Thermal expansion of materials at high temperature exhibits a nonlinear change with temperature. Therefore, it is difficult to design the tolerances of the shaft and bearing, bearing and body, valve disc and metal seat of traditional marine triple eccentric butterfly valves and similar nested structures, which can jam or leak. Tolerances are generally calculated through simple empirical formulas combined with experiments, which is uneconomical and time-consuming. COMSOL is used to analyze the contact stress and friction torque corresponding to different tolerances when each nested structure is subjected to thermal expansion and structural coupling at different temperatures, to obtain the optimal tolerance. Quantile Regression Neural Network (QRNN) is used to analyze the optimal tolerances, which can meet the friction torque based on the minimization of the dimension chain without jam. The optimal tolerances are sorted into a dynamic tolerance diagram of the bearing, forming an efficient design methodology. Results from a high temperature switching test of 1000 tolerances under different working conditions demonstrate that the tolerances obtained by this method are reliable.
The triple eccentric (tritec) butterfly valves can realize flow regulation and fluid truncation at high temperature (up to 973.15 k), which is widely applied in offshore platforms and ships for high temperature steam transmission. The particular structure of the tritec butterfly valve reduces the friction drag of the valve disc and the valve seat. The second is that the pressure becomes tighter and tighter when the valve is closed, so as to achieve reliable sealing requirements. However, this is also the difficulty and key point of design and manufacture, and the reason that friction torque of each part are cannot be accurately calculated. Bearing is a vital structure, and its tolerance optimization design includes the establishment and solution of tolerance optimization model. However, the existing tolerance optimization model mostly considers the dimension chain and price of sealing structure and valve disc for numerical solution, such as genetic algorithm (Ding et al., 2009, Jiao and Li, 2013, Zhang et al., 2011).
