Agent-base multiple supplier tool management system

Introduction

Recently, a new software architecture for managing supply networks at the tactical and operational levels has emerged.
The supply network is viewed as composed of a set of intelligent (software) agents, each responsible for one or more activities in the supply network and interacting with other agents in planning and executing their responsibilities
An agent is an autonomous, goal-oriented software paradigm that operates asynchronously, communicating and coordinating with other agents as needed
A multi-agent system consists of a group of different types of agents that can take on specific roles within an organizational structure

Introduction (2)

Most research on multi-agent systems focuses on the coordinative intelligent behaviour among a collection of autonomous intelligent agents considering that the group of agents provides more than the sum of the capabilities of its members
The adoption of multi-agent technology is based on 3 fundamental system domain characteristics:

data, control, expertise or resources are inherently distributed
the system is naturally regarded as a society of autonomous cooperating components
the system contains legacy components that must interact with other, possibly new software components.

Introduction (3)

A supply network consists of suppliers, factories, warehouses, etc., working together to fabricate products and deliver them to customers.
They cooperate with each other autonomously to serve common goals but also have their own interests
A supply network is dynamic and involves the constant flows of information and materials across multiple functional areas both within and between network members
Multi-agent technology, therefore, appears particularly suitable to support collaboration in supply network management

Introduction (4)

The development of a multi-agent tool management system (MATMS), operating in the framework of a multiple-supplier network, is presented in this paper
The aim of the MATMS is the optimum tool inventory sizing and control, including on-time delivery and cost and stock-out risk minimisation
The tools under consideration are the CBN grinding wheels for Nickel base alloy turbine blade fabrication

Turbine blade manufacturing

Turbine blades are manufactured along several production lines, each for one aircraft engine model requiring a set of CBN grinding wheel types (part-numbers)
Each part-number is planned to work a maximum number of blades: when it reaches its end of life, it is sent for dressing to an external supplier in a supply network and remains unavailable for a time defined as dressing cycle time
For each part-number, a sufficient number of CBN grinding wheels (serial-numbers) must be always available (on-hand inventory) to prevent production breakage due to tool run-out

Turbine blade manufacturing

The part-number on-hand inventory size, I, depends on:

number of pieces per month, P
number of pieces per wheel, G
number of months required without new or dressed wheel supply, C (coverage period) heuristically selected

The wheel demand, D, for each part-number is given by:

D = (P/G) * C – I₀

where: P/G = tool demand rate (number of wheels per month); I₀ = initial part-number inventory size

Turbine blade manufacturing

Traditional tool management consists in the strategic planning of the wheel inventory size based on the selection of a coverage on-hand inventory for each part-number (number of wheels for production needs in the coverage period C)
This procedure does not always prove adequate; the producer, aware of this drawback, increases or reduces the part-number inventory level on the basis of experience
The results of this policy, founded on skilled staff knowledge, are the historical inventory size trends: in some cases, the expected trend matches the historical one; in other cases, it is underestimated, with risk of stock-out, or overestimated, with excessive capital investment
The historical trend is a fit solution that prevents tool run-out and useless investment: it can be used as a reference for assessing alternative tool management strategies

Turbine blade manufacturing

Traditional tool management consists in the strategic planning of the wheel inventory size based on the selection of a coverage on-hand inventory for each part-number (number of wheels for production needs in the coverage period C)
This procedure does not always prove adequate; the producer, aware of this drawback, increases or reduces the part-number inventory level on the basis of experience
The results of this policy, founded on skilled staff knowledge, are the historical inventory size trends: in some cases, the expected trend matches the historical one; in other cases, it is underestimated, with risk of stock-out, or overestimated, with excessive capital investment
The historical trend is a fit solution that prevents tool run-out and useless investment: it can be used as a reference for assessing alternative tool management strategies

Generic agent shell

The MATMS was developed by making use of agent development tools integrated in the FIPA Agent Management Reference Model that provides for the creation, registration, location, communication, migration and retirement of agents
The entities contained in the reference model are logical capability sets (i.e. services) and do not imply any physical configuration
These can be combined in the physical implementation of Agent Platforms (AP) defined by FIPA as the environment where agents can physically exist and operate
FIPA agents located on an AP utilise the facilities offered by the AP for realising their functionalities
In this context, an agent, as a physical software process, has a physical life that has to be managed by the AP

Generic agent shell

Agent management reference model

Multi-agent tool management system

The MATMS activities are carried out according to the multi-agent interaction and cooperation protocols
The block scheme of the developed MATMS, subdivided into three functional levels is reported:

the Supplier Network Level, including the external tool manufacturers in the supply network
the Enterprise Level, including the logistics of the turbine blade producer
the Plant Level, including the production lines of the turbine blade producer

MATMS agent communication

Communication is a fundamental aspect of a multi-agent system activity and takes place through exchange of messages between agents
The agents use a common language, the Agent Communication Language (ACL), to transfer information, share knowledge and negotiate with each other
The most widely used ACL is the one developed by the Foundation for Intelligent Physical Agents (FIPA), FIPA ACL, based on the “Speech Act Theory” derived from the linguistic analysis of human communication
The unit for communication analysis is the message, called communicative act or performative

MATMS functioning (1)

The functioning of the MATMS is described in terms of agent communications and activities in the three system levels (see the scheme of MATMS)
Initially, the RA in the Enterprise Level receives information on CBN grinding wheel end-of-life events from the PAi in the Plant Level through Pj-R inform communicative acts containing the part-number and serial-number of the worn out CBN grinding wheels
The RA starts collecting the necessary information to decide whether or not the worn out CBN grinding wheels should be delivered to an external supplier for dressing
For each worn out CBN grinding wheel, the RA obtains information on current monthly production, P, part-number tool life, G, and part-number inventory level, I, by regularly interrogating the K&DBA through a R-K subscribe act, followed by a K-R inform reply

MATMS functioning (2)

The RA asks the DTPA, through a R-D request act followed by a D-R inform reply, for a supplier-independent dressing cycle time prediction
To issue a dressing cycle time prediction, the DTPA needs the updated historical dressing cycle times that it obtains from the K&dba through a D-K subscribe act, followed by a K-D inform reply
On the basis of the supplier-independent dressing cycle time prediction and the values of P and G for the relevant part-number, the RA evaluates the demand for wheel dressing to provide for part-number inventory sizing and control through its domain problem solving function: the FTMS

MATMS functioning (3)

If the RA does not consider necessary the dressing operation for a certain part-number, it informs the K&DBA through a R-K inform act that the worn out CBN grinding wheel serial-number is kept on-hold in the enterprise warehouse
If the RA deems necessary to issue a dressing job order, the Ra sends a R-O request act to the ODA asking to start a procedure for dressing job order allocation; this request is followed by an O-R agree reply
The task of the ODA consists in allocating the required dressing job order to one of the external tool manufacturers in the supply network on the basis of negotiations and constraints
To this purpose, the ODA needs to gather information necessary for supplier selection from the relevant agents: DTPA, K&DBA, SOAA_i.

MATMS functioning (4)

The ODA starts negotiating with the SOAA_i in the Supply Network Level to obtain from each of them the dressing price and time (offers) for the required worn out CBN grinding wheel
The negotiation is initiated by an O-S_i call for proposals act to which the Soaa_i respond with their offers through S_i-O propose acts
Simultaneously, the ODA obtains from the DTPA the supplier-dependent dressing cycle time prediction and from the K&DBA the dressing job reference price, due-time, etc.
On this basis, the ODA ranks the responding suppliers and allocates the dressing job order to the first supplier in the rank through an O-S_i accept proposal
The selected SOAA_i accepts (refuses) the dressing job order through a S_i-O agree (refuse) act. In case of order refusal, the ODA contacts the second supplier in the rank list, and so on.

MATMS functioning (5)

The ODA informs the K&DBA about the order allocation results through an O-K inform act and requests the Wta to send the worn out CBN grinding wheel for dressing to the selected supplier using an O-W request act, followed by an W-O agree reply
The Wta records the delivery and reception dates of each CBN grinding wheel for actual dressing cycle time evaluation
These data are regularly fed to the K&dba that makes them available for further requests and interrogations by the relevant agents; this is obtained through a K-W subscribe act, followed by W-K inform replies

Conclusions

The development of a multi-agent tool management system (MATMS) for automatic tool procurement, operating in the framework of a negotiation-based multiple-supplier network, was presented
The MATMS was developed by making use of agent development tools integrated in the FIPA Agent Management Reference Model
The fundamental aspect of agent communication for information transfer, knowledge share, and negotiation was realised in the MATMS through the FIPA Agent Communication Language (ACL)
The MATMS functioning was illustrated with reference to the real industrial case of a turbine blade producer requiring dressing jobs on worn out CBN grinding wheels from different tool manufacturers in a supply network