Browse All Digital Items

The concept of passive aeroelastic tailoring is explored to maximize the performance of the NREL 5-MW wind turbine blade in a uniform flow. Variable-angle tow composite materials model the spanwise-variable wind turbine blade design to allow material-adaptive bend-twist coupling under static aerodynamic loading. A constrained optimization algorithm determines the composite fiber angles along the blade span for four inflow conditions ranging from cut-in to rated wind speeds.

In recent studies, machine learning has become a trending field involving mathematics, statistics, computer science and so on. However, when we were using machine learning to predicate in real world problems, the large amount of data usually make negative

Inspired by the advanced capabilities of fish and other aquatic swimmers, in this thesis, a greater understanding of fish-like propulsion has been sought in terms of morphology and kinematics. Unsteady potential flow simulations on real cetacean flukes reveal that the effect of shape and gait on the swimming performance are not intertwined and are in fact independent. There is one fluke shape that maximizes the propulsive efficiency regardless of the gait and vice versa.

Given a set of locations (i.e. bridges, bays, docks, etc.) that must be inspected and a set ofwaypoints, we design and implement a model to route a fleet of unmanned surface vehicles via a set of waypoints that allow the aforementioned locations to be surveilled. Furthermore, the velocity at which the vehicles traverse each part of the route is dependent upon the level of surveillance required for each site.

When an extreme event occurs, such as an earthquake or a tsunami, the amount of socioeconomic losses due to reduced functionality of infrastructure systems over time is comparable to or even higher than the immediate loss due to the extreme event itself.

Gaussian process (GP) regression is based on the context of Bayesian theory and statistical learning theory. It uses Gaussian process priors to perform regression analysis on observation points. In certain cases, the regression models may contain more information such as gradients and boundaries on specific points. Standard GP regression only relied on observation points will lead to infeasible function values. For example, a function y = sin^2?x is strictly non-negative, but the GPR model is possible to give a negative interval.

Renewable power generation resources are one of the biggest trends emerging thepower systems world. The inherent variability of these power sources brings chal-lenges in terms of planning, reliability and feasibility factors.

Rapid Pressure Swing Adsorption (RPSA) is a gas separation technology with an important commercial application for Medical Oxygen Concentrators (MOCs). MOCs use RPSA technology to produce high purity oxygen (O2) from ambient air, and provide medical oxygen therapy to Chronic Obstructive Pulmonary Disease (COPD) patients. COPD is a lung disease which prevents O2 from entering a patient's blood, and reduces the blood oxygen level. The standard therapy for COPD is to provide the patient with high purity (~90%) O2.

Invariant set is an important concept in the theory of dynamical systems and it has a wide range of applications in control and engineering. This thesis has four parts, each of which studies a fundamental problem arising in this field.

Medical oxygen concentrators (MOCs) are widely used to produce high purity oxygen forpatients suered from the Chronic Obstructive Pulmonary Disease (COPD), and mostcommercial MOCs use rapid pressure swing adsorption (RPSA) technology. Previousstudies proposed a multi-variable model predictive control (MPC) strategy applied tothe RPSA system which has been proved to have a better performance than controlusing only single manipulated variable. In practice, the RPSA system will experiencedunknown disturbances and parameter uncertainties.

This thesis proposes several optimization methods that utilize parallel algorithms for large-scale machine learning problems. The overall theme is network-based machine learning algorithms; in particular, we consider two machine learning models: graphical models and neural networks. Graphical models are methods categorized under unsupervised machine learning, aiming at recovering conditional dependencies among random variables from observed samples of a multivariable distribution.

Modern machine learning practices at the interface of big data, distributed environment and complex learning objectives post great challenges to designing scalable optimization algorithms with theoretical guarantees. This thesis, built on the recent advances in randomized algorithms, concerns development of such methods in practice and the analysis of their theoretical implications in the context of large-scale structured learning problems, such as regularized regression/classification, matrix completion, hierarchical multi-label learning, etc.

The feasibility of knitted fabric reinforcement for highly flexible composites has been investigated for the thermoforming process. The composite sheets were made through compression molding before being shaped. We used thermoplastic elastomers as matrices: Thermoplastic Elastomers and Thermoplastic Olefins. The knit reinforcement was provided by jersey knitted fabrics of polyester fibers. We first introduced the fundamentals involved in the study. The manufacturing is presented through compression molding and thermoforming.

Nonlocal modeling has been a subject undergoing intense study in recent years due to its capabilities to describe the physical system that the classic local model has difficulty in modeling. In this dissertation, we introduce the benefit of utilizing nonlocal models in some applications such as crack propagation and provide resolutions to some existing theoretical and numerical challenges in the nonlocal modeling. First, a meshfree quadrature rule is proposed to ensure the asymptotic compatibility in discretization.

This thesis consists of two parts, with two linkage mechanisms investigated: A spring-loaded four-bar linkage mechanism and the compound bow mechanism (a band-drive mechanism). In the first part of this thesis, a spring-loaded four-bar linkage mechanism is designed in one-step. This design includes kinematic synthesis, analysis, virtual work, static balancing and optimization.

Abstract Power plant generates electricity as well as high temperature waste heat. NOx is main pollutant in power plant flue gas. NOx reduction reaction is expected to happen in given duct section. Selective non-catalytic reduction (SNCR) method is selected to control NOx level. SNCR requires no modification on duct geometry and this method is more cost-effective compared to other NOx reduction systems. Computational fluid dynamics (CFD) technique is employed to simulate flue gas flow and mass and heat transports with an injection of droplets.