326 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 45, NO.

2, FEBRUARY 2015

A Robust Replenishment and Production Control

Policy for a Single-Stage Production/Inventory
System with Inventory Inaccuracy
Zheng Wang and Felix T. S. Chan

Abstract—Inaccurate inventory data are often found in pro- both researchers in academia and practitioners in industry.
duction/inventory systems, and have a significant negative impact DeHoratius and Raman [1] revealed the main reasons that
on the performance of raw materials replenishment and produc- lead to inventory inaccuracy: theft, miscounting, misplacing,
tion control. In this paper, robust replenishment and production
control problem for a single-stage production/inventory system deterioration, and so on. Inventory inaccuracy may worsen the
with inventory inaccuracy is investigated, with the objective of performance of inventory management and production con-
minimizing the average inventory and production cost. In addi- trol very significantly. Actually, in the early 1980s, researchers
tion to the inventory inaccuracy, the lead-time of raw materials found that WIP inventory inaccuracy distorts the effectiveness
ordering and machine unreliability are also taken into consider- of the material requirement planning (MRP) system [2]. More
ation. The problem is solved in two steps. First, for a single-stage
production/inventory system with ordering lead-time and uncer- recently, Kang and Gershwin [3] also pointed out that very
tain production capacity, but without inventory inaccuracy, an severe out-of-stock situations might be caused by a very small
approximate optimal replenishment, and production control pol- rate of undetected stock loss.
icy is constructed. Second, we reveal that it is the discontinuity of Therefore, inventory inaccuracy has a significant negative
the control variables in different domains of the state space that impact on the control of a production/inventory system, which
makes the policy sensitive to inventory data error. Consequently,
we develop a robust replenishment and production control pol- is a great challenge we are facing. To overcome this negative
icy to eliminate this sensitivity. This approach maintains the impact, we need to develop a robust replenishment and produc-
structural characteristics of the optimal policy, but allows the tion control policy that is insensitive to inventory record errors.
control variables to change smoothly when the state variables Existing research on replenishment and production control pol-
cross boundaries between different domains in the state space. icy usually focuses on a production/inventory system whose
Numerical experiments are conducted to examine the perfor-
mance of the proposed robust policy for hedging against inventory inventory data is completely accurate. The replenishment and
inaccuracy, and a sensitivity analysis is used to study how the production control problem for a production/inventory system
parameters of the production/inventory system and the robust with inaccurate inventory data has not been sufficiently inves-
policy impact on the average inventory and production cost. tigated. In this research, this new and important problem will
Index Terms—Inventory inaccuracy, optimal production con- be studied in depth, and we will develop a robust replenish-
trol, production/inventory system, robust replenishment and ment and production control policy for the simplest case, i.e., a
production control. single-stage production/inventory system with inventory inac-
curacy. This research will help us to find the method of hedg-
I. I NTRODUCTION ing against inventory inaccuracy and consequently improve
the robustness of production/inventory systems. It is also an
NVENTORY inaccuracy often exists in enterprise resource
I planning (ERP) systems or supply chain management
(SCM) systems, which means that the records of inven-
extension of our previous research on the robust production
control problem for a single machine and single part-type
manufacturing system with inaccurate WIP inventory [4].
tories of raw materials, work-in-process (WIP) or final
The rest of this paper is organized as follows. In Section II,
goods inventories might deviate from the actual quanti-
we review the literature of existing research on inventory
ties. This phenomenon has attracted much attention from
inaccuracy and control theory applied to production/inventory
Manuscript received July 26, 2013; revised January 7, 2014; accepted systems. In Section III, the problem of robust replenishment
May 14, 2014. Date of publication June 20, 2014; date of current version and production control for a single-stage production/inventory
January 13, 2015. This work was supported in part by the Research Grants system with inventory inaccuracy is defined and its mathe-
Council of the Hong Kong Special Administrative Region, China, under
Project PolyU510311, and in part by the National Natural Science Foundation matical model is constructed. In Section IV, we propose a
of China under Grant 61374198, Grant 60974096, and Grant 60934008. This robust replenishment and production control policy to hedge
paper was recommended by Associate Editor A. Janiak. against inventory inaccuracy for such a single-stage produc-
Z. Wang is with the Key Laboratory of Measure and Control of Complex
Engineering Systems, Ministry of Education, and the School of Automation, tion/inventory system. The results of some numerical experi-
Southeast University, Nanjing 210096, China (e-mail: wangz@seu.edu.cn). ments for examining the performance of the proposed policy
F. T. S. Chan is with the Department of Industrial and Systems are reported in Section V. In Section VI, the main contribu-
Engineering, Hong Kong Polytechnic University, Hong Kong (e-mail:
f.chan@polyu.edu.hk). tions of this research are summarized and some problems for
Digital Object Identifier 10.1109/TSMC.2014.2329284 future research are presented.
2168-2216 c 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
WANG AND CHAN: ROBUST REPLENISHMENT AND PRODUCTION CONTROL POLICY 327

II. L ITERATURE R EVIEW with fixed replenishment interval and batch size. Babai and
In this section, we give a brief review on the work relevant to Dallery [13] did a comparison study on the standard static
this research from two perspectives: inventory inaccuracy and reorder point policy (and order-up-to policy) and the forecast-
control theory applied to production/inventory systems, includ- based dynamic reorder point policy (and order-up-to pol-
ing traditional production/inventory theory, classical control icy), and found that if the forecast is reliable, the dynamic
theory in production/inventory systems, optimal control the- policy outperforms the static policy; otherwise, the static
ory in production/inventory systems, and robust production policy performs better. Axsäter [14] studied a single-stage
scheduling theory. production/inventory system with Poisson demand, gamma
distributed processing time and constrained capacity, and pro-
A. Inventory Inaccuracy posed an order-up-to policy to minimize the holding and
backorder cost. Kulkarni and Yan [15] investigated a produc-
In existing research on inventory inaccuracy, more attention
tion/inventory system with a stochastic production rate, a
is paid to this inaccuracy at the raw materials ordering stage
fluctuating demand rate, and a variable lead-time, modeled it
or the final goods distribution stage. However, less attention
by a continuous-time Markov chain, and analyzed the long-run
has been paid to inventory inaccuracy at the production stage.
holding, backlog, and ordering cost.
For example, Fleisch and Tellkamp [5] did a simulation study
Classical control theory, such as transfer function, stabil-
on this issue and found that eliminating inventory inaccuracy
ity analysis, feedback control, feed-forward compensation, or
is helpful in decreasing the out-of-stock level and therefore in
proportional-integral-derivative (PID) control, is often applied
reducing supply chain cost. To overcome the negative impact
to the control of a production/inventory system or a supply
of inventory inaccuracy, radio-frequency identification (RFID)
chain [16]. Disney and Towill [17] developed a discrete trans-
technology can be applied to eliminate inventory record
fer function of a vendor managed inventory (VMI) supply
errors [6], [7]. However, this will increase the investment on
chain integrated with an automatic pipeline, inventory, and
RFID devices and therefore needs to be evaluated carefully [8].
order-based production control system. They analyzed its sta-
Kang and Gershwin [3] found that the inventory inaccuracy
bility and robustness, and identified the boundary of stability.
problem could be solved by improving inventory management
Based on the classical control theory, Dejonckheere et al. [18]
strategies. Subsequently, Kok and Shang [9] proposed a joint
analyzed the factors that can cause the bullwhip effect at the
inspection and replenishment policy that could minimize the
forecasting stage in a supply chain. Also based on classi-
inventory cost, the backorder cost, and the purchase cost, so as
cal control theory, Disney et al. [19] analyzed the stability,
to hedge against inventory inaccuracy. To overcome the short-
steady-state performance, and transient performance of the
comings of inaccurate inventory data, DeHoratius et al. [10]
production and inventory control policy of a supply chain,
developed a Bayesian inventory record accounting method
for both continuous and discrete time models, and found that
and a corresponding replenishment policy so that the phe-
the management insights gained from both models are equiv-
nomenon of “freezing” could be avoided and much of the
alent. The WIP inventory inaccuracy on a shop floor might
cost of inventory inaccuracy could be compensated.
cause a time-delay in the WIP feedback loop in the produc-
Compared to the research on the inventory inaccuracy issue
tion and inventory control system, which makes it difficult to
in inventory management, research on this issue in production
construct a good control policy [20]. Riddalls and Bennett [21]
control has not been investigated sufficiently. In addition, pro-
compared the approximation of first-order lag and pure time-
duction control and inventory management are closely linked
delay component, and found that the latter is more realistic,
to each other. Therefore, research on the integrated replenish-
based on which they find the stability condition for a produc-
ment and production control problem is of great significance.
tion/inventory system with time-delay, and further proposed a
For a production/inventory system without inventory inaccu-
method of designing a production/inventory system controller
racy, this issue has been investigated for many years. However,
to ensure its stability [22].
for a production/inventory system with inventory inaccuracy,
Optimal control theory is often applied to produc-
the integrated replenishment and production control problem
tion/inventory systems in order to solve the optimal produc-
has not been investigated sufficiently, although it is also very
tion/inventory control problem. Gershwin [23] gave a sum-
important and interesting. This is the issue to be addressed in
mary of the hedging point production control policies based
this research.
on the continuous time model of manufacturing systems. For a
two-station production/inventory system with Poisson demand
B. Control Theory in Production/Inventory Systems and exponential service time, Veatch and Wein [24] developed
As mentioned in the previous subsection, the replenishment an optimal control policy by using dynamic programming to
and production control problem for a production/inventory sys- minimize the total inventory holding and backordering costs.
tem with accurate inventory record has been adequately stud- For a production/inventory system with both backorders and
ied. Traditional production/inventory theory usually focuses lost sales, Benjaafar et al. [25] proposed an optimal control
on the ordering and replenishment policy, in which replen- policy captured by three state-dependent thresholds: a produc-
ishment intervals and ordering batches are often taken into tion base-stock level and two order-admission levels, which
consideration. For example, Chao and Zhou [12] proposed determine whether or not to satisfy demand, backorder it or
an optimal ordering and replenishment policy, i.e., the (R, reject it. For a production/inventory system with imperfect
nQ) policy, for a multiechelon production/inventory system advance demand information, Benjaafar et al. [26] developed
328 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 45, NO. 2, FEBRUARY 2015

an optimal state-dependent base-stock control policy based on

a continuous time Markov decision model.
To deal with the uncertainties in manufacturing systems
(e.g., variable processing time, failure of machines, or fluc-
tuation of demands), robust production scheduling is often
applied. Kouvelis et al. [27] measured the uncertainty of pro-
cessing time by the range of its fluctuation, and modeled the
robust scheduling problem by an integer programming model
to minimize the gap between the scheduling performance in
the worst case and that of the optimal scheduling. For the Fig. 1. Single-stage production/inventory system.
uncertainty from machine failure, Yellig and Mackulak [28]
developed a robust scheduling policy to avoid frequent
rescheduling. For uncertainties from both processing time vari- is composed of a replenishment process, a production process,
ability and machine breakdown, Goren and Sabuncuoglu [29] and a demand satisfying process. In the replenishment pro-
defined three robustness and two stability measures and pro- cess, the system orders raw materials from outside to replenish
posed a branch and bound algorithm and a beam search its raw materials stock, which is located between the replen-
heuristic to solve the problem. Robust optimization is an ishment process and the production process and has infinite
often-used technique for solving robust production scheduling capacity. In the production process, the system consumes raw
problems, where uncertain parameters in production schedul- materials and produces final goods, which are stored in the
ing are usually captured by variables in a known interval [30], final goods storage with infinite capacity. In the demand satis-
or random variables with a known probability distribution [31]. fying process, the final goods are delivered to customers. The
This technique has been applied to robust production schedul- structure of this production inventory system is depicted in
ing problems in process industry [32], [33]. It is also applied Fig. 1.
to solve the production planning and scheduling problem for The dynamics of this single-stage production/inventory
complex manufacturing and supply chain systems with ran- system can be captured by two differential equations: one is
dom market demand, uncertain yield or inaccurate production for the replenishment process, and the other is for the produc-
data [34]–[36]. tion and demand satisfying process. The state variables are
the raw materials inventory level and the final goods inven-
C. Summary tory level. It is necessary to note that the inventories of raw
From the above survey on the related work, we find as materials and final goods are inaccurate. We assume that the
follows. errors between observed inventory level and actual inventory
1) Existing research on inventory inaccuracy pays more level for the raw materials and the final goods are normally
attention to inventory control policy for hedging against distributed with known means and variances. The control vari-
the negative impact caused by this inaccuracy. However, ables are the replenishment rate and the production rate. The
little attention has been paid to the inventory inaccuracy lead-time between placing an order for raw materials and the
at the production stage of a supply chain and robust pro- arrival of raw materials are taken into consideration. However,
duction control policies again this inaccuracy. The robust we assume that the demand can be satisfied immediately when
replenishment and production control problem of a pro- the final goods stock is not empty, namely, there is no time-
duction/inventory system with inventory inaccuracy has delay between production and final goods delivery. We also
not been sufficiently studied. assume that the demand rate is constant, the production capac-
2) Existing research on the control of a produc- ity is finite, and the machines are failure-prone. We assume
tion/inventory system has not taken inventory inaccuracy that the time to failure and the time to repair of the machine
into consideration, although many other uncertainties are exponentially distributed. The objective of this problem is
(e.g., stochastic processing time, random demand, fluc- to minimize the average inventory and production cost, which
tuation of production capacity, or variable lead-time) includes the holding cost of raw materials, the holding cost of
have been investigated, and many control policies (e.g., final goods, and the backlog cost of demand.
the policies based on classical control theory, opti- The mathematical model of this problem can be constructed
mal production and inventory control policies, or robust as follows:
production scheduling theory) were developed.
 T 
− − 
Due to the reasons given from the above two perspec- 1
tives, the robust replenishment and production control problem Minimize lim E c1 x1 (t) + c+ +
2 x2 (t) + c2 x2 (t) dt
T→∞ T 0
for a production/inventory system with inventory inaccuracy
(1)
becomes the main issue of this research.

III. P ROBLEM D ESCRIPTION 

dx1 (t) −u(t), if 0 ≤ t < L
Subject to = (2)
In this paper, we try to solve the robust replenishment dt r(t − L) − u(t), if t ≥ L
and production control problem for a single-stage produc- dx2 (t)
tion/inventory system with inventory inaccuracy. This system = u(t) − d(t) (3)
dt
WANG AND CHAN: ROBUST REPLENISHMENT AND PRODUCTION CONTROL POLICY 329

α(t) state of the machine in the production/inventory

yi (t) = xi (t) + εi (t), i = 1, 2 (4) system at time t; α(t) = 1 means that the machine
 
works at time t; α(t) = 0 means that the machine
εi (t) ∼ N 0, σi2 , i = 1, 2 (5)
is down at time t; we assume that the length of the
r(t) = π1 (y(t)) (6) period of α(t) = 1 is exponentially distributed
with the parameter p, and the length of the
u(t) = π2 (y(t)) (7) period of α(t) = 0 is exponentially distributed
r(t) ≤ μ1 (8) with the parameter q.
u(t) ≤ μ2 α(t) (9) Equation (1) is the objective function of the robust replen-
ishment and production control problem, i.e., to minimize the
average inventory and production cost, which is composed of
where
the holding cost of the demand, the holding cost of the final
t index of time; goods, and the backlog cost of the final goods. Equation (2)
T finishing time of the production process under captures the dynamics of the replenishment process. Note that
consideration; the lead-time (or time-delay) of raw materials ordering is taken
x1 (t) inventory level of the raw materials; into account. Equation (3) describes the dynamics of the pro-
x2 (t) production surplus of final goods; if x2 is pos- duction and demand satisfying process. Equation (4) captures
itive, it means the final goods inventory level, the relationship between the observed raw materials inventory
which is denoted by x+ 2 = max(x2 , 0); if x2 level or final goods production surplus and their actual levels.
is negative, its absolute value is the back- In (5), we assume that the observation error of the inventory
log of final goods, which is denoted by x− 2 = or production surplus levels is normally distributed with the
max(−x2 , 0); mean 0 and a known variance. Equations (6) and (7) mean
yi (t) observed inventory level of raw materials for that the replenishment rate and the production rate are func-
i = 1, or observed production surplus of final tions of the observed inventory level of the raw materials and
goods for i = 2; the observed production surplus of final goods. Equations (8)
y = (y1 , y2 )T vector of observed inventory level of raw and (9) are the constraints of replenishment and production
materials and observed production surplus of capacities.
final goods;
εi (t) observation error of the inventory level of raw
materials for i = 1, or observation error of the IV. ROBUST R EPLENISHMENT AND P RODUCTION
production surplus of final goods for i = 2; C ONTROL P OLICY
we assume that εi (t) is normally distributed To solve the robust production control problem (1)–(9),
with the mean 0 and the variance σi 2 , i.e., we proceed with the following steps. First, we solve the
εi (t)∼N(0, σi 2 ) for i = 1, 2; optimal replenishment and production control problem for
c1 unit holding cost of raw materials; a single-stage production/inventory system without inventory
c+
2 unit holding cost of final goods; inaccuracy, and analyze the structure of the control policy.
c−
2 unit backlog cost of demand; Next, we aim to identify the reasons that make the perfor-
r(t) replenishment rate of raw materials at time t, mance of the policy become worse in the situation where
which is a function of the observed inventory inventory inaccuracy exists, and based on these reasons we
level of raw materials and observed produc- develop a robust replenishment and production control policy
tion surplus of final goods; against inventory inaccuracy.
μ1 maximum replenishment rate of the produc-
tion process; A. Replenishment and Production Control Policy Without
π1 the replenishment policy; Inventory Inaccuracy
u(t) production rate of final goods at time t, which
For a single-stage production/inventory system without
is a function of the observed inventory level of
inventory inaccuracy, the mathematical model of its optimal
raw materials and observed production surplus
replenishment and production control problem can be captured
of final goods;
by (1)–(3), (8), and (9). Dynamic programming can be applied
μ2 maximum production rate of the production
to solve this problem.
process;
Let g [x1 (t), x2 (t)] = c1 x1 (t) + c+ + − −
2 x2 (t) + c2 x2 (t), then the
π2 production control policy;
cost-to-go function can be defined as
d(t) demand rate of final goods at time t; we
assume that d(t) is a normally distributed J [x1 (t), x2 (t), α(t), t]
random variable;  T 

L lead-time of raw material replenishment, i.e., = min E g [x1 (τ ), x2 (τ )] dτ x1 (t), x2 (t), α(t) .
r,u t
the length of the time between the plac-
(10)
ing of an order and the arrival of raw
materials; So we have
330 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 45, NO. 2, FEBRUARY 2015

J [x1 (t), x2 (t), α(t), t]

  t+δt 

= min E g [x1 (τ ), x2 (τ )] dτ x1 (t), x2 (t), α(t) + J [x1 (t + δt), x2 (t + δt), α(t + δt), t + δt] + o(δt)
r,u t
= min Eg [x1 (t), x2 (t)] δt + J [x1 (t + δt), x2 (t + δt), α(t + δt), t + δt] + o(δt)
r,u

∂J [x1 (t), x2 (t), α(t), t]
= min Eg [x1 (t), x2 (t)] + J [x1 (t), x2 (t), α(t), t] + δx1 (t)
r,u ∂x1
∂J [x1 (t), x2 (t), α(t), t] ∂J [x1 (t), x2 (t), α(t), t]
+ δx2 (t) + λ1−α,α δt · J [x1 (t), x2 (t), α(t), t] + δt + o(δt)
∂x2 ∂t

where λ1−α,α δt is the probability that the state of the machine From (12) to (17) we know that the optimal replenishment and
changes from 1 − α to α. Dividing both sides of the above production control policy has the following form:
equation by δt and letting δt approach zero, we can obtain the ⎧
⎪ ∂J [x1 (t), x2 (t), , t]
Hamilton–Jacob–Bellman (HJB) equation as follows: ⎪
⎪ μ1 , if <0
⎪
⎪ ∂x1
⎨ ∂J [x1 (t), x2 (t), , t]
∂J [x1 (t), x2 (t), α(t), t] r(t − L) = d, if =0 (18)
− ⎪
⎪ ∂x1
⎪
⎪ ∂J [x1 (t), x2 (t), , t]
 ∂t ⎪
⎩ 0, if >0
∂J [x1 (t), x2 (t), α(t), t] ∂x1
= min Eg [x1 (t), x2 (t)] − u(t) ⎧
r,u ∂x1 0, if α(t) = 0
⎪
⎪
∂J [x1 (t), x2 (t), α(t), t] ⎪
⎪ ∂J [x1 (t), x2 (t), α(t), t]
+ [u(t) − d(t)] ⎪
⎪ μ2 , if α(t) = 1 and
∂x2 ⎪
⎪ ∂x2

⎪
⎪
+ λ1−α,α J [x1 (t), x2 (t), α(t), t] , for 0 ≤ t < L (11) ⎪
⎪ ∂J [x1 (t), x2 (t), , t]
⎪
⎪ − <0
⎪
⎪ ∂x1
∂J [x1 (t), x2 (t), α(t), t] ⎪
⎪
− ⎪
⎨ ∂J [x1 (t), x2 (t), α(t), t]
∂t d, if α(t) = 1 and
u(t) = ∂x2 (19)
⎪
⎪ ∂J [x1 (t), x2 (t), , t]
= min {Eg [x1 (t), x2 (t)] ⎪
⎪ − =0
⎪
⎪
r,u ⎪
⎪ ∂x1
∂J [x1 (t), x2 (t), α(t), t] ⎪
⎪
⎪
⎪ 0, if α(t) = 1 and ∂J [x1 (t), x2 (t), α(t), t]
+ [r(t − L) − u(t)] ⎪
⎪
∂x1 ⎪
⎪ ∂x2
⎪
⎪ ∂J [x1 (t), x2 (t), , t]
∂J [x1 (t), x2 (t), α(t), t] ⎪
⎩
+ [u(t) − d(t)] − > 0.
∂x2 ∂x1

+ λ1−α,α J [x1 (t), x2 (t), α(t), t] , for t ≥ L. (12) However, it is difficult to know the values of ∂J/∂x1 and
∂J/∂x2 − ∂J/∂x1 . For the convenience of mathematical manip-
Therefore, the optimal replenishment and production control ulation, we assume that J is a convex function with respect
problem becomes the following two mathematical program- to x1 and x2 , and further assume that it is quadratic of the
ming problems: form J(x) = (1/2)xT x + bT x + c, where x = (x1 , x2 )T and b =
(P1) For 0 ≤ t < L (b1 , b2 )T . This assumption of convexity can be justified by
  the empirical production data and theoretical analysis given in
∂J [x1 (t), x2 (t), α(t), t] ∂J [x1 (t), x2 (t), α(t), t] the existing literature [37]. Based on this assumption, we have
min − u(t)
r,u ∂x2 ∂x1 ∂J/∂x1 = x1 + b1 and ∂J/∂x2 − ∂J/∂x1 = (x1 − x2 ) + (b1 − b2 ).
(13) So we can construct an approximate optimal replenishment
and production control policy that has a similar structure to
s.t. u(t) ≤ μ2 α(t) (14) (18) and (19). The approximate optimal replenishment and
production control policy has the following form:
(P2) For t ≥ L ⎧
⎨ μ1 , if x1 (t) < Z1
∂J [x1 (t), x2 (t), α(t), t] r(t − L) = d, if x1 (t) = Z1 (20)
r(t − L) ⎩
min 0, if x1 (t) > Z1
r,u
 ∂x1 
∂J [x1 (t), x2 (t), α(t), t] ∂J [x1 (t), x2 (t), α(t), t] ⎧
+ − u(t) ⎪
⎪ 0, if α(t) = 0
∂x2 ∂x1 ⎨
μ2 , if α(t) = 1 and x2 (t) − x1 (t) < Z2
(15) u(t) = (21)
⎪
⎪ d, if α(t) = 1 and x2 (t) − x1 (t) = Z2
⎩
s.t. r(t) ≤ μ1 (16) 0, if α(t) = 1 and x2 (t) − x1 (t) > Z2 .
In (21) and (22), Z 1 and Z 2 are hedging points. Based on the
u(t) ≤ μ2 α(t). (17) aforementioned assumption of the function J, we know that
WANG AND CHAN: ROBUST REPLENISHMENT AND PRODUCTION CONTROL POLICY 331

Z 1 = −b1 and Z 2 = b2 − b1 . From (21), we can see that the

production rate depends not only on the production surplus of the
final goods, but also on the inventory level of the raw materials.
Equation (20) can be rewritten as
⎧
⎨ μ1 , if x1 (t + L) < Z1
r(t) = d, if x1 (t + L) = Z1 (22)
⎩
0, if x1 (t + L) > Z1 .
Equation (22) indicates that the replenishment rate at time t
is the feedback of the raw materials inventory level at time
t + L, i.e., the future state x1 (t + L), which is due to the lead-
time of the raw materials ordering. At time t, it is impossible
to observe the state at time t + L, however, it is possible to
predict it. From (2), we know that
⎧  t+L
⎪
⎪ x (t) − u(τ )dτ, if 0 ≤ t < L
⎨ 1
t
t
x1 (t + L) = x1 (t) + t−L r(τ )dτ (23)
⎪
⎪  t+L
⎩
− t u(τ )dτ, if t ≥ L
where r(τ ) for τ ∈[t − L, L] is known, and u(τ ) for τ ∈[t, t + L]
is unknown. The total production in the future time interval [t, Fig. 2. Robust replenishment policy π 1 and the robust production control
t + L] can be estimated from the production data of the past policy π 2 .
 t+L
several periods. For example, t u(τ )dτ can be estimated by
t
t−mL u(τ )dτ/m. Therefore, the prediction of the future state inventory level at time t + L, which is denoted by
x1 (t + L) is given by ŷ1 (t + L), based on y1 (t). Namely
⎧ t ⎧ t
⎪
⎪ x (t) − m1 t−mL u(τ )dτ, if 0 ≤ t < L ⎪
⎪ y (t) − m1 t−mL u(τ )dτ, if 0 ≤ t < L
⎨ 1 t ⎨ 1 t
x̂1 (t + L) = x1 (t) + t−L r(τ )dτ (24) ŷ1 (t + L) = y1 (t) + t−L r(τ )dτ
⎪ ⎪
⎩ −1 t
⎪ ⎩ −1 t
⎪
m t−mL u(τ )dτ, if t ≥ L. m t−mL u(τ )dτ, if t ≥ L.
(26)
According to (22) and (24), the approximate optimal replen- Then the robust replenishment policy is (27), shown at
ishment policy becomes the top of next page, where
is the probability distri-
⎧
⎨ μ1 , if x̂1 (t + L) < Z1 bution function of the standard normal distribution and
r(t) = d, if x̂1 (t + L) = Z1 (25) β is a parameter that satisfies the following.
⎩
0, if x̂1 (t + L) > Z1 . a) If the observed state (i.e., y1 ) is greater than Z1 +
σ12
−1 (β), then the probability that the true state
Up to now, we have obtained the replenishment and produc-
(i.e., x1 ) is greater than the hedging point Z 1 is β,
tion control policy (25) and (21) for a single-stage produc-
or equivalently.
tion/inventory system without inventory inaccuracy.
b) If the observed state is less than Z1 − σ12
−1 (β),
then the probability that true state is less than Z 1
B. Robust Replenishment and Production Control Policy is β.
Against Inventory Inaccuracy 2) At time t, the raw materials inventory level and the final
From (25) and (21), we find that the values of the control goods production surplus can be observed, which are
variables change suddenly when the values of the state variables denoted by y1 (t) and y2 (t), respectively. The robust pro-
x1 and x2 − x1 cross the hedging points Z 1 and Z 2 , respectively; duction control policy based on the two observed state
or in other words, the control variables are discontinuous. It variables is (28), shown at the top of next page, where
is the discontinuity that may worsen the performance of the β is a parameter that satisfies the following.
 state (i.e., y2 − y1 ) is greater than
replenishment and production control policy when inventory a) If the observed
inaccuracy exists. To hedge against inventory inaccuracy, we Z2 + σ12 + σ22
−1 (β), then the probability that the
design a robust replenishment and production control policy that true state (i.e., x2 − x1 ) is greater than the hedging
is continuous and changes smoothly when the values of the state point Z 2 is β, or equivalently.
variables x1 and x2 − x1 cross the hedging points Z 1 and Z 2 , b) If the observed state is less than Z2 −
respectively. This type of robust replenishment and production σ12 + σ22
−1 (β), then the probability that
control policy is constructed as follows. true state is less than Z 2 is β.
1) Since at time t error exists between the observed raw The curve of the robust replenishment policy π 1 and
materials inventory level y1 (t) and its actual inventory the robust production control policy π 2 are shown
level x1 (t), we can only predict the raw materials in Fig. 2.
332 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 45, NO. 2, FEBRUARY 2015

⎧
⎪
⎪ μ1 , if ŷ1 (t + L) < Z1 − σ12
−1 (β)
⎪
⎪  3
⎪
⎪ Z1 − ŷ1 (t + L)
⎪
⎪
⎪
⎨ (μ1 − d) + d, if Z1 − σ12
−1 (β) ≤ ŷ1 (t + L) < Z1
  σ12
−1 (β)
r(t) = π1 ŷ1 (t + L) =  3 (27)
⎪
⎪
⎪ d Z1 − ŷ1 (t + L) + d,
⎪ if Z1 ≤ ŷ1 (t + L) ≤ Z1 + σ12
−1 (β)
⎪
⎪
⎪
⎪ σ12
−1 (β)
⎪
⎩
0, if ŷ1 (t + L) > Z1 + σ12
−1 (β)

⎧
⎪
⎪ 0, if α(t) = 0  
⎪
⎪
⎪
⎪ μ2 , if α(t) = 1 and y2 (t) − y1 (t) < Z2 − σ12 + σ22
−1 (β)
⎪
⎪    3
⎪
⎪ Z2 − y2 (t) − y1 (t)
⎪
⎪
⎪
⎪ (μ2 − d)  2  +d
⎪ σ1 + σ22
−1 (β)
  ⎨  2 
u(t) = π2 y2 (t) − y1 (t) = if α(t) = 1 and Z2 − σ1 + σ22
−1 (β) ≤ y2 (t) − y1 (t) < Z2 (28)
⎪
⎪    3
⎪
⎪ Z2 − y2 (t) − y1 (t)
⎪
⎪ d  2  +d
⎪
⎪
⎪
⎪ σ1 + σ22
−1 (β)
⎪
⎪  
⎪
⎪ if α(t) = 1 and Z2 ≤ y2 (t) − y1 (t) ≤ Z 2 + σ12 + σ22
−1 (β)
⎪
⎩
0, if α(t) = 1 and y2 (t) − y1 (t) > Z2 + σ12 + σ22
−1 (β)

Equations (27) and (28) is the robust replenishment and pro- demand rate is normally distributed with the mean 1 kg/h and
duction control policy for a single-stage production/inventory variance 1 (kg/h)2 . The lead-time for raw material ordering is
system with inventory inaccuracy. Comparing the robust replen- L = 5 h. The observation errors of the raw materials inventory
ishment and production control policy [i.e., (27) and (28)] with and the final goods surplus are normally distributed with the
the approximate optimal replenishment and production control mean 0 kg and variance 1 kg2 , i.e., σ 1 2 = σ 2 2 = 1 kg2 . The
policy [i.e., (21) and (25)], we can see that the robust policy unit holding cost for the raw materials is c1 = 1 U.S.$/(kg·h),
changes continuously and smoothly when the observed states the unit holding cost of final goods is c+ 2 = 2 U.S.$/(kg·h), and
(i.e., y1 or y2 − y1 ) cross the hedging points (i.e., Z 1 or Z 2 ), the unit backlog cost of demand is c− 2 = 10 U.S.$/(kg·h). The
which makes the policy insensitive to the observation error failure rate of the machine is p = 0.001 and the repair rate of
of inventory and therefore reduces the fluctuation of physical the machine is q = 0.1. In the robust policy, i.e., (27) and (28),
inventory level. Less fluctuation of physical inventory level will the parameter β is set to be 0.9987 so that
−1 (β) = 3.
allow lower values of the hedging point and decrease the prob- Based on these data, we conduct a simulation study for
ability of backlogging the demand, so that the inventory and both policies, and the simulation time is chosen to be T =
production cost can be decreased. The next section describes 1000 h. The parameters Z 1 and Z 2 (i.e., the hedging points)
numerical experiments conducted to examine the performance of each policy is well chosen so that the average inventory and
of the robust policy in hedging against inventory inaccuracy. production cost under the corresponding policy is minimized.
The results are as follows.
V. N UMERICAL E XPERIMENTS 1) The average inventory and production cost under the
In this section, seven numerical experiments are conducted approximate optimal replenishment and production con-
to examine the performance of the proposed robust replen- trol policy [i.e., (21) and (25)] is 14.8734 U.S.$, and the
ishment and production control policy for a single-stage corresponding Z 1 = 0.1 kg.
production/inventory system with inventory inaccuracy. 2) The average inventory and production cost under the
robust replenishment and production control policy [i.e.,
(27) and (28)] is 10.8779 U.S.$, and the corresponding
A. Numerical Experiment 1: Performance of the Robust Z 2 = 1.5 kg.
Replenishment and Production Control Policy
Obviously, the average inventory and production cost when
In the first numerical experiment, we conduct a com- the robust policy is applied is less than the average cost when
parison study on the robust replenishment and production the approximate optimal policy is applied.
control policy (i.e., (27) and (28), which is called robust
policy hereafter), and the approximate optimal replenishment
and production control policy [i.e., (21) and (25), which is B. Numerical Experiment 2: Sensitivity Analysis for
called approximate optimal policy hereafter] for a single-stage Hedging Point
production/inventory system with inventory inaccuracy. The In the second numerical experiment, we examine how the
parameters of the system in this experiment are as follows. hedging points Z 1 and Z 2 influence the average inventory and
The maximum replenishment rate μ1 = 5 kilograms per hour production costs under the two policies. Firstly, Z 1 varies from
(kg/h), the maximum production rate μ2 = 2.0 kg/h, and the 0 to 10 kg with increments of 0.1 kg, and Z 2 is fixed at 1.5 kg.
WANG AND CHAN: ROBUST REPLENISHMENT AND PRODUCTION CONTROL POLICY 333

Fig. 4. Impact of the variances (σ1 2 and σ2 2 ) of the observation errors

of raw materials inventory level and final goods production surplus on the
average inventory and production cost under the two policies.

Fig. 3. Impact of the hedging points Z 1 and Z 2 on the average inventory

and production cost under the approximate optimal replenishment and pro-
duction control policy and the robust replenishment and production control
policy. (a) Impact of the hedging point Z 1 on the average inventory and pro-
duction cost. (b) Impact of the hedging point Z 2 on the average inventory and Fig. 5. Impact of the lead-time (L) of raw materials ordering on the average
production cost. inventory and production cost under the approximate optimal policy and the
robust policy.

Secondly, we let Z 2 vary from 0 to 10 kg with increments of 0.1

kg, while keeping Z 1 = 0.1 kg unchanged. Other parameters are surplus influence the average inventory and production cost
the same as those in Experiment 1. The results of this experiment under the two policies. We let σ 1 2 = σ 2 2 and their values
are depicted in Fig. 3. From Fig. 3, we can see that: 1) the average change from 0 to 10 kg2 with increments of 1 kg2 . Other
costs under both policies increase if the hedging point of the parameters are the same as those in Experiment 1. The results
raw material inventory (i.e., Z 1 ) increases while the hedging of this experiment are depicted in Fig. 4. From Fig. 4, we can
point Z 2 is fixed, and the average cost under the robust policy is see that the average cost under the robust replenishment and
always less than the average cost under the approximate optimal production control policy is always less than the average cost
policy; and 2) if the hedging point of the raw material inventory under the approximate optimal replenishment and production
(i.e., Z 1 ) is fixed, then the average costs under both policies control policy for all the values of the variance of inventory
increase when the hedging point Z 2 increases, and the average error. Especially, when the variance of inventory error becomes
cost under the robust policy is always less than the average cost larger, the advantage of the robust policy is more obvious, i.e.,
under the approximate optimal policy. the average cost under the robust policy is much less that the
cost under the approximate optimal policy.
C. Numerical Experiment 3: Sensitivity Analysis for the
Variance of Inventory Error D. Numerical Experiment 4: Sensitivity Analysis
In the third numerical experiment, we examine how the for Lead-Time
variances of inventory errors (i.e., the parameters σ 1 2 and σ 2 2 ) In the fourth numerical experiment, we examine how the
of raw materials inventory level and final goods production lead-time (L) for raw materials ordering influences the average
334 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 45, NO. 2, FEBRUARY 2015

Fig. 6. Impact of the machine failure rate (p) on the average inventory and
production cost under the approximate optimal replenishment and production
control policy and the robust replenishment and production control policy.

inventory and production cost under the two policies. We let L

vary from 1 to 10 h with increments of 1 h, while keeping other
parameters the same as those in Experiment 1. The results
of this experiment are depicted in Fig. 5. Fig. 5 shows that
for different values of lead-times, the average inventory and
production cost under the robust replenishment and produc-
tion control policy is always less that the average cost under
the approximate optimal replenishment and production control
policy. Furthermore, for each given policy, the average costs
increase when the lead-time increases.

E. Numerical Experiment 5: Sensitivity Analysis for Machine

Failure Rate
In the fifth numerical experiment, we examine how the Fig. 7. Impact of the cost parameters on the average inventory and produc-
machine failure rate (p) influences the average inventory and tion cost under the approximate optimal replenishment and production control
policy and the robust replenishment and production control policy. (a) Impact
production cost under the two policies. We let p vary from 0 to of the unit final goods holding cost on the average inventory and produc-
0.1 with increments of 0.001, while keeping other parameters tion cost. (b) Impact of the unit backlog cost on the average inventory and
the same as those in Experiment 1. The results of this exper- production cost.
iment are depicted in Fig. 6. From Fig. 6, we know that: 1) and let c+2 change from 1 to 10 U.S.$/(kg·h) with increments
if the machine failure rate is relatively small, say, p < 0.07, of 1 U.S.$/(kg·h), while keeping other parameters the same as
then the average cost under the robust policy is less than the those in Experiment 1. We examine the average replenishment
average cost under the approximate optimal policy; and 2) if and production costs under the approximate optimal policy and
the machine failure rate is relatively large, say, p > 0.07, then the robust policy for different values of c+ 2 , and the results are
the average cost under the two policies are very close to each depicted in Fig. 7(a). Next, we keep c1 = 1 U.S.$/(kg·h) and
other, which means that the robust policy does not outperform c+ −
2 = 2 U.S.$/(kg·h) unchanged, and let c2 change from 2 to 20
the approximate optimal policy. This experiment reveals a qual- U.S.$/(kg·h) with increments of 2 U.S.$/(kg·h). We examine
itative condition of applying the robust policy, i.e., the robust the average replenishment and production costs under the two
policy can be applied only when the machine failure rate is policies for different values of c− 2 , and the results are shown
relatively small. in Fig. 7(b). Fig. 7 shows that the average replenishment and
production cost under the two policies increases when c+ 2 or
F. Numerical Experiment 6: Sensitivity Analysis for c−
2 increases. Moreover, the cost under the robust policy is
Cost Parameters always less than the cost under the approximate optimal policy
In the sixth numerical experiment, we examine how the unit for different values of c+ −
2 or c2 .
holding cost of the final goods (i.e., c+2 ) and the unit backlog
cost of demand (i.e., c−
2 ) influence the average inventory and G. Numerical Experiment 7: Sensitivity Analysis for the
production cost under the two policies, given that the unit Variance of Demand Rate
holding cost of the raw materials (i.e., c1 ) is constant. First, we In the seventh numerical experiment, we evaluate the sen-
keep c1 = 1 U.S.$/(kg·h) and c− 2 = 10 U.S.$/(kg·h) unchanged, sitivity of the average inventory and production cost to the
WANG AND CHAN: ROBUST REPLENISHMENT AND PRODUCTION CONTROL POLICY 335

different variances of the demand rate, which is true for both

the normal distribution case and the uniform distribution case.

H. Summary
Based on the results of the seven numerical experiments,
we can draw following conclusions.
1) The average inventory and production cost under the
robust replenishment and production control policy is
usually less than the average cost under the approximate
optimal replenishment and production control policy.
This is true for different values of hedging points, lead
time of raw materials ordering, unit holding cost, unit
backlog cost, and variance of demand rate.
2) The advantage of applying the robust replenishment and
production control policy is more obvious if the variance
of the observation error of the raw materials inventory
level or the final goods production surplus is large, or the
lead-time of raw materials ordering is relatively long.
3) If the machine failure rate is relatively small, the
robust replenishment and production control policy out-
performs the approximate optimal replenishment and
production control policy in producing lower average
inventory and production cost. However, if the machine
failure rate becomes large, the costs under the two
policies are close to each other.

VI. C ONCLUSION
In this paper, we study the robust replenishment and produc-
tion control problem for a single-stage production/inventory
system with inventory inaccuracy. Firstly, the mathematical
model of this problem is constructed. In the model, the dynam-
Fig. 8. Impact of the variance of demand rate on the average inventory ics of the replenishment process and the production process are
and production cost under the approximate optimal replenishment and pro- captured by two differential equations, in which the raw mate-
duction control policy and the robust replenishment and production control rials inventory level and the final goods production surplus
policy. (a) Demand rate is normally distributed. (b) Demand rate is uniformly
distributed. are state variables, and the replenishment rate and produc-
tion rate are control variables. Particularly, the lead-time of
the raw materials ordering is taken into consideration, which
variance of the demand rate. First, we conduct the numerical brings a time-delay to the dynamic equation of the replen-
experiment for the case where the demand rate is normally dis- ishment process. The inaccuracy of inventory is captured by
tributed, with the mean 1 (kg/h), and let its variance change the difference between the observed and actual raw materials
from 0 to 5 (kg/h)2 with increments of 0.5 (kg/h)2 , while inventory levels, as well as the difference between the observed
keeping other parameters the same as those in Experiment 1. and actual final goods surpluses. We assume that the obser-
The average replenishment and production costs under the vation errors are normally distributed with known means and
approximate optimal policy and the robust policy for differ- variances. The objective of this problem is to minimize the
ent variances of demand rate are evaluated, and the results are average inventory and production cost.
illustrated in Fig. 8(a). Next, we consider the case in which the This problem is solved in two steps. In the first step, we
demand rate is uniformly distributed, with the mean 1 (kg/h), study the optimal replenishment and production control prob-
and also let its variance change from 0 to 5 (kg/h)2 with lem for a single-stage production/inventory problem without
increments of 0.5 (kg/h)2 . The average replenishment and pro- inventory inaccuracy. Dynamic programming is applied to
duction costs under the two policies for different variances of solve the problem and a hedging point-based approximate opti-
demand rate are depicted in Fig. 8(b). From Fig. 8, we can see mal replenishment and production control policy is proposed.
that the average replenishment and production costs under the In the second step, a robust replenishment and production con-
two policies increase if the variance of demand rate increases, trol policy is developed based on the approximate optimal
no matter what probability distribution the demand rate fol- policy. In the robust policy, the replenishment and produc-
lows. Moreover, the cost under the robust policy is always tion rate change continuously and smoothly when the state is
less than the cost under the approximate optimal policy for crossing the hedging point.
336 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 45, NO. 2, FEBRUARY 2015

Numerical experiments are conducted to examine the per- also be considered in controlling production/inventory systems
formance of the proposed robust replenishment and production with inaccurate inventory data.
control policy; and sensitivity analysis is also carried out to
evaluate how the hedging point, the variance of inventory error, ACKNOWLEDGMENT
the lead-time of raw materials ordering, the machine failure
rate, the unit cost holding and backlog costs, and the variance The authors would like to thank the Hong Kong Polytechnic
of demand rate influence the average inventory and production University Research Committee for financial and technical
cost. The results of numerical experiments are summarized as support. They would also like to thank the editor and the
follows. reviewers for their valuable comments and suggestions that
If the machine failure rate is relatively small, it is better to have led to the substantial improvement of the paper.
apply the robust replenishment and production control policy,
because it can produce lower average inventory and production R EFERENCES
cost than the approximate optimal replenishment and produc-
[1] N. DeHoratius and A. Raman, “Inventory record inaccuracy: An empir-
tion control policy. Moreover, the advantage of applying the ical analysis,” Manag. Sci., vol. 54, no. 4, pp. 627–641, 2008.
robust policy is more obvious when the variance of the obser- [2] R. L. French, “Accurate WIP inventory—A critical MRP system require-
vation errors of the raw materials inventory level or the final ment,” Prod. Invent. Manag., vol. 21, no. 1, pp. 17–22, 1980.
[3] Y. Kang and S. B. Gershwin, “Information inaccuracy in inventory sys-
goods production surplus becomes large, or the length of the tems: Stock loss and stockout,” IIE Trans., vol. 37, no. 9, pp. 843–859,
lead-time of raw materials ordering becomes large. 2005.
The robust replenishment and production control policy pro- [4] Z. Wang and Y. Yu, “Robust production control policy for a single
machine and single product-type manufacturing system with inaccu-
posed in this paper has good potential for being applied in rate observation of production surplus,” IIE Trans., vol. 44, no. 12,
real-world manufacturing systems. For example, as reported pp. 1061–1082, 2012.
by Bose et al. [38], in the ERP system of a valve manufacture [5] E. Fleisch and C. Tellkamp, “Inventory inaccuracy and supply chain
performance: A simulation study of a retail supply chain,” Int. J. Prod.
in China, the record of the inventory of raw materials, WIP Econ., vol. 95, no. 3, pp. 373–385, 2005.
and finished products are inaccurate because of misplacing, [6] H. S. Heese, “Inventory record inaccuracy, double marginalization, and
miscounting, and loss or damage of documents, which makes RFID adoption,” Prod. Oper. Manag., vol. 16, no. 5, pp. 542–553, 2007.
[7] Y. Rekik, E. Sahin, and Y. Dallery, “Inventory inaccuracy in retail stores
the performance of the ERP system become worse, especially due to theft: An analysis of the benefits of RFID,” Int. J. Prod. Econ.,
in production planning and scheduling. To solve this problem, vol. 118, no. 1, pp. 189–198, 2009.
in addition to the deploying the RFID system or integrating the [8] S. Chang, D. Klabjan, and T. Vossen, “Optimal radio frequency iden-
tification deployment in a supply chain network,” Int. J. Prod. Econ.,
ERP and SCM systems in this enterprise, the robust replenish- vol. 125, no. 1, pp. 71–83, 2010.
ment and production control policy proposed in this paper can [9] A. G. Kok and K. H. Shang, “Inspection and replenishment policies for
also be adopted. This policy can be implemented in three steps: systems with inventory record inaccuracy,” Manuf. Serv. Oper. Manag.,
vol. 9, no. 2, pp. 185–205, 2007.
1) analyzing the inventory data and estimating the probability [10] N. DeHoratius, A. J. Mersereau, and L. Schrage, “Retail inventory
distribution of inventory error; 2) developing the replenish- management when records are inaccurate,” Manuf. Serv. Oper. Manag.,
ment and production control software and integrating it to the vol. 10, no. 2, pp. 257–277, 2008.
[11] S. C. Graves, “A multiechelon inventory model with fixed replenishment
ERP and SCM systems, where the robust replenishment pol- intervals,” Manag. Sci., vol. 42, no. 1, pp. 1–18, 1996.
icy is applied to overcome the inaccuracy in raw materials [12] X. Chao and S. X. Zhou, “Optimal policy for a multiechelon inventory
inventory and the robust production control policy is applied system with batch ordering and fixed replenishment intervals,” Oper.
Res., vol. 57, no. 2, pp. 377–390, 2009.
to overcome the inaccuracy in WIP and finished production [13] M. Z. Babai and Y. Dallery, “Dynamic versus static control policies in
inventories; and 3) developing a middleware to coordinate the single stage production-inventory systems,” Int. J. Prod. Res., vol. 47,
replenishment and production control software and the RFID no. 2, pp. 415–433, 2009.
[14] S. Axsäter, “A capacity constrained production-inventory system with
system so that the inventory inaccuracy can be eliminated as stochastic demand and production times,” Int. J. Prod. Res., vol. 48,
far as possible. no. 20, pp. 6203–6209, 2010.
This paper reports a start-up research on the robust [15] V. Kulkarni and K. Yan, “Production-inventory systems in stochastic
environment and stochastic lead times,” Queueing Syst., vol. 70, no. 3,
replenishment and production control problem of a produc- pp. 207–231, 2012.
tion/inventory system with inventory inaccuracy. In the future, [16] M. Ortega and L. Lin, “Control theory applications to the production–
we will go further to study the robust replenishment and pro- inventory problem: A review,” Int. J. Prod. Res., vol. 42, no. 11,
duction policy for multiechelon production/inventory systems pp. 2303–2322, 2004.
[17] S. M. Disney and D. R. Towill, “A discrete transfer function model to
with inventory inaccuracy, and also take the distribution stage determine the dynamic stability of a vendor managed inventory supply
of a supply chain into consideration. In this research, there is chain,” Int. J. Prod. Res., vol. 40, no. 1, pp. 179–204, 2002.
an implicit assumption that the capacities of the raw materi- [18] J. Dejonckheere, S. M. Disney, M. R. Lambrecht, and D. R. Towill,
“Transfer function analysis of forecasting induced bullwhip in supply
als stock and the final goods stock are infinite. Actually, in chains,” Int. J. Prod. Econ., vol. 78, no. 2, pp. 133–144, 2002.
real-world production/inventory systems, the capacities of raw [19] S. M. Disney, D. R. Towill, and R. D. H. Warburton, “On the equivalence
materials and final goods are finite, which makes the replen- of control theoretic, differential, and difference equation approaches
to modeling supply chains,” Int. J. Prod. Econ., vol. 101, no. 1,
ishment and production control policy more complicated. So pp. 194–208, 2006.
another research issue in the future is the robust replenishment [20] S. M. Disney, M. M. Naim, and D. R. Towill, “Genetic algorithm opti-
and production control policy for finite raw materials and final misation of a class of inventory control systems, ” Int. J. Prod. Econ.,
vol. 68, no. 1, pp. 259–278, 2000.
goods stocks with inaccurate inventories records. Other char- [21] C. E. Riddalls and S. Bennett, “The stability of supply chains,” Int.
acteristics such as batch size and replenishment intervals will J. Prod. Res., vol. 40, no. 2, pp. 459–475, 2002.
WANG AND CHAN: ROBUST REPLENISHMENT AND PRODUCTION CONTROL POLICY 337

[22] C. E. Riddalls and S. Bennett, “Production-inventory system controller Zheng Wang received the B.S. degree in mechanical
design and supply chain dynamics,” Int. J. Syst. Sci., vol. 33, no. 3, engineering from the Changchun Institute of Optics
pp. 181–195, 2002. and Fine Mechanics, Changchun, China, in 1993,
[23] S. B. Gershwin, “Design and operation of manufacturing systems: The and the M.S. and Ph.D. degrees in control theory
control-point policy,” IIE Trans., vol. 32, no. 10, pp. 891–906, 2000. and control engineering from Southeast University,
[24] M. H. Veatch and L. M. Wein, “Optimal control of a two-station tandem Nanjing, China, in 1998 and 2001, respectively.
production/inventory system,” Oper. Res., vol. 42, no. 2, pp. 337–350, He is a Professor of the School of Automation,
1994. Southeast University. His current research inter-
[25] S. Benjaafar, M. ElHafsi, and T. Huang, “Optimal control of a ests include manufacturing systems control, pro-
production-inventory system with both backorders and lost sales,” Nav. duction and inventory control, and supply chain
Res. Logist., vol. 57, no. 3, pp. 252–265, 2010. management. He has published papers on interna-
[26] S. Benjaafar, W. L. Cooper, and S. Mardan, “Production-inventory sys- tional refereed journals such as IEEE T RANSACTIONS ON E NGINEERING
tems with imperfect advance demand information and updating,” Nav. M ANAGEMENT, the IIE Transactions, the European Journal of Operational
Res. Logist., vol. 58, no. 2, pp. 88–106, 2011. Research, the International Journal of Production Research, Journal of the
[27] P. Kouvelis, R. L. Daniels, and G. Vairaktarakis, “Robust scheduling of Operational Research Society, the Journal of Intelligent Manufacturing, the
a two-machine flow shop with uncertain processing times,” IIE Trans., Concurrent Engineering: Research and Applications, and the International
vol. 32, no. 5, pp. 421–432, 2000. Journal of Advanced Manufacturing Technology.
[28] E. J. Yellig and G. T. Mackulak, “Robust deterministic scheduling in Dr. Wang is a member of INFORMS.
stochastic environments: The method of capacity hedge points,” Int.
J. Prod. Res., vol. 35, no. 2, pp. 369–379, 1997.
[29] S. Goren and I. Sabuncuoglu, “Optimization of schedule robustness
and stability under random machine breakdowns and processing time
variability,” IIE Trans., vol. 42, no. 3, pp. 203–220, 2010.
[30] X. Lin, S. Janak, and C. A. Floudas, “A new robust optimization
approach for scheduling under uncertainty: I. Bounded uncertainty,”
Comput. Chem. Eng., vol. 28, nos. 6–7, pp. 1069–1085, 2004.
[31] S. Janak, X. Lin, and C. A. Floudas, “A new robust optimiza-
tion approach for scheduling under uncertainty: II. Uncertainty with
known probability distribution,” Comput. Chem. Eng., vol. 31, no. 3,
pp. 171–195, 2007.
[32] Z. Li and M. Ierapetritou, “Process scheduling under uncertainty:
Review and challenges,” Comput. Chem. Eng., vol. 32, nos. 4–5,
pp. 715–727, 2008.
[33] Z. Li and M. G. Ierapetritou, “Robust optimization for process
scheduling under uncertainty,” Ind. Eng. Chem. Res., vol. 47, no. 12,
pp. 4148–4157, 2008.
[34] S. C. H. Leung, S. O. S. Tsang, W. L. Ng, and Y. Wu, “A robust Felix T. S. Chan received the B.Sc. degree in
optimization model for multi-site production planning problem in an mechanical engineering from Brighton Polytechnic
uncertain environment,” Eur. J. Oper. Res., vol. 181, no. 1, pp. 224–238, (now University), Brighton, U.K., and the M.Sc. and
2007. Ph.D. degrees in manufacturing engineering from
[35] M. K. Zanjani, D. Ait-Kadi, and M. Nourelfath, “Robust production the Imperial College of Science and Technology,
planning in a manufacturing environment with random yield: A case University of London, London, U.K.
in sawmill production planning,” Eur. J. Oper. Res., vol. 201, no. 3, He is currently with the Department of Industrial
pp. 882–891, 2010. and Systems Engineering, Hong Kong Polytechnic
[36] A. P. Kanyalkar and G. K. Adil, “A robust optimisation model for University, Hong Kong. His current research inter-
aggregate and detailed planning of a multi-site procurement-production- ests include logistics and supply chain management,
distribution system,” Int. J. Prod. Res., vol. 48, no. 3, pp. 635–656, operations management, distribution coordination,
2010. systems modeling and simulation, and supplier selection. He has published
[37] S. B. Gershwin, Manufacturing Systems Engineering. Englewood Cliffs, 15 book chapters, over 280 refereed international journal papers, 240 peer
NJ, USA: Prentice Hall, 1994. reviewed international conference papers, and h index = 27 under the Web of
[38] I. Bose, R. Pal, and A. Ye, “ERP and SCM systems integration: The Science.
case of a valve manufacturer in China,” Inf. Manag., vol. 45, no. 4, Prof. Chan is a Chartered Member of the Chartered Institute of Logistics
pp. 233–241, 2008. and Transport, Hong Kong.