Horizontal Pressure Vessels are supported by two saddle supports near the ends. There is a local stress concentration in the vessel near the saddle horn of the saddle support. The parameters of a saddle support that can be changed are the angle of contact of the saddle, the width of saddle, the width of wear plate, and extension of the wear plate over the saddle horn. These parameters directly govern the values and location of maximum stress in the vessel. The aim of this study is to perform FEA analysis and find the effects of different configurations of the saddle and its effect on the maximum stress in the vessel and thus find the most optimum configuration of the saddle to pass the ASME requirements while keeping the material costs at the minimum.
Stress distribution in a horizontal pressure vessel.
Stiffening rings are very commonly used in horizontal pressure vessels where external pressure is applied to give protection against circumferential buckling but, also in the case of thin and long vessels, where internal pressure is applied, use of stiffening rings become very crucial to avoid stress concentrations at certain parts of the vessel and also give extra stiffness to the shell so that it can resist the load of the bending due to the dead weight and the fluid load. In this study, we performed FE analysis on different configurations of stiffening rings by varying the cross-sectional shapes, number and locations of stiffening rings and studied how the different configurations affect the maximum stress concentration in the pressure vessel. This study is important because knowing the right configuration of stiffening rings will lead to a more optimized way of using stiffening rings leading to savings in material cost.
ANSYS Mechanical was used to perform the 3D FEA Simulations for this project. In the beginning of the FEA study we performed a mesh refinement study on a benchmark simulation to get the mesh size so that the results are mesh indifferent. The final node count that we selected to go with was ~2M elements in the entire domain. We also went up to ~4M nodes, but the results became mesh indifferent by the 2M node point.