Professional Documents
Culture Documents
Supply Chain PN Optimization
Supply Chain PN Optimization
Supply Chain PN Optimization
Supply Chain
L. AMODEO
Professor
University of Technology of Troyes
References
Application
!" !$
Performance
Modelization Evaluation Optimization
!#
(cost, service)
!%
Real Case or
project
(Supply chain) Linear programming Direct Exact Methods
State model Or indirecte Approach Methods
Markov Process Discrete (Heuristics -
Queuing files Event Metaheuristics)
Petri Nets Simulation
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 72
Introduction
Material flow
Financial flow
Information flow
Distributors
Producers
Suppliers (wholesaler, retailer)
Manufactories customers
Batch DSPN
BDSPN- Marking
p1 p3
§ µ- Marking
t1
ìï n Î IN si p Î Pd
µ( p) = í 2 2
{
ïî µ µ Î IN
*
} si p Î Pb
p2
4
p4
§ M- Marking ì 3 ü é3ù
ï{2,2,4}ï ê8ú
ì µ ( p) si p Î Pd ï ï M0 = ê ú
ï µ0 = í ý ê0ú
M ( p) = í å µ
ïî µÎµ( p )
si p Î Pb ï 0 ï ê ú
ïî Æ ïþ ë0û
p1 2 p2 p1 p2 !"#$%"
Firing
1 1 1 1
2-enabling
2 t 3 2 t 3
&''
p3 p4 p3 6 p4
Discrete Transition • t Ç P = Æ
b
P1 P2 P1 P2
2 2
Firing
2 2
1 T 3 1 T 3
P3 P4 P3 3 P4
Batch Transition • t Ç P ¹ Æ
b
P1 P2
P1 2 P2 Firing
1 1
1 1
2 T 3
2 T 3
P3 6 P4
P3 P4
P1
P2
2 P3
Batch
3 process
Batch orders
of products
T1
P1
2
P2
Firing of de T1 with q = 2 2 P3
Batch firing
P1
Stock P4
of products: Type A T1 T2
P2
Stock 2 Batch
of products:Type B 3 assembling
3
P3
Customer
Batch orders
4
2 P5 P1
P4
T1 T2
2
P2
2
3
3
Firing of T1 with q = 2
P3
4
2 P5
µ k +1 = µ k + W ´ F t1
4
p1
4 2
t3
4
p2
2 2
§ Caracteristic vector
t2
Firing of t3
ì 0 if i ¹ j With q= 4/2 = 2
F[]
i i=1à m =í k
îqj if i = j t3[2]
k
q Firing indice of Tj from µk
é ù é ù é ù
ê0ú
{}
j
é ù
ê{2,4} ú ê4 2 -2ú ê0ú ê2ú
ê ú + ê
4 úûú
´ ê ú= ê ú
ì 1 if upstream places are discrete ê
ë 0 ú
û ëê0 0 êê2úú ê8ú
qkj í
îb / W (P, Tj ) if upstream places are batch places ë û ë û
m : nb of transitions
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 84
Other developments
§ Performance evaluation
§ Analytical methods based on reachability graph (Markov
process)
§ Discrete event simulation
Exercice
P1
P4
T1 T2
2
P2
3
2
2
P3
3
2 P5
1 T2(q=2) 1
T1(q=2)
3 3
3 {3} {3}
µ1 {2,2} µ3 {2}
9
µ0 {3,2} 0
4
{2} T2(q=2) 0
0 0
T1(q=3) 0
0
µ4 {2}
µ2 {2}
{3}
{2,3}
2
6
T2(q=3)
0
0
µ5
{2}
{2}
0
Production policies
Upstream Downstream
Stock Stock
Customer Demand
Upstream Downstream
Stock Stock
Customer demand
Lead time
Production policies
Assembling
Forecast Machine 1 Machine Exit
T1 P1 T2 P2 T3 P3 T4 P4 T5
Upstream Downstream
stock stock
Customer
demand
Lead time
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 91
Assembly line
machine 1
Assembly machine 3
machine 2
T1 P1
T3 P3 T4
T2 P2 d
Demand
P4 T5
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 92
Inventory system
§ Stock FIFO
3 2 1
l µ
Arrival Exit
T1 P1 T2 P2 T3 P3 T4
l P4 P5 P6 µ
§ Stock LIFO
3 2 1
l µ
Arrival Exit
Inventory system
P4
§ Stock LIFO P5
T5 T7
P6
T1 P1 T2 P2 T3 P3 T4
l µ
P7
P8
T6 T8
Zhou, Di Cesare,1996
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 95
Queueing file M/M/1
System
§ Arrivals are Poisson process with
parameter: l Arrival Exit
§ Service time is a random process
(expo) with parameter: µ Queue Service
§ Single server
§ Unlimited queue capacity
Queue Service
T1 P1 T2 P2 T3
l P3 µ
§ 2 servers
§ Capacity of the queue: 5
§ Population of 6 customers
T1 Queue T2 T3
Service
l P3 µ
Replenishment Delivery
T3 P2 T2 P1 T1
Material
Flow
P4 T4 P3 P5 T5
Q Information
Outstanding Q
Flow
order
R+M(P5)
Replenishment Delivery
T3 P2 T2 P1 T1
P4 T4 P3 P5 T5
Outstanding 4 4
order
4 3
2+M(P5)
junction
Junction with
with the pylon
electrical wire
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 100
Upstream Downstream
Supplier 2 Manufacturer Customer i
transporter 2 transporter
Upstream
Supplier 3
transporter 3 Customer n
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 101
Logistic Process
Welding by friction
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 102
BDSPN Model
Stock 1
Supplier 1
Demand
Supplier 2
Information
Assembling Flow
Supplier 3 Material
flow
Paiement
Producer –
supplier
Financial
flow
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 103
Material flow
Supplier 1 Transporter 1
Supplier 2 Transporter 2
Assembly Packaging
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 104
Information flow
Inventory policy
Transporter Stock 2 Assembly
Stock S2 : flats
Produccer
to manufacturer
Performance evaluation
!"#$%&'"(#)&*+,&-
Average number of tokens in place P7
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 108
Service level
Inventory cost
!"
!$ Optimization
!#
Tool
Inventory Parameters (EA)
!%
S3(R3,Q3) R Î [Rmin, Rmax]
Q Î [Qmin, Q Qmax].
Discrete Event Simulator
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 109
Optimization
§ 2 approaches
§ Mono-objective: 1 solution is given to the decision maker
§ Aggregative Method: F = aF1 + bF2
§ e-constraint method
§ To achieve a minimum Service Level ( > 95%)
§ To minimize the Total Inventory cost
§ Multi-objectifve: front of non dominated solutions
§ To maximize the Service Level
§ To minimize the Total Inventory Cost
§ The decision maker has to chose between several
best non-dominated solutions
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 110
Multi-objectif case
Service Level
Ideal Point
1
To maximize
Pareto
front
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 111
Mono-objective optimization : Genetic algorithm
Individual evaluation
Genetic operations
new
Itération
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 112
Optimization: GA
§ Coding R1 Q1 R2 Q2 R3 Q3 R4 Q4
§ Crossover (Pc)
R1 Q1 R2 Q2 R3 Q3 R4 Q4
2 Parents
R1’ Q1’ R2’ Q2’ R3’ Q3’ R4’ Q4’
1 pt crossover
R1 Q1 R2 Q2 R3’ Q3’ R4’ Q4’
2 Children
R1’ Q1’ R2’ Q2’ R3 Q3 R4 Q4
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 113
Results – Mono-objective
2000
1000
0
S1 S2 S3 S4
Before opti
After opti
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 114
Results: Mono-objective
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 115
Multi-objective optimization
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 116
Multi-objective optimization
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 117
Multi-objective optimization
§ Incorporates elitism
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 118
Multi-objective optimization
Non dominated
sorting
Repeat until
!" !# !$ !% !& stopping
conditions
Diversity with
crowding distance
'()(*+,-.)/,0"
Parentt+1 Offspringt+1
SPEA-2
§ Works with an external archive.
§ Fitness of solution X: its strength (number of solutions it dominates).
§ Selection, crossover, mutation : like in NSGA-2.
§ Archive truncated using a density measure (like spacing).
§ We also modified SPEA-2 to get a variant SPEA-2b, using the gene-
wise reproduction method of Daniel & Rajendran (2005) and the
crowding distance of NSGA-2.
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 120
Multi-objective optimization
MOPSO
§ First PSO used for supply chain optimization.
§ A particle (solution)
§ has a position (R1,Q1,…, Ri,Qi,…,Rn,Qn), a velocity and an
experience (best score found).
§ Moves in solution space controlled by time-varying inertia and
acceleration coefficients.
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 121
Computational Experiments
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 122
Computational Experiments
!"#$%"&'#(
– )*()+,-'.(/0(.'123&%$3/"# 41'.(#3,+2%$3/"5*(67(%"8(677
– )+,-'.(/0(9'"'.%$3/"#* 67:(;<:(<7:(677:(;<7:(<77:(=<7:(6777
>./-2',*(?/@($/(&/,1%.'(#'A'.%2(0./"$#B
;($C1'#(/0(,'%#+.'#('D3#$(3"($E'(23$'.%$+.'*(4,'$.3
F'%#+.'#(&/,1+$'8(@3$E(/"2C(/"'(0./"$
)- /0("/"(8/,3"%$'8(#/2+$3/"#:(Spacing metric (Schott, 1995),
Hypervolume (Zitzler and Thiele, 1998), Spread metric (D):
F'%#+.'#(/-$%3"'8(@3$E(;(0./"$#(46(.'0'.'"&'(0./"$5
G(H3#$%"&'(4I33#'5:(J3$K2'. F'%#+.'(:(L
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 123
Computational Experiments
!"#$%&'%()*+",-)(#-)% %-#",.*/#+"0(*",*12%-#")+/(%+,#/3
– 4)(5&#-"6#%/'-#" 7089:"8!3
– ;"*<"/*&'()*+/"0893"=*1)+%(#= >?"%("&#%/("*+#"/*&'()*+"*<"8!
– @"A)/(%+,#"0B))/#3"µ C"S) =)
– =)"C"=)/(%+,#">#(D##+"%"/*&'()*+")"0893"%+=")("*-(.*E*+%&"
2-*F#,()*+"*+"8!
– G+"*'-",%/#
– H>"*<"+*+I=*1)+%(#="/*&'()*+/"C"/)5#"*<"(.#"2*2'&%()*+
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 124
Computational Experiments
SL = 99 % Industrial solution
SL = 83%
Cost = 510 €
Pareto
front
!"'#"$$%"$$&
!"#"$%"$&
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 126
Computational Experiments
*#%+,)-./)&$0-) µ!"#$%&'()
Na N Ng Time(s) |F| C(Fi; F1) C(F1; Fi) µ µ̅
F1 10 10 71 100 0 0 0.00 0.0000
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 127
Computational Experiments
"&'$%
"&'$%( Rapid
convergence
for SPEA2
and
!"#$% SPEA2b
25 generations
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 128
Computational Experiments
Industrial
Solution
(.83, 510)
"&'$%(
)*&"*
"&'$%
!"#$%
Best Pareto
front for
SPEA2b
1000 generations
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 129
Computational Experiments
§ Ziztler measure
§ NSGA2 front is taken as reference
C(NSGA-2, i) C(i, NSGA-2)
MOPSO 90 0 70 68 58 20.5 49 17 7 10
Computational Experiments
NSGA 2 has the most uniform spacing (may be due to its crowding distance)
Not much difference between the three other Metaheuristics
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 131
Computational Experiments
More than 4h
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 132
Computational Experiments
Industrial
NS = 90% Solution
SL = 83%
Cost = 517 €
Pareto
R Q
Front S1 2300 5000
S2 590 3000
R Q
S3 2000 9300
S1 1150 2510
S4 400 2000
S2 521 2009
S3 1918 5031 Cost = 250 €
S4 622 1004
Prof Lionel Amodeo – Inventory Optimization– All rights reserved – 2017 133