Reduce Simulation Time Research Articles

An Efficient Computational Model for Magnetic Pulse Forming of Thin Structures.

Electromagnetic forming (EMF) is one of the most popular high-speed forming processes for sheet metals. However, modeling this process in 3D often requires huge computational time since it deals with a strongly coupled multi-physics problem. The numerical tools that are capable of modeling this process rely either on shell elements-based approaches or on full 3D elements-based approaches. The former leads to reduced computational time at the expense of the accuracy, while the latter favors accuracy over computation time. Herein, a novel approach was developed to reduce CPU time while maintaining reasonable accuracy through building upon a 3D finite element analysis toolbox which was developed in CEMEF. This toolbox was used to solve magnetic pulse forming (MPF) of thin sheets. The problem was simulated under different conditions and the results were analyzed in-depth. Innovative techniques, such as developing a termination criterion and using adaptive re-meshing, were devised to overcome the encountered problems. Moreover, a solid shell element was implemented and tested for thin structure problems and its applicability was verified. The results of this element type were comparable to the results of the standard tetrahedral MINI element but with reduced simulation time.

Order-to-Disorder Transition of Cylinder-Forming Block Copolymer Films Confined within Neutral Interfaces

We present the phase behavior of cylinder-forming block copolymer (BCP) films confined within neutral interfaces through theory and experiment. The Langevin field-theoretic simulation (L-FTS) is utilized to account for the compositional fluctuation effect on the order-to-disorder transition (ODT), and the ultraviolet divergence in the L-FTS result is eliminated by renormalization of the Flory–Huggins parameter χ, allowing accurate comparison of thickness-dependent (χN)ODT, where N is the polymerization index. The L-FTS is accelerated by utilizing a deep learning method, and we managed to achieve a 50–60% reduction in simulation time. The L-FTS results show that (χN)ODT decreases as the film thickness decreases at a very large value of the invariant polymerization index N̅, and BCPs undergo a phase transition from disordered to cylindrical phases through a spherical phase as predicted by the self-consistent field theory calculation. By performing the L-FTS at experimentally relevant low N̅, we show that the BCPs undergo a direct transition from disordered to cylindrical phases, and the (χN)ODT value of cylinder-forming BCPs confined within neutral interfaces increases as the film thickness decreases. To verify and complement the theoretical results, we investigate the ODT temperature (TODT) of cylinder-forming polystyrene-b-poly(2-vinylpyridine) (PS-b-P2VP) films supported on a neutral mat of P(S-r-2VP) random copolymers using grazing-incidence small-angle X-ray scattering. Our film experiments with low-N̅ PS-b-P2VP confirm the direct transition from disordered to cylindrical phases and the decrease in TODT with decreasing thickness below the onset thickness above which the TODT levels off.

Eddy3D: A toolkit for decoupled outdoor thermal comfort simulations in urban areas

The architectural community needs holistic, evidence-based planning tools to promote urban resilience in the face of global warming. To ensure maximum impact, simulation-driven microclimate analysis methods must be integrated early in the design process. With Eddy3D, we present a toolkit to simulate outdoor thermal comfort metrics with a decoupled approach. We motivate the decoupled systems framework with meteorological measurements and local and global sensitivity analyses of three different climates. For a real-world case study on a university campus, we present results for both wind velocity and mean radiant temperature simulations. Finally, we discuss the advantages and disadvantages of a decoupled simulation approach that considers design aiding for the architectural practice. Our findings support reduced simulation time and flexibility, with the caveat of reduced accuracy due to neglecting forced convection, albeit this being less relevant in the early stages of design. The framework presented in this manuscript has been implemented and released as Eddy3D, a plugin for Rhino & Grasshopper.

Novel time‐domain average model for harmonic current prediction in photovoltaic and wind power units

Computer models designed to assess harmonic levels created by renewable energy sources (RESs) need to reproduce the converter operation effect. However, only a few simplified time-domain models have been developed to predict harmonics while reducing simulation time, which is one of the main limitations of detailed models. This paper presents a novel time-domain model, called the harmonic average model (HM), to estimate harmonic currents generated by inverters. The superposition principle is applied to include dead time and switching effects in the average model. In addition, the snubber effect is considered by introducing snubber circuits in the HM. Time-domain simulations of a type-IV wind turbine are executed with existing models and then with the HM to evaluate the proposed model. Field measurements in a wind power plant and in a photovoltaic plant are used to validate the HM under different power injection scenarios. The results indicate that the HM follows reference currents while accurately estimating the harmonic spectra. Further, the HM is relatively simple to deploy and reduces computation time. A comparison between the HM results and field measurements suggests that the simulated spectra are an adequate approximation of the measured spectra. These results confirm that the HM can be used by transmission system operators to support the prediction of harmonic currents generated by inverters and to estimate the impact of RESs on voltage quality.

GATE Monte Carlo dosimetry in 90Y TARE planning: influence of simulation parameters and image resampling on dosimetric accuracy and optimization of computational times

Direct Monte Carlo (MC) simulation is considered the gold standard approach for internal dosimetry in nuclear medicine, and it is increasingly used in planning Trans-Arterial Radio-Embolization (TARE) of HepatoCellular Carcinoma (HCC) {and hepatic metastases}. However its computational times, longer with respect to other simplified approaches, constitute a limiting factor, especially when dealing with {large size and finely discretized voxelized volumes}. Aim of this work was the investigation of the influence of cuts on the production of secondary particles and of input CT images resamplings on dosimetric accuracy and computational time in patient-specific voxel-level MC simulations of 90 Y-labelled glass microspheres TARE treatment, to find optimal combinations of settings for speeding up such simulations. GATE GEANT4 interface was used to perform simulations employing CT and 99m Tc SPECT as input data, examining multiple CT resolutions (via CT resamplings characterized by voxel volume factors 2, 8, and 64 with respect to native one, and a CT resampling with SPECT resolution) and production cuts (0.01 mm, 0.05 mm, 0.1 mm, 0.5 mm and some more, specific for each resampling). Increasing cut length and reducing CT resolution produces an early rapid decrease followed by a late slow decrease of simulation time as a function of this two parameters. A reduction up to 30% with respect to reference simulation time, while preserving acceptable dosimetric accuracy, was obtained. The best combination of settings among the examined ones resulted the choice of CT resampling with 8 times the native voxel volume and of 0.1-0.5 mm cut, ensuring dosimetric agreement within 1% in liver-related VOIs, while reducing simulation time to 45%.

Robustification of the H[formula omitted] controller combined with fuzzy logic and PI&PID-F[formula omitted] for hybrid control of Wind Energy Conversion System Connected to the Power Grid Based on DFIG

This paper proposes a strategy for hybrid control of variable speed wind turbines connected to the electrical grid, using a Double Fed Induction Generator (DFIG). An efficient validation approach is presented in three steps. The first step concerns the comparison of performances between Fuzzy Logic (FL), H infinity (H∞) and Integral Proportional action (PI) controllers. In a system context based on the DFIG direct vector control structure, the strengths of H∞ are reinforced in the second step by implementing two new advanced mixed controllers, namely the FL-H∞ and the combination of the PI, the Proportional Integral Derivative (PID) action, the Filter Derivative (Fd) action and the H∞ (PI&PID-Fd)-H∞. The second step, based on a simplified system of indirect vector control structure of the DFIG, allows the performance of the (PI&PID-Fd)-H∞ to be compared with that of the FL-H∞. In contrast to the latter, it emerges that, in addition to its simplicity of implementation and the reduction in calculation and simulation time, the (PI&PID-Fd)-H∞ is very reassuring in terms of robustness against mixed uncertainties, stability, precision, tracking and decoupling between active (PS) and reactive (QS)power. This makes (PI&PID-Fd)-H∞ a very advantageous choice for the design of the proposed hybrid control system. The final step gives a more real-life picture of hybrid control of the wind system as follows, a random wind model is considered, an FL regulates the Maximum Power Point Tracking (MPPT) of the turbine model, the PI-H∞ and (PID-Fd)-H∞ control the Ps and Qs on the nominal DFIG model via a Rotor Side Converter (RSC), an FL regulates the DC-BUS voltage and the H∞ control the currents via a Grid Side Converter (GSC) and filter. According to the results, and in contrast to a conventional PI control system, the hybrid control system offered has very good performance, particularly with regard to, for example, optimal extraction of energy from the wind, the currents fed into the grid are well balanced, the Total Harmonic Distortion (THD) is reduced and the cos ϕ is almost equal to unity. In addition, a good compromise between stability and expected performance is ensured. Consequently, the proposed strategy has reasonable and very favourable advantages for the optimisation of energy conversion in grid-connected wind energy systems. The results are developed in the MATLAB/Simulink environment.

Microgrid Oriented modeling of space heating system based on neural networks

Balancing the energy production and consumption is a huge challenge for future smart grids . In this context, many demand-side management programs are being developed to achieve flexibility from different loads like space heating. As space heating models for flexibility simulations are an interdisciplinary field of work, complex civil engineering thermal models need to be combined with complex electrical engineering control simulations in different software frameworks. Traditionally used methods have shortcomings in one of those two domains as the publications that provide complex control strategies for demand response are lacking complex thermal models and vice versa. Co-simulations overcome this problem but are computationally expensive and have compatibility limitations. Thus, the aim of this work is to develop a methodology for designing space heating/cooling models, intended for positive energy district- or smart city simulations, which provide high accuracy at low computational expense. This could be achieved by synthesizing neural network object models from IDA-ICE civil engineering models in Matlab. These machine learning models showed improvements of more than 30% in different error metrics and a simulation time reduction of more than 80% compared to other methods, making them suitable for use in microgrid simulations, including flexibility analyses. • Description of IDA-ICE thermal building models for dataset creation. • Development of novel neural network-based approach for space heating models. • Error metrics comparison of different space heating models. • Comparison of different modeling methods, incl. co-simulation and machine learning models.

Model order reduction of building energy simulation models using a convolutional neural network autoencoder

Building energy simulation (BES) tools are fundamental for predicting energy performance and comfort. However, detailed models are computationally complex and demand high simulation times. These lead to difficulties in parametric runs and numerical optimizations. Performing numerous retrofit scenarios is hardly feasible, especially in multi-zone buildings with complex geometries.This paper introduces a novel approach to model order reduction (MOR) of BES models. The approach utilizes a deep learning-based unsupervised convolutional neural network autoencoder (CNN-AE). The method decomposes complex time series data derived from detailed simulations into lower dimension features. The low-dimension representations can be grouped through clustering algorithms to build a reduced-order model (ROM). The approach in this study automatically finds archetype zones of the original model that represent the energy behavior of a group of rooms, and removes redundant ones. The energy demand of the whole building can be estimated through these archetype zones.Our investigation shows that CNN-AE can be efficiently applied to reduce complex building energy simulation models. As proof of concept, a detailed model of a multi-zone campus building with 889 thermal zones is compared to the ROM derived from the CNN-AE. Comprehensive autoencoder hyperparameter training to optimize the accuracy of the model is provided. The ROM supports different purposes, such as energy scenario developments, with a total error of less than 1% compared to the original model, and reduced simulation times by a factor of more than 16.

Rapid stability analysis of serrated end mills using graphical-frequency domain methods

Serrated tools result in an irregular distribution of chip thickness and varying time delays along their flutes that disturb the regenerative chatter mechanism and maximize productivity in rough machining operations. The chatter stability simulation with such tools is complex due to the delay differential equations with multiple and distributed delays. Although the time domain numerical integration methods and the semi-discretization method are accurate, they are computationally expensive and hence inhibit the optimal design of serrated cutters. Despite the advantages that serrated tools offer, there has been little attention paid to rapidly predict stability lobes diagram for these tools to help instruct their optimal designs. To address this need, new graphical-frequency methods are presented to solve for stability of serrated tools with arbitrary geometry. The graphical method averages the number of teeth in cut to make the system time-invariant and results in a closed-form analytical solution. It reduces the simulation time by up to ∼99% compared to the semi-discretization method. Generalized solutions to the classical zero-order approximation and the multi-frequency methods are also presented by using the Nyquist stability criterion. The use of the zero-order approximation method and the multi-frequency method reduces simulation times by up to ∼97% and ∼79% respectively compared to the semi-discretization method. These are significant improvements over reports in the previous literature. Stability predictions are validated experimentally and benchmarked with high-fidelity time domain methods. The presented methods are general and can be applied to any end mill with non-uniform pitch and helix angles as well.

A Review of Parallel Heterogeneous Computing Algorithms in Power Systems

The power system expansion and the integration of technologies, such as renewable generation, distributed generation, high voltage direct current, and energy storage, have made power system simulation challenging in multiple applications. The current computing platforms employed for planning, operation, studies, visualization, and the analysis of power systems are reaching their operational limit since the complexity and size of modern power systems results in long simulation times and high computational demand. Time reductions in simulation and analysis lead to the better and further optimized performance of power systems. Heterogeneous computing—where different processing units interact—has shown that power system applications can take advantage of the unique strengths of each type of processing unit, such as central processing units, graphics processing units, and field-programmable gate arrays interacting in on-premise or cloud environments. Parallel Heterogeneous Computing appears as an alternative to reduce simulation times by optimizing multitask execution in parallel computing architectures with different processing units working together. This paper presents a review of Parallel Heterogeneous Computing techniques, how these techniques have been applied in a wide variety of power system applications, how they help reduce the computational time of modern power system simulation and analysis, and the current tendency regarding each application. We present a wide variety of approaches classified by technique and application.

Denoising proton therapy Monte Carlo dose distributions in multiple tumor sites: A comparative neural networks architecture study.

Monte Carlo (MC) algorithms provide accurate modeling of dose calculation by simulating the delivery and interaction of many particles through patient geometry. Fast MC simulations using large number of particles are desirable as they can lead to reliable clinical decisions. In this work, we assume that faster simulations with fewer particles can approximate slower ones by denoising them with deep learning. We use mean squared error (MSE) as loss function to train networks (sNet and dUNet), with 2.5D and 3D setups considering volumes of 7 and 24 slices. Our models are trained on proton therapy MC dose distributions of six different tumor sites acquired from 50 patients. We provide networks with input MC dose distributions simulated using 1×106 particles while keeping 1×109 particles as reference. On average over 10 new patients with different tumor sites, in 2.5D and 3D, our models recover relative residual error on target volume, ΔD95TV of 0.67±0.43% and 1.32±0.87% for sNet vs. 0.83±0.53% and 1.66±0.98% for dUNet, compared to the noisy input at 12.40±4.06%. Moreover, the denoising time for a dose distribution is: < 9s and <1s for sNet vs. < 16s and <1.5s for dUNet in 2.5D and 3D, in comparison to about 100min (MC simulation using 1×109 particles). We propose a fast framework that can successfully denoise MC dose distributions. Starting from MC doses with 1×106 particles only, the networks provide comparable results as MC doses with1×109 particles, reducing simulation time significantly.

Reducing molecular simulation time for AFM images based on super-resolution methods.

Atomic force microscopy (AFM) has been an important tool for nanoscale imaging and characterization with atomic and subatomic resolution. Theoretical investigations are getting highly important for the interpretation of AFM images. Researchers have used molecular simulation to examine the AFM imaging mechanism. With a recent flurry of researches applying machine learning to AFM, AFM images obtained from molecular simulation have also been used as training data. However, the simulation is incredibly time consuming. In this paper, we apply super-resolution methods, including compressed sensing and deep learning methods, to reconstruct simulated images and to reduce simulation time. Several molecular simulation energy maps under different conditions are presented to demonstrate the performance of reconstruction algorithms. Through the analysis of reconstructed results, we find that both presented algorithms could complete the reconstruction with good quality and greatly reduce simulation time. Moreover, the super-resolution methods can be used to speed up the generation of training data and vary simulation resolution for AFM machine learning.

FastCAT: Fast cone beam CT (CBCT) simulation.

To develop fastCAT, a fast cone-beam computed tomography (CBCT) simulator. fastCAT uses pre-calculated Monte Carlo (MC) CBCT phantom-specific scatter and detector response functions to reduce simulation time for megavoltage (MV) and kilovoltage (kV) CBCT imaging. Pre-calculated x-ray beam energy spectra, detector optical spread functions and energy deposition, and phantom scatter kernels are combined with GPU raytracing to produce CBCT volumes. MV x-ray beam spectra are simulated with EGSnrc for 2.5- and 6MeV electron beams incident on a variety of target materials and kV x-ray beam spectra are calculated analytically for an x-ray tube with a tungsten anode. Detectors were modeled in Geant4 extended by Topas and included optical transport in the scintillators. Two MV detectors were modeled-a standard Varian AS1200 GOS detector and a novel CWO high detective quantum efficiency detector. A kV CsI detector was also modeled. Energy-dependent scatter kernels were created in Topas for two 16cm diameter phantoms: A Catphan 515 contrast phantom and an anthropomorphic head phantom. The Catphan phantom contained inserts of 1-5mm in diameter of six different tissue types: brain, deflated lung, compact and cortical bone, adipose, and B-100. fastCAT simulations retain high fidelity to measurements and MC simulations: MTF curves were within 3.5% and 1.2% of measured values for the CWO and GOS detectors, respectively. HU values and CNR in a fastCAT Catphan 515 simulation were seen to be within 95% confidence intervals of an equivalent MC simulation for all of the tissues with root mean squared errors less than 16 HU and 1.6 in HU values and CNR comparisons, respectively. The anthropomorphic head phantom CWO detector CBCT image resulted in much higher tissue contrast and lower noise compared to the GOS detector CBCT image. A fastCAT simulation of the Catphan 515 module with an image size of 1024 1024 10 voxels took 61s on a GPU while the equivalent Topas MC was estimated to take more than 0.3CPU years. We present an open source fast CBCT simulation with high fidelity to MC simulations. The fastCAT python package can be found at https://github.com/jerichooconnell/fastCAT.git.

Conversion of AC wind energy system model to DC model for integration to optimization software with large scales

The work presented in this paper deals with an AC model of wind turbine system conversion to a DC model with reduced simulation time, for possible integration to optimization software with larges scales permitting a multi-objective optimization, such as the constrained optimization conjointly of the cost and power losses of the wind turbine energy system. The DC model is based on average calculation of the DC voltage recharging the battery energy accumulator used for recovering the converted wind energy and the electromagnetic torque. Indeed, classical model of wind turbine using generally electric generator associated to a PD3 rectifier to convert the alternative energy on DC energy recoverable on battery energy accumulator with need a large simulation time and thereafter it is non integrable to optimization software for multi-objective optimization problem resolution. The two models are implemented under Matlab-Simulink simulation environment. Simulation results valid entirely the wind turbine system DC model. Finally, as perspective it is interesting to use a booster chopper as an interface between the rectifier and the battery to optimize the recovered energy. An average model of the booster chopper can be integrated into the DC model for performance improvement of the conversion chain.

Capabilities and Limitations of 3D-CFD Simulation of Anode Flow Fields of High-Pressure PEM Water Electrolysis

In this work, single-phase (liquid water) and two-phase (liquid water and gaseous oxygen) 3D-CFD flow analysis of the anode of a high pressure PEM electrolysis cell was conducted. 3D-CFD simulation models of the anode side porous transport layer of a PEM electrolyzer cell were created for the flow analysis. For the geometrical modelling of the PTL, two approaches were used: (a) modelling the exact geometry and (b) modelling a simplified geometry using a porosity model. Before conducting two-phase simulations, the model was validated using a single-phase approach. The Eulerian multiphase and the volume-of-fluid approaches were used for the two-phase modelling and the results were compared. Furthermore, a small section of the PTL was isolated to focus on the gas bubble flow and behaviour in more detail. The results showed plausible tendencies regarding pressure drop, velocity distribution and gas volume fraction distribution. The simplified geometry using the porous model could adequately replicate the results of the exact geometry model with a significant reduction in simulation time. The developed simulation model can be used for further investigations and gives insight into two-phase flow phenomena in the PTL. Additionally, the information obtained from simulation can aid the design and evaluation of new PTL structures.

Device-Level Parallel-in-Time Simulation of MMC-Based Energy System for Electric Vehicles

The device-level electromagnetic transient (EMT) simulation with the nonlinear behaviour model (NBM) of insulated-gate bipolar transistors (IGBTs) and diodes can provide an accurate insight into the power converters from the perspective of thermal performance and energy efficiency. However, device-level simulation is rarely implemented in electric vehicles (EVs) due to its extreme computation complexity natively introduced by the device models. To solve this problem, an interpolation strategy is designed based on the parallel-in-time algorithm for the device-level simulation of the modular multilevel converter (MMC) connected with the induction machine in EV applications. The MMC is mathematically separated as multiple submodules with the same attributes, which can be processed in a parallel manner in the graphics processing unit (GPU). By implementing the device-level simulation in the different time-step in GPU, the interpolation strategy provides the precise initial values for the nonlinear solution process iteratively. The accuracy of the proposed simulation scheme is validated by commercial simulation tools at the device level. In addition, the system-level simulation of EVs is carried out at different driving cycles, and the results demonstrate a significant reduction in simulation time.

Quasistatic Solutions versus Full-Wave Solutions of Single-Channel Circular RF Receive Coils on Phantoms of Varying Conductivities at 3 Tesla.

Purpose.Although full-wave simulations could be used to aid in RF coil design, the algorithms may be too slow for an iterative optimization algorithm. If quasistatic simulations are accurate within the design tolerance, then their use could reduce simulation time by orders of magnitude compared to full-wave simulations. This paper examines the accuracy of quasistatic and full-wave simulations at 3 Tesla.Methods.Three sets of eight coils ranging from 3–10 cm (24 total) were used to measure SNR on three phantoms with conductivities of 0.3, 0.6, and 0.9 S/m. The phantom conductivities were chosen to represent those typically found in human tissues. The range of coil element sizes represents the sizes of coil elements seen in typical coil designs. SNR was determined using the magnetic and electric fields calculated by quasistatic and full-wave simulations. Each simulated SNR dataset was scaled to minimize the root mean squared error (RMSE) when compared against measured SNR data. In addition, the noise values calculated by each simulation were compared against benchtop measured noise values.Results.The RMSE was 0.285 and 0.087 for the quasistatic and full-wave simulations, respectively. The maximum and minimum quotient values, when taking the ratio of simulated to measured SNR values, were 1.69 and 0.20 for the quasistatic simulations and 1.29 and 0.75 for the full-wave simulations, respectively. The ratio ranges, for the calculated quasistatic and full-wave total noise values compared to benchtop measured noise values, were 0.83–1.06 and 0.27–3.02, respectively.Conclusions.Full-wave simulations were on average 3x more accurate than the quasistatic simulations. Full-wave simulations were more accurate in characterizing the wave effects within the sample, though they were not able to fully account for the skin effect when calculating coil noise.

Multi-Area Aggregation of Multi-Grounded Unbalanced Distribution Systems

Currently, several methods applied in distribution systems need power flow solutions, such as evolutionary algorithms, exhaustive search, neural network training, continuation power flow, quasi-static time-series, among others. A properly modeled distribution system requires detailing and expects a three-phase network with neutrals and groundings, which help increase simulation times. For a thorough modeling, but with reduced simulation time, distribution system equivalency can be determined using a multi-area aggregation method in three-phase four-wire multi-grounded feeders, where even the modeling of neutral cables and groundings are allowed. The proposed method enables the user to represent complex network components as a reduced size subnet, achieving a considerable computational time reduction. The proposed method can be inserted into OpenDSS , or multiphase methods, using Newton - Raphson and Backward-Forward Sweep algorithms. Results show that the method is capable of mathematically finding the correct solution for all loading levels. The exactness of the proposition is evaluated using the simple NEV test feeder, IEEE 123, the IEEE 8500 Node Test Feeder, and a meshed network having unbalanced feeders with mutual coupling between different voltage levels.

TxSim: Modeling Training of Deep Neural Networks on Resistive Crossbar Systems

Deep neural networks (DNNs) have gained tremendous popularity in recent years due to their ability to achieve superhuman accuracy in a wide variety of machine learning tasks. However, the compute and memory requirements of DNNs have grown rapidly, creating a need for energy-efficient hardware. Resistive crossbars have attracted significant interest in the design of the next generation of DNN accelerators due to their ability to natively execute massively parallel vector-matrix multiplications within dense memory arrays. However, crossbar-based computations face a major challenge due to device and circuit-level nonidealities, which manifest as errors in the vector-matrix multiplications and eventually degrade DNN accuracy. To address this challenge, there is a need for tools that can model the functional impact of nonidealities on DNN training and inference. Existing efforts toward this goal are either limited to inference or are too slow to be used for large-scale DNN training. We propose TxSim, a fast and customizable modeling framework to functionally evaluate DNN training on crossbar-based hardware considering the impact of nonidealities. The key features of TxSim that differentiate it from prior efforts are: 1) it comprehensively models nonidealities during all training operations (forward propagation, backward propagation, and weight update) and 2) it achieves computational efficiency by mapping crossbar evaluations to well-optimized Basic Linear Algebra Subprograms (BLAS) routines and incorporates speedup techniques to further reduce simulation time with minimal impact on accuracy. TxSim achieves 6×- 108× improvement in simulation speed over prior works, and thereby makes it feasible to evaluate the training of large-scale DNNs on crossbars. Our experiments using TxSim reveal that the accuracy degradation in DNN training due to nonidealities can be substantial (3%-36.4%) for large-scale DNNs and data sets, underscoring the need for further research in mitigation techniques. We also analyze the impact of various device and circuit-level parameters and the associated nonidealities to provide key insights that can guide the design of crossbar-based DNN training accelerators.

Surfer: A fast simulation algorithm to predict surface temperatures and mean radiant temperatures in large urban models

Outdoor thermal comfort simulation simulations rely on the mean radiant temperature (MRT) seen by pedestrians as an important input that remains difficult to compute. Especially for large urban models, computing relevant surface temperatures and radiation fluxes that make up the MRT is a daunting task in terms of simulation setup and the computational overhead. We propose a new algorithm to estimate exterior surface temperatures of building facades, roofs, and ground surfaces in an arbitrary urban 3D model. The algorithm discretizes all model surfaces and clusters them by material properties and sky and sun exposure to reduce computational complexity. The model setup is fully automated, and the algorithm is implemented in the popular Rhino3d CAD environment. We demonstrate the accuracy of the algorithm by comparing both the resulting external surface temperatures against a high-fidelity simulation and the final MRT against real-world measurements. We report an RMSE of 1.8 °C and 2.0 °C, respectively, while reducing simulation times by a factor of ~80. Envisioned applications of the algorithm range from rapid microclimate simulations in fast-paced urban design processes to large scale urban comfort evaluation of existing cities.