You are on page 1of 10

Proceedings of the 2013 Industrial and Systems Engineering Research Conference

A. Krishnamurthy and W.K.V. Chan, eds.


Online Inventory Control Using a Multiobjective Six Sigma
Approach


Emad H. Abualsauod, Christopher D. Geiger
Department of Industrial Engineering and Management Systems
University of Central Florida
Orlando, FL 32816 USA



Abstract

In industrial manufacturing and service environments, it is important to develop or modify objectives for operating a
process in order to incorporate the changes that contributed to the improvement; however, these objectives are
usually conflict with each other. Additionally, in order to maintain consistency in process performance and sustain
improvements, a reliable process control plan is required. This paper utilizes an innovative integrated multiobjective
Six Sigma optimization approach for online inventory management. The approach provides automated feedback for
out-of-control events while attempting to balance two common, relevant and conflicting objectives minimization
of the inventory holding cost and minimization of the inventory ordering cost.

Keywords
Six Sigma, multiobjective optimization, online process control, inventory control

1. Background
In inventory management, it is important in this process to achieve satisfactory levels of customer service while
keeping inventory costs within reasonable bounds. There is a need to provide inventory managers with a mechanism
for correcting the differences between customer demand and on-hand inventory in real-time. Effective inventory
management is inherently challenging as it involves simultaneously satisfying random customer demand and
minimizing on-hand inventory, often viewed as conflicting objectives [1, 2]. Over the past two decades, the Six
Sigma Define-Measure-Analyze-Improve-Control (DMAIC) framework and a host of statistical tools have been
brought to bear on process improvement efforts, including the process of managing inventory. The traditional way to
control a process is analyzing historical performance values and proposing action plans accordingly. Thus,
simultaneous improvement feedback is needed especially when multiple objectives are considered. As a result, non-
traditional techniques are needed. These approaches could offer significant benefits over the traditional process
control techniques, where the main benefits are improved process control and decision-making. This paper explores
the integration of multiobjective optimization within the DMAIC framework for online inventory control.

2. Previous Related Work
Several studies have integrated Six Sigma methodologies with several other approaches. However, limited work
exists that integrate multiobjective optimization with the Six Sigma methodology. [e.g., 3-6]. Existing work include
case studies that seek to optimize the design of liquid packaging pump in order to provide a flow rate within the
specification limits so that minimal defects are produced for the least cost per part, improve the design efficiency of
high temperature superconducting [7], select projects portfolio to implement Lean and Six Sigma concepts [8],
design the airfoil design for a Mars aircraft [9], and design a welded beam in order to carry a certain load with
minimum total fabricating cost [10, 11].

3375
Abualsauod and Geiger

Other studies seek to improve decision-making at manufacturing operations in order to improve quality of products
and reduce associated cost [12], improve the process of penicillin production [13], minimize the material flow intra-
and inter-loops and minimization of the maximum amount of inter-cell flow, considering the limitation of tandem
automated guided vehicle work loading [14], minimize the total cost involved in supply chain processes to ensure
high delivery probability within customer-specified delivery windows [15], optimize the process parameters of the
deep drawing operations [16], optimize the design of a sliding rack to achieve the maximum performance and the
minimum weight [17], and generally finding solutions that are moderately good in terms of optimality and also good
in terms of robustness against small perturbations of values [18].

Additionally, existing research using multiobjective optimization techniques to improve process control has been
used to optimize Exponentially-Weighted Moving Average parameters to improve detecting magnitude shifts and to
reduce process control cost [19, 20], determine the optimal sample size for control charts [21], enhance the testing
power of

and control charts to minimize Type I errors [22], determine the economic design of the

control
charts with a realistic embedded error monitoring model [23], design a multivariate control scheme consisting of two
or three

charts to optimize the charts parameters [24], optimize the performance of attribute control charts using
genetic algorithm [25]. However, these studies are neither conducted within the Six Sigma methodological
frameworks, nor are 6 levels of process control considered during the optimization process.

Although recent studies integrate diverse statistical techniques to improve the quality of solutions of the Six Sigma
methodology, there is limited work that addresses the use of multiobjective optimization techniques that maintain
process quality at 6 levels. Furthermore, existing work that integrates multiobjective optimization with Six Sigma
scalarizes the set of multiple objectives into a single objective using a vector of user-specified weights.
2.1. Multiobjective Optimization and Six Sigma Framework for Online Inventory Control
In this research, an algorithm is proposed to optimize online process performance (see Figure 1). The proposed study
implements the first three phases of DMAIC as in traditional Six Sigma implementations. It starts with the Define
phase by defining the existing process sigma level and the critical-to-quality characteristics (CTQs) from the
customers. Then, during the Measure phase, evaluation of the current status of the process is performed.

After performing root cause analysis in the Analyze phase and implementing the improvement plan, input data such
as the relevant objective functions, constraints, and design variables and associated values are used for online
process monitoring. After the online monitoring starts, it detects any unwanted process behavior such as unnatural
patterns and mean shifts. When a mean shift is detected, it triggers an optimization procedure and provides the
decision-maker with a set of Pareto optima that include new values for the set of the decision variables at the
Improve phase. After the Improve phase is updated with the new decision variable values, online monitoring starts
again until unwanted process behavior is encountered. Then, it re-optimizes the objective functions subject to all
relevant constraints. This cycle continues until significant changes to the process are desired and the set of objective
functions, constraints, and design variables are changed. The subsequent sections provide the details of the proposed
procedure and components therein.

3376
Abualsauod and Geiger


Figure 1. Multiobjective optimization and Six Sigma framework for online process control.

3. Computational Study
3.1. Inventory Problem Case Description
In inventory management, there is a need to provide decision-makers with a mechanism for dynamically adapting to
the differences between customer demand and on-hand inventory. In this study, two conflicting inventory
management objectives are considered minimizing holding cost and minimizing ordering cost [2, 26]. The
objective functions represent the desired CTQs that need to be monitored during the process. Previous studies show
that the Economic Order Quantity (EOQ) model and variations thereof considers this tradeoff between holding cost
and ordering cost when choosing the quantity Q to use to replenish items in inventory. A larger order quantity Q to
meet random demand D results in holding a large amount of inventory, which increases holding cost. On the other
hand, a smaller order quantity Q to meet random demand D eventually reduces the average amount of inventory
thereby increasing ordering cost in order to meet demand. The value of the decision variable Q is varied to find
multiple tradeoff solutions for the objectives given a random demand D. Managing this tradeoff between holding
cost and ordering cost makes the model useful for management decision-making even if the decision variable Q is
known to be within a range of possible values [27, 28].

Specifically, the two objective functions considered for this investigation are:

1. Minimize average annual holding cost H. This cost is an important consideration for inventory management,
and it refers to the cost of storing inventory over a period of time. For this study, the definition of the annual
average holding cost is based on the total annual cost of the EOQ model [27]:
=


2

(1)
where c
h
is the unit holding cost when available inventory exceeds demand D which varies over time and Q is
the order quantity to meet demand D.

2. Minimize ordering cost. Ordering cost (O) refers to the cost of ordering units of inventory.
=

(2)
3377
Abualsauod and Geiger

where c
o
is the fixed ordering cost no matter the size of Q.

3.2. Description of the Decision Variables
In this research investigation, Q is the critical decision variable that influences the values of the objective functions.
The initial range of D is identified from empirical results published in the existing research literature. The range of
demand, and the holding and ordering unit cost values are listed in Table 1 and Table 2, respectively. Specifically,
the ranges are approximated based on experimental results based on a previous study [28].
Table 1. Customer demand for the inventory problem.
Parameter Lower Bound Upper Bound
Demand D (units) 70 160

Table 2. Costs for the inventory problem.
Parameter Value
Holding Cost

$10
Ordering Cost

$70

3.3. Implementation
3.3.1 Online Control Settings for the Inventory Case Study
There are two types of variation that affect the quality of a process. The first type is called a special cause variation,
which is also called assignable cause variation. This type of variation is a result of causes that are not normally
present in the process but they can be traced, identified, and eliminated. The second type of variation is called the
common cause variation this is the result of numerous, ever-present differences in the process. Control charts, which
are commonly used for process monitoring, help in identifying these two types of variation while monitoring the
behavior of the process. Specifically, 3-based control charts are commonly used to perform the online monitoring
of processes, and control charts are integral in the proposed methodology. Control charts are used to detect
simultaneously assignable causes of variations and if shifts in the mean performance values occur. Two control
charts are used, one for each of the two inventory cost functions. Table 3 lists the control chart parameters, which
are determined based on a previous study [28].

Table 3. Initial control chart parameters values for inventory cost problem.
Control Chart Parameter Holding Cost (H) Ordering Cost (O)
Center Line (CL) $85.00 $10.57
Upper Control Limit (UCL) $231.21 $30.10
Lower Control Limit (LCL) $0.00 $0.00

During process monitoring, change-point tests are used to determine whether a change in the mean or standard
deviation has occurred. Since the ordering and holding cost functions averages are likely to shift over the long run,
the Six Sigma DMAIC framework accounts for this mean shift by assuming 1.5 shift. Therefore, detecting shifts in
the inventory management process is important not only for monitoring changes but also to ensure that the cost
functions operate at or close to the desired sigma level. For proof-of-concept, in solving the inventory problem, the
bootstrap method is used to maintain objective functions within 3 level by detecting shifts in mean over time for
both the H and O inventory cost objectives. The theoretical formulation of the bootstrap method is as follows:
CL% = 100

%
(3)


(4)

+ (

) for = 0, . (5)
where CL% represents the percentage of mean shifts occurrence in the observed objective function values, is the
number of bootstrap samples performed, x is the number of bootstrap for which

<

where,

is the
difference of the reorder order for the observed objective function values,

is the difference of the original order


for the observed objective function values,

is the maximum value calculated from

where,

is the cumulative
sums of the straps,

is the minimum value calculated from

, and is the average of the observed objective


function values.
3378
Abualsauod and Geiger

3.3.2 Optimize Settings for the Case Study
In solving multiobjective optimization problems, there are several approaches. However, the most two common
approaches are classical methods and Evolutionary Algorithms (EAs). In the classical methods, the objective
functions are converted into a single objective function. Classical methods have the drawbacks of modeling the
original problem in that there is a single solution. On the other hand, the outcome of an EA is a population of
solutions. This advantage of using EAs over classical methods provides the ability to find multiple solutions. For the
purpose of this research, the second generation Non-dominated Sorting Genetic Algorithm (NSGA II) is used. The
search control parameter settings of NSGA II are selected based on previous studies that use the optimization
algorithm [29]. The set of optimization parameter values for this computational study is summarized in Table 4.

Table 4. Optimization parameters range and values for inventory cost problem.
Optimization Parameter Value
Population Size (P) 100
Generations (G)
15, 30, 100
Crossover Probability (P
c
) 85%
Mutation Probability (P
m
) 5%

3.4. Results and Discussion
The experiments are performed using the control and optimization parameters settings that are evaluated in the
DMAIC Improve phase. Initial solutions when the online process control is started are shown in Figure 2 (a) and
Figure 2 (b). Figure 2 (a) and Figure 2 (b) represent real-time values for the ordering cost and holding cost objective
functions, respectively. Figure (a) illustrates the early stage of running the online control after 38 daily realizations
of demand D. At the early stage, both objective functions are in control and no shifts in the mean are observed
during the series demand realizations. After 40 observations, outlier events are detected when both objective
functions values exceed their respective UCLs. At this point, when an out-of-control event is detected, re-
optimization starts.


Figure 2. Results of online observations for inventory cost problem.

Figure 3 (a) and Figure 3 (b) show the process after 60 observations when an unnatural pattern and a shift in the
mean performance appear. Figure (a) illustrates the appearance of an unnatural pattern while monitoring the
ordering and holding costs. An unnatural pattern appears when six points decrease continuously for both objective
functions starting from sample observation 60. Furthermore, at the same sample observation, as shown in Figure 3
(b), shifts in the mean performance value for both objective functions are found with an 86% confidence level
(CL%). Figure 3 (b) illustrates the cumulative sum bootstraps for the original order of the sample observations S
i

drawn (in black) and compared with the other cumulative sum bootstraps after the observations are reordered. Thus,
this aids the decision-maker in detecting shifts in the mean value of both inventory cost functions. A mean shift
automatically invokes the re-optimization of the inventory cost functions. Then, the decision-maker selects from the
set of Pareto optima the new values for order quantity Q for the random customer demand D.
3379
Abualsauod and Geiger

At present, the online process control using control charts involves set of tests to detect unnatural patterns. Only a
selected set of these tests have been shown in this section; however, these selected tests are used for proof-of-
concept. In other words, the bootstrap methods shown in this research involves a large number of straps generated
after reordering the sample observations. All straps are used for calculating the CL% but only a select set of straps
are used and shown graphically for illustration in this paper.

Figure 3. Results of online process control and detection for unnatural patterns and shifts in the mean performance
value.

Finally, the proposed online process optimization and control approach returns an out-of-control process to an in-
control state. In other words, the optimization procedure generates a new set of Pareto optima that helps the
decision-maker to readjust the process. Figure 4 (a) through Figure 4 (d) show the Pareto frontier for the initial
solutions, after 15, 30 and 100 generations, respectively. From the figure, it can be seen that solutions are
converging fairly quickly and Pareto front is somewhat smooth, especially in Figure 4 (d).

3380
Abualsauod and Geiger



Figure 4. Multiple objective optimization results for the case inventory management problem.

At the Improve phase, the set of Pareto optima generated from the optimization are provided to the decision-maker
in order to select the new values for the decision variable Q.

4. Summary and Future Work
Multiobjective optimization is a viable approach for managing the tradeoffs involved with improving process
performance, including using it to enhance the Six Sigma DMAIC framework. Although several researchers
integrate multiobjective optimization to improve the quality of solutions of the Six Sigma methodology, there is
limited work that addresses the use of multiobjective optimization techniques to sustain improvement after
implementing the DMAIC framework [30, 31]. The objective of this study is to present and implement a framework
that has the ability to provide an online feedback from the Control phase to Improve phase of the Six Sigma DMAIC
framework for multiple objectives. When a process is out of control, the proposed approach generates a set of
tradeoff decision variable values that assist the decision-maker to sustain the process improvement from Six Sigma
project implementation.

The proposed framework is divided into three steps: 1) Improve phase, 2) Online process control, and 3) Online
optimization for multiple objectives. Furthermore, this study proposes that the implementation of the results from
the online process control feedback for out-of-control events are obtained by integrating control and optimization
procedures to improve inventory management process. Finally, it can be found from this study that, at present, the
control and optimization of inventory costs problem involves a set of decision variables.
3381
Abualsauod and Geiger


Future work may expand the proposed methodology implementation and experimental design investigation. Also,
performing online control based on six sigma operating level, which has not, yet, been conducted in the current
study.

References
1. Jacobs, F. R., Berry, W., Whybark, D. C., & Vollmann, T. (2011). Manufacturing Planning and Control for
Supply Chain Management (1st ed.). McGraw-Hill Professional.
2. Teng, J. T. (2002). On the economic order quantity under conditions of permissible delay in payments.
Journal of the Operational Research Society, 53(8), 915918.
3. Ferrin, D., & Muthler, D. (2002). Six Sigma and simulation, so what's the correlation? In Simulation
Conference, 2002 Proceedings of the Winter Simulation Conference, 2, 1439-1443.
4. Brady, J.E., & Allen, T.T. (2006). Six sigma literature: A review and agenda for future research. Quality and
reliability engineering International, 22(3), 335367.
5. Luce, K., Trepanier, L., Ciochetto, F., & Goldman, L. (2005). Simulation and optimization as effective
DFSS tools. In Simulation Conference, 2005 Proceedings of the Winter Simulation Conference, Retrieved
from 10.1109/WSC.2005.1574478
6. Tjahjono, B., Ball, P., Vitanov, V., Scorzafave, C., Nogueira, J., Calleja, J., Minguet, M. (2010). Six Sigma:
a literature review. International Journal of Lean Six Sigma, 1(3), 216-233.
7. Xinying Liu, Shuhong Wang, Jie Qiu, Jian Guo Zhu, Youguang Guo, & Zhi Wei Lin. (2008). Robust
Optimization in HTS Cable Based on Design for Six Sigma. IEEE Transactions on Magnetics, 44(6), 978-
981.
8. Hu, G., Wang, L., Fetch, S., & Bidanda, B. (2008). A multi-objective model for project portfolio selection to
implement lean and Six Sigma concepts. International Journal of Production Research, 46(23), 66116625.
9. Shimoyama, K., Oyama, A., & Fujii, K. (2007). Multi-objective six sigma approach applied to robust airfoil
design for Mars airplane. AIAA Paper, 1966.
10. Shimoyama, K., Oyama, A., & Fujii, K. (2005). A new efficient and useful robust optimization approach
design for multi-objective six sigma. In Proceedings of the 2005 IEEE Congress on Evolutionary
Computation, 1, 950957.
11. Shimoyama, K., Oyama, A., & Fujii, K. (2008). Development of Multiobjective Six-Sigma Approach for
Robust Design Optimization. Journal of Aerospace Computing, Information, and Communication, 5, 1-27.
12. Anand, R.B., & Shukla, S.K. (2007). Six sigma-based approach to optimize deep drawing operation
variables. International Journal of Production Research, 45(10), 23652385.
13. Azzabi, L., Ayadi, D., Boujelbenne, Y., Kobi, A., Robledo, C., & Chabchoub, H. (2009). Six Sigma based
multicriteria approach to improve decision settings. International Journal of Quality Engineering and
Technology, 1(1), 99123.
14. Dassau, E., & Lewin, D. R. (2006). Optimization-based Root Cause Analysis (RCA). Computer Aided
Chemical Engineering, 21, 943-948.
15. Knowles, G., Whicker, L., Femat, J.H., & Canales, F.D. (2005). A conceptual model for the application of
Six Sigma methodologies to supply chain improvement. International Journal of Logistics Research and
Applications, 8(1), 5165.
16. Shirazi, B., Fazlollahtabar, H., & Mahdavi, I. (2010). A six sigma based multi-objective optimization for
machine grouping control in flexible cellular manufacturing systems with guide-path flexibility. Advances in
Engineering Software, 41(6), 865-873.
17. Anand, R.B., & Shukla, S.K. (2007). Six sigma-based approach to optimize deep drawing operation
variables. International Journal of Production Research, 45(10), 23652385.
18. Cong, M., Han, T., & Zhao, Q. (2010). Fuzzy multi-objective optimization of sliding rack based on six
sigma and goal driven. In Proceedings Mechanic Automation and Control Engineering (MACE), 2010
International Conference, 556-559.
19. Ono, S., Yoshitake, Y., & Nakayama, S. (2009). Robust optimization using multi-objective particle swarm
optimization. Artificial Life and Robotics, 14(2), 174177.
20. Aparisi, F., de Luna, M. A., & Epprecht, E. (2010). Optimisation of a set of or principal components control
charts using genetic algorithms. International Journal of Production Research, 48(18), 5345.
doi:10.1080/00207540903095426
21. Epprecht, E., Aparisi, F., & others. (2010). A computer optimization of a set of EWMA quality control
charts. In Proceedings of the 9th WSEAS international conference on Applications of Computer
3382
Abualsauod and Geiger

Engineering, 9398.
22. Kaya, I. (2009). A genetic algorithm approach to determine the sample size for attribute control charts.
Information Sciences, 179(10), 1552-1566. doi:10.1016/j.ins.2008.09.024
23. Bakir, M., & Altunkaynak, B. (2004). The Optimization with the Genetic Algorithm Approach of the Multi-
Objective, Joint Economical Design of the mean and R Control Charts. Journal of Applied Statistics, 31(7),
753-772. doi:10.1080/0266476042000214484
24. Shiau, Y., Lin, M., Chung, I., & Pan, Y. (2006). Monitoring capability study and genetic economic design of
X-bar control chart. The International Journal of Advanced Manufacturing Technology, 30(3), 273-278.
25. Perez, E., Carrion, A., Jabaloyes, J., & Aparisi, F. (2010). Optimization of the new DS-u control chart: an
application of genetic algorithms. In Proceedings of the 9th WSEAS international conference on
Applications of Computer Engineering, 105109.
26. Jacobs, F. R., Berry, W., Whybark, D. C., & Vollmann, T. (2011). Manufacturing Planning and Control for
Supply Chain Management (1st ed.). McGraw-Hill Professional.
27. Cheng, T. C. E. (1991). An economic order quantity model with demand-dependent unit production cost and
imperfect production processes. IIE transactions, 23(1), 2328.
28. Stevenson, W. J., & Hojati, M. (2002). Operations management (Vol. 8). McGraw-Hill/Irwin New York.
Retrieved from http://www.eng.uwi.tt/depts/mech/ugrad/courses/meng3006/Week09b.pdf
29. Belgasmi, N., Ben Sad, L., & Ghdira, K. (2008). Evolutionary multiobjective optimization of the multi-
location transshipment problem. Operational Research, 8(2), 167183.
30. Sahoo, A. K., Tiwari, M. K., & Mileham, A. R. (2008). Six Sigma based approach to optimize radial forging
operation variables. Journal of materials processing technology, 202(1-3), 125136.
31. Aboelmaged, M. G. (2010). Six Sigma quality: a structured review and implications for future research.
International Journal of Quality & Reliability Management, 27(3), 268317.
doi:10.1108/02656711011023294.
3383
Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.

You might also like