Sunteți pe pagina 1din 10

4.1.

WIP - MAD Dynamic, Intelligent Heuristic

This algorithm has been developed to pursue a scheduling strategy searching for the WIP (Work In Progress) and the Due Date MAD (Mean Absolute Deviation) minimization . Being WIP - MAD a dynamic, intelligent, multi-criterion heuristic, we must provide sufficient system state control tools able to evaluate both scheduling objectives degree of satisfation . Thats why we have introduced two control variables which must be updated at every decision-point, giving informations about the criticality degree of waiting jobs in a given queue. The first variable has been named CrMFT (Critical Mean Flow Time). It provides informations about jobs criticality regard to the WIP. Its espression is: CrMFT =
i =1 N

T Ii N t ij
j =1 M

where: T = Real Time. Ii


=

Inlet Time of i th job in a queue. in a generic queue routing.

t ij = Manufacturing time on j th machining center of i th job N = Number of all the jobs in queue. M = Routing lenght of i th job in queue.

So, regard to all waiting jobs the CrMFT is the mean value of ratio between i real flow time and its operation schedule time still elapsed. Why is that ?

th

As we know, an increase of the mean flowtime causes a WIP increase as well; thats why we can say the CrMFT expression well synthesizes the queue state regard to the WIP. Infact, the greater is the CrMFT value, the greater is the incidence of waiting times on the MFT value; this fact means that a generical job in the queue spends a longer time into the system causing a degree of satisfaction lowering regard to the WIP minimization. The second control variable has the aim to evaluate the mean absolute deviation respect to the jobs due dates (MAD) and it was named NTE.

Before its explitation, we have to say that the jobs have been divided into three groups: 1) 2) 3) EARLY JOBS, those jobs having a Slack and Op Slack > 0.0 , so their TARDY JOBS, those jobs having both their slacks < 0.0 , so their OPERATIONAL JOBS, those jobs having OpSlack <0.0 and Slack >

completion is possible before their fixed due date and their next operation duedate. lateness is not fixable anymore. 0.0 , so they are still in time regard to the global manufacturing cycle. Being NTE espression : NTE = Jtardy + Jearly

It clearly synthesizes the Jobs condition regard to the MAD. Infact the MAD is as bigger as more numerous are late and early jobs; so, we can figure out how a great NTE value is indicative of a MAD critical situation. After having defined CrMFT and NTE, we have to face the problem of defining their criticality thresold. This issue is up to the decision makers experience whom must evaluate - into a given system configuration which ones have to be the criticality thresold values.

In order to define these thresoldes we can explane some criterions:

1) As in the considerations of due dates respect affect as the anticipations as the delays and since tightness growin sees growing up the anticipations while the delays reducing and on a reduction of the tightness the situation is reversed, is reasonable regard this compesation rule, choose only a thresold value for NTE whichever tightness value (in the experiments we have NTE = 0.66). 2) Growin medium utilization system, it is required a growth of CrMFT criticality thresold because happens an increase of waiting time incidence on mean flow time; so, if the thresold value keeps low, probably the system would result always critical regard to WIP, and if even MAD was critical, the first was more one.

Called CRIT 1 and CRIT2 the thresold criticality values rispectively of CrMFT and NTE, we can begin to explane the WIP- MAD algorithm dynamic:
IF

{ [ CrMFT < CRIT 1 ] and [ (NTE < 0.50) or ( ( NTE > 0.50 ) and

(the majority of queue jobs is of operationally type) ) ] }

None objective is critical , then the algorithm uses a compound rule. This rule has been carefully developed to synthetizes in own inside both objectives consideration. Its form is: C = min { [ 1 * 1( x) + 2 * 2( x)] } x S

where 1 , 2 [ 0 , 1 ] OPCR x OPCRmin OPCRmax OPCRmin t px t p min t p max t p min

2( x) =

1( x) =

where: tpx is the xth Job in the queue operation time on the actual machine. tpmin is the lower operation time among the jobs in the queue. tpmax is the higher operation time among the jobs. OPCRX is the xth job in the queue operation critical ratio. OPCRmin the lower operation critical ratio among the jobs in the queue.

The newness of this compound rule is its dynamic evaluation of every scheduling objective weight in the compound rule sinthesys process. This problem, very debated in literature, has been easily solved by considering these relationships:

1) 1 + 2 = 1.0 2) 1 / 2 = Ris1 / Ris2

Where

Ris1 and Ris2 are criticality degrees of our two scheduling

objectives,defined as:

Ris1 =

CrMFT Crit1

Ris2 =

NTE Crit2

The 1) says that however are distributed the weights 1 and 2 their sum must be equal 1.0

10

The 2) says that the weights ratio must be equal to the one between the degrees of criticality whom the weights refer to.

IF

( ( CrMFT > CRIT1)

and

(NTE > CRIT2 ) )

IF Ris1 < Ris2 then The algorithm schedules by OPCR rule. ELSIF Ris1 >= Ris2 The algorithm schedules by SPT rule.

*0

IF { [CrMFT > = CRIT 1 ] and [( NTE < 0.50) or (NTE > 0.50

and The relative queueing jobs relative majority is Oplly type) ] } then the WIP - MAD algorithm schedules by SPT.

*1

IF { [CrMFT < CRIT 1 ] and [( NTE > Crit2) or (NTE < = Crit2

and The relative queueing jobs relative majority is not Oplly type) ] } then the WIP - MAD algorithm schedules by OPCR.

4.2

WIP - TARDINESS Dynamic Intelligent Heuristic

11

The motive we have developep this second algorithm for, takes place from some production scenaries where the early shipments are not considered to avoid or anyway are considered less expensive than the delays. For example, if the single product stock cost is not very incident on the global costs the factory has to face to, given a great storehouse capacity, we can consider not important early shipments costs. Then, it would be dispersive concentrate scheduling operators attention to the early shipments but it would be rather useful that scheduling operators concentrate their efforts to solve delays problem. For that concern its charatheristics WIP TARDINESS algorithm is similar to WIP - MAD one, but here NTE is equal to Jtardy. Its necessary to note that the considerations about how fixxing criticality thresold for NTE change respect WIP - MAD ones. In this case is possible reduce this thresold with the growth of the tightness because as much tightness value grows, as much lower will be delay jobs; then, when tightness value is high, we have to reduce NTE criticality thresold to avoid the situation in which TARDINESS is never critical. The algorithm operation description is the same of the WIP - MAD one but: 1) 2) NTE = Jtardy CRIT2 is different.

5.1

The simulation experiments

The main objective of the experiments has been verifing, by discrete event simulation, new two algorithms performances. So, we have studied the perfomances of 11 static eurystics and 4 dynamic ones, known in literature, for having the possibility to use sufficient data to compare with the results obtained by the use of two new eurystics proposed.

12

After having well balanced the manufacturing system in loading and sequencing procedures, for defining the jobs due dates we have used TWK ( Proportional to total work) approach in which the proportionality factor is called TIGHTNESS. For each rule implemented and for each tightness value, we have throwed 10 simulations . So, we have considered two scenaries: 1) 2) Tightness = 4.0 ( In this scenary there is a low operative margin) Tightness = 7.0 (In this scenary there is a medium operative margin)

In managing the simulations, we have used the syncronized simulation approach. The 10 simulations throwed for each rule are different for that concern the arrives order of the jobs in the system. This difference based on the initialization seed of the random integer numbers generator; we have correlated the job type with the integer number extracted by generator. The performance system statistics used in the analisys are: 1) 2) 3) 4) 5) 6) Mean Flow Time The Fraction of the late jobs Mean Lateness Mean Tardiness MAD ( Mean Absolute Deviation) Makespan

In order to new algorithms proposed, we have to fix the criticality thresolds values (Crit 1 , Crit 2) to determine Crit1 value for each production scenary we have evaluate in which ratio are the better mean flow time (among implemented rules) and the operation schedule time mean value among six job types considered. For that concern WIP - TARDINESS algorithm, being not difficult having late jobs, we have evaluate Crit 1 value could be fixed equal to this ratio for escaping the situation in which Due Date respect objective was not the only critical one.

13

Regard to Crit 2 value, it has to lower respect to a growth of the tightness value. When tightness = 4.0 we have fixed Crit 2 value equal to 0.35 while for tightness = 7.0 its equal to 0.15. For that concern WIP -MAD algorithm the early shipments consideration require a growth of Crit 1 value respect to WIP - TARDINESS testing while CRIT 2 has been fixed equal to 0.66.

WIP - MAD Tightness = 4.0 Tightness = 7.0

Crit 1 6.7 6.7

Crit 2 0. 66 0. 66

WIP - TARDINESS
TIGHTN ESS 4.0 TIGHTNESS 7.0

CRIT 1

CRIT 2

4.0

0. 35

4.0

0. 15

Scenery 1 ( Tightness = 4.0)

14

Manufacturing System Machining Centers Number Jobs Types

Job Shop 5 6 900 15

The number of jobs processed Number of eurystics tested

For evaluating the performances of two new eurystics we have: 1) 2) selected, for each performance statistic, the best rule among those tested. built, for each statistic, an interval which extremes were the best and the

worse results among the selected rules. So, if new two algorithms performances regard to the main statistics falls into these intervals, we can affirm that new rules behauvior is considerable. The results are shown in the tab.13 and Tab. 15 . How we can easily see, the performance of new algorithms has been considerable regard to all the performance statistics.

Scenery 2 ( Tightness = 7.0)

Manufacturing System

Job Shop

15

Machining Centers Number Jobs Types

5 6 900 15

The number of jobs processed Number of eurystics tested

For evaluating the performances of these two new eurystics we have: 1) Selected for each performance statistic, the best rule among the tested ones. 2) Built - for each statistic - an interval whose extremes are the best and the worst result among the selected rules. So, if these new two algorithms performance regard to the main statistics will fall into these intervals, we can state that the new rules behauvior is very outstanding. The results are shown in the tab.14 and Tab. 16 . How we can easily see, the efficacy of these new algorithms has been very high regard to all the performance statistics.

Conclusions We can say that by the present work, it has been reached the proposed goal to give a meaningful contribute in order to improve the operative instruments for the job shop scheduling. In fact, the new two algorithms have shown some good realibility and flexibility in both two scenerios considered. It seems reasonable that these features are going to keep their realiability into different tightness scenerios too. But surely, we can say to have override the rigidity and low flexibility of no multi-objective and dynamic euristics.

16

S-ar putea să vă placă și