Thesis (Ph. D.)--University of Hong Kong, 1991.

A robot wanders around an unfamiliar environment, performing actions and observing their perceptual consequences. The robot's task is to construct a model of its environment that will allow it to predict the outcome of its actions and to determine what action sequences take it to particular goal states. In any reasonably complex situation, a robot that aims to manipulate its environment toward some desired end requires an internal representation of the environment because the robot can directly perceive only a small fraction of the global environmental state at any time; some portion of the rest must be stored internally if the robot is to act effectively. Rivest and Schapire (72, 74, 87) have studied this problem and have designed a symbolic algorithm to strategically explore and infer the structure of finite-state environments. At the heart of this algorithm is a clever representation of the environment called an update graph. This dissertation presents a connectionist implementation of the update graph using a highly specialized network architecture and a technique for using the connectionist update graph to guide the robot from an arbitrary starting state to a goal state. This technique requires a critic that associates the update graph's current state with the expected time to reach the goal state. At each time step, the robot selects the action that minimizes the output of the critic. The basic control acquisition technique is demonstrated on several environments, and it is generalized to handle a navigation task involving a more realistic environment characterized by a high-dimensional continuous state-space with real-valued actions and sensations in which a simulated cylindrical robot with a sensor belt operates in a planar environment. The task is short-range homing in the presence of obstacles. Unlike many approaches to robot navigation, our approach assumes no prior map of the environment. Instead, the robot has to use its limited sensory information to construct a model of its environment. A connectionist architecture is presented for building such a model. It incorporates a large amount of a priori knowledge in the form of hard-wired networks, architectural constraints, and initial weights. This navigation example demonstrates the use of a large modular architecture on a difficult task.

The large number of schedules which must be evaluated to develop an optimal solution as opposed to the time required to evaluate a single schedule of the Three Machine Flow-Shop Scheduling Problem when minimizing Makespan, the n/3/F/Cmax problem, causes this problem to be considered unsolvable. The objective of this research is to demonstrate the existence of exploitable structural properties which can be used to construct optimal or near-optimal solutions to the n/3/F/Cmax problem. The interaction of two structural properties identified, the Job Class Decomposition and the Complementary Makespan Paths, has led to theoretical results which reduce the previously smallest solution space known to exist for the n/3/F/Cmax problem; the permutation sequences. These theoretical results indicate that certain subsets of the permutation sequences are always dominated in Makespan Value while other subsets will generally dominate all other permutation sequences in Makespan Value. Each of these subsets are shown to have definable structural properties which exist among the 6 Job Classes defined by the Job Class Decomposition, such as Symmetry 1 and Symmetry 2. Symmetry 1 and Symmetry 2 relate to some strong dependencies which appear to exist among the 6 Job Classes. These strong dependencies suggest that certain sequencing patterns, defined as the Preferred Sequences, generally lead to optimal or near optimal Makespan sequences when implemented in algorithmic form; while other sequencing patterns, defined as the Dominance Categories, lead to sequences that are always dominated in Makespan Value. Eleven Sequencing Rules based upon the properties of the Preferred Sequences and the 6 Job Classes are developed. Four simple constructive algorithms are based upon these sequencing rules. A fifth algorithm is designed to capture all the interrelationships of the properties identified in this thesis. Seven instances of two simple constructive algorithms, one instance each of the two other constructive algorithms, three instances of the fifth algorithm and Palmer's Method are implemented to solve 100 job sets defined in each of six experiments. The results of the six experiments indicate that the fifth algorithm and several instances of the first two simple constructive algorithms, in general, construct a superior Makespan sequence to the one constructed by Palmer's 'Slope index' Method.

M. Tech. Chemical Engineering. / In this work the development of a mathematical model capable of determining suitable operating conditions and reactor configuration to produce nanoparticle products of specified average diameter as well as their deposition to the reactor are presented. The developed model consists of mass, momentum and energy conservation equations. The aim of this model was to predict the effect of inlet gas flow rate, temperature, reactant concentration on the average diameter, surface area, number and volume concentration of product TiO2 nanoparticles and most importantly particle deposition on the average diameter of TiO2 nanoparticle products. Possible formulations of such a model are discussed on the basis of efficiency, cost effectiveness, amount of data generated, inclusion into the computational fluid dynamics (CFD) technique and the best option chosen thereof. The performance of the present monodisperse model was validated by comparing its predicted results with previously published numerical data. Good correlations between the predicted and numerical results were achieved.

No description available.

The computational continua (C2) framework, which is the focus of the present thesis, is a coarse-scale continuum description coupled with an underlying fine-scale description of material heterogeneity of finite size. It is intended to account for a variation of the coarse-scale stresses (strains) over a unit cell (UC) domain. It was originally developed to overcome the theoretical and computational limitations of higher order homogenization models and generalized continuum theories, namely the need for higher order finite element continuity, additional degrees-of-freedom, and non-classical boundary conditions. The key feature of the C2 is so-called nonlocal quadrature scheme (NLQS) defined over a computational continua domain consisting of a disjoint union of so-called computational unit cells (CUC). The CUCs, which are merely computational entities, have a shape and size of the physical periodic microstructure, but their positions depend on the size of the unit cell domain and is determined to reproduce the weak form of the governing equations on the fine scale. In the original C2 formulation the unit cell domains when mapped onto the parent element domains preserved their original shape. Thus, the nonlocal quadrature scheme was limited to structured meshes or meshes with slightly distorted elements. In the present thesis, it is accounted for that the CUCs when mapped onto the parent element domain, may no longer preserve their initial shape. Towards this end, an exact nonlocal quadrature scheme for distorted elements, which matches the two-dimensional monomials of the element, and an approximate tensor-product based nonlocal quadrature that eliminates the need for costly evaluation of the quadrature points for each element were developed. The performance of both nonlocal quadrature schemes is demonstrated in two-dimensional linear elasticity problems on several meshes and microstructures and compared with the classical first-order (O(1)) homogenization theory and the direct numerical simulation (DNS). The error in the overall behavior (total strain energy stored and L2 norm error in von Mises stress) of the C2 formulations offers a 10-20 (%) improvement over the O(1) theory. More substantial is the gain of the C2 formulations over the O(1) theory in the accuracy of the local stresses in critical locations. Finally, the performance of the tensor-product based approximate quadrature is comparable to that of the computationally costly exact nonlocal quadrature in terms of both the global and local error measures making it more attractive. In the wave propagation regime, the computational continua formulation showed strikingly accurate dispersion curves. Unlike classical dispersive methods pioneered more than a half a century ago where the unit cell is quasi-static and provides effective mechanical and dispersive properties to the coarse-scale problem, the dispersive C2 gives rise to transient problems at all scales and for all microphases involved. An efficient block time-integration scheme is proposed that takes advantage of the fact that the transient unit cell problems are not coupled to each other, but rather to a single coarse-scale finite element they are positioned in. It is showed that the computational cost of the method is comparable to the classical dispersive methods for short load durations. The scheme is proved to be stable. Finally, accuracy analysis on a wave propagation model problem demonstrates that the proposed scheme is substantially more accurate when compared with a O(1) homogenization scheme with microinertia effects.

published_or_final_version / abstract / toc / Mechanical Engineering / Master / Master of Philosophy

No description available.

No description available.

No description available.