A model for co-operative planning within a virtual enterprise

L.Ouzizi, D. Anciaux, M. C. Portmann and F. Vernadat MACSI Project, INRIA-Lorraine & LGIPM

Introduction Objectives of industrial enterprises: Minimise different costs (production, transport, holding…)

♦

Establish a strong co-ordination and commitment with other enterprises ♦

♦ Share

benefits and losses with partners


New forms of organisations appear: Extended Enterprise and Virtual Enterprise (VE)

Several definitions for the Virtual Enterprise (VE) An instance, Makatsoris et al (1996): “A virtual enterprise is a subset of units and processes within the supply chain network, consisting of a matrix of largely co-operating manufacturing, store, and transport units of mixed ownership, which behave like a single enterprise through strong co-ordination and cooperation towards mutual goals”.

Virtual Enterprise

Ent. A

Ent. D Ent. B Ent. E

Ent. C

Final assemblers

Retailers

Problem Planning a virtual enterprise in a dynamic environment

Contents 1- Architecture for the Virtual Enterprise and its planning problems 2- A multi-agent model for the VE 3- The negotiation and planning process within each VE node 4- Production planning in the Virtual Enterprise 5- Conclusion

Architecture of the Virtual Enterprise Raw materials suppliers Last level NEVs

. . .

NEV1

NEV1

Second level NEVs

First level NEVs

NEVi

NEV2

NEVn

NEV2

NEVj

NEVk NEV1

NEV1

NEVm NEVi

Customers of the VE Flow of products Flow of information

NEV: Virtual Enterprise Node

Planning problems in a VE • Forecasting evolutions (rolling horizon) Forcastings and Contracts FCold

t0

FCnew We are here

t1

t2

t3

t0

t1

t2

tH

t3

New period

tH

New planning obtained by negotiation

• Definition and negotiation of costs A change of planning ⇒ change of production costs (new delays, working overtime, subcontracting…)


To satisfy the customers of the VE, some NEVs can produce without assuring their benefit We propose that: 1- The first objective is to assure the global benefit of the virtual enterprise:

∑

selling

− ∑ NEVs

costs

≥0

all NEVs

2- Costs will be a function of the quantities purchased or delivered

3- Each NEV is able to establish direct and backward planning iterations 4- Each NEV establishes a mathematical model that takes into account different costs and penalties of delay 5- The global benefit can be transformed into compensations to NEVs with negative benefit

A multi-agent model for the Virtual Enterprise Local architecture of a NEV: Negotiator agent Orders

Planning and costs

Planner agent

Multi-agent architecture of the VE Level i

Level i+1

Negotiator agent Ord ers

Planning and costs

Negotiator agent

Planner agent

Ord ers

Planning and costs Planner agent

…

….

Negotiator agent Ord ers

Negotiator agent

Planning and costs Ord ers

Planner agent

Planning and costs Planner agent

Planning parameters

. . .

Negotiator agent for level i+1

Negotiator agent for level i

Mediator agent of the VE

Planning parameters

. . .

Due date deliveries and costs

Customers of VE

The negotiation and planning process within each NEV Scenarii with costs

Planning tools - Product costs - Purchasing lead times -…

- Deliver without problem - Deliver with efforts - Deliver with compensation of over costs - Advance products and delay others ….

- Rules of the game with partners - Contracts with customers

Negotiator agent

Planner agent Internal constraints : - Stock level - Management policies - Capacities -… Variation of demand Change of planning

Negotiator agent of the level (NAL)

Statechart specification of the Negotiator Agent Negotiator Agent NA OFF Planning mission from the NAL + Reply from PA (Planner Agent) NA ON S

PA reply

F_send

Analyse replies and test if a solution is acceptable SA

Send Reply to NAL

Planning mission from the NAL Specify planning conditions to PA

SA/

FM

Change planning conditions

Send planning missions to PA

FM Send back a planning mission to the PA

F_send

F_send

Production planning in the Virtual Enterprise Variation of demands Plannings are negotiated and coherent

Problem of NEVs of the last level - Correct forecasts of finished products - Add a new period for the period T - Update stocks of components

Problem of each NEV - Is it sufficient to add one period to the planning ? - Is it necessary to change more or less the previous planning ?

Algorithm proposed: A0: we affect a row for each level NR: the row of the last level A1: For each level from NR down to 1 do Preliminary work for each level For each NEV of the level do The due date of products manufactured by the NEV are fixed For each combination of parameters (costs, penalties, possible delays...) do Construct the backward planning corresponding to the due date deliveries fixed Endfor each combination of parameters Endfor each NEV EndFor each level

A2: If each NEV does not need to place orders in past periods then - The negotiator agent of the level (NAL) has a set of plannings established in A1 - It selects plannings on the basis of the two local and global objectives If it is OK for every one Else NEVs concerned or the NAL can change weights of some penalties Else the NEVs establish direct plannings, by replacing negative times by 0 ⇒ the due date for purchased products from VE’s suppliers are fixed. A3: For each level from 1 to NR do Preliminary work for each level For each NEV of the level do The due dates of purchased products are fixed For each combination of parameters chosen by the NEV or the NAL do Construct the direct planning (taking into account products manufactured and the values of the two criteria (costs and penalties) Endfor each combination Endfor each NEV EndFor each level

A4: At the end of A3, consistent plannings of products to be delivered to customers are obtained with delays (acceptable or not) and possible overcosts at some NEVs. If each one is satisfied (possible compensation for NEVs) then it is OK Else start again in A3 or in A1 and change penalties or other parameters.

Advantages of the algorithm: - A global and a local benefit for the virtual enterprise is ensured - The computing time of each direct or backward planning is bounded by a polynomial function of the number of NEV’s and the number of periods

Conclusion • Overall planning of a SC is a complex task • A model and execution infrastructure for the planning of a type of VE presented • Perspectives: enhancement of the model and application to other types of VE • Work done within EU IST V-CHAIN Project