In this paper, a structural design of the knowledge network for the identification and classification fault diagnosis model is introduced together with the detailed reasoning algorithm and fault diagnosis algorithm. Reasoning with uncertainty is realized through the message propagation among the nodes, which makes it have the characteristic of a distributed system. In addition, the machine learning algorithm for the judgement function is also presented. This algorithm adopts the package wrapping algorithm to select the border points sets, which makes the speed of curve approach more quickly.