Using agent techology to manage activities in the manufacturing department.

Popa, Cicerone Laurentiu ; Parpala, Lidia Florentina

1. INTRODUCTION

Agent technologies are suitable for domains where:

* highly complex problems need to be solved or highly complex systems to be controlled;

* the information required for solving problems or controlling systems is distributed and is not available centrally, because: it is distributed geographically and the agents do not want to share it.

* in domains with dynamically changing environment and problem specification.

* different pieces of software need to be integrated in a run time and in the design times. (Vokrinek & Rollo, 2004)

In agent-based programming, the agent is the basic element of distribution. Each agent serves as an independent component with its own local state and execution model. The agent designer can choose to assign a particular set of functionalities to an agent, specify the types of events and messages the agent may invoke or respond to, and implement those triggers and/or responses. In building an agent-based system, each agent behaves independently, interacts with other agents through the events and messages that are communicated from one agent to another. (Weiss, 1999). The agents involved must be able to take decisions, to offer possibilities of cooperation and to negotiate relations between different domains involved in the project. (Popa&Hadar 2006). Agents have the folowing characteristics: autonomy, reactivity, proactiveness, social abilities, mobility, coordination, communication, and cooperation. (Wooldrige, 2002).

2. USING AGENTS TO MANAGE THE ACTIVITIES

Manufacturing planning represents a matter difficult to achieve, especially if it's part of an open and distributed application, from a dynamic environment. In such an application rarely will happen that planning will be up to date with the schedule, the system may require fulfillment of tasks that were not originally foreseen, and sometimes it may be decided to omit certain tasks or unexpected tasks may occur. Also, some resources may not be available at the appropriate time or other types of resources may be needed. Differences may occur concerning execution times, some activities may take longer than was expected, and these delays could affect other activities in progress. In such a system an agent can be responsible for managing a machine, a tool or a human resource.

The agent must have all the information concerning the machine: all the technical data, processing systems, types of tools that can be used, types of tools available in the store, types of workpieces that can be processed. This agent must permanently interact with other agents from the system in order to obtain information regarding workpieces to be processed, the desired shape and characteristics of the processed part etc.

The agent must also obtain information about the activity of the other machines in real time, especially when certain processing operations of the part should initially be made on another machine or even after. The agent must be able to manage a complex database where information is permanently recorded. For example, the agent will know when a tool is used and replaced because the database contains information about the work time of that tool. The supervising agent is in permanent connection with the agent that manages the conveyor or the robot that handles the workpieces and the parts. He should know all these times precisely, in order to avoid a pieces blockage. In fact, all these agents are supervised at the level of the multiagent system by a "supervising agent" who has the role to manage the entire data flow and the agent's activities in the system that is organized hierarchically.

The supervising agent must manage efficiently the crisis; he will communicate to the other agents involved in the process that a problem occurred in the system. The supervising agent can decide: to interrupt all the activities from the system, or to overlying the operations. The supervising agent will offer a solution by planning-programming-reprogramming the activities affected. (Popa et al, 2007)

3. CASE STUDY

Two languages were imposed as the most used: KQML and FIPA-ACL, but generally each creator of a multiagent system uses its own language of communication. FIPA language describes in detail each participant's: name, address, its class, and presents in detail the message itself that is meant to be sent. In conclusion, it gives details of the language used, protocols etc. In our example we will use a very simplified communication model between agents. It will contain only the receiver agent, the sender agent, commands and information defined by the user etc. In this case study will use the system architecture described in Figure 1, in which a possible dialogue between the agents that manage the activities of the machines from the manufacturing department will be exemplified.

The adapted system for this case study is composed of: store for workpieces and parts; MS32 CNC multi-axis turning machine, AFC 105S CNC boring and milling machine, Blohm Precimat CNC grinding machine, four conveyors for parts transport; five robots for the transfer of parts between machines and conveyors.

[FIGURE 1 OMITTED]

The agents defined in this case study are:

* "superviz" is the system's supervising agent in the manufacturing department;

* "turning_machine" the agent that manages the multi-axis turning machine activity in the department;

* "machine_AF" the agent that manages the boring and milling machine in the department;

* "grinding_machine" the agent that manages the grinding machine in the department;

* "store" is the agent who manages the entrances, exits of the store. This agent owns a database that contains all existing tools, workpieces and parts;

* "depan" is the agent who manages the human resource for troubleshooting;

* "conv1" is the agent that manages the activity of the first conveyor of the system; this conveyor is the link between the store and turning machine and so on;

* "rob_transf1" is the agent that manages the activity of robot 1, which makes the transfer between the store and conveyor 1 and so on;

The supervising agent announces the start of dialogue with all the agents in the system:

To all From superviz Hello

It also ensures that each agent of the system is online and ready to begin the activity:

To conv1 From superviz Ping

To rob_transf1 From superviz Ping

To turning_machine From superviz Ping

Such verification is done for each agent part of the system, if the agent is "online" it will reply:

To superviz From store Online

To superviz From turningmachine Online

[FIGURE 1 OMITTED]

We can have a situation in which an agent does not respond, in our case the agent of conveyor 1. The supervising agent will repeat the message and if in a given period of time he will not receive a response he will assume that there is a problem with that machine or with the way the agent that manages that machine was programmed. The supervising agent will notify the agent that manages the human resource to intervene on the machine and the application that sustains that agent in order to identify the error and the activity of that agent:

To depan From superviz conv1 goto

To depan From superviz conv1 71f390b problem

The "depan" agent receives the code of the problem, in our case the conveyor's agent does not respond to the supervising agent's request and the application will automatically contact a qualified human resource for this kind of problem and to which it will submit a report on the previous activities of agent and the machine managed by him. Also, time estimation for solving the problem will be required; when it will be possible the "depan" agent will send this estimation:

To depan From superviz conv1 time estimation

To superviz From depan 0.15.21 time

The supervisor announces the agent that manages the store to prepare 50 parts to be transported to the turning machine in order to be machined:

To store From superviz 50 part1 prepare

Once the parts are ready, the agents that manage the robots which make the transfer are notified, as well as the conveyor's agent who will carry the part to its destination. They will confirm that they are ready to begin the activity:

To rob_transf1 From store part1 ready

To store From rob_transf1 begin activity

To conv1 From rob_transf1 part1 ready

To rob_transf1 From conv1 part1 begin activity

The supervising agent announces the agent that manages the turning machine that part 1 must be processed and the number of such parts that will arrive:

To turning_machine From superviz part1 arrive

To turning_machine From superviz 50 parts arrive

It requires an estimation of the time needed for processing a part:

To turning_machine From superviz part1 time estimation

In order to make time estimation, the machine's agent requests the part's data and full information about it. The agent has access to the complete database of the system, it extracts comprehensive information regarding the part: types of surfaces, the material of the part, the operation plan, necessary tools and delivers them to the requesting agent.

To turning_machine From superviz part1 information

Based on the obtained information, the agent makes a list of the order of necessary operations, necessary tools. Based on this analysis he makes an estimation of the time necessary for processing and transmits it to the supervising agent.

To superviz From turning_machine 0.10.30 time

The supervising agent examines this time, compares it with the initially estimated time in the design stage of the part and of its material flow. If the processing time falls within the limits accepted by him he forwards the approval for starting the processing.

To turning_machine From superviz time ok

4. CONCLUSIONS

This simple system model can be extended towards a distributed architecture, because the agent oriented technology facilitates a good administration of the system. Thus, the presented system can be integrated in a much more complex one. Our further work will concentrate on the analysis and the simulation of a distributed system, based on agent orientated technology.

5. REFERENCES

Popa, C.L. & Hadar, A (2006). A cooperative system design based on agent technology, Proceedings of 17th International DAAAM Symposium, Katalinic, B. (Ed.), pp. 311-312, ISBN 3-901509-57-7, Vienna-Austria, Published by DAAAM International, Vienna

Popa, C.L., Parpala, L.F. & Aurite, T (2007). Algortihm for problems rezolving in distributed system using agent technology, Proceedings of 18th International DAAAM Symposium, Katalinic, B. (Ed.), pp. 575-576, ISBN 3901509-58-5, Zadar-Croatia, Published by DAAAM International, Vienna

Vokrinek, J.; Rollo, M. (2004). Agent System Design. Available from: http://agents.felk.cvut.cz/Accessed:2008-04-20

Weiss G. (1999). Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence, The MIT Press, ISBN 0-26223-203-6, Cambridge

Wooldrige, M. (2002). An Introduction to MultiAgent Systems, John Wiley&Sons