262 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING , VOL. 2, NO. 3, JULY 2005
Machine Learning Approach for Determining
Feasible Plans of a Remanufacturing System
Chen Song, Student Member, IEEE, Xiaohong Guan, Senior Member, IEEE, Qianchuan Zhao, and
Yu-Chi Ho, Life Fellow, IEEE
Abstract—Resource planning for a complex remanufacturing
system is in general extremely difficult in terms of, e.g., problem
size and uncertainties. In many cases, simulation is the only way
to select a good plan among a great number of candidates. When
there exist complicated constraints, direct selection could be very
inefficient since many candidates may not be feasible but cannot
be excluded beforehand. To meet the challenge, a machine learning
method is introduced in this paper to perform feasibility analysis.
The rough set theory is first applied to establish the relationship
between a plan and its feasibility and an iterative reinforcement
process is applied to enhance confidence. The numerical testing
results show that this method is promising and scalable for the
large-scale problems. The research lays a basis for developing an
efficient simulation-based optimization method with complicated
constraints.
Note to Practitioners—This paper was motivated by the resource
planning problem for a complex remanufacturing system, which
is very important but in general extremely difficult to deal with in
terms of, e.g., problem size and uncertainties. Simulation is prob-
ably the only way available to select a good plan among a number
of candidates. When there exist complicated constraints, simula-
tion becomes even more difficult and selection through simulation
could be very inefficient since many candidates may not be fea-
sible but cannot be excluded before simulation. Determining fea-
sibility beforehand is extremely difficult by analytical or numer-
ical methods. This paper suggests a new method using a machine
learning-based approach to predict the plan feasibility required in
practical applications and can be considered as the first step for op-
timization-based planning. By applying the rough set theory, the
prediction rules are obtained or learned from a training dataset
generated by simulation. Then, an iterative reinforcement process
is applied to enhance the confidence of learning and to perform
iterative retraining on new datasets by the rough set method to
Manuscript received May 15, 2004; revised November 22, 2004. This paper
was recommended by Associate Editor M. Zhou and Editor N. Viswanadham
upon evaluation of the reviewers’ comments. This work was supported in part
by the National Outstanding Young Investigator Grant 6970025, the National
Natural Science Foundation (60274011, 60243001), the 863 High Tech Devel-
opment Plan (2001AA413910) and the NCET Program of China, and by the
Fundamental Research Funds and Funds from National Key Lab. for Power Sys-
tems in Tsinghua University. The work of Y.-C. Ho was supported in part by the
U.S. Army Research Office under Contract DAAD19-01-1-0610 and the U.S.
Air Force Office of Scientific Research under Contract F49620-01-1-0288.
C. Song is with the SKLMS Laboratory and System Engineering Institute,
Xian Jiaotong University, Xian 710049, China (e-mail: csong@sei.xjtu.edu.cn).
X. Guan is with the SKLMS Laboratory and the System Engineering Insti-
tute, Xian Jiaotong University, Xian 710049, China and also with the Center
for Intelligent and Networked Systems, Tsinghua University, Beijing 100084,
China (e-mail: xhguan@tsinghua.edu.cn).
Q. Zhao is with the Center for Intelligent and Networked Systems, Tsinghua
University, Beijing 100084, China (e-mail: zhaoqc@tsinghua.edu.cn).
Y.-C. Ho is with the Center for Intelligent and Networked Systems, Tsinghua
University, Beijing 100084, China and also with the Division of Engineering
and Applied Science, Harvard University, Cambridge, MA 02138 USA (e-mail:
ho@hrl.harvard.edu).
Digital Object Identifier 10.1109/TASE.2005.849090
generate new rules to add to the knowledge base until the preset
threshold is satisfied. The numerical testing results show that the
above method is capable of determining the feasible plans for a re-
manufacturing system with good accuracy. The method is efficient
and scalable for the large-scale problems. The method developed in
the paper is being incorporated in the framework of ordinal opti-
mization, and a new constrained ordinal optimization method has
been developed for remanufacturing planning.
Index Terms—Manufacturing planning, remanufacturing sys-
tems, rough set theory, simulation-based optimization.
I. INTRODUCTION
I
N recent years, there has been escalating interest in re-
manufacturing system: an industrial process that restores
worn-out products. Typically, a unit for remanufacturing is
disassembled with its usable parts cleaned, refurbished, and
put into inventory and unusable parts replaced by new ones. It
is then reassembled to become a unit fully equivalent to, and
sometimes superior to, a new unit in terms of performance and
expected lifetime. The remanufacturing system is a typical dis-
crete event dynamic system. It includes four basic ingredients:
a set of tasks or jobs (dissembling, repairing, and assembling
of products), a set of resources (e.g., machines, tools, and
labor), and routing and planning of jobs among resources [32].
A production planner/scheduler must decide how to allocate
suitable capacity and inventory to minimize the cost while
satisfying constraints such that the repair cycling time with a
certain probability must be within the required due date.
Many problems in managing the production planning of re-
manufacturing systems have been identified in the literature.
Critical issues includes: inventory availability and management
[4], [15], [16], shop floor scheduling [15], [17]–[19], and ca-
pacity planning [20].
Production planning is a well-studied area in manufacturing
resource planning [45], [46]. The main issues include material
requirement planning [47], inventory planning [49], [48], and
capacity planning [51], [41], [25]. The methods include: math-
ematical programming, simulation, statistics, network analysis,
queueing,stochastic processes, inventory theory, probability, re-
gression analysis, forecasting models, heuristics, artificial intel-
ligence, expert systems, visual interactive modeling [50].
For the planning problems of remanufacturing systems, one
or more segments of a remanufacturing system such as disas-
sembly were usually analyzed and optimized [22], [23]. Few ex-
isting methods addressed the planning and operation control of
a remanufacturing system as a whole and the interactions among
the activities of the segments. Some researchers took advantages
1545-5955/$20.00 © 2005 IEEE
Authorized licensed use limited to: Tsinghua University Library. Downloaded on May 19, 2009 at 22:45 from IEEE Xplore. Restrictions apply.