operations management in the fruit industrycepac.cheme.cmu.edu/pasilectures/bandoni/pasi 2005 -...

Planta Piloto de Ingeniería QuímicaPlanta Piloto de Ingeniería QuímicaCamino La Carrindanga, Km. 7 (8000) Bahía BlancaArgentina

Camino La Carrindanga, Km. 7 (8000) Bahía BlancaArgentina

J. Alberto BandoniJ. Alberto Bandoni

OPERATIONS MANAGEMENT IN THE FRUIT INDUSTRY

OPERATIONS MANAGEMENT IN THE FRUIT INDUSTRY

PASI Program on Process System Engineering, August 16-25, 2005, Iguazú Falls, ArgentinaPASI Program on Process System Engineering, August 16-25, 2005, Iguazú Falls, Argentina

2

Introduction

Argentinean Fruit Industry Supply Chain Optimization

Frutas & Jugos ARG Co.: Case Study

Optimal Operation of a Packaging Plant

A Novel Method to Reduce Event Variables in Continuous-time Formulation for Short-term Scheduling

PRESENTATION OUTLINEPRESENTATION OUTLINE


3

ARGENTINE IN THE WORLDARGENTINE IN THE WORLD

NeuquénNeuquén

Iguazú FallsIguazú Falls

Buenos Aires

Buenos Aires

Bahía BlancaBahía Blanca




Noemí Petracci, Guillermo Massini, J. Alberto BandoniNoemí Petracci, Guillermo Massini, J. Alberto Bandoni

Supply Chain Optimization in the Fruit Industry

Supply Chain Optimization in the Fruit Industry


FOCAPO 2003, Florida, USA, January 12-15, 2003FOCAPO 2003, Florida, USA, January 12-15, 2003

5

BLACK AND NEUQUEN RIVERS HIGH VALLEY FRUIT INDUSTRY BLACK AND NEUQUEN RIVERS HIGH VALLEY FRUIT INDUSTRY

Neuquén and Black Rivers High Valley

Neuquén and Black Rivers High Valley

NW Region

NW Region

CuyoRegionCuyo

RegionMesopotamian

RegionMesopotamian

Region

The High Valley of RThe High Valley of Ríío Negro o Negro and Rand Ríío Neuquo Neuquéén,n,placed across two states placed across two states southwest of the country, is the southwest of the country, is the area of our country where the area of our country where the apples and pears are grown.apples and pears are grown.


6

Argentinean fruit industry relevant figures

• Fruit Industry : 700.000.000 u$s/year export value

• Apples and Peers : 350.000.000 u$s (50 %)

• HV Region : 330.000.000 u$s (95 %)

ARGENTINEAN FRUIT INDUSTRY IN FIGURESARGENTINEAN FRUIT INDUSTRY IN FIGURES

Apples: Apples: 10.086.000 10.086.000 tnstns

Pears: Pears: 520.000 520.000 tnstns

23 % for industrialization

(concentrate juice)

50 % apples for industrialization

(concentrate juice)


7

During the 90During the 90’’s, companies made important capital investments s, companies made important capital investments on new machinery for efficiency improvement. on new machinery for efficiency improvement.

ARGENTINEAN FRUIT INDUSTRY DESCRIPTIONARGENTINEAN FRUIT INDUSTRY DESCRIPTION

In the last few years, due to new worldwide competitors from AsiIn the last few years, due to new worldwide competitors from Asia a South West, local economic problems and volatile international South West, local economic problems and volatile international markets, companies are compelled to improve even more their markets, companies are compelled to improve even more their competitiveness to keep on business. competitiveness to keep on business.

In this context, they have a need for better decision tools to In this context, they have a need for better decision tools to manage the whole supply chain.manage the whole supply chain.

There are a few large companies that operate along the entire frThere are a few large companies that operate along the entire fruit uit supply chain, and concentrate the largest part of the business isupply chain, and concentrate the largest part of the business in the HV n the HV region.region.


8

Packaging plant

Juice plant

Own farms

Third party farms

Fruit markets

Juice markets

Third party cold

storage

Third party packaging

plant

G

B

H

N

E

A

F

CD

H: Fruit from storage to PPF: Fresh fruit from own farms to PPC: Fresh fruit from own farms to JPA: Fresh fruit from fruit suppliers to PP.G: Fresh fruit from fruit suppliers to JPN: Fruit from supplier’s PP.

B: Fresh fruit prepared and packed in different ways.

D: Fruit from PP that do not fulfill quality specification transferred to JP

E: Product streams of Concentrate Juice of 72°Brix and aroma.

ARGENTINEAN FRUIT INDUSTRY SUPPLY CHAINARGENTINEAN FRUIT INDUSTRY SUPPLY CHAIN


9

Packaging warehouse

J

B

H

AFCA

F

K

D

Refrigeration chamber

AFG

H : Fruit from storage to PPF : Fresh fruit from own farms to PPA : Fresh fruit from fruit suppliers to PPAFG : Fruit sent to the processing line.AFC : Fruit directly sent to cold storage for later processing.K : Fruit to keep in cold storage for later selling.

D : Fruit from PP that do not fulfill quality specification transferred to JP

J : Fruit from cold storage to processing line.B : Fresh fruit prepared and packed in different

ways.

PACKAGING PLANTPACKAGING PLANT


10

G

CConcentrate

juice chamberIndustrial process

M

F

E

Fruit reception

N

D

L

D : Fruit from PP that do not fulfill quality specification transferred to JP.C : Fresh fruit from own farms to JPG : Fresh fruit from fruit suppliers to JPN : Fruit from supplier’s PP.E : Product streams of Concentrate Juice of 72°Brix and aroma.

CONCENTRATE JUICE PLANTCONCENTRATE JUICE PLANT


11

Formulated as a Formulated as a midterm tactical planning problem, over , over oneone--year time horizon (divided in 12 monthly periods.year time horizon (divided in 12 monthly periods.

FRUIT SUPPLY CHAIN PLANNING MODELFRUIT SUPPLY CHAIN PLANNING MODEL

Time horizon coincides with the business cycle from harvest Time horizon coincides with the business cycle from harvest to harvest. During this cycle, many decisions have to be to harvest. During this cycle, many decisions have to be made along the SCmade along the SC..

Model parameters:Model parameters: cost of each variety of raw material, cost of each variety of raw material, selling prices for each product, fruit production and fruit selling prices for each product, fruit production and fruit variety, distances, packaging and juice plant capacities, variety, distances, packaging and juice plant capacities, demands for each product and market, cooling storage of demands for each product and market, cooling storage of fresh fruit and final products, etc.fresh fruit and final products, etc.


12

Model description:Objective Function: max. of benefit along one-year operation.Model equations : mass balance and production equations at

farms, packaging plants and juice plantsConstraints : production limits at proprietary farms, bounds

on fruit supply, bounds on internal processingcapacities, demand satisfaction, stock limits atcold storage and storage at juice plants

MTHEMATICAL MODEL: MILPMTHEMATICAL MODEL: MILP

Model statistics: 14335 continuous variables and 3372 binary variables, 4421 equality and 7524 inequality constraints.

Implementation environment:The GAMS (Brooke et al., 1998) was used in order to implement the MILP optimization model and generate their solutions.


13

57618.69

16864.8127481.23

5591.58

107556.3

0

20000

40000

60000

80000

100000

120000

Tn/

year

From own farms(C+F)

From suppliers'farms (A+G)

From coldschamber (H)

From suppliers'packaging plants

(N)

Total ProcessedFruit

Fruit Processed

NUMERICAL RESULTS: Total fruit processedNUMERICAL RESULTS: Total fruit processed


14

Packed Fruit SoldArgentina

44%

Europe/USA15%

Brazil41%

Argentina44%

Europe/USA15%

Brazil41%

Packed Fruit Sold

Juice Sold

Argentina48%

Europe / USA52%

Argentina48%

Europe/USA52%

Concentrate Juice Sold

NUMERICAL RESULTS: Fruit and juice sold in different marketsNUMERICAL RESULTS: Fruit and juice sold in different markets


15

0

10

20

30

40

50

60

Gal

x 1

03

1 2 3 4 5 6 7 8 9 10 11 12

Period

Europe / USA Local Customers

NUMERICAL RESULTS: Optimal plan for juice commercializationNUMERICAL RESULTS: Optimal plan for juice commercialization


16

0500

10001500200025003000350040004500

Kg

x 10

3

1 2 3 4 5 6 7 8 9 10 11 12

Period

Europe / USA Brazil Argentina

NUMERICAL RESULTS: Optimal plan for fresh fruit commercializationNUMERICAL RESULTS: Optimal plan for fresh fruit commercialization


17

0

20

40

60

80

100

120

Gal

x 1

03

1 2 3 4 5 6 7 8 9 10 11 12

Period

JP1 JP2

NUMERICAL RESULTS: Optimal juice storage profileNUMERICAL RESULTS: Optimal juice storage profile


18

0

2000

4000

6000

8000

10000

12000

14000

16000

Kg

x 10

3

1 2 3 4 5 6 7 8 9 10 11 12

Period

PP1 PP2 PP3 PP4

NUMERICAL RESULTS: Optimal fresh fruit storageNUMERICAL RESULTS: Optimal fresh fruit storage


19

A model was developed to optimize the SC planning A model was developed to optimize the SC planning of a typical fruit company in the HV area of of a typical fruit company in the HV area of Argentina.Argentina.

It optimally decides production plans for fruit and It optimally decides production plans for fruit and juice to satisfy customer orders, while it optimally juice to satisfy customer orders, while it optimally allocates sources or raw material, based on allocates sources or raw material, based on capacities and costs.capacities and costs.

The model realistically represents the current The model realistically represents the current economic scenario in the country.economic scenario in the country.

Current research is under way to complete Current research is under way to complete sensitivity analysis and incorporates uncertainties in sensitivity analysis and incorporates uncertainties in demands and harvest estimations.demands and harvest estimations.

NUMERICAL RESULTS: Optimal fresh fruit storageNUMERICAL RESULTS: Optimal fresh fruit storage


N. Petracci, A. BandoniN. Petracci, A. Bandoni

Case Study:

FRUTAS & JUGOS ARG co.

Case Study:

FRUTAS & JUGOS ARG co.




21

FarmsPlantsSea Port

FRUTAS Y JUGOS ARG Co. : Supply ChainFRUTAS Y JUGOS ARG Co. : Supply Chain


22

•• Villa ReginaVilla Regina

Packaging Plant, PPPackaging Plant, PP11

Concentrate Juice Plant, CJPConcentrate Juice Plant, CJP11

•• GralGral. . RocaRoca

Packaging Plant, PPPackaging Plant, PP22

Concentrate Juice Plant, CJPConcentrate Juice Plant, CJP22

FRUTAS Y JUGOS ARG Co. : Supply ChainFRUTAS Y JUGOS ARG Co. : Supply Chain


23

PP1V. Regina

CJP1V. Regina

PP2Gral. Roca

CJP2Gral. Roca

Plant 1

Plant 2

Market1Brazil

(Sao Pablo Seaport)

Market2USA

(San Antonio Seaport)

Market3Argentina

(Buenos Aires)

Farm 1 Cipolletti

Farm 2 Neuquén

Farm 3 Cinco Saltos

RC

RC

FFi,j,k

FOSfFi,j,k

PFFi,k,

m

PCJi,k,m

FOSi

,k

FFtSi,

k

CJtSi,

k

FFfSi,

k

FRUTAS Y JUGOS ARG Co. : Supply Chain SketchFRUTAS Y JUGOS ARG Co. : Supply Chain Sketch


24

Goal of the studyGoal of the study: :

To design the supply chain that maximizes To design the supply chain that maximizes the gross benefit of the company, analyzing the gross benefit of the company, analyzing several possible scenarios. several possible scenarios.

FRUTAS Y JUGOS ARG Co. : Goal of the studyFRUTAS Y JUGOS ARG Co. : Goal of the study


25

Each site has a refrigeration chamber which may storage preEach site has a refrigeration chamber which may storage pre--classified classified fresh fruit and raw fruit to be processed by concentrated juice fresh fruit and raw fruit to be processed by concentrated juice plant.plant.

Global mass balances of the plants (packaging and concentrate juGlobal mass balances of the plants (packaging and concentrate juice ice plants).plants).

The operating cost evaluation involves the complete supply chainThe operating cost evaluation involves the complete supply chain, from , from farms to markets. Furthermore, terms like raw material cost, farms to markets. Furthermore, terms like raw material cost, transportation cost, production cost, etc., must be consideredtransportation cost, production cost, etc., must be considered

The The Plant’s production may be split in different ways to satisfy the demands, so it must be considered at the time to evaluate sales.

Company benefit is evaluated as a gross benefit (profit), i.e. the difference between sales and operating costs.

FRUTAS Y JUGOS ARG Co. : Bases for the studyFRUTAS Y JUGOS ARG Co. : Bases for the study


26

To produce packed fresh fruit and/or concentrate juice after croTo produce packed fresh fruit and/or concentrate juice after crop p time to fulfill the unsatisfied demand, using as raw material frtime to fulfill the unsatisfied demand, using as raw material fruit from uit from refrigeration chambers.refrigeration chambers.

The Company is committed to receive all the farm production. The Company is committed to receive all the farm production.

Considering the farm production uncertainty. Analyze the supply Considering the farm production uncertainty. Analyze the supply chain profit when one, two or three farms loose a given maximum chain profit when one, two or three farms loose a given maximum production percentage.production percentage.

According to the most important term of the total cost equation,According to the most important term of the total cost equation,suggest possible actions to improve the global profitsuggest possible actions to improve the global profit

Analyze the product price uncertainty.Analyze the product price uncertainty.

FRUTAS Y JUGOS ARG Co. : Possible scenariosFRUTAS Y JUGOS ARG Co. : Possible scenarios


27

A linear model has been developed to maximize the A linear model has been developed to maximize the gross profit of the company along the harvest time, gross profit of the company along the harvest time, January to May.January to May.

The model assign:The model assign:

plant operation levelsplant operation levels

amount and place where raw material should be amount and place where raw material should be obtainedobtained

and final product deliveryand final product delivery

FRUTAS Y JUGOS ARG Co. : SC ModelFRUTAS Y JUGOS ARG Co. : SC Model


28

Based on:Based on:

demands from three major marketsdemands from three major markets

estimated fruit production estimated fruit production

economic information economic information

yield and availability of processing plantsyield and availability of processing plants



29

Objective Function:Objective Function:Maximize Gross ProfitMaximize Gross Profit

Model equations:Model equations:mass balance and production equations at mass balance and production equations at farms, packaging plants and juice plantsfarms, packaging plants and juice plants

Constraints:Constraints:production limits at proprietary farms, bounds on fruit supply, bounds on internal processing capacities, demand satisfaction.



kiPP

kikikiki

FFfSPFFFFtSFF

,

,,,, η

−+=

kiFFISLki ISLFFfSkiFF ,, *

,η=

kiPFFki MaxPFFFFtSki ,, *

,η=

PP Mass Balance

PP global balance


PPi,k Initial Stock

Fresh fruit send to stock

PPk Yield

PPk Maximum Production

( ) kiPPki FFFOSki ,, *1

,η−=

kiMaxPFFki MaxPFFPFFki ,, *

,η=

kikiCJP

kikiki FOStSPCJFOSfSFOSFOSfFki

,,,,, *1

,

+=++η

kiCJISLki ISLFOSfSkiCJ ,, *

,η=

kiki CJPkiPCJki MaxPCJFOStS,, ,, * ηη=

kiMaxPCJki MaxPCJPCJki ,, *

,η=

Out of Spec. fruit to Stock

PCJk Maximum Production

CJP Mass Balance

CJP global balance

PCJi,k initial stock



kikjiPPkji FFFF ,,,,, ∗=α

kikjiCJPkji FOSfFFOSfF ,,,,, ∗=α

1,, =∑j

kjiPPα

1,,=∑

jCJP kji

α

Fresh fr. out of Spec. from farm j to CJPk

Fruit Supplier’s Distribution

Fresh fr. from farm j to PPk

Normalization

kimkiPPmki PFFPFF ,,,,, ∗= β

kimkiCJPmki PCJPCJ ,,,,, ∗= β 1,, =∑m

mkiPPβ 1,,=∑

mCJP mki

β

mik

mkimi DPFFPFFusDPFF ,,,, −= ∑

mik

mkimi DPCJPCJusDPCJ ,,,, −= ∑

Packed fesh fr. delivered by plant k to market m.

Packed conc. juice delivered by plant k to market m.

Unsatisfied demand of packed fresh fruit

Unsat. demand of conc. juice

Demand distribution


( ) ( )∑∑ +=

=+=

kikii

kikii

FOSFF

FOSfFcFOSFFcFF TRFCTRFCTRFC

,,

,, **

( ) ( )∑∑ +=

=+=

kikiki

kikiki

FOSFF

FOSfScFOSfSFFfScFFfS TFfSCTFfSCTFfSC

,,,

,,, **

( ) ( )∑∑ +=

=+=

kjikjikj

kjikjikj

FOSFF

FOSfFdjtcFFFFdjtcFF TFPTrCTFPTrCTFPTrC

,,,,,

,,,,, ****

Raw fruit cost

Fruit from stock cost

Farms to plants fruit trans. cost


Costs

( ) ( )∑ ∑+=

=+=

ki kikikikiki

CJPPP

PCJpcCJPPFFpcPP TPCTPCTPC

, ,,,,, **

( ) ( )∑∑ +=

=+=

kiki

kiki

CJFF

FOStSrcCJFFtSrcFF TRCTRCTRC

,,

,, **

( ) ( )∑∑ +=

=+=

mkimkimk

mkimkimk

PCJPFF

PCJdmtcPCJPFFdmtcPFF TPMTrCTPMTrCTPMTrC

,,,,,

,,,,, ****

Total production cost

Total refrigeration cost

Total plants to market products transport cost



Costs

( )∑ ∗∗=mi

PFFmimi prfpPFFusDPFFusDPFFC,

,,

( )∑ ∗∗=mi

PCJmimi prfpPFFusDPCJusDPCJC,

,,

usDPCJCusDPFFCTPMTrCTRCTPCTFPTrCTFfSCTRFCCostTotal

+++++++=

Unsatisfied Demand PFF Cost

Unsatisfied Demand of Packed Conc. Juice

( ) ( )∑ ∑+=

=+=

mki mkimkimimkimi

CJPPP

PCJpPCJPFFpPFF SalesSalesSales

,, ,,,,,,,, **

TotalCostSalesProfit Gross −=

Gross Profit

Sales

Total plant cost

kiki MaxPFFPFF ,, ≤

kiki MaxPCJPCJ ,, ≤

kiFFki ISLFFfS ,, ≤

kiki MaxPFFFFtS ,, ≤

kiCJki ISLFOSfS ,, ≤

kiCJPkiki MaxPCJFOStS,,, η≤

mik

mki DPFFPFF ,,, ≤∑

mik

mki DPCJPCJ ,,, ≤∑

( )∑ ≤+k

jiFkjikji MaxFPFOSfFFFji ,,,,, *

,η

Inequality constraints




A. Blanco , G. Masini, N. Petracci, A. BandoniA. Blanco , G. Masini, N. Petracci, A. Bandoni

Operations Management of a Packaging Plant in the Fruit Industry

Operations Management of a Packaging Plant in the Fruit Industry


Journal of Food Engineering 70, 297-307 (2005)Journal of Food Engineering 70, 297-307 (2005)

35

DR PC WA WX QC GC PK

NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11



36


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Fruit from farms

Drencher(washing)


37


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


To processing line

To cold storage


38


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Non ProcessedFruit Storage

Own Cold Storage

Third Party Cold Storage


39


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Pre classification(damaged fruit)

Waste(to juice production)


40


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Washing Waxing


41


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Quality Classification(1st, 2nd, 3rd quality)



42


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Gauge Classification(weight or size)



43


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Out of Specification Fruit

Processed Fruit

Packaging


44


NPFS PFSOSF

TPFS

W1 W2 W3

X1

X3

X2

X10

X4

X9X5 X7X6 X8

XOS

X11


Overseas, regionaland local markets


45

• Regular income of fruit (apples and pears) during the whole harvest period (day 12 to day 157)

• Based on historical records it is possible to forecast an income profile in terms of amount, quality, waste and gauge (average value and standard deviation)



46

v Variety SD HP A(kg/day) SDev(kg/day) CR($/kg)v1 Williams pear 12 12 to 36 45560 2000 0.07v2 Beurre D´Anjou pear 45 45 to 69 27400 1500 0.1v3 Beurre Bosc pear 72 72 to 96 35960 2200 0.07v4 Red apple 1 78 78 to 102 24240 780 0.08v5 Packams Triumph pear 91 91 to 115 33200 2100 0.09v6 Red Delicious apple 95 95 to 119 38600 3000 0.08v7 Red apple 2 123 123 to 147 39040 2800 0.07v8 Granny Smith apple 133 133 to 157 48360 3500 0.08

FRUIT INCOMEFRUIT INCOME


47

v q1 q2 q3 q1 q2 Q3

v1 0.58 0.27 0.15 0.063 0.045 0.025v2 0.58 0.26 0.16 0.094 0.045 0.02v3 0.5 0.35 0.16 0.077 0.045 0.019v4 0.58 0.27 0.15 0.07 0.03 0.029v5 0.58 0.34 0.08 0.106 0.057 0.013v6 0.4 0.27 0.33 0.043 0.049 0.038v7 0.58 0.29 0.13 0.092 0.041 0.023v8 0.45 0.32 0.23 0.091 0.055 0.023

A(%) SDev(%)

FRUIT QUALITYFRUIT QUALITY


48

v DPC DQC DGC DPC DQC DGCv1 0.11 0.05 0.06 0.014 0.003 0.007v2 0.14 0.03 0.06 0.011 0.004 0.006v3 0.12 0.03 0.07 0.012 0.005 0.005v4 0.13 0.04 0.06 0.014 0.004 0.007v5 0.1 0.05 0.07 0.011 0.003 0.006v6 0.11 0.04 0.07 0.011 0.005 0.006v7 0.13 0.04 0.05 0.013 0.004 0.007v8 0.12 0.03 0.07 0.013 0.003 0.006

A(%) SDev(%)

WASTEWASTE


49

v g1 g2 g3 g4 g5 g1 g2 g3 g4 g5

v1 0.12 0.11 0.23 0.14 0.4 0.004 0.004 0.008 0.005 0.015v2 0.11 0.16 0.19 0.24 0.3 0.005 0.008 0.009 0.009 0.012v3 0.2 0.12 0.18 0.21 0.29 0.009 0.005 0.008 0.009 0.009v4 0.18 0.12 0.22 0.22 0.26 0.008 0.005 0.008 0.009 0.012v5 0.3 0.22 0.2 0.21 0.09 0.009 0.007 0.008 0.007 0.003v6 0.21 0.26 0.12 0.26 0.16 0.01 0.012 0.005 0.011 0.007v7 0.1 0.23 0.16 0.13 0.38 0.004 0.008 0.005 0.005 0.012v8 0.27 0.16 0.16 0.14 0.28 0.012 0.007 0.005 0.005 0.011

A(%) SDev(%)

GAUGEGAUGE


50

• One permanent labor staff the whole year (eight hour working shift)

• Temporary labor staff may be required to cover two or three additional eight hour working shifts during certain periods in order to satisfy commercial commitments

LABOR POLICYLABOR POLICY


51

• Different markets demand different products

• Products are classified according to:– Fruit variety (v1, …, v8)– Waxing (waxed or not waxed)– Fruit quality (1st, 2nd, 3rd))– Gauge (g1, …, g5)– Crate (only one in the present work)

FINAL PRODUCTSFINAL PRODUCTS


52

v1 v2 v3 v4 v5 v6 v7 v8 w1 w2 q1 q2 q3 g1 g2 g3 g4 g5 p1 PP ($/kg)P1 1 1 1 1 1 0.33P2 1 1 1 1 1 0.39P3 1 1 1 1 1 0.26P4 1 1 1 1 1 0.28P5 1 1 1 1 1 0.24P6 1 1 1 1 1 0.34P7 1 1 1 1 1 0.27P8 1 1 1 1 1 0.29P9 1 1 1 1 1 0.32P10 1 1 1 1 1 0.36P11 1 1 1 1 1 0.28P12 1 1 1 1 1 0.31P13 1 1 1 1 1 0.28P14 1 1 1 1 1 0.29P15 1 1 1 1 1 0.29P16 1 1 1 1 1 0.29P17 1 1 1 1 1 0.29P18 1 1 1 1 1 0.26P19 1 1 1 1 1 0.31P20 1 1 1 1 1 0.31Set SPVP WP QP GP

PACKAGING PRODUCTSPACKAGING PRODUCTS


53

Overseas P1, P2, P6, P9, P14, P15, P19, P20Regional P4, P7, P8, P13, P16, P17Local P3, P5, P10, P11, P12, P18

PRODUCTS AND DELIVERY DATES FOR DIFFERENT MARKETSPRODUCTS AND DELIVERY DATES FOR DIFFERENT MARKETS

Overseas Regional Local25 30 3150 60 5480 90 78

110 120 102140 150 126170 180 150190 210 174215 240 198

270 222300 246330 270360 294

318342360


54

Deterministic income fruit scenario generation by means of Monte Carlo Simulation for each parameter:

PLANNING MODELPLANNING MODEL

amount of fruit per varietyquality of fruit per varietygauge of fruit per varietywaste of fruit per variety

Mass Balances

DrencherPreclassificationWaxing moduleQuality classification moduleGauge classification modulePackaging stage

Packaging plant Cold storageNon processed fruit (total and per variety)Processed fruit (total and per product)Out of specification fruitTotal mass balanceThird party storage


55

• Maximum Processing Capacity is given by the amount of fruit that can be handled at the Pre Classification module

• It depends on the number of working shifts and the processing capacity per shift

• Objective function : profit oriented mode

PLANNING MODELPLANNING MODEL

Total Profit = sales income - raw material cost -labor costs - cooling costs


56

• A plan that maximizes the production of each particular product is obtained

• The resulting plan constitutes a “forecast”of the processing capacity of the facility

• Valuable for managers to establish next year sales commitments

PLANNING MODEL: Profit oriented modePLANNING MODEL: Profit oriented mode


57

PLANNING MODEL: ResultsPLANNING MODEL: Results


58

• Profit = $ 3,331,623• Sales income = $ 58,423,785• Raw material cost = $ 33,242,063• Operating cost = $ 21,783,926• Cooling cost = $ 111,171

PLANNING MODEL: ResultsPLANNING MODEL: Results


59

Provided a historical profile of fruit income and an established sales program, generate a processing plan in order to maximize total profit, while penalizing non satisfaction of sales commitments in terms of volume of fruit and delivery deadlines.

PLANNING MODEL: Sales oriented modePLANNING MODEL: Sales oriented mode




G. Durand, A. BandoniG. Durand, A. Bandoni

A Novel Method to Reduce Event Variables in Continuous-time

Formulation for Short-term Scheduling

A Novel Method to Reduce Event Variables in Continuous-time

Formulation for Short-term Scheduling


7th World Congress of Chemical EngineeringGlasgow, Scotland

July 10-14, 2005

7th World Congress of Chemical EngineeringGlasgow, Scotland

July 10-14, 2005

61

INTRODUCTION – MULTIPURPOSE BATCH PLANTS OPTIMIZATIONINTRODUCTION – MULTIPURPOSE BATCH PLANTS OPTIMIZATION

How to achieve the maximum production, the maximum profits and/or the minimum costs?

How to achieve the maximum production, the maximum profits and/or the minimum costs?

• total operation time (time horizon)

• process recipe

• quantity and capacity of units

• products’ demand

• total operation time (time horizon)

• process recipe

• quantity and capacity of units

• products’ demand

Knowing:Knowing:

SOLUTION: Schedule optimizationSOLUTION: Schedule optimizationDetermining:Determining: • the sequence and the timing of tasks taking place in

each unit

• the batch size of tasks (i.e. the processing time and the required resources/utilities)

• the amount of final products sold and raw materials consumed

• the sequence and the timing of tasks taking place in each unit

• the batch size of tasks (i.e. the processing time and the required resources/utilities)

• the amount of final products sold and raw materials consumed


62

SCHEDULE OPTIMIZATION – STATE TASK NETWORKSCHEDULE OPTIMIZATION – STATE TASK NETWORK

T1 T2 T3 T4 T5F1 S1 S2 INT1 S3 P1

T6P2

WS

0.98

0.02

T10P3

T7S5F2

T8

S4

INT2T9

S60.9

0.10.50.5

A framework for graphical representation and mathematical formulation of recipes (processes) for product manufacturingA framework for graphical representation and mathematical formulation of recipes (processes) for product manufacturing

Uses materials (states) and tasks as building blocks for the process description, with each task consuming and producing materials while using equipment

Uses materials (states) and tasks as building blocks for the process description, with each task consuming and producing materials while using equipment

raw materialsraw materialsintermediatesintermediates

final productsfinal productstaskstasks


63

MODELLING FOR SCHEDULE OPTIMISATION – STN/RTN FORMULATIONSMODELLING FOR SCHEDULE OPTIMISATION – STN/RTN FORMULATIONS

Gantt chartGantt chart

Unit1

t

units

Tf2BTf

1BTf1A,Ts

1B

Ts1A

Tf2ATs

2A

Unit2

Task A Task B

Gives smaller problemsand better LP relaxationsGives smaller problems

and better LP relaxationsDecoupling task events

and unit events(Ierapetritou & Floudas, 1998)


Periods of constantand equal lengthPeriods of constantand equal length

MILP formulation.Too many periods neededto model in a realistic way

MILP formulation.Too many periods neededto model in a realistic way

Discrete time representation (STN)(Kondili et al., 1993)

Task1A

Uni

t2

t

tasks

Uni

t1

Task1B

Task2A

Task2B

Gantt chartGantt chartEvent: period of different lengthfor each occurrence of a taskEvent: period of different lengthfor each occurrence of a task

Produces smaller problems andmodels the time in a more realistic way

Produces smaller problems andmodels the time in a more realistic way

Continuous time representation (RTN)(Schilling & Pantelides, 1996)

t

tasks

Tf2BTf

1BTf1A,Ts

1B

Ts1A

Tf2ATs

2A

Task1A

Uni

t1

Task1B

Uni

t2

Task2A

Task2B


64

SCHEDULE OPTIMISATION IS AN NP-HARD PROBLEMSCHEDULE OPTIMISATION IS AN NP-HARD PROBLEM

Solution performance vs. Problem sizeCase Study I

0

100200

300400

500

600700

800900

1000

8 9 10 11 12 13 14 15

Qty. of events

CPU

Tim

e [s

]

0

1000020000

3000040000

50000

6000070000

8000090000

100000

Itera

tions

CPU time [s] Iterations

STN/RTN formulations have, in general, a practical(*) size limit of 15-20 events

STN/RTN formulations have, in general, a practical(*) size limit of 15-20 events

(*)Practical: solvable in less than 3600 seconds in a 1GHz/256MB system(*)Practical: solvable in less than 3600 seconds in a 1GHz/256MB system


65

IMPROVEMENTS FOR RTN FORMULATION – 1IMPROVEMENTS FOR RTN FORMULATION – 1

Decomposition approaches(Basset, Pekny and Reklaitis, 1996 – Khmelnitsky, Kogan, and Maimon, 2000 – Gupta and Maranas,

1999 – Wu and Ierapetritou, 2003)

One large problemOne large problem

Unit1

t

units

Unit2

Task A Task B

Unit1

t

units

Unit2

Task A Task B

Unit1

t

units

Unit2

Several smaller subproblemsSeveral smaller subproblems

How to model this?OR

How to give to the solver the possibility of choosing this solution?


66

IMPROVEMENTS FOR RTN FORMULATION – 2IMPROVEMENTS FOR RTN FORMULATION – 2

Reducing problem size in other dimensions(Maravelias & Grossmann , 2003)

• Tasks’ starting times can only take place at determined time points (TN), thus reducing the number of time ordering equations.

• Storage is not modelled as a task, therefore reducing the number of binary variables.

• Tasks’ starting times can only take place at determined time points (TN), thus reducing the number of time ordering equations.

• Storage is not modelled as a task, therefore reducing the number of binary variables.

Task A Task B Task C


Unit1

t

units

Unit2

Unit3

TN1 TN2 TN3 TN4Tf3C Tf

2B Tf2B


67

PROCESS RECIPE MODELLING IN RTNPROCESS RECIPE MODELLING IN RTN

Two sets of constraints (equations) model the process recipe in RTN:Two sets of constraints (equations) model the process recipe in RTN:

Material balances

Each task can only consume states that are being produced or were produced by its corresponding preceding task/s (and/or raw materials).

Material balances

Each task can only consume states that are being produced or were produced by its corresponding preceding task/s (and/or raw materials).

Time ordering

If a task consumes states produced by other tasks, it has to start after those tasks have started (continuous) or after they have finished (batch)

Time ordering

If a task consumes states produced by other tasks, it has to start after those tasks have started (continuous) or after they have finished (batch)

( ) ( ) SsNnnjiBpnjiBps is i Ii Jj

cccsi

Ii Jj

pppsi ∈∈∀=−− ∑ ∑∑ ∑

∈ ∈∈ ∈

,0,,1,,

( ) ( ) ( ) ( )[ ]NnIiIiJjj

niwvniwvHnjiTnjiT

pc jp

jcpc

pcppfccs

∈∈∈∈∀

−−−−−≥

,,,,1,,21,,,,

Bs and Ts are continuous variables


68

MODELLING THE PROCESS RECIPE THRU BINARY VARIABLESMODELLING THE PROCESS RECIPE THRU BINARY VARIABLES

task i task i’

intermediatestates

intermediatestates

NnYY niin ∈∀= '

One-to-one pairs of tasksOne-to-one pairs of tasks

Both tasks continuous OR task i continuous/task i’ batchBoth tasks continuous OR task i continuous/task i’ batch

NnNnYY niin ≠∈∀= + ,1'

Task i batch/task i’ continuousTask i batch/task i’ continuous


69


One-to-many pairs of tasksOne-to-many pairs of tasks

All tasks continuous ORtask i continuous/tasks i’ batch

All tasks continuous ORtask i continuous/tasks i’ batch

Task i batch/tasks i’ continuousTask i batch/tasks i’ continuous

task i

task i’1intermediatestates

intermediatestates

task i’2

task i’k

Nn

YY

YYYY

YYYY

niin

niin

niin

nininiin

k

K

∈∀

⎪⎪⎪

⎭

⎪⎪⎪

⎬

⎫

≥

≥≥

+++≤

'

'

'

'''

2

1

21

M

K

NnNn

YY

YYYY

YYYY

niin

niin

niin

nininiin

k

K

≠∈∀

⎪⎪⎪

⎭

⎪⎪⎪

⎬

⎫

≥

≥≥

+++≤

+

+

+

+++

,

1'

1'

1'

1'1'1'

2

1

21

M

K


70

Nn

YYYYYY

YYYY

nininini

nini

nini

nini

K

K

∈∀

⎪⎪⎪

⎭

⎪⎪⎪

⎬

⎫

≥+++≤

≤≤

'

'

'

'

21

2

1

K

M


Many-to-one pairs of tasksMany-to-one pairs of tasks

All tasks continuous ORtasks i continuous/task i’ batch

All tasks continuous ORtasks i continuous/task i’ batch

Tasks i batch/task i’ continuousTasks i batch/task i’ continuous

task i’

task i1 intermediatestates

intermediatestates

task i2

task ik

NnNn

YYYYYY

YYYY

nininini

nini

nini

nini

K

K

≠∈∀

⎪⎪⎪

⎭

⎪⎪⎪

⎬

⎫

≥+++≤

≤≤

+

+

+

+

,

1'

1'

1'

1'

21

2

1

K

M


71

CASE STUDY I – Multiproduct Plant (15 products/11 units/34 tasks)CASE STUDY I – Multiproduct Plant (15 products/11 units/34 tasks)

Modelled with:Ierapetritou & Floudas formulation (1998)I & F form. with proposed constraints

Modelled with:Ierapetritou & Floudas formulation (1998)I & F form. with proposed constraints─ One-to-one pairs─ One-to-one pairs ─ One-to-many pairs─ One-to-many pairs


72

Stopping criterion: Integrality gap <10%Hardware: AMD Duron 996Mhz – 240Mb RAM memory Software: GAMS 21.2/CPLEX 8.1Stopping criterion: Integrality gap <10%Hardware: AMD Duron 996Mhz – 240Mb RAM memory Software: GAMS 21.2/CPLEX 8.1

CASE STUDY I – RESULTSCASE STUDY I – RESULTS

Optimal schedule (obtained with proposed modifications)Optimal schedule (obtained with proposed modifications)


73

CASE STUDY II – Multiproduct Plant (3+1 products/6 units/10 tasks)CASE STUDY II – Multiproduct Plant (3+1 products/6 units/10 tasks)

Modelled with:Maravelias & Grossmann formulation (2003)M & G form. with proposed constraints

Modelled with:Maravelias & Grossmann formulation (2003)M & G form. with proposed constraints─ One-to-one pairs─ One-to-one pairs

T1 T2 T3 T4 T5F1 S1 S2 INT1 S3 P1

T6P2

WS

0.98

0.02

T10P3

T7S5F2

T8

S4

INT2T9

S60.9

0.10.50.5

BMAX in tons, α in hr, γ in kg/hr, δ in kg/hr*tonBMAX in tons, α in hr, γ in kg/hr, δ in kg/hr*ton


74

CASE STUDY II – RESULTSCASE STUDY II – RESULTS

Stopping criterion: Integrality gap <10%Hardware: AMD Duron 996Mhz – 240Mb RAM memory Software: GAMS 21.2/CPLEX 8.1Stopping criterion: Integrality gap <10%Hardware: AMD Duron 996Mhz – 240Mb RAM memory Software: GAMS 21.2/CPLEX 8.1

Optimal schedule (obtained with proposed modifications)Optimal schedule (obtained with proposed modifications)


75

CONCLUSIONSCONCLUSIONS

A new set of constraints for modeling the process recipe for scheduling problems is presented.

They are an additional method, reinforcing the task relations expressed thru material balances and time ordering.

They can be applied to several RTN representations.

The proposed modifications allow finding better values of the objective function in less time than the original formulation where they are applied.

The improved performance comes from the effective elimination of binary variables and faster integer cuts.

A new set of constraints for modeling the process recipe for scheduling problems is presented.

They are an additional method, reinforcing the task relations expressed thru material balances and time ordering.

They can be applied to several RTN representations.

The proposed modifications allow finding better values of the objective function in less time than the original formulation where they are applied.

The improved performance comes from the effective elimination of binary variables and faster integer cuts.


76

Planta Piloto deIngeniería QuímicaPlanta Piloto deIngeniería Química


operations management in the fruit industrycepac.cheme.cmu.edu/pasilectures/bandoni/pasi 2005 -...

Documents