Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “simulation modeling training analysis”

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.

At least 595 records · Page 33
All sourcesNASA NTRSDOE OSTI
NASA NTRS2017

Modeling Microgravity Induced Fluid Redistribution Autoregulatory and Hydrostatic Enhancements

Space flight induces a marked cephalad (headward) redistribution of blood and interstitial fluid potentially resulting in a loss of venous tone and reduction in heart muscle efficiency upon introduction into the microgravity environment. Using various types of computational models, we are investigating how this fluid redistribution may induce intracranial pressure changes, relevant to reported reductions in astronaut visual acuity, part of the Visual Impairment and Intracranial Pressure (VIIP) syndrome. Methods: We utilize a lumped parameter cardiovascular system (CVS) model, augmented by compartments comprising the cerebral spinal fluid (CSF) space, as the primary tool to describe how microgravity, and the associated lack of hydrostatic gradient, impacts fluid redistribution. Models of ocular fluid pressures and biomechanics then accept the output of the above model as boundary condition input to allow more detailed, local analysis (see IWS Abstract by Ethier et al.). Recently, we enhanced the capabilities our previously reported CVS model through the implementation of robust autoregulatory mechanisms and a more fundamental approach to the implementation of hydrostatic mechanisms. Modifying the approach of Blanco et al., we implemented auto-regulation in a quasi-static manner, as an averaged effect across the span of one heartbeat. This approach reduced the higher frequency perturbations from the regulatory mechanism and was intended to allow longer simulation times (days) than models that implement within-beat regulatory mechanisms (minutes). A more fundamental approach to hydrostatics was implemented by a quasi-1D approach, in which compartment descriptions include compartment length, orientation and relative position, allowed for modeling of body orientation, relative body positioning and, in the future, alternative gravity environments. At this time the inclusion of hydrostatic mechanisms supplies additional capabilities to train and validate the CVS model with terrestrial data. Results and Conclusions: With the implementation of auto-regulation and hydrostatic modeling capabilities, the model performs as expected in the maintaining the CA (Central Artery) compartment pressure when simulating orientations ranging from supine to standing. The model appears to generally overpredict heart rate and thus cardiac output, possibly indicating sensitivity to the nominal heart rate, which is used as an initial set point of the regulation mechanisms. Despite this sensitivity, the model performs consistently for many hours of simulation time, indicating the success of our quasi-static implementation approach.

biomechanics↗
NASA NTRS2016

Fatigue Analysis of Notched Laminates: A Time-Efficient Macro-Mechanical Approach

A coupled transversely isotropic deformation and damage fatigue model is implemented within the finite element method and was utilized along with a static progressive damage model to predict the fatigue life, stiffness degradation as a function of number of cycles, and post-fatigue tension and compression response of notched, multidirectional laminates. Initially, the material parameters for the fatigue model were obtained utilizing micromechanics simulations and the provided [0], [90] and [plus or minus 45] experimental composite laminate S-N (stress-cycle) data. Within the fatigue damage model, the transverse and shear properties of the plies were degraded with an isotropic scalar damage variable. The damage in the longitudinal (fiber) ply direction was suppressed, and only the strength of the fiber was degraded as a function of fatigue cycles. A maximum strain criterion was used to capture the failure in each element, and once this criterion was satisfied, the longitudinal stiffness of the element was decreased by a factor of 10 (sup 4). The resulting, degraded properties were then used to calculate the new stress state. This procedure was repeated until final failure of the composite laminate was achieved or a specified number of cycles reached. For post-fatigue tension and compression behavior, four internal state variables were used to control the damage and failure. The predictive capability of the above-mentioned approach was assessed by performing blind predictions of the notched multidirectional IM7/977-3 composite laminates response under fatigue and post-fatigue tensile and compressive loading, followed by a recalibration phase. Although three different multidirectional laminates were analyzed in the course of this study, only detailed results (i.e., stiffness degradation and post-fatigue stress-train curves as well as damage evolution states for a single laminate ([30/60/90/minus 30/minus 60] (sub 2s)) are discussed in detail here.

Composite Fatigue analysis↗
DOE OSTI2017

Feature Selection and Inferential Procedures for Video Data [Slides]

With the rise of electronic and high-dimensional data, new and innovative feature detection and statistical methods are required to perform accurate and meaningful statistical analysis of these datasets that provide unique statistical challenges. In the area of feature detection, much of the recent feature detection research in the computer vision community has focused on deep learning methods, which require large amounts of labeled training data. However, in many application areas, training data is very limited and often difficult to obtain. We develop methods for fast, unsupervised, precise feature detection for video data based on optical flows, edge detection, and clustering methods. We also use pretrained neural networks and interpretable linear models to extract features using very limited training data. In the area of statistics, while high-dimensional data analysis has been a main focus of recent statistical methodological research, much focus has been on populations of high-dimensional vectors, rather than populations of high-dimensional tensors, which are three-dimensional arrays that can be used to model dependent images, such as images taken of the same person or ripped video frames. Our feature detection method is a non-model-based method that fusses information from dense optical flow, raw image pixels, and frame differences to generate detections. Our hypothesis testing methods are based on the assumption that dependent images are concatenated into a tensor that follows a tensor normal distribution, and from this assumption, we derive likelihood-ratio, score, and regression-based tests for one- and multiple-sample testing problems. Our methods will be illustrated on simulated and real datasets. We conclude this report with comments on the relationship between feature detection and hypothesis testing methods.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

MLSPICE: Machine Learning based SPICE Modeling Platform for Power Magnetics

Electrical power converters are critical to a wide range of applications ranging from renewable integration to transportation electrification, and can be a key factor determining the size, weight, and efficiency of energy conversion systems. Magnetic components are typically the largest and least efficient components in power electronics. While there have been major strides in the modeling and analysis of power semiconductor devices and circuit simulations, the necessary advances in the design of power magnetics have lagged. In this project, we have transformed the modeling and design of power magnetics with machine learning enabled methods and catalyze simultaneous disruptive improvements for ML-based power electronics design tools. A fully automated open-source machine learning based magnetics modeling platform – the MagNet project - with innovations in full stack have been developed to greatly accelerate the design process and provide new insights to magnetic material and geometry design. The ARPA-E funded MagNet platform contains three major building blocks: 1) a ML-Integrated Data Acquisition System (MIDAS): a highly automated data acquisition testbed which is capable of measuring a large number of magnetic cores with a wide range of electrical circuit excitations; 2) a ML-integrated Core Loss Model (MICLM): a machine-learning trained modeling method for modeling the core loss and saturation effects of magnetic materials for arbitrary excitation waveforms; 3) ML-guided Magnetics SPICE Simulation Tool (PMSPICE): a fully integrated CAD tool which can simulate the magnetics in SPICE. It can help the designers to quickly model the linear and non-linear characteristics of magnetic components and evaluate their behavior in SPICE simulations. The developed MagNet system has fully demonstrated the proposed performance target and has been open sourced to the entire power electronics community to advance the modeling and design of power magnetics from many different angles.

36 MATERIALS SCIENCE↗
DOE OSTI2016

Artificial Neural Network Representation of Dynamic Criticality Excursion Experiment Data

This paper outlines time dependent considerations in applying the artificial neural network (ANN) paradigm to dynamic nuclear criticality excursion experiment data. The ANN amongst a variety of uses, can apply to function fitting of multiple input-multiple output data sets, pattern recognition, cluster analysis, and time series prediction. The ANN is a component of Machine Learning (ML) data analysis. The practical application of ML to criticality safety has been funded through the Y-12 Plant Directed Research and Demonstration (PDRD) program to support criticality safety and emergency planning efforts. The initial effort considered function fitting of criticality excursion data. Recent efforts successfully extended the ANN to criticality safety code validation. The initial ANN work applied to criticality excursions centered around multivariate function fitting of static (tabular) data taken from the French Consequences Radiologiques d'un Accident de Criticite (CRAC) and SILENE reactor campaigns. The former report (LeCorche) summarized important tabular data for the first 44 CRAC experiments. The initial ANN was trained to specific time-independent inputs, then applied to predict peak fission rate and total fissions for a wide variety of fissile solutions process conditions. This work extends that effort to apply the ANN paradigm to dynamic prediction of time-dependent criticality excursion experiment data. A dynamic criticality excursion in fissile solution represents a complex multiphysics phenomenon. Factors include among other things, coupled neutronic, multiphase fluid flow, temperature, pressure, acoustic shock, boiling heat transfer, reactivity feedback, and radiolytic gas considerations. There have been many efforts to model (over time) as coupled ordinary and (space) as partial differential equations (ODE, PDE). The ANN representation noted here does not rely on solving explicit ODE or PDEs. Rather the intent is to demonstrate that as a time series, the criticality excursion can be predicted by training specific ANN architectures to experimental input and output time series exclusively. The rationale for extending this work to the time-dependent regime is important in developing a 'criticality excursion emulator' solely on the basis of learning characteristics from experiment dynamic data. An emulator is different than a computer code simulator driven by the solution of ODEs and PDEs given select 'tuning parameters' that match outputs to experiments. A criticality excursion emulator, derived exclusively from an ANN or ML, could predict most likely excursion characteristics, trained from historic dynamic experiment data, presented with input values within the range of those experiments. The work noted shows that generic time-series ANN architectures can be satisfactorily applied to the prediction of dynamic, extremely chaotic criticality excursion power given an initial time dependent input series. In reality, an excursion emulator could be driven by user-supplied initial (t=0), time independent or constant values to generate time dependent outputs. Thus ongoing work includes developing a custom ANN such that initial dynamic conditions would be predicted. Those results then serve as inputs to a customized ANN architecture currently in development. (author)

42 ENGINEERING↗
DOE OSTI2022

Monotonic Gaussian Process for Physics-Constrained Machine Learning With Materials Science Applications

Physics-constrained machine learning is emerging as an important topic in the field of machine learning for physics. One of the most significant advantages of incorporating physics constraints into machine learning methods is that the resulting model requires significantly less data to train. By incorporating physical rules into the machine learning formulation itself, the predictions are expected to be physically plausible. Gaussian process (GP) is perhaps one of the most common methods in machine learning for small datasets. In this paper, we investigate the possibility of constraining a GP formulation with monotonicity on three different material datasets, where one experimental and two computational datasets are used. The monotonic GP is compared against the regular GP, where a significant reduction in the posterior variance is observed. The monotonic GP is strictly monotonic in the interpolation regime, but in the extrapolation regime, the monotonic effect starts fading away as one goes beyond the training dataset. Imposing monotonicity on the GP comes at a small accuracy cost, compared to the regular GP. The monotonic GP is perhaps most useful in applications where data are scarce and noisy, and monotonicity is supported by strong physical evidence.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Automated prediction of lattice parameters from X-ray powder diffraction patterns

A key step in the analysis of powder X-ray diffraction (PXRD) data is the accurate determination of unit-cell lattice parameters. This step often requires significant human intervention and is a bottleneck that hinders efforts towards automated analysis. This work develops a series of one-dimensional convolutional neural networks (1D-CNNs) trained to provide lattice parameter estimates for each crystal system. A mean absolute percentage error of approximately 10% is achieved for each crystal system, which corresponds to a 100- to 1000-fold reduction in lattice parameter search space volume. The models learn from nearly one million crystal structures contained within the Inorganic Crystal Structure Database and the Cambridge Structural Database and, due to the nature of these two complimentary databases, the models generalize well across chemistries. A key component of this work is a systematic analysis of the effect of different realistic experimental non-idealities on model performance. It is found that the addition of impurity phases, baseline noise and peak broadening present the greatest challenges to learning, while zero-offset error and random intensity modulations have little effect. However, appropriate data modification schemes can be used to bolster model performance and yield reasonable predictions, even for data which simulate realistic experimental non-idealities. In order to obtain accurate results, a new approach is introduced which uses the initial machine learning estimates with existing iterative whole-pattern refinement schemes to tackle automated unit-cell solution.

42 ENGINEERING↗
NASA NTRS1998

Applications of Artificial Neural Networks in Structural Engineering with Emphasis on Continuum Models

The use of continuum models for the analysis of discrete built-up complex aerospace structures is an attractive idea especially at the conceptual and preliminary design stages. But the diversity of available continuum models and hard-to-use qualities of these models have prevented them from finding wide applications. In this regard, Artificial Neural Networks (ANN or NN) may have a great potential as these networks are universal approximators that can realize any continuous mapping, and can provide general mechanisms for building models from data whose input-output relationship can be highly nonlinear. The ultimate aim of the present work is to be able to build high fidelity continuum models for complex aerospace structures using the ANN. As a first step, the concepts and features of ANN are familiarized through the MATLAB NN Toolbox by simulating some representative mapping examples, including some problems in structural engineering. Then some further aspects and lessons learned about the NN training are discussed, including the performances of Feed-Forward and Radial Basis Function NN when dealing with noise-polluted data and the technique of cross-validation. Finally, as an example of using NN in continuum models, a lattice structure with repeating cells is represented by a continuum beam whose properties are provided by neural networks.

Kapania, Rakesh K.↗
NASA NTRS2012

Mission Assurance Modeling and Simulation: A Cyber Security Roadmap

This paper proposes a cyber security modeling and simulation roadmap to enhance mission assurance governance and establish risk reduction processes within constrained budgets. The term mission assurance stems from risk management work by Carnegie Mellon's Software Engineering Institute in the late 19905. By 2010, the Defense Information Systems Agency revised its cyber strategy and established the Program Executive Officer-Mission Assurance. This highlights a shift from simply protecting data to balancing risk and begins a necessary dialogue to establish a cyber security roadmap. The Military Operations Research Society has recommended a cyber community of practice, recognizing there are too few professionals having both cyber and analytic experience. The authors characterize the limited body of knowledge in this symbiotic relationship. This paper identifies operational and research requirements for mission assurance M&S supporting defense and homeland security. M&S techniques are needed for enterprise oversight of cyber investments, test and evaluation, policy, training, and analysis.

Gendron, Gerald↗
DOE OSTI2018

Using gap-filled MAIAC AOD and WRF-Chem to estimate daily PM 2.5 concentrations at 1 km resolution in the Eastern United States

To link short-term exposures of air pollutants to health outcomes, scientists must use high temporal and spatial resolution estimates of PM 2.5 concentrations. We develop a daily PM 2.5 product at 1 × 1 km 2 spatial resolution across the eastern United States (east of 90° W) with the aid of 1 × 1 km 2 MAIAC aerosol optical depth (AOD) data, 36 × 36 km 2 WRF-Chem output, 1 × 1 km 2 land-use type from the National Land Cover Database, and 0.125° × 0.125° ERA-Interim re-analysis meteorology. A gap-filling technique is applied to MAIAC AOD data to construct robust daily estimates of AOD when the satellite data are missing (e.g., areas obstructed by clouds or snow cover). The input data are incorporated into a multiple-linear regression model trained to surface observations of PM 2.5 from the EPA Air Quality System (AQS) monitoring network. The model generates a high-fidelity estimate (r 2 = 0.75 using a 10-fold random cross-validation) of daily PM 2.5 throughout the eastern United States. Of the inputs to the statistical model, WRF-Chem output (r 2 = 0.66) is the most important contributor to the skill of the model. MAIAC AOD is also a strong contributor (r 2 = 0.52). Daily PM 2.5 output from our statistical model can be easily integrated into county-level epidemiological studies. The novelty of this project is that we are able to simulate PM 2.5 in a computationally efficient manner that is constrained to ground monitors, satellite data, and chemical transport model output at high spatial resolution (1 × 1 km 2 ) without sacrificing the temporal resolution (daily) or spatial coverage (>2,000,000 km 2 ).

47 OTHER INSTRUMENTATION↗
DOE OSTI2020

Approach to Quality Assurance for Complex Environmental Modeling - 20407

A strong Quality Assurance (QA) program for complex environmental modeling is essential for regulatory and public acceptance and trust, but it does not have to be onerous. Neptune and Company, Inc. (Neptune) has developed a strong QA program that improves transparency, traceability, reproducibility, and therefore, defensibility and trust. Neptune's QA program has evolved over the past 27 years, transitioning from an ad hoc QA program, to a program that is currently Nuclear Quality Assurance-1 (NQA-1) compliant, and will soon be NQA-1/DOE approved for DOE EM modeling work. In addition, Neptune is a qualified laboratory assessor and qualified auditor for reference materials, a proficiency testing provider, and fully complies with the Environmental Protection Agency (EPA) QA program. Neptune's President and CEO, Kelly Black, was recently named the Chairperson of the International Organization for Standardization (ISO) Technical Committee 69, Application of Statistical Methods. A strong QA program has been developed for Neptune's radiological performance assessment (PA) program. Although pieces of their QA program are currently in development, their current program includes document control using Subversion, issue tracking and work flow tracking using JIRA, and transparency and traceability via a system of 'calc sheets' for documentation of all data analysis and modeling combined with 'check print' documentation of QA checking, and rigorous model testing and configuration control. All work is reviewed by an independent subject matter expert who is not associated with the collection and assembly of information, for an internal peer review. The program is enforced with a handful of Standard Operation Procedures (SOPs), Work Instructions, QA Project Plans (QAPPs), and Quality Management Plans (QMPs) that are updated frequently, with required annual training and acknowledgment. In addition, effective communication of modeling approaches and results to clients and stakeholders are integral to their QA program. Many of their models are built using the GoldSim modeling platform, for which their models are well-known for their level of transparency, documentation, and QA traceability. This level of QA is also applied to their process-level models. Neptune has taken some lessons learned from the extremely rigorous QA program required for the Yucca Mountain Project (YMP), and imposed the YMP requirements of complete traceability, transparency, and reproducibility, but avoided the inflexibility of a QA program that likely contributed to the suspect e-mails in 2005 that resulted in loss of trust in the YMP, and in nuclear waste disposal (and nuclear power) in general. There are far too many examples of loss of public trust due to poor QA that could have been easily avoided with a simple and straight-forward QA program combined with stakeholder engagement. The purpose of this paper is to share features of Neptune's Radiological Performance Assessment program QA program, and some of their QA related lessons learned over the past 27 years of QA for complex environmental modeling. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2024

Random forest prediction of crystal structure from electron diffraction patterns incorporating multiple scattering

Diffraction is the most common method to solve for unknown or partially known crystal structures. However, it remains a challenge to determine the crystal structure of a new material that may have nanoscale size or heterogeneities. Here, in this study, we train an architecture of hierarchical random forest models capable of predicting the crystal system, space group, and lattice parameters from one or more unknown two-dimensional electron diffraction patterns. Our initial model correctly identifies the crystal system of a simulated electron diffraction pattern from a 20-nm-thick specimen of arbitrary orientation 67% of the time. We achieve a topline accuracy of 79% when aggregating predictions from ten patterns of the same material but different zone axes. The space group and lattice predictions range from 70% to 90% accuracy and median errors of 0.01-0.5Å, respectively, for cubic, hexagonal, trigonal, and tetragonal crystal systems while being less reliable on orthorhombic and monoclinic systems. We apply this architecture to a four-dimensional scanning transmission electron microscopy scan of gold nanoparticles, where it accurately predicts the crystal structure and lattice constants. These random forest models can be used to significantly accelerate the analysis of electron diffraction patterns, particularly in the case of unknown crystal structures. Additionally, due to the speed of inference, these models could be integrated into live transmission electron microscopy experiments, allowing real-Time labeling of a specimen.

36 MATERIALS SCIENCE↗
DOE OSTI2023

DQ Admittance Model Extraction for IBRs via Gaussian Pulse Excitation

While $dq$ admittance models have shown to be very useful for stability analysis, extracting admittance models of inverter-based resources (IBRs) from the electromagnetic transient (EMT) simulation environment using frequency scans takes time. In this letter, a new perturbation method based on Gaussian pulses in combination with the system identification algorithms shows great promise for parametric $dq$ admittance model extraction. We present the $dq$ admittance model extracting method for a type-4 wind turbine. Challenges in implementing Gaussian pulse excitation are also pointed out. In conclusion, the extracted $dq$ admittance model via the new method shows to have a high matching degree with the measurements obtained from frequency scans.

24 POWER TRANSMISSION AND DISTRIBUTION↗
NASA NTRS2000

Equivalent Skin Analysis of Wing Structures Using Neural Networks

An efficient method of modeling trapezoidal built-up wing structures is developed by coupling. in an indirect way, an Equivalent Plate Analysis (EPA) with Neural Networks (NN). Being assumed to behave like a Mindlin-plate, the wing is solved using the Ritz method with Legendre polynomials employed as the trial functions. This analysis method can be made more efficient by avoiding most of the computational effort spent on calculating contributions to the stiffness and mass matrices from each spar and rib. This is accomplished by replacing the wing inner-structure with an "equivalent" material that combines to the skin and whose properties are simulated by neural networks. The constitutive matrix, which relates the stress vector to the strain vector, and the density of the equivalent material are obtained by enforcing mass and stiffness matrix equities with rec,ard to the EPA in a least-square sense. Neural networks for the material properties are trained in terms of the design variables of the wing structure. Examples show that the present method, which can be called an Equivalent Skin Analysis (ESA) of the wing structure, is more efficient than the EPA and still fairly good results can be obtained. The present ESA is very promising to be used at the early stages of wing structure design.

Liu, Youhua↗
NASA NTRSDate not supplied

Derivation of Capabilities and Resources for Treating Medical Conditions in Deep Space

BACKGROUND: Medical care in spaceflight requires the adaptation of terrestrial standards to the constraints of the space environment. However, there is often conflict between the ideal resources required for treatment and diagnosis of a medical condition, and the constraints on their mass, volume, data needs, etc. This optimization of trades between medical risk and resources will be a significant challenge for deep space medical system design. METHODS: A team of physicians with a broad range of expertise reviewed the 120 medical conditions on the condition list for NASA’s Informing Mission Planning via analysis of Complex Tradespaces Medical Database (IMPACT-MD). Each team member was assigned a subset of conditions based on their experience and training. The assigned physician generated a proposed list of capabilities, definitions, and resources required to treat each condition based on terrestrial practice guidelines, medical literature, and subject matter expertise. Additional specialists were consulted for conditions where expertise was not present within the main group. The proposed list was then reviewed by the broader team and modified as needed to achieve consensus. Each capability and resource was then assigned parameters to define quantity required per medical event, necessity, training level required, and primacy order of any alternative resources. These capabilities and resources were placed into condition specific tables and delivered to a team of engineers who added mass and volume data for each specific resource and converted the table into a database for use as input to a computational model to simulate spaceflight (IMPACT) DISCUSSION: This process ensures that a minimum of three subject matter experts review and agree upon the scope of practice, medical diagnostic tools, and treatment modalities that would be necessary to address emergent and non-emergent conditions that may arise during spaceflight to inform requirements during the vehicle design phase. The method is scalable to any design reference mission and permits modification of the existing database as information, conditions or experience are added. By including specialists, generalists, and those with subject matter expertise in the spaceflight environment on the team, we ensure that the included capabilities represent a realistic and actionable foundation for planning deep space missions.

D R Levin↗
DOE OSTI2020

Modeling gas–shell mixing in ICF with separated reactants

Mixing between fuel and shell materials in ICF implosions can affect implosion dynamics and even prevent ignition. We use data from a series of separated reactant experiments on the National Ignition Facility to calibrate and test the predictive power of gas–shell mix models. Two models are used to estimate fuel–shell mix: a Reynolds-averaged turbulence model and molecular diffusion. Minor uncertainties in capsule manufacture, experimental conditions, and values for mix model parameters produce significant variation in simulation results. Using input/output pairs from 1D simulations, we train Gaussian process surrogate models to predict experimental quantities of interest. The surrogates are used to construct posteriors for mix model parameters by marginalizing over uncertainties in capsule manufacture and experimental conditions. Mix models are calibrated with a subset of experimental data (neutron yields, ion temperature, and bang time) and tested using the remaining data. In general, both the diffusion and turbulence model correctly predict experimental DT and TT neutron yields. Despite having more free parameters, the turbulence model underpredicts ion temperature at high convergence ratio. Furthermore, the simpler diffusion model correctly predicts these temperatures, suggesting nonhydrodynamic gas–shell mix. The computational model consistently overpredicts DD neutron yield, indicating possible shortcomings outside of the mix model.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2018

Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks

We report that the ability to accurately and efficiently compute quantum-mechanical partial atomistic charges has many practical applications, such as calculations of IR spectra, analysis of chemical bonding, and classical force field parametrization. Machine learning (ML) techniques provide a possible avenue for the efficient prediction of atomic partial charges. Modern ML advances in the prediction of molecular energies [i.e., the hierarchical interacting particle neural network (HIP-NN)] has provided the necessary model framework and architecture to predict transferable, extensible, and conformationally dynamic atomic partial charges based on reference density functional theory (DFT) simulations. Utilizing HIP-NN, we show that ML charge prediction can be highly accurate over a wide range of molecules (both small and large) across a variety of charge partitioning schemes such as the Hirshfeld, CM5, MSK, and NBO methods. To demonstrate transferability and size extensibility, we compare ML results with reference DFT calculations on the COMP6 benchmark, achieving errors of 0.004e – (elementary charge). This is remarkable since this benchmark contains two proteins that are multiple times larger than the largest molecules in the training set. An application of our atomic charge predictions on nonequilibrium geometries is the generation of IR spectra for organic molecules from dynamical trajectories on a variety of organic molecules, which show good agreement with calculated IR spectra with reference method. Critically, HIP-NN charge predictions are many orders of magnitude faster than direct DFT calculations. Lastly, these combined results provide further evidence that ML (specifically HIP-NN) provides a pathway to greatly increase the range of feasible simulations while retaining quantum-level accuracy.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2014

WE-G-BRF-04: Robust Real-Time Volumetric Imaging Based On One Single Projection

Purpose: Real-time volumetric imaging is highly desirable to provide instantaneous image guidance for lung radiation therapy. This study proposes a scheme to achieve this goal using one single projection by utilizing sparse learning and a principal component analysis (PCA) based lung motion model. Methods: A patient-specific PCA-based lung motion model is first constructed by analyzing deformable vector fields (DVFs) between a reference image and 4DCT images at each phase. At the training stage, we “learn” the relationship between the DVFs and the projection using sparse learning. Specifically, we first partition the projections into patches, and then apply sparse learning to automatically identify patches that best correlate with the principal components of the DVFs. Once the relationship is established, at the application stage, we first employ a patchbased intensity correction method to overcome the problem of different intensity scale between the calculated projection in the training stage and the measured projection in the application stage. The corrected projection image is then fed to the trained model to derive a DVF, which is applied to the reference image, yielding a volumetric image corresponding to the projection. We have validated our method through a NCAT phantom simulation case and one experiment case. Results: Sparse learning can automatically select those patches containing motion information, such as those around diaphragm. For the simulation case, over 98% of the lung region pass the generalized gamma test (10HU/1mm), indicating combined accuracy in both intensity and spatial domain. For the experimental case, the average tumor localization errors projected to the imager are 0.68 mm and 0.4 mm on the axial and tangential direction, respectively. Conclusion: The proposed method is capable of accurately generating a volumetric image using one single projection. It will potentially offer real-time volumetric image guidance to facilitate lung radiotherapy. This study is supported in part by NIH (1R01CA154747-01), The Core Technology Research in Strategic Emerging Industry, Guangdong, China(2011A081402003)

60 APPLIED LIFE SCIENCES↗