Abstract: Conventional design of cylindrical interference fit treats the joint diameter and joint width as constants and doesnt consider the stress concentration caused by edge effect. For the solution of this question and better research of mechanical properties and more suitable design method, based on the advantages of finite element method and BP neural network, the finite element model of cylindrical interference fit is established using ABAQUS, which is used to analyze the properties of contact stress and its influencing factors such as joint diameter, joint width, outer diameter of the container, and interference. The model supplies plenty of training samples for the neural network. Based on that, the BP neural network calculation model for the maximum contact edge stress is established. Accounting for all kinds of factors treated as adjustable variables, a BP neural network dynamic adjusting algorithm for cylindrical interference fit design based on the maximum contact edge stress model which aims at the optimization of the maximum contact edge stress is proposed. The case analysis shows, the algorithm is more reasonable than conventional design method.

Key words: BP neural network, Finite element, Interference fit, Mechanical property