Georgia Tech Institutional Repository
https://smartech.gatech.edu:443
The SMARTech digital repository system captures, stores, indexes, preserves, and distributes digital research material.2021-06-14T19:08:14ZService Network Design for Parcel Trucking
http://hdl.handle.net/1853/64814
Service Network Design for Parcel Trucking
Rocco Rocco, Adolfo Antonio
We develop a large-scale package express service network design methods using integer programming optimization models specified on flat network models that capture important timing constraints to ensure that package flows meet service constraints. In the first part, we focus on shuttle activities and develop optimization technology for the design of shuttle services using novel rate-based models to determine package flow paths as well as vehicle routes. A computational study using data from a large Chinese package company demonstrates that the technology produces a cost-effective service network design for shuttle schedules with excellent on-time performance. The second part presents a strategic hub selection problem developing a cost-effective greedy heuristic approach that solves tractable integer programming models to add a single intermediate hub on each iteration. A computational study shows that the greedy approach selects geographically-distributed and cost-effective hubs for package transfer, and moreover, the heuristic outperforms the full optimization model by a 20% gap difference for the relevant test instances. In the last part, we develop a new approach for solving the flow planning problem of service network design for large-scale networks with timing constraints. We introduce a so-called generalized in-tree, referred to as GIT, which has useful operational benefits. We demonstrate, via a computational study, that imposing a discretized GIT structure that groups remaining times into fixed-width buckets of 2 hours or 4 hours leads to solutions that are only 2% to 4% more costly than those that do not require GIT structure but significantly simpler to operationalize.
2021-05-06T00:00:00ZRocco Rocco, Adolfo AntonioWe develop a large-scale package express service network design methods using integer programming optimization models specified on flat network models that capture important timing constraints to ensure that package flows meet service constraints. In the first part, we focus on shuttle activities and develop optimization technology for the design of shuttle services using novel rate-based models to determine package flow paths as well as vehicle routes. A computational study using data from a large Chinese package company demonstrates that the technology produces a cost-effective service network design for shuttle schedules with excellent on-time performance. The second part presents a strategic hub selection problem developing a cost-effective greedy heuristic approach that solves tractable integer programming models to add a single intermediate hub on each iteration. A computational study shows that the greedy approach selects geographically-distributed and cost-effective hubs for package transfer, and moreover, the heuristic outperforms the full optimization model by a 20% gap difference for the relevant test instances. In the last part, we develop a new approach for solving the flow planning problem of service network design for large-scale networks with timing constraints. We introduce a so-called generalized in-tree, referred to as GIT, which has useful operational benefits. We demonstrate, via a computational study, that imposing a discretized GIT structure that groups remaining times into fixed-width buckets of 2 hours or 4 hours leads to solutions that are only 2% to 4% more costly than those that do not require GIT structure but significantly simpler to operationalize.REACTIVITY AND MECHANISM OF CATALYTIC METHANE CONVERSION OVER CERIA-ZIRCONIA SUPPORTED METAL/METAL OXIDE CATALYSTS
http://hdl.handle.net/1853/64813
REACTIVITY AND MECHANISM OF CATALYTIC METHANE CONVERSION OVER CERIA-ZIRCONIA SUPPORTED METAL/METAL OXIDE CATALYSTS
Lyu, Yimeng
Methane, the major component of natural gas, is vastly available on Earth, but is not used effectively. Methane conversion using heterogeneous catalysts, such as CZ supported metal/metal oxide materials, provides a promising solution to the challenge. The synthesis-structure relationship for such catalyst materials was revealed by thorough characterization of NiO/CZ catalysts synthesized differently. The structural properties are further related to methane dry reforming performances, suggesting that the Ni speciation and the oxygen storage capacity are the key factors shaping the deactivation behavior. Understanding methane activation over NiO/CZ catalysts provides important insights for rational catalyst design and was carried out using in-situ FTIR spectroscopy. Using a novel data analysis algorithm based on non-linear regression fitting, the evolution of different surface species is deconvoluted to elucidate the surface reaction pathways during methane activation, showing that different reactivity can be achieved by fine tuning the nature of active sites and reaction conditions. The selective oxidation of methane to methanol was realized over MOx/CZ (M=Ni, Cu, Fe) catalysts. The strong LAS concentration controls the formation of methoxy intermediates during methane activation, which in turn governs the selectivity towards methanol. Taken together, these examples reveal how systematic investigations of catalysts provide design principles for catalytic methane conversions.
2021-05-04T00:00:00ZLyu, YimengMethane, the major component of natural gas, is vastly available on Earth, but is not used effectively. Methane conversion using heterogeneous catalysts, such as CZ supported metal/metal oxide materials, provides a promising solution to the challenge. The synthesis-structure relationship for such catalyst materials was revealed by thorough characterization of NiO/CZ catalysts synthesized differently. The structural properties are further related to methane dry reforming performances, suggesting that the Ni speciation and the oxygen storage capacity are the key factors shaping the deactivation behavior. Understanding methane activation over NiO/CZ catalysts provides important insights for rational catalyst design and was carried out using in-situ FTIR spectroscopy. Using a novel data analysis algorithm based on non-linear regression fitting, the evolution of different surface species is deconvoluted to elucidate the surface reaction pathways during methane activation, showing that different reactivity can be achieved by fine tuning the nature of active sites and reaction conditions. The selective oxidation of methane to methanol was realized over MOx/CZ (M=Ni, Cu, Fe) catalysts. The strong LAS concentration controls the formation of methoxy intermediates during methane activation, which in turn governs the selectivity towards methanol. Taken together, these examples reveal how systematic investigations of catalysts provide design principles for catalytic methane conversions.An Architecting Methodology for Thermal Management Systems of Commercial Aircraft at the Conceptual Design Phase
http://hdl.handle.net/1853/64812
An Architecting Methodology for Thermal Management Systems of Commercial Aircraft at the Conceptual Design Phase
Shi, Mingxuan
An aircraft thermal management system (TMS) is a subsystem to handle the cooling and heating requirements of the whole aircraft. Traditionally, the TMS is architected based on experience. Its impacts on aircraft conceptual design are estimated using empirical data. However, the heating problem becomes much more serious because of more applications of the electrical systems and the increasing use of composite material. Moreover, novel aircraft concepts that generate much larger amount of heat are emerging. Therefore, there is a lack of historical data for such applications, which makes the conventional TMS architecting approaches inapplicable. To tackle such thermal management challenges, this research proposed the overarching research objective: to develop a TMS architecting methodology suitable for conceptual design phase of commercial aircraft, which is capable of handling increasing cooling loads and emerging aircraft concepts with limited historical data and only information available during early design stage.
The existing TMS architecting methods that generate architecture candidates highly rely on the intuition and experience of the researchers, potentially ignoring other innovative and non-intuitive architectures. Besides, to overcome the lack of data, physics-based modeling and simulation are heavily used for evaluation of TMS designs. However, if the number of candidates is too large, it is impractical to perform physics-based sizing, optimization, and analysis. Thus, an approach to narrow down the architecture space is required. Moreover, the exiting research focus on the evaluation of TMS designs based on fixed aircraft design. The interactions between designs of TMS and aircraft are not studied yet.
To fill these gaps, a backtracking architecting methodology that is guided by behaviors of fundamental physics is implemented to populate the TMS architecture space. To further narrow down the design space and perform optimal down-selection, a filtering process based on feasibility and clustering of key performance indicators. The interactions between designs of TMS and aircraft are studied by the integration of the TMS architecting process into the aircraft design loop.
The primary contributions of this dissertation are: 1. developed an architecting methodology that can systematically populate both intuitive and non-intuitive TMS architectures; 2. developed a filtering method based on feasibility and clustering of clustering of key performance indicators, which enables rapid narrowing down of the architecture candidate space; 3. developed an integrated design framework of the aircraft to incorporate TMS designs.
2021-05-01T00:00:00ZShi, MingxuanAn aircraft thermal management system (TMS) is a subsystem to handle the cooling and heating requirements of the whole aircraft. Traditionally, the TMS is architected based on experience. Its impacts on aircraft conceptual design are estimated using empirical data. However, the heating problem becomes much more serious because of more applications of the electrical systems and the increasing use of composite material. Moreover, novel aircraft concepts that generate much larger amount of heat are emerging. Therefore, there is a lack of historical data for such applications, which makes the conventional TMS architecting approaches inapplicable. To tackle such thermal management challenges, this research proposed the overarching research objective: to develop a TMS architecting methodology suitable for conceptual design phase of commercial aircraft, which is capable of handling increasing cooling loads and emerging aircraft concepts with limited historical data and only information available during early design stage.
The existing TMS architecting methods that generate architecture candidates highly rely on the intuition and experience of the researchers, potentially ignoring other innovative and non-intuitive architectures. Besides, to overcome the lack of data, physics-based modeling and simulation are heavily used for evaluation of TMS designs. However, if the number of candidates is too large, it is impractical to perform physics-based sizing, optimization, and analysis. Thus, an approach to narrow down the architecture space is required. Moreover, the exiting research focus on the evaluation of TMS designs based on fixed aircraft design. The interactions between designs of TMS and aircraft are not studied yet.
To fill these gaps, a backtracking architecting methodology that is guided by behaviors of fundamental physics is implemented to populate the TMS architecture space. To further narrow down the design space and perform optimal down-selection, a filtering process based on feasibility and clustering of key performance indicators. The interactions between designs of TMS and aircraft are studied by the integration of the TMS architecting process into the aircraft design loop.
The primary contributions of this dissertation are: 1. developed an architecting methodology that can systematically populate both intuitive and non-intuitive TMS architectures; 2. developed a filtering method based on feasibility and clustering of clustering of key performance indicators, which enables rapid narrowing down of the architecture candidate space; 3. developed an integrated design framework of the aircraft to incorporate TMS designs.Frank-Wolfe Methods for Optimization and Machine Learning
http://hdl.handle.net/1853/64811
Frank-Wolfe Methods for Optimization and Machine Learning
Combettes, Cyrille W.
In Chapter 2, we present the Frank-Wolfe algorithm (FW) and all necessary background material. We explain the projection-free and sparsity properties of the algorithm, provide motivation for real-world problems, and analyze the convergence rates and a lower bound on the complexity.
In Chapter 3, we review the complexity bounds of linear minimizations and projections on several sets commonly used in optimization, providing a rigorous support to the use of FW. We also propose two methods for projecting onto the lp-ball and the Birkhoff polytope respectively, and we analyze their complexity. Computational experiments for the l1-ball and the nuclear norm-ball are presented.
In Chapter 4, we identify the well-known drawback in FW, a naive zig-zagging phenomenon that slows down the algorithm. In response to this issue, we propose a boosting procedure generating descent directions better aligned with the negative gradients and preserving the projection-free property. Although the method is relatively simple and intuitive, it provides significant computational speedups over the state of the art on a variety of experiments.
In Chapter 5, we address the large-scale finite-sum optimization setting arising in many tasks of machine learning. Based on a sliding technique, we propose a generic template to integrate adaptive gradients into stochastic Frank-Wolfe algorithms in a practical way. Computational experiments on standard convex optimization problems and on the nonconvex training of neural networks demonstrate that the blend of the two methods is successful.
Both developments in Chapters 4 and 5 are motivated by the projection-free property of FW. In Chapter 6, we leverage the natural sparsity of the iterates generated by FW and study an application to the approximate Carathéodory problem. We show that FW generates a simple solution to the problem and that with no modification of the algorithm, better cardinality bounds can be established using existing convergence analysis of FW in different scenarios. We also consider a nonsmooth variant of FW.
In Chapter 7, we carry on with the sparsity property and we consider an extension of the Frank-Wolfe algorithm to the unconstrained setting. It addresses smooth convex optimization problems over the linear span of a given set and resembles the matching pursuit algorithm. We propose a blending method that combines fast convergence and high sparsity of the iterates. Computational experiments validate the purpose of our method.
2021-05-01T00:00:00ZCombettes, Cyrille W.In Chapter 2, we present the Frank-Wolfe algorithm (FW) and all necessary background material. We explain the projection-free and sparsity properties of the algorithm, provide motivation for real-world problems, and analyze the convergence rates and a lower bound on the complexity.
In Chapter 3, we review the complexity bounds of linear minimizations and projections on several sets commonly used in optimization, providing a rigorous support to the use of FW. We also propose two methods for projecting onto the lp-ball and the Birkhoff polytope respectively, and we analyze their complexity. Computational experiments for the l1-ball and the nuclear norm-ball are presented.
In Chapter 4, we identify the well-known drawback in FW, a naive zig-zagging phenomenon that slows down the algorithm. In response to this issue, we propose a boosting procedure generating descent directions better aligned with the negative gradients and preserving the projection-free property. Although the method is relatively simple and intuitive, it provides significant computational speedups over the state of the art on a variety of experiments.
In Chapter 5, we address the large-scale finite-sum optimization setting arising in many tasks of machine learning. Based on a sliding technique, we propose a generic template to integrate adaptive gradients into stochastic Frank-Wolfe algorithms in a practical way. Computational experiments on standard convex optimization problems and on the nonconvex training of neural networks demonstrate that the blend of the two methods is successful.
Both developments in Chapters 4 and 5 are motivated by the projection-free property of FW. In Chapter 6, we leverage the natural sparsity of the iterates generated by FW and study an application to the approximate Carathéodory problem. We show that FW generates a simple solution to the problem and that with no modification of the algorithm, better cardinality bounds can be established using existing convergence analysis of FW in different scenarios. We also consider a nonsmooth variant of FW.
In Chapter 7, we carry on with the sparsity property and we consider an extension of the Frank-Wolfe algorithm to the unconstrained setting. It addresses smooth convex optimization problems over the linear span of a given set and resembles the matching pursuit algorithm. We propose a blending method that combines fast convergence and high sparsity of the iterates. Computational experiments validate the purpose of our method.