Parallel Computational Framework Research Articles

Incorporation of adaptive compression into a GPU parallel computing framework for analyzing large-scale vessel trajectories

Automatic Identification System (AIS) offers a wealth of vessel navigation data, which underpins research in maritime data mining, situational awareness, and knowledge discovery within the realm of intelligent transportation systems. The flourishing marine industry has prompted AIS satellites and base stations to generate massive amounts of vessel trajectory data, escalating both data storage and calculation costs. The conventional Douglas-Peucker (DP) algorithm used for trajectory compression sets a uniform threshold, which hampers effective compression. Additionally, compressing and accelerating the computation of large datasets poses a significant challenge in real-world applications. To address these limitations, this paper aims to develop a new Graphics Processing Unit (GPU) parallel computing and compression framework that enables the acceleration of the optimal threshold calculation for each trajectory automatically in maritime big data mining. It achieves this by incorporating a new Adaptive DP with Speed and Course (ADPSC) algorithm, which utilizes the dynamic navigation characteristics of different vessels. It can effectively solve the associated computational time cost concern when using the ADPSC algorithm to compress vast trajectory datasets in the real world. Additionally, this paper proposes a novel evaluation metric for assessing compression efficacy based on the Dynamic Time Warping (DTW) method. Comprehensive experiments encompass vessel trajectory datasets from three representative research areas: Tianjin Port, Chengshan Jiao Promontory, and Caofeidian Port. The experimental results demonstrate that 1) the newly developed ADPSC method outperforms in terms of compression, and 2) the designed GPU parallel computing framework can significantly shorten the compression time for extensive datasets. The GPU-accelerated compression methodology not only minimizes storage and transmission costs for data from both manned and unmanned vessels but also enhances data processing speed, supporting real-time decision-making. From a theoretical perspective, it provides the key to the puzzle of realizing the real-time anti-collision of manned and unmanned ships, particularly in complex waters. It hence makes significant contributions to maritime safety in the autonomous shipping era.

Predictive Control for Unknown Dynamics With Observation Loss: A Temporal Game-Theoretic Approach

This paper is concerned with a model-free predictive control problem on systems with unknown dynamics. Different from existing predictive control, the predictive feedback strategy is designed to take control inputs to cover the infinite horizon when the system exists the observation loss. To predict future control inputs, a temporal game-theoretic approach is presented to model such a predictive control issue as an optimization problem and ensure the optimality of the system performance. Moreover, a predictive algebraic Riccati equation (PARE) is constructed to solve such an optimization problem. By leveraging offline datasets and the real-time data of state and input, a data-driven parallel computational framework is developed to iteratively solve the PARE. In this way, the prior knowledge of the systems is avoided and the computational complexity of the proposed algorithm is reduced. Finally, numerical and practical examples are presented to show the applicability of the proposed results.

Direct numerical simulation of electrokinetic transport phenomena in fluids: Variational multi-scale stabilization and octree-based mesh refinement

Computational modeling of charged species transport has enabled the analysis, design, and optimization of a diverse array of electrochemical and electrokinetic devices. These systems are represented by the Poisson-Nernst-Planck (PNP) equations coupled with the Navier-Stokes (NS) equation. Direct numerical simulation (DNS) to accurately capture the spatio-temporal variation of ion concentration and current flux remains challenging due to the (a) small critical dimension of the diffuse charge layer (DCL), (b) stiff coupling due to fast charge relaxation times, large advective effects, and steep gradients close to boundaries, and (c) complex geometries exhibited by electrochemical devices.In the current study, we address these challenges by presenting a direct numerical simulation framework that incorporates (a) a variational multiscale (VMS) treatment, (b) a block-iterative strategy in conjunction with semi-implicit (for NS) and implicit (for PNP) time integrators, and (c) octree based adaptive mesh refinement. The VMS formulation provides numerical stabilization critical for capturing the electro-convective flows often observed in engineered devices. The block-iterative strategy decouples the difficulty of non-linear coupling between the NS and PNP equations and allows the use of tailored numerical schemes separately for NS and PNP equations. The carefully designed second-order, hybrid implicit methods circumvent the harsh timestep requirements of explicit time steppers, thus enabling simulations over longer time horizons. Finally, the octree-based meshing allows efficient and targeted spatial resolution of the DCL. These features are incorporated into a massively parallel computational framework, enabling the simulation of realistic engineering electrochemical devices. The numerical framework is illustrated using several challenging canonical examples.

Convolutional neural networks for compressible turbulent flow reconstruction

This paper investigates deep learning methods in the framework of convolutional neural networks for reconstructing compressible turbulent flow fields. The aim is to develop methods capable of up-scaling coarse turbulent data into fine-resolution images. The method is based on a parallel computational framework that accepts five image sets of various resolutions, trained to correspond to the respective fine resolution. The network architecture mainly consists of convolutional layers, constructing an encoder/decoder network. Based on the U-Net scheme, three different implementations are presented, with residual and skip connections. The methods are implemented in a supersonic shock-boundary-layer interaction problem. The results suggest that simple networks perform better when trained on limited data, and this can be a practical and fast solution when dealing with turbulent flow data, where the computational burden is most of the time difficult to decrease. In such a way, a coarse simulation grid can be upscaled to a fine grid.

GEM5: DEVELOPMENT OF COMPUTER SYSTEM MODELS WITH PURPOSE TO EVALUATE THE EFFICIENCY OF PARALLEL CODE

In this paper we consider the possibility of using the gem5 computer system simulator to determine such metrics of parallel program code as its acceleration and the average workload of cpu cores on computer systems with shared memory. The working process in gem5 consists of three stages: building a model of a computer system, loading a software environment into the built model and executing a user program within this environment. The considered approach is used to evaluate the efficiency of a parallel program for modeling the transport of pollutants in the atmosphere. The program assumes both intensive work with memory and strong interaction between computational threads within the framework of parallel computing. For this purpose, two models of computer systems were created in the gem5 simulator: the simplest functional model of the “cores–RAM ” type and the detailed clock model of the “cores–three-level cache–RAM ” type. By running the program on both computer models and comparing obtained results with those observed on a real computer, it is shown that the computer system models created in gem5 are operable, correctly reproduce the functionality of real computer systems, and can be used to obtain fairly accurate estimates of such metrics as the acceleration of the program and the average workload of cpu cores. It should be noted that the accuracy of the estimates slightly depends on the number of simulated cores in computer models. For example, the acceleration accuracy obtained on the detailed clock model is 0.7 % on five cores and might increase up to 10 % on 32 cores.

A concurrent multiscale method based on smoothed molecular dynamics for large-scale parallel computation at finite temperature

A concurrent atomic-to-continuum multiscale method for finite-temperature simulation with large-scale parallel computation is developed. Seamless and stable coupling between molecular dynamics (MD) method and continuum-based material point method (MPM) is achieved with the aid of mesoscopic smoothed molecular dynamics (SMD) method. Novel techniques are proposed to filter spurious high-frequency reflections and bridge different thermal descriptions in atomistic and continuum models by adopting the efficient Markovian generalized Langevin equation (GLE) with newly designed drift matrix. Besides, an effective relaxation strategy is developed to fully relax the multiscale model to thermodynamic equilibrium state. The concurrent multiscale method can be effectively parallelized within the parallel computation framework of particle-grid dual discretization, and its efficiency when executing on large-scale computation clusters is analyzed in detail. The contact-sliding and particle impact examples are calculated with the proposed method to demonstrate its superiority as an effective numerical tool in the practical applications.

PRETTY: A parallel transgenerational learning-assisted evolutionary algorithm for computationally expensive multi-objective optimization

Many real-world optimization problems, called computationally expensive optimization problems, often require a time-consuming fitness evaluation through computer simulation or neural network training. A persistent challenge is to obtain reasonable solutions to these problems quickly. In this study, we design a parallel transgenerational learning-assisted evolutionary algorithm, called PRETTY, to accelerate the search process using three novel mechanisms. First, we fuse the transgenerational learning mechanism with metaheuristic algorithms to reduce unnecessary fitness evaluations. Then, a Hypervolume-Cauchy (HV-Cauchy) algorithm is proposed to balance population convergence and diversity adaptively. Finally, a parallel-computing framework is proposed to fuse the transgenerational learning mechanism and the HV-Cauchy algorithm into a multipopulation evolutionary algorithm. To verify the performance of the proposed algorithm, empirical experiments are conducted on nine benchmarks and four real-world problems, including feature selection, critical node detection, flocking, and neural network training. In terms of execution time, the experimental results demonstrate that PRETTY has significant advantages over several expensive multi-objective algorithms. Furthermore, PRETTY can maintain population diversity and fast convergence while accelerating the computation. In particular, in tackling real-world problems that require expensive fitness calculations, PRETTY can reduce the execution time by more than five times while ensuring high-quality solutions.

A robust computational framework for simulating the dynamics of large assemblies of highly-flexible fibers immersed in viscous flow

The dynamic response of flexible filaments immersed in viscous fluids is important in cell mechanics, as well as other biological and industrial processes. In this paper, we propose a parallel computational framework to simulate the fluid-structure interactions in large assemblies of highly-flexible filaments immersed in a viscous fluid. We model the deformation of each filament in 3D with a C1 geometrically-exact large-deformation finite-element beam formulation and we describe the hydrodynamic interactions by a boundary element discretization of the Stokeslet model. We incorporate a contact algorithm that prevents fiber interpenetration and avoids previously reported numerical instabilities in the flow, thus providing the ability to describe the complex evolution of large clouds of fibers over long time spans. In order to support the required long-term integration, we use implicit integration of the solid-fluid-contact coupling. We address the challenges associated with the solution of the large and dense linear system for the hydrodynamic interactions by taking advantage of the massive parallelization offered by Graphic Processing Units (GPUs), which we test up to 1000 fibers and 45000 degrees of freedom.We validate the framework against the well-established response of the sedimentation of a single fiber under gravity in the low to moderate flexibility range. We then reproduce previous results and provide additional insights in the large to extreme flexibility range. Finally, we apply the framework to the analysis of the sedimentation of large clouds of filaments under gravity, as a function of fiber flexibility. Owing to the long time spans afforded by our computational framework, our simulations reproduce the breakup response observed experimentally in the lower flexibility range and provide new insights into the breakup of the initial clouds in the higher flexibility range.

A finite element level set method based on adaptive octree meshes for thermal free‐surface flows

AbstractWe develop a parallel finite element computational framework based on the level set method and adaptive octree meshes for free‐surface flows. The discretized governing formulations are stabilized using a residual‐based variational multiscale method. We redistance the level set field and restore mass conservation after each time step. The octree mesh is adaptively refined according to the interface location, and the load is rebalanced across processes. The adaptive octree‐based level set method is verified and validated using a canonical problem of dam break without obstacle, and a mesh convergence study is carried out in this problem. The surface tension is further considered via a continuum surface force model in the method and validated using a problem of single bubble rising in ambient liquid. A parallel scaling analysis of this method is also performed for the bubble rising problem. The energy equation and Marangoni effect are finally considered in the method and validated using a problem of thermocapillary droplet migration with and without gravity. This work illustrates the ability of the framework to accurately model and predict free‐surface flows under various scenarios.

High-speed liquid crystal display simulation using parallel reservoir computing approach

The performance of liquid crystal displays (LCDs) is expected to be improved further with the emergence of their new applications. Numerical simulations such as the finite differential method (FDM) and the finite element method are useful in optimum design. However, they take a long time because dynamical systems in LCDs are nonlinear multiphysics composed of electromagnetism, fluid dynamics, and elastic mechanics. A machine learning method is one of the solutions to reduce computational cost. In this paper, we have extended the parallel reservoir computing framework and applied it to LCD simulation. We have discussed how to implement each natural feature of liquid crystal cells, namely, non-autonomy, multiphysics and long-range orientational order, into the framework of parallel reservoir computing. Sufficient higher accuracy was obtained with several display patterns and driving frequencies at computational speeds more than 100 times higher than FDM.

Semi-Definite Relaxation-Based ADMM for Cooperative Planning and Control of Connected Autonomous Vehicles

This paper investigates the cooperative planning and control problem for multiple connected autonomous vehicles (CAVs) in different scenarios. In the existing literature, most of the methods suffer from significant problems in computational efficiency. Furthermore, as the optimization problem is nonlinear and nonconvex, it typically poses great difficulty in determining the optimal solution. To address this issue, this work proposes a novel and completely parallel computation framework by leveraging the alternating direction method of multipliers (ADMM). The nonlinear and nonconvex optimization problem in the autonomous driving problem can be divided into two manageable sub-problems; and the resulting sub-problems can be solved by using effective optimization methods in a parallel framework. Here, the differential dynamic programming (DDP) algorithm is capable of addressing the nonlinearity of the system dynamics rather effectively; and the nonconvex coupling constraints with small dimensions can be resolved by invoking the notion of semi-definite relaxation (SDR), which can also be solved in a very short time. Due to the parallel computation and efficient relaxation of nonconvex constraints, our proposed approach effectively realizes real-time implementation; and thus extra assurance of driving safety is provided. In addition, two transportation scenarios for multiple CAVs are used to illustrate the effectiveness and efficiency of the proposed method.

Development of a Single-Phase, Transient, Subchannel Code, within the MOOSE Multi-Physics Computational Framework

Subchannel codes have been widely used for thermal-hydraulics analyses in nuclear reactors. This paper details the development of a novel subchannel code within the Idaho National Laboratory’s (INL) Multi-physics Object Oriented Simulation Environment (MOOSE). MOOSE is a parallel computational framework targeted at the solution of systems of coupled, nonlinear partial differential equations, that often arise in the simulation of nuclear processes. As such, it includes codes/modules able to solve the multiple linear and nonlinear physics that describe a nuclear reactor, under normal operation conditions or accidents. This includes thermal-hydraulics, fuel performance, and neutronics codes, between others. A MOOSE-based subchannel code is a new addition to the fleet of INL-developed codes, based on the MOOSE framework. In this work, we present the derivation of the subchannel equations for a single-phase fluid, we proceed with the description of the algorithm that is used to solve these equations and describe how this algorithm was implemented within MOOSE. We also present how this code can be coupled to the BISON fuel performance code. Next, we verify the friction model and the turbulent mixing model. We calibrate the turbulent modeling parameters for momentum mixing and enthalpy mixing, CT,β. We validate the code using experimental results and last demonstrate the coupling capabilities using a simple example.

Development of a GPU parallel computational framework for impact debonding of coating–substrate interfaces

Single-particle tests are normally used to evaluate the impact resistance performance of automotive coatings. Impact-induced debonding is one of the major failure patterns for coating–substrate structures of a vehicle. Currently, it remains a challenging task to accurately and efficiently simulate progressive debonding of automotive coating–substrate interfaces under single particle impact. The main purpose of this work is to develop a graphics processing unit (GPU)-based parallel computational framework to achieve this end. An efficient coating finite element model in three dimensions is established, where solid elements with good aspect ratios and solid-shell elements are respectively used to discretize the impact contact region and the rest region of a coating. An intrinsic cohesive zone model coupled with a mortar-based contact algorithm is used to accurately describe the progressive debonding behavior of coating–substrate interfaces. The computational efficiency of the developed method is dramatically enhanced by recourse to the GPU parallel computing technique. Three benchmark tests are carried out to validate the effectiveness and efficiency of the developed computational framework. Finally, the novel computational framework is successfully applied to the debonding analysis of an organic coating bonded with an aluminum substrate under single-particle impact. Results show that the GPU parallel computational framework can achieve a total speedup of 136.5, which provides a powerful tool for coating debonding analysis.

Modelling saline intrusion using dynamic mesh optimization with parallel processing

Saline intrusion (SI) in coastal aquifers is a global problem with the potential to contaminate groundwater used by over a billion people. Numerical modelling of SI in coastal aquifers is a key tool for risk assessment, aquifer management and resource regulation, but is extremely challenging because the mixing zone across the saline front is often very narrow, extending over metres or 10’s metres, yet the saline front itself may extend laterally over a large (i.e. many km) three-dimensional (3D) domain. Moreover, the aquifer may be highly heterogeneous, further complicating the movement and geometry of the front. We test here the use of dynamic mesh optimization (DMO) in a parallel computational framework to simulate SI with higher accuracy and lower computational cost compared to fixed-mesh approaches. The framework uses a double control-volume-finite-element (DCVFE) method and is implemented in the open-source Imperial College Finite Element Reservoir SimulaTor (IC-FERST), but could be implemented in other FE-based simulators. We confirm accuracy and convergence using test cases based on the classic ’Henry’ SI problem, demonstrating that solutions obtained using DMO require significantly fewer elements and therefore have much lower computational cost compared to equivalent fixed mesh solutions. We apply the framework to a realistic 3D case study simulating saline intrusion in a heterogeneous chalk aquifer, demonstrating simulation speed-up in excess of 120×. We suggest that parallelized DMO offers significant advantages over existing methods to simulate SI.

Less can be more: Insights on the role of electrode microstructure in redox flow batteries from two-dimensional direct numerical simulations

Understanding how to structure a porous electrode to facilitate fluid, mass, and charge transport is key to enhancing the performance of electrochemical devices, such as fuel cells, electrolyzers, and redox flow batteries (RFBs). Using a parallel computational framework, direct numerical simulations are carried out on idealized porous electrode microstructures for RFBs. Strategies to improve an electrode design starting from a regular lattice are explored. By introducing vacancies in the ordered arrangement, it is possible to achieve higher voltage efficiency at a given current density, thanks to improved mixing of reactive species, despite reducing the total reactive surface. Careful engineering of the location of vacancies, resulting in a density gradient, outperforms disordered configurations. Our simulation framework is a new tool to explore transport phenomena in RFBs, and our findings suggest new ways to design performant electrodes.

Optimizing Urban LiDAR Flight Path Planning Using a Genetic Algorithm and a Dual Parallel Computing Framework

This paper introduces a genetic algorithm (GA) and a beam tracing algorithm incorporated within a dual parallel computing framework to optimize urban aerial laser scanning (ALS) missions to maximize vertical façade data capture, as needed for many three-dimensional reconstruction and modeling workflows. The optimization employs a low-density point cloud from the site of interest as a spatial representation of the urban scene. The GA is suitable for LiDAR flight path optimization due to its capability of handling open-ended problems that have many solutions. However, GAs require evaluating a very large number of candidates. The use of an initial point cloud allows realistic modeling of the urban environment in the optimization at the cost of high data input volumes. To cope with the computational and data demands, a dual parallel computing framework was devised. The parallel computing framework consists of two layers of parallelization. In the upper layer, multiple evaluators work in parallel and in conjunction with a main multi-threading GA optimizer to perform GA operations and evaluate the flight paths. In the lower layer, to evaluate assigned flight paths, each evaluator distributes its data and computation to multiple executors, which can reside on multiple physical nodes of a distributed-memory computing cluster. In addition to parallelism, the data partitioning on the lower layer allows out-of-core computation. Namely, data partitions are efficiently transferred between disks and memory so that only relevant subsets of data are kept in the main memory. The objective of the proposed method is threefold: (1) search for flight paths that yield the highest numbers of vertical points, (2) create a means to explicitly consider the detailed spatial configuration of urban environments, and (3) assure that the proposed optimization strategy is fast and can scale to large problem sizes. Multiple experiments were conducted and demonstrated the success of the proposed method. Converged results were achieved after dozens of generations within two hours. Two flight paths identified by the GA as the most and the least optimal candidates were deployed in real flight missions. The optimal flight path captured 16% more vertical points than the least optimal one, slightly higher than the 13% predicted. Both layers of parallelization were efficient: 13.1/16 for the lower layer and 3.2/4 for the upper layer. The two complementary layers of parallelization allowed flexible and efficient use of distributed computing resources to reduce the runtime. The scalability of the proposed approach was successfully demonstrated up to a data size of 460 million points. The optimization results were realistic and aligned well with the test flight results.

Finite Element Method for Interconnect Model Extraction

A 3D finite element method full-wave solver is introduced. It has been successfully applied for the interconnect model extraction for printed circuit boards, packages, and integrated circuits. A massively parallel computational framework is designed to address the issue of capacity and scalability due to the increase in system size and complexity. In addition, the method of field-circuit co-simulation is discussed, and it is critical to real-world signal integrity and power integrity applications.

An SPH framework for fluid\u2013solid and contact interaction problems including thermo-mechanical coupling and reversible phase transitions

The present work proposes an approach for fluid–solid and contact interaction problems including thermo-mechanical coupling and reversible phase transitions. The solid field is assumed to consist of several arbitrarily-shaped, undeformable but mobile rigid bodies, that are evolved in time individually and allowed to get into mechanical contact with each other. The fluid field generally consists of multiple liquid or gas phases. All fields are spatially discretized using the method of smoothed particle hydrodynamics (SPH). This approach is especially suitable in the context of continually changing interface topologies and dynamic phase transitions without the need for additional methodological and computational effort for interface tracking as compared to mesh- or grid-based methods. Proposing a concept for the parallelization of the computational framework, in particular concerning a computationally efficient evaluation of rigid body motion, is an essential part of this work. Finally, the accuracy and robustness of the proposed framework is demonstrated by several numerical examples in two and three dimensions, involving multiple rigid bodies, two-phase flow, and reversible phase transitions, with a focus on two potential application scenarios in the fields of engineering and biomechanics: powder bed fusion additive manufacturing (PBFAM) and disintegration of food boluses in the human stomach. The efficiency of the parallel computational framework is demonstrated by a strong scaling analysis.

TopADD: a 2D/3D integrated topology optimization parallel-computing framework for arbitrary design domains

TopADD: a 2D/3D integrated topology optimization parallel-computing framework for arbitrary design domains

A consistent and parallelized height function based scheme for applying contact angle to 3D volume-of-fluid simulations

A height function based numerical scheme has been developed to apply contact angle in the parallel computational framework of 3D volume-of-fluid method. Given that most of the reported height function methods are implemented in 2D, the differences between the implementations of 2D and 3D height function methods are first focused. Then, a novel scheme is proposed to preserve the accuracy and consistency of height function for the calculation of interface curvature near the contact line. Finally, the extension to parallel computation is presented in details because it is not a trivial task as it appears, e.g., the information transmission among the cells neighboring two intersecting boundaries requires careful treatment. Numerical tests reveal that with the proposed scheme, accurate estimations of interface curvature and normal vector are obtained for a spherical interface, and such a numerical strategy allows the spurious currents to decrease to machine accuracy for a 3D stationary droplet. Moreover, by simulating a droplet spreading on a solid wall with 3D Cartesian grid, either from a static state driven by capillary force or with an initial impact velocity, the numerical results verify that the contact line can maintain an axi-symmetric shape during the dynamic spreading.