Multiphase Fluid-Structure Interaction Algorithm And Complex Interface Effects In Droplet Dynamics

Interaction between the droplet and solid object contains various flow phenomena.An in-depth understanding of the complex flow mechanism is of great significance for understanding nature and guiding the development of engineering technology.In this thesis,we developed a three-dimensional multiphase fluid-structure interaction algorithm,proposed an interfacial flow prediction method based on deep learning,then explored the effect of complex geometry and complex microstructure of solid on droplet dynamics.The results and conclusions are briefly given as follows:(1)A three-dimensional diffuse interface immersed boundary method is developed,which can accurately simulate multiphase fluid-structure interaction with dynamic wetting and contact angle hysteresis.The interface tracking and fluid-structure interaction are achieved by the three-phase diffuse interface model and the threedimensional immersed boundary method,respectively.The solid boundary is represented by a series of Lagrangian patches.The motion of rigid solid objects with arbitrary complex geometry is allowed to have six degrees of freedom in multiphase flow.A simple model of contact angle hysteresis is used to ensure pinned contact lines if the local contact angle is within the window of contact angle hysteresis.The hybrid capillary force model is used to accurately calculate the capillary force and corresponding torque on the solid.The program which is optimized based on parallel frameworks such as MPI and CUDA achieves a speedup of about 20 times.The accuracy and robustness of the method are verified by a series of numerical cases.(2)The impact of droplets on a movable cylindrical solid object is studied using the three-dimensional numerical method.On impact at low velocity,the droplet exhibits two rebound modes:kettle-like rebound and wing-like rebound.Increasing the impact velocity or the mass ratio of the cylinder to the droplet λm induces the droplet to convert from the kettle-like rebound to the wing-like rebound mode.Two scaling laws are proposed to predict the spreading length of the droplet in the azimuthal and axial directions,respectively,which are in good agreement with the numerical simulation results.On this basis,it is found that the modal transition boundary of the two rebound modes is We1/2～1+1/λm.The time scale of droplet spreading and retraction is theoretically analyzed,and it is pointed out that the difference in contact time between the two rebound modes is caused by the difference in retraction time.(3)The directional movement of droplets on a wetted anisotropic surface driven by horizontal vibration is studied using the three-dimensional numerical method.It is found that the directional motion is determined by the contact angle,vibration,physical properties and volume of the droplet.The mechanism of droplet directional motion is explained.When the difference in contact angle hysteresis is large,the droplet shifts towards the direction with smaller contact angle hysteresis,dominated by asymmetric capillary action.When the difference in contact angle hysteresis is small,the droplet shifts towards the direction with a larger average contact angle under the influence of asymmetric viscosity.The scaling law of droplets’ directional motion for the two mechanisms is proposed respectively.The critical condition of the two mechanisms is theoretically analyzed,which is in good comparison with the numerical simulation results.(4)A deep learning-based interfacial flow prediction method is developed.A neural network model with embedded fluid governing equations is constructed so that the predicted results of the model satisfy the conservation laws of fluid mechanics.The training and validation dataset is obtained from the high-precision direct numerical simulations.Once the model is trained,features can be extracted from the motion of the fluid interface,and the flow field information such as velocity can be automatically reconstructed by the model.With the characteristics of droplet dynamics in consideration,two models are designed,that is SNN and TSNN models.Through three typical multiphase interfacial flows,the accuracy and robustness of the method are validated.We also explored the effect of the number of sampling points on the model and the result is that the model performance was not sensitive to the number of sampling points.