• Journal of the Korea Society for Simulation
• /
• v.30 no.1
• /
• pp.75-85
• /
• 2021

In this paper, we present a deep neural network-based prediction model that processes and analyzes the corporate credit and personal credit information of individual business owners as a new method to predict the default rate of individual business more accurately. In modeling research in various fields, feature selection techniques have been actively studied as a method for improving performance, especially in predictive models including many features. In this paper, after statistical verification of macroeconomic indicators (macro variables) and credit information (micro variables), which are input variables used in the default rate prediction model, additionally, through the credit information feature selection method, the final feature set that improves prediction performance was identified. The proposed credit information feature selection method as an iterative & hybrid method that combines the filter-based and wrapper-based method builds submodels, constructs subsets by extracting important variables of the maximum performance submodels, and determines the final feature set through prediction performance analysis of the subset and the subset combined set.

• Journal of the Korea Society for Simulation
• /
• v.18 no.4
• /
• pp.219-225
• /
• 2009

The primary purpose of this study is to build computer models referring overall flow of complex and various semiconductor wafer manufacturing process and to implement a educational model which operates with a presentation tool showing device design. It is important that Korean semiconductor industries secure high competitive power on efficient manufacturing management and to develop technology continuously. Models representing the FAB processes and the functions of each process are developed for Seoul National University Semiconductor Research Center. However, it is expected that the models are effective as visually educational tools in Korean semiconductor industries. In addition, it is anticipated that these models are useful for semiconductor process courses in academia. Scalability and flexibility allow semiconductor manufacturers to customize the models and perform simulation education. Subsequently, manufacturers save budget.

• Journal of the Society of Naval Architects of Korea
• /
• v.39 no.3
• /
• pp.75-80
• /
• 2002

In these days of severe struggle for existence, the world has changed a great deal to global and digital oriented period. The enterprises try to introduce new management and production system to adapt such a change. But, if the only new technologies are applied to an enterprise without definite analysis about manufacturing, failure fellows as a logical consequence. Hence, enterprise must analyze manufacturing system definitely and needs new methodologies to mitigate risk. This study suggests that the new approach, which is systems approach for process improvement, is organized to systems analysis, systems diagnosis, and systems verification. Systems analysis analyzes manufacturing systems with object-oriented methodology-UML(Unified Modeling language) from a point of product, process, and resource view. Systems diagnosis identifies the constraints to optimize the system through scientific management or TOC(Theory of constraints). Systems verification shows the solution with virtual manufacturing technique applied to the core problem which emerged from systems diagnosis. This research shows the artifacts to improve the productivity with the above methodology applied to forming shop. UML provides the definite tool for analysis and re-usability to adapt itself to environment easily. The logical tree of TOC represents logical tool to optimize the forming shop. Discrete event simulator-QUEST suggests the tool for making a decision to verify the optimized forming shop.

• Proceedings of the KIEE Conference
• /
• 2001.07d
• /
• pp.2274-2276
• /
• 2001

본 논문에서는 미세 구조물 가공 기술(MEMS)로 제작된 실리콘 자이로스코프를 위한 폐루프 제어기를 주파수 가중 $H_{\infty}$ 제어 기법과 주파수 가중 모델 축소 기법을 이용하여 설계하였다. 실리콘 자이로스코프는 수 kHz 대의 공진 특성을 이용하여 각속도를 측정하는 센서로서, 공진 주파수 영역의 특성이 매우 중요하므로 주파수 영역을 고려하고 공정 오차를 감안한 주파수 가중 $H_{\infty}$ 제어기가 필요하다. 본 논문에서는 고차 강인 제어기의 회로적 구현과 ASIC화가 가능하도록 하기 위해, 주파수 가중 모델 축소 기법을 이용하여 공진 주파수 영역에서 성능을 유지하는 저차 제어기를 설계하였으며, 시뮬레이션을 통해 기존의 제어기 및 고차 제어기와 성능을 비교하였다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.101-104
• /
• 2023

본 연구는 현대 사회에서 빅데이터의 중요성이 강조되는 가운데, 온라인 시장의 확장과 소비자들의 다양한 소비 행태 변화를 반영한 가격지표 개발을 목표로 한다. 통계청의 기존 통계조사 방법론에 대한 한계를 극복하고, 온라인 쇼핑몰 데이터에서 필요한 정보를 추출하고 가공하기 위해 대규모 언어 모델(LLM)을 활용한 인공지능 기술을 적용해보고자 한다. 초기 연구 결과로 공개 Polyglot을 활용하여 비정형 자료 처리와 품목분류에 응용해 보았으며, 제한된 학습 데이터를 사용하여도 높은 정확도의 처리 결과를 얻을 수 있었으며, 현재는 적용 품목을 확장하여 더욱 다양한 품목에 방법론을 적용하는 연구를 진행 중이다.

• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 2017.04a
• /
• pp.98-98
• /
• 2017

대부분의 가정과 요식업체, 식품가공업계에서 이용하고 있는 식용유지는 저장 및 가공과정 중에 산패가 빈번하게 일어나게 된다. 기존에는 유지 산패를 측정하기 위해 산가, 과산화물가 등을 측정하는 이화학적인 적정방법을 이용하였는데 실험자의 숙련도에 따라 결과의 오차가 발생할 수 있고, 반복실험으로 인한 시간과 비용이 많이 소모되는 등 여러 제약사항을 포함하고 있어 식용유지의 산패를 실시간 비파괴적으로 분석할 수 있는 기술의 개발에 많은 관심이 모아지고 있다. 따라서, 본 연구에서는 식용유지의 저장조건에 따른 산패정도를 비파괴적으로 평가하기 위한 근적외선 분광분석과 인공신경망 분석기술을 개발하여 그 실효성을 평가하였다. 식물성 식용유지인 들기름을 특정 온도에서 일정한 시간동안 저장하면서 이화학적 적정방법을 통해 산가와 과산화물가를 측정하였으며 동일한 시료의 근적외선 투과스펙트럼을 획득하였다. 수집된 정보를 이용하여 유지 산패 예측 모델을 개발하기 위해 다변량 분석기법 (주성분 회귀분석, 최소자승 회귀분석과 인공신경망 분석)을 적용하였다. 분석 결과, 인공신경망 분석모델이 산가 ($R^2_{tra}:0.9037$, $R^2_{val}:0.8175$, $R^2_{test}:0.8555$)와 과산화물가 ($R^2_{tra}:0.9210$, $R^2_{val}:0.9341$, $R^2_{test}:0.8286$)의 예측 성능이 가장 우수한 것으로 확인되었다. 본 연구의 결과들은 농산물과 식품의 성분 측정뿐만 아니라 다른 산업분야에서도 유용하게 활용될 수 있을 것으로 기대되어진다.

• Journal of the Society of Naval Architects of Korea
• /
• v.39 no.2
• /
• pp.34-44
• /
• 2002

Approximately 70 percent of shop's hull plate consists of three-dimensional curved shell. Concerning with the research on the automation of plate forming many studies have been carried out for the last decade. The purpose of this study is to develop the simulator of heating on the basis of the reasonable mechanical model representing a heating phenomenon. The beating experiment has been carried out with varying parameters influencing on the results of heating information at the kinematics analysis, simulatorestimate the shape of deformed plate that process along the processing information. When we get the initial shape and the object shape, we calculate the processing information first, using kinematics analysis. In a simulator we estimate deformed shape from the processing information. After this we compare deformed shape and object shape. If the error of deformed shape and object shape is in the proper limits, that information is determined the final processing information. Else we repeat the process changing variable.

• Proceedings of the Korean Society for Technology of Plasticity Conference
• /
• 2006.06a
• /
• pp.55-60
• /
• 2006

Draw-bead is applied to control the material flow in a stamping process and improve the product quality by controlling the draw-bead restraining force (DBRF). Actual die design depends mostly on the trial-and-error method without calculating the optimum DBRF. Die design with the predicted value of DBRF can be utilized at the tryout stage effectively reducing the cost of the product development. For the prediction of DBRF, a simulation-based prediction model of the circular draw-bead is developed using the Box-Behnken design with selected shape parameters such as the bead height, the shoulder radius and the sheet thickness. The value of DBRF obtained from each design case by analysis is approximated by a second order regression equation. This equation can be utilized to the calculation of the restraining force and the determination of the draw-bead shape as a prediction model. For the evaluation of the prediction model, the optimum design of DBRF in sheet metal forming is carried out using response surface methodology. The suitable type of the draw-bead is suggested based on the optimum values of DBRF. The prediction model of the circular draw-bead proposes the design method of the draw-bead shape. The present procedure provides a guideline in the tool design stage for sheet metal forming to reduce the cost of the product development.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.17 no.4
• /
• pp.39-46
• /
• 2018

Two surfaces that have relative motion show different characteristics according to surface roughness or surface patterns in all lubrication areas. For two rough surfaces with mixed lubrication, this paper proposes a new approach that includes the contact characteristics of the surfaces and a probabilistic method for a numerical analysis of lubrication. As the contact area of the two surfaces changes according to the loading conditions, asperity contact is very important. An average flow model developed by Patir-Cheng is central to the study of lubrication for rough surfaces. This average flow model also refers to a multi-asperity contact model for deriving a modified Reynolds equation and calculating the lubricant characteristics of a bearing surface with random roughness during fluid flow. Based on the average flow model, this paper carried out a numerical analysis of lubrication using a contact model by considering a load change made by the actual contact of asperities between two surfaces. Lubrication properties show different characteristics according to the surface patterns. This study modeled various geometric surface patterns and calculated the characteristics of lubrication.

• Journal of IKEEE
• /
• v.23 no.1
• /
• pp.127-133
• /
• 2019

Electric power demand forecasting is one of the important research areas for future smart grid introduction. However, It is difficult to predict because it is affected by many external factors. Traditional methods of forecasting power demand have been limited in making accurate prediction because they use raw power data. In this paper, a probability-based CRBM is proposed to solve the problem of electric power demand prediction using raw power data. The stochastic model is suitable to capture the probabilistic characteristics of electric power data. In order to compare the mid-term power demand forecasting performance of the proposed model, we compared the performance with Recurrent Neural Network(RNN). Performance comparison using electric power data provided by the University of Massachusetts showed that the proposed algorithm results in better performance in mid-term energy demand forecasting.