An overview of design and operational issues of kanban systems M. S. AKT

1
An overview of design and operational issues of
kanban systemsM. S. AKTÜRK and F. ERHUN

• Presented by
• Y. Levent KOÇAGA

2
CONTENT

• Introduction to JIT and kanban
• Literature Overview
• A model for sequencing production kanbans
• Conclusion

3
Introduction to JIT

• JIT is a manuf. policy with a very simple goal
• produce the required items
• at the required quality
• in the required quantities
• at the precise time they are required

4
JIT

• An ideal of having the necessary amount of
  material available where it is needed and when it
  is needed
• A pull system
• Effective in environments of high process
  reliability, low demand variability and setups

5
JIT benefits

• Reduced WIP and FGI
• Reduced lead times
• Higher quality, reduced scrap and rework
• Ability to keep schedules
• Increased flexibility
• Easier automation
• Higher utilization

6
Limitations of JIT

• Applicable mostly to repetetive manufacturing
• Final assembly schedule must be very level and
  stable
• Large information lead times

7
Just in Time

• JIT philosophy
• JIT techniques
• JIT shop floor control systems

8
Kanban

• Dual-card
• production kanban transportation kanban
• Single-card
• a schedule instead of production kanban
• Instantenous vs Periodic review
• Periodic review fixed quantity or fixed
  withdrawal cycle

9
Literature review

• Mathematical programming
• Markov Chain
• Simulation
• Other approaches

10
Solution methodology

• Solution approach is either exact or heuristic
• Exact approaches include dynamic programming, LP,
  IP, MIP or NIP

11
Model details (analytical)

• Decision variables are mainly
• kanban sizes
• number of kanbans
• withdrawal cycle length
• safety stock
• Objective is to minimize cost or inventories
• (maximizing throughput for stochastic models)

12
Model details (simulation)

• Performance measures used
• number of kanbans
• machine utilizations
• inventor holding cost
• backorder cost
• fill rate (probability that an order will be
  satisfied through inventory)

13
Settings of the models

• Production settings include
• layout
• number of time periods
• number of items
• number of stages
• capacity

14
Kanban system

• Singlecard or dual-card

15
Assumptions

• Kanban size (empty cell for decision variable)
• Nature of the system
• deterministic vs stochastic
• Production cycle
• continuous vs fixed intervals
• Material handling
• instantaneous vs periodic
• Backorders and reliability

16
Determining kanban sequences

• FAS determines prodn orders for all stages
• Once assembly line is scheduled it is assumed
  that the sequences propagate back
• Rest of kanbans scheduled by FCFS
• Some studies use simple dispatching rules

17
Determining kanban sequences

• Production levelling through scheduling is
  crucial
• Sequencing more complex because
• kanbans may not have specific due dates
• kanban controlled shops can have station
  blocking
• Sophisticated scheduling rules needed

18
Computational analysis

• Close interaction between design parameters
• such as
• number of kanbans
• kanban sizes
• kanban sequences

19
Computational analysis

• Thus an experimental design developed to
  determine
• the withdrawal cycle length
• number of kanbans
• kanban sizes
• and kanban sequences at each stage
  simultaneously for aperiodic review multi-item,
  multi-stage, multi-period kanban system

20
Computational analysis

• Objective is to minimize total production cost
• that is the sum of inventory holding and
  backorder costs over all stages
• Impact of operating issues such as sequencing and
  lead times on design parameters
• four sequencincing rules considered
• (SPT, SPT-F,FCFS,FCFS-F)
• Family based rules of FCFS andSPT/LATE

21
Model
22
Algorithms
23
Algorithms
24
Algorithms
25
Experimental factors
26
Toyota formula

• maximum inventory levelnaDL(1s)
• Lead time is not an attribute of the part
• Rather it is dependent on the shop floor
• Work-in-queue rule used for lead time estimation
• As lead times are estimated the maximum inventory
  level at each stage will change
• Thus the solution space increases

27
Results

• Effects of kanban sizes and number of kanbans and
  their interaction significant
• Therefore they are chosen so that MINVijm remains
  constant
• There decision variables
• withdrawal cycle lenth, T
• number of kanbans for part i of family j,
  nijT
• kanban size, aijT
• Six alternatives for T from 8,4,1,0.5,0.25 in
  hours or
• 480,240,60,30,15 in minutes
• number of kanbans as powers of two, thus kanban
  sizes given by

28
Results

• Therfore each sequencing rule evaluates 36
  alternatives and finds the kanban sequences at
  each stage with minimum sum of inventory holding
  and backorder costs

29
Resultscomparison of the number of instances of
best withdrawal cycle lengths
30
Resultscomparison of the maximum inventory
levels of sequencing rules
31
Resultscomparison of inventory holding costs of
sequencing rules
32
Results

• Smaller setup to processing time ratio results in
  withdrawal shorter cycle lengths
• Thus FCFS produces longer cycles
• Withdrawal cycle length not robust to scheduling
  rules
• Item based rules perform well when withdrawal
  cycles are long
• FCFS-F prefers shorter cycles compared to FCFS

33
Results

• Minimum value for maximum inventory via SPT/LATE
• Highest for FCFS
• Maximum value for all rules given by 1110111
• Minimum avg. inv. Holding cost by SPT-F
• 55.88 of inventories full for SPT/LATE
• All these point to the necessity of sophisticated
  scheduling algorithms

34
Conclusions

• About existing studies
• very few sizes consider kanban sizes explicitly
• (but of kanbans depends on it)
• the scheduling algorithms should go beyond the
  scope of smoothing
• Periodic review systems should be considered

35
Conclusions

• About the experimental study
• Withdrawal cycle lenghts not robust to scheduling
  algorithms
• Item-based rules outperform family-based ones if
  system load is loose (opposite if system loaded)
• When setups increase system performance decreases
• For high setups family-based rules perform better
  
• Finally, more sophisticated scheduling algorithms
  must be cosidered

36

• Q A