Abstract: At present,the traditional expert system technology have problems in reasoning process aspect:incomplete part information extraction,knowledge acquisition and reasoning ability weakly.Use hybrid reasoning mechanism based on neural network and rule instead of traditional expert system can solve the problems.First,use feature technology to extract part information,converting it to characteristic matrix,as input of neural network expert system.Then,according to the characteristic matrix search reasoning strategy,and classify the network as accuracy,shape and heat treatment three types based on shaft parts characters,using the momentum vector-adaptive BP algorithm training network.Finally,elaborate the design and implementation of the hybrid system technology reasoning.

Key words: reasoning process, expert system, neural network, feature matrix