• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.29B no.1
• /
• pp.94-101
• /
• 1992

The multilayer perceptron, trained by the error back-propagation learning rule, has been known as a mapping network which can represent arbitrary functions. However depending on the complexity of a function and the initial weights of the multilayer perceptron, the error back-propagation learning may fall into a local minimum or a flat area which may require a long learning time or lead to unsuccessful learning. To solve such difficulties in training the multilayer perceptron by standard error back-propagation learning rule, the paper proposes a learning method which progressively enlarges the learning domain from a small area to the entire region. The proposed method is devised from the investigation on the roles of hidden nodes and connection weights in the multilayer perceptron which approximates a function of one variable. The validity of the proposed method was illustrated through simulations for a function of one variable and a function of two variable with many extremal points.

• Journal of information and communication convergence engineering
• /
• v.21 no.3
• /
• pp.233-238
• /
• 2023

This study presents a method for identifying partial discharge defects in an eco-friendly gas insulated system using a backpropagation algorithm. Four partial discharge (PD) electrode systems, namely, a free-moving particle, protrusion on the conductor, protrusion on the enclosure, and voids, were designed to simulate PD defects that can occur during the operation of eco-friendly gas-insulated switchgear. The PD signals were measured using an ultrahigh-frequency sensor as a nonconventional method based on IEC 62478. To identify the types of PD defects, the PD parameters of single PD pulses in the time and frequency domains and the phase-resolved partial discharge patterns were extracted, and a back-propagation algorithm in the artificial neural network was designed using a virtual instrument based on LabVIEW. The backpropagation algorithm proposed in this paper has an accuracy rate of over 90% for identifying the types of PD defects, and the result is expected to be used as a reference database for asset management and maintenance work for eco-friendly gas-insulated power equipment.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.5 no.4
• /
• pp.357-362
• /
• 2010

This paper proposes the speech and noise recognition system by using a neural network in order to detect the speech and noise sections at each frame. The proposed neural network consists of a layered neural network training by back-propagation algorithm. First, a power spectrum obtained by fast Fourier transform and linear predictive coefficients are used as the input to the neural network for each frame, then the neural network is trained using these power spectrum and linear predictive coefficients. Therefore, the proposed neural network can train using clean speech and noise. The performance of the proposed recognition system was evaluated based on the recognition rate using various speeches and white, printer, road, and car noises. In this experiment, the recognition rates were 92% or more for such speech and noise when training data and evaluation data were the different.

• Proceedings of the KIEE Conference
• /
• 2006.07d
• /
• pp.1868-1870
• /
• 2006

Neural network is used in many fields of control systems, but input-output patterns of a control system are not easy to be obtained and by using as single feedback neural network controller. And also it is difficult to get a satisfied performance when the changes of rapid load and disturbance are applied. To resolve those problems, this paper proposes a new algorithm which is the neural network controller. The new algorithm uses the neural network instead of activation function to control object at the output node. Therefore, control object is composed of neural network controller unifying activation function, and it supplies the error back propagation path to calculate the error at the output node. As a result, the input-output pattern problem of the controller which is resigned by the simple structure of neural network is solved, and real-time learning can be possible in general back propagation algorithm. Application of the new algorithm of neural network controller gives excellent performance for initial and tracking response and it shows the robust performance for rapid load change and disturbance. The proposed control algorithm is implemented on a high speed DSP, TMS320C32, for the speed of 3-phase induction motor. Enhanced performance is shown in the test of the speed control.

• The KIPS Transactions:PartB
• /
• v.14B no.4
• /
• pp.255-262
• /
• 2007

This paper proposes an enhanced skin color-based detection method to find a region of human face in color images. The proposed detection method combines three color spaces, RGB, $YC_bC_r$, YIQ and builds color union histograms of luminance and chrominance components respectively. Combined color union histograms are then fed in to the back-propagation neural network for training and Levenberg-Marquadt algorithm is applied to the iteration process of training. Proposed method with Levenberg-Marquadt algorithm applied to training process of neural network contributes to solve a local minimum problem of back-propagation neural network, one of common methods of training for face detection, and lead to make lower a detection error rate. Further, proposed color-based detection method using combined color union histograms which give emphasis to chrominance components divided from luminance components inputs more confident values at the neural network and shows higher detection accuracy in comparison to the histogram of single color space. The experiments show that these approaches perform a good capability for face region detection, and these are robust to illumination conditions.

• Korean Journal of Cognitive Science
• /
• v.7 no.4
• /
• pp.141-162
• /
• 1996

• The Journal of Engineering Geology
• /
• v.15 no.1
• /
• pp.67-76
• /
• 2005

This paper presents the application of a neural network for prediction of the unconfined compressive strength from physical properties and schmidt hardness number on rock samples. To investigate the suitability of this approach, the results of analysis using a neural network are compared to predictions obtained by statistical relations. The data sets containing 55 rock sample records which are composed of sandstone and shale were assembled in Daegu area. They were used to learn the neural network model with the back-propagation teaming algorithm. The rock characteristics as the teaming input of the neural network are: schmidt hardness number, specific gravity, absorption, porosity, p-wave velocity and S-wave velocity, while the corresponding unconfined compressive strength value functions as the teaming output of the neural network. A data set containing 45 test results was used to train the networks with the back-propagation teaming algorithm. Another data set of 10 test results was used to validate the generalization and prediction capabilities of the neural network.

• Transactions of Materials Processing
• /
• v.4 no.3
• /
• pp.214-221
• /
• 1995

In the paper, we have proposed a new method to determine the initial billet for the forged products using a function approximation in the neural network. The architecture of neural network is a three-layer neural network and the back propagation algorithm is employed to train the network. By utilizing the ability of function approximation of a neural network, an optimal billet is determined by applying the nonlinear mathematical relationship between the aspect ratios in the initial billet and the final products. The amount of incomplete filling in the die is measured by the rigid-plastic finite element method. The neural network is trained with the initial billet aspect ratios and those of the unfilled volumes. After learning, the system is able to predict the filling regions which are exactly the same or slightly different to the results of finite element simulation. This new method is applied to find the optimal billet size for the plane strain rib-web product in cold forging. This would reduce the number of finite element simulation for determining the optimal billet size of forging product, further it is usefully adapted to physical modeling for the forging design.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.21 no.5
• /
• pp.727-734
• /
• 1997

In practice, the design of forging processes is performed based on an experience-oriented technology, that is designer's experience and expensive trial and errors. Using the finite element simulation and the artificial neural network, we propose an optimal die geometry satisfying the design conditions of final product. A three-layer neural network is used and the back propagation algorithm is employed to train the network. An optimal die geometry that satisfied the same between inner extruded rib and outer extruded one is determined by applying the ability of function approximation of neural network. The neural networks may reduce the number of finite element simulation for determine the optimal die geometry of forging products and further they are usefully applied to physical modelling for the forging design.

• Journal of Advanced Marine Engineering and Technology
• /
• v.26 no.6
• /
• pp.695-704
• /
• 2002

Generally, induction motor controller requires rotor speed sensor for commutation and current control, but it increases cost and size of the motor. So in these days, various researches including speed sensorless vector control have been reported and some of them have been put to practical use. In this paper a new speed estimation method using neural networks is proposed. The optimal neural network structure was tracked down by trial and error, and it was found that the 8-16-1 neural network has given correct results for the instantaneous rotor speed. Supervised learning methods, through which the neural network is trained to learn the input/output pattern presented, are typically used. The back-propagation technique is used to adjust the neural network weights during training. The rotor speed is calculated by weights and eight inputs to the neural network. Also, the proposed method has advantages such as the independency on machine parameters, the insensitivity to the load condition, and the stability in the low speed operation.