You are on page 1of 5

2018 10th International Conference on Communication Software and Networks

Research on Power Battery Formation Production Scheduling Problem with


Limited Waiting Time Constraints

Ning Zhou, Menghan Wu, Jianxin Zhou *


School of Information Engineering, Wuhan University of Technology, Wu Han 430070, China
e-mail: zhouning@whut.edu.cn, wumenghan@whut.edu.cn, zjx@whut.edu.cn

Abstract—The formation production scheduling problem of two processing; Chang X K and Dong M [9] established a
power battery is a combinatorial optimization problem with two-stage stochastic programming model for solving the
limited waiting time constraints. The time of waiting for the mitigation flow shop scheduling problem with wait-time-
start of the formation processing is limited after power constrained constraints, and designed an algorithm for
batteries completed the high temperature standby processing. solving the L-section.
In this paper, to minimize the makespan and the minimum Power battery formation system is continuous operation
number of changes to machines production parameters, we production under high temperature environment. The
proposed a scheduling model with waiting time constraints and formation processing is processed after the high temperature
put forward a particle swarm optimization algorithm based on
standby processing, and the products of the standby
the formation production scheduling. The validity of the
processing are the raw material of the formation processing.
proposed model and the algorithm were verified by the
example-based numerical simulation and comparison with the Since the performance of the batteries during the processing
current scheduling methods which commonly used on the may change over time, the quality of the batteries may be
production line. affected after if the waiting time for the start of the formation
processing exceeds the specified maximum waiting time. In
Keywords-power battery formation production; scheduling; this paper, we proposed a particle swarm optimization
waiting time constraints; particle swarm optimization algorithm based on the formation production scheduling with
limited waiting time constraints and analyze the problem in
I. INTRODUCTION actual production. On the premise of satisfying each
constraint condition, select the appropriate processing start
Production scheduling problems are classical scheduling time and processing machines for each job to improve the
problems in the field of industrial production and belong to quality and efficiency of the battery formation processing.
the NP-hard combinatorial optimization problems [1]. The
production scheduling problem limited with waiting time II. FORMATION PRODUCTION SCHEDULING
constraints exists extensively in the actual production system. MATHEMATICAL MODEL
It requires that the waiting time of jobs for the next process
after the completion of a certain process should not exceed A. Description of The Problem
the value of pre-set time, otherwise the quality of the After the batteries was filled with liquid and sealed, they
products may be affected [2]. The goals of scheduling are to enter the automated production system as a temporary whole
determine the processing start time and processing machines together with the trays and are processed on the standby
for all jobs based on satisfying the constraints and racks and the formation machines in sequence until all the
assumptions, as to obtain the overall optimal scheduling processing steps are completed, and then exit the formation
target value [3] [4]. system. Because there is a latency constraint, the waiting
The study of production scheduling with limited waiting time for the batteries to wait for the formation processing to
time constraints is a research hotspot in recent years. Karoui start cannot exceed a defined upper limit. Otherwise, the
W, Huguet M J, Lopez P et al. [5] studied the flow-shop and defective rate of subsequent operations would be increased.
job-shop scheduling problems with limited waiting time In an industrialized automatic production line, there are
constraints, and proposed a variation algorithm for climbing m sets of formation machines (hereafter referred to as
difference search; Gicquel C, Hege L, Minoux M, et al [6] machines) having the same function, and a total of n trays
proposed an exact solution approach for this optimization (hereafter referred to as jobs) with full-size batteries for i
problem, based on a discrete time representation and a types need to be completed formation processing.
mixed-integer linear programming formulation.; Cao Z C, 1) Waiting time constraint: To ensure the high quality of
Lin C R, Huang R. [7] proposed a Copula function-based production as much as possible, the start time of batteries
distribution estimation algorithm for a class of uncorrelated formation processing should be in specified time, it is the
parallel machine scheduling problems with waiting time most important constraint in formation production
constraints; Condotta A, Shakhlevich N V [8] proposed a scheduling.
heuristic algorithm based on taboo search for re-inserting 2) Change of machines production parameters: Different
artifacts, experiments show that the algorithm can solve the types of batteries have different recipes. Each type of
scheduling problem of jobs that are processed two times batteries should be processed by its own recipe. The
between machines and have a certain time interval between

978-1-5386-7223-5/18/$31.00 ©2018 IEEE 497


production parameters need to be manually adjusted if the 6) The machines can be used directly when it is used
machine processes different types of batteries before and initially.
after, and it takes a period of time, which is called setup 7) All the jobs can be processed on any machine. The
time. processing time is pre-set and remains unchanged
during the whole processing.
B. Notations and Definition
j is the job number, and jęJJ={j1,j2,…,jn}; D. Objective Function
Based on the above model descriptions and assumptions,
r is the machine number, and ręRR={r1,r2,…,rm}; the objective functions selected in this paper are as follows:
k is the type of the job, and kęKK={k1,k2,…,ki}; 1) Minimize the makespan
In the normal working conditions of the machines, the
ui is the required quantity of each type, and uęU shorter makespan, the higher the production efficiency and
the machines utilization, the lower the idle rate of the
U={u1,u2,ಹ,ui}
machine. Hence, minimize the makespan is to meet the core
Ti is the formation processing time of the i-type jobs; objectives of the scheduling. The objective function is shown
in Equation (6).
T j is the limited waiting time constraint for the job j; f1 min(maxTk , j ) (6)

Ts j is the start time for job j to be processed; 2) Minimum number of changes of machines production
parameters
The changes of production parameters requires manual
Te j is the arrival time of job j, namely the upper limit of the operation and it takes a setup time. It would be highly
time that j can be processed; dependent on people and disrupt production rate, these are
contrary to the concept of automated production. Therefore,
Twj is the lower limit of the time that j can be processed; the continuity of production should be ensured in practical
production. Due to the number of replacements of machines
Tr , j is the end time of the last j is processed on the machine production parameters relatively small, it is easily
overlooked in the calculation procedures. So, the value of the
r; number of changes are multiplied by 10 to increase the effect
Cr is the number of changes production parameters for of the objective on the overall fitness value.
m
machines r. f2 min ¦ Cr *10 (7)
r 1
The constraints are as follows:
(1) ui t 0; (2) Ti t 0; (3) Tj t 0; (4) Twj Te j  Tj In summary, the objective functions are shown in
Equation (8):
­1 if job j is processed on machine r at time t ­ f1 min(maxTk , j )
(5) xtr, j ®
¯0 otherwise °
® m (8)
Equation (1) indicates that the jobs demand for each type ° f 2 min ¦ Cr *10
¯ r 1
is greater than 0. Equation (2) indicates that each type jobs
formation processing time is greater than 0. Equation (3) From Equation (8), we can see that the optimization of
indicates that the limited waiting time constraint for each job formation production scheduling is a multi-objective model,
is greater than 0. Equation (4) shows that the lower limit of we can set different weights for different targets and turn the
the time for starting the battery formation process is the sum multi-objective model into a single-objective model, which is
of the limited waiting time constraints and the upper limit of beneficial for solving. Here, weights are set for two objective
time. Equation (5) is the allocation constraint, the machines r functions, respectively, as shown in Equation (9), and
can only produce one job or be idle at the same time. Z1 +Z2 =1 
C. The Assumption of the Problem f min Z1 f1  Z2 f 2 (9)
1) Failure of the machines during the whole processing is
not considered. III. PROPOSED ALGORITHM
2) There is no restriction of priority constraints among all In this section, we propose a particle swarm optimization
jobs. (PSO) to solve formation production scheduling problem.
3) For each job, the processing cannot be interrupted once
it has started. A. The Principle of PSO
4) A job can only be processed on one machine at a time, The PSO was inspired by the coordinated search for food
and a machine can only process one job at a time. which lets a swarm of bird’s land at a certain place where
5) The range of the limited waiting time for each job is food can be found. The PSO is a kind of evolutionary
fixed. algorithm that based on the swarm intelligence, whose

498
mathematical of PSO model is presented in explicit equation Mapping the above particles into the machines and
and can be conducted and operated easily, shown as processing sequence, the scheduling results are shown in
Equation (10) and (11)[10]. Table II
vidt 1 Zvidt  c1r1 ( pidt  xidt )  c2 r2 ( pgd
t
 xidt ) (10)
TABLE II. JOBS ASSIGNMENT
xidt 1 xidt  vidt 1 (11)
Sequence
1 2 3
Z it is the inertia factor; c1 , c2 are the acceleration factors
Machines
M1 A1 A3 0
represent the weight of each particle to the optimal position
M2 C6 B4 A2
itself pbest and the position of the global optimal position
M3 B5 C7 0
gbest respectively; r1 , r2 are random numbers distributed
2) Evaluating the individual fitness function
between [0,1]; vidt is the velocity of the particle i in the d-th In the PSO, the choice of the optimal solution is
dimension of the t-th iteration; pidt is the position of the determined by its fitness function value, and the solution
individual optimal value in the d-th dimension of the particle selection direction should be consistent with the fitness
i after the t-th iteration. pgd t
is the position of the global function value. Since each particle represents a scheduling
scheme, the makespan and the number of changes of the
optimal value in the d-th dimension of the whole particle machines production parameters could be calculated from
group after the t-th iteration, and xidt is the current position Equation (6) and (7).
of the particle i in the d-th dimension in the t-th iteration[11]. Considering that the power battery formation system is a
B. PSO Algorithm Based on Formation Production continuous processing production industry under high
Scheduling temperature environment, the battery should enter the next
process within a specified time after completing a certain
Combined with the design of the mathematical model process to ensure the stability of the battery performance.
with waiting time constraints in this paper, the PSO based on j j
formation production scheduling is designed as follows: Here, the job j whose Ts exceeds Tw is obviously not a
1) Mapping the application solution space into the satisfactory solution, so it should be punished. We replace
particle this objective function value of the individual with the
The production scheduling problem belongs to the smallest fitness in the contemporary population, and then add
combinatorial optimization problem, and its coding method the constructed penalty value to the objective function. It’s
has a great influence on the solution of the problem. The more obvious between the feasible solution and the
formation production scheduling problem is that there are n infeasible solution. Penalty function is constructed as shown
jobs need to be processed on m machines, we need to select in Equation (12).

^
the proper processing start time and machine for each job n

within a given waiting time interval. For this problem, we P=M ¦ max 0, Ts j  Twj ` (12)
j 1
use an n-dimensional real number vector X to identify the
state of the particle. The integer part of each dimension in the In summary, the evaluation function of the individual
vector represents the machine which is selected. For particles fitness in this paper is shown in Equation (13).
with the same integer part, the corresponding jobs would be fitness f  P (13)
processed on the same machines; the fractional part of each
dimension in the vector represents the processing sequence Among them, M is a constant penalty coefficient, its
of the jobs on the same machines, the smaller the value, the value should be determined in combination with the actual
higher the priority of the job processing. The value of each situation.
dimension should be within the range of [1, m+1), avoiding 3) PSO algorithm’s parameters
the absence of selected machines.
For example, here are 7 jobs belong to 3 types {A1, A2, In this paper, the settings of various parameters are
A3, B4, B5, C6, C7}, which need to be pprocessed on 3 shown in Table III.
machines {M1, M2, M3}. X [ x1 , x2 , , x7 ] is a particle TABLE III. PSO PARAMETER SETTING
that is randomly generated. After the X is rounded, new
Parameter name Setting number
natural number encoded particle is X c [ x1c , x2c , , x7c ] , as Population size 100
shown in Table I. Number of
500
iterations
Zmax -Zmin
TABLE I. PARTICLE CODING Z = Zmax  u gi
g max
Jobs A1 A2 A3 B4 B5 C6 C7
Z Zmax =0.9; Zmin =0.4; g max is the
X 1.47 2.39 1.84 2.36 3.34 2.18 3.51
X’ 1 2 1 2 3 2 3 maximum number of iterations; g i is the
number of current iterations

499
c1 1.5 293, the optimal value was 734.4, and remained stable until
the end of the algorithm. The value of the optimal solution
c2 1.5
varies with the number of iterations as shown in Figure 2.
>vmin , vmax @ [-1,1] The output scheduling scheme is shown in Figure 3.
> xmin , xmax @ [1,m+1]

IV. SIMULATION EXPERIMENTS AND ANALYSIS


Taking a power battery into an automated production line
as an example, a power battery factory chemical conversion
production line has 12 formation machines. There are 3 types
of batteries: A, B, C, the specific information is given in
Table IV. Among them, the time interval between two
adjacent trays in each batch is 2 min. It takes 10 minutes to
change the production parameters of the machine. The
waiting time interval corresponding to each job is shown in
Figure 1.

TABLE IV. EXPERIMENTAL DATA

Battery type
A B C B A
Parameters Figure 2. The change of the optimal solution with the number of iterations
Formation
210 360 180 360 210
time(min)
Quantity(tray) 8 10 14 8 10
Tray number 1-8 9-18 19-32 33-40 40-50
The time of first
0 120 240 480 600
tray arrive(min)
Waiting time
240 240 240 240 240
constraint(min)

Figure 3. The Gantt graph of we proposed scheduling optimization


method

At present, the production line generally adopts a


scheduling method in which the battery trays are sequentially
Figure 1. The waiting time interval of each job placed on an idle device. Compared with the scheduling
optimization method presented in this paper, the detailed
According to research and analysis of the processing of a data is shown in Table VI.
power battery into automatic production, here the relevant
parameters in the fitness function are defined as shown in TABLE VI. COMPARISON OF THE RESULTS OF TWO SCHEDULING
TableV. STRATEGIES

The scheduling
TABLE V. RELATED PARAMETERS IN FITNESS FUNCTION Processed in
method proposed in
sequence
this paper
Z1 Z2 M f1 1072 1064

Value 0.6 0.4 5


f2 36 24

The number of jobs that


The PSO program written in MATLAB is used to solve have exceeded the 2 0
the examples. When the number of population iterations was waiting time

500
Although the first method is simple and easy, it can be [3] Yan J G, Xing L N, Zhang Z S, et al. Dual Time Window Constrained
seen from Table VI, that the makespan and the changes of Job-shop Scheduling Algorithm[J]. Science Technology &
Engineering, 2016.
machines production parameters and the waiting time for the
[4] Joo B J, Kim Y D. A Branch-and-Bound Algorithm for a Two-
tray to exceed the limit are all more than the second method, Machine Flowshop Scheduling Problem with Limited Waiting Time
which would increase the production cost of the enterprise Constraints[J]. Journal of the Operational Research Society, 2009,
and reduces the quality of the batteries. Therefore, the result 60(4):572-582.
of the study verifies the feasibility and superiority of the [5] Karoui W, Huguet M J, Lopez P, et al. Climbing discrepancy search
synthetic production scheduling model based on PSO. for flowshop and jobshop scheduling with time lags[J]. Electronic
Notes in Discrete Mathematics, 2010, 36:821-828.
V. CONCLUSIONS [6] Gicquel C, Hege L, Minoux M, et al. A discrete time exact solution
This paper aimed at the production scheduling problem approach for a complex hybrid flow-shop scheduling problem with
limited-wait constraints[J]. Computers & Operations Research, 2012,
with limited waiting time constraints in the formation 39(3):629-636.
production of power battery and determined the scheduling [7] Cao Z C, Lin C R, Huang R. An Estimation of Distribution Algorithm
objective that minimizes the makespan and the minimum Based on Copula for Parallel Machine Scheduling with Constrained
number of changes of machines production parameters. We Waiting Time[J]. Acta Electronica Sinica, 2017.
established the optimization of production scheduling for the [8] Condotta A, Shakhlevich N V. Scheduling coupled-operation jobs
power battery formation processing, and then introduced the with exact time-lags[J]. Discrete Applied Mathematics, 2012,
particle swarm algorithm to solve the model, and finally 160(16–17):2370-2388.
compared the example results with the current scheduling [9] Chang X K, Dong M. Solving the Hybrid Flowshop Problem with
Limited Waiting Times in a Stochastic Environment[J]. Industrial
methods commonly used on the production line to verify the Engineering & Management, 2016.
feasibility of the proposed model and algorithm. [10] Sha D Y, Lin H H. A multi-objective PSO for job-shop scheduling
problems[J]. Expert Systems with Applications, 2010, 37(2):1065-
REFERENCES 1070.
[1] Mansouri S A, Aktas E, Besikci U. Green scheduling of a two- [11] Singh M R, Singh M, Mahapatra S S, et al. Particle swarm
machine flowshop: Trade-off between makespan and energy optimization algorithm embedded with maximum deviation theory for
consumpt254ion[J]. European Journal of Operational Research, 2016, solving multi-objective flexible job shop scheduling problem[J].
248(3):772-788. International Journal of Advanced Manufacturing Technology, 2016,
[2] An Y J, Kim Y D, Choi S W. Minimizing makespan in a two- 85(9-12):2353-2366.
machine flowshop with a limited waiting time constraint and
sequence-dependent setup times[M]. Elsevier Science Ltd. 2016.

501

You might also like