• 소성∙가공
• /
• 제27권1호
• /
• pp.28-36
• /
• 2018

Electromagnetic forming is a type of high-speed forming process to deform a workpiece through a Lorentz force. As the high strain rate in an electromagnetic-forming simulation causes infeasibility in determining constitutive parameters, we employed inverse parameter estimation in the previous study. However, the inverse parameter estimation process required us to spend considerable time, which leads to an increase in computational cost. To overcome the computational obstacle, in this research, we applied two types of surrogate modeling methods and compared them to each other to evaluate which model is best for the electromagnetic-forming simulation. We exploited an artificial neural network and we reduced-order modeling methods. During the construction of a reduced-order model, we extracted orthogonal bases with proper orthogonal decomposition and predicted basis coefficients by utilizing an artificial neural network. After the construction of the surrogate models, we verified the artificial neural network and reduced-order models through training and testing samples. As a result, we determined the artificial neural network model is slightly more accurate than the reduced-order model. However, the construction of the artificial neural network model requires a considerably larger amount of time than that of the reduced-order model. Thus, a reduced order modeling method is more efficient than an artificial neural network for estimating the electromagnetic forming and for the rapid approximation of structural simulations which needs repetitive runs.

• 한국지능정보시스템학회:학술대회논문집
• /
• 한국지능정보시스템학회 2001년도 The Pacific Aisan Confrence On Intelligent Systems 2001
• /
• pp.133-141
• /
• 2001

Since the computing environment changes very rapidly, the estimation of software effort is very difficult because it is not easy to collect a sufficient number of relevant cases from the historical data. If we pinpoint the cases, the number of cases becomes too small. However is we adopt too many cases, the relevance declines. So in this paper we attempt to balance the number of cases and relevance. Since many researches on software effort estimation showed that the neural network models perform at least as well as the other approaches, so we selected the neural network model as the basic estimator. We propose a search method that finds the right level of relevant cases for the neural network model. For the selected case set. eliminating the qualitative input factors with the same values can reduce the scale of the neural network model. Since there exists a multitude of combinations of case sets, we need to search for the optimal reduced neural network model and corresponding case, set. To find the quasi-optimal model from the hierarchy of reduced neural network models, we adopted the beam search technique and devised the Case-Set Selection Algorithm. This algorithm can be adopted in the case-adaptive software effort estimation systems.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2001년도 추계학술발표논문집 (상)
• /
• pp.363-366
• /
• 2001

It is very difficult to maintain the performance of estimation models for the new breed of projects since the computing environment changes so rapidly in terms of programming languages, development tools, and methodologies. So, we propose to use the relevant cases for a neural network model, whose cost is the decreased number of cases. To balance the relevance and data availability, the qualitative input factors are used as criteria of data classification. With the data sets that have the same value for certain qualitative input factors, we can eliminate the factors from the model making reduced neural network models. So we need to seek the optimally reduced neural network model among them. To find the optimally case-selective neural network, we propose the search techniques and sensitivity analysis between data points and search space.

• 한국IT서비스학회:학술대회논문집
• /
• 한국IT서비스학회 2003년도 추계학술대회
• /
• pp.616-623
• /
• 2003

Since the computing environment changes very rapidly, the estimation of software effort is very difficult because it is not easy to collect a sufficient number of relevant cases from the historical data. If we pinpoint the cases, the number of cases becomes too small. However if we adopt too many cases, the relevance declines. So in this paper we attempt to balance the number of cases and relevance. Since many researches on software effort estimation showed that the neural network models perform at least as well as the other approaches, so we selected the neural network model as the basic estimator. We propose a search method that finds the right level of relevant cases for the neural network model. For the selected case set, eliminating the qualitative input factors with the same values can reduce the scale of the neural network model. Since there exists a multitude of combinations of case sets, we need to search for the optimal reduced neural network model and corresponding case set. To find the quasi-optimal model from the hierarchy of reduced neural network models, we adopted the beam search technique and devised the Case-Set Selection Algorithm. This algorithm can be adopted in the case-adaptive software effort estimation systems.

• 말소리와 음성과학
• /
• 제7권4호
• /
• pp.3-8
• /
• 2015

In this paper, we investigate an input dimension reduction method using continuous word vector in deep neural network language model. In the proposed method, continuous word vectors were generated by using Google's Word2Vec from a large training corpus to satisfy distributional hypothesis. 1-of-${\left|V\right|}$ coding discrete word vectors were replaced with their corresponding continuous word vectors. In our implementation, the input dimension was successfully reduced from 20,000 to 600 when a tri-gram language model is used with a vocabulary of 20,000 words. The total amount of time in training was reduced from 30 days to 14 days for Wall Street Journal training corpus (corpus length: 37M words).

• 한국정밀공학회지
• /
• 제24권4호
• /
• pp.93-101
• /
• 2007

Aluminum alloy which is one of the light materials has been tried to apply to light weight vehicle body. In order to do that, welding technology is very important. In case of the aluminum laser welding, the strength of welded part is reduced due to porosity, underfill, and magnesium loss. To overcome these problems, laser welding of aluminum with filler wire was suggested. In this study, experiment about laser welding of AA5182 aluminum alloy with AA5356 filler wire was performed according to process parameters such as laser power, welding speed and wire feed rate. The tensile strength was measured to find the weldability of laser welding with filler wire. The models to estimate tensile strength were suggested using three regression models and one neural network model. For regression models, one was the multiple linear regression model, another was the second order polynomial regression model, and the other was the multiple nonlinear regression model. Neural network model with 2 hidden layers which had 5 and 3 nodes respectively was investigated to find the most suitable model for the system. Estimation performance was evaluated for each model using the average error rate. Among the three regression models, the second order polynomial regression model had the best estimation performance. For all models, neural network model has the best estimation performance.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1996년도 한국자동제어학술회의논문집(국내학술편); 포항공과대학교, 포항; 24-26 Oct. 1996
• /
• pp.1083-1086
• /
• 1996

Cold rolling mill process in steel works uses stands of rolls to flatten a strip to a desired thickness. At cold rolling mill process, precalculation determines the mill settings before a strip actually enters the mill and is done by an outdated mathematical model. A corrective neural network model is proposed to improve the accuracy of the roll force prediction. Additional variables to be fed to the network include the chemical composition of the coil, its coiling temperature and the aggregated amount of processed strips of each roll. The network was trained using a standard backpropagation with 4,944 process data collected from no.1 cold rolling mill process from March 1995 through December 1995, then was tested on the unseen 1,586 data from Jan 1996 through April 1996. The combined model reduced the prediction error by 32.8% on average.

• Smart Structures and Systems
• /
• 제26권2호
• /
• pp.175-184
• /
• 2020

Conventional Monte Carlo simulation-based methods for seismic risk assessment of water networks often require excessive computational time costs due to the hydraulic analysis. In this study, an Artificial Neural Network-based surrogate model was proposed to efficiently evaluate the flow-based system reliability of water distribution networks. The surrogate model was constructed with appropriate training parameters through trial-and-error procedures. Furthermore, a deep neural network with hidden layers and neurons was composed for the high-dimensional network. For network training, the input of the neural network was defined as the damage states of the k-dimensional network facilities, and the output was defined as the network system performance. To generate training data, random sampling was performed between earthquake magnitudes of 5.0 and 7.5, and hydraulic analyses were conducted to evaluate network performance. For a hydraulic simulation, EPANET-based MATLAB code was developed, and a pressure-driven analysis approach was adopted to represent an unsteady-state network. To demonstrate the constructed surrogate model, the actual water distribution network of A-city, South Korea, was adopted, and the network map was reconstructed from the geographic information system data. The surrogate model was able to predict network performance within a 3% relative error at trained epicenters in drastically reduced time. In addition, the accuracy of the surrogate model was estimated to within 3% relative error (5% for network performance lower than 0.2) at different epicenters to verify the robustness of the epicenter location. Therefore, it is concluded that ANN-based surrogate model can be utilized as an alternative model for efficient seismic risk assessment to within 5% of relative error.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2000년도 제15차 학술회의논문집
• /
• pp.212-212
• /
• 2000

This study was performed to construct a hot strip speed measuring system and check over whether the measured speed can be used for improving the mass flow of the head-end part of a hot strip in the 7-stand finishing mill. Because the mass flow in hot rolling mill affects the looper operation and the thickness and width control of a strip, accurate measurement of strip speed ie important. The measured speed was compared with the roll speeds of No. 6 and No.7 stand to check the performance of the system and analyzed to find how to apply the speed. As a result, it is shown that the accuracy of the system is enough, strip thickness error can be reduced by -275∼+200$\mu\textrm{m}$ using the measured speed and the existing FSU model has low accuracy for predicting forward slip rate. A neural network was developed to calculate forward slip rate instead of FSU model. The test result of the neural network shows that the neural network is more accurate than the FSU model.

• International Journal of Quality Innovation
• /
• 제3권2호
• /
• pp.113-131
• /
• 2002

The relationships among multi-dimensional data (such as medical examination data) with ambiguity and variation are difficult to explore. The traditional approach to building a data classification system requires the formulation of rules by which the input data can be analyzed. The formulation of such rules is very difficult with large sets of input data. This paper first describes two classification approaches using back-propagation (BP) neural network and Mahalanobis distance (MD) classifier, and then proposes two classification approaches for multi-dimensional feature selection. The first one proposed is a feature selection procedure from the trained back-propagation (BP) neural network. The basic idea of this procedure is to compare the multiplication weights between input and hidden layer and hidden and output layer. In order to simplify the structure, only the multiplication weights of large absolute values are used. The second approach is Mahalanobis-Taguchi system (MTS) originally suggested by Dr. Taguchi. The MTS performs Taguchi's fractional factorial design based on the Mahalanobis distance as a performance metric. We combine the automatic thresholding with MD: it can deal with a reduced model, which is the focus of this paper In this work, two case studies will be used as examples to compare and discuss the complete and reduced models employing BP neural network and MD classifier. The implementation results show that proposed approaches are effective and powerful for the classification.