• Computers and Concrete
• /
• v.12 no.3
• /
• pp.285-301
• /
• 2013

This paper concentrates on the results of experimental work on tensile strength of self-compacting concrete (SCC) caused by flexure, which is called rupture modulus. The work focused on concrete mixes having water/binder ratios of 0.35 and 0.45, which contained constant total binder contents of 500 $kg/m^3$ and 400 $kg/m^3$, respectively. The concrete mixes had four different dosages of a superplasticizer based on polycarboxylic with and without silica fume. The percentage of silica fume that replaced cement in this research was 10%. Based upon the experimental results, the existing equations for anticipating the rupture modulus of SCC according to its compressive strength were not exact enough. Therefore, it is decided to use artificial neural networks (ANN) for anticipating the rupture modulus of SCC from its compressive strength and workability. The conclusion was that the multi layer perceptron (MLP) networks could predict the tensile strength in all conditions, but radial basis (RB) networks were not exact enough in some circumstances. On the other hand, RB networks were more users friendly and they converged to the final networks quicker.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.12
• /
• pp.2418-2425
• /
• 1994

Although the error backpropagation(EBP) algorithm based on the gradient descent method is a widely-used learning algorithm of neural networks, learning sometimes takes a long time to acquire accuracy. This paper develops a novel learning method to alleviate the problems of EBP algorithm such as local minima, slow speed, and size of structure and thus to improve performance by adopting other new networks. Gaussian Potential Function networks(GPFN), in parallel with multilayer neural networks. Empirical simulations show the efficacy of the proposed algorithm in function approximation, which enables us to train networks faster with the better generalization capabilities.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2008.05a
• /
• pp.205-208
• /
• 2008

In numerical analysis numerical values of thermodynamic properties such as temperature, pressure, specific volume, enthalpy and entropy are required. But most of the thermodynamic properties of the steam table are determined by experiment. Therefore they are supposed to have measurement errors. In order to make noised thermodynamic properties corresponding to errors, random numbers are generated, adjusted to appropriate magnitudes and added to original thermodynamic properties. the neural networks and quadratic spline interpolation method are introduced for function approximation of these modified thermodynamic properties in the saturated water based on pressure. It was proved that the neural networks give smaller percentage error compared with quadratic spline interpolation. From this fact it was confirmed that the neural networks trace the original values of thermodynamic properties better than the quadratic interpolation method.

• International Journal of Quality Innovation
• /
• v.6 no.1
• /
• pp.109-118
• /
• 2005

The parameter design is the most emphasized measure by researchers for a new products development. It is critical for makers to achieve simultaneously in both the time-to-market production and the quality enhancement. However, there are difficulties in practical application, such as (1) complexity and nonlinear relationships co-existed among the system's inputs, outputs and control parameters, (2) interactions occurred among parameters, (3) where the adjustment factors of Taguchi's two-phase optimization procedure cannot be sure to exist in practice, and (4) for some reasons, the data became lost or were never available. For these incomplete data, the Taguchi methods cannot treat them well. Neural networks have a learning capability of fault tolerance and model free characteristics. These characteristics support the neural networks as a competitive tool in processing multivariable input-output implementation. The successful fields include diagnostics, robotics, scheduling, decision-making, prediction, etc. This research is a case study of spherical annealing model. In the beginning, an original model is used to pre-fix a model of parameter design. Then neural networks are introduced to achieve another model. Study results showed both of them could perform the highest spherical level of quality.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.5
• /
• pp.728-733
• /
• 2018

Deep learning has been actively studied for predicting protein secondary structure based only on the sequence information of the amino acids constituting the protein. In this paper, we compared the performances of the convolutional neural networks of various structures to predict the protein secondary structure. To investigate the optimal depth of the layer of neural network for the prediction of protein secondary structure, the performance according to the number of layers was investigated. We also applied the structure of GoogLeNet and ResNet which constitute building blocks of many image classification methods. These methods extract various features from input data, and smooth the gradient transmission in the learning process even using the deep layer. These architectures of convolutional neural networks were modified to suit the characteristics of protein data to improve performance.

• Smart Structures and Systems
• /
• v.4 no.5
• /
• pp.583-603
• /
• 2008

In this study, a real-time damage detection method using output-only acceleration signals and artificial neural networks (ANN) is developed to monitor the occurrence of damage and the location of damage in structures. A theoretical approach of an ANN algorithm that uses acceleration signals to detect changes in structural parameters in real-time is newly designed. Cross-covariance functions of two acceleration responses measured before and after damage at two different sensor locations are selected as the features representing the structural conditions. By means of the acceleration features, multiple neural networks are trained for a series of potential loading patterns and damage scenarios of the target structure for which its actual loading history and structural conditions are unknown. The feasibility of the proposed method is evaluated using a numerical beam model under the effect of model uncertainty due to the variability of impulse excitation patterns used for training neural networks. The practicality of the method is also evaluated from laboratory-model tests on free-free beams for which acceleration responses were measured for several damage cases.

• The KIPS Transactions:PartB
• /
• v.11B no.7 s.96
• /
• pp.831-840
• /
• 2004

In this paper, we propose a scheme to implement intelligent game characters based on neural networks. Neural networks that implement in-telligent game character receive the action of an opponent character and the distance between them, decide intelligent character's action, and output the decision. The neural networks are trained by reinforcement learning using the scores acquired by the actions of two characters as reinforcement values. To show the usefulness of the proposed scheme, a simple fighting action game is implemented and various experiments are performed. Experimental results show that proposed intelligent characters can learn the rule of the game. The proposed scheme can be ap-plied to massively multiple online games as well as fighting action games.

• Proceedings of the KIEE Conference
• /
• 2005.10b
• /
• pp.149-151
• /
• 2005

In this paper, we propose the Optimal Polynomial Neural Networks(PNN) for nonlinear process. The PNN is based on Group Method of Data Handling(GMDH) method and its structure is similar to feedforward Neural Networks. But the structure of PNN is not fixed like in conventional Neural Networks and can be generated. The each node of PNN structure uses several types of high-order polynomial such as linear, quadratic and modified quadratic, and is connected as various kinds of multi-variable inputs. The conventional PNN depends on experience of a designer that select No. of input variable, input variable and polynomial type. Therefore it is very difficult a organizing of optimized network. The proposed algorithm identified and selected No. of input variable, input variable and polynomial type by using Genetic Algorithms(GAs). In the sequel the proposed model shows not only superior results to the existing models, but also pliability in organizing of optimal network. Medical Imaging System(MIS) data is simulated in order to confirm the efficiency and feasibility of the proposed approach in this paper.

• ETRI Journal
• /
• v.42 no.3
• /
• pp.311-323
• /
• 2020

Encrypted traffic classification plays a vital role in cybersecurity as network traffic encryption becomes prevalent. First, we briefly introduce three traffic encryption mechanisms: IPsec, SSL/TLS, and SRTP. After evaluating the performances of support vector machine, random forest, naïve Bayes, and logistic regression for traffic classification, we propose the combined approach of entropy estimation and artificial neural networks. First, network traffic is classified as encrypted or plaintext with entropy estimation. Encrypted traffic is then further classified using neural networks. We propose using traffic packet's sizes, packet's inter-arrival time, and direction as the neural network's input. Our combined approach was evaluated with the dataset obtained from the Canadian Institute for Cybersecurity. Results show an improved precision (from 1 to 7 percentage points), and some application classification metrics improved nearly by 30 percentage points.

• Computers and Concrete
• /
• v.18 no.6
• /
• pp.1113-1134
• /
• 2016

A methodology using neural networks has been proposed for rapid prediction of inelastic bending moments in reinforced concrete continuous beams subjected to service load. The closed form expressions obtained from the trained neural networks take into account cracking in concrete at in-span and at near the internal supports and tension stiffening effect. The expressions predict the inelastic moments (considering the concrete cracking) from the elastic moments (neglecting the concrete cracking) at supports. Three separate neural networks are trained since these have been postulated to represent all the beams having any number of spans. The training, validating, and testing data sets for the neural networks are generated using an analytical-numerical procedure of analysis. The proposed expressions are verified for example beams of different number of spans and cross-section properties and the errors are found to be small. The proposed expressions, at minimal input data and computation effort, yield results that are close to FEM results. The expressions can be used in preliminary every day design as they enable a rapid prediction of inelastic moments and require a computational effort that is a fraction of that required for the available methods in literature.