Fast Performance Evaluation Research Articles

Linear and nonlinear analytical equations for fast design of three types of triangular-amplified compliant mechanisms

Compliant amplification mechanisms based on the triangular principle have attracted considerable applications in precision and other engineering fields due to their compactness and efficient amplification capacity. However, a fast engineering design for geometric nonlinearity of large strokes is still much challenging. We report herein a series of analytical equations of nonlinear amplification ratio, input and output stiffness for three commonly-used triangular-amplified compliant mechanisms, namely the rhombus, diamond and bridge types. The pronounced geometric nonlinearities of axially-loaded stiffening and kinematics-arching effects are explicitly formulated. The proposed nonlinear formulas can be directly degenerated as the linear models by vanishing the nonlinear terms, which enables a comprehensive analysis of linear and nonlinear kinetostatics, and hence offers a straightforward way for the fast performance evaluation and size synthesis/optimization. This relieves an engineer's experience and knowledge for an ab initio modeling process. The prediction error is discussed and some insights into the linear and nonlinear characteristics of the three commonly-used triangular-amplified compliant mechanisms are outlined that confirms the advantage of analytical equations bringing to a fast engineering analysis and design.

Evaluation of stochastic flow lines with provisioning of auxiliary material

Flow lines are often used to perform assembly operations in multi-stage processes. During these assembly operations, components that are relatively small, compared to the work pieces travelling down the flow line, are mounted to the work pieces at a given stage. Those components, or more generally, any kind of auxiliary material, are provisioned to the corresponding production stage in a repetitive but not necessarily deterministic manner using a certain delivery frequency, each time filling the local storage up to a predetermined order-up-to level. Just like random processing times, machine failures, and repairs, the randomness of the provisioning process can impact the long-term throughput of such a flow line. In this paper, we develop a fast and accurate analytical performance evaluation method to estimate the long-term throughput of a Markovian flow line of this type for the practically important case of limited buffer capacities between the production stages. We first give an exact characterization of a two-machine line of that type and show how to determine system state probabilities and aggregate performance measures. Furthermore, we show how to use this two-machine model as the building block of an approximate decomposition approach for longer flow lines. As opposed to previous decomposition approaches, even the state space of the two-machine lines can become so large that an exact solution of the Markov chains can become impractical. We hence show how to set up, train, and use an artificial neural network to replace the Markov chain solver embedded in the decomposition approach, which then leads to an accurate and extremely fast flow line evaluation tool. The proposed methodology is evaluated by a comparison with simulation results and used to characterize the structural patterns describing the behaviour of flow lines of this type. The method can be used to systematically consider the combined impact of the delivery frequency and the local order-up-to levels for the auxiliary material when designing a flow line of this type.

Fast Stray Light Performance Evaluation Based on BSDF and Radiative Transfer Theory.

Evaluating the stray light cancellation performance of an optical system is an essential step in the search for superior optical systems. However, the existing evaluation methods, such as the Monte Carlo method and the ray tracing method, suffer from the problems of vast arithmetic and cumbersome processes. In this paper, a method for a rapid stray light performance evaluation model and quantitatively determining high-magnitude stray light outside the field of view are proposed by adopting the radiative transfer theory based on the scattering property of the bidirectional scattering distribution function (BSDF). Under the global coordinates, based on the derivation of the light vector variation relationship in the near-linear system, the specific structural properties of the off-axis reflective optical system, and the specular scattering properties, a fast quantitative evaluation model of the optical system's stray light elimination capability is constructed. A loop nesting procedure was designed based on this model, and its validity was verified by an off-axis reflective optical system. It successfully fitted the point source transmittance (PST) curve in the range of specular radiation reception angles and quantitatively predicted the prominence due to incident stray light outside the field of view. This method does not require multiple software to work in concert and requires only 10-5 orders of magnitude of computing time, which is suitable for the rapid stray light assessment and structural screening of off-axis reflective optical systems with a good symmetry. The method is promising for improving imaging radiation accuracy and developing lightweight space cameras with low stray light effects.

Production Forecasting of Unruly Geoenergy Extraction Wells Using Gaussian Decline Curve Analysis

Fast and rigorous well performance evaluation is made possible by new solutions of the pressure diffusion equation. The derived Gaussian pressure transient (GPT) solutions can be practically formulated as a decline curve analysis (DCA) equation for history matching of historic well rates to then forecast the future well performance and estimate the remaining reserves. Application in rate transient analysis (RTA) mode is also possible to estimate fracture half-lengths. Because GPT solutions are physics-based, these can be used for production forecasting as well as in reservoir simulation mode (by computing the spatial and temporal pressure gradients everywhere in the reservoir section drained by either an existing or a planned well). The present paper focuses on the physics-based production forecasting of so-called “unruly” wells, which at first seem to have production behavior noncompliant with any DCA curve. Four shale wells (one from the Utica, Ohio; one from the Eagle Ford Formation, East Texas; and two from the Wolfcamp Formation, West Texas) are analyzed in detail. Physics-based adjustments are made to the Gaussian DCA history matching process, showing how the production rate of these wells is fully compliant with the rate implied by the hydraulic diffusivity of the reservoir sections where these wells drain from.

Multistage Op Amp design space exploration by gm/ID sampling and symbolic design equations

SummaryMultiple‐stage complementary metal‐oxide‐semiconductor (CMOS) operational amplifiers (Op Amps) have been extensively studied in the literature; often, each new design reports several performance advancements over the existing ones. However, the design space boundaries of those new proposals were rarely explored. Predicting the design space boundary of a new circuit topology is usually challenging due to the dependency on many aspects of technical consideration regarding the process technology, sizing priority adopted in practice, and balancing of performance metrics on power, area, noise immunity, and so on. Knowing the design space boundary could be beneficial to a fair appraisal of new designs. In this paper, we propose a gm/ID‐based method for the design space exploration of multistage Op Amps. By introducing a sampling technique while taking advantage of the device grouping property with gm/ID, it is possible to explore the circuit performance boundary by combining approximate symbolic modeling and gm/ID table lookup. Due to the introduction of a fast performance evaluation method, a large ensemble of circuit sizing samples can be evaluated in parallel, by which an efficient data mining procedure could be incorporated to deduce the circuit samples that can achieve extremal performance at Pareto boundaries. The fidelity of this proposal has been validated via simulation program with integrated circuit emphasis (SPICE) . The sampling method is also compared with a multi‐objective genetic algorithm to show the superiority in capturing Pareto boundaries.

Toward a Fast but Reliable Energy Performance Evaluation Method for Existing Residential Building Stock

Building a reliable energy model for old residential buildings with insufficient documentation and user assistance is a challenging and time-consuming task. Nevertheless, the ambitious European decarbonization targets require this building stock to be renovated, making energy assessment a key priority. In line with this goal, the following study explores a more simplified and automatic framework to generate a residential building energy model (BEM). The paper’s approach is based on the concept of urban building energy modelling (UBEM) archetypes or building prototypes and is customized according to the principles of dynamic simulations performed in the existing BEM software, Integrated Environmental Solutions Virtual Environment IES VE, and Solemma Open Studio. Therefore, based on three real starting inputs, a prototype database (DB) of assigned inputs is generated, i.e., an input matrix, using Google Maps as a geometry source. Other data are drawn from tabular DB. The proposed approach is evaluated by benchmarking the simulation results with precise models and monitoring the data that come from the Horizon2020 project REZBUILD. Nevertheless, a level of simplification is introduced that creates less accurate results for total or system-level energy consumption; this is compensated for using a set of simple calibration steps. The approach gives promising results for daily indoor temperature, making it a suitable indicator for evaluating further retrofitting alternatives.

When Database Meets New Storage Devices: Understanding and Exposing Performance Mismatches via Configurations

NVMe SSD hugely boosts the I/O speed, with up to GB/s throughput and microsecond-level latency. Unfortunately, DBMS users can often find their high-performanced storage devices tend to deliver less-than-expected or even worse performance when compared to their traditional peers. While many works focus on proposing new DBMS designs to fully exploit NVMe SSDs, few systematically study the symptoms, root causes and possible detection methods of such performance mismatches on existing databases. In this paper, we start with an empirical study where we systematically expose and analyze the performance mismatches on six popular databases via controlled configuration tuning. From the study, we find that all six databases can suffer from performance mismatches. Moreover, we conclude that the root causes can be categorized as databases' unawareness of new storage devices characteristics in I/O size, I/O parallelism and I/O sequentiality. We report 17 mismatches to developers and 15 are confirmed. Additionally, we realize testing all configuration knobs yields low efficiency. Therefore, we propose a fast performance mismatch detection framework and evaluation shows that our framework brings two orders of magnitude speedup than baseline without sacrificing effectiveness.

A New Fast Deterministic Economic Dispatch Method and Statistical Performance Evaluation for the Cascaded Short-Term Hydrothermal Scheduling Problem

The Cascaded Short-Term Hydrothermal Scheduling (CSTHTS) problem is a highly non-linear, multi-modal, non-convex, and NP-hard optimization problem that has been solved by conventional and metaheuristic algorithms in the past. As the CSTHTS problem falls under the category of applied operational research, therefore, the work is still on-going to find new algorithms and variants of the existing algorithms that would better approximate the optimal global solution in a shorter computational time. This article proposes a novel deterministic thermal economic dispatch method embedded with the improved Accelerated Particle Swarm Optimization (APSO) algorithm to infinitesimally reduce the Big O time complexity for the standard benchmark test case of the CSTHTS optimization problem. Then, it discusses and presents the importance of performing standard statistical tests to establish the supremacy of one metaheuristic algorithm over the other one in solving the CSTHTS problem. The results obtained are better than the results of the many state-of-the-art algorithms applied to solve the considered test case of the CSTHTS problem in the literature, and the superiority of the improved APSO algorithm has been established statistically using the parametric independent samples t-test and the non-parametric Mann–Whitney U-test over the other metaheuristic algorithms such as particle swarm optimization in solving the chosen test case of the CSTHTS problem.

Measurement and Prediction of Micronewton Class Thrust of Electric Propulsion Based on the Torsional Pendulum and Machine Learning Technique

Accurate and direct measurement of thrust is the most fundamental requirement in the evaluation of electric thruster performance. This article performed the measurement and prediction of the micronewton class thrust of micro gridded ion thrusters (μGITs) based on a novel design of torsional pendulum and high precision neural network models. The developed torsional pendulum adopts specially designed liquid metal connectors to eliminate the interference caused by the power supply cables and propellant feeding lines, realizing an accurate thrust measurement of radio frequency (RF)- and dc-μGIT from micronewton level to millinewton level with a measuring range of 0–10 mN and a resolution of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$6.4~\mu \text{N}$ </tex-math></inline-formula> . Results show that the RF-μGIT generates a thrust of 0.46–2.43 mN for the RF power from 100 to 120 W and the xenon flow rate from 0.75 to 1.50 sccm. The dc-μGIT generates a thrust of 0.016–0.310 mN for the acceleration voltage from 700 to 1050 V and the flow rate from 0.2 to 2.0 sccm. Both the artificial neural network (ANN) and the radial basis function neural network (RBF-NN) are adopted to predict the relationship between the thrust and input parameters of RF-μGIT. The predicted thrusts by ANN deviate about 10% from the measured data, and the maximum relative deviation using RBF-NN model is about 2%. Furthermore, the RBF-NN model is used to obtain the distribution maps of thrust and specific impulse under 600 different working conditions and provides a solution for intelligent regulation of thruster performance. For the first time, the combination of thrust measurement and machine learning (ML) provides a new approach for the fast performance evaluation, prediction and regulation of electric thrusters.

Fast Performance Evaluation for Switched-Capacitor Converter Based on Convolutional Neural Networks

Switched-capacitor converters (SCCs) have been developed for decades and applied successfully in many applications thanks to their merits of high power density, high efficiency, and low cost. Existing SCC evaluation methods do not provide an efficient and effective way to estimate their performance. This paper proposes a new evaluation performance method for SCCs that transforms the topology optimization issue into a more typical non-linear regression problem. The SCC topologies are described by the adjacent matrix. A CNN model is then selected to solve the regression problem. Experimental results confirm that this CNN model greatly shortens the calculation time while performing better than other machine learning algorithms.

Rotor Slot Design of Squirrel Cage Induction Motors With Improved Rated Efficiency and Starting Capability

Among the electro-mechanical devices transforming energy from electrical to mechanical, the squirrel cage induction motor can be surely considered a workhorse of the industry due to its robustness, low cost and good performance when directly fed by the a.c. grid. Being the most influencing motor topology in terms of energy consumption, optimizing the efficiency of squirrel cage induction motors could lead to a great impact towards the reduction of the human environmental footprint. The induction motor design aided by finite element analysis presents significant challenges because an accurate performance prediction requires a considerable computational burden. This paper makes use of an innovative fast and accurate performance evaluation method embedded into an automatic design procedure to optimize different rotor slot geometries. After introducing the performance estimation approach, its advantages and limits are discussed comparing its prediction with the experimental tests carried out on an off-the-shelf induction motor. Different rotor cage structures with increasing geometrical complexity are then optimized in terms of starting and rated performance adopting the same design optimization process, the same stator geometry and constituent materials. The analysis of the optimal solutions shows how it is possible to improve the rated efficiency without compromising other performance indexes. The presented results can be used as general design guidelines of squirrel cage induction motors for industrial applications.

An efficient multi-fidelity simulation method for adaptive cycle engine ejector nozzle performance evaluation

The three-bypass duct structure of Adaptive Cycle Engine (ACE) provides unique performance advantages while also challenging the design of its exhaust system. The separated exhaust nozzles widely adopted in current research may not meet the performance requirement of ACE. The feature of combining two exhaust streams to discharge makes the ejector nozzle suitable as the exhaust system of ACE. However, due to the connection between the third bypass duct and the ejector nozzle, the ejector nozzle and ACE form a highly coupled system, thus it is necessary to perform an integrated simulation of the ejector nozzle and ACE. The multi-fidelity simulation could offer accurate performance simulation as well as consider the complicated component matching working relations. In this paper, a multi-fidelity simulation method for ACE ejector nozzle simulation is developed for fast and accurate performance evaluation, and the design requirements of ACE ejector nozzle are initially explored. A multi-fidelity joint simulation method for the (0-D) ACE performance model and the (2-D) ejector nozzle Computational Fluid Dynamics (CFD) model is constructed through a dynamically optimized and updated surrogate model. The result shows, when ensuring the ejector nozzle does not affect the ACE working point, the peak gross thrust coefficient occurs in subsonic cruise condition as 0.9895. The gross thrust coefficients of acceleration conditions decrease with the increase of flight Mach number due to the increase of shroud divergent angle. Only 350 times CFD simulations are required to complete multi-fidelity simulations of 29 ACE operating points (As a comparison, the de-coupled method requires about 105 times CFD simulations for the whole characteristics, and the fully integrated method requires about 100 times CFD simulations for one ACE operating point). The diameter of the ejector shroud shoulder must be variable, otherwise it will cause a maximum thrust loss of 22.6% at dry acceleration condition and fan surge at cruise and reheat acceleration conditions. This method can further realize the evaluation of the installed performance and the control law optimal design of the ACE ejector nozzle.

Fast explosive performance prediction via small-dose energetic materials based on time-resolved imaging combined with machine learning

Fast, reproducible, and quantitative performance evaluation of monomolecular energetic materials (EMs) is a significant challenge that limits the tailored applications of EMs and the development of new high-energy-density materials.

Optimization of solar receivers for high-temperature solar conversion processes: Direct vs. Indirect illumination designs

High-temperature solar receivers are core components in concentrated solar energy utilization systems. The rational design of solar receivers relies on the fundamental understanding of coupled physics across various scales. We developed a numerical model to assess the solar thermal conversion behaviour under various porous structure designs, material choices, geometrical designs, as well as operational conditions. The 3D model solves a coupled set of equations accounting for mass, momentum, and energy conservations in the cavity and porous media domains. We have optimized the radiative transfer model to resolve the participating behaviour only in the porous domain and non-participating behaviour only for cavity and window surfaces. This cost-effective model enables fast performance evaluation for various design parameters. The windowless receiver, in general, shows higher solar-to-thermal efficiency compared to the window receiver. This superiority in optical efficiency for the windowless case can be surpassed by the window case at high incident solar power with a fixed outlet temperature. For porous media structure, both window and windowless receivers are optimized by the porosity and pore diameter. The solar-to-thermal efficiency was slightly influenced by the cavity inclination. Large porous layer thickness leads to reduced performance for the windowless receivers, while the efficiency reaches a plateau for the window receiver at large thickness. We have provided a comprehensive modeling framework for the design of solar receivers for high-temperature applications using concentrated solar energy which can be easily adapted for receiver under a wide range of designs and operation conditions.

TileNET: Hardware accelerator for ternary Convolutional Neural Networks

Convolutional Neural Networks (CNNs) are popular in Advanced Driver Assistance Systems (ADAS) for camera perception. The versatility of the algorithm makes it applicable in multiple applications like object detection, lane detection and semantic segmentation. For image processing to be viable in driver assistance systems, the throughput requirement ranges in the order of a few tens of TeraMACs per second (TMACs). In addition, high accuracy levels of image detection and recognition cannot be compromised for the need for high throughput.In this paper, we present TileNET, a novel tiled architecture for ternary-weighted CNNs. TileNET is modular and scalable across variations in network organization and device configurations. Two modes of the implementation are presented, viz., systolic and streaming. A high-level estimation technique has been developed that facilitates fast performance evaluation through design space exploration among a range of target devices and varying CNN models.Performance has been verified for area and throughput estimation for Xilinx Virtex, Artix, Kintex and Zynq devices. TileNET implemented on Virtex-7 (XC7VX1140T) results in a throughput of about 16 Tera-operations per second (TOPs) for LeNet, AlexNet, ResNet-50 and VGG-16. In addition, the 45nm standard cell implementation of TileNet shows a throughput of about 30 TOPs respectively.

Fast Screening Method for Nitrogen Reduction Reaction (NRR) Electrocatalytic Activity with Rotating Ring‐Disc Electrode (RRDE) Analysis in Alkaline Environment

AbstractThe Haber‐Bosch process for NH3 production leads to a considerable greenhouse gas release due to the remarkable use of fossil fuels. Therefore, there is an increasing interest in developing alternative and environmental friendly approaches. Among the possible solutions, the electrocatalytic conversion of N2 has recently gained significant attention; on the other hand, not only scientific but also important technical aspects remain fundamental issues to be clarified. Particularly relevant is the need to improve the analytical protocols to ascertain that any detected NH3 is actually produced from N2 rather than from any external contaminations or partial decomposition of the catalyst itself. Here, a rotating ring‐disc electrode (RRDE) setup is used for the first time to study the N2 electroreduction process with the aim to recognize the product species formed at the disc and detected at the ring electrodes, respectively. We demonstrated that this experimental approach is effective to discern also a low‐level ammonium concentration through monitoring the ammonia oxidation peak at the ring electrode for a fast and preliminary electrocatalytic performance evaluation and to prevent false positives. The versatility of the RRDE method employed as a fingerprint of new electrocatalyst candidates could allow to reserve time and cost.

Artificial Neural Network Modeling of Cyclic Steam Injection Process in Naturally Fractured Reservoirs

Summary The enhanced oil recovery (EOR) technology can be instrumental in achieving the maximum rate of return from a hydrocarbon reservoir. One of the widely implemented EOR methodologies is the cyclic steam injection (CSI) which is a thermal recovery process aiming to reduce the oil viscosity and increase the production in naturally fractured heavy-oil reservoirs. However, commercial software used for CSI modeling can be difficult to learn and implement, also can be time-consuming and costly. This paper describes three artificial neural network (ANN) based models that have been trained for accurate and fast CSI performance evaluation with easy implementation. In this study, to model the CSI process, a commercial numerical model is used synchronously with the ANN models. Our goal is to discuss smart proxy models’ mimicking ability of the commonly used numerical models. Three ANN models have been trained with different network topologies, and transfer functions using a data set consists of 1,428 cases: Forward model: to predict performance indicators and viscosity contours. Inverse Model 1: to predict CSI design parameters. Inverse Model 2: to predict significant reservoir properties. The results from the trained ANN models have been compared with the results generated by the available commercial software. It was observed that ANN models can provide the results within a few seconds while it takes more than 30 minutes in some cases for the commercial software. The computational time for the numerical model being extensive, the number of trials to be conducted to find the optimum parameters for the CSI operations can be prohibitively expensive. The trained ANN-based models are capable of providing results within a rather low error margin. The developed ANN-based models are controlled by a user-friendly graphical user interface (GUI), which decreases the time expended on learning and executing the software.

An energy-based approach for kinetostatic modeling of general compliant mechanisms

Kinetostatic analysis for general compliant mechanisms requires simultaneous considerations of kinematic and elasto-mechanical behaviors. Therefore, it will be difficult and low-efficiency to design and analyze general compliant mechanisms. In this paper, a kinetostatic modeling approach that integrates the screw theory with the energy method is proposed to provide an accurate and efficient solution. The coordinate transformations of stiffness and compliance matrices are deduced by using the screw theory. Then, the kinetostatic modeling of general compliant mechanisms is presented based on the energy method and the screw theory. The advantage of the proposed modeling approach is that there is no need to solve the laborsome equilibrium equations of nodal force when it is applied to the kinetostatic analysis of general compliant mechanisms. Performance comparisons of the proposed approach with the matrix displacement approach and the finite element analysis are respectively conducted for three exemplary mechanisms to reveal the prediction accuracy. The results indicate that the proposed modeling approach is applicable for fast performance evaluation of general compliant mechanisms at the early design stage.

Model-based platform for design space exploration in single-microring-based silicon photonic interconnects on chip

A large number of optical routers have been proposed for optical network-on-chips (ONoCs). The performance of optical routers is usually evaluated by measurement after fabrication or simulation before fabrication. In simulations, a common method is to set the component performance, such as insertion loss and crosstalk, to fixed values. The performance of optical routers is obtained by combining these values. In this way, the design parameters are not taken into consideration for evaluation. In this work, we propose and establish a model-based platform for single-microring(MR)-based silicon photonic interconnect on chip, taking the impact of design parameters into consideration. The application of our platform is characteristic of fast and effective performance evaluation. We study the influence of the design parameters, and evaluate the performance of MR-based optical routers and ONoCs on the platform, including insertion loss, crosstalk, MR tuning costs, the required laser power and BER. The simulation results provide a more accurate reference for router design before fabrication

Single-Trial Decoding of Scalp EEG under Natural Conditions.

There is significant current interest in decoding mental states from electroencephalography (EEG) recordings. EEG signals are subject-specific, are sensitive to disturbances, and have a low signal-to-noise ratio, which has been mitigated by the use of laboratory-grade EEG acquisition equipment under highly controlled conditions. In the present study, we investigate single-trial decoding of natural, complex stimuli based on scalp EEG acquired with a portable, 32 dry-electrode sensor system in a typical office setting. We probe generalizability by a leave-one-subject-out cross-validation approach. We demonstrate that support vector machine (SVM) classifiers trained on a relatively small set of denoised (averaged) pseudotrials perform on par with classifiers trained on a large set of noisy single-trial samples. We propose a novel method for computing sensitivity maps of EEG-based SVM classifiers for visualization of EEG signatures exploited by the SVM classifiers. Moreover, we apply an NPAIRS resampling framework for estimation of map uncertainty, and thus show that effect sizes of sensitivity maps for classifiers trained on small samples of denoised data and large samples of noisy data are similar. Finally, we demonstrate that the average pseudotrial classifier can successfully predict the class of single trials from withheld subjects, which allows for fast classifier training, parameter optimization, and unbiased performance evaluation in machine learning approaches for brain decoding.