Faster Than Real-Time Prediction of Ocean Waves
Left: The spatiotemporal data of 3minutes of sea surface of 4500 m as the input to the model. Right) The prediction along with the ground truth for the next 3 minutes for the area.
Presented in American Physical Society Division of Fluids as Dynamic in 2019 with title: Phase-Resolved Ocean Waves Prediction via Machine Learning
The phase-resolved prediction of ocean waves suffers from two major challenges that have so far hindered solutions in real time or faster: (i) The reconstruction of the nonlinear and dispersive waves: how to calculate the state of the wavefield from measurable wave amplitude data, and (ii) Rapid solution: how to integrate equations in a timely manner. Here, we demonstrate that these two challenges can be overcome at once through advanced machine learning techniques based on spatiotemporal patches of the time history of wave height data in the domain. Specifically, as a benchmark here we consider equations that govern the evolution of weakly nonlinear surface gravity waves such as those propagating on the surface of the oceans. For the case of oceanic surface waves considered here, we demonstrate that the proposed methodology, while maintaining a high accuracy, can make phase-resolved predictions more than two orders of magnitude faster than numerically integrating governing equations.
Modeling collisions of arbitrary-shaped particles in simulations of particulate flows
Top: Collision of a spherical particle with a solid stationary wall at Re=164 similar to the experiment of [Gondret, P., M. Lance, and L. Petit, Bouncing motion of spherical particles in fluids. Physics of Fluids (1994-present), 2002. 14(2): p. 643-652]. Steel particle with Dp=3 mm collides with a glass wall in silicon oil RV10.
Left: Frictional collision of a falling brick. Sequence of particle motion for the collision of a 2×0.5 rectangular copper particle exposed to a unit freestream velocity and contact with a steel wall at Re=100 and Fr=0.102.
Right: Frictionless collision of falling brick. Sequence of particle motion for the collision of a 2×0.5 rectangular copper particle exposed to a unit freestream velocity and contact with a steel wall at Re=100 and Fr=0.102.
Simulations of dense suspensions containing arbitrary-shaped particles must deal with particle collision models. This study deals with the wet collisional interactions between particles in a carrier fluid, in the context of immersed boundary treatment of fully resolved particles. When collisions of non-spherical particles occur, several parameters that govern collisional mechanics are challenging to obtain: these include collision point, normal and tangential forces, degree of overlapping between particles, the local relative velocity and the local curvatures at the collision point. These challenges are overcome in this study by defining a field variable, i.e. a distance function that describes the particle on a regular Cartesian grid. A Simplified Spring Collision Method (SSCM) is proposed to model wet collision forces. The Stokes number at the collision point is computed from the particle's relative velocities and local curvatures and is used to calculate the restitution coefficient; damping is incorporated into the model by modifying the restitution coefficient when the rebound phase starts. The methods are validated against experiments and other benchmarks and shown to provide good results over a wide range of Stokes numbers. The capability of the method is also tested successfully for non-circular particle shapes.
Simulation of Blood as a Particulate Flow in The Hinge Gap Region of a Mechanical Heart Valve
For the direct numerical simulation of blood flow containing numerous rigid finite size particles, I implemented the smoothed profile method (SPM) with a new efficient parameter base-particles collisions method for blood cells. By studying a flow over a constricted region we showed that the platelets' (spherical object) activation is more likely to happen for the particles that pass from the middle of the upper bump region because those particles will have longer exposure time to the high shear flow behind the bump.
Vorticity generation on an oscillating flat plate airfoil at Re=10000, St = 0.3 and the reduced frequency of κ=πfc/U =π/2. We show that the updated smoothed profile method as a diffuse interface solver is able to model complex behaviors at high Re.