Dissertations / Theses on the topic 'Self-Organizing Feature Maps (SOFM)'

The Self-Organizing Map (SOM) is a type of neural network that forms a regular grid of neurons where clusters of neurons represent different classes of targets. The aim of this thesis is to design electromagnetic target classifiers by using the Self-Organizing Map (SOM) type artificial neural networks for dielectric and conducting objects with simple or complex geometries. Design simulations will be realized for perfect dielectric spheres and also for small-scaled aircraft targets modeled by thin conducting wires. The SOM classifiers will be designed by target features extracted from the scattered signals of targets at various aspects by using the Wigner distribution. Noise performance of classifiers will be improved by using slightly noisy input data in SOM training.

碩士<br>臺南師範學院<br>資訊教育研究所<br>87<br>The self-organizing feature maps（SOFM）net is one kind of unsupervised learning neural network. When the SOFM had learned the features of training patterns, the neurons, contain similar features, are close together. In this paper, we propose a multi-dimensional Self-Organizing Feature Maps (SOFM) to construct the relationship of Chinese characters. The proposed model extends the traditional one-dimensional relationship of Chinese characters （e.g. in a Chinese dictionary, Chinese characters with one same part are put together.）and then becomes a useful tool for orders of characters’education and measurement. Especially, it is useful to teach stroke orders of Chinese characters, Chinese characters with similar characteristics can be retrieved effectively. This technique can enhance the power of computer-assisted instruction（CAI）system and then the system becomes more “intelligent”. Keywords：self-organizing feature maps（SOFM）,stroke orders of Chinese characters, computer-assisted instruction （CAI）.

The rising demand for exacting performances from manufacturing systems has led to new challenges for the development of complex tool condition monitoring techniques. Although a wide range of monitoring methods have been investigated and developed, there has been very little migration of these innovations into industrial practice. The principal factor behind this phenomenon is the stochastic nature of the environment in which the system must function. A truly universal application has yet to be developed. The work presented here centres around the application of an unsupervised neural network model to the said problem. These networks learn without the aid of a human teacher or supervisor and learn to organise and re-organise themselves in accordance to the input data. This leads to the network structure reflecting the given input distribution more precisely than a predefined model, which generally follows a decay schedule. The dynamic nature of the process provides an evaluation of the underlying connectivity and topology in the original data space. This makes the network far more capable of capturing details in the target space. These networks have been successfully used in speech recognition applications and various pattern recognition tasks involving very noisy signals. Work is in progress on their application to robotics, process control and telecommunications. The procedure followed here has been to conduct experimental drilling trials using solid carbide drills on a Duplex Stainless Steel workpiece. Duplex Stainless Steel was chosen as a preferred metal for drilling experiments because of this high strength, good resistance to corrosion, low thermal expansion and good fatigue resistance. During the drilling trials, forces on the workpiece along the x, y and z axes were captured in real time and moments of the forces were calculated using these values. These three axial forces, along with their power spectral densities and moments were used as input parameters to the Artificial Neural Network model which followed the Self-Organising Map algorithm to classify this data. After the network was able to adapt itself to classify this real world data, the generated model was tested against a different set of data values captured during the drilling trials. The network was able to correctly identify a worn out drill from a new drill from this previously unseen set of data. This autonomous classification of the drill wear state by the neural network is a step towards creating a “universal” application that will eventually be able to predict tool wear in any machining operation without prior training.

碩士<br>逢甲大學<br>交通工程與管理所<br>93<br>Traffic accidents can be resulted from various factors. Consequently, authentication on accident liabilities can be very tedious and difficult. Due to the fact that information collected in traffic accident reports are normally incomplete and are varied from case to case. Similar cases can sometimes be authenticated with different liabilities. On the other hand, knowledge and experiences of committee members can also bias his/her judgment on similar cases at different time. Such variance of authentication on similar cases can easily be criticized by general public, and may hurt the image of government-established investigation committees. Consequently, it is apparent that there is a need to unify liability authentication of all traffic accident investigation committees. In this study, a liability authentication support system was constructed by using self-organizing feature maps. This system is intended to provide accident records and liability authentication results similar to inquiries as supplementary information to committee members. Hopefully, righteousness and fairness can be better reached with the help of this system. Due to the fact that accidents involving more than three cars can be very complicated, this study was thus limited to two-car accidents. The first step to construct the proposed system is to establish a self-organizing feature map (SOM) model for two-car crashes. Effectiveness of SOM models were checked by using the Silhouette coefficients (SC). After SC value for every cluster being determined, the best clusters were chosen to be the proposed SOM models. Grey relation analysis was then employed to decide order of referable cases. Traffic accident information adopted in this study is abstracted from the database constructed by the center for traffic accident authentication in Feng Chia university. The grey relational values between new cases and reference cases calculated from the selected SOM models were found range between 0.6458 and 1. Average grey relational value of same crash was approximately 0.8208. Average grey relational value of crosswise crash was approximately 0.8668. Average grey relational value of opposite crash was approximately 0.8641. These values indicated that the proposed models do have ability to provide similar accident cases as the inquiry. With the selected SOM models, a decision support system for traffic accident liability authentication is constructed using Active Sever Pages (ASP). The system is designed to provide characteristics and liability authentication results of cases similar to user input inquiries. Meanwhile, traffic safety rules related to the input inquiry can also be provided to the users for reference. Although initial results appeared to be acceptable, the system is still under development. In this paper, basic example is provided for better understanding of the system. Any comment or suggestion will certainly be sincerely appreciated.

M.Ing.<br>This project investigates the use of Self-Organizing Maps (SOM) for feature recognition and analysis in 3D objects. Object data was generated to simulate data obtained from 3D scanning and trained using SOM. The trained data was analysed using speci cally developed software. The feature recognition and analysis process can be summarized as follows: a 3D object le is converted to a pure 3D data le, this data le is trained using the SOM algorithm after which the output is analyzed using a 3D object viewer and SOM data display.

碩士<br>國立雲林科技大學<br>工業工程與管理研究所<br>83<br>The Vehicle Routing Problem(VRP) is the problem of designing a minimum cost set of routes for a fleet of delivery vehicles of fixed capacity. Since the VRP is considered as the class of NP- hard problems, optimum solutions can only be found in polynomial time. Kohonen''s Self-Organizing Feature Maps(SOFM) has the topo- logical characteristics that can be effectively used in solving combinatorial optimization problem such as Traveling Salesman Problem(TSP). In this research, a neural network based approach which implements SOFM model is proposed to solve the VRP. The VRP is presented by a M one-dimensional ring topology structure and add capacity constraint in learning process of selecting winner neuron. Finally, a simulation study is conducted and the results show good performances on computation time and solution quality.

碩士<br>國立臺北科技大學<br>生產系統工程與管理研究所<br>89<br>Data mining has become popular and is a rapidly emerging field. Various data mining techniques can be used to improve decision support systems. In this research, clustering technology for market segmentation is the main theme. Genetic algorithms (GA) are theoretically and empirically found to provide global near-optimal solutions for various complex optimization problems. Since GA is good at searching, GA can cluster the data according to their similarities. This research proposes a two-stage method, which first uses the Self-Organizing Feature Maps (SOM) to determine the number of clusters and then employs GA-based clustering method to find the final solution (it is defined as S+G in this research). Besides, a modified two-stage method, which first uses the self-organizing feature maps to determine the number of clusters and the starting points and then employs the K-means method to find the final solution, is proposed by Kuo [1] for market segmentation (it is defined as S+K in this research). The computational performance of K-means, S+K, and S+G is compared via a Monte Carlo study. As this research, S+G gets better computational performance than the other two methods based on within-cluster variations (SSW). In order to further testify the proposed approach, S+G, a real-world problem, the wireless telecommunications industry market segmentation, is employed. The questionnaire is designed, and factor analysis technique is used to extract the factors, as the basis of market segmentation. It is expected to make some suggestions about market strategy for the wireless telecommunications industry.