Hybrid Kanban/Conwip Control System Simulation

and Optimization Based on Theory of Constraints

Yanhong Wang Jun Cao Lixin Kong

School of Information Science & Engineering School of Information Science & Engineering
Shenyang University of Technology Shenyang University of Technology
Control Theory and Control Engineering Control Theory and Control Engineering
Shenyang, People's Republic of China Shenyang, People's Republic of China

Abstract—the Pull type production control system has been of a However, since the inventory levels inside the Conwip system
continuing interest and challenge for researchers and are not controlled individually, if there is a bottleneck machine,
practitioners since its advent. However, the bottleneck problems high inventories can appear in front of the bottleneck, just like
under this mechanism have not yet been studied in–depth. This a push type system. To solve this problem, Bonvik in 1997,
approach proposed a Hybrid Knaban/Conwip control system, proposed a hybrid Kanban/Conwip control strategy [2]. In
which combined the advantages of Conwip and Kanban control hybrid Kanban/Conwip, as in Conwip, an overall cap is placed
strategy, as well as the theory of constraints (TOC). The hybrid on the amount of inventory allowed in the production system.
control system model is formulated with respect to DBR (Drum- In addition, inventory is controlled using Kanbans in all stages
Buffer-Rope) technique and a novel TOC based bottleneck
except the last stage. The hybrid system combines the
resources discovery and scheduling mechanism is presented. To
validate its effectiveness, simulations under witness environment
advantages of Conwip, i.e. high throughput with a low overall
were done and the comparison among the Conwip, Conwip/pull WIP level, with those of Kanban, i.e. control of inventory
and integrated control system was given. The simulation results levels at each stage, so it can solve a large number of
showed that the method was indeed able to solve the bottleneck inventories in the products as a result of mechanical failure in
problem and greatly improve the productivity and reduced the the Conwip or other pull type control system. After Bonvik,
delay time of the production system. several models that implement hybrid control strategies have
been developed. However, most of the literatures has merely
The simulation and comparison results showed that the proposed introduced or explained the concept of the hybrid model, the
hybrid system can solve the bottleneck problem effectively with system parameters optimization or other related problem have
enhancing of the system and reducing the delay time. not yet been studied in–depth.
The bottleneck resource is one of the crucial factors for
controlling the production line/system [3]. It is well know that
if the bottleneck resources wasted an hour, it will equivalent to
a waste of the whole system about one hour. However, it is not
easy to control this bottleneck resource due to the complexity
Keywords-component; bottleneck problem; TOC; Hybrid
and randomness of the actual system. While TOC (Theory of
Conwip/pull control system; WITNESS; improved simulated
annealing algorithm (key words) Constant) has the advantage of bottleneck allocation and
control, it will contribute to balance the production logistics
and improve the throughput of the critical resource. In this
I. INTRODUCTION article, a TOC based hybrid system is proposed to focuses on
critical resource optimization in the production control
Due to the technological advancement, the conventional
processes. The TOC control mechanism was described and a
method of push production system was changed to pull type
model is formulated with respect to DBR (Drum-Buffer-Rope)
production system to meet out the global competition, where
technique. An improved simulated annealing algorithm
the work-in-process (WIP) can be managed and controlled
more accurately than the push-production system. Conwip approach is also applied for the TOC based Kanban/Conwip
(Constant work in process) has received a lot of research system to find the optimum parameter setting. Besides
attention since it was first proposed by Spearman in 1990[1]. simulation result of the proposed TOC based hybrid system
Conwip is a generalization of the Kanban system, but it uses a under Witness environment was showed out and the
single card type to control the total amount of WIP permitted in comparison among the Conwip, Conwip/pull and integrated
the entire production line. So Conwip is applicable to a wider control system was given.
variety of production environments, while Kanban is just for II. TOC BASED HYBRID CONTROL SYSTEM MODEL
repetitive manufacturing. Furthermore, it has been
demonstrated by many studies that Conwip has preferable TOC (theory of constraints) originated from the OPT
performance than Kanban or other types of pull systems. (Optimized Production Technology). The central issues of
_____________________________
978-1-4244-4738-1/09/$25.00 ©2009 IEEE


TOC were how to identify and break through the bottleneck so model and methods based on an improved simulated annealing
as to make the manufacturing enterprises the maximization of algorithm are applied in this section. The traditional simulated
the production rate. For enhancing the productivity of the annealing algorithm is a global optimum algorithm, its
system, the basic tool for production management of TOC convergence rate is much slower, in this paper a fast simulated
theory is DBR (Drum-Buffer-Rope) technique. The main annealing algorithm (VFSA) is used, which was inspired his
functions of DBR includes: 1) To identify and control Ingber’s original approach[4]. We made use of Cauchy
bottlenecks (drum), 2) To protect the bottleneck from the distribution function to replace Gaussian distribution function
random fluctuations by setting up a buffer, 3) To use control of traditional simulated annealing algorithm by Witness
mechanisms of ropes and make of other aspects software own .it generated a new model with the help of the
synchronization with bottlenecks in processes. Based on TOC new distribution function by disturbance .Similarly, we utilized
and DBR, a Kanban/Conwip hybrid control system is proposed the acceptance probability formula and the annealing algorithm
in this paper, which is shown in Fig.1. of VFSA inside to achieve the reception and annealing
scheme .This can effectively improve the system simulation
efficiency .Here, Cauchy distribution is adopted, which
depended on the temperature.
mic mi  y i Bi  Ai  (1)

yi >
T u sgn u  0.51  1 / T 
2 u 1
@
1 (2)

Where, mi is variables i at the current model, u is a random

number created in even distribution [0, 1], the range of mi is
Figure 1. TOC based Hybrid Control System
[Ai, Bi], mci is variables i in the new model after disturbance

Here, we assume there is an assembly production process in variables of the first i, in addition, mic belongs to [Ai, Bi]
which the assembly station is fed by two fabrication lines. Each The new model is generated according to Cauchy
workstation is a production (or assembly)/inventory system distribution. Thus the advantages of this method is that it only
made up of a manufacturing (or assembly) process. The search near the current model at the lower temperatures, and
manufacturing process may consist of a single machine or a the search range get larger at the high temperatures. Moreover,
sub-network of several machines. In the above figure, it has a “flat tail”, which makes it easy to quickly jump out of
manufacturing/assembling processes are drawn as circles; the local extremum.
buffers are drawn as rectangles. While solid lines represent A new acceptance probability by Boltzmann Gibbs
material flows and the dotted lines represent the kanban cards distribution is obtained. The formulation is as follows.
movement; the short curve denotes the Rope in TOC/DBR.
The workstation wi is supported as a bottleneck resource in
the production/assembly line. Whereupon, a Time buffer TB
P >1  1  h 'E y T @1y1h  (3)
was located before the bottleneck in order to relieve the 'E E m   E m0  (4)
material congestion. Meanwhile an assembly buffer AB is also
set before the last but one workstation, just before the final Where, h is real number. h  >0,1@ .
assembly workstation, in another sub-line. The control Finally, we come to achieve the annealing scheme, the
mechanisms of the rope were adopted to control the delivery of formulation is as follows
materials before the bottleneck resource, and the push-type
control would be applied on the downstream station of the

T K  T0 exp  ck 1 / N  (4)
bottleneck in order to coordinate the production processes of Where, T0 is the initial temperature; k is iteration number, c
the two sub-lines. Thus the hybrid Conwip/pull control is a constant, N is the number of inversion parameters. We used
mechanism is attached to the preceding workstations which are equation 1 and equation 2 to generate a new model by
located before the bottleneck resource in the sub-line. In our disturbance. Then, the new model received dates by equation 3
hybrid system, the rope is carried out by Conwip cards so as to and 4, and then dropped temperature slowly by equation 5.
control the production process efficiently and effectively. This
is a novel production control mode of push-pull hybrid control IV. SIMULATION AND COMPARISON AMONG THE HYBRID
strategy, which is capable to manage bottleneck resources and SYSTEM AND OTHERS
control production steps and make the material flow moved
smoothly and swimmingly. Computer-aided simulation is a powerful tool to verify the
validity for a method especially in the area of performance
III. PARAMETERS SETTING OF THE HYBRID SYSTEMS evaluation or improvement in manufacturing environments. We
made a large of simulations to optimal the number of cards
The main parameters for the single-loop and the multi-loop under a discrete system simulation tool Witness in order to
Conwip systems are numbers of the cards. Respect to the goal verify the validity of the proposed approach.
of the production system, approaches that enable the
parameters setting of the systems is needed. An optimization


A. Simulation and optimization strategies was written by Yachao Wang[6]. The data of production
A three-step approach was adopted to identify the system derived from the above-mentioned book. The hybrid
bottleneck and quantify the proposed TOC base hybrid Kanban/Conwip control production system considered in this
production control techniques on system performance and research is for a mechanical parts assembly line, which consists
optimization. of twelve processing centers (as shown as figure 3).The
assembly station are WS11 and WS12, which are fed by two
Step I: Simulation modeling fabrication sub-lines, one sub-line is made up of five
Step II: Bottleneck identifying and buffer position locating workstations, defined from WS1 to WS5, and the other is
Step III: Model Optimization composed of three a workstations, named from WS7 to WS9.
In step I, simulation model will be established and random In the following figure, workstations are drawn as rectangles;
parameters will be introduced, according to the capacity data the buffers are drawn as rectangles and the two cylindrical
and procedure data of a given production system and its task- containers represent Standard containers. While solid lines
specific combinations. represent material flows and the dotted lines represent the
In step II, the bottleneck will be determined and buffer kanban cards movement.
position will be located by running the simulation model via
setting up certain load rate and busy rate of various resources,
where the randomness of real manufacturing system will be
taken into account.
In step III, the bottleneck resources and their related
resources will be re-scheduling in order to make the production
line work in an ideal state. Scheduling issue of bottleneck
resources played a vital role in the TOC. For production
resources scheduling, the objective function and schedule rules
should be set. In the actual production environment, the
manufacturing objectives and demands will change from time
to time, and the optimal scheduling goals and strategies will
changed accordingly. While in order to simplify the model and
taking into account the general situation, the average delay time
of orders, the system output, the WIP, and the operation rate
were set as the objective function for the bottleneck scheduling.
In addition, some well-known scheduling rules were used such
as earliest due-date (EDD), shortest processing time (SPT) and Figure 3. Simulation model for the assembly system
so on in the simulating processes [5].
Details of the above three steps of the simulation was We established the simulation model through discrete event
shown as the following diagram. system modeling and simulation software Witness. The
simulation assumptions are set as follows:
x The cycle time of the experiment was set as 120000
min (250*8*60=120000 min) to simulate the working
period of the production line for one year, and the
simulation warm-up time was 310 min.
x To simplify the description, we assumed that the
production process was discrete event process and
subjected to¤ distribution.
x The transportation time between workstations is
negligible, and raw materials and parts are always
available at stations. Besides, information, such as
kanbans, is transmitted instantly.
x Jobs authorized for loading follow a first-come, first-
serve (FCFS) dispatching policy at all stations. And the
system objective function was the average waiting time
of orders (dmdwaittime), WIP, operation rate, system
output, and output rate.

Figure 2. Simulation Flow Diagram C. Simulation results

We do three classes simulation under the above assembly
B. Problem definition and Assumptions of the model system model which was specified as the following three cases.
The hybrid Conwip/pull control system prototype Case_1: The proposed hybrid Kanban/Conwip control
originated lj production logistics system modeling and system optimization relate to the number of cards.
simulation system- Witness system and applicationNJ,which


 Case_2: The proposed hybrid system optimization, under buffer coefficient R was acquired with the help of the improved
different bottleneck resources scheduling rule. simulated annealing algorithm and buffer time formulation, it
Case_3: Performance comparison among Conwip, was 2.8. Through the above-mentioned method of setup buffer
Conwip/pull and our hybrid system. time, one time buffer was set up between WS4 and WS5, and
In the simulation case_1, to optimize the number of cards in the other was set between WS9 and WS10, while we
the proposed hybrid system, an improved simulated annealing established the assembly buffer before WS11 and WS12.
algorithm was used to obtain the maximum production rate for Finally, several schedule rules were applied to find which rule
corresponding card number. The curve of card numbers was best was by identical the average delay time and the average
plotted as shown in Fig. 4. It is obviously shown that when the flow time of different orders. The average flow time under
card number increases, the production rate also increases. different rules on bottleneck resources were followed in the
However, when the card number is over 486, the rate did not Table 1. As was shown in the table, we found the average delay
increase any more and any additional card only results in time of orders was least under the rule of EDD, while it was
higher WIP. Thus, 486 is used as the rational number of cards largest under SPT rule; however, the average flow time was
for the hybrid system. least under the rule FCFS. Hence, we should choose different
scheduling strategies according to different optimal objective.

TABLE 1. Simulation results about different rules

In the last case_3, simulations were done to compare the

performances among Conwip, Kanban/Conwip and our hybrid
system under the rational card number by simulation. Table 2
shows the simulation results for cases_3 which gives the
optimal performance measures. From the table it can be
inferred that the hybrid system provides better performance
compared with Conwip and Kanban/Conwip with a lowest WIP
and highest output rate. It shows that for an assembly system,
the proposed hybrid system which introduced TOC to tackle
bottleneck resources can have the best performance among the
three. It is an effective system for the assembly production line.

Figure 4. Production feature curve TABLE 2. Three systems simulation results

comparison table

From large amount of simulation, we find that even though

the conventional hybrid Kanban/Conwip control system
maintained lower inventories, higher system output rate and
operation rate, and shorter average delay time of orders than
Conwip system, however, the delay time was much higher for
enterprise yet. Besides, the hybrid control system can reduce
inventory in the production line, but it can not effectively
improve the output rate. We also found that conventional
Kanban/Conwip failed to effectively handle the bottlenecks
resource through analysis. So in the simulation case_2, the
bottleneck resources were detected out and some scheduling
rule were used to control the production processes of the entire
assemble line. Firstly, bottleneck resources were determined by
examine the load rate and busy rate of all resources. After CONCLUSION
simulation, WS5 and WS10 which had more higher load rate
In this paper, we proposed a TOC based Kanban/Conwip
and busy rate ˈ therefore, these two workstations were
hybrid system to tackle the production line control and
identified as the bottleneck process. Then, some buffers would optimization problems, which pay more attention to the
be set to enhance the utilization. The initial value scope of the bottleneck resources, a situation that differs from that of other
buffer coefficient R was set as [1, 4] and then the optimal the


current researches. In order to accomplish this, a model was REFERENCES
formulated with respect to the output rate and the average delay
time. Taking into consideration that TOC had advantages of [1] Spearman, M. L., Woodruff, D. L., & Hopp, W. J. (1990).CONWIP:A
dealing with bottleneck problems, we introduced TOC to the pull alternative to kanban. International Journal of Production Research,
hybrid control system to establish hybrid control integrated 28, 879–894.
system based on TOC. The simulation results showed that the [2] Bonvik A M,Couch C E,and Gershwin S B, “A comparison of production
proposed hybrid system can solve the bottleneck problem line control mechanism,” Int. J. Production Economics. 1997,vol
effectively and enhanced the productivity of the system and 35,pp.789-804.
reduced the delay time greatly by the TOC based novel [3] Rui Zhang and Cheng Wu, “Bottleneck identification procedures for the
job shop scheduling problem with application to genetic algorithms,” Int
bottleneck resources discovery and scheduling mechanism. J Adv Manuf Technol.2007,vol 5,pp.87-92.
In conclusion, the results have clearly illustrated that the [4] Ingber L, “Very Fast simulation annealing,” Math Conput
TOC based hybrid Kanban/CONWIP system offers better Modeling.1989,pp.867-973
performances than the others for the assembly production line [5] Hua Yu and Suying Yang,“Production Scheduling Problem Based on
with respect to output rate and WIP level. The average flow Theory of Constraints and Arithmetic Intelligence,”Computer Integrated
time is lower and the output rate is guaranteed in the proposed Manufacturing System.2005,vol 8,pp.165-182
hybrid system. Therefore, the TOC based hybrid system is a [6] Yachao Wang and Hanwu Ma, “production logistics system modeling
more effective production control system for the assembly and simulation system- Witness system and application,” 2006,pp.237-
270
production line.