Study On CNC Remote Fault Diagnosis System Based On Multi-agent

Computer Numeric Control machine tools (CNC) plays the role of act foot importance in national economic development. The development of modern big industry makes the structure of CNC increasingly complex, automation level is more and more high. The loss is huge because stopping machine during the production caused by fault, and the maintenance ability of single enterprise is limited. It is necessary that establishing the remote detection and diagnosis of real time system. Equipment offer business, enterprise technical people and domain experts cooperate in time to realize the prevention and diagnose of fault accurately. Multi-Agent System (MAS) has autonomy, distribution and coordinativeness, and have self-organization, learning and inference ability. Actual application system with MAS technical has very strong adaptability, robust and reliability, and is efficient.Combining Programs for Science and Technology Development of Hebei"the development of CNC fault diagnosis system based on WEB"(number: 05213507D), this paper lays out the researches of theory and empirical on the key technology of CNC remote fault diagnosis. This paper analyses comprehensively the present state of CNC remote fault diagnosis in last few years, as well as distributed artificial intelligent, Multi-Agent theory, expert system theory, the mechanical vibration signal processing theory and artificial immunity network theory. A new conception, CNC distributed intelligent fault diagnosis system based on Multi-Agent, is established in order to realize the accumulation of data and knowledge, resource share, and to realize coordinately fault diagnosis in different places during heterogeneous Network Environments. This system remedies at some extent the shortcoming of single version system, where domain knowledge is non adequate. In the Java language and environment, based on open JADE 3.5 (Java Agent Development Framework) development platform, followed FIPA (Foundation for Intelligent Physical Agent) standard, using Java, SQL Server, and other technologies to achieve a CNC machine tool prototype remote fault diagnosis system, this paper explores out a way from Multi-Agent theory to a practical technology.The main contents were as follows:1. Research on knowledge representation and acquisition of CNCRDSThe diagnosis level and quality of an intelligent diagnostic system depends largely on the number and quality of knowledge in the knowledge base. In CNCRDS it is very important to apply appropriate knowledge representation and acquisition because knowledge in large-scale knowledge bases own complex structures including both static knowledge and dynamic knowledge, and including both rule-based knowledge and principle-based knowledge from all kinds of manuals.This paper researched the knowledge representation and knowledge acquisition method in CNCRDS, analyzed the knowledge architecture of CNC fault diagnosis system, and constructed CNC fault diagnosis architectural model. According to the knowledge model including shallow knowledge model, deep knowledge model, and the combination of the two knowledge model, knowledge from different components were represented with different knowledge representation. As a very important interface between electrical and mechanical in CNC, PLC plays a very important role. The paper discussed the knowledge representation of PLC, presented a new method converting the PLC ladder diagram into rules used by inference, and established two diagnosis model based on PLC logic and time series respectively.2. Research on diagnosis strategy of CNCRDSThe selection and application of diagnostic strategy in an intelligent diagnostic system affect the efficiency of the diagnosis and decision accuracy. Diagnostic reasoning models include shallow reasoning model, deep reasoning model and mixed reasoning model in CNCRDS. In order to finish fault diagnosis and to locate fault accurately, different knowledge base and diagnostic strategy will be applied according to fault symptom during diagnosis. Shallow reasoning can only be used in case-based reasoning because it reason efficiently, but of limited capacity. Deep reasoning will take a very long time, but it can locate almost all the fault type because of reasoning based on the design specification. CNCRDS applied hybrid reasoning model combining shallow reasoning with deep reasoning. The hybrid reasoning model can overcome the weaknesses of a single model and play a more powerful force. Shallow model can be applied to simple fault diagnosis and define the searching border of deep reasoning models. In order to reduce the frequency of testing, heuristic rules can be used to find the search borders for deep reasoning models when the testing process is more difficult or testing need a long time. The use of hybrid reasoning can guarantee the efficiency of reasoning, but also guarantee the accuracy and validity of inference.3. Build the general frame of CNCRDS based on Multi-AgentAfter having analyzed the Agent and MAS theory and the decomposition characteristics of CNC remote fault diagnosis task, this paper first introduced the hierarchy and structure model of CNC remote fault diagnosis system based on Multi-Agent in our country. The formalization, the internal logical structure and functions of all kinds of agents in system was presented. The results showed that the level of this intelligent system was clear, the agents in system presented modular structure, both independent of each other and linked organically together, and finally the robustness and scalability of this system was very good.4. Research on communication mechanism, coordination strategy assessment algorithm between agents in CNCRDSIt is one of the important issues to communicate between agents in MAS because a MAS is loosely coupled system composed of multiple agents and communication is foundation to achieve interaction between agents and to constitute social organization. After analyzed the existing agent communication languages, this system used FIPA ACL as the communication language. This paper established XML-based communication framework.Competition among agents in MAS is an important feature. Agents owning strong competition have more opportunities. For fault diagnosis Agents in CNCRDS, the rate of accurate diagnosis and the time it taking to diagnose are two important indexes used to evaluate the effectiveness of a diagnosis agent. Therefore, two effectiveness measurement methods of an agent were discussed in this paper based on the diagnostic accuracy of the agent and time needed. Finally this paper presented a diagnostic task negotiation strategy based on above effectiveness measurement methods. The results showed that this intelligent system achieved efficient communication between agents. 5. Research and develop the Toolbox of mechanical faults signalsVibration signals can be characterized the typical characteristics of mechanical failure, this paper researched the collection, extraction and analysis of the symptoms and the relationship between symptoms and faults in CNC vibration signals. A signal-processing tool have been developed including the pretreatment of signals, the time-domain signal analysis, frequency analysis and joint time-frequency signal analysis methods. Experiments comparative was laid out. The characteristics of traditional and modern signal processing methods were analyzed, this paper focused on the empirical mode decomposition (EMD) method. Aiming at the characteristic that early prediction of mechanical faults is difficult in CNCRDS such as the spindle gearbox failures and bearing faults, the paper proposed an improved EMD algorithm applied to analyze the gear crack fault. The results showed that the proposed method was very effective to predict mechanical earlier failures.6. Research on the learning mechanism in CNCRDSThis paper researched two learning mechanism in order to accumulate and perfect knowledge in the knowledge base of CNCRDS system. Aimed at the knowledge model of CNCRDS system, this paper proposed a learning algorithm based on qualitative knowledge such as rules first; after analyzed the present state of artificial immune network, this paper proposed a fault pattern learning algorithm based on artificial immune network to overcome the shortcoming of mechanical faults. An example of bearing fault was applied to verify the proposed algorithm, the experimental result being compared with which of artificial neural network. The result showed that the proposed learning algorithm is efficient.7. Implement of CNCRDSAfter having researched the FIPA standard, Java technology, JADE Agent development platform and SQL Server database technology, we have designed and finished the prototype system of CNC remote fault diagnosis system based on Multi-Agent to guarantee the openness of this intelligent system making it possible to interact between agents in heterogeneous network, which explored out a way from Multi-Agent theory to a practical technology.