The optimization of natural frequency on the cross flow-induced vibration and heat transfer in a circular cylinder with LSTM deep learning model

Abstract

BackgroundIn this study, the fluid-induced vibrations of a long cylinder placed in the cross flow of a fluid with Reynolds number 325 have been investigated using ANSYS CFX 11. For this purpose, a middle section of the cylinder is considered in two-dimensional geometry and the damping and stiffness of the tube are modeled with appropriate springs and dampers. Since the frequency of the vortex shedding around the cylinder can exert an oscillating force on the cylinder, the effect of the applied force on the drag, lift and vibration path of the cylinder has been investigated. Also, by changing the natural frequency of the structure, different vibration modes and lock-in phenomenon have been investigated. MethodsVarious machine learning approaches have been used for predictive analysis where numerical data are generated to train intelligent algorithms for prediction errors. Different machine learning approaches such as long short-term memory (LSTM), multilayer perceptron (MLP) and support vector regression (SVR) are employed to optimize the prediction of drag and lift coefficient, vibration path of the cylinder and average Nusselt on the cylinder in terms of the natural frequency of the structure. This work proves that machine learning models are capable to predict the complex behavior of fluid flow in systems where flow-induced (FIV) vibration is coupled with heat transfer. FindingsIn the presented results, the Lissajous pattern in the form of number 8 was observed in the lock-in mode and the heat transfer in this mode shows an increase of about 13 percent compared to the fixed cylinder mode