• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.365-365
• /
• 2000

In this paper, we describe the system to recognize the six digit postal number of mails using neural network. Our zip-code recognition system consists of a preprocessing procedure for the original captured image, a segmentation procedure for separating an address block area with a shape, and recognition procedure for the cognition of a postal number. we extract the feature vectors that are the input of a neural network for the recognition process based on an area optimizing and an image thinning processing. The neural network classifies the zip-code in the mail and the recognized zip-code is verified through the zip-code database.

• Korean Management Science Review
• /
• v.18 no.2
• /
• pp.25-38
• /
• 2001

The well-known back-propagation algorithm for multi-layered neural network has successfully been applied to pattern c1assification problems with remarkable flexibility. Recently. the multi-layered neural network is used as a powerful data mining tool. Nevertheless, in many cases with complex boundary of classification, the successful learning is not guaranteed and the problems of long learning time and local minimum attraction restrict the field application. In this paper, an Improved learning procedure of multi-layered neural network is proposed. The procedure is based on the generalized delta rule but it is particular in the point that the architecture of network is not fixed but enlarged during learning. That is, the number of hidden nodes or hidden layers are increased to help finding the classification boundary and such procedure is controlled by entropy evaluation. The learning speed and the pattern classification performance are analyzed and compared with the back-propagation algorithm.

• Proceedings of the ESK Conference
• /
• 1996.10a
• /
• pp.197-205
• /
• 1996

Unification of neural network with a hierarchical pattern recognition is presented for recognizing large set of objects. A two-step identification procedure is developed for pattern recognition: coarse and fine identification. The coarse identification is designed for finding a class of object while the fine identification procedure is to identify a specific object. During the training phase a course neural network is trained for clustering larger set of reference objects into a number of groups. For training a fine neural network, expert neural network is also trained to identify a specific object within a group. The presented idea can be interpreted as two step identification. Experimental results are given to verify the proposed methodology.

• The Journal of Information Technology
• /
• v.8 no.1
• /
• pp.119-124
• /
• 2005

Learning procedure in the neural network is updating of weights between neurons. Unadequate initial learning coefficient causes excessive iterations of learning process or incorrect learning results and degrades learning efficiency. In this paper, adaptive learning algorithm is proposed to increase the efficient in the learning algorithms of Kohonens Self-Organization Neural networks. The algorithm updates the weights adaptively when learning procedure runs. To prove the efficiency the algorithm is experimented to clustering of the random weight. The result shows improved learning rate about 42~55% ; less iteration counts with correct answer.

• International Journal of Korean Welding Society
• /
• v.1 no.2
• /
• pp.12-17
• /
• 2001

With the advance of the robotic welding process, procedure optimization that selects the welding procedure and predicts bead width that will be deposited is increased. A major concern involving procedure optimization should define a welding procedure that can be shown to be the best with respect to some standard and chosen combination of process parameters, which give an acceptable balance between production rate and the scope of defects for a given situation. This paper presents a new algorithm to establish a mathematical model f3r predicting bead width through a neural network and multiple regression methods, to understand relationships between process parameters and bead width, and to predict process parameters on bead width for GMA welding process. Using a series of robotic arc welding, additional multi-pass butt welds were carried out in order to verify the performance of the neural network estimator and multiple regression methods as well as to select the most suitable model. The results show that not only the proposed models can predict the bead width with reasonable accuracy and guarantee the uniform weld quality, but also a neural network model could be better than the empirical models.

• Smart Structures and Systems
• /
• v.5 no.5
• /
• pp.507-515
• /
• 2009

The present work is an alternative methodology in order to balance a nonlinear highly flexible rotor by using neural networks. This procedure was developed aiming at improving the performance of classical balancing methods, which are developed in the context of linearity between acting forces and resulting displacements and are not well adapted to these situations. In this paper a fully experimental procedure using neural networks is implemented for dealing with the adaptive balancing of nonlinear rotors. The nonlinearity results from the large displacements measured due to the high flexibility of the foundation. A neural network based meta-model was developed to represent the system. The initialization of the learning procedure of the network is performed by using the influence coefficient method and the adaptive balancing strategy is prone to converge rapidly to a satisfactory solution. The methodology is tested successfully experimentally.

• Journal of the Korean Data and Information Science Society
• /
• v.20 no.5
• /
• pp.829-835
• /
• 2009

Parameter design is an approach to reducing performance variation of quality characteristic value in products and processes. Taguchi has used the signal-to-noise ratio to achieve the appropriate set of operating conditions where variability around target is low in the Taguchi parameter design. However, there are difficulties in practical application, such as complexity and nonlinear relationships among quality characteristics and control factors (design factors), and interactions occurred among control factors. Neural networks have a learning capability and model free characteristics. There characteristics support neural networks as a competitive tool in processing multivariable input-output implementation. In this paper we propose a substantially simpler optimization procedure for parameter design using neural network. An example is illustrated to compare the difference between the Taguchi method and neural network method.

• Journal of Mechanical Science and Technology
• /
• v.18 no.8
• /
• pp.1327-1337
• /
• 2004

An efficient multi-objective optimization method is presented making use of neural network and a systematic satisficing trade-off method (STOM), in order to simultaneously improve both maneuverability and durability of tire. Objective functions are defined as follows: the sidewall-carcass tension distribution for the former performance while the belt-edge strain energy density for the latter. A back-propagation neural network model approximates the objective functions to reduce the total CPU time required for the sensitivity analysis using finite difference scheme. The satisficing trade-off process between the objective functions showing the remarkably conflicting trends each other is systematically carried out according to our aspiration-level adjustment procedure. The optimization procedure presented is illustrated through the optimum design simulation of a representative automobile tire. The assessment of its numerical merit as well as the optimization results is also presented.

• Structural Engineering and Mechanics
• /
• v.32 no.1
• /
• pp.71-94
• /
• 2009

In this paper a procedure for Monte Carlo simulation of univariate stationary stochastic processes with the aid of neural networks is presented. Neural networks operate model-free and, thus, circumvent the need of specifying a priori statistical properties of the process, as needed traditionally. This is particularly advantageous when only limited data are available. A neural network can capture the "pattern" of a short observed time series. Afterwards, it can directly generate stochastic process realizations which capture the properties of the underlying data. In the present study a simple feed-forward network with focused time-memory is utilized. The proposed procedure is demonstrated by examples of Monte Carlo simulation, by synthesis of future values of an initially short single process record.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.15 no.5
• /
• pp.48-55
• /
• 2007

A bushing component of a vehicle suspension system is tested to capture the nonlinear behavior of rubber bushing element using the MTS 3-axes rubber test machine. The results of the tests are used to model the artificial neural network bushing model. The performances from the neural network model usually are dependent on the structure of the neural network. In this paper, maximum error, peak error, root mean square error, and error-to-signal ratio are employed to evaluate the performances of the neural network bushing model. A simple simulation is carried out to show the usefulness of the developed procedure.