A local basis approximation approach for nonlinear parametric model order reduction
Not Available
SEARCH · Search NASA
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.
Not Available
Abstract not provided.
Dynamic Mode Decomposition (DMD) is a model-order reduction approach, whereby spatial modes of fixed temporal frequencies are extracted from numerical or experimental data sets. The DMD low-rank or reduced operator is typically obtained by singular value decomposition of the temporal data sets. For parameter-dependent models, as found in many multi-query applications such as uncertainty quantification or design optimization, the only parametric DMD technique developed was a stacked approach, with data sets at multiple parameter values were aggregated together, increasing the computational work needed to devise low-rank dynamical reduced-order models. Here in this paper, we present two novel approach to carry out parametric DMD: one based on the interpolation of the reduced-order DMD eigen-pair and the other based on the interpolation of the reduced DMD (Koopman) operator. Numerical results are presented for diffusion-dominated nonlinear dynamical problems, including a multiphysics radiative transfer example. All three parametric DMD approaches are compared.
Reduced-order models are often used to describe the behavior of complex systems, whose simulation with a full model is too expensive, or to extract salient features from the full model’s output. We introduce a new model-reduction framework DRIPS (dimension reduction and interpolation in parameter space) that combines the offline local model reduction with the online parameter interpolation of reduced-order bases (ROBs). The offline step of this framework relies on dynamic mode decomposition (DMD) to build a low-rank linear surrogate model, equipped with a local ROB, for quantities of interest derived from the training data generated by repeatedly solving the (nonlinear) high-fidelity model for multiple parameter points. The online step consists of the construction of a parametric reduced-order model for each target/test point in the parameter space, with the interpolation of ROBs done on a Grassman manifold and the interpolation of reduced-order operators done on a matrix manifold. The DMD component enables DRIPS to model (typically low-dimensional) quantities of interest directly, without having to access the (typically high-dimensional and possibly nonlinear) operators in a high-fidelity model that governs the dynamics of the underlying high-dimensional state variables, as required in projection-based reduced-order modeling. A series of numerical experiments suggests that DRIPS yields a model reduction, which is computationally more efficient than the commonly used projection-based proper orthogonal decomposition; it does so without requiring a prior knowledge of the governing equation for quantities of interest. Furthermore, for the nonlinear systems considered, DRIPS is more accurate than Gaussian-process interpolation (Kriging).
The mechanical characterization of materials relies more and more often on sophisticated experimental methods that permit to acquire a large amount of data and, contemporarily, to reduce the invasiveness of the tests. This evolution accompanies the growing demand of non-destructive diagnostic tools that assess the safety level of components in use in structures and infrastructures, for instance in the strategic energy sector. Advanced material systems and properties that are not amenable to traditional techniques, for instance thin layered structures and their adhesion on the relevant substrates, can be also characterized by means of combined experimental-numerical tools elaborating data acquired by full-field measurement techniques. In this context, parameter identification procedures involve the repeated simulation of the laboratory or in situ tests by sophisticated and usually expensive non-linear analyses while, in some situation, reliable and accurate results would be required in real time. The effectiveness and the filtering capabilities of reduced models based on decomposition and interpolation techniques can be profitably used to meet these conflicting requirements. This communication intends to summarize some results recently achieved in this field by the author and her co-workers. The aim is to foster further interaction between engineering and mathematical communities.
Images acquired with passive sensing techniques suffer from illumination variations and poor local contrasts that create major difficulties in interpretation and identification tasks. On the other hand, images acquired with active sensing techniques based on monochromatic illumination are degraded with speckle noise. Mathematical morphology offers elegant techniques to handle a wide range of image degradation problems. Unlike linear filters, morphological filters do not blur the edges and hence maintain higher image resolution. Their rich mathematical framework facilitates the design and analysis of these filters as well as their hardware implementation. Morphological filters are easier to implement and are more cost effective and efficient than several conventional linear filters. Morphological filters to remove speckle noise while maintaining high resolution and preserving thin image regions that are particularly vulnerable to speckle noise were developed and applied to SAR imagery. These filters used combination of linear (one-dimensional) structuring elements in different (typically four) orientations. Although this approach preserves more details than the simple morphological filters using two-dimensional structuring elements, the limited orientations of one-dimensional elements approximate the fine details of the region boundaries. A more robust filter designed recently overcomes the limitation of the fixed orientations. This filter uses a combination of concave and convex structuring elements. Morphological operators are also useful in extracting features from visible and infrared imagery. A multiresolution image pyramid obtained with successive filtering and a subsampling process aids in the removal of the illumination variations and enhances local contrasts. A morphology-based interpolation scheme was also introduced to reduce intensity discontinuities created in any morphological filtering task. The generality of morphological filtering techniques in extracting information from a wide variety of images obtained with active and passive sensing techniques is discussed. Such techniques are particularly useful in obtaining more information from fusion of complex images by different sensors such as SAR, visible, and infrared.
Non-intrusive reduced-order models (ROMs) have recently generated considerable interest for constructing computationally efficient counterparts of nonlinear dynamical systems emerging from various domain sciences. They provide a low-dimensional emulation framework for systems that may be intrinsically high-dimensional. This is accomplished by utilizing a construction algorithm that is purely data-driven. It is no surprise, therefore, that the algorithmic advances of machine learning have led to non-intrusive ROMs with greater accuracy and computational gains. However, in bypassing the utilization of an equation-based evolution, it is often seen that the interpretability of the ROM framework suffers. This becomes more problematic when black-box deep learning methods are used which are notorious for lacking robustness outside the physical regime of the observed data. In this article, we propose the use of a novel latent-space interpolation algorithm based on Gaussian process regression. Notably, this reduced-order evolution of the system is parameterized by control parameters to allow for interpolation in space. The use of this procedure also allows for a continuous interpretation of time which allows for temporal interpolation. The latter aspect provides information, with quantified uncertainty, about full-state evolution at a finer resolution than that utilized for training the ROMs. This research assesses the viability of this algorithm for an advection-dominated system given by the inviscid shallow water equations.
The ab initio description of the spectral interior of the absorption spectrum poses both a theoretical and computational challenge for modern electronic structure theory. Due to the often spectrally dense character of this domain in the quantum propagator’s eigenspectrum for medium-to-large sized systems, traditional approaches based on the partial diagonalization of the propagator often encounter oscillatory and stagnating convergence. Electronic structure methods which solve the molecular response problem through the solution of spectrally shifted linear systems, such as the complex polarization propagator, offer an alternative approach which is agnostic to the underlying spectral density or domain location. This generality comes at a seemingly high computational cost associated with solving a large linear system for each spectral shift in some discretization of the spectral domain of interest. In this work, we present a novel, adaptive solution to this high computational overhead based on model order reduction techniques via interpolation. Model order reduction reduces the computational complexity of mathematical models and is ubiquitous in the simulation of dynamical systems and control theory. The efficiency and effectiveness of the proposed algorithm in the ab initio prediction of X-ray absorption spectra is demonstrated using a test set of challenging water clusters which are spectrally dense in the neighborhood of the oxygen K-edge. On the basis of a single, user defined tolerance we automatically determine the order of the reduced models and approximate the absorption spectrum up to the given tolerance. We also illustrate that, for the systems studied, the automatically determined model order increases logarithmically with the problem dimension, compared to a linear increase of the number of eigenvalues within the energy window. Furthermore, we observed that the computational cost of the proposed algorithm only scales quadratically with respect to the problem dimension.
Nonautonomous dynamical systems are characterized by time-dependent inputs, which complicates the discovery of predictive models describing the spatiotemporal evolution of the state variables of quantities of interest from their temporal snapshots. When dynamic mode decomposition (DMD) is used to infer a linear model, this difficulty manifests itself in the need to approximate the time-dependent Koopman operators. Our approach is to approximate the original nonautonomous system with a modified system derived via a local parameterization of the time-dependent inputs. The modified system comprises a sequence of local parametric systems, which are subsequently approximated by a parametric surrogate model using the DRIPS (dimension reduction and interpolation in parameter space) framework. The offline step of DRIPS relies on DMD to build a linear surrogate model, endowed with reduced-order bases for the observables mapped from training data. The online step interpolates on suitable manifolds to construct a sequence of iterative parametric surrogate models; the target/test parameter points on these manifolds are specified by a local parameterization of the test time-dependent inputs. Here, we use numerical experimentation to demonstrate the robustness of our method and compare its performance with that of deep neural networks.
We present a similarity transformation theory based on a polynomial form of a particle-hole pair excitation operator. In the weakly correlated limit, this polynomial becomes an exponential, leading to coupled cluster doubles. In the opposite strongly correlated limit, the polynomial becomes an extended Bessel expansion and yields the projected BCS wavefunction. In between, we interpolate using a single parameter. The e ective Hamiltonian is non-hermitian and this Polynomial Similarity Transformation Theory follows the philosophy of traditional coupled cluster, left projecting the transformed Hamiltonian onto subspaces of the Hilbert space in which the wave function variance is forced to be zero. Similarly, the interpolation parameter is obtained through minimizing the next residual in the projective hierarchy. We rationalize and demonstrate how and why coupled cluster doubles is ill suited to the strongly correlated limit whereas the Bessel expansion remains well behaved. The model provides accurate wave functions with energy errors that in its best variant are smaller than 1% across all interaction stengths. The numerical cost is polynomial in system size and the theory can be straightforwardly applied to any realistic Hamiltonian.
In earlier works, a mathematical procedure for invertible microstructure-property linkages was developed using computationally efficient spectral methods for polycrystalline cubic and hexagonal metals. This paper formulates such invertible microstructure–property linkages for orthorhombic polycrystalline metals relying on the generalized spherical harmonics (GSH) spectral basis. The procedure is used to compute property closures of orthorhombic polycrystals. The closures represent the complete set of theoretically possible combinations of effective properties for a selected material. The procedure relies on the first-order bounding theories and considers orientation distribution functions (ODFs) as the main microstructural descriptor influencing homogenized properties. Numerous examples of these closures involving second-rank thermal expansion and fourth-rank elastic stiffness tensorial properties over a broad range of temperatures are presented for α-uranium (α-U). In doing so, certain key properties of these closures are exploited to facilitate their computation with drastically reduced computational effort. Along with the recently developed GSH-based interpolation procedure for ODFs from coarsely spaced experimental measurement grids to finely spaced finite element mesh resolution grids presented in Barrett et al., the developed computationally efficient ODF-effective property linkages are used to establish a crystal mechanics-based simulation framework coupled with the finite element method (FEM). The ODF dependent thermal expansion and elastic stiffness tensors are efficiently calculated at every integration point and used by the FEM to predict the overall distortion of a hemispherical part made of α-U during heating. In conclusion, it is shown that the developed framework can be used to simulate microstructurally heterogeneous components under thermo-mechanical loadings in a computationally efficient manner.
Purpose: Four-dimensional computer tomography (4D-CT) has been widely used in lung cancer radiotherapy due to its capability in providing important tumor motion information. However, the prolonged scanning duration required by 4D-CT causes considerable increase in radiation dose. To minimize the radiation-related health risk, radiation dose is often reduced at the expense of interslice spatial resolution. However, inadequate resolution in 4D-CT causes artifacts and increases uncertainty in tumor localization, which eventually results in extra damages of healthy tissues during radiotherapy. In this paper, the authors propose a novel postprocessing algorithm to enhance the resolution of lung 4D-CT data. Methods: The authors' premise is that anatomical information missing in one phase can be recovered from the complementary information embedded in other phases. The authors employ a patch-based mechanism to propagate information across phases for the reconstruction of intermediate slices in the longitudinal direction, where resolution is normally the lowest. Specifically, the structurally matching and spatially nearby patches are combined for reconstruction of each patch. For greater sensitivity to anatomical details, the authors employ a quad-tree technique to adaptively partition the image for more fine-grained refinement. The authors further devise an iterative strategy for significant enhancement of anatomical details. Results: The authors evaluated their algorithm using a publicly available lung data that consist of 10 4D-CT cases. The authors' algorithm gives very promising results with significantly enhanced image structures and much less artifacts. Quantitative analysis shows that the authors' algorithm increases peak signal-to-noise ratio by 3-4 dB and the structural similarity index by 3%-5% when compared with the standard interpolation-based algorithms. Conclusions: The authors have developed a new algorithm to improve the resolution of 4D-CT. It outperforms the conventional interpolation-based approaches by producing images with the markedly improved structural clarity and greatly reduced artifacts.
Purpose: Synchronized moving grid (SMOG) is a promising technique to reduce scatter and ghost artifacts in cone beam computed tomography (CBCT). However, the grid blocks part of image information in each projection, and multiple projections at the same gantry angle have to been taken to obtain full information. Because of the continuity of a patient's anatomy in the projection, the blocked information may be estimated by interpolation. This study aims to evaluate an inpainting-based interpolation approach to recover the missing information for CBCT reconstruction. Method: We used a simple region-based inpainting approach to interpolate the missing information. For a pixel to be interpolated, we divided the nearby regions having image information into 6 sub-regions: up-left, up-middle, up-right, down-left, down-middle, and down-right, each with 9 pixels. The average pixel value of each sub-region was calculated. These average values, along with the pixel location, were used to determine the interpolated pixel value. We compared our approach with the Criminisi Exemplar (CE) and total variation (TV) based inpainting techniques. Projection images of Catphan and a head phantom were used for the comparison. The SMOG was simulated by erasing the information (filling with “0”) of the areas in each projection corresponding to the grid. Results: For the Catphan, the processing time was 178, 45 and 0.98 minutes for CE, TV and our approach, respectively. The signal to noise ratio (SNR) was 19.4, 18.5 and 26.4 db, correspondingly. For the head phantom, the processing time was 222, 45 and 0.93 minutes for CE, TV and our approach, respectively. The SNR was 24.6, 20.2 and 26.2db correspondingly. Conclusion: We have developed a simple inpainting based interpolation approach, which can recover some of the image information for the SMOG-based CBCT imaging. This study is supported by NIH/NCI grant 1R01CA166948-01.
Self-tuning aircraft engine models can be applied for control and health management applications. The self-tuning feature of these models minimizes the mismatch between any given engine and the underlying engineering model describing an engine family. This paper provides details of the construction of a self-tuning engine model centered on a piecewise linear Kalman filter design. Starting from a nonlinear transient aerothermal model, a piecewise linear representation is first extracted. The linearization procedure creates a database of trim vectors and state-space matrices that are subsequently scheduled for interpolation based on engine operating point. A series of steady-state Kalman gains can next be constructed from a reduced-order form of the piecewise linear model. Reduction of the piecewise linear model to an observable dimension with respect to available sensed engine measurements can be achieved using either a subset or an optimal linear combination of "health" parameters, which describe engine performance. The resulting piecewise linear Kalman filter is then implemented for faster-than-real-time processing of sensed engine measurements, generating outputs appropriate for trending engine performance, estimating both measured and unmeasured parameters for control purposes, and performing on-board gas-path fault diagnostics. Computational efficiency is achieved by designing multidimensional interpolation algorithms that exploit the shared scheduling of multiple trim vectors and system matrices. An example application illustrates the accuracy of a self-tuning piecewise linear Kalman filter model when applied to a nonlinear turbofan engine simulation. Additional discussions focus on the issue of transient response accuracy and the advantages of a piecewise linear Kalman filter in the context of validation and verification. The techniques described provide a framework for constructing efficient self-tuning aircraft engine models from complex nonlinear simulations.Self-tuning aircraft engine models can be applied for control and health management applications. The self-tuning feature of these models minimizes the mismatch between any given engine and the underlying engineering model describing an engine family. This paper provides details of the construction of a self-tuning engine model centered on a piecewise linear Kalman filter design. Starting from a nonlinear transient aerothermal model, a piecewise linear representation is first extracted. The linearization procedure creates a database of trim vectors and state-space matrices that are subsequently scheduled for interpolation based on engine operating point. A series of steady-state Kalman gains can next be constructed from a reduced-order form of the piecewise linear model. Reduction of the piecewise linear model to an observable dimension with respect to available sensed engine measurements can be achieved using either a subset or an optimal linear combination of "health" parameters, which describe engine performance. The resulting piecewise linear Kalman filter is then implemented for faster-than-real-time processing of sensed engine measurements, generating outputs appropriate for trending engine performance, estimating both measured and unmeasured parameters for control purposes, and performing on-board gas-path fault diagnostics. Computational efficiency is achieved by designing multidimensional interpolation algorithms that exploit the shared scheduling of multiple trim vectors and system matrices. An example application illustrates the accuracy of a self-tuning piecewise linear Kalman filter model when applied to a nonlinear turbofan engine simulation. Additional discussions focus on the issue of transient response accuracy and the advantages of a piecewise linear Kalman filter in the context of validation and verification. The techniques described provide a framework for constructing efficient self-tuning aircraft engine models from complex nonlinear simulatns.
This presentation reviews the development, integration, and testing of Near Real Time (NRT) MODIS forest % maximum NDVI change products resident to the USDA Forest Service (USFS) ForWarn System. ForWarn is an Early Warning System (EWS) tool for detection and tracking of regionally evident forest change, which includes the U.S. Forest Change Assessment Viewer (FCAV) (a publically available on-line geospatial data viewer for visualizing and assessing the context of this apparent forest change). NASA Stennis Space Center (SSC) is working collaboratively with the USFS, ORNL, and USGS to contribute MODIS forest change products to ForWarn. These change products compare current NDVI derived from expedited eMODIS data, to historical NDVI products derived from MODIS MOD13 data. A new suite of forest change products are computed every 8 days and posted to the ForWarn system; this includes three different forest change products computed using three different historical baselines: 1) previous year; 2) previous three years; and 3) all previous years in the MODIS record going back to 2000. The change product inputs are maximum value NDVI that are composited across a 24 day interval and refreshed every 8 days so that resulting images for the conterminous U.S. are predominantly cloud-free yet still retain temporally relevant fresh information on changes in forest canopy greenness. These forest change products are computed at the native nominal resolution of the input reflectance bands at 231.66 meters, which equates to approx 5.4 hectares or 13.3 acres per pixel. The Time Series Product Tool, a MATLAB-based software package developed at NASA SSC, is used to temporally process, fuse, reduce noise, interpolate data voids, and re-aggregate the historical NDVI into 24 day composites, and then custom MATLAB scripts are used to temporally process the eMODIS NDVIs so that they are in synch with the historical NDVI products. Prior to posting, an in-house snow mask classification product is computed for the current compositing period and integrated into the change images to account for snow related NDVI drops. The supplemental snow classification product was needed because other available QA cloud/snow mask typically underestimates snow cover. MODIS true and false color composites were also computed from eMODIS reflectance data and the true color RGBs are also posted on ForWarn?s FCAV; this data is used for assessing apparent occasional quality issues on the change products due to residual unmasked cloud cover. New forest change products are posted with typical latencies of 1-2 days after the last input eMODIS data collection date for a given 24 day compositing period.
The calculation of potential energy surfaces for quantum dynamics can be a time consuming task—especially when a high level of theory for the electronic structure calculation is required. We propose an adaptive interpolation algorithm based on polyharmonic splines combined with a partition of unity approach. The adaptive node refinement allows to greatly reduce the number of sample points by employing a local error estimate. The algorithm and its scaling behavior are evaluated for a model function in 2, 3, and 4 dimensions. The developed algorithm allows for a more rapid and reliable interpolation of a potential energy surface within a given accuracy compared to the non-adaptive version.
Numerically solving partial differential equations (PDEs) can be challenging and computationally expensive. This has led to the development of reduced-order models (ROMs) that are accurate but faster than full order models (FOMs). Recently, machine learning advances have enabled the creation of non-linear projection methods, such as Latent Space Dynamics Identification (LaSDI). LaSDI maps full-order PDE solutions to a latent space using autoencoders and learns the system of ODEs governing the latent space dynamics. By interpolating and solving the ODE system in the reduced latent space, fast and accurate ROM predictions can be made by feeding the predicted latent space dynamics into the decoder. In this paper, we introduce GPLaSDI, a novel LaSDI-based framework that relies on Gaussian process (GP) for latent space ODE interpolations. Using GPs offers two significant advantages. First, it enables the quantification of uncertainty over the ROM predictions. Second, leveraging this prediction uncertainty allows for efficient adaptive training through a greedy selection of additional training data points. This approach does not require prior knowledge of the underlying PDEs. Consequently, GPLaSDI is inherently non-intrusive and can be applied to problems without a known PDE or its residual. Here we demonstrate the effectiveness of our approach on the Burgers equation, Vlasov equation for plasma physics, and a rising thermal bubble problem. Our proposed method achieves between 200 and 100,000 times speed-up, with up to 7% relative error.
We propose a new method for obtaining potential energy surfaces in sum-of-products (SOP) form. If the number of terms is small enough, a SOP potential surface significantly reduces the cost of quantum dynamics calculations by obviating the need to do multidimensional integrals by quadrature. The method is based on a Smolyak interpolation technique and uses polynomial-like or spectral basis functions and 1D Lagrange-type functions. When written in terms of the basis functions from which the Lagrange-type functions are built, the Smolyak interpolant has only a modest number of terms. The ideas are tested for HONO (nitrous acid)