Abstract: To overcome the limits of present network fault diagnosis systems and the difficulty of different fault patterns, a distributed network fault diagnosis framework based on muiti-agent was proposed. In the framework, combined fuzzy neural network with rough sets theory, a new algorithm—RNFNN was presented, which greatly improves the performance of the system, with a nonlinear mapping realized by using topological structures and weights of the FNN. Also, the state-checking and authentication mechanism ensure the security of the agents and their communication. The framework is environment- independent, thus providing a general- purpose heterogeneous network fault diagnosis. The experiment result of this framework shows a good diagnostic ability.

Key words: communication technique , agent , fault diagnosis , fuzzy neural network , network performance