a power aware solution for multi- machine job shop schedulingxs3d.kaist.ac.kr/tc-sma/9 - tang, gina...
TRANSCRIPT
A Power Aware Solution for Multi-
machine Job Shop Scheduling
Congbo Li
The State Key Lab of Mechanical Transmission
Chongqing University, Chongqing, China
Email: [email protected]
Ying (Gina) Tang
Electrical and Computer Engineering
Rowan University, Glassboro, NJ 08028
Email: [email protected]
Introduction
Multi-machine Job Shop Scheduling
(MMJSS) Model and Optimization
Case Study
Conclusion
The energy consumption by the industrial sector has
almost doubled and the number continues to increase
that accounts for one-half of the world’s total energy
consumption
Introduction
Background
The rising severity of energy shortage together with
other environmental problems results in great efforts
towards energy saving and emission reduction
Manufacturing sector is responsible for 1/3 energy
consumption and 28% greenhouse gas in USA
Similarly, it is responsible for 59% energy consumption
in China – 20% higher than the international level
Introduction
Background (continued)
Machine tools are the main sources of energy
consumption
Existing methods in reducing energy consumption of
machine tools and improving their efficiency:
Change properties of machine tools
Choose better fit process parameters or opting out for a
lower energy consumption machine
However, the above methods require a large amount
of capital investments for hardware changes
Introduction
Solution
MMJSS is a well-studied optimization problem that
often minimize makespan, prodcution cost and other
factors of interest. And very few studies address the
problem with environmental consideration.
Develop an optimal scheduling procedure to achieve
the goal of energy reduction:
A MMJSS is established where both energy consumption
and makespan are considered
A modified multi-objective Genetic Algorithm (GA) is
applied to solve the optimization problem
MMJSS Model
Problem Formulation
There are a set of machines M = {Mj, j=1, 2, …, m} to
process a set of jobs J={Ji, i=1, 2, …, n}. Each job Ji
consists of a sequence of operations, which must be
accomplished according to a given manufacturing
process dependent order.
The route for each job in which the sequence of
operations are processed by the machines is known a
priori.
A schedule is a function S: Oij + {0} that for
each operation of the ith job at the jth machine, Oij ,
defines start time stij.
MMJSS Model
Problem Assumptions
Each job is nonpreemptive, requiring one and only
one machine at a time.
Each job does not visit the same machine twice.
Each machine can process at most one job at a time.
There are no precedence constraints among
operations of different jobs.
Setup time is included in processing time
Two Objectives
Energy consumption
MMJSS Model
ej is accumulated when the jth machine is on for its first
operation of the n jobs until it is off after its last operation
of the n jobs
ppj - The total power usage of the machine j when it is
processing jobs
ptj - The total processing time of the machine j itj - The total idle time of the machine j
ipj - The total power usage of the machine j when it is
idle
CTi - The completion time of the job i, Ji
m
j
jjjj
m
j
jitipptppeSE
11
)()(
Makespan of a schedule T(S) }{max)(i
i
CTST
MMJSS Model
The MMJSS Optimizing Total Energy and Makespan
))(),(min( STSE
ih
i
hjihij sttst )1(
kj
j
kikjijstxtst )1(
n
i
ijj tpt
1
n
i
ijijjij sttstit
1
)( '
Subject to:
Where: Oi’j is operated after Oij ;
is a very large integer constant.
is the precedence relationship parameter that equals
one if Ji is processed at Mh before Mj;
is the precedence relationship parameter that equals
one if Jk is processed at Mj before Ji
i
hj
j
kix
MMJSS Optimization
Multi-Objective Genetic Algorithm
Single-point
crossover
Vector evaluated
GA
Operation-based encoding scheme
Two JSS models, the proposed model and the one
considering makespan only, are applied to a 6X6
benchmark problem
The modified multi-objective GA is then applied to both
models 10 times and the best results are presented
Case Study
Correlation of Energy Consumption and Makespan
Is it normally the case that a shorter makespan yields
less energy consumption?
(m, t) (m, t) (m, t) (m, t) (m, t) (m, t)
Job 1: 3,1 1,3 2,6 4,7 6,3 5,6
Job 2: 2,8 3,5 5,10 6,10 1,10 4,4
Job 3: 3,5 4,4 6,8 1,9 2,1 5,7
Job 4: 2,5 1,5 3,5 4,3 5,8 6,9
Job 5: 3,9 2,3 5,5 6,4 1,3 4,1
Job 6: 2,3 4,3 6,9 1,10 5,4 3,1
Note: (m,t) means which machine processes the job and the
number of time units needed
Machine Number Idle power(KW) Processing power(KW)
1、3 3.1 12.5
2、4 0.7 3
3、6 1.8 8.8
Table 1: Benchmark problem Table 2: The idle and processing power of machines
Note: the energy data is adapted from D.N. Kordonowy, A
Power Assessment of Machining Tools. Bachelor of Science
Thesis in Mechanical Engineering, Massachusetts Institute of
Technology, Massachusetts, USA: MIT Press, 2002
Case Study
Correlation of Energy Consumption and Makespan (cont.)
Experiment No. The proposed model
Makespan (Day) Energy consumption (109J)
1 57 49.3128
2 59 48.744
3 58 48.7512
4 59 48.8484
5 55 49.3416
6 59 49.1004
7 60 48.7224
8 59 48.78
9 58 48.7512
10 59 49.1688
Average 58.3 48.95208
Experiment No. The baseline model
Makespan (Day) Energy consumption (109J)
1 55 50.0184
2 55 50.04
3 55 49.518
4 57 50.7204
5 55 50.4504
6 55 49.7304
7 55 50.0112
8 57 50.7528
9 55 49.7052
10 57 49.7412
Average 55.6 50.08212
Schedules with the same makespan may vary
considerably in energy consumption
It is not necessarily true that a shorter makespan yields
less energy use
The average value of energy consumption of the proposed
model is 2.25 percent lower than that of the baseline
model
Case Study
Correlation of Energy Consumption and Makespan (cont.)
(a) (b)
For the baseline model as shown in (a), when the best and
mean values of makespan are decreasingly convergent over
iterations, those of energy consumption are fluctuating with
no pattern to seek.
The proposed model takes into consideration of both energy
consumption and makespan in MMJSS, the mean and best
values of both energy consumption and makespan are
convergent over iterations as shown in Fig. (b)
Case Study
Performance of Modified GA
The modified multi-objective GA is applied to several
benchmark problems with different complexity from Fisher
and Thompson (1963) [1] and Lawrence (1984) [3]
The results are then compared to the ones using 2ST-GA [2]
and 2ST-PSO [4]
1. H. Fisher, and G.L. Thompson, “Probabilistic learning combinations of local job shop scheduling
rules”. in Industrial scheduling. New Jersey, Prentice-Hall, pp. 225-251
2. V. Kachitvichyanukul, and S. Sitthitham, “A two-stage genetic algorithm for multi-objective job shop
scheduling problems,” Journal of Intelligent Manufacturing, vol. 22, no. 3, pp. 355-365, 2011
3. S. Lawrence, “Supplement to resource constrained project scheduling: An experimental
investigation of heuristic scheduling techniques,” in Technical report, GSIA, Carnegie Mellon
University. 1984
4. T. Pratchayaborirak, and V. Kachitvichyanukul, “A two-stage PSO algorithm for job shop scheduling
problem,” International Journal of Management Science and Engineering Management, vol. 6, no. 2,
pp. 84-93, 2011
Case Study
Performance of Modified GA (cont.)
The modified multi-objective GA improves on the
performance of the other two methods for the small-scale
benchmark problems
When the problem size increases, 2ST-PSO beats the
others, but the modified multi-objective GA still can deliver
fairly good results with way much less time in comparison to
2ST-GA (average 89% time reduction).
Problem Size nXm BKS Modified multi-objective GA 2ST-GA 2ST-PSO
Mean Time (s) Mean Time (s) Mean Time(s)
FT06 6X6 55 55.7 7.36 55 833 55 9
LA01 10X5 666 666 9.6 666 1273 666 15
LA06 15X5 926 926 17.2 926 2480 926 34
LA11 20X5 1222 1222 27.2 1222 4254 1222 59
LA16 10X10 945 993.8 105.5 955.3 567.9 945 31
LA26
20X10 1218 1304.8 658 1288.6 17345 1266 111
Note: FT problems are designed by Fisher and Thompson (1963) and LA problems are designed by Lawrence (1984).
Contribution:
A power-aware multi-machine job shop scheduling
model with the objectives of minimizing both energy
consumption and makespan is proposed.
The relationship of energy consumption and
makesan of a schedule is thoroughly analyzed, from
which a MMJSS solution with relatively better energy
consumption and makespan is ensured.
The simulation experiments concluded that a shorter
makespan does not necessarily yield less energy
consumption
Conclusion
Future Work:
Comprehensive energy models with respect to processing
loads and speeds are worth of investigation for more
complex industrial cases
More sophisticated power-aware scheduling models are
needed for the practical flexible job shop scheduling
problem
The extension of this work for large-scale real industrial
MMJSS problems to investigate the intrinsic links between
scheduling and energy consumption is necessary
Conclusion