Quality has become a central concern as it has been observed that reducing defects will lower the cost of production. Hence, companies generate and store vast amounts of quality related data. Analysis of this data is critical in order to understand the quality problems and their causes, and to take preventive actions. In this thesis, we propose a methodology for this analysis based on one of the data mining techniques, association rules. The methodology is applied for quality related data of an electronics company. Apriori algorithm used in this application generates an excessively large number of rules most of which are redundant. Therefore we implement a three phase elimination process on the generated rules to come up with a reasonably small set of interesting rules. The approach is applied for two different data sets of the company, one for production defects and one for raw material non-conformities. We then validate the resultant rules using a test data set for each problem type and analyze the final set of rules.

In industrial quality improvement and design studies, an important aim is to improve the product or process quality by determining factor levels that would result in satisfactory quality results. In these studies, quality characteristics that are qualitative are often encountered. Although there are many effective methods proposed for parameter optimization (robust design) with continuous responses, the methods available for qualitative responses are limited. In this study, a parameter optimization method for solving binary response robust design problems is proposed. The proposed method uses Mahalanobis Taguchi System to form a classification model that provides a distance function to separate the two response classes. Then, it finds the product/process variable settings that minimize the distance from the desired response class using quadratic programming. The proposed method is applied on two cases previously studied using Logistic Regression. The classification models are formed and the parameter optimization is conducted using the formed MTS models. The results are compared with those of the Logistic Regression. Conclusions and suggestions for future work are given.

In this study, we propose a bi-objective facility location model that considers both partial coverage and service to uncovered demands. In this model, it is assumed that the demand nodes within the predefined distance of opened facilities are fully covered and after that distance the coverage level linearly decreases. The objectives are the maximization of the sum of full and partial coverage the minimization of the maximum distance between uncovered demand nodes and their closest opened facilities. We apply two existing Multi Objective Genetic Algorithms (MOGAs), NSGA-II and SPEA-II to the problem. We determine the drawbacks of these MOGAs and develop a new MOGA called modified SPEA-II (mSPEA-II) to avoid the drawbacks. In this method, the fitness function of SPEA-II is modified and the crowding distance calculation of NSGA-II is used. The performance of mSPEA-II is tested on randomly generated problems of different sizes. The results are compared with the solutions resulting from NSGA-II and SPEA-II. Our experiments show that mSPEA-II outperforms both NSGA-II and SPEA-II.

In this study we focus on two-sided mixed-model assembly line balancing type-I problem. There is a production target for a fixed time horizon and the objective is to produce this amount with the minimum level of workforce. A mathematical model is developed to solve this problem in an optimal manner. For large scale problems, the mathematical model fails to give the optimal solution within reasonable computational times. Thus, a heuristic approach based on threshold accepting algorithm is presented. Both the mathematical model and the heuristic approach are executed to solve several example problems from the literature and a case study problem which is derived from the refrigerator production. Computational experiments are carried out using both approaches. It is observed that the heuristic procedure finds good solutions within very reasonable computational times.

In large data, set of mined association rules are typically large in number and hard to interpret. Some grouping and pruning methods have been developed to make rules more understandable. In this study, one of these methods is modified to be more effective and more efficient in applications including low thresholds for support or confidence, such as association analysis of product/process quality improvement. Results of experiments on benchmark datasets show that the proposed method groups and prunes more rules. In the literature, many rule reduction methods, including grouping and pruning methods, have been proposed for different applications. The variety in methods makes it hard to select the right method for applications such those of quality improvement. In this study a novel performance comparison basis is introduced to address this problem. It is applied here to compare the improved method to the original one. The introduced basis is tailored for quality data, but is flexible and can be changed to be applicable in other application domains.

In this study, a metaheuristic procedure, particularly a variable neighborhood search procedure, is proposed to solve the combined location and selective traveling salesman problem in glass recycling. The collection of used glass is done by a collecting vehicle that visits a number of predefined collection centers, like restaurants and hospitals that are going to be referred to as compulsory points. Meanwhile, it is desired to locate a predetermined number of bottle banks to residential areas. The aim is to determine the location of these bottle banks and the route of the collecting vehicle so that all compulsory points and all bottle banks are visited and the maximum profit is obtained. Population zones are defined in residential areas and it is assumed that the people in a particular population zone will recycle their used glass to the closest bottle bank that fully or partially covers their zone. A Variable Neighborhood Search algorithm and its variant have been utilized for the solution of the problem. Computational experiments have been made on small and medium scale test problems, randomly generated and adapted from the literature.

In this dissertation, we consider the clustering problem in data sets with unknown number of clusters having arbitrary shapes, intracluster and intercluster density variations. We introduce a clustering methodology which is composed of three methods that ensures extraction of local density and connectivity properties, data set reduction, and clustering. The first method constructs a unique neighborhood for each data point using the connectivity and density relations among the points based upon the graph theoretical concepts, mainly Gabriel Graphs. Neighborhoods subsequently connected form subclusters (closures) which constitute the skeleton of the clusters. In the second method, the external shape concept in computational geometry is adapted for data set reduction and cluster visualization. This method extracts the external shape of a non-convex n-dimensional data set using Delaunay triangulation. In the third method, we inquire the applicability of Swarm Intelligence to clustering using Ant Colony Optimization (ACO). Ants explore the data set so that the clusters are detected using density break-offs, connectivity and distance information. The proposed ACO-based algorithm uses the outputs of the neighborhood construction (NC) and the external shape formation. In addition, we propose a three-phase clustering algorithm that consists of NC, outlier detection and merging phases. We test the strengths and the weaknesses of the proposed approaches by extensive experimentation with data sets borrowed from literature and generated in a controlled manner. NC is found to be effective for arbitrary shaped clusters, intracluster and intercluster density variations. The external shape formation algorithm achieves significant reductions for convex clusters. The ACO-based and the three-phase clustering algorithms have promising results for the data sets having well-separated clusters.

Sorting is the problem of assignment of alternatives into predefined ordinal classes according to multiple criteria. A new distance function based solution approach is developed for sorting problems in this study. The distance to the ideal point is used as the criteria disaggregation function to determine the values of alternatives. These values are used to sort them into the predefined classes. The distance function is provided in general distance norm. The criteria disaggregation function is determined according to the sample preference set provided by decision maker. Two mathematical models are used in order to determine the optimal values and assign classes. The method also proposes an approach for handling alternative opt imal solutions, which are widely seen in sorting problems. Probabilities of belonging to each class for an alternative are calculated using the alternative optimal solutions and provided as the outputs of the model. Decision maker assigns the alternatives into classes according to these probabilities. The method is applied to five data sets and results are provided for different performance measures. Different distance norms are tried for each data set and their performances are evaluated for each data set. The probabilistic approach is also applied to UTADIS. The performance of the distance based model and modified UTADIS are compared with the previous sorting methods such as UTADIS and classification tree. The developed method has new aspects such as using distances to ideal point for sorting purpose and providing probabilities of belonging to classes. The handling of alternative optimal solutions within the method instead of a post-optimality analysis is another new and c ritical aspect of the study.

The thesis considers one of the central problems of corporate planning for a construction company, the project selection and bidding decision, and a model is developed for the entire decision environment. The nature of decision systems is examined and considered to consist of the identification, evaluation and selection from a range of options. Corporate decisions are discussed leading to the conclusion that a suitable model is needed. A basic model is proposed in which three outcome criteria consisting of people, money and property are required to be assessed, the values of the outcome criteria being determined by four project characteristics. Some approaches to the solution of multiple criteria problems are examined. The implications of time are next considered and the use of Gottinger's sequential machines examined as a means of modelling the complexities involved. Non-deterministic aspects of the problem are introduced which, together with dynamical considerations, suggest a model of intermediate complexity to be appropriate. The final chapters of the thesis concentrate on some ways in which the computational burden associated with the model can be reduced. The role of decision strategies is examined as a means of identifying the most suitable options. The suitability of probabilistic approaches to modelling non-deterministic aspects is investigated and an empirical analysis of three sets of bidding data is made to examine some possible simplifying assumptions.

This study focuses on two types of routing problems based on standard Traveling Salesman Problem, which are TSP with pickup and delivery (TSPPD) and TSP with backhauls (TSPB). In both of these problems, there are two types of customers, i.e. &ldquo / delivery customers&rdquo / demanding goods from depot and &ldquo / pickup customers&rdquo / sending goods to depot. The objective is to minimize the cost of the tour that visits every customer once without violating the side constraints. In TSPB, delivery customers should precede the pickup customers, whereas the vehicle capacity should not be exceeded in TSPPD. The aim of the study is to propose good Evolutionary Algorithms (EA) for these two problems and also analyze the adaptability of an EA, originally designed for the standard TSP, to the problems with side constraints. This effort includes commenting on the importance of feasibility of the solutions in the population with respect to these side constraints. Having this in mind, different EA strategies involving feasible or infeasible solutions are designed. These strategies are compared by quantitative experiments realized over a set of problem instances and the results are given.