• Journal of Institute of Control, Robotics and Systems
• /
• v.17 no.3
• /
• pp.211-217
• /
• 2011

This paper proposes a quasi-optimal linear DOA (Direction-of-Arrival) estimator which is necessary for the development of a real-time robot auditory system tracking moving acoustic source. It is well known that the use of conventional nonlinear filtering schemes may result in the severe performance degradation of DOA estimation and not be preferable for real-time implementation. These are mainly due to the inherent nonlinearity of the acoustic signal model used for DOA estimation. This motivates us to consider a new uncertain linear acoustic signal model based on the linear prediction relation of a noisy sinusoid. Using the suggested measurement model, it is shown that the resultant DOA estimation problem is cast into the NCRKF (Non-Conservative Robust Kalman Filtering) problem [12]. NCRKF-based DOA estimator provides reliable DOA estimates of a fast moving acoustic source in spite of using the noise-corrupted measurement matrix in the filter recursion and, as well, it is suitable for real-time implementation because of its linear recursive filter structure. The computational efficiency and DOA estimation performance of the proposed method are evaluated through the computer simulations.

• Journal of Korean Institute of Industrial Engineers
• /
• v.40 no.3
• /
• pp.283-290
• /
• 2014

In the flat panel display industry, to meet production target quantities and the deadline of production, the scheduler and dispatching systems are major production management systems which control the order of facility production and the distribution of WIP (Work In Process). Especially the delivery time is a key factor of the dispatching system for the time when a lot can be supplied to the facility. In this paper, we use survival analysis methods to identify main factors of the delivery time and to build the delivery time forecasting model. To select important explanatory variables, the cox proportional hazard model is used to. To make a prediction model, the accelerated failure time (AFT) model was used. Performance comparisons were conducted with two other models, which are the technical statistics model based on transfer history and the linear regression model using same explanatory variables with AFT model. As a result, the mean square error (MSE) criteria, the AFT model decreased by 33.8% compared to the statistics prediction model, decreased by 5.3% compared to the linear regression model. This survival analysis approach is applicable to implementing the delivery time estimator in display manufacturing. And it can contribute to improve the productivity and reliability of production management system.

• Wind and Structures
• /
• v.36 no.6
• /
• pp.355-366
• /
• 2023

Although machine learning (ML) techniques have been widely used in various fields of engineering practice, their applications in the field of wind engineering are still at the initial stage. In order to evaluate the feasibility of machine learning algorithms for prediction of wind loads on high-rise buildings, this study took the exposure category type, wind direction and the height of local wind force as the input features and adopted four different machine learning algorithms including k-nearest neighbor (KNN), support vector machine (SVM), gradient boosting regression tree (GBRT) and extreme gradient (XG) boosting to predict wind force coefficients of CAARC standard tall building model. All the hyper-parameters of four ML algorithms are optimized by tree-structured Parzen estimator (TPE). The result shows that mean drag force coefficients and RMS lift force coefficients can be well predicted by the GBRT algorithm model while the RMS drag force coefficients can be forecasted preferably by the XG boosting algorithm model. The proposed machine learning based algorithms for wind loads prediction can be an alternative of traditional wind tunnel tests and computational fluid dynamic simulations.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 1999.10a
• /
• pp.365-373
• /
• 1999

Recently, numerous studies have demonstrated that artificial intelligence such as neural networks can be an alternative methodology for classification problems to which traditional statistical methods have long been applied. In building neural network model, the selection of independent and dependent variables should be approached with great care and should be treated as a model construction process. Irrespective of the efficiency of a learning procedure in terms of convergence, generalization and stability, the ultimate performance of the estimator will depend on the relevance of the selected input variables and the quality of the data used. Approaches developed in statistical methods such as correlation analysis and stepwise selection method are often very useful. These methods, however, may not be the optimal ones for the development of neural network models. In this paper, we propose a genetic algorithms approach to find an optimal or near optimal input variables for neural network modeling. The proposed approach is demonstrated by applications to bankruptcy prediction modeling. Our experimental results show that this approach increases overall classification accuracy rate significantly.

• Transactions of the Korean Society of Machine Tool Engineers
• /
• v.17 no.6
• /
• pp.49-55
• /
• 2008

Arc welding process is one of the most important technologies to join metal plates. Robotic welding offers the reduced manufacturing cost sought, but its widespread use demands a means of sensing and correcting for inaccuracies in the part, the fixturing and the robot. A number of problems that need to be addressed in robotic arc welding processes include sensing, joint tracking, and lack of adequate models for process parameter prediction and quality control. Problems with parameter settings and quality control occur frequently in the GMA(Gas Metal Arc) welding process due to the large number of interactive process parameters that must be set and accurately controlled. The objectives of this paper are to realize the mapping characteristics of bead width using a sensitivity analysis and develop the neural network and multiple regression method, and finally select the most accurate model in order to control the weld quality(bead width) for fillet welding. The experimental results show that the proposed neural network estimator can predict bead width with reasonable accuracy, and guarantee the uniform weld quality.

• Journal of Korean Association for Spatial Structures
• /
• v.21 no.2
• /
• pp.99-110
• /
• 2021

This study presents the estimation of crack depth by analyzing temperatures extracted from thermal images and environmental parameters such as air temperature, air humidity, illumination. The statistics of all acquired features and the correlation coefficient among thermal images and environmental parameters are presented. The concrete crack depths were predicted by four different machine learning models: Multi-Layer Perceptron (MLP), Random Forest (RF), Gradient Boosting (GB), and AdaBoost (AB). The machine learning algorithms are validated by the coefficient of determination, accuracy, and Mean Absolute Percentage Error (MAPE). The AB model had a great performance among the four models due to the non-linearity of features and weak learner aggregation with weights on misclassified data. The maximum depth 11 of the base estimator in the AB model is efficient with high performance with 97.6% of accuracy and 0.07% of MAPE. Feature importances, permutation importance, and partial dependence are analyzed in the AB model. The results show that the marginal effect of air humidity, crack depth, and crack temperature in order is higher than that of the others.

• Journal of Institute of Control, Robotics and Systems
• /
• v.18 no.7
• /
• pp.680-687
• /
• 2012

One of the frequent problems in the stabilized gimbal system is the rejection of disturbances associated with moving components. Very often such disturbances have non-linear characteristics. In a typical gimbal system, each gimbal and platform are connected by a mutual bearing which induces inevitable friction. Particularly, the non-linear Coulomb friction causes position errors as well as slow responses that lead to unfavorable performance. In this paper, a modified PID controller that is augmented by Coulomb friction estimator is presented. Through constantly estimating the Coulomb friction torque, it is applied to the output of the existing PID controller. The effectiveness of the proposed controller is evaluated through a series of experiments.

• ETRI Journal
• /
• v.22 no.4
• /
• pp.40-50
• /
• 2000

We apply a "sliding-window" Maximum Likelihood(ML) estimator to estimate traffic parameters On-Off source and develop a method for estimating stochastic predicted individual cell arrival rates. Based on these results, we propose a simple Connection Admission Control(CAC)scheme for delay sensitive services in broadband onboard packet switching satellite systems. The algorithms are motivated by the limited onboard satellite buffer, the large propagation delay, and low computational capabilities inherent in satellite communication systems. We develop an algorithm using the predicted individual cell loss ratio instead of using steady state cell loss ratios. We demonstrate the CAC benefits of this approach over using steady state cell loss ratios as well as predicted total cell loss ratios. We also derive the predictive saturation probability and the predictive cell loss ratio and use them to control the total number of connections. Predictive congestion control mechanisms allow a satellite network to operate in the optimum region of low delay and high throughput. This is different from the traditional reactive congestion control mechanism that allows the network to recover from the congested state. Numerical and simulation results obtained suggest that the proposed predictive scheme is a promising approach for real time CAC.

• Proceedings of the Computational Structural Engineering Institute Conference
• /
• 1995.04a
• /
• pp.136-143
• /
• 1995

The problem of the structural identification becomes important, particularly with relation to the rapid increase of the number of the damaged or deteriorated structures, such as highway bridges, buildings, and industrial facilities. This paper summarizes the recent studies related to those problems by the present authors. The system identfication methods are generally classified as the time domain and the frequency domain methods. As time doamin methods, the sequential algorithms such as the extended Kalman filter and the sequential prediction error method are studied. Several techniques for improving the convergences are incorporated. As frequency domain methods, a new frequency response function estimator is introduced. For damage estimation of existing structures, the modal perturbation and the sensitivity matrix methods are studied. From the example analysis, it has been found that the combined utilization of the measurement data for the static response and the dynamic (modal) properties are very effictive for the damage estimation.

• The Korean Journal of Applied Statistics
• /
• v.23 no.3
• /
• pp.609-619
• /
• 2010

Geostatistical data among spatial data is analyzed in three stages: (1) variogram estimation, (2) model fitting for the estimated variograms and (3) spatial prediction using the fitted variogram model. It is very important to estimate the variograms properly as the first stage(i.e., variogram estimation) affects the next two stages. In general, the variogram is estimated with the moment estimator. To estimate the variogram, we have to decide the 'lag increment' or the 'number of lags'. However, there is no established rule for selecting the number of lags in estimating the variogram. The present paper proposes a method of choosing the optimal number of lags based on the PRESS statistic. To show the usefulness of the proposed method, we perform a small simulation study and show an empirical example with with air pollution data from Korea.