• Journal of Power System Engineering
• /
• v.17 no.3
• /
• pp.72-81
• /
• 2013

Many articles have been published about a 2-degree of freedom model that includes the lateral and yaw motions for controller synthesis in intelligent transport system applications. In this paper, a 3-degree of freedom linear model that includes the roll motion is developed to design a robust steering controller for lane following maneuvers using ${\mu}$-synthesis. This linear perturbed system includes a set of parametric uncertainties in cornering stiffness and unmodelled dynamics in steering actuators. The state-space model with parametric uncertainties is represented in linear fractional transformation form. Design purpose can be obtained by properly choosing the frequency dependent weighting functions. The objective of this study is to keep the tracking error and steering input energy small in the presence of variations of the cornering stiffness coefficients. Furthermore, good ride quality has to be achieved against these uncertainties. Frequency-domain analyses and time-domain numerical simulations are carried out in order to evaluate these performance specifications of a given vehicle system. Finally, the simulation results indicate that the proposed robust controller achieves good performance over a wide range of uncertainty for the given maneuvers.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.1
• /
• pp.121-128
• /
• 2017

In this paper, we propose a dual-channel acoustic event detection (AED) method using nonnegative tensor factorization (NTF) and hidden Markov model (HMM) in order to improve detection accuracy of AED in multisource environments. The proposed method first detects multiple acoustic events by utilizing channel gains obtained from the NTF technique applied to dual-channel input signals. After that, an HMM-based likelihood ratio test is carried out to verify the detected events by using channel gains. The detection accuracy of the proposed method is measured by F-measures under 9 different multisource conditions. Then, it is also compared with those of conventional AED methods such as Gaussian mixture model and nonnegative matrix factorization. It is shown from the experiments that the proposed method outperforms the convectional methods under all the multisource conditions.

• Korean Journal of Clinical Pharmacy
• /
• v.21 no.2
• /
• pp.145-155
• /
• 2011

The objective of this study was to determine whether any pretreatment parameters were associated with pharmacological effect or toxicity parameters after vincristine administration and to describe a mathematical model, which explains the interpatient pharmacodynamic variability. The relationship between patient characteristics and vincristine dose and hematological toxicity were evaluated. 68 pediatric and adolescence patients and 107 adults with acute lymphoblastic leukemia were treated with vincristine $1.5mg/m^2/day$ IV and other anticancer drugs as scheduled. Complete blood counts and other blood test results were obtained. The input variables were age, gender, weight, lean body weight (LBW), height, body surface area, vincristine dose and total vincristine dose. The outcome measures were nadir values (white blood cells, absolute neutrophil counts, hemoglobin, and platelets); the absolute decrease, relative decrease, and survival fraction of blood cells. Polynomial regression analysis was carried out to determine the other significant covariates. The variability of $WBC_{nadir}$ was modeled with good precision and accuracy with a two-covariate model. This model should be validated and improved on with further clinical data. We believe that such pharmacodynamic modeling should be explored further to determine its performance and clinical relevance compared with modeling using pharmacokinetic parameter.

• Journal of KIISE
• /
• v.44 no.8
• /
• pp.774-781
• /
• 2017

Mention detection systems use nouns or noun phrases as a head and construct a chunk of text that defines any meaning, including a modifier. The term "mention detection" relates to the extraction of mentions in a document. In the mentions, a coreference resolution pertains to finding out if various mentions have the same meaning to each other. A pointer network is a model based on a recurrent neural network (RNN) encoder-decoder, and outputs a list of elements that correspond to input sequence. In this paper, we propose the use of mention detection using pointer networks. Our proposed model can solve the problem of overlapped mention detection, an issue that could not be solved by sequence labeling when applying the pointer network to the mention detection. As a result of this experiment, performance of the proposed mention detection model showed an F1 of 80.07%, a 7.65%p higher than rule-based mention detection; a co-reference resolution performance using this mention detection model showed a CoNLL F1 of 52.67% (mention boundary), and a CoNLL F1 of 60.11% (head boundary) that is high, 7.68%p, or 1.5%p more than coreference resolution using rule-based mention detection.

• Journal of Korea Multimedia Society
• /
• v.14 no.11
• /
• pp.1373-1382
• /
• 2011

In this study, we propose an automated smile analysis system for self smile training. The proposed system detects the face area from the input image with the AdaBoost algorithm, followed by identifying facial features based on the face shape model generated by using an ASM(active shpae model). Once facial features are identified, the lip line and teeth area necessary for smile analysis are detected. It is necessary to judge the relationship between the lip line and teeth for smiling degree analysis, and to this end, the second differentiation of the teeth image is carried out, and then individual the teeth areas are identified by means of histogram projection on the vertical axis and horizontal axis. An analysis of the lip line and individual the teeth areas allows for an automated analysis of smiling degree of users, enabling users to check their smiling degree on a real time basis. The developed system in this study exhibited an error of 8.6% or below, compared to previous smile analysis results released by dental clinics for smile training, and it is expected to be used directly by users for smile training.

• Journal of Korean Society of Rural Planning
• /
• v.7 no.1 s.13
• /
• pp.61-76
• /
• 2001

During the performance of large scale tidal land reclamation project along the coast line, the construction of large scale structures such as sea-dikes, closing estuaries will induce big changes of near-shore hydraulic behavior. In this paper, its is aimed to verify the change of tide and currents after the construction of sea-dike of the Saemangeum project along the coast line. Numerical scale model "TRISULA" which development by Delft Hydraulics in the Netherlands was used. "TRISULA" is adopting the finite difference numerical scheme, and mostly using for hydro-dynamic solution along the sea and estuaries. Model boundary is covering $100{\times}170$ Km and constructed with $133{\times}337$ grids. Outer side boundary is divided 48 sections, and input 37 tidal components are gained from another big scale numerical "Yellow Sea" model. Model calibrations & verifications were performed th field tide & current datas which were measured along sea-dike alignment during Aug. $1997{\sim}Apr$. 1999. And then, numerical simulation with the tide condition dated 17 Apr. 1999 was performed with & without sea-dike construction condition for the comparison. Evaluated boundary is 20 km out-side from sea-dike alignment. Four cross lanes were set up, each of lane contains 3 points, for the comparison of sea-dike construction effects. Results showing the tidal amplitude is reducing approximately 20 cm after the construction of sea-dike during spring tide condition, amplitude 6.9m. Currents after construction of sea-dikes along the alignment, the northern part shows 50%(inner), 90%(outer) and the southern part shows 10%(inner) 50%(outer) of the currents before construction.

• Journal of Korean Port Research
• /
• v.13 no.2
• /
• pp.399-408
• /
• 1999

Copeland’s(1985) hyperbolic mild-slope equation including diffraction refraction and reflection in the wave field is used as a governing equation in this study. The result of Maruyama & Kajima(1985) is used to calculate wave direction and that of Watanabe & Maruyama(1986) is used as a energy dissipation formula. Numerical solutions are obtained by the Leap-Frog scheme and compared with Watanabe & Maruyama’s (1984) hydraulic experimental results and numerical simulation results for the detached breakwater. This wave model is applied to a detached breakwater and compared with Watanabe and Maruyama’s (1984) hydraulic model results to check the characteristics of reflected wave field around a detached breakwater. The distribution of wave height and we phase in front of a detached breakwater is more accurate than the Watanabe and Maruyama’s numerical results. The results from our wave model show good agreements with the others and also show nonlinear effects around the detached breakwater. This model is applied to the Gamcheon harbor of pusan. the field observations were carried out at Pusan harbor wave station in 1986-1995 and the results were accepted as a design wave condition in this study. The wave height and wave period was measured by Dong-A university at one station in the Gamcheon harbor in 1996-1997 and used as a calibration criterion. The measured data were used as input data for the numerical simulation and also compared with simulated results. The numerical simulation shows a fairly good results which considering the effect of topographic characteristics and effect of narrow entrance due to two separated breakwaters in Gamcheon harbor. The wave distribution characteristics inside Gamcheon harbor is quite different with the offshore wave direction and wave period.

• Journal of Ocean Engineering and Technology
• /
• v.23 no.4
• /
• pp.69-77
• /
• 2009

Structural analysis of a brake shoe for commercial vehicle was performed using finite element method. Since the strength of a brake shoe is affected by the magnitude and distribution shape of the contact pressure with the drum, the contact pressure between the shoe friction material and drum was calculated using a 2-Dimensional non-linear contact analysis in a state. And the brake was actuated by input air pressure and the drum of it was calculated both stationary and dynamic based on forced torque applied to the drum during the static state analysis. The results of the above analysis were then used as the load boundary conditions for a 3-Dimensional shoe model analysis to determine the maximum strain on the shoes. In the analysis model, the values of tensile test were used for the material properties of the brake shoes and drum, while the values of compression test were used for the friction material. We assumed it as linear variation, even though the properties of friction material were actually non-linear. The experiments were carried out under the same analysis conditions used for fatigue test and under the same brake system which equipped with a brake drum based on the actual axle state in a vehicle. The strains were measured at the same locations where the analysis was performed on the shoes. The obtained results of the experiment matched well with those from the analysis. Consequently, the model used in this study was able to determine the stress at the maximum air pressure at the braking system, thereby a modified shoe model in facilitating was satisfied with the required endurance strength in the vehicle.

• Smart Media Journal
• /
• v.7 no.4
• /
• pp.17-23
• /
• 2018

We proposed a LSTM-based RNN model that can effectively perform the automatic spacing characteristics. For those long or noisy sentences which are known to be difficult to handle within Neural Network Learning, we defined a proper input data format and decoding data format, and added dropout, bidirectional multi-layer LSTM, layer normalization, and attention mechanism to improve the performance. Despite of the fact that Sejong corpus contains some spacing errors, a noise-robust learning model developed in this study with no overfitting through a dropout method helped training and returned meaningful results of Korean word spacing and its patterns. The experimental results showed that the performance of LSTM sequence-to-sequence model is 0.94 in F1-measure, which is better than the rule-based deep-learning method of GRU-CRF.

• Korean Journal of Remote Sensing
• /
• v.35 no.3
• /
• pp.447-455
• /
• 2019

Since the forest type map in Korea has been mostly constructed every five years, the forest information from the map lacks up-to-date information. Forest research has been carried out by aerial photogrammetry and field surveys, and hence it took a lot of times and money. The vertical structure of forests is an important factor in evaluating forest diversity and environment. The vertical structure is essential information, but the observation of the vertical structure is not easy because the vertical structure indicates the internal structure of forests. In this study, the index map and texture map produced from KOMPSAT-3/3A/5 satellite images and the canopy information generated by the difference between DSM (Digital Surface Model) and DTM (Digital Terrain Model) were classified using the artificial neural network. The vertical structure of forests of single and multi-layer forests was classified to identify 81.59% of the final classification result.