China
Africa
Explore chapters and articles related to this topic
Failure Detection Application in Autonomous Vehicles
Published in Diego Galar, Uday Kumar, Dammika Seneviratne, Robots, Drones, UAVs and UGVs for Operation and Maintenance, 2020
Diego Galar, Uday Kumar, Dammika Seneviratne
The pixel relationship of the objects is represented by a vector composed of six features. The first three features are Boolean values that represent the origin of the object (reference, LF1, LF2), and the next three features are the distance field values that show the distance of the corresponding pixel to the closest object. Three extra features representing the weight of each sensor are added to create a training vector (Table 9.1). The MF feature vector is trained offline with an SVM algorithm using positive vectors from a group of pixels that have been manually marked as detected objects and using negative vectors selected randomly from the other pixels (no objects) (Realpe, Vintimilla, & Vlaci, 2015b).
CFD simulation of loadings on circular duct in calm water and waves
Published in Ships and Offshore Structures, 2020
Andro Bakica, Nikola Vladimir, Inno Gatin, Hrvoje Jasak
In order to model the free surface waves in the gravitational field certain differences are made in the mathematical model. Solution is achieved with the Spectral Wave Implicit Navier Stoked Equations (SWENSE) approach described in Vukčević (2016). Major difference is in the modelling of the two phase flow. Instead of the indicator function the two non-mixing fluids are modelled as a Phase-Field (PF) which is bounded between and 1 with a hyperbolic tangent profile. The PF is written as a function of a signed distance field which is usually called the Level Set (LS) and is the shortest Euclidean distance from the interface. For more details regarding the model the reader is referred to Sun and Beckermann (2007). The PF profile is defined as:where ψ is the signed distance field and is the interface smearing parameter. The resulting LS transport equation without the curvature-driven motion is expressed as Sun and Beckermann (2007):where b is the numerical parameter to prevent excessive smearing of the interface. Further explanations can be found in Vukčević (2016).
On the Contribution of Wall Distance Fields to the Adjoint of a RANS Model
Published in International Journal of Computational Fluid Dynamics, 2022
Matteo Ugolotti, Paul Orkwis, Nathan Wukie
The computation of the distance field is a critical step in a CFD simulation involving turbulence modelling, and needs to be computed efficiently and with sufficient accuracy. Poor quality of the distance field can cause a slower non-linear convergence, divergence of the primal flow solver (Galbraith, Allmaras, and Darmofal 2018), or even incorrect results. The methods for obtaining the wall distance field d can be classified as direct search or equation-based. The direct search class, sometimes defined as ‘crude search methods’, encompasses a vast set of algorithms that span from the very simple brute-force minimum distance calculation between volume and surface nodes to more efficient and advanced techniques using alternating digital tree (ADT) (Boger 2001) or voxelized marching spheres (Roget and Sitaraman 2013). Among the equation-based approaches, the Fast-Marching Method (Sethian 1999) and the Fast Sweeping method (Zhao 2005) reformulate the wall proximity problem as the first time of arrival of a unit-speed wave propagating from the solid surface, which translates into solving the Eikonal equation (). Other PDE-based alternatives include solutions of the Hamilton-Jacobi equation (Tucker 2003), the Poisson equation (Tucker et al. 2005), and a mixed Hamilton-Jacobi-Poisson equation (Tucker 2011). A good description of these approaches can be found in Tucker et al. (2005) and Wukie (2018). Following Belyaev and Fayolle (2015), the p-Poisson equation in Equation (3) with boundary conditions defined in Equations (4) and (5) is used in this work and solved with the staged approach described by Wukie and Orkwis (2017). The content of the square bracket in Equation (3) is the diffusion coefficient and n in Equation (5) is the normal to the domain boundary. The solution to Equation (3) tends to the Euclidean distance field as as p tends to ∞. For this study, the Poisson parameter p is sequentially and gradually increased from 2 to 6 in three steps with each solution re-initialised with the previous one. In this way, the wall distance equation is well converged at each step and sufficiently accurate at the last step. Even though the computational advantage can really be appreciated for large grids, the method is simple to implement since existing numerical techniques and infrastructure of the numerical framework can be reutilised to solve this auxiliary problem. For the next section, it is convenient to define Equation (3) in residual form as follows. where is the distance field solution vector and are the node coordinates of the computational domain.