155

Chapter 6 On Just-In-Time
Production Leveling
6.1 Publishing Info
Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

1 1
Francesco Giordano and Massimiliano M. Schiraldi

[1]
Department of Enterprise Engineering, “Tor Vergata” University of Rome, Italy

6.2 Introduction
Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

Since the 80’s, the Japanese production techniques and philosophies spread among
the Western manufacturing companies. This was possible because the Toyota Motor
Company experience was indeed a success. The so-called “Toyota Production System”
(TPS) seemed to be the “one best way” to manage a manufacturing production site.

On the other side, it is also well known that not every implementation of Lean
Production was a success, especially in Western companies: some enterprises –
together with the consultancy firms that should have supported them – forgot that
there are some main hypotheses and issues to comply with, in order to achieve
Toyota-like results. On top of this, certain requisites are not related to a mere
managerial approach, but depend on exogenous conditions, e.g. market behavior or
supplier location; thus, not every company can successfully implement a TPS system.

One critical requirement for a TPS approach to be effective is that the production plan
should be leveled both in quantity and in mix. This is indicated by the Japanese term
heijunka (平準化), which stands for “leveling” or “smoothing”. Here, we will focus our
attention on why leveled production is a key factor for JIT implementation, and
specifically we will describe and analyze some approaches to deal with the leveling
problem.

At first, the original Toyota Production System is briefly recalled, with specific regard to
the Just In Time (JIT) approach to manage inventories in production. JIT is a stock
replenishment policy that aims to reduce final product stocks and work-in-process
(WIP); it coordinates requirements and replenishments in order to minimize stock-
buffer needs, and it has reversed the old make-to-stock production approach, leading
most companies to adopt "pull” instead of “push” policies to manage material and
finished product flows. However, in case of unleveled demand, stock levels in JIT may
grow uncontrolled.
156

Secondly, kanban-based production is described: kanban, a Japanese word meaning

“visual record”, is a card that contains information on a product in a given stage of the
manufacturing process, and details on its path of completion. It is acknowledged as
one of the most famous technique for material management in the JIT approach. Here
we will present some common algorithms for managing kanban queues, along with
their criticalities in terms of production smoothing requirements and reduced
demand stochasticity. Some of the JIT-derivative approaches will be recalled as well:
CONWIP, Quick Response Manufacturing, Theory of Constraints and the Just-In-
Sequence approach.

Then, a review on the mixed-model JIT scheduling problem (MMJIT), along with the
related solving approaches, is presented. Despite the huge literature on MMJIT
mathematical programming approaches, here it will be described why the real-world
production systems still prefer the simpler kanban approach and the old (1983) Goal
Chasing Method algorithm. In the end, an overview on simulators advantages to test
alternative heuristics to manage JIT production is presented.

6.3 Managing Just-In-Time production systems

Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

Just-in-Time was first proposed within the Toyota Production System (TPS) by Taiichi
Ohno after the 50’s when he conceived a more convenient way to manage inventory
1
and control production systems . Lean Production – the un-branded name of TPS – is a
mix of a philosophy for production systems management and a collection of tools to
2
improve the enterprise performances . Its cornerstones are the reduction of muda
3
(wastes), mura (unevenness) and muri (overburden). Ohno identified seven wastes
that should be reduced to maximize the return of investment of a production site:

• transportation;
• inventory;
• motion;
• waiting;
• over-processing;
• over-producing;
• defects.

The TPS catchphrase emphasizes the “zero” concept: zero machine changeovers (“set-
ups”), zero defects in the finished products, zero inventories, zero production stops,
zero bureaucracy, zero misalignments. This result may be reached through a
continuous improvement activity, which takes cue from Deming’s Plan-Do-Check-Act
4
cycle : the kaizen approach.

1. J. P. Womack, D. T. Jones & D. Roos, The machine that changed the world: The Story of Lean Production, New York (NY):
HarperPerennial, 1991.
2. M. Rother & J. Shook, Learning to See: Value Stream Mapping to Add Value and Eliminate Muda, Brookline (MA): The
Lean Enterprise Institute, 1999.
3. T. Ohno, Toyota Production System: Beyond Large-Scale Production, Productivity Press, 1988.
4. J. P. Womack, D. T. Jones & D. Roos, The machine that changed the world: The Story of Lean Production, New York (NY):
HarperPerennial, 1991.
 157

Just-In-Time is the TPS solution to reduce inventory and waiting times. Its name,
5
according to , was coined by Toyota managers to indicate a method aimed to ensure
“the right products, in the right quantities, just in time, where they are needed”.
Differently from Orlicky’s Material Requirement Planning (MRP) – which schedules the
6
production run in advance compared to the moment in which a product is required –
JIT approach will replenish a stock only after its depletion. Among its pillars there are:

• one-piece flow;
• mixed-model production;
• demand-pull production;
• takt time;

Indeed, generally speaking, processing a 10 product-batch requires one tenth of the

time needed for a 100 product-batch. Thus, reducing the batch value (up to “one
piece”) would generate benefits in reducing either time-to-market or inventory level.
This rule must come along with mixed-model production, which is the ability of
manufacture different products alternating very small batches on shared resources.
Demand-pull production indicates that the system is activated only after an order
receipt; thus, no semi-finished product is processed if no downstream workstation
asks for it. On top of this, in order to smooth out the material flow, the process
operations should be organized to let each workstation complete different jobs in
similar cycle times. The base reference is, thus, the takt time, a term derived from the
German word taktzeit (cycle time), which is computed as a rapport between the net
operating time, available for production, and the demand in terms of units required.
These are the main differences between the look-ahead MRP and the look-back JIT
system. For example, the MRP algorithm includes a lot-sizing phase, which results in
product batching; this tends to generate higher stock levels compared to the JIT
7
approach. Several studies have been carried out on MRP lot-sizing and trying to
8 9 10
improve the algorithm performance , , ; however, it seems that JIT can outperform
MRP given the heijunka condition, in case of leveled production both in quantity and in
mix. The traditional JIT technique to manage production flow is named kanban.

6.4 The kanban technique

Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

11
A kanban system is a multistage production scheduling and inventory control system .
Kanban cards are used to control production flow and inventories, keeping a reduced
production lead time and work-in-process. Clearly, a kanban is not necessarily a

5. R. J. Schonberg, Japanese manufacturing techniques: Nine hidden lessons in simplicity, New York (NY): Free Press, 1982.
6. J. Orlicky, Material Requirement Planning, New York (NY): McGraw-Hill, 1975.
7. L. Baciarello, M. D’Avino, R. Onori & M. Schiraldi, « Lot-Sizing Heuristics Performance» working paper, 2013.
8. A. Bregni, M. D’Avino e M. Schiraldi, «A revised and improved version of the MRP algorithm: Rev MRP,» Advanced
Materials Research (forthcoming), 2013.
9. M. D’Avino, V. De Simone & M. Schiraldi, «Revised MRP for reducing inventory level and smoothing order releases: a case
in manufacturing industry,» Production Planning & Control, 2013 (forthcoming).
10. M. D’Avino, M. Correale & M. Schiraldi, «No news, good news: positive impacts of delayed information in MRP» working
paper, 2013.
11. C. Sendil Kumar & R. Panneerselvam, «Literature review of JIT-KANBAN system,» International Journal of Advanced
Manufacturing Technologies, pp. 393-408, 2007.
158

physical paper/plastic card, as it can be either electronic or represented by the

container itself.

Since it was conceived as an easy and cheap way to control inventory levels, many
different implementations of kanban systems have been experimented in
manufacturing companies all over the world. In the following paragraphs, the most
commonly used “one/two cards” kanban systems are described.

6.4.1 One-card kanban system

Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

The “one-card” is the simplest implementation of kanban systems. This approach is

used when the upstream and downstream workstations (respectively, the preceding
and succeeding processes) are physically close to each other, so they can share the
12 13
same stock buffer. The card is called “Production Order Kanban” (POK) , . The stock
buffer acts either as the outbound buffer for the first (A) workstation or as the
inbound buffer for the second (B) workstation. A schematic diagram of a one-card
system is shown in Figure 6.1.

Fig. 6.1: A one-card kanban system

Here, each container (the JIT unit load) has a POK attached, indicating the quantity of a
certain material contained, along with eventual complementary information. The POK
also represents a production order for the Workstation A, indicating to replenish the

12. B. J. Berkley, «A review of the kanban production control research literature,» Production and Operations Management,
vol. 1, n. 4, pp. 393-411, 1992.
13. B. Sharadapriyadarshini & R. Chandrasekharan, «Heuristics for scheduling in a Kanban system with dual blocking
mechanisms,» European Journal of Operational Research, vol. 103, n. 3, pp. 439-452, 1997.
 159

container with the same quantity. When a B operator withdraws a container from the
buffer, he removes the POK from the container and posts it on a board. Hence, A
operator knows that one container with a specific part-number must be replenished in
the stock buffer.

6.4.2 Two-card kanban system

Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

In the two-card system, each workstation has separate inbound and outbound
14 15
buffers , . Two different types of cards are used: Production Order Kanbans (POK)
and Withdrawal Kanbans (WK). A WK contains information on how much material (raw
materials / semi-finished materials) the succeeding process should withdraw. A
schematic diagram of a two-card system is shown in Figure 6.2.

Each work-in-progress (WIP) container in the inbound buffer has a WK attached, as

well as each WIP in the outbound buffer has a POK. WK and POK are paired, i.e. each
given part number is always reported both in n POK and n WK. When a container is
withdrawn from the inbound buffer, the B operator posts the WK on the WK board.
Then, a warehouse-keeper operator uses the WK board as a picking list to replenish
the inbound buffer: he takes the WK off the board and look for the paired POK in the
outbound buffer. Then, he moves the corresponding quantity of the indicated
material from the A outbound to the B inbound buffer, while exchanging the related
POK with the WK on the container, restoring the initial situation. Finally, he posts the
left POK on the POK board. Hence, like in the previous scenario, A workstation
operator knows that one container of that kind must be replenished in the outbound
stock buffer. The effectiveness of this simple technique – which was described in
16 17 18 19
details by several authors , , , – is significantly influenced by the policy followed
to determine the kanban processing order, in the boards.

14. O. Kimura & H. Terada, «Design and analysis of pull system, a method of multistage production control,» International
Journal of Production Research, n. 19, pp. 241-253, 1981.
15. B. Hemamalini & C. Rajendran, «Determination of the number of containers, production kanbans and withdrawal
kanbans; and scheduling in kanban flowshops,» International Journal of Production Research, vol. 38, n. 11, pp.
2549-2572, 2000.
16. T. Ohno, Toyota Production System: Beyond Large-Scale Production, Productivity Press, 1988.
17. B. Hemamalini & C. Rajendran, «Determination of the number of containers, production kanbans and withdrawal
kanbans; and scheduling in kanban flowshops,» International Journal of Production Research, vol. 38, n. 11, pp.
2549-2572, 2000.
18. R. Panneerselvam, Production and Operations Management, New Delhi: Prentice Hall of India, 1999.
19. H. Wang & H.-P. B. Wang, «Optimum number of kanbans between two adjacent workstations in a JIT system,»
International Journal of Production Economics, vol. 22, n. 3, pp. 179-188, 1991.
160

Fig. 6.2: A two-card kanban system

6.4.3 Standard approaches to manage the kanban board

Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

From the previously described procedure, it is clear that the each workstation bases
its production sequence on kanban cards posted on the POK board. In literature, few
traditional ways to manage the board are reported: each of them is quite easy to
implement and does not require significant investments in technology or other
expensive assets.
20
The most commonly used policy requires having a board for each station, and this
should be managed as a single First-In-First-Out (FIFO) queue. The board is usually
structured as one vector (one column, multiple rows): POK are posted on the board in
the last row. Rows are grouped in three zones (red/yellow/green) which indicate three
levels of urgency (respectively, high/medium/low). Kanban are progressively moved
from the green to the red zone and the workstation operator will process the topmost
kanban. If a kanban reaches the red rows, it means that the correspondent material is
likely to be requested soon, by the succeeding process. Thus, it should be urgently
replenished in the outbound buffer, in order to avoid stock-outs.

Although this policy does not rely on any optimized procedure, it may ensure a leveled
production rate in each workstation, given the fact that other TPS pillars are
implemented, e.g. setup time reduction and mixed model scheduling. Indeed, if the
final downstream demand is leveled, the production plan of the workstations will be
leveled as well. Clearly, this policy is vulnerable to high setup times and differences
among workstations cycle times: in this latter case, indeed, the ideal jobs sequence for
a workstation may be far from optimal for the preceding. It is noticeable that the
colored zones on the board only provide a visual support for the operators and do not
influence the jobs processing order.

20. T. Ohno, Toyota Production System: Beyond Large-Scale Production, Productivity Press, 1988.
 161

A heijunka box is a sort of enhanced kanban board: it still acts as a visual scheduling
tool to obtain production leveling at the workstations. However, differently from the
traditional board, it manages to keep evidence of materials distinctions. Usually, it is
represented as a grid-shaped wall schedule. Analogously to the simpler board, each
row represents a time interval (usually, 30-60 minutes), but multiple columns are
present, each one associated to a different material. POKs are placed in the so-called
“pigeon-holes” within the box, based on number of items to be processed in the job
and on the material type. Workstation operators will process all the kanban placed in
the current period row, removing them from the box. Hence, heijunka box not only
provides a representation for each job queued for production, but for its scheduled
time as well, and allows operators to pursue production leveling when inserting new
POKs in the boxes.

6.4.4 Criticism on JIT

Available under Creative Commons-ShareAlike 4.0 International License (http://
creativecommons.org/licenses/by-sa/4.0/).

21
During the last decades, Just-In-Time has been criticized from different authors .
Indeed, certain specific conditions – which, though, are not uncommon in
manufacturing companies – can put in evidence some well-known weak points of the
Japanese approach. Specifically, un-steady demand in multi-product environments
where differences in processing lead times are not negligible represent a scenario
where JIT would miserably fail, despite the commitment of the operations managers.

First, we have to keep in mind that one pillar of Lean Production is the “one-piece-
flow” diktat. A one-piece batch would comply with the Economic Production Quantity
22
theory only when order cost (i.e. setup time) is zero. Having non-negligible setup
times hampers JIT implementation and makes the production leveling problem even
more complicated. It is peculiar that, originally, operations researchers concentrated
on finding the best jobs sequence considering negligible setups time. This bound was
introduced into the mixed model kanban scheduling problem only since 2000. Setups
are inevitable in the Lean Production philosophy, but are considered already
optimized as well. Given that setup times are muda, TPS approach focuses on
quickening the setup time, e.g. through technical interventions on workstations or on
the setup process with SMED techniques, not on reducing their frequency: the
increased performance gained through setups frequency reduction is not worth the
flexibility loss that the system may suffer as a consequence. Indeed, the standard
kanban management system, ignoring the job sequencing, does not aim at reducing
setup wastes at all. Analogously, the Heijunka box was developed for leveling
production and can only assure that the product mix in the very short term
reproduces that in the long term; in its original application, the decision on the job
sequence is left to the operator. Only in some enhanced version, the sequence is pre-
defined applying some scheduling algorithm.

21. D. Y. Golhar & C. L. Stamm, «The just in time philosophy: a literature review,» International Journal of Production
Research,vol. 29, n. 4, pp. 657-676, 1991.
22. F. W. Harris, «How many parts to make at once,» Factory, The Magazine of Management, vol. 10, n. 2, pp. 135-136, 1913.
162

Given the fact that JIT is based on stock replenishment, constant production and
withdrawal rates should be ensured in order to avoid either stock outs or stock
proliferation. Mixed-model production requires a leveled Master Production Schedule
23
(MPS) , but this is not sufficient to smooth the production rate in a short time period.
While it is easy to obtain a leveled production in a medium or even medium-short
period, it is difficult to do it in each hour, for each workstation and each material.

Indeed, demand is typically unstable under two points of view: random frequency,
which is the chance that production orders are irregularly received, and random
quantities, which is related to product mix changes. Indeed, since TPS assume minimal
stock levels, the only chance to cope with demand peak is to recur to extra production
capacity. However, available production capacity should be higher than required as
the average (as TPS requires), but for sure cannot be limitless. Thus, the JIT
management system should anyway be able to consider the opportunity of varying
the maintenance plan as well as the setup scheduling, in case of need. On the other
hand, if the production site faces a leveled production, changes in product mix should
not represent a problem; however, they increase sequencing problem complexity.
Most of the operational research solutions for JIT scheduling are designed for a fixed
product mix, thus its changes can greatly affect the optimality of solutions, up to make
them useless.

On the contrary, kanban board mechanism is not influenced by demand randomness:

as long as demand variations are contained into a certain (small) interval, kanban-
managed workstations will handle their production almost without any problem.
Therefore, in case of unstable demand, in order to prevent stock-outs, inventory
managers can only increase the kanban number for each product: the greater are the
variations, the greater is the need of kanban cards and, thus, the higher is the stock
24 25
level. In order to prevent stock level raise, some authors , proposed to adopt a
frozen schedule to implement JIT production in real companies, where demand may
clearly be unstable. Anyway, this solution goes in the opposite direction compared to
JIT foundations.

Moreover, one-piece-flow conflicts with demand variability: the batch size should be
chosen as its processing time exceeds the inter-arrival time of materials requests.
Thus, the leveling algorithm must find the proper sequencing policy that, at the same
time, reduces the batch size and minimize the inter-arrival time of each material
request. This sequence clearly depends on the total demand of each material in the
planning horizon. However, JIT does not use forecasting, except during system design;
thus, scheduling may be refreshed daily. From a computational point of view, this is a
non-linear integer optimization problem (defined mixed-model just-in-time scheduling
problem, MMJIT), which has non-polynomial complexity and it currently cannot be
solved in an acceptable time. Thus, reliable suppliers and a clockwork supply chain are
absolutely required to implement JIT. Toyota faced this issue using various
26
approaches :

23. R. B. Chase, N. J. Aquilano & R. F. Jacobs, Operations management for competitive advantage, McGraw-Hill/Irwin, 2006.
24. R. Suri, «QRM and Polca: A winning combination for manufacturing enterprises in the 21st century,» Center for Quick
Response Manufacturing, Madison, 2003.
25. P. Ericksen, R. Suri, B. El-Jawhari & A. Armstrong, «Filling the Gap,» APICS Magazine, vol. 15, n. 2, pp. 27-31, 2005.
26. J. Liker, The Toyota Way: 14 Management principles from the world's greatest manufacturer, McGraw-Hill, 2003.