Fast Computation Research Articles

BiFSMNv2: Pushing Binary Neural Networks for Keyword Spotting to Real-Network Performance.

Deep neural networks, such as the deep-FSMN, have been widely studied for keyword spotting (KWS) applications while suffering expensive computation and storage. Therefore, network compression technologies such as binarization are studied to deploy KWS models on edge. In this article, we present a strong yet efficient binary neural network for KWS, namely, BiFSMNv2, pushing it to the real-network accuracy performance. First, we present a dual-scale thinnable 1-bit-architecture (DTA) to recover the representation capability of the binarized computation units by dual-scale activation binarization and liberate the speedup potential from an overall architecture perspective. Second, we also construct a frequency-independent distillation (FID) scheme for KWS binarization-aware training, which distills the high- and low-frequency components independently to mitigate the information mismatch between full-precision and binarized representations. Moreover, we propose the learning propagation binarizer (LPB), a general and efficient binarizer that enables the forward and backward propagation of binary KWS networks to be continuously improved through learning. We implement and deploy BiFSMNv2 on ARMv8 real-world hardware with a novel fast bitwise computation kernel (FBCK), which is proposed to fully use registers and increase instruction throughput. Comprehensive experiments show our BiFSMNv2 outperforms the existing binary networks for KWS by convincing margins across different datasets and achieves comparable accuracy with the full-precision networks (only a tiny 1.51% drop on Speech Commands V1-12). We highlight that benefiting from the compact architecture and optimized hardware kernel, BiFSMNv2 can achieve an impressive 25.1× speedup and 20.2× storage-saving on edge hardware.

A Fast ion Flow Field Calculation Method of UHVDC Transmission Lines Based on Adaptive Iteration Factor and Half-Error Term

A Fast ion Flow Field Calculation Method of UHVDC Transmission Lines Based on Adaptive Iteration Factor and Half-Error Term

Resolving Strain Localization in Frictional and Time‐Dependent Plasticity: Two‐ and Three‐Dimensional Numerical Modeling Study Using Graphical Processing Units (GPUs)

AbstractShear strain localization refers to the phenomenon of accumulation of material deformation in narrow slip zones. Many materials exhibit strain localization under different spatial and temporal scales, particularly rocks, metals, soils, and concrete. In the Earth's crust, irreversible deformation can occur in brittle as well as in ductile regimes. Modeling of shear zones is essential in the geodynamic framework. Numerical modeling of strain localization remains challenging due to the non‐linearity and multi‐scale nature of the problem. We develop a numerical approach based on graphical processing units (GPU) to resolve the strain localization in two and three dimensions of a (visco)‐hypoelastic‐perfectly plastic medium. Our approach allows modeling both the compressible and incompressible visco‐elasto‐plastic flows. In contrast to symmetric shear bands frequently observed in the literature, we demonstrate that using sufficiently small strain or strain rate increments, a non‐symmetric strain localization pattern is resolved in two‐ and three‐dimensions, highlighting the importance of high spatial and temporal resolution. We show that elasto‐plastic and visco‐plastic models yield similar strain localization patterns for material properties relevant to applications in geodynamics. We achieve fast computations using three‐dimensional high‐resolution models involving more than 1.3 billion degrees of freedom. We propose a new physics‐based approach explaining spontaneous stress drops in a deforming medium.

A fast calculation method for magnetic field of 110 kV oil-immersed transformers based on the combination of order reduction and response surface methodology

A fast calculation method for magnetic field of 110 kV oil-immersed transformers based on the combination of order reduction and response surface methodology

Globally optimal sequencing of optimal reactive dispatch control adjustments to minimize operational losses in transmission systems by graph shortest path, parallel computing, and dynamic programming

Minimizing operational losses in transmission systems through the Optimal Reactive Dispatch (ORD), a non-convex mixed-integer nonlinear programming problem, is crucial for operational cost reduction, resource optimization, and greenhouse gas emission mitigation. Besides all intricacies associated with solving ORDs, transmission system operators encounter the challenge of determining sequences in which ORD control adjustments must be implemented before significant changes occur in generators scheduled power output and system loading. Sequencing ORD control adjustments, in spite of not being novel, remains modestly scrutinized in the literature. This paper introduces a two-phase framework that tackles the globally optimal sequencing of n ORD control adjustments over n! potential paths by solving the ORD to minimize operational losses in transmission systems in the first phase, and optimally sequencing ORD control adjustments employing fast power flow calculations, graph shortest path, parallel computing, and dynamic programming in the second phase. We discuss the framework’s second phase asymptotic time complexity, which is exponential over factorial for brute-force approaches, and its capability to guarantee globally optimal paths toward minimal operational losses determined in the framework’s first phase. ORD control adjustments for transmission systems with up to 27 controllable variables are benchmarked against two mixed-integer nonlinear programming solvers: BARON, a global non-convex solver, and Knitro, a local solver (assuming convexity around local optima). Globally optimal sequences of ORD control adjustments over n! potential paths (more than 1028 for sequencing 27 control adjustments) and average algorithm runtimes validate the straightforward application and, more importantly, effectiveness of such a comprehensive framework.

NNEoS : Neural network-based thermodynamically consistent equation of state for fast and accurate flash calculations

Equations of state (EOS) correlate thermodynamic properties and are essential for flash calculations. However, solving for an EOS can be time-consuming, and EOS do not precisely represent physical reality, causing the deviation of flash results from phase equilibrium data. In this work, we propose a neural network-based EOS (NNEoS) inherently satisfying thermodynamic consistency. NNEoS first predicts the residual Gibbs energy and then derives other thermodynamic properties through differentiation. NNEoS can be trained using an analytical EOS and then serve as a reliable, computationally efficient substitute. NNEoS can also be fine-tuned with experimental data to better match flash results to experimental data. We evaluate the performance of NNEoS against analytical EOS on three case studies, including binary and multicomponent mixtures with and without cross-association. The results show that NNEoS achieves significantly faster flash calculations via GPU-based parallel computing and offers superior predictive accuracy after fine-tuning compared to analytical EOS.

Multi-Criteria Decision-Making Method for Simple and Fast Dimensioning and Selection of Glass Tube Collector Type Based on the Iterative Thermal Resistance Calculation Algorithm with Experimental Validation

This paper presents an analytical method for the dimensioning and selection of the four glass tube collector types: single-glazed with an air layer, single-glazed with a vacuum layer, double-glazed with an air layer, and double-glazed with a vacuum layer. In the first part of the paper (dimensioning phase), the iterative thermal resistance calculation algorithms were developed for all glass tube collector types, whereby the iterative thermal resistance calculation algorithm of the single-glazed tube collector with an air layer was experimentally tested and validated. The second part of the paper (selection phase) uses a multi-criteria decision-making method to determine the optimal glass tube collector design. Unlike other papers, three indicator groups are taken into account in this case: geometric (mass, surface occupation, total surface occupation, volume occupation), economic (manufacturing and exploitation costs), and ecological (embodied energy and greenhouse gas emission). The proposed method is characterized by simple and fast calculations with satisfactory accuracy, which avoids high investment costs (experimental research), approximation and discretization of physical models (numerical research), and a large number of input parameters with boundary conditions (theoretical research). It should be noted that, with certain additions and changes, it can also be applied to other solar thermal collectors, so the authors believe such tools are handy for the global scientific public.

Fast computation of radio wave diffraction effects

SummaryUnmanned aerial vehicle operations are quickly gaining ground due to rapid global market penetration. While on one hand, novel technologies that bridge communication networks to aviation industry are yet to be explored, on the other hand, their development requires highly scalable systems to enable beyond visual line‐of‐sight missions. This requirement imposes a big bottleneck in terms of computation complexity. This paper presents a method for fast computation of multiple diffraction of radio waves over knife‐edge obstacles based on the Deygout technique and some offline computation steps, including a ground profile analysis. We prove that this algorithm is equivalent to the original Deygout algorithm for all non‐line‐of‐sight points, show heuristics confirming that it is mostly applicable in the line‐of‐sight case. The computational and memory complexity of our algorithm is approximately , compared to for the original Deygout algorithm. Finally we discuss how to apply the approach to the Epstein‐Peterson technique and the Giovanelli technique, and how to use it to compute clutter‐loss.

Rocking block simulation based on numerical dissipation.

In this paper, a computational approach based on numerical dissipation is proposed to simulate rocking blocks. A rocking block is idealized as a solid body interacting with its foundation through a contact-based formulation. An implicit time integration scheme with numerical dissipation, set to optimally treat dissipation in contact problems, is employed. The numerical dissipation is ruled by the time step and the rocking dissipative phenomenon at impacts is accurately predicted without any damping model. A broad numerical campaign is conducted to define a regression law in analytic form for the setting of the time step, depending on the block size and aspect ratio, the contact stiffness, as well as the coefficient of restitution selected. The so-obtained regression law appears accurate and an a posteriori validation with cases not in the training dataset confirms the effectiveness of the approach. Finally, the comparison with available experimental tests highlights the approach efficacy for free rocking and harmonic loading cases (in a deterministic sense), and for earthquake-like loading cases (in a statistical sense). It is found that rocking blocks with sizes of interest for structural engineering (e.g., cultural heritage structures) can be simulated with time steps within 10-3 ÷ 10-1s, so allowing very fast computations.

Pseudo-Normalization via Integer Fast Inverse Square Root and Its Application to Fast Computation without Division

Pseudo-Normalization via Integer Fast Inverse Square Root and Its Application to Fast Computation without Division

Two-dimensional electromagnetic scattering analysis based on the boundary element method

An effective formula for the shape-sensitivity analysis of electromagnetic scattering is presented in this paper. First, based on the boundary element method, a new electromagnetic scattering formula is derived by combining the traditional electromagnetic scattering formula with the non-uniform rational B-spline (NURBS) curve, and the geometric model is represented by NURBS, which ensures the geometric accuracy, avoids the heavy grid division in the optimization process, and realizes the fast calculation of high-fidelity numerical solutions. Second, by deducing the sensitivity variables, the electromagnetic scattering equation of shape optimization is obtained, which can provide reliable data references for shape optimization. Finally, the effectiveness and accuracy of the algorithm are demonstrated by an example, and the sensitivity data of some examples are given.

A fast numerical method for calculating SPND space charge effect at steady state

To address the issue of low efficiency and long computation times in calculating the space charge effect in Self-Powered Neutron Detectors (SPNDs), we developed a fast and efficient numerical method to calculate the steady-state electric fields in SPND insulators. Previous calculations have shown that even when the electric field in the SPND insulator reaches 1 GV/m, its effect on electron transport is negligible. Therefore, we assume that the electric field has a negligible effect on electron transport. Based on this assumption, we developed a one-step numerical method for calculating the steady-state electric field in the insulator. This fast method is based on the charge conservation equation in equilibrium, thus requiring only a single charge deposition result without an electric field to obtain the distribution of the electric field under steady-state conditions. We validated our method by comparing it with precise iterative calculation results and experiments. The computational results indicate that, for three typical experimental scenarios, the fast method only takes up to one percent, or even less, of the time required by the iterative method. At the same time, the Mean Relative Error (MRE) in the electric field distribution between the fast method and iterative calculations are 1.48%, 0.31%, and 5.12%, respectively; whereas the relative errors in sensitivity calculations based on the electric field compared to iterative calculations are all less than 2%. Therefore, the fast method can significantly reduce computation time and ensure sufficient accuracy of steady state electric field distribution when the electric field is less than 1 GV/m. The introduction of this method has made it possible to conduct space charge effect assessments and precise sensitivity calculations for the placement of detectors throughout the entire reactor core.

Estimating the synthetic accessibility of molecules with building block and reaction-aware SAScore

Synthetic accessibility prediction is a task to estimate how easily a given molecule might be synthesizable in the laboratory, playing a crucial role in computer-aided molecular design. Although synthesis planning programs can determine synthesis routes, their slow processing times make them impractical for large-scale molecule screening. On the other hand, existing rapid synthesis accessibility estimation methods offer speed but typically lack integration with actual synthesis routes and building block information. In this work, we introduce BR-SAScore, an enhanced version of SAScore that integrates the available building block information (B) and reaction knowledge (R) from synthesis planning programs into the scoring process. In particular, we differentiate fragments inherent in building blocks and fragments to be derived from synthesis (reactions) when scoring synthetic accessibility. Compared to existing methods, our experimental findings demonstrate that BR-SAScore offers more accurate and precise identification of a molecule's synthetic accessibility by the synthesis planning program with a fast calculation time. Moreover, we illustrate how BR-SAScore provides chemically interpretable results, aligning with the capability of the synthesis planning program embedded with the same reaction knowledge and available building blocks.Scientific contributionWe introduce BR-SAScore, an extension of SAScore, to estimate the synthetic accessibility of molecules by leveraging known building-block and reactivity information. In our experiments, BR-SAScore shows superior prediction performance on predicting molecule synthetic accessibility compared to previous methods, including SAScore and deep-learning models, while requiring significantly less computation time. In addition, we show that BR-SAScore is able to precisely identify the chemical fragment contributing to the synthetic infeasibility, holding great potential for future molecule synthesizability optimization.

Fault diagnosis of monoblock centrifugal pumps using pre-trained deep learning models and scalogram images

The monoblock centrifugal pump (MCP) is widely utilized in a diverse range of applications encompassing residential and industrial usage. Sectors such as agriculture, civil projects, mine dewatering, and numerous other industrial applications have employed centrifugal pumps. Despite their extensive usage, these pumps are susceptible to faults and failures due to the presence of critical components that are prone to issues such as bearing faults, sealing problems, cavitation and impeller faults. Therefore, conducting timely fault diagnosis becomes crucial to ensure uninterrupted operation. To address this, the technique of transfer learning, a form of deep learning, is employed. This method entails utilizing prior knowledge from previous operations to improve fault diagnostic performance in monoblock centrifugal pumps. Specifically, scalogram images derived from vibration signals collected during experimental setups were used in fault diagnosis. The study classified faults in monoblock centrifugal pumps using fifteen pre-trained networks including DenseNet-201, GoogLeNet, VGG-19, InceptionResNetV2, Xception, ShuffleNet, VGG-16, InceptionV3, ResNet101, ResNet-50, EfficientNetb0, NasNetmobile, ResNet-18, AlexNet and MobileNet-v2. The highest classification accuracy was obtained by carefully adjusting the hyperparameters which were subsequently employed in the fault classification process. AlexNet, one of the pre-trained network models, showcased remarkable capabilities by achieving a perfect classification accuracy of 100% within a relatively fast computation time of 18 s. This approach employs a reliable and effective process for discovering defects from the start, lowering the risk of possible damage and ensuring the seamless operation of the system.

A new class of efficient high order semi-Lagrangian IMEX discontinuous Galerkin methods on staggered unstructured meshes

In this paper we present a new high order semi-implicit discontinuous-Galerkin (DG) scheme on two-dimensional staggered triangular meshes applied to different nonlinear systems of hyperbolic conservation laws such as advection-diffusion models, incompressible Navier-Stokes equations and natural convection problems. While the temperature and pressure field are defined on a triangular main grid, the velocity field is defined on a quadrilateral edge-based staggered mesh. A semi-implicit time discretization is proposed, which separates slow and fast time scales by treating them explicitly and implicitly, respectively. The nonlinear convection terms are evolved explicitly using a semi-Lagrangian approach, whereas we consider an implicit discretization for the diffusion terms and the pressure contribution. High-order of accuracy in time is achieved using a new flexible and general framework of IMplicit-EXplicit (IMEX) Runge-Kutta schemes specifically designed to operate with semi-Lagrangian methods. To improve the efficiency in the computation of the DG divergence operator and the mass matrix, we propose to approximate the numerical solution with a less regular polynomial space on the edge-based mesh, which is defined on two sub-triangles that split the staggered quadrilateral elements. This allows for a fast computation of the DG operators and matrices without any need to store them for each mesh element. Due to the implicit treatment of the fast scale terms, the resulting numerical scheme is unconditionally stable for the considered governing equations because the semi-Lagrangian approach is the only explicit discretization for convection terms that does not need any stability restriction on the maximum admissible time step. Contrarily to a genuinely space-time discontinuous-Galerkin scheme, the IMEX discretization permits to preserve the symmetry and the positive semi-definiteness of the arising linear system for the pressure that can be solved at the aid of an efficient matrix-free implementation of the conjugate gradient method. We present several convergence results, including nonlinear transport and density currents, up to third order of accuracy in both space and time.

Prediction of Soot in an RQL Burner Using a Semi-Detailed Jeta-1 Chemistry

Abstract The present work proposes a methodology to include accurate kinetics for soot modeling taking into account real fuel complexity in Large Eddy Simulation of aeronautical engines at a reasonable computational cost. The methodology is based on the construction of an analytically reduced kinetic mechanism describing both combustion and gaseous soot precursors growth with sufficient accuracy on selected target properties. This is achieved in several steps, starting from the selection of the detailed kinetic model for combustion and soot precursors growth, followed by the determination of a fuel surrogate model describing the complex real fuel blend. Finally the selected kinetic model is analytically reduced with the code ARCANE while controlling the error on flame properties and soot prediction for the considered fuel surrogate. To perform all evaluation and reduction tests on canonical sooting flames, a Discrete Sectional Model for soot has been implemented in Cantera. The resulting code (Cantera-soot) is now available for the fast calculation of soot production in laminar flames for any fuel. The obtained reduced kinetic scheme is finally validated in a Rich-Quench-Lean burner of the literature in terms of soot prediction capabilities by comparison of LES coupled to the Lagrangian Soot Tracking model with measurements. Results show a significant improvement of the soot level prediction when using the reduced more realistic kinetics, which also allows a more detailed analysis of the soot emission mechanisms. This demonstrates the gain in accuracy obtained with improved reduced kinetics, and validates the methodology to build such schemes.

Differentiable Voronoi Diagrams for Simulation of Cell-Based Mechanical Systems

Navigating topological transitions in cellular mechanical systems is a significant challenge for existing simulation methods. While abstract models lack predictive capabilities at the cellular level, explicit network representations struggle with topology changes, and per-cell representations are computationally too demanding for large-scale simulations. To address these challenges, we propose a novel cell-centered approach based on differentiable Voronoi diagrams. Representing each cell with a Voronoi site, our method defines shape and topology of the interface network implicitly. In this way, we substantially reduce the number of problem variables, eliminate the need for explicit contact handling, and ensure continuous geometry changes during topological transitions. Closed-form derivatives of network positions facilitate simulation with Newton-type methods for a wide range of per-cell energies. Finally, we extend our differentiable Voronoi diagrams to enable coupling with arbitrary rigid and deformable boundaries. We apply our approach to a diverse set of examples, highlighting splitting and merging of cells as well as neighborhood changes. We illustrate applications to inverse problems by matching soap foam simulations to real-world images. Comparative analysis with explicit cell models reveals that our method achieves qualitatively comparable results at significantly faster computation times.

Technical note: A GPU-based shared Monte Carlo method for fast photon transport in multi-energy x-ray exposures.

The Monte Carlo (MC) method is an accurate technique for particle transport calculation due to the precise modeling of physical interactions. Nevertheless, the MC method still suffers from the problem of expensive computational cost, even with graphics processing unit (GPU) acceleration. Our previous works have investigated the acceleration strategies of photon transport simulation for single-energy CT. But for multi-energy CT, conventional individual simulation leads to unnecessary redundant calculation, consuming more time. This work proposes a novel GPU-based shared MC scheme (gSMC) to reduce unnecessary repeated simulations of similar photons between different spectra, thereby enhancing the efficiency of scatter estimation in multi-energy x-ray exposures. The shared MC method selects shared photons between different spectra using two strategies. Specifically, we introduce spectral region classification strategy to select photons with the same initial energy from different spectra, thus generating energy-shared photon groups. Subsequently, the multi-directional sampling strategy is utilized to select energy-and-direction-shared photons, which have the same initial direction, from energy-shared photon groups. Energy-and-direction-shared photons perform shared simulations, while others are simulated individually. Finally, all results are integrated to obtain scatter distribution estimations for different spectral cases. The efficiency and accuracy of the proposed gSMC are evaluated on the digital phantom and clinical case. The experimental results demonstrate that gSMC can speed up the simulation in the digital case by ∼37.8% and the one in the clinical case by ∼20.6%, while keeping the differences in total scatter results within 0.09%, compared to the conventional MC package, which performs an individual simulation. The proposed GPU-based shared MC simulation method can achieve fast photon transport calculation for multi-energy x-ray exposures.

The effect of switching and cycle-to-cycle variations of RRAM on 4-bit encryption/decryption process

The resistive RAM (RRAM) based in-memory computation is a promising technology to overcome the Von-Neumann bottleneck to provide fast and efficient computation. The RRAM is the most appropriate choice for cryptographic applications like encryption/decryption in which the data is computed and stored in the memory itself which enhances the security. The variability issue of RRAM namely switching or device parameter variations and cycle-to-cycle variations deteriorates the functionality of RRAM based circuits. In this paper, the XOR gate with V/R-R logic and a 4-bit encryption/decryption process are implemented using the RRAM Stanford model integrated in the Cadence circuit simulator. The output voltage variations of XOR gate and the encryption/decryption by varying switching and cycle-to-cycle parameters are analyzed. The range of switching parameters of the model that provides the accurate outputs of XOR gate and encryption/decryption is determined.

Molecular Interaction Fields Describing Halogen Bond Formable Areas on Protein Surfaces.

Molecular interaction fields (MIFs) are three-dimensional interaction maps that describe the intermolecular interactions expected to be formed around target molecules. In this paper, a method for the fast computation of MIFs using the approximation functions of quantum mechanics-level MIFs of small model molecules is proposed. MIF functions of N-methylacetamide with chlorobenzene, bromobenzene, and iodobenzene probes were precisely approximated and used to calculate the MIFs on protein surfaces. This method appropriately reproduced halogen-bond-formable areas around the ligand-binding sites of proteins, where halogen bond formation was suggested in a previous study.