• 디지털융복합연구
• /
• 제19권10호
• /
• pp.253-263
• /
• 2021

기계학습은 학습에 이용되는 학습 데이터와 데이터를 예측할 인공신경망을 이용하여 비용함수를 만들고, 비용함수를 최소화시키는 파라미터들을 찾는 과정이다. 파라미터들은 비용함수의 그래디언트 기반 방법들을 이용하여 변화하게 된다. 디지털 신호가 복잡할수록, 학습하고자 하는 문제가 복잡할수록, 인공신경망의 구조는 더욱 복잡해지고 깊어진다. 복잡하고, 깊어지는 인공신경망 구조는 과적합(Over-fitting) 문제를 발생시킨다. 과적합 문제를 해결하기 위하여 파라미터의 가중치 감소 정규화 방법이 사용되고 있다. 우리는 이러한 방법에서 추가로 비용함수의 값을 이용한다. 이러한 방법으로 기계학습의 정확도가 향상되는 결과를 얻었으며 이는 수치 실험을 통하여 우수성이 확인된다. 이러한 결과는 기계학습을 통한 인공지능의 폭넓은 데이터에 대한 정확한 값을 도출한다.

• 한국소음진동공학회논문집
• /
• 제13권5호
• /
• pp.334-340
• /
• 2003

Exponential window function Is widely used In impact hammer testing to reduce leakage error as well as to get a good S/N ratio. The larger its decaying rate is, the more effectively the leakage errors are reduced. But if the decay rate of the exponential window is too large, the FRF is distorted. And the modal parameters of the system can not be exactly identified by modal analysis technique. Therefore, it is a difficult problem to determine proper decay rate in impact hammer testing. In this paper, amount of the FRF distortion caused by exponential window is theoretically uncovered. A new circle fitting method is also proposed so that the modal parameters are directly extracted from impulse response spectrum distorted by the exponential-windowed impulse response data. The results by the conventional and proposed circle fitting method are compared through a numerical example.

• 산업경영시스템학회지
• /
• 제27권2호
• /
• pp.37-43
• /
• 2004

High speed machining is a machining process which cuts materials with the fast movement and rotation of a spindle in a machine tool. It reduces machining time because of the high feed and the high speed of a spindle. In addition it gets rid of post processes for high precision machining. When the high speed machining is applied to especially hardened steel, operators should select the proper parameters of machining. This can produce machining surfaces which is qualified with good surface roughness. This paper presents a method for selecting machining parameters to minimize surface roughness with high speed machining in cutting the hardened steels. Experimental data for surface roughness are collected in a machining shop based on the cutting feed and the spindle rotation. The data fits in hi-cubic polynomial surface of mathematical form. From the model this research minimize the surface roughness to find the optimal values of the feed and the spindle speed. This paper presents a program which automatically generates optimal solutions from the raw data of experiments.

• Geomechanics and Engineering
• /
• 제20권5호
• /
• pp.411-420
• /
• 2020

Numbers fitting-curve equations have been proposed to predict soil-water retention curve (SWRC) whose parameters have no definitude physical meaning. And these methods with precondition of measuring SWRC data is time-consuming. A simplified directly method to estimate SWRC without parameters obtained by fitting-curve is proposed. Firstly, the total SWRC can be discretized into linear segments respectively. Every segment can be represented by linear formulation and every turning point can be determined by the pore-size distribution (PSD) of Mercury Intrusion Porosimetry (MIP) tests. The pore diameters governing the air-entry condition (AEC) and residual condition (RC) can be determined by the PSDs of MIP test. The PSD changes significantly during drying in SWR test, so the determination of AEC and RC should use the PSD under corresponding suction conditions. Every parameter in proposed equations can be determined directly by PSD without curve-fitting procedure and has definitude physical meaning. The proposed equations give a good estimation of both unimodal and bimodal SWRCs.

• 한국음향학회지
• /
• 제19권3호
• /
• pp.86-91
• /
• 2000

본 논문에서는 피아노 음을 FFT(Fast Fourier Transform)를 이용하여 주파수 영역으로 변환한 후, 크기(magnitude)와 위상(phase)에 대한 특성들을 분석한다. 이 분석한 결과로 크기와 위상을 모델링하여 파라미터를 생성한다. 크기의 특성에서 기본주파수와 고조파 부분은 다른 부분에 비해 크기가 매우 크다. 그래서 이 부분의 모델링은 오차를 줄이기 위해 곡선의 적합성(Curve Fitting) 방법을 이용하였고, 노이즈 부분의 모델링은 기본주파수 전후의 노이즈가 피아노 음색 특성에 중요한 역할을 하고 불규칙하므로 선분 근사법을 이용하였으며, 크기가 작고 완만하게 변하는 나머지 노이즈 부분은 곡선의 적합성을 이용하였다. 위상은 크기에 적응한 방법과 동일하게 적용하였다. 생성된 파라미터로 크기와 위상을 만들어 역변환 FFT를 하면 피아노 원음에 매우 근접한 음을 얻을 수 있다.

• 한국가시화정보학회지
• /
• 제19권1호
• /
• pp.28-35
• /
• 2021

In this study, an image processing method for the measurement of two-phase bubbly flow is developed. Shadowgraphy images obtained by high-speed camera are used for analysis. Some bubbles are generated as single unit and others are overlapped or clustered. Single bubbles can be easily analyzed using parameters such as bubble shape, centroid, and area. But overlapped bubbles are difficult to transform clustered bubbles into segmented bubbles. Several approaches were proposed for the bubble segmentation such as Hough transform, connection point method and watershed. These methods are not enough for bubble segmentation. In order to obtain the size distribution of bubbles, we present a method of splitting overlapping bubbles using watershed and approximating them to ellipse. There is only 5% error difference between manual and automatic analysis. Furthermore, the error can be reduced down to 1.2% when a correction factor is used. The ellipse fitting algorithm developed in this study can be used to measure bubble parameters accurately by reflecting the shape of the bubbles.

• Journal of Astronomy and Space Sciences
• /
• 제7권2호
• /
• pp.73-103
• /
• 1990

Morphological luminosity parameters$(\mu_e,\;r_e,\;\mu_0,\;\alpha^{-1})$ and D/B were estimated from the decomposition of surface brightness distributions of 28 extra galaxies. Decomposition was made using the standard non-linear least square fitting method and we used the seeing convolved model to get the central brightness of these galaxies. Masses and $M/L_B$ were calculated using rotational velocities of these galaxies from the fitting to the generalized Toomre's mass model.

• Journal of electromagnetic engineering and science
• /
• 제6권3호
• /
• pp.155-159
• /
• 2006

An efficient method is presented to design filters without the similarity transform of their coupling coefficient matrix as circuit parameters, which is very tedious due to pivoting and deciding rotation angles needed during the iterations. The transfer function of a filter is directly used for the design and its desired form is derived by the optimized rational-function fitting technique. A 3rd order coaxial lowpass filter is taken as an example to validate the proposed method.

• 대한전자공학회논문지
• /
• 제24권1호
• /
• pp.67-72
• /
• 1987

InSb TFT is fabricated by the vacuum evaporation method and I-V characteristics are measured. Employing Davidon Fletcher-Powell algorithm, the device parameters are extracted. The current-voltage relations calculated by extracdted parameters are in good agreement with experimental results. It is found that optimization technique may be more simple and accurate than curve fitting method in device parameters extration.

• Smart Structures and Systems
• /
• 제20권4호
• /
• pp.461-472
• /
• 2017

Explicit design formulae of liquid column vibration absorber (LCVA) for suppressing harmonic vibration of structures with small inherent structural damping are developed in this study. The developed design formulae are also applicable to the design of a tuned mass damper (TMD) and a tuned liquid column damper (TLCD) for damped structures under harmonic force excitation. The optimum parameters of LCVA for suppressing harmonic vibration of undamped structures are first derived. Numerical searching of the optimum parameters of tuned vibration absorber system for suppressing harmonic vibration of damped structure is conducted. Explicit formulae for these optimum parameters are then obtained by a series of curve fitting techniques. The analytical result shows that the control performance of TLCD for reducing harmonic vibration of undamped structure is always better than that of non-uniform LCVA for same mass and length ratios. As for the effects of structural damping on the optimum parameters, it is found that the optimum tuning ratio decreases and the optimum damping ratio increases as the structural damping is increased. Furthermore, the optimum head loss coefficient is inversely proportional to the amplitude of excitation force and increases as the structural damping is increased. Numerical verification of the developed explicit design expressions is also conducted and the developed expressions are demonstrated to be reasonably accurate for design purposes.