On top of machine learning (ML) models, uncertainty quantification (UQ) functions as an essential layer of safety assurance that could lead to more principled decision making by enabling sound risk assessment and management. The safety and reliability improvement of ML models empowered by UQ has the potential to significantly facilitate the broad adoption of ML solutions in high-stakes decision settings, such as healthcare, manufacturing, and aviation, to name a few. In this tutorial, we aim to provide a holistic lens on emerging UQ methods for ML models with a particular focus on neural networks and the applications of these UQ methods in tackling engineering design as well as prognostics and health management problems. Towards this goal, we start with a comprehensive classification of uncertainty types, sources, and causes pertaining to UQ of ML models. Next, we provide a tutorial-style description of several state-of-the-art UQ methods: Gaussian process regression, Bayesian neural network, neural network ensemble, and deterministic UQ methods focusing on spectral-normalized neural Gaussian process. Established upon the mathematical formulations, we subsequently examine the soundness of these UQ methods quantitatively and qualitatively (by a toy regression example) to examine their strengths and shortcomings from different dimensions. Then, we review quantitative metrics commonly used to assess the quality of predictive uncertainty in classification and regression problems. Afterward, we discuss the increasingly important role of UQ of ML models in solving challenging problems in engineering design and health prognostics. Two case studies with source codes available on GitHub are used to demonstrate these UQ methods and compare their performance in the life prediction of lithium-ion batteries at the early stage (case study 1) and the remaining useful life prediction of turbofan engines (case study 2).
A key challenge in inverse problems is the selection of sensors to gather the most effective data. In this paper, we consider the problem of inferring the initial condition to a linear dynamical system and develop an efficient control-theoretical approach for greedily selecting sensors. Our method employs a Galerkin projection to reduce the size of the inverse problem, resulting in a computationally efficient algorithm for sensor selection. As a byproduct of our algorithm, we obtain a preconditioner for the inverse problem that enables the rapid recovery of the initial condition. We analyze the theoretical performance of our greedy sensor selection algorithm as well as the performance of the associated preconditioner. Finally, we verify our theoretical results on various inverse problems involving partial differential equations.
This paper explores the concept of predictive maturity for non-linear concrete constitutive models employed in the computational prediction of the structural response of reinforced concrete structures to impact from free-flying missiles. Such concrete constitutive models are widely varied in complexity. Three constitutive models were utilized within the same finite element structural model to simulate the response of the IRIS III experiment. Each of the models were individually calibrated with available material testing data and also re-calibrated assuming limited availability of test data. When full calibration is possible, more sophisticated constitutive models appear to provide more predictive maturity; however, when this data is not available (e.g. for an existing structure where representative test specimens may not be available), the expected maturity is reduced. Indeed, this hypothesis is supported by the simulations that indicate good agreement with measured experimental response quantities from the IRIS III tests with complex constitutive models and full calibration, and accordingly poor predictions when less complex models are used or when the more sophisticated models are poorly calibrated. Thus, predictions of structural response where complete material testing data is not obtainable should be understood as less predictive.
A major hurdle in utilizing carbon dioxide (CO2) lies in separating it from industrial flue gas mixtures and finding suitable storage methods that enable its application in various industries. To address this issue, we utilized a combination of molecular dynamics simulations and experiments to investigate the behavior of CO2 in common room-temperature ionic liquids (RTIL) when in contact with aqueous interfaces. Our investigation of RTILs, [EMIM][TFSI] and [OMIM][TFSI], and their interaction with a pure water layer mimics the environment of a previously developed ultrathin enzymatic liquid membrane for CO2 separation. We analyzed diffusion constants and viscosity, which reveals that CO2 molecules exhibit faster mobility within the selected ILs compared to what would be predicted solely based on the viscosity of the liquids using the standard Einstein-Stokes relation. Moreover, we calculated the free energy of translocation for various species across the aqueous-IL interface, including CO2 and HCO3-. Free energy profiles demonstrate that CO2 exhibits a more favorable partitioning behavior in the RTILs compared to that in pure water, while a significant barrier hinders the movement of HCO3- from the aqueous layer. Experimental measurement of the CO2 transport in the RTILs corroborates the model. These findings strongly suggest that hydrophobic RTILs could serve as a promising option for selectively transporting CO2 from aqueous media and concentrating it as a preliminary step toward storage.
This article characterises the effects of cathode photoemission leading to electrical discharges in an argon gas. We perform breakdown experiments under pulsed laser illumination of a flat cathode and observe Townsend to glow discharge transitions. The breakdown process is recorded by high-speed imaging, and time-dependent voltage and current across the electrode gap are measured for different reduced electric fields and laser intensities. We employ a 0D transient discharge model to interpret the experimental measurements. The fitted values of transferred photoelectron charge are compared with calculations from a quantum model of photoemission. The breakdown voltage is found to be lower with photoemission than without. When the applied voltage is insufficient for ion-induced secondary electron emission to sustain the plasma, laser driven photoemission can still create a breakdown where a sheath (i.e. a region near the electrode surfaces consisting of positive ions and neutrals) is formed. This photoemission induced plasma persists and decays on a much longer time scale ( ∼ 10 s μ s) than the laser pulse length ( 30 ps). The effects of different applied voltages and laser energies on the breakdown voltage and current waveforms are investigated. The discharge model can accurately predict the measured breakdown voltage curves, despite the existence of discrepancy in quantitatively describing the transient discharge current and voltage waveforms.
Drinking water infrastructure in urban settings is increasingly affected by population growth and disruptions like extreme weather events. This study explores how the integration of direct wastewater reuse can help to maintain drinking water service when the system is compromised.
Here this study investigates the nonlinear frequency response of a shaft-bearing assembly with vibro-impacts occurring at the bearing clearances. The formation of nonlinear behavior as system parameters change is examined, along with the effects of asymmetries in the nominal, inherently symmetric system. The primary effect of increasing the forcing magnitude or decreasing the contact gap sizes is the formation of grazing-induced chaotic solution branches occurring over a wide frequency range near each system resonance. The system's nominal setup has very hard contact stiffness and shows no evidence of isolas or superharmonic resonances over the frequency ranges of interest. Moderate contact stiffnesses cause symmetry breaking and introduce superharmonic resonance branches of primary resonances. Even if some primary resonances are not present due to the system's inherent symmetry, their superharmonic resonances still manifest. Branches of quasiperiodic isolas (isolated resonance branches) are also discovered, along with a cloud of isolas near a high-frequency resonance. Parameter asymmetries are found to produce a few significant changes in behavior: asymmetric linear stiffness, contact stiffness, and gap size could affect the behavior of primary resonant frequencies and isolas.
Aqueous electrolytes composed of 0.1 M zinc bis-(trifluoromethyl-sulfonyl)-imide (Zn-(TFSI)2) and acetonitrile (ACN) were studied using combined experimental and simulation techniques. The electrolyte was found to be electrochemically stable when the ACN V% is higher than 74.4. In addition, it was found that the ionic conductivity of the mixed solvent electrolytes changes as a function of ACN composition, and a maximum was observed at 91.7 V% of ACN although the salt concentration is the same. This behavior was qualitatively reproduced by molecular dynamics (MD) simulations. Detailed analyses based on experiments and MD simulations show that at high ACN composition the water network existing in the high water composition solutions breaks. As a result, the screening effect of the solvent weakens and the correlation among ions increases, which causes a decrease in ionic conductivity at high ACN V%. Furthermore, this study provides a fundamental understanding of this complex mixed solvent electrolyte system.
A two-step solar thermochemical looping cycle based on Co3Mo3N/Co6Mo6N reduction/nitridation reactions offers a pathway for green NH3 production that utilizes concentrated solar irradiation, H2O, and air as feedstocks. The NH3 production cycle steps both derive process heat from concentrated solar irradiation and encompass 1) the reduction of Co3Mo3N in H2 to Co6Mo6N and NH3; and 2) nitridation of Co6Mo6N to Co3Mo3N with N2. Co3Mo3N reduction/nitridation reactions are examined at different H2 and/or N2 partial pressures and temperatures. NH3 production is quantified in situ using liquid conductivity measurements coupled with mass spectrometry (MS). Solid-state characterization is performed to identify a surface oxygen layer that necessitates the addition of H2 during cycling to prevent surface oxidation by trace amounts of O2. H2 concentrations of > 5% H2/Ar and temperatures >500 °C are required to reduce Co3Mo3N to Co6Mo6N and form NH3 at 1 bar. Complete regeneration of Co3Mo3N from Co6Mo6N is achieved at conditions of 700 °C under 25–75% H2/N2. H2 pressure-swings are observed to increase NH3 production during Co3Mo3N reduction. In conclusion, the results represent the first comprehensive characterization of and definitive non-catalytic production of NH3 via chemical looping with metal nitrides and provide insights for technology development.
Before residential photovoltaic (PV) systems are interconnected with the grid, various planning and impact studies are conducted on detailed models of the system to ensure safety and reliability are maintained. However, these model-based analyses can be time-consuming and error-prone, representing a potential bottleneck as the pace of PV installations accelerates. Data-driven tools and analyses provide an alternate pathway to supplement or replace their model-based counterparts. In this article, a data-driven algorithm is presented for assessing the thermal limitations of PV interconnections. Using input data from residential smart meters, and without any grid models or topology information, the algorithm can determine the nameplate capacity of the service transformer supplying those customers. The algorithm was tested on multiple datasets and predicted service transformer capacity with >98% accuracy, regardless of existing PV installations. This algorithm has various applications from model-free thermal impact analysis for hosting capacity studies to error detection and calibration of existing grid models.
Recent developments integrating micromechanics and neural networks offer promising paths for rapid predictions of the response of heterogeneous materials with similar accuracy as direct numerical simulations. The deep material network is one such approaches, featuring a multi-layer network and micromechanics building blocks trained on anisotropic linear elastic properties. Once trained, the network acts as a reduced-order model, which can extrapolate the material’s behavior to more general constitutive laws, including nonlinear behaviors, without the need to be retrained. However, current training methods initialize network parameters randomly, incurring inevitable training and calibration errors. Here, we introduce a way to visualize the network parameters as an analogous unit cell and use this visualization to “quilt” patches of shallower networks to initialize deeper networks for a recursive training strategy. The result is an improvement in the accuracy and calibration performance of the network and an intuitive visual representation of the network for better explainability.
An interpretable machine learning method, physics-informed genetic programming-based symbolic regression (P-GPSR), is integrated into a continuum thermodynamic approach to developing constitutive models. The proposed strategy for combining a thermodynamic analysis with P-GPSR is demonstrated by generating a yield function for an idealized material with voids, i.e., the Gurson yield function. First, a thermodynamic-based analysis is used to derive model requirements that are exploited in a custom P-GPSR implementation as fitness criteria or are strongly enforced in the solution. The P-GPSR implementation improved accuracy, generalizability, and training time compared to the same GPSR code without physics-informed fitness criteria. The yield function generated through the P-GPSR framework is in the form of a composite function that describes a class of materials and is characteristically more interpretable than GPSR-derived equations. The physical significance of the input functions learned by P-GPSR within the composite function is acquired from the thermodynamic analysis. Fundamental explanations of why the implemented P-GPSR capabilities improve results over a conventional GPSR algorithm are provided.
R. toruloides is an oleaginous yeast, with diverse metabolic capacities and high tolerance for inhibitory compounds abundant in plant biomass hydrolysates. While R. toruloides grows on several pentose sugars and alcohols, further engineering of the native pathway is required for efficient conversion of biomass-derived sugars to higher value bioproducts. A previous high-throughput study inferred that R. toruloides possesses a non-canonical l-arabinose and d-xylose metabolism proceeding through d-arabitol and d-ribulose. In this study, we present a combination of genetic and metabolite data that refine and extend that model. Chiral separations definitively illustrate that d-arabitol is the enantiomer that accumulates under pentose metabolism. Deletion of putative d-arabitol-2-dehydrogenase (RTO4_9990) results in > 75% conversion of d-xylose to d-arabitol, and is growth-complemented on pentoses by heterologous xylulose kinase expression. Deletion of putative d-ribulose kinase (RTO4_14368) arrests all growth on any pentose tested. Analysis of several pentose dehydrogenase mutants elucidates a complex pathway with multiple enzymes mediating multiple different reactions in differing combinations, from which we also inferred a putative l-ribulose utilization pathway. Our results suggest that we have identified enzymes responsible for the majority of pathway flux, with additional unknown enzymes providing accessory activity at multiple steps. Further biochemical characterization of the enzymes described here will enable a more complete and quantitative understanding of R. toruloides pentose metabolism. These findings add to a growing understanding of the diversity and complexity of microbial pentose metabolism.
Emerging and re-emerging viral pathogens present a unique challenge for anti-viral therapeutic development. Anti-viral approaches with high flexibility and rapid production times are essential for combating these high-pandemic risk viruses. CRISPR-Cas technologies have been extensively repurposed to treat a variety of diseases, with recent work expanding into potential applications against viral infections. However, delivery still presents a major challenge for these technologies. Lipid-coated mesoporous silica nanoparticles (LCMSNs) offer an attractive delivery vehicle for a variety of cargos due to their high biocompatibility, tractable synthesis, and amenability to chemical functionalization. Here, we report the use of LCMSNs to deliver CRISPR-Cas9 ribonucleoproteins (RNPs) that target the Niemann–Pick disease type C1 gene, an essential host factor required for entry of the high-pandemic risk pathogen Ebola virus, demonstrating an efficient reduction in viral infection. We further highlight successful in vivo delivery of the RNP-LCMSN platform to the mouse liver via systemic administration.
The properties of electrons in matter are of fundamental importance. They give rise to virtually all material properties and determine the physics at play in objects ranging from semiconductor devices to the interior of giant gas planets. Modeling and simulation of such diverse applications rely primarily on density functional theory (DFT), which has become the principal method for predicting the electronic structure of matter. While DFT calculations have proven to be very useful, their computational scaling limits them to small systems. We have developed a machine learning framework for predicting the electronic structure on any length scale. It shows up to three orders of magnitude speedup on systems where DFT is tractable and, more importantly, enables predictions on scales where DFT calculations are infeasible. Our work demonstrates how machine learning circumvents a long-standing computational bottleneck and advances materials science to frontiers intractable with any current solutions.
Across many industries and engineering disciplines, systems of components are designed and deployed into their operational environments. It is the desire of the engineer to be able to predict if the component or system will survive its operational environment or if the component will fail due to mechanical stresses. One method to determine if the component will survive the operational environment is to expose the component to a simulation of the environment in a laboratory. One difficulty in executing such a test is that the component may not have the same boundary condition in both the laboratory and operational configurations. This paper presents a novel method of quantifying the error in the modal domain that occurs from the impedance difference between the laboratory test fixture and the operational configuration. The error is calculated from the projection from one mode shape space to the other, and the error is in terms of each mode of the operational configuration. The error provides insight into the effectiveness of the test fixture with respect to the ability to recreate the individual mode shapes of the operational configuration. A case study is presented to show the error in the modal projection between two configurations is a lower limit for the error that can be achieved by a laboratory test.
We report on the substantial advancement of long wavelength InAs-based interband cascade lasers (ICLs) utilizing advanced waveguides formed from hybrid cladding layers and targeting the 10-12μm wavelength region. Modifications in the hole injector have improved carrier transport in these ICLs, resulting in significantly reduced threshold voltages (Vth) as low as 3.62 V at 80 K. Consequently, much higher voltage efficiencies were observed, peaking at about 73% at 10.3μm and allowing for large output powers of more than 100 mW/facet. Also, low threshold current densities (Jth) of 8.8 A/cm2 in cw mode and 7.6 A/cm2 in pulsed mode near 10μm were observed; a result of adjustments in the GaInSb hole well composition intended to reduce the overall strain accumulation in the ICL. Furthermore, an ICL from the second wafer operating at a longer wavelength achieved a peak voltage efficiency of 57% at 11.7μm, with a peak output power of more than 27 mW/facet. This ICL went on to lase beyond 12μm in both cw and pulsed modes, representing a new milestone in long wavelength coverage for ICLs with the standard W-QW active region.
Interface resistance has become a significant bottleneck for solid-state batteries (SSBs). Most studies of interface resistance have focused on extrinsic mechanisms such as interface reactions and imperfect contact between electrodes and solid electrolytes. Interface potentials are an important intrinsic mechanism that is often ignored. Here, we highlight Kelvin probe force microscopy (KPFM) as a tool to image the local potential at interfaces inside SSBs, examining the existing literature and discussing challenges in interpretation. Drawing analogies with electron transport in metal/semiconductor interfaces, we showcase a formalism that predicts intrinsic ionic resistance based on the properties of the contacting phases, and we emphasize that future battery designs should start from material pairs with low intrinsic resistance. We conclude by outlining future directions in the study of interface potentials through both theory and experiment. Graphic abstract: [Figure not available: see fulltext.]
The fraction of tritium converted to the water form in a fire scenario is one of the metrics of greatest interest for radiological safety assessments. The conversion fraction is one of the prime variables contributing to the hazard assessment. This paper presents measurements of oxidation rates for the non-radioactive hydrogen isotopes (protium and deuterium) at sub-flammable concentrations that are typical of many of the most likely tritium release scenarios. These measurements are fit to a simplified 1-step kinetic rate expression, and the isotopic trends for protium and deuterium are extrapolated to produce a model appropriate for tritium. The effects of the new kinetic models are evaluated via CFD simulations of an ISO-9705 standard room fire that includes a trace release of hydrogen isotope (tritium), illustrating the high importance of the correct (measurement-based) kinetics to the outcome of the simulated conversion.
Quantum cascade lasers (QCLs) have emerged as promising candidates for generating chip-scale frequency combs in mid-infrared and terahertz wavelengths. In this work, we demonstrate frequency comb formation in ring terahertz QCLs using the injection of light from a distributed feedback (DFB) laser. The DFB design frequency is chosen to match the modes of the ring cavity (near 3.3 THz), and light from the DFB is injected into the ring QCL via a bus waveguide. By controlling the power and frequency of the optical injection, we show that combs can be selectively formed and controlled in the ring cavity. Numerical modeling suggests that this comb is primarily frequency-modulated in character, with the injection serving to trigger comb formation. We also show that the ring can be used as a filter to control the output of the DFB QCL, potentially being of interest in terahertz photonic integrated circuits. Our work demonstrates that waveguide couplers are a compelling approach for injecting and extracting radiation from ring terahertz combs and offer exciting possibilities for the generation of new comb states in terahertz, such as frequency-modulated waves, solitons, and more.
Perspectives for understanding the brain vary across disciplines and this has challenged our ability to describe the brain’s functions. In this comment, we discuss how emerging theoretical computing frameworks that bridge top-down algorithm and bottom-up physics approaches may be ideally suited for guiding the development of neural computing technologies such as neuromorphic hardware and artificial intelligence. Furthermore, we discuss how this balanced perspective may be necessary to incorporate the neurobiological details that are critical for describing the neural computational disruptions within mental health and neurological disorders.
Behind-the-meter (BTM) battery energy storage systems (BESS) are undergoing rapid deployment. Simple equations to estimate the installed cost of BTM BESS are often necessary when a rigorous, bottom-up cost estimate is not available or not appropriate, in applications such as energy system modeling, informing a BESS sizing decision, and cost benchmarking. Drawing on project-level data from California, I estimate several predictive regression models of the installed cost of a BTM BESS as a function of energy capacity and power capacity. The models are evaluated for in-sample goodness-of-fit and out-of-sample predictive accuracy. The results of these analyses indicate stronger empirical support for models with natural log transformations of installed cost, energy, and power as compared against widely-used models that posit a linear relationship among the untransformed versions of these variables. Building on these results, I present a logarithmic model that can predict installed cost conditional on energy capacity, power capacity, AC or DC coupling with distributed generation, customer sector, and local wages for electricians. I document how the model can be easily extrapolated to future years, either with forecasts from other sources or by re-estimating the parameters with the latest data.
Efficient conversion of pentose sugars remains a significant barrier to the replacement of petroleum-derived chemicals with plant biomass-derived bioproducts. While the oleaginous yeast Rhodosporidium toruloides (also known as Rhodotorula toruloides) has a relatively robust native metabolism of pentose sugars compared to other wild yeasts, faster assimilation of those sugars will be required for industrial utilization of pentoses. To increase the rate of pentose assimilation in R. toruloides, we leveraged previously reported high-throughput fitness data to identify potential regulators of pentose catabolism. Two genes were selected for further investigation, a putative transcription factor (RTO4_12978, Pnt1) and a homolog of a glucose transceptor involved in carbon catabolite repression (RTO4_11990). Overexpression of Pnt1 increased the specific growth rate approximately twofold early in cultures on xylose and increased the maximum specific growth by 18% while decreasing accumulation of arabitol and xylitol in fast-growing cultures. Improved growth dynamics on xylose translated to a 120% increase in the overall rate of xylose conversion to fatty alcohols in batch culture. Proteomic analysis confirmed that Pnt1 is a major regulator of pentose catabolism in R. toruloides. Deletion of RTO4_11990 increased the growth rate on xylose, but did not relieve carbon catabolite repression in the presence of glucose. Carbon catabolite repression signaling networks remain poorly characterized in R. toruloides and likely comprise a different set of proteins than those mainly characterized in ascomycete fungi.
Access to accurate solar resource data is critical for numerous applications, including estimating the yield of solar energy systems, developing radiation models, and validating irradiance datasets. However, lack of standardization in data formats and access interfaces across providers constitutes a major barrier to entry for new users. pvlib python's iotools subpackage aims to solve this issue by providing standardized Python functions for reading local files and retrieving data from external providers. All functions follow a uniform pattern and return convenient data outputs, allowing users to seamlessly switch between data providers and explore alternative datasets. The pvlib package is community-developed on GitHub: https://github.com/pvlib/pvlib-python. As of pvlib python version 0.9.5, the iotools subpackage supports 12 different datasets, including ground measurement, reanalysis, and satellite-derived irradiance data. The supported ground measurement networks include the Baseline Surface Radiation Network (BSRN), NREL MIDC, SRML, SOLRAD, SURFRAD, and the US Climate Reference Network (CRN). Additionally, satellite-derived and reanalysis irradiance data from the following sources are supported: PVGIS (SARAH & ERA5), NSRDB PSM3, and CAMS Radiation Service (including McClear clear-sky irradiance).
Future machine learning strategies for materials process optimization will likely replace human capital-intensive artisan research with autonomous and/or accelerated approaches. Such automation enables accelerated multimodal characterization that simultaneously minimizes human errors, lowers costs, enhances statistical sampling, and allows scientists to allocate their time to critical thinking instead of repetitive manual tasks. Previous acceleration efforts to synthesize and evaluate materials have often employed elaborate robotic self-driving laboratories or used specialized strategies that are difficult to generalize. Herein we describe an implemented workflow for accelerating the multimodal characterization of a combinatorial set of 915 electroplated Ni and Ni–Fe thin films resulting in a data cube with over 160,000 individual data files. Our acceleration strategies do not require manufacturing-scale resources and are thus amenable to typical materials research facilities in academic, government, or commercial laboratories. The workflow demonstrated the acceleration of six characterization modalities: optical microscopy, laser profilometry, X-ray diffraction, X-ray fluorescence, nanoindentation, and tribological (friction and wear) testing, each with speedup factors ranging from 13–46x. In addition, automated data upload to a repository using FAIR data principles was accelerated by 64x.
