Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “LIME”

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 343 records · Page 19

Glass Strengthening via High-Intensity Plasma-Arc Heating

The use of a high-intensity plasma-arc lamp was used to irradiate the surface of soda-lime silicate glass tiles to determine if an increase in strength could be achieved. The lamp had a power density of 3500 W/cm2, a processing area of 1 cm x 10 cm, irradiated near-infrared heating at a wavelength between 0.2 1.4 m, and was controlled to unidirectionally sweep across 50-mm-square tiles at a constant speed of 8 mm/s. Ring-on-ring (RoR) equibiaxial flexure and 4 pt uni-directional flexure testings of entire tiles were used to measure and compare failure stress distributions of treated and untreated glass. Even with non-optimized processing conditions, RoR failure stress increased by approximately 25% and the 4 pt bend failure stress increased by approximately 65%. Strengthening was due to a fire-polishing-like mechanism. The arc-lamp heat-treatment caused the location of the strength-limiting flaws in the 4-pt-bend tiles to change; namely, failure initiation occurred on the gage section surface for the treated glass whereas it occurred at a gage section edge for the untreated. Arc-lamp heat-treatment is attractive not only because it provides strengthening, but because it can (non-contact) process large amounts of glass quickly and inexpensively, and is a process that either a glass manufacturer or end-user can readily employ.

36 MATERIALS SCIENCE↗

Interpretation of Wild 2 Dust Fine Structure: Comparison of Stardust Aluminium Foil Craters to the Three-Dimensional Shape of Experimental Impacts by Artificial Aggregate Particles and Meteorite Powders

New experimental results show that Stardust crater morphology is consistent with interpretation of many larger Wild 2 dust grains being aggregates, albeit most of low porosity and therefore relatively high density. The majority of large Stardust grains (i.e. those carrying most of the cometary dust mass) probably had density of 2.4 g cm{sup -3} (similar to soda-lime glass used in earlier calibration experiments) or greater, and porosity of 25% or less, akin to consolidated carbonaceous chondrite meteorites, and much lower than the 80% suggested for fractal dust aggregates. Although better size calibration is required for interpretation of the very smallest impacting grains, we suggest that aggregates could have dense components dominated by {micro}m-scale and smaller sub-grains. If porosity of the Wild 2 nucleus is high, with similar bulk density to other comets, much of the pore-space may be at a scale of tens of micrometers, between coarser, denser grains. Successful demonstration of aggregate projectile impacts in the laboratory now opens the possibility of experiments to further constrain the conditions for creation of bulbous (Type C) tracks in aerogel, which we have observed in recent shots. We are also using mixed mineral aggregates to document differential survival of pristine composition and crystalline structure in diverse fine-grained components of aggregate cometary dust analogues, impacted onto both foil and aerogel under Stardust encounter conditions.

36 MATERIALS SCIENCE↗

Reactive binder and aggregate interfacial zones in the mortar of Tomb of Caecilia Metella concrete, 1C BCE, Rome

Integrated spectroscopic analyses and synchrotron X-ray microdiffraction investigations provide insights into the long-term reactivity of volcanic aggregate components and calcium-aluminum-silicate-hydrate (C-A-S-H) binder in mortar samples from the robust concrete of the sepulchral corridor of the Tomb of Caecilia Metella, 1st C BCE, Rome. The results of innovative micrometer-scale analytical maps indicate that Pozzolane Rosse tephra components–scoria groundmass, clinopyroxene, and leucite crystals–contributed to pozzolanic production of C-A-S-H binder and then remained reactive long after hydrated lime (Ca(OH) 2 ) was fully consumed. The C-A-S-H binding phase is reorganized into wispy halos and tendril-like strands, some with nanocrystalline preferred orientation or, alternatively, split into elongate features with short silicate chain lengths. These microstructures apparently record chemical and structural destabilization of C-A-S-H during excessive incorporation of Al 3+ and K + released through leucite dissolution. Resistance to failure may result from the intermittent toughening of interfacial zones of scoriae and clinopyroxene crystals with post-pozzolanic strätlingite and Al-tobermorite mineral cements and from long-term remodeling of the pozzolanic C-A-S-H binding phase. Roman builders’ selection of a leucite-rich facies of Pozzolane Rosse tephra as aggregate and construction of the tomb in an environment with high surface and ground water exposure apparently increased beneficial hydrologic activity and reactivity in the concrete.

36 MATERIALS SCIENCE↗

Effect of calcite filler on carbonation behavior in a synthesized C‐S‐H binder

Widespread implementation of portland limestone cements (PLC) in industry has raised new questions about their carbonation resistance due to higher initial limestone content than ordinary portland cement. While carbonation of portland cement and each of its hydrate phases has been studied at a fundamental level, the integral role of how limestone fillers interact chemically or physically during carbonation necessitates further study. In this study, a C-S-H binder comprised of a highly reactive zeolite pozzolan and lime was used to evaluate the effect of varying calcite additions on carbonation behavior at a microstructural level. Fundamental insight into hydration kinetics and carbonation mechanisms of hydraulic binders was obtained via analysis of the C-S-H and calcite microstructure. X-ray microcomputed tomography (X-ray CT) was used to quantify microstructural changes, and the results showed a decrease in shrinkage for increased calcite dosages, with a ∼50% reduction observed for a 45% wt. calcite replacement level. Crystallographic and thermal analyses measured compositional changes in the binder and confirmed that shrinkage was related to the decalcification of C-S-H. Furthermore, the addition of calcite to the C-S-H binder effectively reduced the liquid-to-solid ratio, porosity, and amount of C-S-H present in these binders. Ultimately, these results help elucidate how the carbonation risks of PLC can be mitigated by proportioning the mixture water only to the reactive component of the binder.

C-S-H↗

Regional analysis of aluminum and steel flows into the American automotive industry

Aluminum and steel represent the two most dominant metals in light duty vehicles, yet the flows of these materials into the American automotive industry have not been well characterized. This study proposes and implements a method for analyzing the flow of these metals into the automotive industry. We create a framework for performing regionally linked, sector specific material flow analyses and use this framework to trace flows of aluminum and steel entering the American automotive industry, focusing on flows downstream from raw material production. We show that automotive aluminum sheet and extrusions are sourced primarily from the NPCC (23%), SERC (20%), MRO (18%), and RFC (13%) NERC regions and a spatially unresolved Local region within the U.S. and Canada (18%). We determine that primary aluminum is largely from Canada (70%), nearly all from Quebec (69%). Further upstream, alumina and bauxite originate mostly from Brazil, Australia, and Jamaica. We also show that finished automotive steel is sourced primarily from the RFC (63%) and SERC (20%) regions. The crude steel supply similarly originates mainly from the RFC (69%) and SERC (7%) regions. Upstream raw materials including coke, coking coal, iron ore, lime, and steel scrap are primarily sourced from the U.S. with only direct reduced iron and pig iron used in electric arc furnace steel production coming mostly from outside the U.S. The framework developed here allows for increased spatial resolution of material flows, which can be used to develop more specific life cycle impact factors for life cycle assessments.

aluminum↗

Influence of copper to indium atomic ratio on the properties of Cu-In-Te based thin-film solar cells prepared by low-temperature co-evaporation

The influence of copper to indium atomic ratio (Cu/In) on the properties of Cu-In-Te based thin films and solar cells was investigated. The films (Cu/In = 0.38-1.17) were grown on both bare and Mo-coated soda-lime glass substrates at 250 Degree-Sign C by single-step co-evaporation using a molecular beam epitaxy system. Highly (112)-oriented CuInTe{sub 2} films were obtained at Cu/In ratios of 0.84-0.99. However, stoichiometric and Cu-rich films showed a poor film structure with high surface roughness. The films consist of polyhedron-shaped grains, which are related to the coexistence of a Cu{sub 2-x}Te phase, and significant evidence for the coexistence of the Cu{sub 2-x}Te phase in the stoichiometric and Cu-rich films is presented. KCN treatment was performed for the films in order to remove the Cu{sub 2-x}Te phase. The stoichiometric CuInTe{sub 2} thin films exhibited a high mobility above 50 cm{sup 2}/V s at room temperature after the KCN treatment. A preliminary solar cell fabricated using a 1.4-{mu}m-thick Cu-poor CuInTe{sub 2} thin film (Cu/In = 0.84, E{sub g} = 0.988 eV) yielded a total-area efficiency of 2.10%. The photovoltaic performance of the cell was improved after long-term ambient aging in dark conditions.

36 MATERIALS SCIENCE↗

Annealing effect for SnS thin films prepared by high-vacuum evaporation

Thin films of SnS are deposited onto molybdenum-coated soda lime glass substrates using the high-vacuum evaporation technique at a substrate temperature of 300 °C. The as-deposited SnS layers are then annealed in three different media: (1) H{sub 2}S, (2) argon, and (3) vacuum, for different periods and temperatures to study the changes in the microstructural properties of the layers and to prepare single-phase SnS photoabsorber films. It is found that annealing the layers in H{sub 2}S at 400 °C changes the stoichiometry of the as-deposited SnS films and leads to the formation of a dominant SnS{sub 2} phase. Annealing in an argon atmosphere for 1 h, however, causes no deviations in the composition of the SnS films, though the surface morphology of the annealed SnS layers changes significantly as a result of a 2 h annealing process. The crystalline structure, surface morphology, and photosensitivity of the as-deposited SnS films improves significantly as the result of annealing in vacuum, and the vacuum-annealed films are found to exhibit promising properties for fabricating complete solar cells based on these single-phase SnS photoabsorber layers.

36 MATERIALS SCIENCE↗

Properties of Cu(In,Ga,Al)Se{sub 2} thin films fabricated by magnetron sputtering

Cu(In,Ga,Al)Se{sub 2} (CIGAS) thin films were studied as an alternative absorber layer material to Cu(In{sub x}Ga{sub 1−x})Se{sub 2}. CIGAS thin films with varying Al content were prepared by magnetron sputtering on Si(100) and soda-lime glass substrates at 350 °C, followed by postdeposition annealing at 520 °C for 5 h in vacuum. The film composition was measured by an electron probe microanalyzer while the elemental depth profiles were determined by secondary ion mass spectrometry. X-ray diffraction studies indicated that CIGAS films are single phase with chalcopyrite structure and that the (112) peak clearly shifts to higher 2θ values with increasing Al content. Scanning electron microscopy images revealed dense and well-defined grains, as well as sharp CIGAS/Si(100) interfaces for all films. Atomic force microscopy analysis indicated that the roughness of CIGAS films decreases with increasing Al content. The bandgap of CIGAS films was determined from the optical transmittance and reflectance spectra and was found to increase as Al content increased.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Effect of the cadmium chloride treatment on RF sputtered Cd{sub 0.6}Zn{sub 0.4}Te films for application in multijunction solar cells

Single phase Cd{sub 0.6}Zn{sub 0.4}Te (CdZnTe) films of 1 μm thickness were deposited by radio frequency planar magnetron sputter deposition on commercial soda lime glass samples coated with fluorine-doped tin oxide and cadmium sulphide (CdS). The stack was then treated with cadmium chloride (CdCl{sub 2}) at different temperatures using a constant treatment time. The effect of the CdCl{sub 2} treatment was studied using optical, materials, and electrical characterization of the samples and compared with the as-deposited CdZnTe film with the same stack configuration. The band gap deduced from Tauc plots on the as-deposited CdZnTe thin film was 1.72 eV. The deposited film had good crystalline quality with a preferred orientation along the {111} plane. After the CdCl{sub 2} treatment, the absorption edge shifted toward longer wavelength region and new peaks corresponding to cadmium telluride (CdTe) emerged in the x-ray diffraction pattern. This suggested loss of zinc after the CdCl{sub 2} treatment. The cross sectional transmission electron microscope images of the sample treated at 400 °C and the energy dispersive elemental maps revealed the absence of chlorine along the grain boundaries of CdZnTe and residual CdTe. The presence of chlorine in the CdTe devices plays a vital role in drastically improving the device performance which was not observed in CdZnTe samples treated with CdCl{sub 2}. The loss of zinc from the surface and incomplete recrystallization of the grains together with the presence of high densities of stacking faults were observed. The surface images using scanning electron microscopy showed that the morphology of the grains changed from small spherical shape to large grains formed due to the fusion of small grains with distinct grain boundaries visible at the higher CdCl{sub 2} treatment temperatures. The absence of chlorine along the grain boundaries, incomplete recrystallization and distinct grain boundaries is understood to cause the poor performance of the fabricated devices.

36 MATERIALS SCIENCE↗

Intense pulsed light annealing of copper zinc tin sulfide nanocrystal coatings

A promising method for forming the absorber layer in copper zinc tin sulfide [Cu{sub 2}ZnSnS{sub 4} (CZTS)] thin film solar cells is thermal annealing of coatings cast from dispersions of CZTS nanocrystals. Intense pulsed light (IPL) annealing utilizing xenon flash lamps is a potential high-throughput, low-cost, roll-to-roll manufacturing compatible alternative to thermal annealing in conventional furnaces. The authors studied the effects of flash energy density (3.9–11.6 J/cm{sup 2}) and number of flashes (1–400) during IPL annealing on the microstructure of CZTS nanocrystal coatings cast on molybdenum-coated soda lime glass substrates (Mo-coated SLG). The annealed coatings exhibited cracks with two distinct linear crack densities, 0.01 and 0.2 μm{sup −1}, depending on the flash intensity and total number of flashes. Low density cracking (0.01 μm{sup −1}, ∼1 crack per 100 μm) is caused by decomposition of CZTS at the Mo-coating interface. Vapor decomposition products at the interface cause blisters as they escape the coating. Residual decomposition products within the blisters were imaged using confocal Raman spectroscopy. In support of this hypothesis, replacing the Mo-coated SLG substrate with quartz eliminated blistering and low-density cracking. High density cracking is caused by rapid thermal expansion and contraction of the coating constricted on the substrate as it is heated and cooled during IPL annealing. Finite element modeling showed that CZTS coatings on low thermal diffusivity materials (i.e., SLG) underwent significant differential heating with respect to the substrate with rapid rises and falls of the coating temperature as the flash is turned on and off, possibly causing a build-up of tensile stress within the coating prompting cracking. Use of a high thermal diffusivity substrate, such as a molybdenum foil (Mo foil), reduces this differential heating and eliminates the high-density cracking. IPL annealing in presence of sulfur vapor prevented both low- and high-density cracking as well as blistering. However, grain growth was limited even after annealing with 400 flashes. This lack of grain growth is attributed to a difficulty of maintaining high sulfur vapor pressure and absence of alkali metal impurities when Mo foil substrates are used.

36 MATERIALS SCIENCE↗

Dehydration of Uranyl Nitrate Hexahydrate to Uranyl Nitrate Trihydrate under Ambient Conditions as Observed via Dynamic Infrared Reflectance Spectroscopy

the hexahydrate [UO 2 (NO 3 ) 2 (H 2 O) 6 ] (UNH) and the trihydrate [UO 2 (NO 3 ) 2 (H 2 O) 3 ] (UNT) forms. Their stabilities depend on both relative humidity and temperature. Both phases have previously been studied by infrared transmission spectroscopy, but the data were limited by both instrumental resolution and the ability to prepare the samples as pellets without desiccating them. We report time-resolved infrared (IR) measurements using an integrating sphere that allow us to observe the transformation from the hexahydrate to the trihydrate simply by flowing dry nitrogen gas over the sample. Hexahydrate samples were prepared and confirmed via known XRD patterns, then measured in reflectance mode. The hexahydrate has a distinct uranyl asymmetric stretch band at 949.0 cm -1 that shifts to shorter wavelengths and broadens as the sample dehydrates and recrystallizes to the trihydrate, first as a blue edge shoulder but ultimately resulting in a doublet band with reflectance peaks at 966 and 957 cm -1 . The data are consistent with transformation from UNH to UNT since UNT has two non-equivalent UO 2 2+ sites. The dehydration of UO 2 (NO 3 ) 2 (H 2 O) 6 to UO 2 (NO 3 ) 2 (H 2 O) 3 is both a morphological and structural change that has the lustrous lime green crystals changing to the dull greenish yellow of the trihydrate. Crystal structures and phase transformation were confirmed theoretically using DFT calculations and experimentally via microscopy methods. Both methods showed a transformation with two distinct sites for the uranyl cation in the trihydrate, as opposed to a single crystallographic site in the hexahydrate.

36 MATERIALS SCIENCE↗

Characterization of Cu2ZnSnS4 Thin Films Prepared by the Sulfurization of Co-sputtered Metal Precursors

In this work, we report on the synthesis and characterization of Cu2ZnSnS4 (CZTS) thin films prepared by annealing of co-sputtered metal precursors in sulfur atmosphere. Radio-frequency magnetron sputtering was applied to deposit the metal layers from single metal targets on Mo-coated soda-lime glass substrates. The chemical composition of the precursors was controlled by varying the sputtering working power, resulting in films with various compositions. X-ray fluorescence was used to determine the elemental concentration of these metal films. The metal precursors were then converted into CZTS in a tube furnace using different sulfurization conditions to investigate the effect of the annealing process on the properties of the final CZTS films. Film structural characterization and phase identification results were supported by X-ray diffraction (XRD) and Raman spectroscopy. Surface and cross-sectional film morphology was carried out by scanning electron microscopy (SEM). For the sulfurized films, significant Sn loss was noticed. However, the loss of Sn was successfully controlled by depositing precursors with an excess of Sn. After optimizing the composition of the metal precursor, XRD and Raman scattering results revealed single-phase CZTS films without clear signs of secondary phases. SEM showed improved morphology in the form of dense structures and smooth surfaces for the films sulfurized at 600 degrees C. Our first solar cell, based on a CZTS film originating from a precursor sulfurized at 550 degrees C for 60 min, showed an open-circuit voltage of 471 mV, a short-circuit current density of 9.92 mA/cm-2, a fill factor of 36.9%, and an efficiency of 1.72%.

14 SOLAR ENERGY↗

Temperature and Excitation Dependence of Recombination in CIGS Thin Films with High Spatial Resolution

Alkali post-deposition treatments significantly improve the performance of CuInGaSe2 (CIGS) devices, but there is still room for improvement. Here, we investigate the effects of potassium fluoride alkali post-deposition treatment on the defect chemistry and recombination at grain boundaries and grain interiors using temperature- and injection-dependent cathodoluminescence (CL) spectrum imaging. We study CIGS thin films grown on alkali-free sapphire substrates to isolate the effects of alkali treatment from alkali metals that can diffuse from standard soda-lime glass substrates. We find that alkali treatment affects the energy and temperature dependence of the luminescence peaks, as well as the defect activation energies. CL spectrum images reveal that the luminescence transitions at grain boundaries have a distinct power dependence after alkali treatment and substantially different defect chemistry. This work shows that temperatureand injection-dependence CL spectrum images can provide unique insight into the defect chemistry and recombination behavior of CIGS thin films.

14 SOLAR ENERGY↗

Stability of CIGS Solar Cells and Component Materials Evaluated by a Step-Stress Accelerated Degradation Test Method: Preprint

A step-stress accelerated degradation testing (SSADT) method was employed for the first time to evaluate the stability of CuInGaSe2 (CIGS) solar cells and device component materials in four Al-framed test structures encapsulated with an edge sealant and three kinds of backsheet or moisture barrier film for moisture ingress control. The SSADT exposure used a 15 degrees C and then a 15% relativehumidity (RH) increment step, beginning from 40 degrees C/40%RH (T/RH = 40/40) to 85 degrees C/70%RH (85/70) as of the moment. The voluminous data acquired and processed as of total DH = 3956 h with 85/70 = 704 h produced the following results. The best CIGS solar cells in sample Set-1 with a moisture-permeable TPT backsheet showed essentially identical I-V degradation trend regardless of theAl-doped ZnO (AZO) layer thickness ranging from standard 0.12 um to 0.50 um on the cells. No clear 'stepwise' feature in the I-V parameter degradation curves corresponding to the SSADT T/RH/time profile was observed. Irregularity in I-V performance degradation pattern was observed with some cells showing early degradation at low T/RH < 55/55 and some showing large Voc, FF, and efficiency degradation due to increased series Rs (ohm-cm2) at T/RH = 70/70. Results of (electrochemical) impedance spectroscopy (ECIS) analysis indicate degradation of the CIGS solar cells corresponded to increased series resistance Rs (ohm) and degraded parallel (minority carrier diffusion/recombination) resistance Rp, capacitance C, overall time constant Rp*C, and 'capacitor quality' factor (CPE-P), whic were related to the cells' p-n junction properties. Heating at 85/70 appeared to benefit the CIGS solar cells as indicated by the largely recovered CPE-P factor. Device component materials, Mo on soda lime glass (Mo/SLG), bilayer ZnO (BZO), AlNi grid contact, and CdS/CIGS/Mo/SLG in test structures with TPT showed notable to significant degradation at T/RH = 70/70. At T/RH = 85/70, substantial blistering of BZO layers on CIGS cell pieces was observed that was not seen on BZO/glass, and a CdS/CIGS sample displayed a small darkening and then flaking feature. Additionally, standard AlNi grid contact was less stable than thin Ni grid contact at T/RH = 70/70. The edge sealant and moisture-blocking films were effective to block moisture ingress, as evidenced by the good stability of most CIGS solar cells and device components at T/RH = 85/70 for 704 h, and by preservation of the initial blue color on the RH indicator strips. The SSADT experiment is ongoing to be completed at T/RH = 85/85.

14 SOLAR ENERGY↗

The influence of substrate temperature on the structural and optical properties of ZnS thin films

Thin films of ZnS were deposited on soda lime glass substrates by a modified close-space sublimation technique. The change in optical and structural properties of the films deposited at various substrate temperatures (150-450 Degree-Sign C) was investigated. X-ray diffraction spectra showed that films were polycrystalline in nature having cubic structure oriented only along (111) plan. The crystallinity of films increased with the substrate temperature up to 250 Degree-Sign C. However, crystallinity decreased with further increase of substrate temperature and films became amorphous at 450 Degree-Sign C. The atomic force microscopy data revealed that the films become more uniform and dense with the increase of substrate temperature. Optical properties of the films were determined from the transmittance data using Swanepoel model. It was observed that the energy band gap is increased from 3.52 to 3.65 eV and refractive index of the films are decreased with the increase of substrate temperature. Moreover, considerable improvement in blue response of the films was noticed with increasing substrate temperature.

36 MATERIALS SCIENCE↗

Deposition and characterization of molybdenum thin films using dc-plasma magnetron sputtering

Molebdenum (Mo) thin films were deposited on well-cleaned soda-lime glass substrates using DC-plasma magnetron sputtering. In the design of experiment deposition was optimized for maximum beneficial characteristics by monitoring effect of process variables such as deposition power (100–200 W). Their electrical, structural and morphological properties were analyzed to study the effect of these variables. The electrical resistivity of Mo thin films could be reduced by increasing deposition power. Within the range of analyzed deposition power, Mo thin films showed a mono crystalline nature and the crystallites were found to have an orientation along [110] direction. The surface morphology of thin films showed that a highly dense micro structure has been obtained. The surface roughness of films increased with deposition power. The adhesion of Mo thin films could be improved by increasing the deposition power. Atomic force microscopy was used for the topographical study of the films and to determine the roughness of the films. X-ray diffractrometer and scanning electron microscopy analysis were used to investigate the crystallinity and surface morphology of the films. Hall effect measurement system was used to find resistivity, carrier mobility and carrier density of deposited films. The adhesion test was performed using scotch hatch tape adhesion test. Mo thin films prepared at deposition power of 200 W, substrate temperature of 23°C and Ar pressure of 0.0123 mbar exhibited a mono crystalline structure with an orientation along (110) direction, thickness of ∼550 nm and electrical resistivity value of 0.57 × 10{sup −4} Ω cm.

36 MATERIALS SCIENCE↗

Physicochemical Properties and Structure of the Bone Matrix in Simulated Tuberculous Osteitis

The physicochemical properties and structure of the bone matrix are studied in a case of simulated tuberculous osteitis without treatment and after a full course of specific antibacterial therapy. Using X-ray diffraction, infrared spectroscopy, thermal analysis, and scanning electron microscopy, we revealed that tuberculous osteitis causes fine disintegration of the bone matrix due to an increase in nonstoichiometric hydroxyapatite, formation of amorphous calcium phosphates, and a decrease of the organic phase, which is accompanied by embrittlement of the bone matrix. Excessive growth of carbonate-hydroxyapatite crystals of a mixed AB substitution type leads to excessive osteogenesis, accompanied by the uncontrolled growth of bone trabeculae. The presence of “liming” regions in the crystal lattice of hydroxyapatite increases the “binding” properties of the bone matrix in which mycobacteria are immobilized and removed in the composition of detritus.

60 APPLIED LIFE SCIENCES↗

Topological Interpretability for Deep Learning

With the growing adoption of AI-based systems across everyday life, the need to understand their decision-making mechanisms is correspondingly increasing. The level at which we can trust the statistical inferences made from AI-based decision systems is an increasing concern, especially in high-risk systems such as criminal justice or medical diagnosis, where incorrect inferences may have tragic consequences. Despite their successes in providing solutions to problems involving real-world data, deep learning (DL) models cannot quantify the certainty of their predictions. These models are frequently quite confident, even when their solutions are incorrect. This work presents a method to infer prominent features in two DL classification models trained on clinical and non-clinical text by employing techniques from topological and geometric data analysis. We create a graph of a model's feature space and cluster the inputs into the graph's vertices by the similarity of features and prediction statistics. We then extract subgraphs demonstrating high-predictive accuracy for a given label. These subgraphs contain a wealth of information about features that the DL model has recognized as relevant to its decisions. We infer these features for a given label using a distance metric between probability measures, and demonstrate the stability of our method compared to the LIME and SHAP interpretability methods. This work establishes that we may gain insights into the decision mechanism of a DL model. This method allows us to ascertain if the model is making its decisions based on information germane to the problem or identifies extraneous patterns within the data.

Spannaus, Adam↗