Search

The discovery of nonlinear variation patterns in high-dimensional profile data is an important task in many quality control and manufacturing settings. We present an automated method for discovering nonlinear variation patterns using deep autoencoders. The approach provides a functional mapping from a low-dimensional representation to the original spatially-dense feature space of the profile data that is both interpretable and efficient with respect to preserving information. We compare our deep autoencoder approach to several other methods for discovering variation patterns in profile data. Our results indicate that deep autoencoders consistently outperform the alternative approaches in reproducing the original profiles from the learned variation sources.

Most machine learning algorithms possess hyperparameters. For example, an artificial neural network requires the determination of the number of hidden layers, nodes, and many other parameters related to the model fitting process. Despite this, there is still no clear consensus on how to tune them. The most popular methodology is an exhaustive grid search, which can be highly inefficient and sometimes infeasible. Another common solution is to change one hyperparameter at a time and measure its effect on the model’s performance. However, this can also be inefficient and does not guarantee optimal results since it ignores interactions between the hyperparameters. In this paper, we propose to use the Design of Experiments (DOE) methodology (factorial designs) for screening and Response Surface Methodology (RSM) to tune a machine learning algorithm’s hyperparameters. An application of our methodology is presented with a detailed discussion of the results of a random forest case-study using a publicly available dataset. Benefits include fewer training runs, better parameter selection, and a disciplined approach based on statistical theory.

Autoassociative neural networks (ANNs) have been proposed as a nonlinear extension of principal component analysis (PCA), which is commonly used to identify linear variation patterns in high-dimensional data. While principal component scores represent uncorrelated features, standard backpropagation methods for training ANNs provide no guarantee of producing distinct features, which is important for interpretability and for discovering the nature of the variation patterns in the data. Here, we present an alternating nonlinear PCA method, which encourages learning of distinct features in ANNs. A new measure motivated by the condition of orthogonal loadings in PCA is proposed for measuring the extent to which the nonlinear principal components represent distinct variation patterns. We demonstrate the effectiveness of our method using a simulated point cloud data set as well as a subset of the MNIST handwritten digits data. The results show that standard ANNs consistently mix the true variation sources in the low-dimensional representation learned by the model, whereas our alternating method produces solutions where the patterns are better separated in the low-dimensional space.

The development of ceramic composites with three-dimensional fiber reinforcement architectures formed by textile methods has led to the potential for active shape-morphing surfaces that can operate in high temperature and variable pressure environments. This technology is of particular interest for hypersonic applications, where SCRAM jet engines require variable inlet geometry to achieve efficient flight over realistic flight profiles and variable flight conditions. The experiments reported here show that significant shape morphing can be achieved and good control of the shape sustained even in the presence of large temperature and pressure gradients. Experiments were carried out using a subscale morphing hypersonic inlet with rectangular cross-section in a Mach 8 wind tunnel facility with a total temperature of 800 K. The upper surface of the inlet consisted of a C–SiC composite plate (0.7 mm thick, 37.5 cm long, and 11 cm wide) connected to five actuators through a triangular truss support structure. The lower surface was a flat plate instrumented with an array of pressure taps along the flow centerline. As the shape varied, the surface contour was reliably controlled for high efficiency, low loss compression. A factor of six inlet area ratio variation was achieved and good agreement with model predictions was observed.

A cDNA (DRC1, differentially regulated clone 1) was obtained from differential-display polymerase chain reaction (PCR) of brook trout ovarian tissue stimulated with phorbol-12-myristate-13-acetate (PMA) and A23187. Using 5' RACE (rapid amplification of cDNA ends), two full-length clones were obtained from DRC1 that were 425 and 660 base pairs long and contained the same open reading frame. On Northern blots, DRC1 hybridized with two ovarian mRNAs of 0.45 and 0.7 kb that were significantly suppressed in the presence of PMA and/or A23187. The mRNAs were not observed in ovaries prior to the resumption of meiosis but were present during ovulation and 24 hr after ovulation. Of other trout tissues tested by Northern blotting, the expression of DRC1-related transcripts also was extremely high in the liver. Based on the full-length cDNAs obtained from RACE, these mRNAs presumably encode an 88-amino-acid protein (DRTP1, differentially regulated trout protein 1) that is homologous to a gene superfamily composed of snake venom neurotoxins, a CD59 complement regulatory protein, Ly-6 alloantigens, and a urokinase-type plasminogen activator receptor. To our knowledge, this is the first description of this type of cDNA from a nonmammalian source other than snake venom. In view of the sequence homology and tissue expression of DRTP1, a possible function of this protein may be to regulate the complement system in trout.

DRS Technologies has designed and delivered Thermal Weapon Site (TWS) and Driver's Viewer Enhancer (DVE) system using its U3500 (320x240) and U6000 (640x480) 1-mil detector arrays. The detectors are modified to enhance its manufacturability, thermal time constant, package life time, and its reliability under shock and vibration to meet TWS and DVE requirements. The U6000 array operates at 30 Hz frame rate with NETD less than 50 mK normalized to F/1.0 optics. At a saving to the system weight and power, these arrays operate from -40°C to +65°C without the aid of a TE cooler. This is accomplished through a combination of sensor calibration and smart ROIC architecture.

The recently considered concepts for a wharf development identified a number of options, including conventional wharf topside modules on steel tubular piled foundations, steel-jacket-type modules anchored to the seabed, concrete caissons, and a hybrid wharf substructure with a Gravity Base Structure (GBS) connected into a steel jacket sub-frame. Due to the unprecedented demand for site-based skilled labour and a marine construction plant on the Australian coast from numerous major resource projects, further consideration was given to the pre-assembled hybrid wharf alternative and the associated cost, fabrication yard availability, and transport issues. To overcome the potential limits and risks of constructing and sea-towing a concrete base structure, the preferred option is to use a multi-cell steel base instead of concrete. The GBS method of construction is to use mature technology in the offshore oil and gas industry and can take advantage of modularisation of the substructure and topsides by fully fitting out larger units in overseas fabrication yards. For alternate wharf applications, the GBS has the potential of allowing pre-assembly and pre-commissioning of equipment and systems, or the ability to enhance the substructure installation in readiness for topsides installation (either floatover integral topsides or modular lift). It also opens up a wide choice of existing fabrication yards and shops in China or Korea that either fabricate wharf or jacket substructure components now, or are in close proximity to existing loading-dock facilities. This can result in considerable schedule and cost savings by reducing site (offshore) labour and plant costs.

A large resource of unpublished technical data exists in the form of reports submitted by industry under the Toxic Substances Control Act (TSCA) to the U.S. Environmental Protection Agency. The TSCA Test Submissions (TSCATS) database was recently developed and made available to the public to allow computerized searching of test submissions according to specific chemical identity or type of study. Studies are categorized into three broad subject areas (health effects, environmental effects, environmental fate) and additional indexing terms are assigned from a controlled vocabulary describing the experimental protocol and certain observations of the test. Records from the on-line database also contain reference information to assist in locating the source document, and to identify the submitting organization and reason for submission of the test data.

The crystal and molecular structure of hexaureatitanium(III) iodide, [TiCCOCNHg)^)^]!^!*^ been determined from a complete set of counter collected, three-dimensional x-ray diffraction intensity data. Refinement by conventional Fourier and least squares techniques was to a final residual of 0.064- The complex crystallizes in the space group R3c with a = 17 67A, c = 14 ISA, and six formula units in the o o unit cell (/> = / o bs / calc 2.06. /) _ = 2.03 g/cm ) The titanium atom 3 occupies a special position of D symmetry and is surrounded by six oxygen atoms at the vertices of a slightly distorted octahedron, the three ligand molecules above the plane of the titanium being rotated through an angle of 5.55° with respect to the three below the plane. The iodide ions form essentially linear chains parallel to the crystallographic three-fold axis Ligand field calculations based on the molecular structure so determined have been used to interpret the spectral and magnetic properties of the complex. It has been found that for the form of trigonal distortion observed in hexaureatltanium(III) iodide; all ligand field splittings can be expressed in terms of the single parameter Dq. The ligand field calculations indicate an orbital singlet (^A ) ground term and the recently published single crystal

NOTE: Text or symbols not renderable in plain ASCII are indicated by [...]. Abstract is included in .pdf document. A cancellation machine has been constructed to automatically produce the cancellation required by the method of Leighton in making Doppler movies. Various adaptations of the machine allow rapid data reduction to the 16 mm print required for cinema viewing. The preliminary data gathered in 1966 and 1967 has been reduced. These data consist mostly of He Doppler movie pairs, and some Zeeman movie pairs, of sunspot regions. A new calibration technique for inferring Doppler velocities from cancellation densities has been developed. This technique extends the calibration to large Doppler shifts with no less in accuracy. Doppler movies taken in H[alpha],.070 nm from the core, show a long lived flow region connecting spots of opposite magnetic polarity in newly developing sunspot regions. The flow regions consist of a series of parallel arched filaments about 2 Mm thick by 23 [...] 8 Mm long which are found also in the H[alpha] core superimposed upon bright plage. The material in the arches is shown to flow downward along the legs of the arch from its center. The average height of the arches is 2 to 4 Mm depending upon the method of calculation. Velocities of 35 km/s inferred in the arches are consistent with those expected from frictionless descent from the top of the arch. An ascent velocity of about 5 km/s of the arch center is also consistent with the data. The material flow is not a mass flow from one leg of the arch to another as previously reported by other observers. A discussion of other minor discrepancies of this work with that of previous observers points out certain possible sources of error in both their work and this. A physical model for the arches is proposed, whereby they consist of material trapped by magnetic lines of force which are emerging from the photosphere in the forming sunspot region. This material then flows downward along the arched lines of force as they rise. The subsurface magnetic field energy is shown to be sufficient to raise the material to the heights inferred for the arches, but insufficient to raise the temperature of the photosphere at the feet of the arches by more than about 7° K. The average lifetime of an arch is about thirty minutes yet the rise and decay times indicated by the data are shown to be about five minutes. Two possible explanations are offered for this discrepancy, one involving a great optical depth and the other a series of arches superimposed. Several suggestions are made for further persuing this phenomenon. The study of Tile of velocity features in the Ha chromosphere of the quiet sun has been extended to active regions. The results for network regions near the sunspots corroborate those of Title. A mean lifetime of 215 ± 9 s was found for upflow events and 371 ± 30 s for downflow. In the regions covered by bright plage in the H[alpha] core photographs the mean lifetime in a majority of both upflow and downflow events was 217±71 s and 211 ± 91 s respectively.