• International Journal of Concrete Structures and Materials
• /
• v.1 no.1
• /
• pp.63-73
• /
• 2007

Optimum multi-layered feed-forward neural network (NN) models using a resilient back-propagation algorithm and early stopping technique are built to predict the shear capacity of reinforced concrete deep and slender beams. The input layer neurons represent geometrical and material properties of reinforced concrete beams and the output layer produces the beam shear capacity. Training, validation and testing of the developed neural network have been achieved using 50%, 25%, and 25%, respectively, of a comprehensive database compiled from 631 deep and 549 slender beam specimens. The predictions obtained from the developed neural network models are in much better agreement with test results than those determined from shear provisions of different codes, such as KBCS, ACI 318-05, and EC2. The mean and standard deviation of the ratio between predicted using the neural network models and measured shear capacities are 1.02 and 0.18, respectively, for deep beams, and 1.04 and 0.17, respectively, for slender beams. In addition, the influence of different parameters on the shear capacity of reinforced concrete beams predicted by the developed neural network shows consistent agreement with those experimentally observed.

• Wind and Structures
• /
• v.12 no.4
• /
• pp.299-312
• /
• 2009

It has long been studied about the flow around bluff bodies, but the effect of aspect ratio on the sharp-edged bodies in thick turbulent boundary layers is still argued. The author investigates the flow characteristics around a series of rectangular bodies ($40^d{\times}80^w{\times}80^h$, $80^d{\times}80^w{\times}80^h$ and $160^d{\times}80^w{\times}80^h$ in mm) placed in a deep turbulent boundary layer. The study is aiming to identify the extant Reynolds number independence of the rectangular bodies and furthermore understand the surface pressure distribution around the bodies such as the suction pressure in the leading edge, when the shape of bodies is changed, responsible for producing extreme suction pressures around the bluff bodies. The experiments are carried out at three different Reynolds numbers, based on the velocity U at the body height h, of 24,000, 46,000 and 67,000, and large enough that the mean boundary layer flow is effectively Reynolds number independent. The experiment includes wind tunnel work with the velocity and surface pressure measurements. The results show that the generation of the deep turbulent boundary layer in the wind tunnel and the surface pressure around the bodies were all independent of Reynolds number and the longitudinal length, but highly dependent of the transverse width.

• Ocean and Polar Research
• /
• v.27 no.1
• /
• pp.1-13
• /
• 2005

The samples for organic carbon analysis were collected between $5^{\circ}\;and\;17^{\circ}N$ along $131.5^{\circ}W$ in the northeast Pacific KODOS (Korea Deep Ocean Study) area. The mean concentration of total organic carbon (TOC) in the surface mixed layer $({\sim}50 m)$ was $100.13{\pm}2.05{\mu}M-C$, while the mean concentration of TOC in the lower 500m of the water column was $50.19{\pm}4.23{\mu}M-C$. A strong linear regression between TOC and temperature $(r^2=0.70)$ showed that TOC distribution was controlled by physical process. Results from the linear regression between chlorophyll-a and TOC, and between chlorophyll-a and particulate organic carbon (POC), decreasing of dissolved organic carbon (DOC) in the surface layer caused by non-biological photo-oxidation process. Below the surface layer, biological production and consumption occurred. DOC accumulation dominated in the depth range of $30{\sim}50m$ and DOC consumption occurred in the depth range of $50{\sim}200m$. TOC was inversely correlated with apparent oxygen utilization (AOU) and TOC/AOU molar ratios ranged from -0.077 to -0.21. These ratios indicated that TOC oxidation was responsible fur $10.9{\sim}30.1%$ (mean 20.2%) of oxygen consumption in the NE Pacific KODOS area. In the euphotic zone, distributions of dissolved and particulate organic matter were controlled by photo-chemical, chemical, biological and physical processes.

• Ocean and Polar Research
• /
• v.36 no.2
• /
• pp.165-177
• /
• 2014

In an effort to investigate the structure and variability of the coastal current in the East Sea, a moored ADCP observation was conducted off Uljin from late May to mid-October 2006. Owing to the transition of season from summer to autumn, the features of the current and wind can be divided into two parts. Until mid-August (Part-I), a southward flow is dominant at all depths with a mean alongshore velocity of 4.2~8.9 cm/s but northward winds are not strong enough to reverse the near-surface current. During Part-II, a strong northward current occurs frequently in the upper layer but winds are predominantly southward including two typhoons that have deep-reaching influence. Profile of mean velocity has three layers with a northward velocity embedded at 12~28 m depth. The near-surface current of Part-II significantly coheres with winds at 4-8 day periods with a phase lag of about 12 hours. The modal structure of the current obtained by EOF analysis is: (1) Mode-1, having 83.6% of total variance, represents the current in the same direction at all depths corresponding to the southward North Korean Cold Current (NKCC). (2) Mode-2 (11.7%) reveals a two-layer structure that can be explained by the northward East Korean Warm Current (EKWC) in the upper layer and NKCC in the lower. (3) Mode-3 (2.6%) has three layers, in which the EKWC is reversed near the surface by opposing winds. This mode is particularly similar to the mean velocity profile of Part-II.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• The Sea:JOURNAL OF THE KOREAN SOCIETY OF OCEANOGRAPHY
• /
• v.13 no.3
• /
• pp.178-189
• /
• 2008

The distribution of inorganic nutrients and their remineralization ratio in the divergence zone ($7^{\circ}{\sim}10.5^{\circ}N$) of the northeastern Pacific were investigated from July 2003 to July 2007. A divergence zone along the boundary of the North Equatorial Counter Current (NECC) and North Equatorial Current (NEC) at $10^{\circ}N$ was observed in July 2007 when the La Nina event and divergence-related upwelling was strong. The mean depth of oligotrophic surface mixed layer in the divergence zone was 46, 61, and 30 m in July 2003, August 2005, and July 2007, respectively. Below the surface mixed layer, a nutricline was clearly observed. The depth integrated value of nitrate including nitrite (DIVn) in the upper layer($0{\sim}100$ m depth) ranged from 5.51 to 21.71 $gN/m^2$(mean 12.82 $gN/m^2$) in July 2003, from 5.62 to 8.46 $gN/m^2$ (mean 7.15 $gN/m^2$) in August 2005, and from 8.98 to 27.80 $gN/m^2$(mean 21.12 $gN/m^2$) in July 2007. The maximum DIVn was observed at the divergence zone. The distributions of phosphate(DIVp) and silicate(DIVsi) were similar to that of DIVn and the DIVn/DIVsi ratio was $0.87{\pm}0.11$ in the upper layer. The limiting nutrient for phytoplankton growth in the study area was identified as nitrogen(N/P ratio=14.6). The nitrate (including nitrite) concentrations were lower in the region mainly affected by NEC than in the region affected by NECC. The study area of low silicate concentrations was also considered to be Si-limiting environment. The remineralization ratios of nutrients were $P/N/-O_2=1/14.6{\pm}1.1/100.4{\pm}8.8(23.44{\leq}Sigma-{\theta}{\leq}26.38)$ in the study area. These ratios suggested remineralization process in the surface layer of divergence zone.

• Structural Engineering and Mechanics
• /
• v.85 no.4
• /
• pp.469-484
• /
• 2023

A deep recursive bidirectional Cuda Deep Neural Network Long Short Term Memory (Bi-CuDNNLSTM) layer is recruited in this paper to predict the entire force time histories, and the corresponding hysteresis and backbone curves of reinforced concrete (RC) bridge piers using experimental fast and slow cyclic tests. The proposed stacked Bi-CuDNNLSTM layers involve multiple uncertain input variables, including horizontal actuator displacements, vertical actuators axial loads, the effective height of the bridge pier, the moment of inertia, and mass. The functional application programming interface in the Keras Python library is utilized to develop a deep learning model considering all the above various input attributes. To have a robust and reliable prediction, the dataset for both the fast and slow cyclic tests is split into three mutually exclusive subsets of training, validation, and testing (unseen). The whole datasets include 17 RC bridge piers tested experimentally ten for fast and seven for slow cyclic tests. The results bring to light that the mean absolute error, as a loss function, is monotonically decreased to zero for both the training and validation datasets after 5000 epochs, and a high level of correlation is observed between the predicted and the experimentally measured values of the force time histories for all the datasets, more than 90%. It can be concluded that the maximum mean of the normalized error, obtained through Box-Whisker plot and Gaussian distribution of normalized error, associated with unseen data is about 10% and 3% for the fast and slow cyclic tests, respectively. In recapitulation, it brings to an end that the stacked Bi-CuDNNLSTM layer implemented in this study has a myriad of benefits in reducing the time and experimental costs for conducting new fast and slow cyclic tests in the future and results in a fast and accurate insight into hysteretic behavior of bridge piers.

• Atmosphere
• /
• v.24 no.2
• /
• pp.207-223
• /
• 2014

The bogussing method was further developed by incorporating the asymmetric component into the symmetric bogus tropical cyclone of the Structure Adjustable Balanced Vortex (SABV). The asymmetric component is separated from the disturbance field associated with the tropical cyclone by establishing local polar coordinates whose center is the location of the tropical cyclone. The relative importance of wave components in azimuthal direction was evaluated, and only two or three wave components with large amplitude are added to the symmetric components. Using the Weather Research and Forecast model (WRF), initialized with the asymmetric bogus vortex, the track and central pressure of tropical cyclones were predicted. Nine tropical cyclones, which passed over Korean peninsula during 2010~2012 were selected to assess the effect of asymmetric components. Compared to the symmetric bogus tropical cyclone, the track forecast error was reduced by about 18.9% and 17.4% for 48 hours and 72 hours forecast, while the central pressure error was not improved significantly. The results suggest that the inclusion of asymmetric component is necessary to improve the track forecast of tropical cyclones.

• Ocean and Polar Research
• /
• v.21 no.2
• /
• pp.99-108
• /
• 1999

With sectional data obtained in September of 1987, 1988 and 1989 by quadrireciprocal ADCP measurement and CTD cast, the current structure, volume transport and vorticity in the Western Channel of the Korea Strait were studied. The characteristics of Tsushima Current water persisted throughout the summer especially in the homogeneous water of temperature $14-16^{\circ}C$ located at the depth of 50-100m below seasonal termocline. Thickness and velocity of the homogeneous layer are about 10-170m and 20-60cm/s. and the relative vorticity for this layer is shown to be nearly constant and it is smaller than the planetary vorticity. Potential vorticity of $2.70-7.10{\times}10^{-6}m^{-1}s^{-1}$ is found to be dependent mainly on planetary rather than on the relative vorticities. The Tsushima Current water represented by the homogeneous layer R14-16^{\circ}C$ may keep the potential vorticity at the area of strong current in the Strait. The ADCP current structure is similar to geostrophic current and the core of the current with the speed of 30-50cm/s is situated in the middle layer over the deep trough. With large tidal fluctuation the volume transport has mean value of 1.17sv which was about 40% larger than that of geostrophic calculation.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.23 no.4
• /
• pp.163-168
• /
• 1987

The underwater observation of the ambient noise and the noise generated by the engine revolution in a ship was carried out in July to August, 1984, 1985 and 1987, near around some ports and in the Eastern Sea of Korea. Vertical distribution of the sound pressure of both noises were observed and the spectrum characteristics were analysed and compared. The results obtained are summarized as follows: 1. Sound pressure level of the ambient noise at 5m deep layer in calm sea condition (wind speed 0-2m/s) near around the ports were observed as 108dB at the eastern part of Pusan port, 106dB at the southern part of Pusan port and 101dB at Kuryongpo port. It shows that the level near around the large port which contains much noisy resources is higher than the small port. The level at 5m deep layer in the open sea, in the mid-region between Korean Peninsula and Ulnung Island was observed as 100dB. It mean that the level in the open sea is lower than that around the ports. The level at 20m and 70m deep layer were 1-2dB lower than that at 5m deep layer, and that at deeper layer than 100m was almost constantly 100dB around. 2. Sound pressure level of the ambient noise at 5m deep layer in windy open sea condition (wind speed 10-15m/s) was 108dB, and was gradually decreased in accordance with the increase of depth with representing 100dB at 70m deep layer and that at deeper layer was almost constantly 100dB. The level of the noise generated by engine revolution was 146, 125, 112, 110, 104dB at 5, 50, 100, 150 and 200m deep layer respectively. It means that the level decrease with the depth. 3. Spectrum level of the ambient noise at 5m deep layer with the frequency band of 10 Hz, 100 Hz, 1 KHz, 10 KHz, in the windy sea condition were 86, 75, 61, 32dB respectively and the level of the noise generated by engine revolution was 105, 95, 86, 55dB respectively. It means that the latter are about 20dB higher than the former. The level of the former at 200m deep layer was 80, 68, 47, 26dB and the latter 82, 70, 59, 31dB. It means that the latter are about 4dB higher than the former.