<p>High-speed winding machine plays a role in stretching and winding the fiber by virtue of the rotation movement of spindle, which directly affects the forming quality of silk cake. Improving and maintaining assembly precision is the key to ensure long-term stable operation of the equipment. Due to the large number of spindle parts and complex connected relation, the assembly precision is difficult to predict and control, let alone its high retainability that promote the winding machine to maintain good dynamic characteristics. Tolerance modeling is an important aspect of ensuring assembly precision. Existing methods typically separate the static precision index from the dynamic characteristics of the assembly, and simplify the matching conditions between parts. This discrepancy between tolerance modeling calculations and engineering reality is considerable. This paper aims to establish a tolerance modeling method that considers precision retainability by analyzing dynamic characteristics. Firstly, assembly structure and requirements of the spindle were analyzed, and the influence of precision index on vibration response was solved by deriving Lagrange equations for dissipative systems; secondly, tolerance modeling based on improved Jacobian-Torsor theory is carried out according to the obtained assembly precision indexes; finally, a multi-objective tolerance optimization was performed by considering cost and quality loss, and a better tolerance allocation scheme was presented using NSGA-II solution. Taking a certain type of winding machine as an example, the results show that the optimized tolerance scheme reduces the cost by 36% and the quality loss by 63%, demonstrating the proposed method is beneficial to the precision retainability improvement.</p>