Cluster-based generalized multiscale finite element method for elliptic PDEs with random coefficients
Not provided.
Computer Science↗
SEARCH · Search NASA
Results for “Randomized methods”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Evaluating the performance of random forest and iterative random forest based methods when applied to gene expression data
Gene-to-gene networks, such as Gene Regulatory Networks (GRN) and Predictive Expression Networks (PEN) capture relationships between genes and are beneficial for use in downstream biological analyses. There exists multiple network inference tools to produce these gene-to-gene networks from matrices of gene expression data. Random Forest-Leave One Out Prediction (RF-LOOP) is a method that has been shown to be efficient at producing these gene-to-gene networks, frequently known as GEne Network Inference with Ensemble of trees (GENIE3). Random Forest can be replaced in this process by iterative Random Forest (iRF), which performs variable selection and boosting. Here we validate that iterative Random Forest-Leave One Out Prediction (iRF-LOOP) produces higher quality networks than GENIE3 (RF-LOOP). We use both synthetic and empirical networks from the Dialogue for Reverse Engineering Assessment and Methods (DREAM) Challenges by Sage Bionetworks, as well as two additional empirical networks created from Arabidopsis thaliana and Populus trichocarpa expression data.
59 BASIC BIOLOGICAL SCIENCES↗
Evaluating the Performance of Random Forest and Iterative Random Forest Based Methods when Applied to Gene Expression Data
Gene-to-gene networks, such as Gene Regulatory Networks (GRN) and Predictive Expression Networks (PEN) capture relationships between genes and are beneficial for use in downstream biological analyses. There exists multiple network inference tools to produce these gene-to-gene networks from matrices of gene expression data. Random Forest-Leave One Out Prediction (RF-LOOP) is a method that has been shown to be efficient at producing these gene-to-gene networks, frequently known as GEne Network Inference with Ensemble of trees (GENIE3). Here we validate that iterative Random Forest-Leave One Out Prediction (iRF-LOOP) produces higher quality networks than GENIE3. We use both synthetic and empirical networks from the Dialogue for Reverse Engineering Assessment and Methods (DREAM) Challenges by Sage Bionetworks, as well as two additional empirical networks created from Arabidopsis thaliana and Populus trichocarpa expression data.
iRF-Loop, expression network, Populus Trichocarpa↗
Randomized Federated Learning Methods for Nonsmooth, Nonconvex, and Hierarchical Optimization (Final Technical Report)
This final technical report summarizes the outcomes of a DOE-funded project on federated scientific machine learning (FL) under nonsmooth, nonconvex, and hierarchical optimization settings. The project develops new mathematical models, algorithms, and theoretical guarantees for decentralized stochastic, bilevel, and minimax optimization problems arising in DOE mission-relevant applications. A unified framework of randomized and zeroth-order federated optimization methods is introduced, providing provable convergence, communication efficiency, and sample-complexity guarantees. The report documents algorithmic design, theoretical analysis, and empirical validation of the proposed federated learning methods. The project also contributes to workforce development through graduate training and dissemination of results via publications and seminars.
97 MATHEMATICS AND COMPUTING↗
Evaluation of three analytical methods for structures under random acoustic excitation
Evaluation of three methods for determining response analysis of plate and shell structures under random acoustic excitation
Hwang, C.↗
A minimalistic approach to static and dynamic electron correlations: Amending generalized valence bond method with extended random phase approximation correlation correction
A perfect-pairing generalized valence bond (GVB) approximation is known to be one of the simplest approximations, which allows one to capture the essence of static correlation in molecular systems. In spite of its attractive feature of being relatively computationally efficient, this approximation misses a large portion of dynamic correlation and does not offer sufficient accuracy to be generally useful for studying electronic structure of molecules. We propose to correct the GVB model and alleviate some of its deficiencies by amending it with the correlation energy correction derived from the recently formulated extended random phase approximation (ERPA). On the examples of systems of diverse electronic structures, we show that the resulting ERPA-GVB method greatly improves upon the GVB model. ERPA-GVB recovers most of the electron correlation and it yields energy barrier heights of excellent accuracy. Thanks to a balanced treatment of static and dynamic correlation, ERPA-GVB stays reliable when one moves from systems dominated by dynamic electron correlation to those for which the static correlation comes into play.
37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
Dynamic response and input identification of MDOF structures subjected to coupled random vector inputs
A method of random dynamic analysis is presented which is based on the classical approach applied to a discretized structure. The method assumes that the system identification is available in the form of natural modes and frequencies. These modes and frequencies can be found from either available solutions or approximately from finite element programs. A computer program has been developed to perform the computations required in the analysis. All computations are performed with transformed modal variables, which results in significant economy since the number of modal degrees of freedom is almost always less than the number of physical degrees of freedom. The program computes the response to force and base inputs which are statistically coupled. A method is also presented for predicting one of the inputs if the second input and the acceleration response at a point on the structure are known. Finally, results are presented for the random response of a rectangular plate subjected to a random pressure and a random base input. The inputs are considered individually and with various degrees of statistical coupling.
Ocallahan, J. C.↗
Dynamic Loads Generation for Multi-Point Vibration Excitation Problems
A random-force method has been developed to predict dynamic loads produced by rocket-engine random vibrations for new rocket-engine designs. The method develops random forces at multiple excitation points based on random vibration environments scaled from accelerometer data obtained during hot-fire tests of existing rocket engines. This random-force method applies random forces to the model and creates expected dynamic response in a manner that simulates the way the operating engine applies self-generated random vibration forces (random pressure acting on an area) with the resulting responses that we measure with accelerometers. This innovation includes the methodology (implementation sequence), the computer code, two methods to generate the random-force vibration spectra, and two methods to reduce some of the inherent conservatism in the dynamic loads. This methodology would be implemented to generate the random-force spectra at excitation nodes without requiring the use of artificial boundary conditions in a finite element model. More accurate random dynamic loads than those predicted by current industry methods can then be generated using the random force spectra. The scaling method used to develop the initial power spectral density (PSD) environments for deriving the random forces for the rocket engine case is based on the Barrett Criteria developed at Marshall Space Flight Center in 1963. This invention approach can be applied in the aerospace, automotive, and other industries to obtain reliable dynamic loads and responses from a finite element model for any structure subject to multipoint random vibration excitations.
Shen, Lawrence↗
Statistical Evaluation and Improvement of Methods for Combining Random and Harmonic Loads
Structures in many environments experience both random and harmonic excitation. A variety of closed-form techniques has been used in the aerospace industry to combine the loads resulting from the two sources. The resulting combined loads are then used to design for both yield/ultimate strength and high- cycle fatigue capability. This Technical Publication examines the cumulative distribution percentiles obtained using each method by integrating the joint probability density function of the sine and random components. A new Microsoft Excel spreadsheet macro that links with the software program Mathematica to calculate the combined value corresponding to any desired percentile is then presented along with a curve tit to this value. Another Excel macro that calculates the combination using Monte Carlo simulation is shown. Unlike the traditional techniques. these methods quantify the calculated load value with a consistent percentile. Using either of the presented methods can be extremely valuable in probabilistic design, which requires a statistical characterization of the loading. Additionally, since the CDF at high probability levels is very flat, the design value is extremely sensitive to the predetermined percentile; therefore, applying the new techniques can substantially lower the design loading without losing any of the identified structural reliability.
Brown, A. M.↗
Statistical Comparison and Improvement of Methods for Combining Random and Harmonic Loads
Structures in many environments experience both random and harmonic excitation. A variety of closed-form techniques has been used in the aerospace industry to combine the loads resulting from the two sources. The resulting combined loads are then used to design for both yield ultimate strength and high cycle fatigue capability. This paper examines the cumulative distribution function (CDF) percentiles obtained using each method by integrating the joint probability density function of the sine and random components. A new Microsoft Excel spreadsheet macro that links with the software program Mathematics is then used to calculate the combined value corresponding to any desired percentile along with a curve fit to this value. Another Excel macro is used to calculate the combination using a Monte Carlo simulation. Unlike the traditional techniques, these methods quantify the calculated load value with a Consistent percentile. Using either of the presented methods can be extremely valuable in probabilistic design, which requires a statistical characterization of the loading. Also, since the CDF at high probability levels is very flat, the design value is extremely sensitive to the predetermined percentile; therefore, applying the new techniques can lower the design loading substantially without losing any of the identified structural reliability.
Brown, Andrew M.↗
MULTILEVEL ACCELERATION OF STOCHASTIC COLLOCATION METHODS FOR PDE WITH RANDOM INPUT DATA
Stochastic Collocation (SC) methods for stochastic partial differential equa- tions (SPDEs) suffer from the curse of dimensionality, whereby increases in the stochastic dimension cause an explosion of computational effort. To combat these challenges, multilevel approximation methods seek to decrease computational complexity by balancing spatial and stochastic discretization errors. As a form of variance reduction, multilevel techniques have been successfully applied to Monte Carlo (MC) methods, but may be extended to accelerate other methods for SPDEs in which the stochastic and spatial degrees of freedom are de- coupled. This article presents general convergence and computational complexity analysis of a multilevel method for SPDEs, demonstrating its advantages with regard to standard, single level approximation. The numerical results will highlight conditions under which multilevel sparse grid SC is preferable to the more traditional MC and SC approaches.
Webster, Clayton G↗
Analytic method for calculating properties of random walks on networks
A method for calculating the properties of discrete random walks on networks is presented. The method divides complex networks into simpler units whose contribution to the mean first-passage time is calculated. The simplified network is then further iterated. The method is demonstrated by calculating mean first-passage times on a segment, a segment with a single dangling bond, a segment with many dangling bonds, and a looplike structure. The results are analyzed and related to the applicability of the Einstein relation between conductance and diffusion.
Goldhirsch, I.↗
Fusion Imaging to Guide Thoracic Endovascular Aortic Repair (TEVAR): A Randomized Comparison of Two Methods, 2D/3D Versus 3D/3D Image Fusion
Purpose: To compare the accuracy of two-dimensional (2D) versus three-dimensional (3D) image fusion for thoracic endovascular aortic repair (TEVAR) image guidance. Materials and Methods: Between December 2016 and March 2018, all eligible patients who underwent TEVAR were prospectively included in a single-center study. Image fusion methods (2D/3D or 3D/3D) were randomly assigned to guide each TEVAR and compared in terms of accuracy, dose area product (DAP), volume of contrast medium injected, fluoroscopy time and procedure time. Results: Thirty-two patients were prospectively included; 18 underwent 2D/3D and 14 underwent 3D/3D TEVAR. The 3D/3D method allowed more accurate positioning of the aortic mask on top of the fluoroscopic images (proximal landing zone error vector: 1.7 ± 3.3 mm) than was achieved by the 2D/3D method (6.1 ± 6.1 mm; p = 0.03). The 3D/3D image fusion method was associated with significantly lower DAP than the 2D/3D method (50.5 ± 30.1 Gy cm{sup 2} for 3D/3D vs. 99.5 ± 79.1 Gy cm{sup 2} for 2D/3D; p = 0.03). The volume of contrast medium injected was significantly lower for the 3D/3D method than for the 2D/3D method (50.6 ± 22.9 ml vs. 98.4 ± 47.9 ml; p = 0.002). Conclusion: Higher image fusion accuracy and lower contrast volume and irradiation dose were observed for 3D/3D image fusion than for 2D/3D during TEVAR. Level of Evidence: II, Randomized trial.
62 RADIOLOGY AND NUCLEAR MEDICINE↗
Extracting Damping Ratio from Dynamic Data and Numerical Solutions
There are many ways to extract damping parameters from data or models. This Technical Memorandum provides a quick reference for some of the more common approaches used in dynamics analysis. Described are six methods of extracting damping from data: the half-power method, logarithmic decrement (decay rate) method, an autocorrelation/power spectral density fitting method, a frequency response fitting method, a random decrement fitting method, and a newly developed half-quadratic gain method. Additionally, state-space models and finite element method modeling tools, such as COMSOL Multiphysics (COMSOL), provide a theoretical damping via complex frequency. Each method has its advantages which are briefly noted. There are also likely many other advanced techniques in extracting damping within the operational modal analysis discipline, where an input excitation is unknown; however, these approaches discussed here are objective, direct, and can be implemented in a consistent manner.
Casiano, M. J.↗
Real-Time Adaptive Drag Minimization Wind Tunnel Investigation of a Flexible Wing with Variable Camber Continuous Trailing Edge Flap System
This paper reports the results of a recently completed real-time adaptive drag minimization wind tunnel investigation of a highly flexible wing wind tunnel model equipped with the Variable Camber Continuous Trailing Flap (VCCTEF) technology at the University of Washington Aeronautical Laboratory (UWAL). The wind tunnel investigation is funded by NASA SBIR Phase II contract with Scientific Systems Company, Inc. (SSCI) and University of Washington (UW) as a subcontractor. The wind tunnel model is a sub-scale Common Research Model (CRM) wing constructed of foam core and fiberglass skin and is aeroelastically scaled to achieve a wing tip deflection of 10% of the wing semi-span which represents a typical wing tip deflection for a modern transport such as Boeing 787. The jig-shape twist of the CRM wing is optimized using a CART3D aero-structural model to achieve the minimum induced drag for the design cruise lift coefficient of 0.5. The wing is equipped with two chord wise cambered segments for each of the six span wise flap sections for a total of 12 individual flap segments that comprise the VCCTEF system. Each of the 12 flap segments is actively controlled by an electric servo-actuator. The real-time adaptive drag optimization strategy includes an on-board aerodynamic model identification, a model excitation, and a real-time drag optimization. The on-board aerodynamic model is constructed parametrically as a function of the angle of attack and flap positions to model the lift and drag coefficients of the wing. The lift coefficient models include a linear model and a second-order model. The drag coefficient models include a quadratic model and a higher-order up to 6th-order model to accurately model the drag coefficient at high angles of attack. The onboard aerodynamic model identification includes a recursive least-squares (RLS) algorithm and a batch least-squares (BLS) algorithm designed to estimate the model parameters. The model excitation method is designed to sample the input set that comprises the angle of attack and the flap positions. Three model excitation methods are developed: random excitation method, sweep method, and iterative angle-of-attack seeking method. The real-time drag optimization includes a generic algorithm developed by SSCI and several optimization methods developed by NASA which include a second-order gradient Newton-Raphson optimization method, an iterative gradient optimization method, a pseudo-inverse optimization method, an analytical optimization method, and an iterative refinement optimization method. The first wind tunnel test entry took place in September 2017. This test revealed major hardware issues and required further redesign of the flap servo mechanisms. The second test entry took place in April 2018. However, the test was not successful due to the issues with the onboard aerodynamic model identification RLS algorithm which incorrectly identified model parameters. This test also provides an experimental comparison study between the VCCTEF and a variable camber discrete trailing edge flap (VCDTEF) without the elastomer transition mechanisms. The experimental result confirms the benefit of the VCCTEF which produces lower drag by 5% than the VCDTEF. The third and final test entry took place in June 2018 after the issues with the RLS algorithm have been identified and corrected. Additional improvements were implemented. These include the BLS algorithm, the iterative angle-of-attack seeking method, the iterative gradient optimization method, and the pseudo-inverse optimization method. The test objectives were successfully demonstrated as the real-time drag optimization identifies several optimal solutions at off-design lift coefficients. The iterative gradient optimization method is found to achieve up to 4.7% drag reduction for the off-design lift coefficient of 0.7. The pseudo-inverse optimization method which does not require the drag coefficient model is found to be quite effective in reducing drag. Up to 9.4% drag reduction for the off-design lift coefficient of 0.7 is achieved with the pseudo-inverse optimization method. The wind tunnel investigation demonstrates the potential of real-time drag optimization technology. Several new capabilities are developed that could enable future adaptive wing technologies for flexible wings equipped with drag control devices such as the VCCTEF.
Drag Optimization↗
Applications of Finite Energy Methods to Transient, Random, and Shock Predictions
Explore the source record for details and available documents.
Daniel Kaufman↗
Quantifying the Effect of Lidar Turbulence Error on Wind Power Prediction
Currently, cup anemometers on meteorological towers are used to measure wind speeds and turbulence intensity to make decisions about wind turbine class and site suitability; however, as modern turbine hub heights increase and wind energy expands to complex and remote sites, it becomes more difficult and costly to install meteorological towers at potential sites. As a result, remote-sensing devices (e.g., lidars) are now commonly used by wind farm managers and researchers to estimate the flow field at heights spanned by a turbine. Although lidars can accurately estimate mean wind speeds and wind directions, there is still a large amount of uncertainty surrounding the measurement of turbulence using these devices. Errors in lidar turbulence estimates are caused by a variety of factors, including instrument noise, volume averaging, and variance contamination, in which the magnitude of these factors is highly dependent on measurement height and atmospheric stability. As turbulence has a large impact on wind power production, errors in turbulence measurements will translate into errors in wind power prediction. The impact of using lidars rather than cup anemometers for wind power prediction must be understood if lidars are to be considered a viable alternative to cup anemometers.In this poster, the sensitivity of power prediction error to typical lidar turbulence measurement errors is assessed. Turbulence estimates from a vertically profiling WINDCUBE v2 lidar are compared to high-resolution sonic anemometer measurements at field sites in Oklahoma and Colorado to determine the degree of lidar turbulence error that can be expected under different atmospheric conditions. These errors are then incorporated into a power prediction model to estimate the sensitivity of power prediction error to turbulence measurement error. Power prediction models, including the standard binning method and a random forest method, were developed using data from the aeroelastic simulator FAST for a 1.5 MW turbine. The impact of lidar turbulence error on the predicted power from these different models is examined to determine the degree of turbulence measurement accuracy needed for accurate power prediction.
17 WIND ENERGY↗