Transcript Stochastic simulation of dispatching rules in the capital goods industry

Stochastic simulation of
dispatching rules in the
capital goods industry
Dr Christian Hicks
University of Newcastle upon Tyne
http://www.staff.ncl.ac.uk/chris.hicks/presindex.htm
IGLS04/1
© C.Hicks, University of Newcastle
Dispatching rule
literature
• Majority of work has focused upon
small problems.
• Work has focused upon the production
of components, mostly in job shops.
• Minimum set-up, machining and
transfer times have been neglected.
• Deterministic process times have been
assumed.
IGLS04/2
© C.Hicks, University of Newcastle
Capital goods
companies
• Design, manufacture and construction
of large products such as turbine
generators, cranes and boilers.
• Complex product structures with many
levels of assembly.
• Highly customised and produced in
low volume on an engineer-to-order
basis.
IGLS04/3
© C.Hicks, University of Newcastle
Manufacturing System Simulation Model
CAPM modules used
System parameters
System dynamics
Planned Schedule
Product information
Resource
information
Operational factors
Logic
Manufacturing Planing &
Control System
Measures of
performance
Manufacturing Facility
Tools
Flow measurement
Cluster Analysis
Layout generation methods
IGLS04/4
© C.Hicks, University of Newcastle
Case Study
• 52 Machine tools
• Three product families competing for
resource (main product, spares and
subcontract)
• Complex product structures
IGLS04/5
© C.Hicks, University of Newcastle
IGLS04/6
© C.Hicks, University of Newcastle
Experimental design
Factors
Minimum setup time
Minimum machining time
Minimum transfer time
Data update period
Capacity constraints
Levels
0, 30 (mins)
0, 60 (mins)
0, 2 days
0, 8 hours
Infinite, finite*
Process times normally distributed with
standard deviation = 0.1 * mean
IGLS04/7
© C.Hicks, University of Newcastle
Dispatching rules
•
•
•
•
•
•
•
•
Earliest due first (EDF)
First event first (FEF)
Longest operation first (LOF)
Least remaining operations first (LRF)
Least remaining slack first (LSF)
Most remaining operations first (MRF)
Shortest operation first (SOF)
Random (RAN)
IGLS04/8
© C.Hicks, University of Newcastle
IGLS04/9
© C.Hicks, University of Newcastle
Performance Metrics
Throughput Efficiency () =
Minimum flow time x 100 (%)
Actual flow time
Tardiness (T) =
completion time – due time
(for completion time > due time)
Tardiness (T) = 0
(for completion time  due time)
Due date performance =
completion time – due time
IGLS04/10
© C.Hicks, University of Newcastle
Infinite capacity
experiments
IGLS04/11
© C.Hicks, University of Newcastle
IGLS04/12
© C.Hicks, University of Newcastle
IGLS04/13
© C.Hicks, University of Newcastle
IGLS04/14
© C.Hicks, University of Newcastle
IGLS04/15
© C.Hicks, University of Newcastle
Infinite Capacity
Experiment Results
• Infinite capacity experiments indicated
that more factors and interactions
were statistically significant at
component level than at product level.
• Minimum transfer time had the
greatest impact upon mean throughput
efficiency and mean tardiness.
• Throughput efficiency was much
higher at component level than
product level suggesting that the
Company’s plans were not well
synchronised.
IGLS04/16
© C.Hicks, University of Newcastle
Finite Capacity
Experiments
IGLS04/17
© C.Hicks, University of Newcastle
IGLS04/18
© C.Hicks, University of Newcastle
IGLS04/19
© C.Hicks, University of Newcastle
IGLS04/20
© C.Hicks, University of Newcastle
IGLS04/21
© C.Hicks, University of Newcastle
IGLS04/22
© C.Hicks, University of Newcastle
Finite Capacity Experiment
Summary
At product level:
• Mean throughput efficiency maximised
by SOF (main and subcontract) and
MRF (spares).
• Mean tardiness minimised by SOF
(subcontract), LSF (main product),
MRF (spares).
• Dispatching rule most important factor
for both measures.
IGLS04/23
© C.Hicks, University of Newcastle
Finite Capacity Experiment
Results
At component level:
• Best rules for mean throughput
efficiency and tardiness were LOF
(subcontract), EDF (main) and SOF
(spares) i.e. different to products
• Minimum transfer time most important
factor for minimising throughput time.
• Dispatching rule most important factor
for minimising tardiness.
IGLS04/24
© C.Hicks, University of Newcastle
Conclusions
• Most dispatching rule research has
focused upon job shops and has
neglected other operational factors
such as minimum setup, machining
and transfer times and the data update
period.
• Dispatching rule research has
investigated deterministic situations.
• This research has included complex
assemblies, stochastic processing
times and a multi-product
environment.
IGLS04/25
© C.Hicks, University of Newcastle
Conclusions
• Performance at product level much
worse than at component level –
probably due to poorly synchronised
plan.
• “Best” dispatching rule varies
according to measure, level and
product family.
• Results for “best” rule under stochastic
conditions different with deterministic
processing times.
• SOF generally best in agreement with
Blackstone.
• Statistical significance of other factors
varies by level, product and measure,
but dispatching rules important in all
cases.
IGLS04/26
© C.Hicks, University of Newcastle