Professional Documents
Culture Documents
net/publication/232809257
Article in ARCHIVE Proceedings of the Institution of Mechanical Engineers Part C Journal of Mechanical Engineering Science 1989-1996 (vols 203-210) · March 2009
DOI: 10.1243/09544062JMES1149
CITATIONS READS
9 23,396
3 authors, including:
All content following this page was uploaded by Xiaoqiang Du on 23 May 2014.
The manuscript was received on 21 April 2008 and was accepted after revision for publication on 28 July 2008.
DOI: 10.1243/09544062JMES1149
Abstract: To decrease the vibration and improve the dynamic performance of traction-type
passenger elevators, an accurate vertical dynamic model with nine degrees of freedom for a gear-
less 2 : 1 traction-type passenger elevator system is presented. Then, an optimization model of
dynamic parameters for this type of passenger elevator is proposed based on the dynamic model.
The optimization model takes the amplitude of vibration acceleration response of the elevator
cage as the objective function. Various exciting forces and various working conditions can be
considered by using corresponding weight coefficients in the objective function. To get higher
efficiency and stronger robustness of optimized value, a dynamic byte coding genetic algorithm
in solving the optimization model by combining the advantages of binary coding method with
dynamic parameter encoding method is proposed, and the optimal solution is verified by sen-
sitivity analysis. A practical engineering optimization for a 2 : 1 traction-type passenger elevator
system shows that the optimization model and method of dynamic parameters proposed in this
article are effective.
Keywords: traction-type elevator, ride quality, dynamic parameter optimization, dynamic byte
coding genetic algorithm, sensitivity verification
JMES1149 © IMechE 2009 Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science
596 D Mei, X Du, and Z Chen
where x is the system displacement vector, x = [x1 , x2 , M, C, and K are the system mass, damping, and
x3 , x4 , x5 , x6 , θ1 , θ2 , θ3 ]T , q the system loading vector, and stiffness matrices, respectively, given by
M = diag[m1 , m2 , m3 , m4 , m5 , m6 , I1 , I2 , I3 ]
⎡ ⎤
c0 + c 1 + c 2 −c1 −c2 0 0 0 (c1 − c2 )r1 c 1 r2 −c2 r3
⎢ c1 + ce1 + c6 0 0 0 −c6 −c1 r1 (−c1 + ce1 )r2 0 ⎥
⎢ c 2 + ce2 + c4 −c4 (c2 − ce2 )r3 ⎥
⎢ 0 0 c1 r1 0 ⎥
⎢ c4 + c5 −c5 ⎥
⎢ 0 0 0 0 ⎥
⎢ c5 0 0 0 0 ⎥
C=⎢ ⎥
⎢ Symmetry c6 0 0 0 ⎥
⎢ ⎥
⎢ (c1 + c2 )r12 + ct c1 r 1 r 2 c 2 r1 r3 ⎥
⎢ ⎥
⎣ (c1 + ce1 )r22 0 ⎦
(c2 + ce2 )r32
⎡ ⎤
k 0 + k1 + k2 −k1 −k2 0 0 0 (k1 − k2 )r1 k1 r2 −k2 r3
⎢ k1 + ke1 + k6 0 0 0 −k6 −k1 r1 (−k1 + ke1 )r2 0 ⎥
⎢ k2 + ke2 + k4 −k4 (k2 − ke2 )r3⎥
⎢ 0 0 k2 r1 0 ⎥
⎢ k4 + k5 −k5 ⎥
⎢ 0 0 0 0 ⎥
⎢ k5 0 0 0 0 ⎥
K=⎢ ⎥
⎢ Symmetry k6 0 0 0 ⎥
⎢ ⎥
⎢ (k1 + k2 )r12 + kt k1 r1 r2 k 2 r1 r3 ⎥
⎢ ⎥
⎣ (k1 + ke1 )r22 0 ⎦
(k2 + ke2 )r32
where ke1 = (k1 k3 )/(k1 + k3 ), ce1 = (c1 c3 )/(c1 + c3 ), 2.2 Optimization model of dynamic parameters
ke2 = (k2 k3 )/(k2 + k3 ), ce2 = (c2 c3 )/(c2 + c3 ), and kt and for a 2 : 1 traction-type passenger elevator with
ct , respectively, denote the rotational stiffness and multiple exciting forces
rotation damping of the driving sheave.
Several significant factors that contribute to the ride
quality of a traction-type passenger elevator can
be sorted mainly into functional parameters, noise
parameters, and flexible dynamic parameters [6].
Compared with other parameters, flexible dynamic
parameters contribute much more to its dynamic
performance and ride quality. They can be changed
in a large range to modify the dynamic characteris-
tics of an elevator system. These parameters include
the stiffness and damping of the rubber under trac-
tion machine and cage, the thimble rod spring, the
car spring, and so on. They may be continuous or
discrete according to practical design requirements.
Thus, the flexible dynamic parameters can be chosen
as the design variables for the optimization of dynamic
performance for a 2 : 1 traction-type passenger
elevator.
Due to the harmonic property of exciting forces of
vertical vibration mostly generated by the asymme-
try of the driving sheave, guide sheave, counterweight
sheave, and so on, the vibration response of the cage
is strongly harmonic, which is obvious in practical
measurements. Otherwise, vibration acceleration is
generally adopted as the criterion for evaluating the
dynamic behaviour of elevator products. Therefore,
the amplitude of the vibration acceleration response
is appropriate to be selected as the objective function
Fig. 3 A nine-DOF vertical dynamic model for a 2 : 1 of the optimization of dynamic parameters for a 2 : 1
traction-type passenger elevator traction-type passenger elevator.
JMES1149 © IMechE 2009 Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science
598 D Mei, X Du, and Z Chen
Based on equation (1), the optimization model of ±10 per cent range to evaluate the deviation of the
dynamic parameters for a 2 : 1 traction-type passenger objective function. The final optimal solution should
elevator with single exciting force can be established have small response amplitude, be insensitive to every
as design variable’s small change, and can satisfy any
practical requirements.
min RSI (u, q) u = {u1 , u2 , . . . , un }T
s.t. ui ∈ {ui,1 , ui,2 , . . . , ui,ki } i = 1, 2, . . . , n1 (2) 3 OPTIMIZATION METHOD USING DYNAMIC
uj,min uj uj,max j = n1 + 1, . . . , n BYTE CODING GA
where RSI (u, q) is the optimization object, and the 3.1 Dynamic byte coding GA
subscript ‘SI’ the amplitude of vibration acceleration
GA, first proposed by Holland [7], is a stochas-
response of the elevator cage excited with a single
tic searching algorithm based on Darwinian models
exciting force; q is the exciting forces vector, in which
of natural selection and evolution. The algorithm
only one element is non-zero; u = {u1 , u2 , . . . , un }T the
encodes a potential solution to a specific problem on a
design variables (flexible dynamic parameters) vec-
simple genome-like data structure and applies recom-
tor, the first n1 components of which are discrete
bination operators to these structures so as to preserve
variables, and the others are continuous variables;
critical information. The operators mainly include
{ui,1 , ui,2 , . . . , ui,ki } is a set composed of ki optional val-
selection, crossover, and mutation, which are called
ues of the discrete variable ui ; and uj,min and uj,max the
as genetic operators. By simulating the processes
lower and upper limits of the continuous variable uj .
of natural selection and mutation in the biological
In practice, several exciting forces usually have sig-
evolution, GA provides a new thought for optimiz-
nificant effect on the vibration of elevator cage, and
ing large-scale and complex real-world problems that
the frequencies of them are distinct from each other. In
are difficult for traditional mathematical method to
these cases, equation 2 cannot precisely describe the
solve [8].
dynamic behaviour of traction-type elevators. Thus,
Encoding is a primary issue in GA. The choice of
the weighted root-mean-square value of all the ampli-
encoding method determines not only the genome
tudes of vibration acceleration response of each single
form of individuals, but also the transformation of
force is taken as the optimization object, and the
individuals from genotype in the search space to
optimization model of dynamic parameters with mul-
representation code in the solution space. It also deter-
tiple exciting forces for a 2 : 1 traction-type passenger
mines the genetic operations of the GA population
elevator can be established as
and the efficiency of GA to a great extent. At present,
there are mainly four types of encoding mechanisms:
min RMU = [wfi Rf2i ,SI (u, q fi )], binary coding, Gray coding, floating-point encoding,
and dynamic parameter encoding [9, 10]. However,
u = {u1 , u2 , . . . , un }T these encoding methods have their own merits and
s.t. ui ∈ {ui,1 , ui,2 , . . . , ui,ki } i = 1, 2, . . . , n1 (3) demerits, such as Hamming cliffs in binary coding and
low global search ability in floating-point encoding.
uj,min uj uj,max j = n1 + 1, . . . , n
Therefore, each of them has limited scope of applica-
tion. To get higher efficiency and stronger robustness
where RMU is the optimization object, the subscript
of optimized value, a new encoding GA optimiza-
‘MU’ the equivalent amplitude of vibration acceler-
tion method (dynamic byte coding GA) was proposed
ation response of elevator cage excited by multiple
in this research, which combines the advantages of
forces, wfi the weight coefficient of the exciting force fi ,
binary coding method and dynamic parameter encod-
Rfi ,SI the amplitude of vibration acceleration response
ing method.
of the exciting force fi , and in exciting forces vector q fi ,
Binary coding method has strong global search abil-
only the exciting force fi is non-zero.
ity, high efficiency, and strong robustness of optimized
Taking advantage of some numerical methods, the
value, whereas dynamic parameter encoding method
optimization model of dynamic parameters in equa-
has good local constringency. Dynamic byte coding
tions 2 or 3 can be solved. However, it is impossible
GA is based on the combination of these two encod-
to adjust the practical dynamic parameters that are
ing methods and its basic mechanism is introduced as
precisely equal to the optimal results. For a feasible
follows.
solution, the objective function should be insensi-
tive to each design variable’s deviation in a small 1. Each design variable is represented by an unsigned
scope, so that it can be verified by sensitivity analy- byte integer (eight bits), whether it is a discrete
sis. In this research, after a series of optimal solutions variable or a continuous one. This representation
were searched, their sensitivity analyses were also con- is another kind of binary coding method, which
ducted by letting every design variable change in a essentially adopts an eight-bit integer to represent
Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science JMES1149 © IMechE 2009
Dynamic parameters for a traction-type passenger elevator 599
the gene since the gene with such coding length adjustment of the search space are critical in
has high convergent precision compared to that in dynamic parameter encoding. Whitley et al. [12]
dynamic parameter encoding [11]. have proposed a method to judge the population
2. With unsigned byte genetic representations, the maturity according to the average number of dif-
crossover operator and mutation operator are ferent genes in all individuals, which is obtained
implemented by the rule of logic operation bit by by comparing the genome between two individ-
bit. Therefore, the efficiency is higher than that uals and counting the amount of different genes
of normal binary coding method that uses single they have. If the average number is smaller than
integer to represent the gene. a certain threshold value, the population should
3. Adjustment strategy of search space: the continuous be considered mature and the search space of
evolution of GA population will drive the popula- the design variable should be adjusted. However,
tion close to maturation. To improve the convergent it will take much computing time to judge the
precision, the search space of parameters should be population maturity. Suppose there are n indi-
adjusted. As for unsigned byte integer encoding, the viduals in the population, each individual has b
decoding method of single gene is given by genes (design variables) and each gene is rep-
resented by a p-bit string, then the total logic
g ·h operation time for the judgment of the population
y =o+ (4) maturity is n(n − 1)bp/2. For unsigned byte inte-
255
ger encoding, there are 256 × 256 = 65 536 kinds
where y denotes the parameter in the search space, of possible combinations. It is applicable to build-
g the gene parameter corresponding to the param- up a look-up table before the iterative evolution
eter y, o the minimum value of the search space of of the population for the comparison of genes.
the parameter, and h the range of the search space. In this way, the computing time will decrease
The search space of each parameter may resize remarkably.
based on the solution, which has the maximum fit-
ness in the asymptotically mature population and
is set as the centre of the new search space. When 3.2 Procedures to solve the optimization model of
the offset distance of one parameter relative to dynamic parameters
the current range centre is estimated, its search Using dynamic byte encoding GA, the optimiza-
region can be adjusted dynamically according to tion model of the elevator’s dynamic parameters for
the magnitude of the distance. The offset distance is traction-type elevator in equations 2 or 3 can be
defined as solved; its flow chart is shown in Fig. 4 and its detailed
solving procedures are as follows.
d = |g − 127.5| (5)
Step 1: preparation. Initialize the search space of
each parameter. Establish a look-up table for gene
and the calculation of the new search space is comparison and the globally optimum fitness is
described as set as a negative with sufficiently large absolute
value.
αhc Step 2: initialization and evaluation. N individuals of
o = max ymin , y −
2 initial population are generated at random. Each indi-
vidual’s fitness is calculated according to the selected
αhc
h = min y + , ymax − o (6) cost function.
2 Step 3: resize. If the population is close to mature,
resize the search space and generate new population
where ymin and ymax are assumed to be the minimum at random. Then, the evaluation is executed again.
and maximum value of the original search space of Step 4: selection. The individual with high fitness has
the parameter, respectively, hc the range of the cur- more chance to be selected.
rent search space, and α the adjustment coefficient Step 5: crossover. Each selected individual does
that is defined as crossover operation.
Step 6: mutation. Some digits of certain individuals in
1.1d population mutate by adding a Gaussian stochastic
α =1+ − 0.9 · rand(0.9, 1.1) (7)
127.5 variable.
Step 7 : evaluation. Each individual’s fitness is calcu-
where the function rand(0.9, 1.1) will return a lated and compared to convergence criteria of the
random number between 0.9 and 1.1. optimization. If the convergence criteria are satisfied,
4. The judgment of the population maturity: the judg- then stop the iterative process. Otherwise, go to Step3.
ment of the population maturity and the dynamical Step 8: end of optimization process.
JMES1149 © IMechE 2009 Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science
600 D Mei, X Du, and Z Chen
g z
d
yd = P (int) (8)
255.0 4 AN APPLICATION EXAMPLE
where yd denotes the discrete parameter, gd the The optimization model of dynamic parameters for
gene integer representation corresponding to the traction-type passenger elevator and optimization
discrete parameter yd , and P an array represent- method by using dynamic byte coding GA pro-
ing the definition domain of the discrete parameter, posed in this article has been applied success-
which has z elements. fully in some practical engineering problems. To
2. Design of the fitness function: directly take the demonstrate the effectiveness, an application exam-
inverse value of the objective function as the fitness ple that was finished in 2004 can be introduced as
function. follows.
Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science JMES1149 © IMechE 2009
Dynamic parameters for a traction-type passenger elevator 601
Table 1 Weights of different working conditions Table 3 Optimal solution of dynamic parameters
Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science JMES1149 © IMechE 2009
Dynamic parameters for a traction-type passenger elevator 603
Fig. 7 Practical measurements of the elevator after optimization: (a) measurements of Z -direction
vibration acceleration after optimization, (b) FFT of Z -direction vibration acceleration
(section B) after optimization
difference between theoretical value and practical validate the optimization model and method of
value. Otherwise, in the modelling process, although dynamic parameters proposed in this article. As for
most components and elements of the traction-type the theoretical calculation result, the theoretical opti-
passenger elevator have been considered, some com- mal result of the vibration response is 12.40 mm/s2 ,
ponents and elements are still simplified or neglected which is much less than 72.61 mm/s2 – the result
since a traction-type passenger elevator is a very calculated with the original parameters. As for the test-
complex system containing both mechanical and elec- ing result, the peak–peak value of vertical vibration
trical elements. Another very important reason is the acceleration of elevator car was decreased from 4.45 ×
installation of the elevator system; in some cases, the 10−1 to 1.55 × 10−1 m/s2 after optimization. Therefore,
dynamic behaviour of some same type elevators are both the theoretical result and testing result show
very different, but the installation errors of the eleva- that the optimization model and method of dynamic
tor system are also difficult to be described precisely parameters proposed in this article are effective.
in the modelling process. Due to the above causes, the
theoretical optimal result of the vibration response,
12.40 mm/s2 , is substantially different from the testing 5 CONCLUSIONS
result, 1.55 × 10−1 m/s2 .
On the contrary, in this research, actually the To suppress the vibration and improve the dynamic
difference between theoretical and testing results performance of traction-type passenger elevators,
is not very pivotal because both can, respectively, an accurate vertical dynamic model with nine-DOF
JMES1149 © IMechE 2009 Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science
604 D Mei, X Du, and Z Chen
for a 2 : 1 traction-type passenger elevator system 2 Gang, X. Y., Mei, D. Q., and Chen, Z. C. A horizontal vibra-
is presented. Then, a new optimization model of tion wave model for high speed elevators. Key Eng. Mater.,
dynamic parameters for this type of passenger ele- 2005, 297–300, 1585–1591.
vator is proposed based on the dynamic model. The 3 Rildova. Seismic performance of rail-counterweight sys-
tem of elevator in buildings. PhD Thesis, Virginia Poly-
optimization model takes the amplitude of vibra-
technic Institute and State University, 2004.
tion acceleration response of the elevator cage as an
4 Nai, K., Forsythe, W., and Goodall, R. M. Improving ride
objective function.Various exciting forces and working quality in high-speed elevators. Elevator World, 1997, 45,
conditions can be considered by using correspond- 88–93.
ing weight coefficients in objective function. To get 5 Nai, K., Forsythe, W., and Goodall, R. M. Modeling and
higher efficiency and stronger robustness of opti- simulation of a lift system. In Proceedings of the Interna-
mized value, a dynamic byte coding GA was proposed tional Conference on Control, Modeling, Computation
in solving the optimization model by combining the and Information, Institute of Maths and its applications,
advantages of binary coding method with dynamic Manchester, September 1992, pp. 6–11.
parameter encoding method, and the optimal solution 6 Lorsbach, G. P. Analysis of elevator ride quality, vibration.
is also verified by sensitivity analysis. The optimiza- Elevator world, 2003, 51, 108–111.
7 Holland, J. H. Adaptation in nature and artificial system,
tion model and method was used in a gearless 2:1
1992 (MIT Press, Cambridge, MA).
traction-type passenger elevator system that was in
8 Chen, G. L. Genetic algorithm and its application, 1996
severe vibration condition. After optimization and (People’s Posts & Telecom Press, Beijing).
modification, the peak–peak value of vertical vibra- 9 David, B. F. An introduction to simulated evolutionary
tion acceleration of the elevator car was decreased optimization. IEEE Trans., Neural Netw., 1994, 5, 3–14.
from 4.45 × 10−1 to 1.55 × 10−1 m/s2 , and the A95 value 10 Schraudolph, N. N. and Belew, R. K. Dynamic parameter
was decreased from 3.96 × 10−1 to 8.2 × 10−2 m/s2 . encoding for genetic algorithms. Mach. Learn., 1992, 9,
The application example shows that the optimization 9–21.
model and method of dynamic parameters proposed 11 Robert, J. S. and Robert, J. M. Dynamic fuzzy control
in this article is effective. of genetic algorithm parameter coding. IEEE Trans.Syst.
Although the optimization model of dynamic Man Cybern. B, 1999, 29, 426–433.
12 Whitley, D., Mathias, K., and Fitzhorn, P. Delta coding:
parameters for a 2 : 1 traction-type passenger eleva-
an iterative search strategy for genetic algorithms. In Pro-
tor and optimization method by using dynamic byte
ceedings of the 4th International Conference on Genetic
coding GA proposed in this article have been applied Algorithms, San Mateo, CA, 1991, pp. 77–84.
successfully in some practical engineering problems, 13 Back, T. The interaction of mutation rate, selection and
some relevant research work still needs to be inves- self-adaptation within a genetic algorithm. In Proceed-
tigated in the future. First, considering more compo- ings of the 2nd Parallel Problem Solving from Nature,
nents and elements, a more accurate dynamic model North Holland, 1992, pp. 85–94.
for a 2 : 1 traction-type passenger elevator system 14 PMT, EVA-625 system operations manual, version 6 for
needs to be investigated. Second, other than consider- Windows™, 2001.
ing the vibration in the vertical direction, the dynamic 15 International Standard ISO 18738. Lifts (elevators) –
modelling in the horizontal direction also should be measurement of lift ride quality, 2003.
considered. Finally, the measuring technologies for
some dynamics parameters, which are very difficult to
be measured precisely need to be investigated in the
future such as the stiffness, damping, exciting force, APPENDIX
and so on, since the measuring errors of these parame-
ters will directly determine the result of the simulation Notation
of the dynamic behaviour for the elevator system. B the number of genes in each
individuals
ACKNOWLEDGEMENTS C damping (N s/m)
ct rotation damping (N s m)
This research work was supported by National Natu- C system damping matrix
ral Science Foundation of China (Grant No. 50775203) D offset distance
and Science and Technology Key Special Program of G the gene parameter corresponding to
Zhejiang Province (Grant No. 2006C11251). the parameter y
hc the range of the current search space
REFERENCES H the range of the search space
I rotary inertia (kg m2 )
1 Swerrie, D. A. The San Francisco earthquake of 1989 and K stiffness (N/m)
an inspector’s reflections. Elevator World, 1990, XXXVIII, kt rotation stiffness (N m)
14–22. K system stiffness matrix
m mass (kg)
Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science JMES1149 © IMechE 2009
Dynamic parameters for a traction-type passenger elevator 605
JMES1149 © IMechE 2009 Proc. IMechE Vol. 223 Part C: J. Mechanical Engineering Science