SEARCH · Engineering Papers
Results for “Network simulation”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Nonlinear thermodynamic computing out of equilibrium
We present the design for a thermodynamic computer that can perform arbitrary nonlinear calculations in or out of equilibrium. Simple thermodynamic circuits, fluctuating degrees of freedom in contact with a thermal bath and confined by a quartic potential, display an activity that is a nonlinear function of their input. Such circuits can therefore be regarded as thermodynamic neurons, and can serve as the building blocks of networked structures that act as thermodynamic neural networks, universal function approximators whose operation is powered by thermal fluctuations. We simulate a digital model of a thermodynamic neural network, and show that its parameters can be adjusted by genetic algorithm to perform nonlinear calculations at specified observation times, regardless of whether the system has attained thermal equilibrium. This work expands the field of thermodynamic computing beyond the regime of thermal equilibrium, enabling fully nonlinear computations, analogous to those performed by classical neural networks, at specified observation times.
Performance analysis and acceleration of explicit integration for large kinetic networks using batched GPU computations
Abstract—We demonstrate the systematic implementation of recently-developed fast explicit kinetic integration algorithms that solve efficiently N coupled ordinary differential equations (subject to initial conditions) on modern GPUs. We take representative test cases (Type Ia supernova explosions) and demonstrate two or more orders of magnitude increase in efficiency for solving such systems (of realistic thermonuclear networks coupled to fluid dynamics). This implies that important coupled, multiphysics problems in various scientific and technical disciplines that were intractable, or could be simulated only with highly schematic kinetic networks, are now computationally feasible. As examples of such applications we present the computational techniques developed for our ongoing deployment of these new methods on modern GPU accelerators. We show that similarly to many other scientific applications, ranging from national security to medical advances, the computation can be split into many independent computational tasks, each of relatively small-size. As the size of each individual task does not provide sufficient parallelism for the underlying hardware, especially for accelerators, these tasks must be computed concurrently as a single routine, that we call batched routine, in order to saturate the hardware with enough work.
On the Need for Communication for Voltage Regulation of Power Distribution Grids
We consider the problem of regulating the voltage profile of a power distribution grid by controlling the reactive power injection of distributed microgenerators. We define a very general class of purely local feedback controllers in which reactive power injection is adjusted based on the local voltage measurements. This class includes most of the solutions proposed in the literature and in the latest grid codes. We show that these strategies do not guarantee the desired regulation, as each of them can have an equilibrium that is not feasible with respect to the desired voltage and power constraints. We then propose a networked feedback law to show that, by adding short-range communication between microgenerators, it is possible to design control strategies that provably converge to the feasible set. This fundamental performance gap between local and networked strategies is finally illustrated via simulations.
CMLM (Co-Optimized Machine-Learned Manifolds) [SWR-23-41]
Co-optimized Machine-Learned Manifolds (CMLM) is a data-driven approach for developing reduced-order manifold models for high-dimensional chemically reacting systems. It involves a specially designed neural network, the training of which simultaneously optimizes linear combinations of species that define the manifold, nonlinear mapping to outputs of interest such as reaction rates, and (optionally) subfilter closure for large eddy simulation. This software package provides an implementation of the CMLM approach in Python using the PyTorch machine learning library. A few example cases are included, showing how the tool can be applied to different types of data from 0D and 1D reacting simulations performed using Cantera. The neural networks can be saved in a format that is readable by the Pele suite of combustion solvers for use in reacting computational fluid dynamics simulations. This software repository contains several python scripts to perform various tasks associated with the Co-optimized Machine Learned Manifolds (CMLM) model, which is described in Perry, Henry de Frahan, and Yellapantula, CNF, 2022 (https://doi.org/10.1016/j.combustflame.2022.112286). This includes not only the code that defines the CMLM model, but also scripts to generate suitable training data, scripts to pre-process the data, scripts to train the CMLM model, and scripts to plot the output, as well as various other helper files. The scripts depend on several commonly used python libraries for data analysis and chemical reaction computations. The trained models that result from this tool are designed to work with the an interface being implemented in the Pele suite of reacting flow solvers (https://github.com/AMReX-Combustion).
Package Equivalent Reactor Networks as Reduced Order Models for Use with CAPE-OPEN Compliant Simulation
Engineering simulations of coal gasifiers are typically performed using computational fluid dynamics (CFD) software, where a 3-D representation of the gasifier equipment is used to model the fluid flow in the gasifier and source terms from the coal gasification process are captured using discrete-phase model source terms. Simulations using this approach can be very time consuming, making it difficult to imbed such models into overall system simulations for plant design and optimization. For such system-level designs, process flowsheet software is typically used, such as Aspen Plus®, where each component where each component is modeled using a reduced-order model. For advanced power-generation systems, such as integrated gasifier/gas-turbine combined-cycle systems (IGCC), the critical components determining overall process efficiency and emissions are usually the gasifier and combustor. Providing more accurate and more computationally efficient reduced-order models for these components, then, enables much more effective plant-level design optimization and design for control.
Investigating the Impact of Connected Vehicle Market Share on the Performance of Reinforcement-Learning Based Traffic Signal Control
We aim to understand and explore the performance of reinforcement learning based signal control algorithms in a mixed environment with less than 100% market share of connected and automated vehicles (CAVs). Within a simulation environment, we have considered partial connectivity—less than 100% market share of CAVs—in the network and investigated the impact on the performance of the signal control algorithm. Two test networks including a four-intersection arterial in Lankershim Boulevard, CA, and a portion of downtown Springfield, IL, with 20 intersections. The first network is calibrated in the micro-simulator PTV Vissim with the US DOT provided NGSIM datasets. The results provide insights regarding the impact of the connectivity and sensing technologies on the practical implementation of traffic signal control algorithms that leverage the data sharing capability of a connected environment. For scenarios with 40% or more market share of CAVs, we observed improvement in the performance metrics—travel time, queue time, and energy consumption—with the increase in market share. Results from our experiments do not indicate any clear trend when the networks have low (less than 40%) market share of CAVs. The higher standard deviations as obtained from the statistical analyses of the performance metrics at low market share may indicate the instability of the RL controller arising from the partial (if not zero) observability of the traffic states. Further, we have conducted simplified scenario analyses to explore the impact of the market share of battery electric vehicles (BEVs) on energy consumption due to the regenerative braking feature. Our results and findings will be the foundation for the future reinforcement learning based control algorithm development that accounts for partial connectivity—less than 100% CAV market share, and the presence of BEVs in a network of connected and automated signalized intersections.
MODELING THE ATOMIC-TO-MOLECULAR TRANSITION AND CHEMICAL DISTRIBUTIONS OF TURBULENT STAR-FORMING CLOUDS
We use 3D-PDR, a three-dimensional astrochemistry code for modeling photodissociation regions (PDRs), to post-process hydrodynamic simulations of turbulent, star-forming clouds. We focus on the transition from atomic to molecular gas, with specific attention to the formation and distribution of H, C{sup +}, C, H{sub 2}, and CO. First, we demonstrate that the details of the cloud chemistry and our conclusions are insensitive to the simulation spatial resolution, to the resolution at the cloud edge, and to the ray angular resolution. We then investigate the effect of geometry and simulation parameters on chemical abundances and find weak dependence on cloud morphology as dictated by gravity and turbulent Mach number. For a uniform external radiation field, we find similar distributions to those derived using a one-dimensional PDR code. However, we demonstrate that a three-dimensional treatment is necessary for a spatially varying external field, and we caution against using one-dimensional treatments for non-symmetric problems. We compare our results with the work of Glover et al., who self-consistently followed the time evolution of molecule formation in hydrodynamic simulations using a reduced chemical network. In general, we find good agreement with this in situ approach for C and CO abundances. However, the temperature and H{sub 2} abundances are discrepant in the boundary regions (A{sub v} {<=} 5), which is due to the different number of rays used by the two approaches.
Thermal Drawdown-Induced Flow Channeling in Fractured Geothermal Reservoirs
In this paper, we investigate the flow-channeling phenomenon caused by thermal drawdown in fractured geothermal reservoirs. A discrete fracture network-based, fully coupled thermal–hydrological–mechanical simulator is used to study the interactions between fluid flow, temperature change, and the associated rock deformation. The responses of a number of randomly generated 2D fracture networks that represent a variety of reservoir characteristics are simulated with various injection-production well distances. We find that flow channeling, namely flow concentration in cooled zones, is the inevitable fate of all the scenarios evaluated. We also identify a secondary geomechanical mechanism caused by the anisotropy in thermal stress that counteracts the primary mechanism of flow channeling. This new mechanism tends, to some extent, to result in a more diffuse flow distribution, although it is generally not strong enough to completely reverse flow channeling. We find that fracture intensity substantially affects the overall hydraulic impedance of the reservoir but increasing fracture intensity generally does not improve heat production performance. Finally, increasing the injection-production well separation appears to be an effective means to prolong the production life of a reservoir.
Towards Improved Real-Time Observability of Behind-Meter PhotoVoltaic Systems: A Data-Driven Approach
The stochasticity in energy supply introduced by increased PV penetration in grid distribution systems can lead to frequent and sudden supply-demand imbalances, thereby causing voltage and frequency fluctuations. Enhanced observability, in real-time, of grid attributes such as real and reactive power flows, power injections/consumption and voltages is becoming increasingly essential to ensure smooth grid operation under these new large-scale PV integration scenarios. Two significant challenges prevent grid operators from achieving this objective: 1) Computational complexity of quasi-static timeseries simulations, and 2) Partial observability of behind-meter grid assets. In this work, we take a step towards this objective by enhancing the observability of smart inverter equipped PV systems located behind AMI meters. Smart inverters convert the DC output of PV systems into real and reactive power, the ratio of which is governed by a power factor setting. This in turn is governed by the real-time voltage at its terminal, as per a pre-set volt/var curve. While it is possible to infer the power factor setting - thereby, real and reactive power injection of the PV system - at each time interval by quasi-static time series simulations, such methods are not scalable to large grid sizes. We propose to reduce the reliance on simulations by developing a deep neural network based data-driven model which predicts the time dependent power factors of the PV systems in the grid. The model uses a small amount of simulated data to learn the dependence of power factor on 1) load and apparent PV power injection (root mean square of real and reactive power) data, and 2) AMI net-load data (for BTM PV assets). Our results show that our model can predict power factors with mean absolute error of less than 6.2% under full observability scenario and under 8.6% under the partial observability scenario. Thus, our data-driven model can reduce the amount of simulations required to compute real and reactive injections from PV systems in the distribution system with or without complete information.
Classification of Semiconductors Using Photoluminescence Spectroscopy and Machine Learning
Photoluminescence spectroscopy is a nondestructive optical method that is widely used to characterize semiconductors. In the photoluminescence process, a substance absorbs photons and emits light with longer wavelengths via electronic transitions. This paper discusses a method for identifying substances from their photoluminescence spectra using machine learning, a technique that is efficient in making classifications. Neural networks were constructed by taking simulated photoluminescence spectra as the input and the identity of the substance as the output. Here, six different semiconductors were chosen as categories: gallium oxide (Ga 2 O 3 ), zinc oxide (ZnO), gallium nitride (GaN), cadmium sulfide (CdS), tungsten disulfide (WS 2 ), and cesium lead bromide (CsPbBr 3 ). The developed algorithm has a high accuracy (>90%) for assigning a substance to one of these six categories from its photoluminescence spectrum and correctly identified a mixed Ga 2 O 3 /ZnO sample.
Advancing graph-based algorithms for predicting flow and transport in fractured rock
Discrete fracture network (DFN) models are a powerful alternative to continuum models for subsurface flow and transport simulations because they explicitly include fracture geometry and network topology, thereby allowing for better characterization of the latter's influence on flow and transport through fractured media. Recent advances in high performance computing have opened the door for flow and transport simulations in large explicit three-dimensional DFN, but this increase in model fidelity and system size comes at a huge computational cost because of the large number of mesh elements required to represent thousands of fractures (with sizes that can range several orders of magnitude, from mm to km). In most subsurface applications, fracture characteristics are only known statistically and numerous realizations are needed to bound uncertainty in flow and transport in the system, thereby exacerbating the computational burden. blackGraphs provide a simple and elegant way to characterize, query, and interrogate fracture network connectivity. We propose a DFN model reduction framework where various graph-representations are used in conjunction with high-fidelity DFN simulations to increase computational efficiency while retaining accuracy of key quantities of interest. The appropriate choice of a graph-representation, namely, which attributes of the DFN are to be represented as nodes and which ones as edges connecting those nodes, depends on the relevant scientific questions. We demonstrate that the proposed DFN model reduction framework provides an efficient means for DFN modeling through both system reduction of the DFN using graph-based properties and combining DFN and graph-based flow and transport simulations.
Experimental studies of adiabatic flow boiling in fractal-like branching microchannels
Experimental results of adiabatic boiling of water flowing through a fractal-like branching microchannel network are presented and compared to numerical model simulations. The goal is to assess the ability of current pressure loss models applied to a bifurcating flow geometry. The fractal-like branching channel network is based on channel length and width ratios between adjacent branching levels of 2{sup -1/2}. There are four branching sections for a total flow length of 18 mm, a channel height of 150 {mu}m and a terminal channel width of 100 {mu}m. The channels were Deep Reactive Ion Etched (DRIE) into a silicon disk. A Pyrex disk was anodically bonded to the silicon to form the channel top to allow visualization of the flow within the channels. The flow rates ranged from 100 to 225 g/min and the inlet subcooling levels varied from 0.5 to 6 C. Pressure drop along the flow network and time averaged void fraction in each branching level were measured for each of the test conditions. The measured pressure drop ranged from 20 to 90 kPa, and the measured void fraction ranged from 0.3 to 0.9. The measured pressure drop results agree well with separated flow model predictions accounting for the varying flow geometry. The measured void fraction results followed the same trends as the model; however, the scatter in the experimental results is rather large. (author)
Real-time evolution of Anderson impurity models via tensor network influence functionals
In this work, we present and analyze two tensor network-based influence functional approaches for simulating the real-time dynamics of quantum impurity models such as the Anderson model. Via comparison with recent numerically exact simulations, we show that such methods accurately capture the long-time nonequilibrium quench dynamics. The two parameters that must be controlled in these tensor network influence functional approaches are a time discretization (Trotter) error and a bond dimension (tensor network truncation) error. We show that the actual numerical uncertainties are controlled by an intricate interplay of these two approximations, which we demonstrate in different regimes. Our work opens the door to using these tensor network influence functional methods as general impurity solvers.
Link Scheduling in Satellite Networks via Machine Learning Over Riemannian Manifolds
Low Earth Orbit (LEO) satellites play a crucial role in enhancing global connectivity, serving a complementary solution to existing terrestrial systems. In wireless networks, scheduling is a vital process that allocates time-frequency resources to users for interference management. However, LEO satellite networks face significant challenges in scheduling their links towards ground users due to the satellites’ mobility and overlapping coverage. This paper addresses the dynamic link scheduling problem in LEO satellite networks by considering spatio-temporal correlations introduced by the satellites’ movements. The first step in the proposed solution involves modeling the network over Riemannian manifolds, thanks to their representation as symmetric positive definite matrices. We introduce two machine learning (ML)-based link scheduling techniques that model the dynamic evolution of satellite positions and link conditions over time and space. To accurately predict satellite link states, we present a recurrent neural network (RNN) over Riemannian manifolds, which captures spatio-temporal characteristics over time. Furthermore, we introduce a separate model, the convolutional neural network (CNN) over Riemannian manifolds, which captures geometric relationships between satellites and users by extracting spatial features from the network topology across all links. Simulation results demonstrate that both RNN and CNN over Riemannian manifolds deliver comparable performance to the fractional programming-based link scheduling (FPLinQ) benchmark. Remarkably, unlike other ML-based models that require extensive training data, both models only need 30 training samples to achieve over 99% of the sum rate while maintaining similar computational complexity relative to the benchmark.
Specification, annotation, visualization and simulation of a large rule-based model for ERBB receptor signaling
Mathematical/computational models are needed to understand cell signaling networks, which are complex. Signaling proteins contain multiple functional components and multiple sites of post-translational modification. The multiplicity of components and sites of modification ensures that interactions among signaling proteins have the potential to generate myriad protein complexes and post-translational modification states. As a result, the number of chemical species that can be populated in a cell signaling network, and hence the number of equations in an ordinary differential equation model required to capture the dynamics of these species, is prohibitively large. To overcome this problem, the rule-based modeling approach has been developed for representing interactions within signaling networks efficiently and compactly through coarse-graining of the chemical kinetics of molecular interactions. Here, we provide a demonstration that the rule-based modeling approach can be used to specify and simulate a large model for ERBB receptor signaling that accounts for site-specific details of protein-protein interactions. The model is considered large because it corresponds to a reaction network containing more reactions than can be practically enumerated. The model encompasses activation of ERK and Akt, and it can be simulated using a network-free simulator, such as NFsim, to generate time courses of phosphorylation for 55 individual serine, threonine, and tyrosine residues. The model is annotated and visualized in the form of an extended contact map. With the development of software that implements novel computational methods for calculating the dynamics of large-scale rule-based representations of cellular signaling networks, it is now possible to build and analyze models that include a significant fraction of the protein interactions that comprise a signaling network, with incorporation of the site-specific details of the interactions. Modeling at this level of detail is important for understanding cellular signaling.
In situ Parallel Training of Analog Neural Network Using Electrochemical Random-Access Memory
In-memory computing based on non-volatile resistive memory can significantly improve the energy efficiency of artificial neural networks. However, accurate in situ training has been challenging due to the nonlinear and stochastic switching of the resistive memory elements. One promising analog memory is the electrochemical random-access memory (ECRAM), also known as the redox transistor. Its low write currents and linear switching properties across hundreds of analog states enable accurate and massively parallel updates of a full crossbar array, which yield rapid and energy-efficient training. While simulations predict that ECRAM based neural networks achieve high training accuracy at significantly higher energy efficiency than digital implementations, these predictions have not been experimentally achieved. In this work, we train a 3 × 3 array of ECRAM devices that learns to discriminate several elementary logic gates (AND, OR, NAND). We record the evolution of the network’s synaptic weights during parallel in situ (on-line) training, with outer product updates. Due to linear and reproducible device switching characteristics, our crossbar simulations not only accurately simulate the epochs to convergence, but also quantitatively capture the evolution of weights in individual devices. The implementation of the first in situ parallel training together with strong agreement with simulation results provides a significant advance toward developing ECRAM into larger crossbar arrays for artificial neural network accelerators, which could enable orders of magnitude improvements in energy efficiency of deep neural networks.
Neuromorphic intermediate representation: A unified instruction set for interoperable brain-inspired computing
Abstract Spiking neural networks and neuromorphic hardware platforms that simulate neuronal dynamics are getting wide attention and are being applied to many relevant problems using Machine Learning. Despite a well-established mathematical foundation for neural dynamics, there exists numerous software and hardware solutions and stacks whose variability makes it difficult to reproduce findings. Here, we establish a common reference frame for computations in digital neuromorphic systems, titled Neuromorphic Intermediate Representation (NIR). NIR defines a set of computational and composable model primitives as hybrid systems combining continuous-time dynamics and discrete events. By abstracting away assumptions around discretization and hardware constraints, NIR faithfully captures the computational model, while bridging differences between the evaluated implementation and the underlying mathematical formalism. NIR supports an unprecedented number of neuromorphic systems, which we demonstrate by reproducing three spiking neural network models of different complexity across 7 neuromorphic simulators and 4 digital hardware platforms. NIR decouples the development of neuromorphic hardware and software, enabling interoperability between platforms and improving accessibility to multiple neuromorphic technologies. We believe that NIR is a key next step in brain-inspired hardware-software co-evolution, enabling research towards the implementation of energy efficient computational principles of nervous systems. NIR is available atneuroir.org