola for la  la < 0, in which la is the optimum value of
la, etc.

+ variation of design parameter

Optimum point
x

Optimum point (a cusp)

design parameter

Fig. 9. Possible congurations of an optimum point (a) smooth and (b) non-smooth objective function.

50

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

12

MSE variation(%) / length variation(%)

MSE variation (%) / length variation (%)

Fig. 10 shows the sensitivity plots obtained using the rst partial derivatives of the objective function for the
trailing CTO conguration, where Richardsons extrapolation algorithm [14] is used here to ensure that the
derivatives are computed reliably.

11
10
9

5
4
0

0.1

0.2
h

0.3

0.4

70

60

1
50

2
3

40

30

20

0.1

0.2
h

0.3

25

1
20

2
3
15

4
10

0.4

70

50

2
3

40

30
20
10
0
0

0.1

0.2
h

0.3

0.2
h

0.3

0.4

160
140

120

100

80

60

0.4

(e) Positive neighborhood of h

0.1

0.2
h

0.3

0.4

(d) Negative neighborhood of lt

MSE variation (%) / lengthvariation (%)

MSE variation (%) / lengthvariation (%)

80

0.1

180

(c) Positive neighborhood of lt

60

(b) Negative neighborhood of la

MSE variation (%) / lengthvariation (%)

MSE variation (%) / lengthvariation (%)

(a) Positive neighborhood of la

30

35
30
25

1
2

20

3
15

4
10
5
0

0.1

0.2
h

0.3

0.4

(f) Negative neighborhood of h

Fig. 10. Sensitivity plots of the steering error for trailing CTO conguration with wb = 1.4, and la of (1) la = 0.14, (2) la = 0.16, (3)
la = 0.19, and (4) la = 0.24.

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

51

16

MSE variation(%) / length variation(%)

MSE variation (%) / length variation (%)

Now, sensitivity with respect to positive or negative variation of every parameter, i.e., in positive and negative neighborhoods, is provided in the plots in term of percentage of the MSE variation to percentage of
length variation. Hence, the percentage of MSE variation due to just the length variation is achieved by
multiplying the sensitivity into the percentage of the length variation. The sensitivity analysis for the leading

14
12
10
8

0
0

0.1

0.2
h

0.3

0.4

45
40
35
30
25
20

15

2
3

10

4
5
0

300

250

200

1
2

150

3
4

100

0.1

0.2
h

0.3

0.4

0.3

0.4

110
100
90
80

70

60

50

40
0

0.1

0.2
h

0.3

0.4

(d) Negative neighborhood of lt

120

50

MSE variation (%) / length variation (%)

MSE variation (%) / length variation (%)

(c) Positive neighborhood of lt

100

40

1
2
3
4

30

20

10

0
0

0.2
h

(b) Negative neighborhood of la

MSE variation (%) / length variation (%)

MSE variation (%) / length variation (%)

(a) Positive neighborhood of la

0.1

0.1

0.2
h

0.3

0.4

(e) Positive neighborhood of h

80

1
2
3
4

60
40
20
0
0

0.1

0.2
h

0.3

0.4

(f) Negative neighborhood of h

Fig. 11. Sensitivity plots of the steering error for leading CTO conguration with wb = 1.4, and la of (1) la = 0.14, (2) la = 0.16, (3)
la = 0.19, and (4) la = 0.24.

52

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

conguration of the steering linkage which was optimized in Fig. 8 is done and the results are presented in
Fig. 11.
The key observation from the plots is: sensitivity in positive and negative neighborhoods of every design
parameter is dierent, as shown in Figs. 10 and 11. Hence, this point should be seriously taken into account
while assigning manufacturing tolerances, etc. This means that plus and minus tolerances to the link lengths
should not preferably be the same. For example, sensitivity in positive neighborhood of lt, Fig. 11c, is almost
twice of that in negative neighborhood of the parameter, Fig. 11d, for a leading conguration. Thus, the plus
tolerance of tie rod length should be tighter than its minus tolerance.
6. Sensitivity and kinematic optimization
Since the optimized sets of the design parameters based on only kinematic consideration are innitive, as
reported in Section 4, and very sensitive, as reported in Section 5, it is preferable to optimize a function that
takes into account both the kinematics and sensitivity aspects, i.e., providing a kinematically optimized values
that are least sensitive. Such multi-objective optimization can be treated in two ways: (a) according to [15], a
new objective function can be dened as the weighted sum of the two objective functions, namely, the MSE
and sensitivity and (b) one of the objective functions is considered as a constraint, while the second one is treated as the objective function [16]. Here, however, a new approach is proposed to attempt optimization in two
stages. According to this approach, the problem is rstly optimized for one of the objective functions, i.e.,
kinematic optimization, and among all the solutions generated in this stage, the algorithm searches for the
optimum one based on the second objective function, i.e., sensitivity optimization. Therefore, the problem
is started with two design variables, namely, la and h at every stage of sensitivity calculation. The third
one, i.e., lt, is obtained from MSE optimization. The three variables are utilized to calculate sensitivity. For
the problem at hand, the second objective function is dened as
Minimize Sensitivity S w1 wla w2 wlt w3 wh

10a

for which the corresponding constraints are considered as

Lal 6 La 6 Lau ;

and

Hl 6 H 6 Hu

10bc

MSE 6 allowable limit

10d

Minimum allowable value 6 rack stroke 6 maximum allowable value

Maximum pressure angle b 6 Allowable pressure angle

10e
10f

User-defined constraints if any

10g

In Eq. (10a), wla , wlt and wh are the sensitivities of the linkage w.r.t. la, lt, and h, respectively. Since, every
one of these variables has two values, which in positive and negative neighborhoods of the parameter, e.g., la,
the bigger one is chosen as the sensitivity. Moreover, in Eq. (10a), w1, w2, and w3 are the weighting factors.
Furthermore, Lal, Lau, Hl, and Hu are the lower and upper limits of La and H, respectively, and Eq. (10d) constrains the MSE to allowable MSE, Eq. (10e) constrains rack stroke between minimum and maximum allowable values. Next constraint is pressure angle constraint, being important for force transmission eciency. As
reported in [2], between pressure angles dened in Section 3, b2 is usually maximum at the maximum steering
angle, which can be depicted from Fig. 5. Also the maximum pressure angle of about 60 is considered suitable
for good force transmission eciency and to avoid jamming the linkage. This constraint is taken into account
in Eq. (10f). Additional constraints, like space limitation, can also be included, but are not considered here.
Now, the optimization methodology is shown in the owchart of Fig. 12.
A MATLAB code based on the proposed methodology has been written. It has three components, namely,
a main program and two function programs, each of kinematic optimization and sensitivity calculations. Out
of three normalized design variables, i.e., la, lt, and h, two, namely, la and h, are kept constant for MSE optimization and are entered in the kinematic optimization program. As discussed in Section 4 and shown in Figs.
7 and 8, the output of this part is optimum lt for the given constant values of la and h. These three design variables are then utilized to calculate sensitivity of the linkage. Finally, the link lengths of the optimized linkage
are achieved by multiplying la, lt, and h into the normalization factor (WtLr).

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

53

Enter initial parameters

La, Lr, H, Wt , and Wb

Normalize La and H by
(Wt - Lr)

Normalize Wb and Wt
by Wt

Kinematics optimization, eqs.

(8a-8e): Find optimum lt
Sensitivity objective
function calculation

Choose
another set
of design
variables
according to
optimization
method

Are the
constraints satisfied?
and
Is the objective
function minimum?

No

Yes
Find optimum La, Lt , and H by denormalizing
optimum la, lt , and ht , respectively
End
Fig. 12. Combined optimization owchart.

7. A case study
The kinematics and post-optimal sensitivity optimization for the rack-and-pinion steering linkage based on
the concepts outlined above is illustrated with the help of an STO trailing linkage used in a commercial hatchback car. Its actual and normalized link lengths are given in the rst and second columns of Table 1. The actual
values are approximate, as obtained from real model measurement and not from manufacturers blueprint. In
the third column, the kinematically optimized link lengths, the MSE, and the sensitivity of the system with
respect to design variables are given. At this stage, sensitivity is not considered during optimization, i.e., the
optimization is based on methodology presented in Section 4. Note here that the real dimensions of the
STO conguration appearing in the 1st column of Table 1 is rst normalized and shown in the 2nd column.
The resulting linkage is then optimized and denormalized to the optimized STO, whose dimensions appear
in the 3rd column of the table. As it is expected, La and H are unchanged and only Lt is changed to optimum
Lt. The optimum dimensions of the linkage considering both MSE and sensitivity based on methodology presented in Section 6 are shown in the 4th column of the table, where w1, w2, and w3 are considered equal to 0.33.
The limits of Eq. (10bc), namely, Lal, Lau, and others are as also shown in Table 1. To solve this constrained
optimization problem, the method of Multiplier (MOM) [15] is applied to the problem to penalize the constraints, and Simplex direct search method [15] is used to nd the optimum point. In addition, the proper initial
values should be found for the optimization problem to converge to optimum without violating the constraints.
For this purpose, the bracketing method [15] is applied to the objective function over the dened span to
nd suitable starting point. The optimum results show good improvement in sensitivity almost without sacricing the optimized steering error. Finally, the kinematic steering error and sensitivity of the error due to
variation of every linkage parameters, namely, La, Lt, and H, is veried by applying 1% variation of every

54

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

Table 1
Dimensions of a STO steering linkage before and after optimization

Steering arm length

Tie rod length
Distance from rack axis to wheel
axis
Wheel base
Wheel track
Max. error, dhO
Rack length, Lr
Sensitivity, S, of Eq. (10a)
MSE variation (%) due to 1% +ve
MSE variation (%) due to 1% ve
MSE variation (%) due to 1% +ve
MSE variation (%) due to 1% ve
MSE variation (%) due to 1% +ve
MSE variation (%) due to 1% ve

Initial
dimensions

Normalized
dimensions

Kinematic optimum
dimensions

Kinematic and
sensitive
Optimum dimensions

La = 110 mm
Lt = 256 mm
H = 177 mm

la = 0.2048
lt = 0.4767
h = 0.3269

La = 110 mm
Lt = 266.2 mm
H = 177 mm

La = 170.5 mm
Lt = 212.3 mm
H = 15 mm

Wb = 2175 mm
Wt = 1215 mm
3.9

wb = 1.7901
wt = 1

Lavariation
Lavariation
Ltvariation
Ltvariation
H variation
H variation

0.7
678 mm
59.32%
11.39%
27.1%
44.6%
112.5%
40.2%
16.32%

Wb = 2175 mm
Wt = 1215 mm
0.75
28.27%
0.42%
0.48%
34.5%
83.2%
0.91%
1.9%

Allowable Ackermann error = 0.75; Lal = 100 mm; Lau = 300 mm; Hl = 0; Hu = 300 mm; minimum rack stroke = 100 mm; maximum
rack stroke = 140 mm; maximum pressure angle = 56.

parameter in separate stages of kinematic calculation and nd the steering error variation. The results closely
match with those obtained from the optimized results. It proves that the optimization is successful and its
results are reliable.
8. Conclusions
In this paper a combined kinematic and sensitivity optimization of a rack-and-pinion steering linkage are
presented for the rst time. Since the variations of the linkage parameters cannot not be avoided in practice,
attention to the sensitivity analysis and optimization is considered essential. The kinematic optimization of
the steering linkage is carried out using three homogenous design parameters all having the unit of length as
a requirement for post-optimal sensitivity analysis. A simple but generalized methodology based on rackand-pinion steering linkage cognates is proposed to analyze the central take-o congurations. The results
are applicable to side take-o conguration of the steering linkage as well. The analysis of the linkage shows
that the system is very sensitive with respect to variations of the links lengths close to optimum and are dierent
left and right of the optimum point. This has implications in assigning tolerances to the links lengths. Next
observation is that the sensitivities of the steering error with respect to the variations of dierent design parameters vary very widely. Hence, it should be taken into account that some alignments during maintenance,
namely toe angle by changing the tie rod length, can increase the steering error extensively. For optimization
of both the kinematic steering error and sensitivity to link length variation, a two-level multi-objective optimization was proposed. The advantage of the proposed method is that it reduces the number of design variables by
one, thus increasing the stability and eciency of the optimization. The proposed methods were successfully
applied to the rack-and-pinion steering linkage of a real vehicle, and the numerical results provided in the paper.
Appendix A
The analytical method used to solve the nonlinear position problem of the central take-o rack-and-pinion
steering linkage is as follows:
Eqs. (3) and (4) are rst rearranged as
lt ch12 wt =2 b  la ch1

and

lt sh12 h  la sh1

A:1

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

55

Squaring and summing the expressions, in Eq. (A.1) one obtains the following:
2

l2t wt =2 b h2 l2a  wt 2bla ch1  2la hsh1

Now, denoting, k 1 

l2t

l2a

A:2

h , k2  [k1 + (wt/2 + b) ] /la, and

1  z21
1 z21

A:3

k 2 wt 2bz21  4hz1 k 2  wt 2b 0

A:4

z1  tan

h1
;
2

sh1 

2z1
1 z21

and

ch1 

Eq. (A.2) becomes

which is a quadratic equation in z1. The solution, z1, is given by

p
2h  D1
z1
k 2 wt 2b

A:5

where
D1  4h2  k 22 wt 2b2

A:6

For the initial condition where the solution, z1, is denoted as z10, is solved from Eq. (A.5) for b = 0. Thus,
Eqs. (A.5) and (A.6) are converted to:
p
2h  D10
A:7
z10
k 2 wt
in which
D10  4h2  k 22 w2t

A:8

The joint angle, h1, is nally obtained from Eq. (A.5) as

h1 atan2sh1 ; ch1

A:9

where atan2 is the function used in any computer programming language to provide unambiguous angle results to tan1. Similar steps can be followed for other side of the linkage of Fig. 5 to nd h6 = atan2(s h6, c h6)
where s h6 and c h6 are obtained from the value of z6 which is solved as
p
2h  D6
A:10
z6
k 3 wt  2b
where
D6  4h2  k 23 wt  2b

and

k 3  k 1  wt =2  b =la

A:11

Note that + and  signs in Eqs. (A.5), (A.7) and (A.10) correspond to the two dierent assembly positions
of the mechanism, where only + sign is applicable for the steering linkages.
References
[1] J. Reimpell, H. Stoll, J.W. Betzler, The Automotive Chassis: Engineering Principles, second ed., SAE International, Butterworth
Heinemann Publisher, Oxford, 2001.
[2] P.A. Simionescu, M.R. Smith, Initial estimates in the design of rack-and-pinion steering linkages, ASME J. Mech. Des. 122 (2) (2000)
194200.
[3] C.E. Zarak, M.A. Townsend, Optimal design of rack-and-pinion steering linkages, ASME J. Mech. Transm. Automat. Des. 105
(1983) 220226.
[4] M.L. Felzien, D.L. Cronin, Steering error optimization of the McPherson strut automotive front suspension, Mech. Mach. Theory 20
(1) (1985) 1726.
[5] P.A. Simionescu, M.R. Smith, Applications of Watt II function generator cognates, Mech. Mach. Theory 35 (2) (2000) 1535
1549.
[6] H. Han, T. Park, Robust optimal design of multi-body systems, Multibody Syst. Dyn. 11 (2004) 167183.

56

A. Rahmani Hanzaki et al. / Mechanism and Machine Theory 44 (2009) 4256

[7] C. Zang, M.I. Friswell, J.E. Mottershead, A review of robust optimal design and its application in dynamics, Comp. Struct. 83 (4-5)
(2005) 15326.
[8] A. Rahmani Hanzaki, S.K. Saha, P.V.M. Rao, Analysis of a six-bar rack-and-pinion steering linkage, in: Proceedings of SAEINDIA
International Mobility Engineering Congress, 2005, pp. 103108.
[9] O.M.A. Shar, M.R. Smith, A simple method for the allocation of appropriate tolerances and clearances in linkage mechanisms,
Mech. Mach. Theory 18 (2) (1983) 123129.
[10] P.E. Gill, W. Murray, M.H. Wright, Practical Optimization, Academic Press, London, 1988, pp. 320-323.
[11] L.F. Knappe, A technique for analyzing mechanism tolerances, Mach. Des. (1963) 155157.
[12] J. Chakraborty, Synthesis of mechanical error in linkages, Mech. Mach. Theory 10 (1975) 155165.
[13] S.G. Dhande, J. Chakraborty, Analysis and synthesis of mechanical error in linkages a stochastic approach, ASME J. Eng. Ind. 95
(3) (1973) 672676.
[14] S. Yakowitz, F. Szidarovszky, An Introduction to Numerical Computations, Macmillan Publishers, New York, 1990, pp. 214216.
[15] K. Deb, Optimization for Engineering Design, Algorithms and Examples, Prentice-Hall of India, New Delhi, 1995, pp. 162163.
[16] A.B. Koteswara Rao, Analysis and Design of Hexaslide Manipulators for Machine Tools Application, Ph.D. Thesis, Mechanical
Engineering Department, Indian Institute of Technology Delhi, 2004, pp. 6470.