To provide technical support for sand control decision-making, it is necessary to predict the production condition at which sand production occurs. Sanding onset prediction involves simulating the stress state on the surface of an oil/gas producing cavity (e.g. borehole, perforation tunnel) and applying appropriate sand production criterion to predict the fluid pressure or pressure gradient at which sand production occurs. In this work, we present numerical and analytical poroelastoplastic stress models describing stress around producing cavity and verify those models against each other. Using those models, we evaluate the stress state on the cavity surface and derive sanding onset prediction models in terms of fluid pressure or pressure gradient based on the given sand production criterion. We then run field case studies and validate the sanding onset prediction models. Rock strength criterion plays important roles in sanding onset prediction. We investigate how the sanding onset prediction results vary with the selection of one or another rock strength criterion. In this work, we present four commonly used rock strength criteria in sanding onset prediction and wellbore stability studies: Mohr-Coulomb, Hoek-Brown, Drucker-Prager, and Modified Lade criteria. In each of the criterion, there are two or more parameters involved. In the literature, a two-step procedure is applied to determine the parameters in the rock strength criterion. First, the Mohr-Coulomb parameters like cohesion So and internal friction angle ff are regressed from the laboratory test data. Then, the parameters in other criteria are calculated using the regressed Mohr-Coulomb parameters. We propose that the best way to evaluate the parameters in a specific rock strength criterion is to perform direct regression of the laboratory test data using that criterion. Using this methodology, we demonstrate that the effect of various rock strength criteria on sanding onset prediction is less dramatic than using the commonly used method. With this methodology, the uncertainties of the effect of rock strength criterion on sanding onset prediction are also reduced.

By taking advantage of the increasing amount of available electronic health data, applications of machine learning in the intensive care unit have the potential to improve medical diagnostics and risk stratification. This thesis proposes an approach for early onset prediction of pneumothorax with such technique, using time series data extracted from a clinical database. The prevalence of pneumothorax among patients is identified through ICD-9 codes, and labels for the onset are obtained by relying on proxies closely related to the condition. Both simple algorithms and deep learning networks are used in a sliding window-based prediction framework, and the importance of each feature is measured with weighted Shapley values. The results proved the feasibility of this approach using Long Short-Term Memory models, although the number of false positives is noticeably high. Mechanical ventilation was the most contributing feature for the outcome. In future work, the focus should be on addressing the large class imbalance that prevails, along with considering more well-founded methods of target labeling.

The ionosphere carries a substantial portion of the electrical current flowing in Earth's space environment. Currents and electric fields in the ionosphere are generated through (1) the interaction of the solar wind with the magnetosphere, i.e. magnetic reconnection and (2) the collision of neutral molecules with ions leading to charged particle motions across the geomagnetic field, i.e. neutral wind dynamo. In this study we applied statistical and deep learning techniques to various datasets to investigate the driving influences of ionospheric electrodynamics at mid- and high-latitudes. In Chapter 2, we analyzed an interval on 12 September 2014 which provided a rare opportunity to examine dynamic variations in the dayside convection throat measured by the RISR-N radar as the IMF transitioned from strong By+ to strong Bz+. We found that the high-latitude plasma convection can have dual flow responses with different lag times to strong dynamic IMF conditions that involve IMF By rotation. We proposed a dual reconnection scenario, one poleward of the cusp and the other at the magnetopause nose, to explain the observed flow behavior. In Chapters 3 and 4, we investigated the driving influences of nightside subauroral convection. We developed new statistical models of nightside subauroral (52 - 60 degree) convection under quiet (Kp <= 2+) to moderately disturbed (Kp = 3) conditions using data from six mid-latitude SuperDARN radars across the continential United States. Our analysis suggests that the quiet-time subauroral flows are due to the combined effects of solar wind-magnetosphere coupling leading to penetration electric field and neutral wind dynamo with the ionospheric conductivity modulating their relative dominance. In Chapter 5, we examined the external drivers of magnetic substorms using machine learning. We presented the first deep learning based approach to directly predict the onset of a magnetic substorm. The model has been trained and tested on a comprehensive list of onsets compiled between 1997 and 2017 and achieves 72 +/- 2% precision and 77 +/- 4% recall rates. Our analysis revealed that the external factors, such as the solar wind and IMF, alone are not sufficient to forecast all substorms, and preconditioning of the magnetotail may be an important factor. / Doctor of Philosophy / The Earth's ionosphere, ranging from about 60 km to 1000 km in altitude, is an electrically conducting region of the upper atmosphere that exists primarily due to ionization by solar ultraviolet radiation. The Earth's magnetosphere is the region of space surrounding the Earth that is dominated by the Earth's magnetic field. The magnetosphere and ionosphere are tightly coupled to each other through the magnetic field lines which act as highly conductive wires. The sun constantly releases a stream of plasma (i.e., gases of ions and free electrons) known as the solar wind, which carries the solar magnetic field known as the interplanetary magnetic field (IMF). The solar wind interacts with the Earth's magnetosphere and ionosphere through a process called magnetic reconnection, which drives currents and electric fields in the coupled magnetosphere and ionosphere. The ionosphere carries a substantial portion of the electrical currents flowing in the Earth's space environment. The interaction of the ionospheric currents and electric fields with plasma and neutral particles is called ionospheric electrodynamics. In this study we utilized statistical and machine learning techniques to study ionospheric electrodynamics in three distinct regions. First, we studied the influence of duskward IMF on plasma convection in the polar region using measurements from the Resolute Bay Incoherent Scatter Radar – North (RISR-N). Specifically, we analyzed an interval on Sep. 12, 2014 when the RISR-N radar made measurements in the high latitude noon sector while the IMF turned from duskward to strongly northward. We found that the high latitude plasma convection can have flow responses with different lag times during strong IMF conditions that involve IMF By rotation. Such phenomena are rarely observed and are not predicted by the antiparallel or the component reconnection models applied to quasi‐static conditions. We propose a dual reconnection scenario, with reconnection occurring poleward of the cusp and also at the dayside subsolar point on the magnetopause, to explain the rarely observed flow behavior. Next, we used measurements from six mid-latitude Super Dual Auroral Radar Network (SuperDARN) radars distributed across the continental United States to investigate the driving influences of plasma convection in the subauroral region, which is equatorward of the region where aurora is normally observed. Previous studies have suggested that plasma motions in the subaruroral region were mainly due to the neutral winds blowing the ions, i.e. the neutral wind dynamo. However, our analysis suggests that subauroral plasma flows are due to the combined effects of solar wind-magnetosphere coupling and neutral wind dynamo with the ionospheric conductivity modulating their relative importance. Finally, we utilized the latest machine learning techniques to examine the external drivers (i.e., solar wind and IMF) of magnetic substorms, which is a physical phenomenon that occurs in the auroral region and causes explosive brightening of the aurora. We developed the first machine learning model that forecasts the onset of a magnetic substorm over the next one hour. The model has been trained and tested on a comprehensive list of onsets compiled between 1997 and 2017 and correctly identify substorm onset ~75% of the time. In contrast, an earlier prediction algorithm correctly identified only ~21% of the substorm onsets in the same dataset. Our analysis revealed that external factors alone are not sufficient to forecast all substorms, and preconditioning of the nightside magnetosphere may be an important factor.