3 The Architecture of the MAS Semi-finishing: at this level the main activities are calendering,

2. 2.3 The Architecture of the MAS Semi-finishing: at this level the main activities are calendering,

extrusion and cutting of mixtures; the number of components Each site in the supply chain is represented by three agents:

manufactured in a day is really huge (250 different items out • of 300).

a Procurement Agent (PA), which supports the functionality of the Purchase Department;

3. Assembling: 24 semi-automatic and 3 fully automated • machines build up green tires.

a Sales Agent (SA), which interacts with customers; a customer might be the procurement agent of another site using the MSP,

4. Curing: green tires are vulcanized and molded in presses, in a or a physical person, working in an internal department (e.g. the

synchronized flow towards finishing units. planning or the logistics department);

5. Finishing: tests are performed on every item coming from •

a Production Management Agent (PMA), which helps the

curing, to detect potential defects.

production management department, and receives information In Carlisle plant production is managed by a pull system: two from the shop-floor and the inbound and outbound logistics. manual kanban loops are present in the process, one between semi-

There are only two kind of interactions among agents belonging finishing and assembly, the other one between assembly and curing. to different plants of a Supply Chain: the SA of a Supplier

Burton deliveries satisfy the most of Carlisle requirements. Burton interacts with the PA of the Customers; and the PA of the nodes at

plant is working 10 shifts in 6 day, out of 21 shift (without the same stage of the supply chain may interact with one another.

stopping) a week performed in Carlisle. Set-up times are also longer Each agent is to a certain extent autonomous; in this way, a

for Burton than for Carlisle process. As a consequence, it is not consistent part of the decision making process can be carried out

possible to establish a pull connection between the two plants. The without the need of user intervention. In particular:

communication of the weekly plan results in a push system. • The visibility provided by such a system is rather poor:

the PA acts as a broker of information, in that it gathers a lot of visibility in semi-finishing and assembling is limited to just one data via communication with the suppliers’ SAs, and then

production unit ahead, while inter-site visibility is limited to the communicates to the PMA only relevant information; weekly communication. The time required to get stock information

• the PMA is the interface among the Production Planning and

is too long and subject to errors.

Control system internal to the node and the MSP; it has The MSP provides accurate on-line information about stock (partial) authority on the plant facilities;

levels and production performance in the main points of the SC. • The access to inter-site stock information is granted only to

the SA is committed towards the customers; in case of authorized sites and people. A web connection between sites problems, it has the authority to decide which order to delay

improves controls on capacity levels and plant productivity. Data (according to some priority criterion set by the user); moreover,

flows (event driven or on demand) could guarantee overall visibility it represents its site in the negotiation process with other

and provide automated demand leveling, reducing problems of agents.

overproduction or stock-out.

Burton plant could better distribute resources and production demand with full on-line visibility of accomplished orders to consider new orders (coming from other SC customers, for

Proceedings of the ECAI 2000 Workshop 13 “Agent Technologies and Their Application Scenarios in Logistics” edited by I. J. Timm et al.

page 58 page 58

change of production in such a continuos process with no need of The production is not at all leveled. While an item run out of

change kanban board.

production process, another is run in production and, as a In the first implementation, the Pirelli Pilot was restricted to consequence, timely stops of production to empty pipeline are not

three sites (Carlisle, Burton, and one external supplier). The MAS managed. The MSP is able to communicate and calculate the

is therefore constituted by 9 different agents: PAS, PAC, PAB, quantity of material to be produced in the different stages to empty

PMAS, PMAC, PMAB, SAS, SAC, SAB.

Burton on

Semi-

Trent

Compounding Finishing

Ot her Suppliers

Figure 3: Pirelli manufacturing process and Supply Chain

10. SAB and SAS evaluate their eligibility by asking to the observes a situation exceeding some pre-defined threshold value.

The automatic re-planning process is initiated whenever a PMA

respective PMAs whether they are able to produce the The objective of this process is to find a solution with the least

auctioned components, and at which costs; negative effect, both in terms of previous plans disruption and additional production, holding, and transport costs. We shall

11. the PAC receives the bids; it then awards the best bidder, and explain the whole process with an example.

communicates the choice to the other bidders also. When an auction is open, the message circulated to all the SAs

1. Due to a machine breakdown, the PMAB has highlighted the need of a re-planning.

indicates the item type, quantity, due date. When submitting a bid, an agent has to indicate the maximum quantity it is able to supply,

2. It communicates the problem to the SAB, which has the and the price. In case the auctioneer required more than one item, authority to decide which job to delay.

the bidder is not compelled to consider all the item in its bid. Finally, the auctioneer selects the set of bids allowing it to

3. The SAB asks to the PAC (its only customer) its priority list minimise an internal cost function. This operation could be (plainly speaking, this is like asking a customer “I’m sending performed in a reasonable amount of time by exploiting a branch- you all the stuff but one component; which one would you

and-bound algorithm.

dislike less to be delivered late?”).

4. Taking into account the plans issued by PMAC, PAC generates a priority list, and then sends it to SAB.

5 CONCLUSIONS

5. Working in collaboration with PMAB, SAB considers several

A set of baselines has been defined by Pirelli in order to measure unfeasible plans, and selects the best (i.e. the less bad) by

the improvement following the introduction of MSP, including considering both the just received priority list and its internal

increased daily output, widened product mix, a higher number of priority list.

mould changes per day, shorter throughput times, lower stock levels and scrap rates, transport savings and so on. Improvement

6. It then communicates to PAC that certain deliveries have been will be assessed by the end of the project (December 2000). put off, and the new (expected) delivery date.

Expected improvements are such, that the investment in the

7. PAC transfers this information to PMAC, which now has MSP system will have a pay-back time shorter than 18 months. updated information on expected delivery terms.

At the moment the authors are conducting further simulations in order to assess the validity of the proposed model in case of three

8. After evaluating the new delivery dates, PMAC may ask to

stages supply chain.

PAC to find an alternative supplier. In this case, an auction procedure is initiated.