Research On The Architecture Of Open CNC System Based On Multi-agent Theory

With the continuous development of information technology and controltechnology, the production scale of mechanical manufacturing industry expandsconstantly and the complexity of mechanical equipment increases continually, whichlead to the diversification and individuation of user requirements. The closedarchitecture of traditional CNC system, which lacks integration, portability andflexibility, has badly restricted the development of CNC system. To meet the need ofthe development of mechanical manufacturing, scholars from both domestic andoverseas carried out many related researches in which a variety of architecture modelsof CNC system have been provided. Most models are built by improving the traditionaldesign approach from the view of engineering. These proposed models, to a certainextent, can meet the demand of high performance and high applicability of CNCsystem. However, these models still didn’t take the demand of intelligence intoconsideration. Multi-agent is an important system modeling theory in distributedartificial intelligence and is originated from the simulation of human social structureand group behaviors. A multi-agent system consists of many independent agents. Eachagent, which can be taken as a physical or abstract entity, needs to communicate witheach other so as to complete the task cooperatively. Multi-agent systems havecharacters of distribution, autonomy and harmony. This paper will focus on themodeling approach and the research of relevant technologies on open architecture ofCNC system based on multi-agent theory. Application of the multi-agent theory in theprocess of open architecture CNC system modeling will not only meet the demand ofopen frame, but also will improve the system integration because of its characteristicsof distributive, flexibility and synchronism. The main works are undertaken mainly infollowing aspects.1. Architectural model of open CNC system based on multi-agent theory isbuilt. Taking the requirement of high intelligence of open CNC system architectureinto consideration, the research of modeling method of open CNC system architecture is carried out in this part. The feasibility and the advancement of applying multi-agenttheory into building architectural model of open CNC system are firstly discussed.After analyzing of the principle and the working flow of CNC system, a model of openCNC system architecture based on multi-agent theory is then provided. In this model,the layered modularized method is adopted which can not only meet demands ofreconfigurability, expandability and portability, but also can improve the systemintegration by using the intellectual technology of distributed artificial intelligence.2. Individual agent modeling methods are studied. After analyzingcharacteristics of human nature of each agent, the study of individual agent modelingmethod is carried out. The social attribute of agent and its influence on group behaviorsin multi-agent environment are expounded. Concerning the disadvantages of existingmodeling methods, the traditional SBDI-agent model is improved by increasing socialattributes of agent. The SBDI-Agent model based on social attributes and theSBDI-Agent based on Petri-Net are then provided. Based on the good ability ofgraphical description of Petri-Net, the mental state of the Agent inspection can then beanalyzed and modeled easily.3. Knowledge model in multi-agent system is studied. After analyzing thecomplexity, isomerism, and fuzzy of knowledge in multi-agent system, a knowledgemodel in multi-agent system based on weight fuzzy petri-net, the corresponding fuzzyreasoning algorithm and learning mechanism based on artificial immune network areproposed. Fuzzy Petri Net (FPN) is a kind of modified fuzzy knowledge modeling toolbased on traditional Petri net. It not only has the same ability of graphic description astraditional Petri net but also is capable of fuzzy reasoning. The combination of positiveand negative bi-directional reasoning algorithm based on WFPN is proposed in thispart. The problem-solving space is reduced by optimizing the WFPN model throughconfirming the rules and conditions that are related to decision objectives by backwardreasoning method. The positive confidence level is calculated by matrix algebraformula, which is helpful to enhance efficiency and search speed of large-scaleknowledge base. Referring to immune regulation mechanism of biological immune system, the immune regulation mechanism based on similarity of antigen is designed.According to those important net parameters obtained by immune regulation, theconcrete structure model of immune net is built. Using the proposed learning algorithmfor extracting rules from knowledge processing process can make the database insystem constantly improved. The experiment result indicates that the proposedreasoning method has significant advantages over the traditional reasoning algorithmin less reasoning time and error rate. The immune algorithm has shown faster learningspeed and higher classification accuracy.4. Communication mechanism in Multi-agent system is studied. Based onanalysis of the characteristics and the process of communication in multi-agent system,the study of communication mechanism in MAS is carried out. First of all, a mixedcommunication mode is presented. Direct communication will be used if there is adirect interaction between agents. Otherwise, indirect communication is implementedby building data shared pool. Secondly, a description method of KQML based on XMLhas been provided. The grammar of KQML message will be standardized by usingXML. The code of XML includes analyzing message, which makes the procedure ofanalysis more convenient. The social characteristic of MAS and the similarity betweenoppressive behaviors and unbalanced states of communication are firstly discussed.Regarding the oppressive behavior, some metrics have been designed to measure thecommunication among agents, and some rules have been devised to classify agentsinto several patterns in the following part. Experiments are finally conducted onCrisis-MAS. Experiment results show that the causes of the unbalancedcommunication design can be figured out by means of oppressive behavior and thattheir metrics, classification rules, and the performance also can be improved byeliminating the oppressive behavior in MAS.5. Negotiation mechanism in multi-agent system is studied. After analyzingthe concept, characteristics and process of negotiation in multi-agent system, the studyof negotiation mechanism in MAS is carried out. By comparing the advantages anddisadvantages of the current mainstream negotiation methods, a concessions negotiation mechanism in multi-agent system based on predicting the negotiationtendency of rival is given. Depending on whether the utility is complementary or not,the utility computing can be divided into two cases in order to avoid the waste ofnegotiation resource caused by low complementary utility value. The negotiationmechanism, using the idea of win-win negotiation strategy for reference, takespsychological attributes of agent as one of the elements that affect negotiation results.In order to solve the negotiation failure problem caused by obsession with utility,negotiation weights can be dynamically modified by predicting the negotiationtendency of rival.