A scalable convolutional neural network approach to fluid flow prediction in complex environments.

We evaluate the capability of convolutional neural networks (CNNs) to predict a velocity field as it relates to fluid flow around various arrangements of obstacles within a two-dimensional, rectangular channel. We base our network architecture on a gated residual U-Net template and train it on velocity fields generated from computational fluid dynamics (CFD) simulations. We then assess the extent to which our model can accurately and efficiently predict steady flows in terms of velocity fields associated with inlet speeds and obstacle configurations not included in our training set. Real-world applications often require fluid-flow predictions in larger and more complex domains that contain more obstacles than used in model training. To address this problem, we propose a method that decomposes a domain into subdomains for which our model can individually and accurately predict the fluid flow, after which we apply smoothness and continuity constraints to reconstruct velocity fields across the whole of the original domain. This piecewise, semicontinuous approach is computationally more efficient than the alternative, which involves generation of CFD datasets required to retrain the model on larger and more spatially complex domains. We introduce a local orientational vector field entropy (LOVE) metric, which quantifies a decorrelation scale for velocity fields in geometric domains with one or more obstacles, and use it to devise a strategy for decomposing complex domains into weakly interacting subsets suitable for application of our modeling approach. We end with an assessment of error propagation across modeled domains of increasing size.

Molecular Methods for Assessing the Morphology, Topology, and Performance of Polyamide Membranes.

The molecular-scale morphology and topology of polyamide composite membranes determine the performance characteristics of these materials. However, molecular-scale simulations are computationally expensive and morphological and topological characterization of molecular structures are not well developed. Molecular dynamics simulation and analysis methods for the polymerization, hydration, and quantification of polyamide membrane structures were developed and compared to elucidate efficient approaches for producing and analyzing the polyamide structure. Polymerization simulations that omitted the reaction-phase solvent did not change the observed hydration, pore-size distribution, or water permeability, while improving the simulation efficiency. Pre-insertion of water into the aggregate pores (radius ≈ 4 Å) of dry domains enabled shorter hydration simulations and improved simulation scaling, without altering pore structure, properties, or performance. Medial axis and Minkowski functional methods were implemented to identify permeation pathways and quantify the polyamide morphology and topology, respectively. Better agreement between simulations and experimentally observed systems was accomplished by increasing the domain size rather than increasing the number of ensemble realizations of smaller systems. The largest domain hydrated was an order of magnitude larger by volume than the largest domain previously reported. This work identifies methods that can enable more efficient and meaningful fundamental modeling of membrane materials.

Microscale modeling of nondilute flow and transport in porous medium systems.

Nondilute transport occurs routinely in porous medium systems. Experimental observations have revealed effects that seemingly depend upon density, viscosity, velocity, and chemical activity. Macroscale models based upon averaged behavior over many pores have been relied upon to describe such systems to date, which require parametrization of important physical phenomena in material coefficients. To advance fundamental understanding of these complex systems, we examine nondilute transport from a fundamental microscale, or pore-scale, continuum modeling perspective. We approximate the solution of a model based upon the variable-density Navier-Stokes equations and a nondilute species transport equation. Known dependencies of the densities, viscosities, chemical activity, and diffusion for a salt solution on chemical composition are included in the model. Microscale model solutions are averaged to the macroscale and compared with extant experimental observations. Investigation of the effects of various physical phenomena on the microscale velocity distribution and the observed macroscale dispersion are considered using dimensional analysis and constrained simulations. Simulation results are used to explain observed experimental results in light of underlying mechanisms. Conditions under which the various physicochemical effects investigated are important are revealed.

Modeling cross model non-Newtonian fluid flow in porous media.

Fluids exhibiting non-Newtonian rheologies are used in a range of applications, including hydraulic fracturing, enhanced oil recovery, remediation, and industrial processes. Hydraulic fracturing in particular has received attention from environmental scientists, policy-makers, and the general public due in part to concerns about the possibility of contamination of groundwater resources by the complex and potentially harmful fluids used in the process. The non-Newtonian nature of many hydraulic fracturing fluids complicates the prediction of their movement, and precludes use of most traditional flow and transport models. To improve understanding of the flow of such fluids in porous media, a series of column experiments was conducted and a pore-scale lattice Boltzmann model (LBM) was developed, verified, and used to simulate analogous systems. Flow experiments were conducted with guar gum solutions of varying concentration and three porous media systems. The LBM was developed for transient, three-dimensional porous medium systems and included a shear rate-dependent dynamic viscosity based on the Cross rheological model. The LBM was verified using a semi-analytical solution for Cross model fluid flow, OpenFOAM simulations, and grid resolution inter-comparisons between two different solution approaches. Simulations were performed on synthetic porous medium systems produced with a sphere packing algorithm to approximate the properties of the experimental systems. The simulations were in good agreement with the experimental results, particularly for systems that exhibited the greatest non-Newtonian character. The modeling approach developed in this work provides a valuable tool for investigating relationships between pore-scale fluid flow and macroscale variables of interest for simulating movement of non-Newtonian fluids at larger scales.

The Toxicology Investigators Consortium Case Registry-the 2015 Experience.

The American College of Medical Toxicology established the Toxicology Investigators Consortium (ToxIC) Case Registry in 2010. The Registry contains all medical toxicology consultations performed at participating sites. The Registry has continued to grow since its inception, and as of December 31, 2015, contains 43,099 cases. This is the sixth annual report of the ToxIC Registry, summarizing the additional 8115 cases entered in 2015. Cases were identified by a query of the Registry for all cases entered between January 1 and December 31, 2015. Specific data reviewed for analysis included demographics (age, race, gender), source of consultation, reason for consultation, agents and agent classes involved in exposures, signs, symptoms, clinical findings, fatalities, and treatment. By the end of 2015, there were 50 active sites, consisting of 101 separate health-care facilities; 51.2 % of cases involved females. Adults between the ages of 19 and 65 made up the majority (64.2 %) of Registry cases. Caucasian race was the most commonly reported (55.6 %); 9.6 % of cases were identified as Hispanic ethnicity. Inpatient and emergency department referrals were by far the most common referral sources (92.9 %). Intentional pharmaceutical exposures remained the most frequent reason for consultation, making up 52.3 % of cases. Of these intentional pharmaceutical exposures, 69 % represented an attempt at self-harm, and 85.6 % of these were a suicide attempt. Nonopioid analgesics, sedative-hypnotics, and antidepressant agents were the most commonly reported agent classes in 2015. Almost one-third of Registry cases involved a diagnosed toxidrome (32.8 %), with a sedative-hypnotic toxidrome being the most frequently described. Significant vital sign abnormalities were recorded in 25.3 % of cases. There were 98 fatalities reported in the Registry (1.2 %). Adverse drug reactions were reported in 4.3 % of cases. Toxicological treatment was given in 65.3 % of cases, with 33.0 % receiving specific antidotal therapy. Exposure characteristics and trends overall were similar to prior years. While treatment interventions were required in the majority of cases, fatalities were rare.