Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “explainability”

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 1,081 records · Page 60

High-temperature Hydrogen Chloride Releases from Mixtures of Sodium Chloride with Sulfates: Implications for the Chlorine-Mineralogy as Determined by the Sample Analysis at Mars Instrument on the Curiosity Rover in Gale Crater, Mars

Hydrogen chloride releases above 500 °C occurred in several samples analyzed by the Sample Analysis at Mars (SAM) evolved gas analyzer on the Curiosity rover in Gale crater. These have been attributed to reactions between chlorides (original or from oxychlorine decomposition) and water. Some of these HCl releases that peaked below the melting temperature of common chlorides did not co-evolve with oxygen or water, and were not explained by laboratory analog work (Figure 1). Therefore, these HCl releases were not caused by MgCl2 or soley due to reactions between water and melting chlorides. The goal of this work was to explain the HCl releases that did not co-evolve with oxygen or water and occurred below the melting point of common chlorides, which have not been explained by previous laboratory analog work. This work specifically evaluates the role of evolved SO2 in the production of HCl.

Clark, J.V.↗

The Cognition and Fine Motor Skills Test Batteries: Normative Data and Interdependencies

Space mission success and safety relies upon astronaut functional state. Since spaceflight stressors affect cognitive processing and fine motor skills, NASA requires that measures of performance of these things remain within clinically accepted values (NASA STD 3001). NASA is in the process of developing two test batteries for the assessment of crew cognitive and fine motor skills before, during and after spaceflight. Toward that goal, the current project collected normative scores in 91 “astronaut-like” military and civilian pilots. The Cognition Test Battery (CTB) contains ten sub-tests that measure a range of cognitive abilities. For five of the ten CTB sub-tests, we propose scores to improve the battery’s sensitivity. Among the ten sub-tests, response times were more highly correlated than accuracy scores. Principle component analysis of the correlations revealed that the first response time factor could explain over 40% of the total variance and appeared to represent the tendency of observers to try to respond more quickly. The first accuracy factor (explaining only 20%) gave a high weight to the higher level cognitive sub-tests and a negative weight to tasks associated with motor and lower level cognitive processing. The Fine Motor Skills (FMS) test battery contains four sub-tests (Tracking, Pointing, Tracing, Rotating) performed on an Apple iPad tablet computer. Principle component analysis on the sub-test response time correlations revealed that the first two factors accounted for ~80% of the variance in performance. The first component captured overall speed on all four of the sub-tests. The second factor separated the sub-tasks into two groups (Drag-Point vs Trace-Rotate). Previous work found the first group response times correlated with that of a standard peg board task, while those of the other group did not. Correlations were computed between the first FMS factor and the response time and accuracy scores from each CTB sub-task. Performing fine motor behaviors rapidly was significantly correlated with the ability to perform many of the CTB sub-tests rapidly. This ability cannot be simple motor speed since scores on the Psychomotor Vigilance Test (PVT) subtask did not correlate with the ability to perform the other tasks rapidly. Speed on fine motor skills correlated significantly with accuracy on the short-term-memory sub-test. We hypothesize that eye movements, which can be regarded as a fine motor skill, may explain this relationship.

cognition↗

Thermodynamic constraints on the textural evolution of eucrite, EET 90020

Basaltic eucrites, which are possibly parts of the Vestan crust, formed as either lava flows or intrusions, and can offer insights into early crust formation. However most eucrites have experienced thermal metamorphism, which exacerbates the challenges of understanding these samples. A variety of petrogenetic models such as, partial melting of a primitive source, fractional crystallization of magmas emplaced in a primitive crust, and/or partial melting of a eucritic source coupled with melt mixing and assimilation have been proposed in order to explain observed petrologic and chemical characteristics of eucrites. EET 90020 is an unbrecciated eucrite that has experienced significant thermal metamorphism, with temperatures of metamorphic equilibration calculated to be ~840-1040°C. Its petrologic history remains contentious, in part due to differences in trace element analyses [10,17, this work], which has led to multiple metamorphic interpretations [8,10-11]. Here, we combined petrologic observations, chemical analyses, and thermodynamic modeling, to interpret micro-domain textures identified in EET 90020 and better constrain its petrologic history. Ultimately, the development of such textures is a direct result of the geologic processes operating on a young Vesta or similar asteroid. Sample description: We identified three textural domains in EET 90020 (Figure 1); (I) A coarse grain domain with granoblastic plagioclase and pyroxene, (II) a medium grain domain with curved grain boundaries between plagioclase and pyroxene, and (III) a fine grain domain dominated by tridymite that fills in interstitial space between spherical plagioclase. Major element chemistry in phases is consistent across domains. Pyroxene have distinctive high Ca lamellae (Wo40.92En28.1Fs31.0) forming from the low-Ca pigeonite host (Wo3.5En33.7Fs62.8). There are small amounts of fayalitic olivine along pyroxene/oxide grain boundaries, where pyroxenes have reacted with ilmenite/chromite clasts. Plagioclase is anorthitic (Avg ~An88), with little variation (An86-92.5). Trace element bulk rock analyses were collected via ICP-MS and individual mineral analyses with LA-ICP-MS (Figure 2). The bulk sample has an observable enrichment in light rare earth elements, with a notable depletion in Eu. Plagioclase is enriched in LREEs, depleted in HREEs, and has a positive Eu anomaly while pyroxene is enriched in HREEs with a negative Eu anomaly. Variations in trace element analyses is likely influenced by the presence of phosphates because are highly enriched in REEs and have a dramatic effect on the bulk domain compositions. Thermodynamic modeling: The software package, Perple_X, was used to calculate mineral phase equilibria over a range of conditions using a Gibbs free energy minimization approach [12]. Models used all major/minor elements except P2O5. Thermodynamic properties from [13] were used to constrain endmember phase stabilities. Activity models were used to describe mixing in phases with solid solution. Isochemical P-T phase diagrams were constructed for the bulk thin section, and the coarse, medium and fine grain domain compositions. Domain compositions were determined using microprobe analyses of phases identified in each domain and observed modal mineral abundancies. Thermodynamically stable pyroxene compositions were extracted from model results and compared to pyroxene compositions collected via EMPA, allowing us to calculate temperatures for metamorphic equilibrium. Results: Results that use the coarse domain composition (Fig. 3a) demonstrate overlap at T ~ 1020°C indicating high T equilibration. There was no overlap for the medium and fine grain domains, or the bulk composition (e.g., Fig. 3b) indicating that these are not equilibrium assemblages. These results were replicated using data from [8,20]. Discussion: The disequilibrium results from the bulk rock, fine grain and medium grain models imply that EET 90020 was modified during and/or after peak metamorphism (open system behavior). However, this is in contrast to the coarse domain where it appears as if equilibrium was maintained. We suggest that this conflict is explained by highly localized equilibrium occurring at millimeter length-scales, which has been observed in terrestrial metamorphism where aqueous fluid infiltrate [14] or melt loss occurs [15]. It is unlikely that aqueous alteration was significant in modifying the bulk composition of EET 90020, because it lacks hydrated minerals. However, the presence of melt is supported by the presence of an abundance of curved grain boundaries and spherical grains in the medium and fine grain domains This contrasts the coarse domain, where grain boundaries are angular. Given the evidence for disequilibrium in the medium and fine grain domains, it would be inappropriate to draw additional conclusions about textural evolution using thermodynamic models of those domains. However, modeling results of the coarse grain domain, where equilibrium was maintained, can provide useful insights into the geologic evolution of EET 90020. We approximate maximum metamorphic temperatures, using the Ca component in pyroxene, at ~1020°C (Fig. 4), which is consistent with previous two-pyroxene thermometry [11]. At this temperature, up to 10 vol % melt can be produced in the coarse grain domain, which is at the boundary of minimum melt needed in order to segregate and form a melt network [16]. Tridymite and plagioclase would be the first minerals to melt out during heating starting ~990°C. This is consistent with the observed melt textures formed by tridymite and plagioclase in the fine grain domain. We suggest that a partial melting model best explains the textural development of EET 90020. Upon heating, melt generated was either trapped, resulting in no net change in the localized rock composition, or migrated, resulting in disequilibrium between pyroxene and the surrounding matrix. A positive Eu anomaly in plagioclase indicates that all of the plagioclase was melted previously. This is consistent with previous studies that have suggested that EET 90020 represents a residual eucrite that experienced partial melting and subsequent melt loss [8]. We speculate that partial melting could have occurred when EET 90020 was heated in the asteroid’s lower crust or adjacent to an intrusion [7]. Future work will study other, texturally heterogeneous samples to see if EET 90020 presents a unique circumstance or whether crustal partial melting was ubiquitous on the eucrite asteroid.

Eucrites↗

Large Contribution of Meteorological Factors to Inter-Decadal Changes in Regional Aerosol Optical Depth

Aerosol optical depth (AOD) has become a crucial metric for assessing global climate change. Although global and regional AOD trends have been studied extensively, it remains unclear what factors are driving the inter-decadal variations in regional AOD and how to quantify the relative contribution of each dominant factor. This study used a long-term (1980–2016) aerosol dataset from the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) reanalysis, along with two satellite-based AOD datasets (MODIS/Terra and MISR) from 2001 to 2016, to investigate the long-term trends in global and regional aerosol loading. Statistical models based on emission factors and meteorological parameters were developed to identify the main factors driving the inter-decadal changes of regional AOD and to quantify their contribution. Evaluation of the MERRA-2 AOD with the ground-based measurements of AERONET indicated significant spatial agreement on the global scale (r= 0.85, root-mean-square error = 0.12, mean fractional error = 38.7 %, fractional gross error = 9.86 % and index of agreement = 0.94). However, when AOD observations from the China Aerosol Remote Sensing Network (CARSNET) were employed for independent verification, the results showed that MERRA-2 AODs generally underestimated CARSNET AODs in China (relative mean bias = 0.72 and fractional gross error =−34.3 %). In general, MERRA-2 was able to quantitatively reproduce the annual and seasonal AOD trends on both regional and global scales, as observed by MODIS/Terra, although some differences were found when compared to MISR. Over the 37-year period in this study, significant decreasing trends were observed over Europe and the eastern United States. In contrast, eastern China and southern Asia showed AOD increases, but the increasing trend of the former reversed sharply in the most recent decade. The statistical analyses suggested that the meteorological parameters explained a larger proportion of the AOD variability (20.4 %–72.8 %) over almost all regions of interest (ROIs) during 1980–2014 when compared with emission factors (0 %–56 %). Further analysis also showed that SO2 was the dominant emission factor, explaining 12.7 %–32.6 % of the variation in AOD over anthropogenic-aerosol-dominant regions, while black carbon or organic carbon was the leading factor over the biomass-burning-dominant (BBD) regions, contributing 24.0 %–27.7 % of the variation. Additionally, wind speed was found to be the leading meteorological parameter, explaining 11.8 %–30.3 % of the variance over the mineral-dust-dominant regions, while ambient humidity (including soil moisture and relative humidity) was the top meteorological parameter over the BBD regions, accounting for 11.7 %–35.5 % of the variation. The results of this study indicate that the variation in meteorological parameters is a key factor in determining the inter-decadal change in regional AOD.

Huizheng Che↗

Revisiting the Global Patterns of Seasonal Cycle in Sea Surface Salinity

Argo profiling floats and L-band passive microwave remote sensing have significantly improved the global sampling of sea surface salinity (SSS) in the past 15 years, allowing the study of the range of SSS seasonal variability using concurrent satellite and in situ platforms. Here, harmonic analysis was applied to four 0.25° satellite products and two 1° in situ products between 2016 and 2018 to determine seasonal harmonic patterns. The 0.25° World Ocean Atlas (WOA) version 2018 was referenced to help assess the harmonic patterns from a long-term perspective based on the 3-year period. The results show that annual harmonic is the most characteristic signal of the seasonal cycle, and semiannual harmonic is important in regions influenced by monsoon and major rivers. The percentage of the observed variance that can be explained by harmonic modes varies with products, with values ranging between 50% and 72% for annual harmonic and between 15% and 19% for semiannual harmonic. The large spread in the explained variance by the annual harmonic reflects the large disparity in nonseasonal variance (or noise) in the different products. Satellite products are capable of capturing sharp SSS features on meso- and frontal scales and the patterns agree well with the WOA 2018. These products are, however, subject to the impacts of radiometric noises and are algorithm dependent. The coarser-resolution in situ products may underrepresent the full range of high-frequency small scale SSS variability when data record is short, which may have enlarged the explained SSS variance by the annual harmonic.

L-band↗

The Effects of Surface Fossil Magnetic Fields on Massive Star Evolution - III. The Case of τSco

τSco, a well-studied magnetic B-type star in the Upper Sco association, has a number of surprising characteristics. It rotates very slowly and shows nitrogen excess. Its surface magnetic field is much more complex than a purely dipolar configuration which is unusual for a magnetic massive star. We employ the CMFGEN radiative transfer code to determine the fundamental parameters and surface CNO and helium abundances. Then, we employ MESA and GENEC stellar evolution models accounting for the effects of surface magnetic fields. To reconcileτSco’s properties with single-star models, an increase is necessary in the efficiency of rotational mixing by a factor of 3–10 and in the efficiency of magnetic braking by a factor of 10. The spin-down could be explained by assuming a magnetic field decay scenario. However, the simultaneous chemical enrichment challenges the single-star scenario. Previous works indeed suggested a stellar merger origin forτSco. However, the merger scenario also faces similar challenges as our magnetic single-star models to explainτSco’s simultaneous slow rotation and nitrogen excess. In conclusion, the single-star channel seems less likely and versatile to explain these discrepancies, while the merger scenario and other potential binary-evolution channels still require further assessment as to whether they may self-consistently explain the observables of τSco.

Z Keszthelyi↗

The Zn, S, and Cl Isotope Compositions of Mare Basalts: Implications for the Effects of Eruption Style and Pressure on Volatile Element Stable Isotope Fractionation on the Moon

We compare the stable isotope compositions of Zn, S, and Cl for Apollo mare basalts to better constrain the sources and timescales of lunar volatile loss. Mare basalts have broadly elevated yet limited ranges in δ(66)Zn, δ(34)S, and δ(37)Cl_(SBC+WSC) values of 1.27 ± 0.71, 0.55 ± 0.18, and 4.1 ± 4.0‰, respectively, compared to the silicate Earth at 0.15, –1.28, and 0‰, respectively. We find that the Zn, S, and Cl isotope compositions are similar between the low- and high-Ti mare basalts, providing evidence of a geochemical signature in the mare basalt source region that is inherited from lunar formation and magma ocean crystallization. The uniformity of these compositions implies mixing following mantle overturn, as well as minimal changes associated with subsequent mare magmatism. Degassing of mare magmas and lavas did not contribute to the large variations in Zn, S, and Cl isotope compositions found in some lunar materials (i.e., 15‰ in δ(66)Zn, 60‰ in δ(34)S, and 30‰ in δ(37)Cl). This reflects magma sources that experienced minimal volatile loss due to high confining pressures that generally exceeded their equilibrium saturation pressures. Alternatively, these data indicate effective isotopic fractionation factors were near unity. Our observations of S isotope compositions in mare basalts contrast to those for picritic glasses (Saal and Hauri 2021), which vary widely in S isotope compositions from –14.0 to 1.3‰, explained by extensive degassing of picritic magmas under high-P/P_(Sat) values (>0.9) during pyroclastic eruptions. The difference in the isotope compositions of picritic glass beads and mare basalts may result from differences in effusive (mare) and explosive (picritic) eruption styles, wherein the high-gas contents necessary for magma fragmentation would result in large effective isotopic fractionation factors during degassing of picritic magmas. Additionally, in highly vesiculated basalts, the δ(34)S and δ(37)Cl values of apatite grains are higher and more variable than the corresponding bulk-rock values. The large isotopic range in the vesiculated samples is explained by late-stage low-pressure “vacuum” degassing (P/P-(Sat) ~ 0) of mare lavas wherein vesicle formation and apatite crystallization took place post-eruption. Bulk-rock mare basalts were seemingly unaffected by vacuum degassing. Degassing of mare lavas only became important in the final stages of crystallization recorded in apatite—potentially facilitated by cracks/fractures in the crystallizing flow. We conclude that samples with wide-ranging volatile element isotope compositions are likely explained by localized processes, which do not represent the bulk Moon.

Halogens↗

Ionization Efficiency in the Dayside Ionosphere of Mars: Structure and Variability

Ionization efficiency, the ratio between photoelectron impact ionization and primary photoionization, is an important quantity from the perspective of ionospheric energy balance and as a quantity to be parameterized for use in global simulations. We investigate ionization efficiency using MAVEN in situ measurements of solar extreme ultraviolet (EUV) flux, suprathermal electron flux, and neutral density. Its behavior with respect to pressure and solar zenith angle (SZA) is explained by ionization cross-sections and photoabsorption, plus photoelectron transport near the terminator. We find similar ionization efficiencies between the species CO 2 , O, N 2 , and CO, with Argon ∼40% higher, explained largely by its higher ionization potential. Efficiency depends positively on solar activity, increasing by ∼50–100% (depending on season) from low to moderate solar EUV levels, before flattening at higher EUV levels. EUV spectral hardening appears to be responsible for only ∼15% of this increase, the remainder best explained by variability in neutral composition. We find a negligible dependence of ionization efficiency on the strength and direction of crustal magnetic fields. Efficiencies agree poorly with the theoretical model of Nicholson et al. (2009), https://doi.org/10.1111/j.1365-2966.2009.15463.x, possibly reflecting differences in model versus measured solar EUV spectrum and neutral densities. We fit the data to the empirical parameterization of Mendillo et al. (2011), https://doi.org/10.1029/2011ja016865 but find that a single fit cannot capture ionization efficiency behavior across all dayside SZA values. We believe that an empirical model of Mars ionization efficiency will require MAVEN data and validated transport modeling to reach to sufficiently high pressures to capture Mars' photochemical peaks M1 and M2

Robert Lillis↗

TruePAL – An AI Assistant for First Responder Safety

This paper presents the development of an AI assistant, Trusted and Explainable Artificial Intelligence for Saving Lives (TruePAL), to provide real-time warning of risks of potential crashes to the first responders. The TruePAL system employs an AI and deep learning technology for saving first responders and roadside crews lives in and around active traffic. A deep neural network (DNN) and a Non-Axiomatic Reasoning System (NARS) are implemented as an AI system. A mobile app with AI interface is developed to perform verbal communication with the first responders. The TruePAL team has developed an explainable AI approach by opening up the DNN blackbox to extract the activation filters of various features and parts of the targeted objects. The combination of DNN and NARS makes the TruePAL system explainable to the users. TruePAL ingests on-board cameras, radar, and other sensor signals, analyzes the environment and traffic patterns to generate timely warning to drivers and roadside crews to avoid crashes. The TruePAL team, in collaboration with the Miami/Dade Police Dept., has designed five use cases and multiple sub-scenarios in a CARLA driving simulator to test the capability of TruePAL in timely warning to the first responder drivers in potential crash scenarios. We have successfully demonstrated its capability of timely warning in over a dozen scenarios based on the use cases. The preliminary test simulation results show that TruePAL could provide the drivers and crew members advanced warning before a crash occurs.

Chow, Edward↗

Linking OH Variability to Observable Variables, Meteorology and Transport

The hydroxyl radical (OH) plays a vital role in tropospheric chemistry, as it provides the dominant sink for a multitude of pollutants and climate-relevant gases such as methane. Observational constraints on the global distribution and temporal variability of OH are limited, and models simulate a wide range of OH distributions. While OH itself has a short atmospheric lifetime, OH is photochemically coupled to longer-lived species that undergo atmospheric transport. Here, we investigate how much of the OH variability within and between models can be explained by differences in observable species to develop diagnostics for OH differences. We find that NO 2 and water vapor together explain much of the spatial and temporal variability in simulated OH, and we use satellite observations to identify biases in these variables. The OH response to ENSO also differs between models, and we investigate potential causes of these differences such as differences in convection or lightning NOx. We also explore the potential of idealized tracers to represent the OH distribution. Within a single model, meteorological variables such as humidity and idealized tracers of transport can explain a significant portion of the OH spatial variability. We use a Gradient Boosted Regression Trees, a type of machine learning, to account for non-linear relationships between OH and the input variables.

Meteorology↗

Seminar 14 - Desiccant Enhanced Air Conditioning: Desiccant Enhanced Evaporative Air Conditioning (Presentation)

This presentation explains how liquid desiccant based coupled with an indirect evaporative cooler can efficiently produce cool, dry air, and how a liquid desiccant membrane air conditioner can efficiently provide cooling and dehumidification without the carryover problems of previous generations of liquid desiccant systems. It provides an overview to a liquid desiccant DX air conditioner that can efficiently provide cooling and dehumidification to high latent loads without the need for reheat, explains how liquid desiccant cooling and dehumidification systems can outperform vapor compression based air conditioning systems in hot and humid climates, explains how liquid desiccant cooling and dehumidification systems work, and describes a refrigerant free liquid desiccant based cooling system.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Critical Infrastructure Classification via CNN-based Modeling and Image Analysis

With recent advances in the fields of satellite imagery and machine learning we now have the ability to develop explainable deep learning models that enhance critical infrastructure analysis. Funded through Idaho National Laboratory’s (INL) Laboratory Directed Research and Development (LDRD) office we are in the process of developing a deep learning model capable of identifying critical infrastructure facilities and embedded features within those facilities. Utilizing current limit of practice techniques in the machine learning areas of explainability and transfer learning our model, once complete, will have the capacity to be used on multiple different imagery data sets and produce results that not only classify critical infrastructure facilities, but also explain why a critical infrastructure facility was classified as a certain type of facility. These advancements eliminate the ‘black box’ approach deep learning models have had in the past, where a user will have to trust the conclusion of a model without understanding what reasoning when into the model’s classification process. They also expand a model’s usefulness, traditionally a deep learning model will have to use the same data set it was originally trained on. Given the long training times of deep learning models this is impractical in a number of scenarios. By utilizing transfer learning advancements, we are eliminating the need to train our model on the same data set it is then run on to classify critical infrastructure facilities. We are also enabling the analysis and classification of data sets that are potentially too small to be divided into a training and testing data set. Once completed this model can provide a foundation to enhanced critical infrastructure analysis, dependency analysis, and potential disaster relief efforts.

97 MATHEMATICS AND COMPUTING↗

A Real-Time ANPC Inverter Digital Twin with Integrated Design-For-Trust

The demand for renewable energy has increased over the last few years, and so has the demand for greater expectations within the energy market. This increasing trend has been accompanied by more significant usage of internet-connected devices (IoT), leading to critical electrical infrastructure being connected to the internet. Implementing internet connectivity with such devices and systems provides benefits such as improving the system's performance, facilitating irregularity and anomaly mitigation, and providing additional situational awareness for enhanced decision-making. However, enhancing the connected system with IoT introduces a drawback – a greater vulnerability to cyber-attacks. Cyber-attacks targeting critical infrastructure in the electrical sector have occurred in the United States and Ukraine. These cyber-attacks highlight and expose vulnerabilities that a system inherits when connecting to the internet. These attacks left thousands of customers without electricity for hours until operators could regain control of the electric utility grid. Therefore, to address the vulnerabilities of an internet-connected power electronic device, this work focused on the hardware layer of the system. Implementing a cyber-control system inside the hardware layer can significantly reduce the possibility of an attacker patching malicious controller firmware into a photovoltaic grid-connected inverter, thus mitigating the likelihood that the inverter becomes inactive a cyber-attack scenario. With this mitigation technique, if a cyberattack is successful and an attacker gains control of the network, a cyber-defense technique is in place to mitigate the impact of the cyber-attack. This additional protection layer was developed based on an innovative concept known as Digital Twin (DT). A DT, in this case, replicates an Active-Neutral Point Clamped (ANPC) inverter and was designed using a hardware language known as VHDL (Very High-Speed Integrated Circuit Hardware Description Language) and applied to Field-Programmable-GateArray (FPGA). The DT is embedded within the FPGA and contained in a controller board, the UCB (Unified Controller Board), developed by the University of Arkansas electrical engineering team. This UCB also contains two Digital Signal Processors (DSPs) responsible for generating associated signals to control an authentic physical inverter. These DSP signals are received and processed by the FPGA that implements the DT of an ANPC; in other words, it simulates in realtime the expected output of an actual ANPC inverter using the signals from the DSP. When a new firmware is ready to be patched, the DT provides output signals simulating behavior that a real ANPC inverter would generate with the new firmware. The new firmware is tested to check if it meets all the operational requirements established using a Design-For-Trust technique (DFTr). If the new firmware fails in at least one of the DFT tests, it is considered malicious and must be rejected. This work is divided into sections, such as Background, which explains the pieces that were used and the strategy behind this work; Process and Procedure, which explains the methodology that was adopted to prove the reliability and effectiveness of this work; Results and Discussion, where the simulations and results are described and explained; followed by Conclusion and Future work section, which concludes this work and adds possible future projects to continue this work further.

do Amaral Custodio, Paulo Vitor↗

A detailed study of interpretability of deep neural network based top taggers

Abstract Recent developments in the methods of explainable artificial intelligence (XAI) allow researchers to explore the inner workings of deep neural networks (DNNs), revealing crucial information about input–output relationships and realizing how data connects with machine learning models. In this paper we explore interpretability of DNN models designed to identify jets coming from top quark decay in high energy proton–proton collisions at the Large Hadron Collider. We review a subset of existing top tagger models and explore different quantitative methods to identify which features play the most important roles in identifying the top jets. We also investigate how and why feature importance varies across different XAI metrics, how correlations among features impact their explainability, and how latent space representations encode information as well as correlate with physically meaningful quantities. Our studies uncover some major pitfalls of existing XAI methods and illustrate how they can be overcome to obtain consistent and meaningful interpretation of these models. We additionally illustrate the activity of hidden layers as neural activation pattern diagrams and demonstrate how they can be used to understand how DNNs relay information across the layers and how this understanding can help to make such models significantly simpler by allowing effective model reoptimization and hyperparameter tuning. These studies not only facilitate a methodological approach to interpreting models but also unveil new insights about what these models learn. Incorporating these observations into augmented model design, we propose the particle flow interaction network model and demonstrate how interpretability-inspired model augmentation can improve top tagging performance.

97 MATHEMATICS AND COMPUTING↗

Case Study: Analysis of Autonomous Center line Tracking Neural Networks

Deep neural networks have gained widespread usage in a number of applications. However, limitations such as lack of explainability and robustness inhibit building trust in their behavior, which is crucial in safety critical applications such as autonomous driving. Therefore, techniques which aid in understanding and providing guarantees for neural network behavior are the need of the hour. In this paper, we present a case study applying a recently proposed technique, Prophecy, to analyze the behavior of a neural network model, provided by our industry partner and used for autonomous guiding of airplanes on taxi runways. This regression model takes as input an image of the runway and produces two outputs, cross-track error and heading error, which represent the position of the plane relative to the center line. We use the Prophecy tool to extract neuron activation patterns for the correctness and safety properties of the model. We show the use of these patterns to identify features of the input that explain correct and incorrect behavior. We also use the patterns to provide guarantees of consistent behavior. We explore a novel idea of using sequences of images (instead of single images) to obtain good explanations and identify regions of consistent behavior.

Deep Neural Networks↗

The Primary and Recycling Sources of OH During the NACHTT-2011 Campaign: HONO as an Important OH Primary Source in the Wintertime

We present OH observation results during the NACHTT-11 field campaign at the Boulder Atmospheric Observatory in Weld County, Colorado. The observed OH levels during the daytime (at noon) were ~ 2.7 × 106 molecules cm-3 at the ground level (2 m above ground level, AGL). HONO and ozone photolysis were the two dominant photochemical OH production pathways during the field campaign. However, alkene ozonolysis, found an important source for OH by two previous winter season OH observations, was a minor contribution to OH primary production (~5 %). To evaluate recycling sources of OH from HO2 and RO2, an observation constrained University of Washington Chemical Mechanism (UWCM) box model is employed to simulated ambient OH levels with different model scenarios. For the base run without constraining observed HONO, the model simulated OH significantly underestimates the observed OH level (20.8 times in the morning and 7.2 times in the daytime). This indicates that the known HONO sources incorporated in the UWCM model cannot explain the observed HONO level. Once HONO is constrained by the observation, the discrepancy between observation and model simulation improves (5.1 times in the morning and 2.1 times in the daytime) but still out of the measurement uncertainty range (35 %). We explore two possible reasons for the observed unexplainably high wintertime OH levels. First, potential roles of Cl atoms produce organic peroxy radicals from the reactions between Cl atmos and alkane compounds. However, the Cl levels during the observation period are estimated very low (~ 103 atoms cm-3) to explain the enhanced OH levels. Second, Impacts of higher HONO levels on the ground was evaluated. Strong HONO gradient towards ground was observed especially during the early morning (6 am to 8 am) was observed and the lowest level available for the HONO observation during the campaign is 5 m AGL. Once we assume the twice of the observed HONO levels averaged between 5 m to 15 m at 2 m AGL, model predicted OH levels agree well within the observation uncertainty range. Wintertime photochemistry has not been investigated as much as the summer season. The results of this study along with a limited number of winter OH observations clearly urge further investigation on tropospheric oxidation capacity in the winter season considering implications of tropospheric oxidation capacity to the short-lived climate forcers especially methane.

Kim, S.↗

Impact of large-scale dynamics on the microphysical properties of midlatitude cirrus

In situ microphysical observations 3 of mid-latitude cirrus collected during the Department of Energy Small Particles in Cirrus (SPAR-TICUS) field campaign are combined with an atmospheric state classification for the Atmospheric Radiation Measurement (ARM) Southern Great Plains (SGP) site to understand statistical relationships between cirrus microphysics and the large-scale meteorology. The atmospheric state classification is informed about the large-scale meteorology and state of cloudiness at the ARM SGP site by combining ECMWF ERA-Interim reanalysis data with 14 years of continuous observations from the millimeter-wavelength cloud radar. Almost half of the cirrus cloud occurrences in the vicinity of the ARM SGP site during SPARTICUS can be explained by three distinct synoptic condi- tions, namely upper-level ridges, mid-latitude cyclones with frontal systems and subtropical flows. Probability density functions (PDFs) of cirrus micro- physical properties such as particle size distributions (PSDs), ice number con- centrations and ice water content (IWC) are examined and exhibit striking differences among the different synoptic regimes. Generally, narrower PSDs with lower IWC but higher ice number concentrations are found in cirrus sam- pled in upper-level ridges whereas cirrus sampled in subtropical flows, fronts and aged anvils show broader PSDs with considerably lower ice number con- centrations but higher IWC. Despite striking contrasts in the cirrus micro- physics for different large-scale environments, the PDFs of vertical velocity are not different, suggesting that vertical velocity PDFs are a poor predic-tor for explaining the microphysical variability in cirrus. Instead, cirrus mi- crophysical contrasts may be driven by differences in ice supersaturations or aerosols.

Muhlbauer, Andreas↗

Evaluation of Cloud-resolving and Limited Area Model Intercomparison Simulations using TWP-ICE Observations. Part 1: Deep Convective Updraft Properties

Ten 3D cloud-resolving model (CRM) simulations and four 3D limited area model (LAM) simulations of an intense mesoscale convective system observed on January 23-24, 2006 during the Tropical Warm Pool – International Cloud Experiment (TWP-ICE) are compared with each other and with observed radar reflectivity fields and dual-Doppler retrievals of vertical wind speeds in an attempt to explain published results showing a high bias in simulated convective radar reflectivity aloft. This high bias results from ice water content being large, which is a product of large, strong convective updrafts, although hydrometeor size distribution assumptions modulate the size of this bias. Snow reflectivity can exceed 40 dBZ in a two-moment scheme when a constant bulk density of 100 kg m-3 is used. Making snow mass more realistically proportional to area rather than volume should somewhat alleviate this problem. Graupel, unlike snow, produces high biased reflectivity in all simulations. This is associated with large amounts of liquid water above the freezing level in updraft cores. Peak vertical velocities in deep convective updrafts are greater than dual-Doppler retrieved values, especially in the upper troposphere. Freezing of large rainwater contents lofted above the freezing level in simulated updraft cores greatly contributes to these excessive upper tropospheric vertical velocities. Strong simulated updraft cores are nearly undiluted, with some showing supercell characteristics. Decreasing horizontal grid spacing from 900 meters to 100 meters weakens strong updrafts, but not enough to match observational retrievals. Therefore, overly intense simulated updrafts may partly be a product of interactions between convective dynamics, parameterized microphysics, and large-scale environmental biases that promote different convective modes and strengths than observed.

clould-resolving↗