• Proceedings of the KIEE Conference
• /
• 2002.07d
• /
• pp.2755-2757
• /
• 2002

In this paper, we propose a now architecture for hardware implementation of digital neural network. By adopting flexible ladder-style bus and internal connection network into traditional SIMD-type digital neural network architecture, the proposed architecture enables fast processing that is based on parallelism, while does not abandon the flexibility and extensibility of the traditional approach. In the proposed architecture, users can change the network topology by setting configuration registers. Such reconfigurability on hardware allows enough usability like software simulation. We implement the proposed design on real FPGA, and configure the chip to multi-layer perceptron with back propagation for alphabet recognition problem. Performance comparison with its software counterpart shows its value in the aspect of performance and flexibility.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.8 no.3
• /
• pp.185-191
• /
• 2008

In this paper, experimental studies of a neural network (NN) control technique for non-model based position control of the x-y table robot are presented. Decentralized neural networks are used to control each axis of the x-y table robot separately. For an each neural network compensator, an inverse control technique is used. The neural network control technique called the reference compensation technique (RCT) is conceptually different from the existing neural controllers in that the NN controller compensates for uncertainties in the dynamical system by modifying desired trajectories. The back-propagation learning algorithm is developed in a real time DSP board for on-line learning. Practical real time position control experiments are conducted on the x-y table robot. Experimental results of using neural networks show more excellent position tracking than that of when PD controllers are used only.

• Journal of Institute of Control, Robotics and Systems
• /
• v.21 no.6
• /
• pp.518-523
• /
• 2015

This paper proposes a gait phase classifier using a Recurrent Neural Network (RNN). Walking is a type of dynamic system, and as such it seems that the classifier made by using a general feed forward neural network structure is not appropriate. It is known that an RNN is suitable to model a dynamic system. Because the proposed RNN is simple, we use a back propagation algorithm to train the weights of the network. The input data of the RNN is the lower body's joint angles and angular velocities which are acquired by using the lower limb exoskeleton robot, ROBIN-H1. The classifier categorizes a gait cycle as two phases, swing and stance. In the experiment for performance verification, we compared the proposed method and general feed forward neural network based method and showed that the proposed method is superior.

• Proceedings of the KIEE Conference
• /
• 1995.07b
• /
• pp.705-709
• /
• 1995

The objective of this study is to design a multi-layer neural network which controls the position of excavator's attachment. In this paper, a dynamic controller has been developed based on an error back-propagation(BP) neural network. Since the neural network can model an arbitrary nonlinear mapping, it was used as a commanded feedforward input generator. A PD feedback controller is used in parallel with the feedforward neural network to train the system. The neural network was trained by the current state of the excavator as well as the PD feedback error. By using the BP network as a feedforward controller, no a priori knowledge on system dynamics is need. Computer simulation results demonstrate such powerful characteristics of the proposed controller as adaptation to changing environment, robustness to disturbancen and performance improvement with the on-line learning in the position control of excavator attachment.

• Nuclear Engineering and Technology
• /
• v.51 no.3
• /
• pp.702-708
• /
• 2019

This paper deals with break size estimation of loss of coolant accidents (LOCA) using a nonlinear autoregressive with exogenous inputs (NARX) neural network. Previous studies used static approaches, requiring time-integrated parameters and independent firing algorithms. NARX neural network is able to directly deal with time-dependent signals for dynamic estimation of break sizes in real-time. The case studied is a LOCA in the primary system of Bushehr nuclear power plant (NPP). In this study, number of hidden layers, neurons, feedbacks, inputs, and training duration of transients are selected by performing parametric studies to determine the network architecture with minimum error. The developed NARX neural network is trained by error back propagation algorithm with different break sizes, covering 5% -100% of main coolant pipeline area. This database of LOCA scenarios is developed using RELAP5 thermal-hydraulic code. The results are satisfactory and indicate feasibility of implementing NARX neural network for break size estimation in NPPs. It is able to find a general solution for break size estimation problem in real-time, using a limited number of training data sets. This study has been performed in the framework of a research project, aiming to develop an appropriate accident management support tool for Bushehr NPP.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.9
• /
• pp.49-61
• /
• 1996

Back-propagation(BP) algorithm needs a lot of time to train the artificial neural network (ANN) to get high accuracy level in classification tasks. So there have been extensive researches to process back-propagation algorithm on parallel processors. This paper prsents a linear systolic array which calculates forward-backward propagation of BP algorithm at the same time using effective space-time transformation and PE structure. First, we analyze data flow of forwared and backward propagations and then, represent the BP algorithm into data dapendency graph (DG) which shows parallelism inherent in the BP algorithm. Next, apply space-time transformation on the DG of ANN is turn with orthogonal direction projection. By doing so, we can get a snakelike systolic array. Also we calculate the interval of input for parallel processing, calculate the indices to make the right datas be used at the right PE when forward and bvackward propagations are processed in the same PE. And then verify the correctness of output when forward and backward propagations are executed at the same time. By doing so, the proposed system maximizes parallelism of BP algorithm, minimizes th enumber of PEs. And it reduces the execution time by 2 times through making idle PEs participate in forward-backward propagation at the same time.

• Bulletin of the Korean Chemical Society
• /
• v.34 no.6
• /
• pp.1643-1650
• /
• 2013

A non linear QSAR model was constructed on a series of 57 xanthone and curcuminoide derivatives as ${\alpha}$-glucosidase inhibitors by back-propagation neural network method. The neural network architecture was optimized to obtain a three-layer neural network, composed of five descriptors, nine hidden neurons and one output neuron. A good predictive determination coefficient was obtained (${R^2}_{Pset}$ = 86.7%), the statistical results being better than those obtained with the same data set using a multiple regression analysis (MLR). As in the MLR model, the descriptor MATS7v weighted by Van der Waals volume was found as the most important independent variable on the ${\alpha}$-glucosidase inhibitory.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2001.12a
• /
• pp.133-136
• /
• 2001

This paper partially presents a Hierachical neural network architecture for providing the intelligent control of complex Automatic Transmission(AJT) system which is usually nonlinear and hard to model mathematically. It consists of the module to apply or release an engine brake at the slope and that to judge the intention of the driver. The HNN architecture simplifies the structure of the overall system and is efficient for the learning time. This paper describes how the sub-neural networks of each module have been constructed and will compare the result of the intelligent hJT control to that of the conventional shift pattern.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2001.12a
• /
• pp.267-270
• /
• 2001

In this paper, we propose the self-autonomous algorithm for mobile robot system. The proposed mobile robot system which is teamed by learning with the neural networks can trace the target at the same distances. The mobile robot can evaluate the distance between robot and target with ultrasonic sensors. By teaming the setup distance, current distance and command velocity, the robot can do intelligent self-autonomous drive. We use the neural network and back-propagation algorithm as a tool of learning. As a result, we confirm the ability of tracing the target with proposed mobile robot.

• Korean Journal of Human Ecology
• /
• v.12 no.1
• /
• pp.63-73
• /
• 2003

The purpose of this study was to quantify the hands of fabrics for the Korean folk clothes using both a KES-FB and an artificial neural network. In order to select the proper input parameters, we calculated the correlation using step-wise regression between mechanical properties and the hand value of fabrics. For the classification, the primary hand values and total hand value, five neural networks with three-layered structure were constructed using the error back propagation algorithm and, in order to reduce errors and to speed up learning, the momentum method was selected. From the analysis of the primary and total hands using a self-constructed artificial intelligence system, the error rates of sleekness, stiffness, silkiness, and roughness compared with the judgement of expert panels were found to be 3.3%, 3.3%, 1.6%, and 4.9%, respectively, while that of the total hand was 9.83%.