Research On Intelligent Fault Diagnosis Of Gas-Monitoring System

As being a complex and dynamic system, gas-monitoring system demands very high safety. Because the limitations of a single fault diagnosis method or expert knowledge, a single fault diagnosis method cannot satisfy the requirements of real-time and accuracy. Therefore, we need much experience from several experts to draw a correct diagnosis conclusion. Moreover, it's required that the solution of fault diagnosis system have characteristics of integration, intelligence, automation and network. But the traditional diagnosis methods cannot have the above characteristics. Therefore, new diagnosis theory and diagnosis structure are needed to carry out fault diagnosis. Agent is a kind of intelligent unit with autonomous behavior. It can deal with multidimensional information in real-time through the mutual communication and cooperation between agents, which makes different methods can be adopted to solve problems under the same environment and makes the adaptability of the diagnosis system. In order to realize the intelligent diagnosis for complex gas-monitoring system, a new fault diagnosis system based on multi-agent is constructed by using the autonomous property of agent in the dissertation.Firstly, some factors of influencing the reliability of gas-monitoring system and the present development of fault diagnosis techniques for electromechanical equipments in coal mine are analyzed. Then the mission and contents of the fault diagnosis for gas-monitoring system are put forward based on the analysis. Aiming at the fault mode of gas sensor and the system complexity, a multi-agent fault diagnosis planning and strategy are proposed.From the point of view of part diagnosis, fault diagnosis Agent1 for gas sensor is built up. A high precision RBF network approximated is constructed by using multi-sensor data fusion technique to fuse a great deal of data provided by related sensors, such as wind velocity sensor, temperature sensor, CO₂ sensor, etc. in a gas-monitoring system .The approximating value of RBF network is then viewed as the reference basis of monitoring the status of gas sensor, and so as to realize the fault diagnosis for gas sensors effectively. The proposed method can diagnose drift fault with higher speed and mutation fault accurately. From the point of view of system diagnosis, the graph theory based on fault tree analysis (FTA) are studied, the FTA graph theory model of underground sub-station in gas-monitoring system are brought forward, the FTA of the commonly running faults in the system are constructed, and the intelligent fault diagnosis system represented by mixed knowledge of frame and rule is designed. Thus, the FTA-based Agent2 for gas-monitoring system is built up.The case-based reasoning (CBR) Agent3 for gas-monitoring system diagnosis is built up and the detailed design process of the method are given. An intelligent clustering method based on adaptive resonance theory (ART) neural net model is proposed to carry out the clustering of cases. In addition, genetic algorithm (GA) is used to optimize the characteristic weights of cases.The multi-Agent (MAS) model of fault diagnosis system is researched and an abstract MAS model is brought forward, as well as how to use the model to realize fault diagnosis in gas-monitoring system is studied. According to the characteristics of the solution of MAS, the probability assignment functions of DS evidence theory are restricted and specified and the complete steps of multi-agent cooperative diagnosis based on DS evidence theory are described in detail. In order to verify the validity of the proposed method, three examples including the same conclusion, the partial conflicting conclusion and complete conflicting one are analyzed comprehensively in the dissertation. Finally, the proposed multi-agent fault diagnosis scheme based on the above three diagnosis methods is applied to the design and the realization of the practical gas-monitoring system.