The Learning and Forgetting Effects of Dynamic Workers on the Performance of
Dispatching Rules in Flow Shops
Horng-Chyi HORNGand
Shu-Pei HUDepartment of Industrial Engineering and Management, Chaoyang University of Technology
Wufong Township, Taichung County 41349, Taiwan
2
Presentation Outline
qIntroductionqLearning and ForgettingqSimulation ConstructionqExperimental DesignqResults and AnalysisqConclusions
3
Introduction
q The problem of worker shortage comes along with the decreasing birth rate for the recent decade in Taiwan.
q One possible solution is training workers to become so-called multitasking workers.
q The flexibility of production scheduling increases when adopting multitasking workers, however, so does the complexity.
q The purpose of this study is to evaluate how the performance of dispatching rules affected by the learning and forgetting effects of workers.
4
Commonly used dispatching rules
q Flow shops
min. Zij = PijSPTmin. Zij = Pij / ( t-ri )PT/TISmin. Zij = rijFIFOmin. Zij = ri - rkiAT-RPTmin. Zij = riAT
DefinitionsDispatching Rules
q Reference: C. Rajendran and O. Holthaus, “A Comparative Study of Dispatching Rules in Dynamic Flow Shops and Job Shops,” European Journal of Operational Research, Vol. 116, 1999, pp. 156-170.
5
Worker assignment rules
qTwo categories: “when” and “where”.qAdopting CEN (centralized): the most flexible
“when” rule that move away a worker whenever he/she completed the current job.qAdopting FCFS (first come first serve): a “where”
rule that prevents jobs waiting in a queue for too long.
q Reference: H.V. Kher, 2000, “Examination of Worker Assignment and Dispatching Rules for Managing Vital Customer Priorities in Dual Resource Constrained Job Shop Environments,” Computer and Operations Research, Vol. 27, 2000, pp. 525-537.
6
Learning and Forgetting
qLearning effectsqThe efficiency of a worker in processing jobs at a
certain station shall improve as the number of jobs he/she continuously completed increases.
qForgetting effectsqThe deteriorating of a worker’s efficiency in processing
jobs at a certain station as he/she, for some reasons was removed away from that station, stopped processing the same jobs for a period of time.
q Reference: J.C. Fisk and D.P. Ballou, “Production Lot Sizing under a Learning Effect,” IIE Transactions, Vol. 14, No. 4, 1982, pp. 257-264.S. Globerson, and N. Levin, “Incorporating Forgetting into Learning Curves,” International Journal of Operations and Production Management, Vol. 7, No. 4, 1987, pp. 80-92.
7
Models for learning and forgetting
qLearning model – Mazur and Hastie (1978)
qAdding forgetting effects – Nembhard and Uzumeri (2000)
xx rpxpx
ky ε+
++
+=
xx
xx
rpxR
pxRky εα
α+
++
+=
8
Simulation Construction
q A hypothetical multi-station, single-server flow shop.
q Performance measurements:q mean flowtime of jobsq maximum flowtime of jobs, and q work-in-process (WIP).
9
Experimental Factors
q number of stations: 10, 20, and 30;q number of multitasking workers: 10, 20, and
30;q processing time: Normal(3,12), Uniform(1, 5),
and Exponential(3) for coefficient of variation(CV) equals to 33.33%, 38.49%, and 100%, respectively; and
q utilization rate: 75%, 82.5%, and 90%.
10
Experimental Design (I)
Learning effects with random forgetting rate5th stage
Learning effects with constant forgetting rate4th stage
Learning effects with complete forgotten3rd stage
Only learning effects 2nd stage
No learning nor forgetting effects1st stage
DescriptionsStages
11
The 2nd stage learning model
Pro
duct
ion
Rat
e
0Time
k
Worker assigned to other station
12
The 3rd stage learning and forgetting model
Pro
duct
ion
Rat
e
0Time
k
Worker assigned to other station
13
The 4th stage learning and forgetting model
Pro
duct
ion
Rat
e
0 Time
k
Worker assigned to other station
14
The 5th stage learning and forgetting model
Pro
duct
ion
Rat
e
0Time
k
Worker assigned to other station
15
Experimental Design (II)
qAt each stage, a 22 factorial design with nc = 3 center points is adopted as guideline for performing simulation experiments. qTwo major factors are number of station and
utilization rate. qAt each simulation run, there are 30 replications
for statistical comparison purposes. Replication length is set to be 12,500 minutes in which the first 2,500 minutes is the system warm-up period.
16
Results and Analysis
qSimulation results from the 1st stage and 2nd stage serve as fundamental basis for comparison with subsequent stages one by one.qThe 1st stage simulation results are validated and
compared to those mentioned in the literature. qMean flowtime ~ SPT and PT/TISqWIP ~ SPT and PT/TISqMaximum flowtime ~ AT, FIFO, and AT-RPT
17
ATFIFO
AT-RPT
SPTPT/TISFIFO
1st
Stage
SPT
FIFOAT
AT-RPTSPTPT/TIS
SPTPT/TIS
AT-RPTFIFOAT
SPTPT/TIS
SPTPT/TISFIFO
AT-RPTFIFOAT
SPTPT/TIS
2nd
Stage
WIPFmaxWIPFmaxWIPFmax
EXPOUNIFNORM
Best rule(s) from the 2nd stage experiment for large system with high utilization rate
FFF
18
ATFIFOSPT
ATAT-RPT
SPTAT
FIFO AT-RPT
1st
Stage
SPTPT/TIS
FIFOSPTATSPT
PT/TISSPT
PT/TIS
AT-RPTAT
SPTPT/TIS
SPTPT/TIS
2nd
Stage
SPTAT
FIFOSPT
SPTSPTAT
AT-RPTSPT
SPTPT/TIS
ATAT-RPT
FIFO
SPT3rd
Stage
WIPFmaxWIPFmaxWIPFmax
EXPOUNIFNORM
Best rule(s) from the 3rd stage experiment for small system with low utilization rate
FFF
19
1st
Stage
SPTPT/TIS
AT-RPT AT
FIFO
SPTPT/TIS
SPTPT/TIS
SPTPT/TIS
SPTPT/TIS
AT-RPTAT
SPTPT/TIS
2nd
Stage
SPTAT-RPT
FIFOAT
SPTSPT
AT-RPTAT
SPTSPTAT-RPT
AT FIFO
SPT4th
Stage
WIPFmaxWIPFmaxWIPFmax
EXPOUNIFNORM
Best rule(s) from the 4th stage experiment for medium-size system with medium utilization rate
FFF
20
1st
Stage
ATAT-RPT
FIFO
SPTPT/TISFIFO
AT-RPTAT
FIFO
SPTPT/TISFIFO
SPTPT/TISFIFO
AT-RPTAT
FIFO
SPTPT/TISFIFO
2nd
Stage
SPTPT/TIS
ATFIFO
AT-RPTSPT
PT/TIS
SPTPT/TIS
AT-RPTAT
SPTPT/TIS
SPTPT/TIS
ATAT-RPT
SPTPT/TIS
5th
Stage
WIPFmaxWIPFmaxWIPFmax
EXPOUNIFNORM
Best rule(s) from the 5th stage experiment for small system with high utilization rate
FFF
21
Conclusionsq In general, SPT is the best dispatching rule for the mean
flowtime and WIP criteria. On the other hand, AT and AT-RPT perform better than others in the maximum flowtime criterion.
q Only when the system is relatively small with lower utilization rate, does the learning effect without forgetting effect produces different ranking of best rules than that of no learning effect case.
q The rule PT/TIS is a good candidate for the mean flowtime and WIP criteria. However, it is significantly influenced by the models for the forgetting effects of workers.
q The incorporation of the forgetting effects of workers to the system actually changes the rankings of best rules in statistical sense.
22
Research Directions
q Considering the effects caused by the shortage of workers;
q Incorporating more worker assignment rules; andq Examining the effects of learning and forgetting effects
on other production systems, such as job shops.
23
Thanks for listening!