A point process model for the packet stream arising in teletraffic processes is the discrete, non-negative integer-valued, stationary process introduced by Neuts and Pearce. In this thesis, we examine an empirical approach to develop a monitoring scheme for that point process. Monitoring is a procedure of tracking a stochastic process to identify quickly the development of anomalous situations in the evolution of that process and detect their assignable causes. Further, a data-analytic scheme to evaluate the order of a Markov chain that quantifies the local dependence embedded in the point process and Walsh spectral techniques are examined.

In this thesis, we consider determining the economic lot sizes for a finite production rate assembly system with n facilities. Costs at each facility consist of a stationary positive echelon holding cost, and a fixed set up cost. The goal is to determine the production lot size at each facility in order to minimize the long-run total average cost of the system. Power-of-two policies, in which the lot size at each facility is a power of two times some base lot size, are considered. A 94%-effective power-of-two policy is determined from the optimal solution to a continuous relaxation problem by an O(n) algorithm, while a 98%-effective power-of-two policy is found using an O(n log n) algorithm. Near optimal solutions to the continuous relaxation problem are found by a subgradient optimization procedure and a cyclic coordinate descent method. Computational results suggest both methods are efficient for very large systems.

This thesis compares three approaches for assigning workers and scheduling the flow of jobs in a Group Technology (GT) cell. The performance measure considered is the throughput time of jobs. The performance of the three strategies is analyzed by varying the following factors: (1) The ratio of Batch processing time to setup time. (2) Utilization. (3) The number of operators assigned to the cell. The scheduling strategies considered are Individual Machine Loading, Cell Dedication and Worker Batch Assignment. Analytical queueing models as well as computer simulations are used to compare the performance of the scheduling strategies. The dedicated cells perform best with large batch sizes and moderate workloads. The Worker Batch Assignment strategy does as well as traditional batch handling techniques with respect to the throughput time measure. This may be important since this kind of job assignment strategy has the advantage of augmenting quality (due to direct accountability) and enhancing morale without adversely impacting throughput times.

No description available.

No description available.

No description available.

No description available.

Approved for public release; distribution is unlimited / The thesis addresses the problem of optimally re-assigning strike aircraft to targets in response to the emergence of "pop-ups" or time-sensitive targets. The first part of thesis develops an automated decision aid to rapidly revise the current air tasking order (ATO), so as to: maximize achievement of target destruction goals (weighted by target priorities), minimize attrition risk to employed assets, and disrupt the current ATO as little as possible. The second part of the thesis develops a detailed test and evaluation plan to conduct a comparison of two competing automated decision aids and the current manual reassignment methods. Critical operational issues, measures of effectiveness and measures of performance were developed to fully evaluate operational performance. The time-sensitive-targeting decision aid was tested and validated during major air strike live exercises at Marine Aviation Weapons and Tactics Squadron One. Careful measurements comparing the re-taskings recommended by the decision aid against actual decisions demonstrated that in every case the model's solutions were of better or equal quality, maximized combat asset utilization, and were achieved significantly faster. / Major, United States Marine Corps

Many military Intelligence Surveillance and Reconnaissance (ISR) operations would benefit greatly from a fleet of disparate sensor-bearing UAVs that are tightly integrated via a communications network, work cooperatively for a common operational objective, enhance situation awareness of the areas of operation, and increase persistence of sensor dwell time on strategic targets. This would enable continuity in the entire target acquisition cycle, from detection to classification to identification and finally localization of targets, in a diverse and dynamic environment. The integration of sensors and development of tactics in a cooperative sensing environment is one of the current focuses among the military intelligence community, and hence motivates this thesis effort. By building models with an existing agent-based simulation platform and using an extremely efficient experimental design methodology, numerous factors which could potentially affect the effectiveness of a cooperative sensing network against two arrays of targets are explored. The factors considered include UAV airspeed, reliability, detection/classification coverage and probability, network latency and degradation, UAV configurations and responsiveness, as well as air space separation. The two arrays of targets are mobile armor concentrations and time critical targets / these vary in their deployment profiles, vulnerability constraints and ease of detectability. Factors characterizing these targets, such as the shoot-and-scoot behavior of time critical targets, are also investigated. The study provides operational insights pertaining to the design and effective use of cooperative sensing for ISR purposes. These include the importance of having good UAV sensor capabilities, the need for a suite of sensors to aid in locating well-camouflaged time-critical targets, and the need for "intelligent" application of UAV cooperation tactics based on the characteristics of recently-classified targets.

This thesis studies "the convoy-path interdiction problem" (CPIP) in which an interdictor uses limited resources to attack and disrupt road segments ("arcs") or road intersections ("nodes") in a road network in order to delay an adversary's convoy from reaching its destination. The convoy will move between a known origin node a and destination node b using a "quickest path." We first show how to compute, using an A* search, the convoy's quickest path under the assumptions that the convoy may occupy several arcs simultaneously, each arc may have a different speed limit, and the convoy maintains constant inter-vehicle spacing. The basic model assumes that the convoy moves in a single lane of traffic; an extension handles arcs that may have multiple lanes. Using that algorithm as a subroutine, a decomposition algorithm solves the optimal interdiction problem. Interdiction of a node or arc makes that node or arc impassable. Computational results are presented on grid networks with up to 629 nodes and 2452 arcs with varying levels of interdiction resource. Using Xpress-MP optimization software and a 2 GHz Pentium IV computer, the largest network problem solves in no more than 360 seconds given that at most 4 arcs can be interdicted.