• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.34C no.6
• /
• pp.71-78
• /
• 1997

In this paper, a new neural network training algorithm will be devised for function approximator with good generalization characteristics and tested with the time series prediction problem using santaFe competition data sets. To enhance the generalization ability a constraint term of hidden neuraon activations is added to the conventional output error, which gives the curvature smoothing characteristics to multi-layer neural networks. A hybrid learning algorithm of the error-back propagation and Hebbian learning algorithm with weight decay constraint will be naturally developed by the steepest decent algorithm minimizing the proposed cost function without much increase of computational requriements.

• Journal of the Korean Society for Precision Engineering
• /
• v.15 no.7
• /
• pp.167-174
• /
• 1998

This paper presents an unoccluded cylindrical object pose measurement using a slit beam laser in which a robot recognizes all of the unoccluded objects from the top of jumbled objects, and picks them up one by one. The elliptical equation parameters of a projected curve edge on a slice are calculated using LSM. The coefficients of standard elliptical equation are compared with these parameters to estimate the object pose. The hamming distances between the estimated coordinates and the calculated ones are extracted as measures to evaluate a local constraint and a smoothing surface curvature. The edges between slices are linked using error function based on the edge types and the hamming distances. The linked edges on slices are compared with the model object's length to recognize the unoccluded object. This proposed method may provide a solution to the automation of part handling in manufacturing environments such as punch press operation or part assembly.

• KIPS Transactions on Computer and Communication Systems
• /
• v.10 no.2
• /
• pp.53-58
• /
• 2021

In this paper, an alternative finite memory structure(FMS) smoothing filter is developed for discrete-time state-space model with a control input. To obtain the FMS smoothing filter, unbiasedness will be required beforehand in addition to a performance criteria of minimum variance. The FMS smoothing filter is obtained by directly solving an optimization problem with the unbiasedness constraint using only finite measurements and inputs on the most recent window. The proposed FMS smoothing filter is shown to have intrinsic good properties such as deadbeat and time-invariance. In addition, the proposed FMS smoothing filter is shown to be equivalent to existing FMS filters according to the delay length between the measurement and the availability of its estimate. Finally, to verify intrinsic robustness of the proposed FMS smoothing filter, computer simulations are performed for a temporary model uncertainty. Simulation results show that the proposed FMS smoothing filter can be better than the standard FMS filter and Kalman filter.

• Journal of Advanced Marine Engineering and Technology
• /
• v.34 no.7
• /
• pp.943-950
• /
• 2010

An efficient meshfree method based on a stabilized conforming nodal integration method is developed for elastoplastic contact analysis of metal forming processes. In this approach, strain smoothing stabilization is introduced to eliminate spatial instability in Galerkin meshfree methods when the weak form is integrated by a nodal integration. The gradient matrix associated with strain smoothing satisfies the integration constraint for linear exactness in the Galerkin approximation. Strain smoothing formulation and numerical procedures for path-dependent problems are introduced. Applications of metal forming analysis are presented, from which the computational efficiency has been improved significantly without loss of accuracy.

• Journal of the Korean Society for Precision Engineering
• /
• v.23 no.9 s.186
• /
• pp.190-199
• /
• 2006

A new approach which combines implicit surface scheme and recursive subdivision method is suggested in order to fill the holes with complex shapes in the polygon model. In the method, a base surface is constructed by creating smooth implicit surface from the points selected in the neighborhood of holes. In order to assure C$^1$ continuity between the newly generated surface and the original polygon model, offset points of same number as the selected points are used as the augmented constraint conditions in the calculation of implicit surface. In this paper the well-known recursive subdivision method is used in order to generate the triangular net with good quality using the hole boundary curve and generated base implicit surface. An efficient anisotropic smoothing algorithm is introduced to eliminate the unwanted noise data and improve the quality of polygon model. The effectiveness and validity of the proposed method are demonstrated by performing numerical experiments for the various types of holes and polygon model.

• Proceedings of the IEEK Conference
• /
• 2003.07e
• /
• pp.1903-1906
• /
• 2003

This paper proposes an adaptive regularized noise smoothing algorithm for range image using the area decreasing flow method, which can preserve meaningful edges during the smoothing process. Although the area decreasing flow method can easily smooth Gaussian noise, it has two problems; ⅰ) it is not easy to remove impulsive noise from observed range data, and ⅱ) it is also difficult to remove noise near edge when the adaptive regularization is used. In the paper, therefore, the second smoothness constraint is addtionally incorporated into the existing regularization algorithm, which minimizes the difference between the median filtered data and the estimated data. As a result, the Proposed algorithm can effectively remove the noise of dense range data with edge preserving.

• Journal of Advanced Marine Engineering and Technology
• /
• v.21 no.1
• /
• pp.66-70
• /
• 1997

In this paper a new learning algorithm for curvature smoothing and improved generalization for multi-layer neural networks is proposed. To enhance the generalization ability a constraint term of hidden neuron activations is added to the conventional output error, which gives the curvature smoothing characteristics to multi-layer neural networks. When the total cost consisted of the output error and hidden error is minimized by gradient-descent methods, the additional descent term gives not only the Hebbian learning but also the synaptic weight decay. Therefore it incorporates error back-propagation, Hebbian, and weight decay, and additional computational requirements to the standard error back-propagation is negligible. From the computer simulation of the time series prediction with Santafe competition data it is shown that the proposed learning algorithm gives much better generalization performance.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.65 no.12
• /
• pp.2046-2052
• /
• 2016

Ramp metering is one of the most efficient and widely used control methods for an intelligent transportation management system on a freeway. Its objective is to control and upgrade freeway traffic by regulating the number of vehicles entering the freeway entrance ramp, in such a way that not only the alleviation of the congestion but also the smoothing of the traffic flow around the desired density level can be achieved for the maintenance of the maximum mainline throughput. When the cycle of the signal detection is larger than that of the system process, the density tracking problem needs to be considered in the form of the discrete-time system. Therefore, a discrete-time sliding mode control method is proposed for the ramp metering problem in the presence of both input constraint in the on-ramp and exogenous disturbance in the off-ramp considering the random behavior of the driver. Simulations were performed using a validated second-order macroscopic traffic flow model in Matlab environment and the simulation results indicate that proposed control method can achieve better performance than previously well-known ALINEA strategy in the sense that mainstream flow throughput is maximized and congestion is alleviated even in the presence of input constraint and exogenous disturbance.

• Industrial Engineering and Management Systems
• /
• v.10 no.1
• /
• pp.54-64
• /
• 2011

The purpose of this study is to optimally allocate the human resources to tasks while minimizing the total daily human resource costs and smoothing the human resource usage. The human resource allocation problem (hRAP) under consideration contains two kinds of special constraints, i.e. operational precedence and skill constraints in addition to the ordinary constraints. To deal with the multiple objectives and the special constraints, first we designed this hRAP as a network problem and then proposed a Pareto multistage decisionbased genetic algorithm (P-mdGA). During the evolutionary process of P-mdGA, a Pareto evaluation procedure called generalized Pareto-based scale-independent fitness function approach is used to evaluate the solutions. Additionally, in order to improve the performance of P-mdGA, we use fuzzy logic controller for fine-tuning of genetic parameters. Finally, in order to demonstrate the applicability and to evaluate the performance of the proposed approach, P-mdGA is applied to solve a case study in a hotel, where the managers usually need helpful automatic support for effectively allocating hotel staff to hotel tasks.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38C no.6
• /
• pp.503-511
• /
• 2013

This paper presents a speech enhancement method using non-negative matrix factorization. In training phase, we can obtain each basis matrix from speech and specific noise database. After training phase, the noisy signal is separated from the speech and noise estimate using basis matrix in enhancement phase. In order to improve the performance, we model the change of encoding matrix from training phase to enhancement phase using independent Gaussian distribution models, and then use the constraint of the objective function almost same as that of the above Gaussian models. Also, we perform a smoothing operation to the encoding matrix by taking into account previous value. Last, we apply the Log-Spectral Amplitude type algorithm as gain function.