Endüstri Mühendisliği Bölümü Tez Koleksiyonu
http://hdl.handle.net/11655/428
2021-04-19T08:43:27ZBaskılı Devre Kartı Montajında Hata Birlikteliklerini ve Örüntülerini Keşfederek Ürün Kalitesini İyleştirmek için Veri Madenciliği
http://hdl.handle.net/11655/23216
Baskılı Devre Kartı Montajında Hata Birlikteliklerini ve Örüntülerini Keşfederek Ürün Kalitesini İyleştirmek için Veri Madenciliği
Parlaktuna, Ayşe Merve Turanlı
To achieve operational excellence, companies must satisfy the customers’ quality expectations while reducing the number of encountered defects and reworks. This can be achieved only if companies produce the product right at the first time. In this study, defects, which are encountered while producing printed circuit board cards are investigated via data mining rather than conventional methods. Both associations and sequential patterns of defects were analyzed to find out a common root cause that can trigger different kind of defects. Two different sets of defect data are investigated in this study. Apriori algorithm was selected for association rules,while Sequential Pattern Discovery using Equivalence Classes – SPADE algorithm was considered for sequential pattern mining. Defects are analyzed in two ways for association mining; card-based and year-based. On the other hand, in sequential pattern mining, associated defects relationship over time were investigated. During the investigations, there were huge number of rule sets to be analyzed and these rule sets are not easy to analyze or have a meaningful conclusion. Thus, meta rules approach was utilized to come up with reliable results. One goal of this thesis is to standardize this analysis method for different companies at different application areas. To achieve this goal, two companies’ data were used to standardize the method and compare the results. R software is used to create an interface for visualization of data, which extends the standardization procedure. Within this thesis, exemplifications for organizing data, implementation methods for analysis, a procedure for generating meaningful rule sets as well as visualization tools for comparison of results is provided.
2020-09-01T00:00:00ZDetermınıng The Best Settıngs for the Operators and Parameters of Genetıc Algorıthms: A Methodology and Its Applıcatıon to Travelıng Salesperson Problem
http://hdl.handle.net/11655/22807
Determınıng The Best Settıngs for the Operators and Parameters of Genetıc Algorıthms: A Methodology and Its Applıcatıon to Travelıng Salesperson Problem
Akduran, Yavuzhan
Genetic Algorithms (GAs) are heuristic algorithms that are used to approximate the optimal solutions of optimization problems. They are inspired by the theory of natural evolution, where a population of solutions evolves through generations and only the fittest individuals survive at the end. GAs perform very well in many optimization problems in terms of approximation quality and run time. However, a typical GA has several operators such as mutation and crossover, and parameters such as population size and generation number that affect the performance of the GA significantly. In the literature, the operators and parameters of GAs are set based on either the previous experiences of users or trial-error experiments since finding optimum settings of GAs is quite difficult.
In this thesis, a methodology is developed for effectively setting the operators and parameters of GAs. Hence, the best settings that will exploit the potential of the used genetic algorithm can be determined. Typically, performance of a GA is evaluated based on two criteria: (1) approximation quality and (2) run time. Approximation quality of an algorithm is determined based on the closeness of the solution found by the algorithm to the optimal solution. Run time is measured by the computational time the algorithm consumes until termination. In general, there are trade-offs between these two criteria, i.e. higher approximation quality requires more run time, and a GA is expected to find a solution with high approximation quality in a short time. Settings of the operators and parameters affect both criteria, and different settings can be advantageous in terms of different criteria. Therefore, we model the problem of effectively setting the parameters of a GA as a multi-objective optimization problem, using approximation quality and run time as the objectives.
In the thesis, we employ a multi-objective evolutionary algorithm (MOEA) to solve the problem and discover the trade-offs between approximation quality and run time of GAs. MOEAs are population-based heuristics that mimic natural evolution process and find a well-converged and well-diversified set of nondominated solutions. In our approach, each solution of MOEA represents a setting for operators and parameters of the GA considered. To evaluate a solution in the population, the GA is run using the settings defined in the solution. The fittest settings in terms of approximation quality and run time survive through generations. At the end, a set of settings, each of which is better on another criterion, is found.
The developed methodology is demonstrated on travelling salesperson problems (TSP). A GA that is used to solve TSP is selected. Several alternatives for operators and parameters are considered for the GA, and the best settings are investigated by experimenting on 31 TSP instances selected from the literature. The set of best settings is searched using NSGA-II, a well-known MOEA. Then a greedy heuristic is developed to help decision makers to reduce the size of the set of final solutions based on their preferences.
2020-01-01T00:00:00ZA Probabılıstıc Project Control Tool For Projects Wıth Hıgh Rısks and Uncertaınty
http://hdl.handle.net/11655/22775
A Probabılıstıc Project Control Tool For Projects Wıth Hıgh Rısks and Uncertaınty
Sü, Yasemin
Project monitoring and control are essential for project success. One of the most commonly used project control methods is Earned Value Management (EVM). EVM ensures that the projects are controlled in terms of cost, time and scope of the work and can make estimates about the completion time and cost according to the progress of the project. However, the common feature of the projects is that they contain risk and uncertainty and since EVM does not take into account uncertainty and risk factors, it is not effective in projects with high risk and uncertainty.
This study aims to develop a project control tool that is capable of effective project control under uncertainty and risk. The tool can control the project in multiple dimensions in terms of cost, time and scope, and it can calculate the uncertainty and causal risk factors related to these parameters. The tool uses Bayesian Networks (BNs) to model uncertainty and risk factors in the project parameters and to make statistical calculations related to them. BNs offer a powerful modeling technique for modeling and calculating probabilistic relationships, allowing expert knowledge and data to be combined.
In order to examine the applicability of the developed tool to different project areas, case studies will be examined in three different areas. Two of these case studies were based on real project data from different sectors. The positive and negative sides of the developed tool will be evaluated.
2020-01-01T00:00:00ZHeurıstıc Approaches for The Multı-Objectıve Routıng Problem for A Fleet of Unmanned Aerıal Vehıcles
http://hdl.handle.net/11655/22747
Heurıstıc Approaches for The Multı-Objectıve Routıng Problem for A Fleet of Unmanned Aerıal Vehıcles
Bişkin, Büşra
Nowadays, Unmanned Aerial Vehicles (UAVs) are extensively employed for various missions with different purposes. In every mission, different goals and problem structures are considered. In this thesis, we study the routing problem of a fleet of identical UAVs under multiple objectives. UA Vs in the fleet, which have limited flight durations, take off from a base, visit a number of targets in a two-dimensional mission area, and return to the base. We assume that the targets have different priorities, and the UAVs try to visit as many targets as possible to collect maximum reward within flight limits. We consider the following three objectives: minimizing the total distance traveled by the fleet, maximizing the total reward collected from the targets, and minimizing the total radar threat. We address two versions of the problem: routing in a radar-free terrain (with distance and reward as objectives) and routing in a radar-monitored terrain (with all three objectives). We aim to find efficient routes for each UAV in the fleet and the trajectory between pairs of targets in each route.
We employ two solution approaches for each version of our problem. First, we model the problem as a Multi-Objective Team Orienteering Problem (MOTOP) and find exact solutions. In our second approach, we utilize an Evolutionary Algorithm, EA-fUAV (Evolutionary Algorithm for routing a fleet of UAVs), to approximate efficient solutions in reasonable time. We test both approaches on three different problem cases. The results show that EA-fUAV approximates the efficient set well in reasonable time.
2019-12-01T00:00:00Z