• 전자공학회논문지CI
• /
• 제44권3호
• /
• pp.98-107
• /
• 2007

얼굴인식 등과 같은 고차원 패턴인식에서 학습패턴의 수가 패턴 차원에 비해 매우 적을 경우 희소성 문제(the Small Sample Size problem)가 발생한다. 최근 이 문제를 해결하기 위하여 LDA, PCA+LDA, Direct-LDA 등을 비롯한 다양한 LDA-확장 법이 제안되었다. 본 논문에서는 LDA-확장 법으로 차원을 축소하기 전에 학습 패턴을 사전 클러스터링하여 서브 클래스 수를 증가시키는 방법으로 LDA-확장에 기반을 둔 식별기의 성능을 향상시키는 방법을 제안한다. LDA (또는 Direct-LDA)에서 축소된 특징공간의 차원은 학습패턴의 클래스 수로 제한되기 때문에 LDA의 식별 성능을 향상시킬 수 있도록 학습패턴을 사전에 클러스터링하여 서브 클래스의 수를 증가시키는 방법이다. 즉, 학습패턴의 특성공간(the eigen space)은 레인지 공간(the range space)과 널 공간(the null space)으로 구성되며, 레인지 공간의 차원은 클래스 수의 증가에 따라 증가한다. 따라서 변환 행렬을 구성할 때 클래스의 수를 늘려 널 공간을 최소화하게 되면 이 공간에 기인한 정보의 손실을 최소화 할 수 있다. 제안 방법을 X-OR 형태의 인공데이터와 AT&T와 Yale 벤취마크 얼굴영상 데이터베이스를 대상으로 실험한 결과 본 방법의 효용성을 확인하였다.

• 한국연소학회지
• /
• 제2권1호
• /
• pp.39-51
• /
• 1997

A review of probability density function(PDF) methodology and direct numerical simulation for the purpose of modeling turbulent combustion are presented in this study where particular attention is focused on the modeling problem of turbulent molecular mixing term appearing in PDF transport equation. Existing mixing models results were compared to those evaluated by direct numerical simulation in a turbulent premixed medium with finite rate chemistry in which the initial scalar field is composed of pockets of partially burnt gases to simulate autoignition. Two traditional mixing models, the least mean square estimations(LMSE) and Curl#s model are examined to see their prediction capability as well as their modeling approach. Test calculations report that the stochastically based Curl#s approach, though qualitatively demonstrates some unphysical behaviors, predicts scalar evolutions which are found to be in good agreement with statistical data of direct numerical simulation.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1997년도 추계학술대회 논문집 학회본부
• /
• pp.69-71
• /
• 1997

The direct drive motor is directly coupled by load. So, it is directly affected by load and disturbances. To control the direct drive motor, a robust controller is need. The main feature of variable structure system is that system trajectories are robust and insensitive to parameter variations and disturbances in the sliding mode. In this paper, adaptive variable structure controller, is used for the BLDD SM(Brushless Direct Drive Servo Motor) control. The chattering problem is reduced by using the saturation function.

• Journal of Power Electronics
• /
• 제18권3호
• /
• pp.736-745
• /
• 2018

This paper presents an extended state observer (ESO) based direct torque control (DTC) for use in induction motor systems to handle the issues of time delays, load torque disturbances and parameter uncertainties. Direct torque control offers an excellent torque response and it does not require a proportion integration (PI) controller in the current loop. However, a PI controller is still adopted in the outer speed loop to generate the torque reference value, which is a slow method. An ESO based compound control scheme is proposed to improve the response rate and accuracy of the torque reference signal, especially when load torque is injected. In addition, the time delay problem is analyzed and compensated for in this paper to reduce torque ripples. The proposed disturbance compensation technique based direct control scheme is shown to have good performance both in the transient and stable states via simulations and experimental results.

• 한국통신학회논문지
• /
• 제31권10C호
• /
• pp.942-947
• /
• 2006

In this paper, a new feature representation technique called Improved 2-directional 2-dimensional direct linear discriminant analysis (Improved $(2D)^2$ DLDA) is proposed. In the case of face recognition, thesmall sample size problem and need for many coefficients are often encountered. In order to solve these problems, the proposed method uses the direct LDA and 2-directional image scatter matrix. Moreover the selection method of feature vector and the method of similarity measure are proposed. The ORL face database is used to evaluate the performance of the proposed method. The experimental results show that the proposed method obtains better recognition rate and requires lesser memory than the direct LDA.

• International Journal of Aeronautical and Space Sciences
• /
• 제9권2호
• /
• pp.79-86
• /
• 2008

High performance direct-iterative hybrid linear solver for large scale finite element problem is developed. Direct solution method is robust but difficult to parallelize, whereas iterative solution method is opposite for direct method. Therefore, combining two solution methods is desired to get both high performance parallel efficiency and numerical robustness for large scale structural analysis problems. Hybrid method mentioned in this paper is based on FETI-DP (Finite Element Tearing and Interconnecting-Dual Primal method) which has good parallel scalability and efficiency. It is suitable for fourth and second order finite element elliptic problems including structural analysis problems. We are using the hybrid concept of theses two solution method categories, combining the multifrontal solver into FETI-DP based iterative solver. Hybrid solver is implemented for our general structural analysis code, IPSAP.

• 한국정밀공학회지
• /
• 제12권2호
• /
• pp.97-111
• /
• 1995

In this paper, we investigate path compensation scheme for the machining errors due to tool deflection in 2D contour machining. The significance of the deflection error is first shown by experiments, and a direct compensation scheme is sought. In the presented scheme, the tool path is evaluated and correcte based on the instantaneous deflection force model, until the desired contour can be obtained under the presence of tool deflection in actual machining. In the sense that the developed method estimates and compensates the machining errors via modifying the tool path, it is distinguished from the previous approach based on geometric simulation and cutting simulation. Further, it can be viewed as a direct and active method toward direct shape control in CNC machining. Simulation results are included to show the validity and adequacy of the path-modification scheme under various cutting conditions.

• 대한기계학회:학술대회논문집
• /
• 대한기계학회 2001년도 춘계학술대회논문집C
• /
• pp.398-401
• /
• 2001

There are two methods to calculate design sensitivity such as direct differentiation method and adjoint method. A sort of direct differentiation method for design sensitivity analysis costs too much when number of design variables is much larger than the number of response functions whose design sensitivity analyses are required. Therefore, an adjoint method is suggested for the case that the dimension of design variables is lager than the number of response function. An adjoint method is required to compute adjoint variables from the simultaneous linear system equation, the so-called adjoint equation, requiring only the eigenvalue and its associated eigenvectors for mode being differentiated. This method has been extended to the repeated eigenvalue problem. In this paper, we propose an adjoint method for deign sensitivity analysis of damped vibratory systems with distinct eigenvalues.

• 한국정보시스템학회지:정보시스템연구
• /
• 제18권3호
• /
• pp.375-399
• /
• 2009

In this study, we propose a integrated model of logistic regression, artificial neural networks, support vector machines(SVM), with case-based reasoning(CBR). To predict respondents in the direct marketing is the binary classification problem as like bankruptcy prediction, IDS, churn management and so on. To solve the binary problems, we employed logistic regression, artificial neural networks, SVM. and CBR. CBR is a problem-solving technique and shows significant promise for improving the effectiveness of complex and unstructured decision making, and we can obtain excellent results through CBR in this study. Experimental results show that the classification accuracy of integration model using CBR is superior to logistic regression, artificial neural networks and SVM. When we apply the customer response model to predict respondents in the direct marketing, we have to consider from the view point of profit/cost about the misclassification.

• 한국소음진동공학회:학술대회논문집
• /
• 한국소음진동공학회 2003년도 추계학술대회논문집
• /
• pp.778-783
• /
• 2003

The problem considered in this paper is about the collocation of sensor and actuator for the active control of sound and vibration. It is well-known that a point collocated sensor-actuator pair offers an unconditional stability with very high performance when it is used with a direct velocity feedback (DVFB) control, because the pair has strictly positive real (SPR) property. In order to utilize this SPR characteristics, a matched piezoelectric sensor and actuator pair is considered, but this pair suffers from the in-plane motion coupling problem with the out-of$.$plane motion due to the piezo sensor and actuator interaction. This coupling phnomenon limits the stability and performance of the matched pair with DVFB control. As a new alternative, a point sensor and piezoelectric actuator pair is also considered, which provides SPR property in all frequency range except at the first resonance in very low frequency. This non-SPR resonance could be minimized by applying a phase lag compensator.