SEARCH · Engineering Papers
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.
In this paper a machine vision approach is applied to Interferometric Synthetic Aperture Radars (IFSAR) data to extract the most relevant built structures in a dense urban environment. The algorithm tries to cluster primitives (line segments) into more complex surfaces (planes) to approximate the 3D shape of these objects. Very interesting results starting from TOPSAR data recorded over S, Monica are presented.
This paper describes the methods used to produce an interactive visualization of a 2 TB computational fluid dynamics (CFD) data set using particle tracing (streaklines). We use the method introduced by Bruckschen et al. [2001] that pre-computes a large number of particles, stores them on disk using a space-filling curve ordering that minimizes seeks, and then retrieves and displays the particles according to the user's command. We describe how the particle computation can be performed using a PC cluster, how the algorithm can be adapted to work with a multi-block curvilinear mesh, and how the out-of-core visualization can be scaled to 296 billion particles while still achieving interactive performance on PG hardware. Compared to the earlier work, our data set size and total number of particles are an order of magnitude larger. We also describe a new compression technique that allows the lossless compression of the particles by 41% and speeds the particle retrieval by about 30%.
Aerogels are of interest to the aerospace community primarily for their thermal properties, notably their low thermal conductivities. While the gels are typically fragile, recent advances in the application of conformal polymer layers to these gels has made them potentially useful as lightweight structural materials as well. In this work, we investigate the strength and fracture behavior of silica aerogels using a molecular statics-based computer simulation technique. The gels' structure is simulated via a Diffusion Limited Cluster Aggregation (DLCA) algorithm, which produces fractal structures representing experimentally observed aggregates of so-called secondary particles, themselves composed of amorphous silica primary particles an order of magnitude smaller. We have performed multi-length-scale simulations of fracture in silica aerogels, in which the interaction b e e n two secondary particles is assumed to be described by a Morse pair potential parameterized such that the potential range is much smaller than the secondary particle size. These Morse parameters are obtained by atomistic simulation of models of the experimentally-observed amorphous silica "bridges," with the fracture behavior of these bridges modeled via molecular statics using a Morse/Coulomb potential for silica. We consider the energetics of the fracture, and compare qualitative features of low-and high-density gel fracture.
In air traffic management systems, airspace is partitioned into regions in part to distribute the tasks associated with managing air traffic among different systems and people. These regions, as well as the systems and people allocated to each, are changed dynamically so that air traffic can be safely and efficiently managed. It is expected that new air traffic control systems will enable greater flexibility in how airspace is partitioned and how resources are allocated to airspace regions. In this talk, I will begin by providing an overview of some previous work and open questions in Dynamic Airspace Configuration research, which is concerned with how to partition airspace and assign resources to regions of airspace. For example, I will introduce airspace partitioning algorithms based on clustering, integer programming optimization, and computational geometry. I will conclude by discussing the development of a tablet-based tool that is intended to help air traffic controller supervisors configure airspace and controllers in current operations.
It is essential to evaluate and refine aerosol classification methods applied to passive satellite remote sensing. We have developed an aerosol classification algorithm (called Specified Clustering and Mahalanobis Classification, SCMC) that assigns an aerosol type to multi-parameter retrievals by spaceborne, airborne or ground-based passive remote sensing instruments [1]. The aerosol types identified by our scheme are pure dust, polluted dust, urban-industrialdeveloped economy, urban-industrialdeveloping economy, dark biomass smoke, light biomass smoke and pure marine. We apply the SCMC method to inversions from the ground-based AErosol RObotic NETwork (AERONET [2]) and retrievals from the space-borne Polarization and Directionality of Earths Reflectances instrument (POLDER, [3]). The POLDER retrievals that we use differ from the standard POLDER retrievals [4] as they make full use of multi-angle, multispectral polarimetric data [5]. We analyze agreement in the aerosol types inferred from both AERONET and POLDER and evaluate GEOS-Chem [6] simulations over the globe. Finally, we use in-situ observations from the SEAC4RS airborne field experiment to bridge the gap between remote sensing-inferred qualitative SCMC aerosol types and their corresponding quantitative chemical speciation. We apply the SCMC method to airborne in-situ observations from the NASA Langley Aerosol Research Group Experiment (LARGE, [7]) and the Differential Aerosol Sizing and Hygroscopicity Spectrometer Probe (DASH-SP, [8]) instruments; we then relate each coarsely defined SCMC type to a sum of percentage of individual aerosol species, using in-situ observations from the Particle Analysis by Laser Mass Spectrometry (PALMS, [9]), the Soluble Acidic Gases and Aerosol (SAGA, [10]), and the High - Resolution Time - of - Flight Aerosol Mass Spectrometer (HR ToF AMS, [11]).
The high complexity of modern aircraft and spacecraft requires elaborate Verification and Validation (V&V) approaches to make sure that such complex systems work properly and reliably. MARGInS is a framework for the analysis, understanding, and prediction of the behavior of a complex, hybrid system. MARGInS contains a set of machine learning and statistical algorithms for multivariate clustering, treatment learning, critical factor determination, time-series analysis, event prediction, and safety-boundary detection and characterization. The framework supports system testing and can be configured to find novel features in test suites, determine classes of behavior, propose new experiments that can efficiently explore and characterize the boundaries between classes of system behavior, and to create visualizations and reports.
An adaptive fuzzy system using the concept of the Adaptive Resonance Theory (ART) type neural network architecture and incorporating fuzzy c-means (FCM) system equations for reclassification of cluster centers was developed. The Adaptive Fuzzy Leader Clustering (AFLC) architecture is a hybrid neural-fuzzy system which learns on-line in a stable and efficient manner. The system uses a control structure similar to that found in the Adaptive Resonance Theory (ART-1) network to identify the cluster centers initially. The initial classification of an input takes place in a two stage process; a simple competitive stage and a distance metric comparison stage. The cluster prototypes are then incrementally updated by relocating the centroid positions from Fuzzy c-Means (FCM) system equations for the centroids and the membership values. The operational characteristics of AFLC and the critical parameters involved in its operation are discussed. The performance of the AFLC algorithm is presented through application of the algorithm to the Anderson Iris data, and laser-luminescent fingerprint image data. The AFLC algorithm successfully classifies features extracted from real data, discrete or continuous, indicating the potential strength of this new clustering algorithm in analyzing complex data sets. The hybrid neuro-fuzzy AFLC algorithm will enhance analysis of a number of difficult recognition and control problems involved with Tethered Satellite Systems and on-orbit space shuttle attitude controller.
We present a polynomial-scaling algorithm for the localized active space unitary selective coupled cluster singles and doubles (LAS-USCCSD) method. In this approach, cluster excitations are selected based on a threshold ϵ determined by the absolute gradients of the LAS-UCCSD energy with respect to cluster amplitudes. Using the generalized Wick’s theorem for multireference wave functions, we derive the gradient expression as a polynomial function of one-, two-, and three-body reduced density matrices and 1- and 2-electron integrals, valid for any multireference wave function. The resulting gradient implementation exhibits a memory scaling of 𝒪(N 6 ), with N spin orbitals in the combined active space of all fragments. The variational quantum eigensolver is used to optimize the selected cluster excitations on a quantum simulator. Furthermore, by plotting the energy error, defined as the difference between the LAS-USCCSD and corresponding CASCI energies, against the inverse cluster amplitude selection threshold (ϵ –1 ) for polyene chains containing 2 to 5 π-bond units, we establish a relationship between the energy error and the threshold. To further validate the accuracy of LAS-USCCSD, we computed the cis–trans isomerization energy of stilbene (a 20-qubit system) and the magnetic coupling constant of the tris-hydroxo-bridged chromium dimer [Cr 2 (OH) 3 (NH 3 ) 6 ] 3+ (evaluated as both 12- and 20-qubit systems) using the Qiskit-Qulacs simulator. Assessing such examples is important to determine the practical feasibility of quantum simulations for chemically realistic systems. Toward this goal, with the LAS-USCCSD algorithm we estimated the quantum resources required for simulating an active space of (30e,22o) in [Cr 2 (OH) 3 (NH 3 ) 6 ] 3+ , a size that remains beyond the reach of current quantum simulators for accurate treatment.
This paper presents a unified communication optimization framework for sparse triangular solve (SpTRSV) algorithms on CPU and GPU clusters. The framework builds upon a 3D communication-avoiding (CA) layout of Px × Py × Pz processes that divides a sparse matrix into Pz submatrices, each handled by a Px × Py 2D grid with block-cyclic distribution. We propose three communication optimization strategies: First, a new 3D SpTRSV algorithm is developed, which trades the inter-grid communication and synchronization with replicated computation. This design requires only one inter-grid synchronization, and the inter-grid communication is efficiently implemented with sparse allreduce operations. Second, broadcast and reduction communication trees are used to reduce message latency of the intra-grid 2D communication on CPU clusters. Finally, we leverage GPU-initiated one-sided communication to implement the communication trees on GPU clusters. With these nested inter- and intra-grid communication optimization strategies, the proposed 3D SpTRSV algorithm can attain up to 3.45x speedups compared to the baseline 3D SpTRSV algorithm using up to 2048 Cori Haswell CPU cores. In addition, the proposed GPU 3D SpTRSV algorithm can achieve up to 6.5x speedups compared to the proposed CPU 3D SpTRSV algorithm with Pz up to 64. Finally it is remarkable that the proposed GPU 3D SpTRSV can scale to 256 GPUs using the Perlmutter system while the existing 2D SpTRSV algorithm can only scale up to 4 GPUs.
Traditionally, nodes in a sensor network simply collect data and then pass it on to a centralized node that archives, distributes, and possibly analyzes the data. However, analysis at the individual nodes could enable faster detection of anomalies or other interesting events, as well as faster responses such as sending out alerts or increasing the data collection rate. There is an additional opportunity for increased performance if individual nodes can communicate directly with their neighbors. Previously, a method was developed by which machine learning classification algorithms could collaborate to achieve high performance autonomously (without requiring human intervention). This method worked for supervised learning algorithms, in which labeled data is used to train models. The learners collaborated by exchanging labels describing the data. The new advance enables clustering algorithms, which do not use labeled data, to also collaborate. This is achieved by defining a new language for collaboration that uses pair-wise constraints to encode useful information for other learners. These constraints specify that two items must, or cannot, be placed into the same cluster. Previous work has shown that clustering with these constraints (in isolation) already improves performance. In the problem formulation, each learner resides at a different node in the sensor network and makes observations (collects data) independently of the other learners. Each learner clusters its data and then selects a pair of items about which it is uncertain and uses them to query its neighbors. The resulting feedback (a must and cannot constraint from each neighbor) is combined by the learner into a consensus constraint, and it then reclusters its data while incorporating the new constraint. A strategy was also proposed for cleaning the resulting constraint sets, which may contain conflicting constraints; this improves performance significantly. This approach has been applied to collaborative clustering of seismic and infrasonic data collected by the Mount Erebus Volcano Observatory in Antarctica. Previous approaches to distributed clustering cannot readily be applied in a sensor network setting, because they assume that each node has the same view of the data set. A view is the set of features used to represent each object. When a single data set is partitioned across several computational nodes, distributed clustering works; all objects have the same view. But when the data is collected from different locations, using different sensors, a more flexible approach is needed. This approach instead operates in situations where the data collected at each node has a different view (e.g., seismic vs. infrasonic sensors), but they observe the same events. This enables them to exchange information about the likely cluster membership relations between objects, even if they do not use the same features to represent the objects.
Traditionally, nodes in a sensor network simply collect data and then pass it on to a centralized node that archives, distributes, and possibly analyzes the data. However, analysis at the individual nodes could enable faster detection of anomalies or other interesting events as well as faster responses, such as sending out alerts or increasing the data collection rate. There is an additional opportunity for increased performance if learners at individual nodes can communicate with their neighbors. In previous work, methods were developed by which classification algorithms deployed at sensor nodes can communicate information about event labels to each other, building on prior work with co-training, self-training, and active learning. The idea of collaborative learning was extended to function for clustering algorithms as well, similar to ideas from penta-training and consensus clustering. However, collaboration between these learner types had not been explored. A new protocol was developed by which classifiers and clusterers can share key information about their observations and conclusions as they learn. This is an active collaboration in which learners of either type can query their neighbors for information that they then use to re-train or re-learn the concept they are studying. The protocol also supports broadcasts from the classifiers and clusterers to the rest of the network to announce new discoveries. Classifiers observe an event and assign it a label (type). Clusterers instead group observations into clusters without assigning them a label, and they collaborate in terms of pairwise constraints between two events [same-cluster (mustlink) or different-cluster (cannot-link)]. Fundamentally, these two learner types speak different languages. To bridge this gap, the new communication protocol provides four types of exchanges: hybrid queries for information, hybrid "broadcasts" of learned information, each specified for classifiers-to-clusterers, and clusterers-to-classifiers. The new capability has the potential to greatly expand the in situ analysis abilities of sensor networks. Classifiers seeking to categorize incoming data into different types of events can operate in tandem with clusterers that are sensitive to the occurrence of new kinds of events not known to the classifiers. In contrast to current approaches that treat these operations as independent components, a hybrid collaborative learning system can enable them to learn from each other.
Adiabatic quantum computers are a promising platform for efficiently solving challenging optimization problems. Therefore, many are interested in using these computers to train computationally expensive machine learning models. We present a quantum approach to solving the balanced k-means clustering training problem on the D-Wave 2000Q adiabatic quantum computer. In order to do this, we formulate the training problem as a quadratic unconstrained binary optimization (QUBO) problem. Unlike existing classical algorithms, our QUBO formulation targets the global solution to the balanced k-means model. We test our approach on a number of small problems and observe that despite the theoretical benefits of the QUBO formulation, the clustering solution obtained by a modern quantum computer is usually inferior to the solution obtained by the best classical clustering algorithms. Nevertheless, the solutions provided by the quantum computer do exhibit some promising characteristics. We also perform a scalability study to estimate the run time of our approach on large problems using future quantum hardware. Finally, as a final proof of concept, we used the quantum approach to cluster random subsets of the Iris benchmark data set.
Not provided.
Liquid argon time projection chambers (LArTPCs) are a class of detectors that produce high resolution images of charged particles within their sensitive volume. In these images, the clustering of distinct particles into superstructures is of central importance to the current and future neutrino physics program. Electromagnetic (EM) activity typically exhibits spatially detached fragments of varying morphology and orientation that are challenging to efficiently assemble using traditional algorithms. Similarly, particles that are spatially removed from each other in the detector may originate from a common interaction. Graph neural networks (GNNs) were developed in recent years to find correlations between objects embedded in an arbitrary space. The graph particle aggregator (GrapPA) first leverages GNNs to predict the adjacency matrix of EM shower fragments and to identify the origin of showers, i.e., primary fragments. On the PILArNet public LArTPC simulation dataset, the algorithm achieves a shower clustering accuracy characterized by a mean purity of 99.4%, a mean efficiency of 99.6% and a primary identification accuracy of 99.8%. It yields a relative shower energy uncertainty of (4.1 + 1.4 / $\sqrt{\text{E(GeV)})}$% and a shower direction uncertainty of (2.1/ $\sqrt{\text{E(GeV)})}$°. Finally, the optimized algorithm is then applied to the related task of clustering particle instances into interactions and yields a mean purity of 99.8% and a mean efficiency of 99.5% for an interaction density of $\mathcal{O}(1)$ m –3 .
Where spatial boundaries between phenomena are diffuse, classification methods which construct mutually exclusive clusters seem inappropriate. The Fuzzy c-means (FCM) algorithm assigns each observation to all clusters, with membership values as a function of distance to the cluster center. The FCM algorithm is applied to AVHRR data for the purpose of classifying polar clouds and surfaces. Careful analysis of the fuzzy sets can provide information on which spectral channels are best suited to the classification of particular features, and can help determine likely areas of misclassification. General agreement in the resulting classes and cloud fraction was found between the FCM algorithm, a manual classification, and an unsupervised maximum likelihood classifier.
Where spatial boundaries between phenomena are diffuse, classification methods which construct mutually exclusive clusters seem inappropriate. The Fuzzy c-means (FCM) algorithm assigns each observation to all clusters, with membership values as a function of distance to the cluster center. The FCM algorithm is applied to AVHRR data for the purpose of classifying polar clouds and surfaces. Careful analysis of the fuzzy sets can provide information on which spectral channels are best suited to the classification of particular features, and can help determine like areas of misclassification. General agreement in the resulting classes and cloud fraction was found between the FCM algorithm, a manual classification, and an unsupervised maximum likelihood classifier.
Many quantum algorithms for machine learning require access to classical data in superposition. However, for many natural data sets and algorithms, the overhead required to load the data set in superposition can erase any potential quantum speedup over classical algorithms. Recent work by Harrow introduces a new paradigm in hybrid quantum-classical computing to address this issue, relying on coresets to minimize the data loading overhead of quantum algorithms. We investigated using this paradigm to perform k-means clustering on near-term quantum computers, by casting it as a QAOA optimization instance over a small coreset. We used numerical simulations to compare the performance of this approach to classical k-means clustering. We were able to find data sets with which coresets work well relative to random sampling and where QAOA could potentially outperform standard k-means on a coreset. However, finding data sets where both coresets and QAOA work well—which is necessary for a quantum advantage over k-means on the entire data set—appears to be challenging.
In an accompanying paper, we present a data-driven method for clustering in ‘integrals of motion’ space and apply it to a large sample of nearby halo stars with 6D phase-space information. The algorithm identified a large number of clusters, many of which could tentatively be merged into larger groups. The goal here is to establish the reality of the clusters and groups through a combined study of their stellar populations (average age, metallicity, and chemical and dynamical properties) to gain more insights into the accretion history of the Milky Way. To this end, we developed a procedure that quantifies the similarity of clusters based on the Kolmogorov–Smirnov test using their metallicity distribution functions, and an isochrone fitting method to determine their average age, which is also used to compare the distribution of stars in the colour–absolute magnitude diagram. Also taking into consideration how the clusters are distributed in integrals of motion space allows us to group clusters into substructures and to compare substructures with one another. We find that the 67 clusters identified by our algorithm can be merged into 12 extended substructures and 8 small clusters that remain as such. The large substructures include the previously known Gaia-Enceladus, Helmi streams, Sequoia, and Thamnos 1 and 2. We identify a few over-densities that can be associated with the hot thick disc and host a small metal-poor population. Especially notable is the largest (by number of member stars) substructure in our sample which, although peaking at the metallicity characteristic of the thick disc, has a very well populated metal-poor component, and dynamics intermediate between the hot thick disc and the halo. We also identify additional debris in the region occupied by Sequoia with clearly distinct kinematics, likely remnants of three different accretion events with progenitors of similar masses. Although only a small subset of the stars in our sample have chemical abundance information, we are able to identify different trends of [Mg/Fe] versus [Fe/H] for the various substructures, confirming our dissection of the nearby halo. We find that at least 20% of the halo near the Sun is associated to substructures. When comparing their global properties, we note that those substructures on retrograde orbits are not only more metal-poor on average but are also older. We provide a table summarising the properties of the substructures, as well as a membership list that can be used for follow-up chemical abundance studies for example.