• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2003.07a
• /
• pp.518-521
• /
• 2003

A new measurement system was developed by fault diagnosis system for traction motor using current signal analysis. The motor current signature analysis method is used for traction motor fault diagnosis. The diagnosis system program is constructed by artificial neural networks algorithm, those results from the program are used to train neural networks. The trained neural networks have the ability to compute adaptive results for non-trained inputs, and to calculate very fast due to original parallel structure of neural networks with high accuracy within destined tolerance. This system suggested that available test for checking, the probable extent of aging, and the rate of which aging is taking place.

• Bulletin of the Korean Chemical Society
• /
• v.28 no.9
• /
• pp.1477-1484
• /
• 2007

Artificial neural networks (ANNs), for a first time, were successfully developed for the prediction partial molar heat capacity of aqueous solutions at infinite dilution for various polar aromatic compounds over wide range of temperatures (303.55-623.20 K) and pressures (0.1-30.2 MPa). Two three-layered feed forward ANNs with back-propagation of error were generated using three (the heat capacity in T = 303.55 K and P = 0.1 MPa, temperature and pressure) and six parameters (four theoretical descriptors, temperature and pressure) as inputs and its output is partial molar heat capacity at infinite dilution. It was found that properly selected and trained neural networks could fairly represent dependence of the heat capacity on the molecular descriptors, temperature and pressure. Mean percentage deviations (MPD) for prediction set by the models are 4.755 and 4.642, respectively.

• Proceedings of the KIEE Conference
• /
• 2000.07d
• /
• pp.2946-2948
• /
• 2000

In this paper, we proposed the Fuzzy Polynomial Neural Networks(FPNN) model with fuzzy activation node. The proposed FPNN structure is generated from the mutual combination of PNN(Polynomial Neural Networks) structure and fuzzy inference system. The premise of fuzzy inference rules defines by triangular and gaussian type membership function. The fuzzy inference method uses simplified and regression polynomial inference method which is based on the consequence of fuzzy rule expressed with a polynomial such as linear, quadratic and modified quadratic equation are used. The structure of FPNN is not fixed like in conventional Neural Networks and can be generated. The design procedure to obtain an optimal model structure utilizing FPNN algorithm is shown in each stage. Gas furnace time series data used to evaluate the performance of our proposed model.

• Proceedings of the KIEE Conference
• /
• 1998.11b
• /
• pp.651-653
• /
• 1998

This thesis propose the prefilter type control strategies using modified chaotic neural networks #or the trajectory control of robotic manipulator. Since the structure of chaotic neural networks and neurons, chaotic neural networks can show the robust characteristics for controlling highly nonlinear dynamics like robotic manipulators. For its application, the trajectory controller of the three-axis PUMA robot is designed by CNN. The CNN controller acts as the compensator of the PD controller. Simulation results show that learning error decrease drastically via on- line learning and the performance is excellent. The CNN controller have much better controllability and shorter calculation time compared to the RNN controller. Another advantage of the proposed controller could be attached to conventional robot controller without hardware changes.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.25 no.4
• /
• pp.271-280
• /
• 2016

Predicting cutting forces in machine tools is essential to productivity improvement and process control in the manufacturing field. Furthermore, milling machining is more complicated than turning machining. Therefore, several studies have been conducted previously to simulate milling forces; this study aims to simulate the cutting forces in milling machines using multi-layered neural networks. In the experiments, the number of layers in these networks was 3 and 4 and the number of neurons in the hidden layers was varied from 20 to 200. The root mean square errors of simulated cutting force components were obtained from taught and untaught data for the various neural networks. Results show that the error trends for untaught data were non-uniform because of the complex nature of the cutting force components, which was caused by different cutting factors and nonlinear characteristics coming into play. However, trends for taught data showed a very good coincidence.

• Journal of Semiconductor Engineering
• /
• v.1 no.1
• /
• pp.31-37
• /
• 2020

We present an approximate DRAM architecture for energy-efficient deep learning. Our key premise is that by bounding memory errors to non-critical information, we can significantly reduce DRAM refresh energy without compromising recognition accuracy of deep neural networks. To validate the key premise, we make extensive Monte-Carlo simulations for several well-known convolutional neural networks such as LeNet, ConvNet and AlexNet with the input of MINIST, CIFAR-10, and ImageNet, respectively. We assume that the highest-order 8-bits (in single precision) and 4-bits (in half precision) are protected from retention errors under the proposed architecture and then, randomly inject bit-errors to unprotected bits with various bit-error-rates. Here, recognition accuracies of the above convolutional neural networks are successfully maintained up to the 10^-5-order bit-error-rate. We simulate DRAM energy during inference of the above convolutional neural networks, where the proposed architecture shows the possibility of considerable energy saving up to 10 ~ 37.5% of total DRAM energy.

• International Journal of Advanced Culture Technology
• /
• v.7 no.4
• /
• pp.283-288
• /
• 2019

Neural networks have been reborn as a Deep Learning thanks to big data, improved processor, and some modification of training methods. Neural networks used to initialize weights in a stupid way, and to choose wrong type activation functions of non-linearity. Weight initialization contributes as a significant factor on the final quality of a network as well as its convergence rate. This paper discusses different approaches to weight initialization. MNIST dataset is used for experiments for comparing their results to find out the best technique that can be employed to achieve higher accuracy in relatively lower duration.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.10a
• /
• pp.474-478
• /
• 1998

A new approach for the decision feedback equalizer(DFE) based on the back-propagation neural networks is described. We propose the method of optimal structure for back-propagation neural networks model. In order to construct an the optimal structure, we first prescribe the bounds of learning procedure, and the, we employ the method of incrementing the number of input neuron by utilizing the derivative of the error with respect to an hidden neuron weights. The structure is applied to the problem of adaptive equalization in the presence of inter symbol interference(ISI), additive white Gaussian noise. From the simulation results, it is observed that the performance of the propose neural networks based decision feedback equalizer outperforms the other two in terms of bit-error rate(BER) and attainable MSE level over a signal ratio and channel nonlinearities.

• Proceedings of the KIEE Conference
• /
• 2001.07d
• /
• pp.2358-2361
• /
• 2001

In recent years there have been increasing interests in real-time object tracking with image information. Since image information is affected by illumination, this paper presents the real-time object tracking method based on neural networks which have robust characteristics under various illuminations. This paper proposes three steps to track the object and the fast tracking method. In the first step the object color is extracted using neural networks. In the second step we detect the object feature information based on invariant moment. Finally the object is tracked through a shape recognition using neural networks. To achieve the fast tracking performance, this paper first has a global search of entire image and tracks the object through local search when the object is recognized.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.50 no.4
• /
• pp.184-191
• /
• 2001

This paper proposes a prefilter type velocity compensating control system using modified chaotic neural networks for the trajectory control of robotic manipulator. Since the structure of modified chaotic neural networks(MCNN) and neurons have highly nonlinear dynamic characteristics, MCNN can show the robust characteristics for controlling highly nonlinear dynamics like robotic manipulators. For its application, the trajectory controller of the three-axis robot manipulator is designed by MCNN. The MCNN controller acts as the compensator of the PD controller. Simulation results show that learning error decrease drastically via on-line learning and the performance is excellent. The MCNN controller showed much better control performance and shorter calculation time compared to the RNN controller, Another advantage of the proposed controller could by attached to conventional robot controller without hardware changes.