Optimization of Pitch Sequencing For Pneumatic Tire by Tabu Search Method
Optimization of Pitch Sequencing For Pneumatic Tire by Tabu Search Method
Optimization of Pitch Sequencing For Pneumatic Tire by Tabu Search Method
The reduction of tire noise has been an important research field in tire industry. The
sequence design of tire tread blocks with different pitches is the primary way to reduce
the objectionable noise level.
In this paper, the optimization of pitch sequence around the tire circumference by
tabu search has been implemented so that the maximum noise index over all harmonics is
minimized. The pitch sequence of two tires has been examined. The result has shown that
the maximum noise index of harmonics and the magnitude of power spectrum density of
tire noise can be reduced remarkably.
1. Introduction
The tread of vehicle tire comprises plurality of load bearing tread block ele-
ments around the circumference of the tire. When the tire is rolling, the tread block
engages with a pavement and causes the disturbance of surrounding air molecules
that generates noise, which is called as "tire noise". If the pitch length of each tire
tread block is constant, the tire noise is dominated by a single frequency and its
harmonics, i.e. the majority of the tire noise is concentrated in a small frequency
range of the audio frequency spectrum with high noise level. Pitch sequencing is a
method to arrange the tread blocks of tire in various pitch sizes, so that the noise
frequencies produced by tread blocks will be distributed over a wide band, and the
noise will be reduced to a small level. The pitch sequencing for tire design has been
a primary method to reduce subjective tire noise for 50 years since Buddenhagen
[1] was awarded a US patent for noiseless tread. Williams [2], Nakajima [3] and
Zakelj [4, 5] have developed different kinds of patent for pitch sequences to reduce
the tire noise.
Williams [6] mentioned that the basic parameters of pitch arrangement for tire
design are: the total number of pitches, the total number of different pitch types,
the pitch size of each type, the number of pitches for each type, and the sequence
of pitches. The last one is the most important parameter for tire design. Nakajuna
and Abe [7] have applied genetic algorithms to optimize the pitch sequence based
400 J.T. CHID, W.C. WENG and C.F. HUNG
on two types of objective functions: the spectrum peak and the spectrum deviation.
Hoffmeister and Bernard [8] examined the application of GA on the optimization
of pitch arrangement problems. These results showed that the noise was reduced
remarkably by optimization of pitch arrangement.
The arrangement of the pitch sequence is an order optimization problem, which
is similar to a job sequencing in scheduling problems. Each pitch number is like a
job, which is not preemptive and has no precedents in a scheduling problem.
In 1986, Glover firstly proposed the tabu search method (TS in short) [9],
which is based on hill-climbing heuristic method [10-11]. After that, TS, one of local
search methods, has been applied to many combinatorial problems successfully, such
as scheduling, planning and other practical problems extensively. Some research
works have shown the excellent performance of tabu search for sequencing problems
[12, 13].
TS performs the search not from point to point but from subset to subset of
solution space sequentially. The subset is termed as a neighborhood, which is a
point set of solutions generated from a reference point by a rule or an algorithm
that is called as the move function. TS starts the search from an initial reference
point. And the best solution ever found is the result after several neighborhood
searches. In each search iteration the neighborhood is examined. The best point in
the neighborhood will become the reference point of the next iteration. A tabu list,
which records the information of move functions relating to the best points found
in search iterations, plays an important role in TS for preventing the cyclic search.
In this paper, the tabu search method with back jump mechanism proposed by
Nowicki and Smutnicki [14, 15] is adopted to find the optimum pitch sequence where
the objective function is to minimize the maximum noise index over all harmonics
of the tire.
Figure 1 shows the tread pattern and different pitch size of a tire. A tire has
n p types of tread blocks with various pitch length, i.e. the tire has nr types of pitch,
and the pitch length of type i is )i. The pitch type i is numbered from 1 to n,
according to pitch length from short to long, i.e. Al <. X2 < . • . < ) np .
The summation of the pitch length of all tread blocks should be equal to the
tire circumference L
np
L = E PZ x \i (1)
i =1
ny
N =EP i (2)
Z —i
When the tire is rolling on the road surface, the noise generated by i -th pitch
Optimization of Pitch Sequencing for Pneumatic Tire 401
I1 (^ Type 3
Type 2
Type 1
f(t)=O t < ti
f2 (t) = c • e - i t -t ji sin(ß(t - ti)) t > ti (3)
Where c, ß and c are the parameters determined by curve fitting from the measured
noise [7]. ti is the time when the i-th pitch engages with ground. Parameter a is a
large number and equation (3) can be approximated by an impulse function
Where M is a selected large number to keep the accuracy of the function approx-
imation. The fundamental period is defined as T = L , where V is the car speed.
And then the fundamental frequency is wo = T . The coefficient aim and bi ra can
be expressed by the following equations:
I
T 6(t_
ai m = T ti) cos mwot dt = T cos mwoti
(7)
The total noise level generated by all pitches within a fundamental period can be
expressed as follows
N N M
F(t) = E fi (t) = E E (ai s,,, cos mwo t + b i , sin mwot) (8)
i—i i =1 m =i
The component of the m-th harmonic in equation (8) can be expressed as:
N
D„ (t) = (ai.m, cos n 0t + bi„ s in ~ot) (9)
i =1
N 2 N 21/2
Dm 1 = « N aim ^ + (bi m ) )
2 2 1 ^ 2(10)
1 cos mwoti)
_sin mwoti) + (
ii
N
msi 2 N
L s2 )
_
)
Difference in type number between two adjacent pitches < n,/2 (13)
Optimization problem
The optimization problem of pitch sequence for tire design is defined as follows:
1. The number of total pitches around the tire circumference N, the number
of pitch types up , the quantities of each pitch type { ql, q2, ... , q n , }, and the
pitch size of each type {A1, \2, ... , b np } are given and defined as constant.
2. The design variable is the pitch sequence of tire.
z-1
si = E A Zj , for i = 2 to N; and si = 0, for i = 1 (15)
j =1
3. The objective function is the maximum noise index for all harmonic numbers:
zl — ZNI < np /2
(17)
zi — z2_1 <n p / 2; i =2,3,...,N
1 <zti<n p ; i = 1 to N (18)
N
L=EA, (19)
2-1
In the solution space X, each point Z is a feasible solution. Unlike the con-
ventional optimization techniques, tabu search does not perform the search from
point to point, but examines the best point from one point set to another point set.
This point set is called a neighborhood, which is generated from a reference point
x following a rule or an algorithm, called as move function. A search iteration is
an exploration of the best point over a neighborhood. The best solution x' of a
neighborhood is called local best point, which will become the reference point of
next search iteration. To prevent the cyclic search a tabu list is introduced. All
points specified by the records in tabu list will be excluded from the neighborhood
for each search iteration. A variable o, termed as overall best point, memorizes
the best result ever found by tabu search. The variable, o,, will be updated by the
local best solution only when it is better than o d,. Until one of the stop criteria is
satisfied, the tabu search is ended and the final o,, is the answer.
At the beginning, some parameters for tabu search has to be initialized: an
empty tabu list T and the number of records in T, represented by LengthT; a start
point x o to be used as the reference point for the first iteration, and the overall best
point o,, which is set to be x 0 . In each iteration, the following four components
will be updated: the reference point x, the neighborhood S(x, T), the tabu list T
and the overall best point o x .
(a) i < j
40
el e2 e3 ... e;_1 ei e; ei+l ... ez -1 ei+l ...... eN
That means m(x, 1, j) can generate n — 1 new sequences and the other m(x, i, j)
can generate n — 2 new sequences from x. Therefore the size of M(x) is (N — 1) 2 .
For example, a reference point x = {5, 3, 2, 4, 1} is composed of 5 elements. The
move function m(x, 3, 4) generates the new sequence {5, 3, 4, 2, 1} by moving the
third element onto the fourth position. On the other hand, function m(x, 3,2) is
not allowed to use, because it will generate as the same point as m(x, 2, 3) does.
Each record in tabu list contains a pair of data, an element and a position
number. A record, (z, J), in tabu list specifies the points whose J -th element is
the pitch type z. If the reference point x of the current iteration is {3, 1, 2, 3, 2},
for example, a record (3,2) in tabu list specifies that all points have pitch type 3
on position 2, e.g. f 3, 3, 1, 2, 2} and f 1, 3, 2, 3, 2}. A subset M'(x, T) where each
point specified in tabu list will be excluded from the current neighborhood, i.e.
S(x, T) = M(x) — M'(x, T).
As long as the neighborhood has been well defined, the examination of local
best point x' can be processed. When the local best point x', which is generated
by m(x, u, v), is found, the new record (z, u) is appended to the tabu list, where
Optimization of Pitch Sequencing for Pneumatic Tire 405
zti is the element on position u in current x. If the list is full, the oldest record will
be deleted before appending the new record. If the objective function value of x' is
better than that of o., the global best point will be updated, i.e. o. = x'.
The search is stopped when one of stop criteria is satisfied, and then the final
overall best result is the answer.
Stop criteria
The search will be stopped when all of the back jumps specified in history list
have been run out, or the number of total iterations reaches the allowed maximum
number, represented as MaxlterTot.
Tabu search, in sum, performs the search in a way of examining one neigh-
borhood to another neighborhood by moving from one reference point to another
reference point respectively. All reference points will form a search track sequen-
tially. The algorithm is briefly shown below:
1. Set search parameters:
The allowed total iteration number: MaxIterTot
The iteration number threshold to activate back jump: MaxlterBJ
The number of records of tabu list: LengthT
The number of records of history list: LengthH
Select an initial reference point x,, E X, and let o,, := x 0 .
2. Set up the point set M(x) and the neighborhood S(x, T) = M(x) — M'(x, T)
3. If S(x, T) is empty, then stop.
Otherwise process neighborhood search
a. Set k := k + 1 and select Z E S(x, T) such that
x' = optimum(Z: Z E S(x, T)).
b. Append the record (z,l , u) to T for x' := m(x, u, v)
4. Let x := x' and kh := kh + 1.
5. Check global best
If C(x') < C(o) then
Let o,, := x' and kh := 0
x" = the second best point in S(x, T)
Append the record (T, x") to H
Otherwise go to Step 6
6. Check back jump
If kh = MaxIterBJ, then start back jump mechanism:
If H = 0, then stop
Otherwise
The last record in H is restored and then removed from H.
x and T are redefined.
Set kh := 0
Otherwise, go to Step 7.
7. If k < MaxIterTot, then go to Step 2.
Otherwise, stop.
5. Numerical Examples
The optimizations of pitch sequences to reduce the noise index for two tires
are selected as case studies. Case 1 is a real tire P155/80R13 with circumference
1813.02 mm as shown in Figure 1, case 2 is referred to the pitch arrangement of a
Goodyear patent [16] with same circumference. The pitch sizes of the tread patterns
of two tires are shown in Table 1 and Table 2, respectively. The searched best result
of case 2 will be compared with that of genetic algorithms in Reference [8].
Table 2. The pitch size of tread patterns for the Goodyear patent [16].
The optimization of pitch sequence has been performed 10 trials for both cases
with different initial solutions. The parameters for tabu search are set as follows.
In general, the MaxIterTot needs not greater than 300. For test purpose, here
a huge number is given so that tabu search will be continued to search wide do-
main until all of records in the history list have been restored and removed. The
LengthT should be not too long for preventing some potential partial sequence
being excluded.
Table 3 and 4 are the searched results of two cases respectively.
The maximum noise index of the best solution searched for case 1, which is
located on the second row with an asterisk in Table 3, is C(o) = 16.65. And the
best sequence of pitch type is shown as follows:
The distribution of noise index versus harmonic number and the power spectrum
density of tire noise are shown in Figure 3 and 4, respectively.
The pitch arrangement of the Goodyear patent has the maximum noise index
of 21.50. The quantities of each pitch type are q l = 18, q2 = 27 and q3 = 18 and
the pitch sequence is as follows:
The result of the best solution searched for case 2, locating at the 7th row in
Table 4, is summarized as follows:
Optimization of Pitch Sequencing for Pneumatic Tire 409
30.00
20.00
:r.
q
____________________________________l^f^l________________________
10.00
Il ll^^lilllllII I
0.00 ^r-^^ ^^ I
0 20 40 60 80 100 120 140 160 180
harmonic number
Fig. 3. Distribution of noise index versus harmonic number (case 1, 2nd trial).
E5
2
N
a.4
U)
Fig. 4. Distribution of PSD of tire noise of the 2nd trial for case 1.
410 J.T. CHID, W.C. WENG and C.F. HUNG
On the other hand, the best sequence for case 2 searched by GA [8] is as follows:
Where the quantities of each pitch types are ql = 29, Q2 = 10 and q3 = 24, which
are different from that of Goodyear patent.
30.00
II
20.00
II
Ili
M II
ór. 'I'll
,^^di^ 11111
^ r^`uI EI!t I1Ii!1i1I I JIII
13^I^tl ! ^M^I^^I^I^^ilpl l'^^l^Il l^lldilI^^fhl
10.00
Fig. 5. The comparison of noise index between the Goodyear patent [16] and the
results of GA [8] and TSBJ.
Optimization of Pitch Sequencing for Pneumatic Tire 411
is little better than that by GA (C(o,,) = 15.3). It can be expected, if the pitch
ratio and the quantity of each pitch type are also included in design variables, the
searched results will be improved a little. It will be the main work for next paper
in advance.
From the results of two study cases, the tabu search method has improved the
initial solution at least 20% and up to 50%. The last rows in Table 3 and 4 show
the mean value of maximum noise index of initial and final solutions, the average
reduction of noise index is above 30%. To find the final results, it needs wider
search, and may take much time. The optimization has been carried out by personal
computer with CPU Pentium I1I-500. The results show that the computing time
is still acceptable. Most of the optimization run is stopped before 250 iterations.
For some trials, it took much search time and the found result may be not so
good as other trials. For some other trials, the search efficiencies are good both
in result and search time. It shows that the initial solution determines the search
track, which connects all reference points in whole search procedure. The initial
solution affects the search result and search time strongly. Moreover, a better initial
solution cannot promise to find a better result. The search time depends on the
problem size and the initial solution. The larger problem size needs longer search
time.
For the purpose of comparison, the tabu search without back jump mechanism
was performed with the same initial point of the second trial for case 1 shown
in Table 3. The allowed total search iteration is set MaxIterTot = 200. The
found best solution has maximum noise index of 17.52. The search histories of the
maximum noise index found by the tabu search with and without back jump are
shown in Figure 6. The result shows that tabu search found a good result after 20
search iterations. Nevertheless, the back jump mechanism provides a wider domain
search to improve the final result further. Figure 6 shows that the overall best point
has not been improved after 60 search iterations.
In general, the overall best point will be hardly improved after some decades of
iterations or after several back jumps. For practical application, the length of tabu
list (LengthT) and MaxlterBJ can be shorter than those given in study cases.
The move function determines the neighborhood of a reference point, and af-
fects the search result in advance. Generally speaking, a large size of neighborhood
will enable the search to examine wider domain. On the contrary, a small neigh-
borhood will constrain the search domain and land the result in a local optimum
point. There are three kinds of move function for dealing with sequencing problems
412 J.T. CHtu, W.C. WENG and C.F. HUNG
28.00
24.00
K
-c^
6
20.00
16.00
0
I
40
IH 80 120 160 200
iteration no.
Fig. 6. The search history of the local best point and overall best point in TS and
TSBJ (case 1, 2nd trial).
in common: to exchange any two adjacent elements, to exchange any two speci-
fied non-adjacent elements and to insert one specific element from old position to
new position. For a sequence with N elements, the size of neighborhood generated
by the three kinds of move functions from the same reference point are (N — 1),
(N — 1)(N — 2)/2 and (N — 1)*(N — 1), respectively. The third one is chosen in
this paper.
The flip-flop algorithm in Reference [7] belongs to second kind of move function.
A test run of flip-flop algorithm for case 1 was performed. The search was started
from the same initial solution of the second trial in Table 3. The search was
ended after 3031 flip or flop operations due to the cyclic search control. The found
maximum noise index 19.93 is a local optimum, which is worse than that of all
trials in Table 3.
6. Conclusions
In this paper the tabu search method has been successfully applied to the
optimization of pitch sequencing problem to reduce the tire noise. For tire design
with large number of tread blocks and plenty different patterns, the optimization
Optimization of Pitch Sequencing for Pneumatic Tire 413
References
[ 1 ] F.E. Buddenhagen, Tire casing with noiseless tread. United States Patent Document, No.
26122928, October, 1952.
[ 2 ] T.A. Williams, Low noise tire tread. United States Patent Document, No. 5759311, Jun. 2,
1998.
[ 3 ] Y. Nakajima, Method of determining a pitch arrangement of a tire. United States Patent
Document, No. 5717613, Feb. 10, 1998.
[ 4 ] P. Zakelj and J.L. Turner, Tire noise treatment. United States Patent Document,
No. 5865921, Feb. 2, 1999.
[ 5 ] P. Zakelj, Tire tread noise treatment. United States Patent Document, No. 6112167, Aug.
29, 2000.
[ 6 ] T.A. Williams, Tire tread pattern noise reduction through the application of pitch sequenc-
ing. Noise and Vibration Conference 1995, Travers, Michigan, Issue 291, Vol.2, 1995.
[ 7 ] Y. Nakajima and A. Abe, Application of genetic algorithms for optimization of tire pitch
sequences. Japan J. Indust. App!. Math., 17 (2000), 404-426.
[ 8 ] K.M. Hoffmeister and J.E. Bernard, Tread pitch arrangement optimization through the use
of a genetic algorithm. Tire Sci. & Technol., TSTCA, 26, No.1 (1998), 2-22.
[ 9 ] F. Glover, Future paths for integer programming and links to artificial intelligence. Com-
puters Opns. Res., 13, No.5 (1986), 533-549.
[10] F. Glover, Tabu search Part I. ORSA J. Comput., 1 (1989), 190-206.
[11] F. Glover, Tabu search Part II. ORSA J. Comput., 2 (1990), 4-32.
[12] M. Sinclair, Comparison of the performance of modern heuristics for combinatorial opti-
mization on real data. Computers Opns. Res., 20, No.7 (1993), 687-695.
[13] R.J.M. Vaessens, E.H.L. Aarts and J.K. Lenstra, Job shop scheduling by local search. IN-
FORMS J. Comput., 8, No.3 (1996), 302-317.
[14] E. Nowicki and C. Smutnicki, A fast taboo search algorithm for the flow shop. Eur. J. Oper.
Res., 91, No.1 (1996), 160-175.
[15] E. Nowicki and C. Smutnicki, A fast taboo search algorithm for the job shop. Management
Sci., 42, No.6 (1996), 797-813.
[16] S.P. Landers, Spreading noise generated by load supporting elements. United State Patent
Document, No. 4327792, The Goodyear Tire and Rubber Company, May 4, 1982.
414 J.T. CHID, W.C. WENG and C.F. HUNG