• 한국해양공학회지
• /
• 제18권2호
• /
• pp.33-38
• /
• 2004

All adaptive neural network controller has been developed for a model of an underwater vehicle. This controller combines a radial basis neural network and sliding mode control techniques. No prior off-line training phase is required, and this scheme exploits the advantages of both neural network control and sliding mode control. An on-line stable adaptive law is derived using Lyapunov theory. The number of neurons and the width of Gaussian function should be chosen carefully. Performance of the controller is demonstrated through computer simulation.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제14권1호
• /
• pp.17-25
• /
• 2014

This paper presents the formulation of an impedance controller for regulating the contact force with the environment. To achieve an accurate force tracking control, uncertainties in both robot dynamics and the environment require to be addressed. As part of the framework of the proposed force tracking formulation, a neural network is introduced at the desired trajectory to compensate for all uncertainties in an on-line manner. Compensation at the input trajectory leads to a remarkable structural advantage in that no modifications of the internal force controllers are required. Minimizing the objective function of the training signal for a neural network satisfies the desired force tracking performance. A neural network actually compensates for uncertainties at the input trajectory level in an on-line fashion. Simulation results confirm the position and force tracking abilities of a robot manipulator.

• 한국지진공학회:학술대회논문집
• /
• 한국지진공학회 2003년도 춘계 학술발표회논문집
• /
• pp.433-441
• /
• 2003

In this paper, a study on the nonlinear modeling of lead rubber bearings(LRBs) by a neural network theory was carried out. The random tests on the LRB were used for a training of neural network model. Numerical simulations using the neural network model were peformed on a scaled structural model with the LRBs excited by three type of seismic loads and compared with the shaking table tests. As a result, it was shown that the neural network model would be useful to a numerical modeling of LRB.

• 한국정밀공학회지
• /
• 제12권12호
• /
• pp.100-108
• /
• 1995

The In-process minitoring of the chatter vibration is necessarily required to an automatic manufacturing system. In this study, we constructed a multi-sensing system using tool dynamoneter, accelerometer and AE(Acoustic Emission) sensor for a more credible detection of chatter vibration. And a new approach using a multiple neural network to extract the features of multi-sensor for the recognition chatter vibration is proposed. With the Back-propagation training process, the neural network memorize and classify the features of multi-sensor signals. As a result, it is shown by multiple neural network that the chatter vibration can be monitored accurately, and it can be widely used in practical unmanned system.

• 한국지반공학회논문집
• /
• 제18권5호
• /
• pp.37-42
• /
• 2002

인공신경망은 복잡한 상호관계를 가지는 문제의 해결을 위한 효과적인 컴퓨터 테크닉으로써 많은 분야에 활발히 활용되고 있다. 본 논문에서는 지반의 액상화 가능성을 판별하기 위하여 인공신경망 이론을 사용하였으며, 이를 위하여 반복삼축압축시험 결과와 토성자료, 지반조사자료 등을 학습인자로 사용하였다. 학습과 검증에 서해안지역의 43개의 반복삼축압축시험 데이터가 사용되었다. 여기서 인공신경망의 학습은 예측된 CSR과 실측한 CSR 사이의 오차가 적어지도록 신경망의 가중치를 수정하는 것으로 이루어진다. 전체 신경망에 대한 평균제곱의 오차가 허용치 이내로 감소할 때까지 학습은 반복되어 진행되며 일반적으로 15,000 이상의 학습이 요구되는 것으로 나타났다. 다양한 노드수를 가지는 신경망에 대한 학습을 수행한 결과, 1번째 은닉층의 수가 20개이고 2번째 은닉층의 수가 10개인 신경망이 72~98%에 해당되는 정밀도를 가지고 해당 전단변형률과 반복횟수에서의 CSR값을 예측할 수 있었다. 여기서 NOC(Number of Cycle)와$D_10$, ($N_1$)$_60$ 등의 입력변수가 지반의 액상화 거동에 주요한 영향인자로 나타났다. 연구결과 인공신경망을 이용한 지반의 액상화 거동의 예측이 비교적 정확하게 산정됨을 알 수 있었으며, CSR과 ($N_1$)$_60$, NOC와의 관계가 기존의 연구 결과에 부합하여 나타남을 알 수 있었다.

• Journal of Electrical Engineering and Technology
• /
• 제13권6호
• /
• pp.2468-2478
• /
• 2018

We propose the Sparse Feature Convolutional Neural Network (SFCNN) to reduce the volume of convolutional neural networks (CNNs). Despite the superior classification performance of CNNs, their enormous network volume requires high computational cost and long processing time, making real-time applications such as online-training difficult. We propose an advanced network that reduces the volume of conventional CNNs by producing a region-based sparse feature map. To produce the sparse feature map, two complementary region-based value extraction methods, cluster max extraction and local value extraction, are proposed. Cluster max is selected as the main function based on experimental results. To evaluate SFCNN, we conduct an experiment with two conventional CNNs. The network trains 59 times faster and tests 81 times faster than the VGG network, with a 1.2% loss of accuracy in multi-class classification using the Caltech101 dataset. In vehicle classification using the GTI Vehicle Image Database, the network trains 88 times faster and tests 94 times faster than the conventional CNNs, with a 0.1% loss of accuracy.

• 한국정보전자통신기술학회논문지
• /
• 제6권4호
• /
• pp.243-248
• /
• 2013

In this thesis, we have designed the indirect adaptive controller using Dynamic Neural Units(DNU) for unknown nonlinear systems. Proposed indirect adaptive controller using Dynamic Neural Unit based upon the topology of a reverberating circuit in a neuronal pool of the central nervous system. In this thesis, we present a genetic DNU-control scheme for unknown nonlinear systems. Our method is different from those using supervised learning algorithms, such as the backpropagation (BP) algorithm, that needs training information in each step. The contributions of this thesis are the new approach to constructing neural network architecture and its training.

• 대한전기학회논문지:전력기술부문A
• /
• 제48권10호
• /
• pp.1231-1238
• /
• 1999

This paper presents an approach for the optimal heating load Identification using Diagonal Recurrent Neural Networks(DRNN). In this paper, the DRNN captures the dynamic nature of a system and since it is not fully connected, training is much faster than a fully connected recurrent neural network. The architecture of DRNN is a modified model of the fully connected recurrent neural network with one hidden layer. The hidden layer is comprised of self-recurrent neurons, each feeding its output only into itself. In this study, A dynamic backpropagation (DBP) with delta-bar-delta learning method is used to train an optimal heating load identifier. Delta-bar-delta learning method is an empirical method to adapt the learning rate gradually during the training period in order to improve accuracy in a short time. The simulation results based on experimental data show that the proposed model is superior to the other methods in most cases, in regard of not only learning speed but also identification accuracy.

• Nuclear Engineering and Technology
• /
• 제33권5호
• /
• pp.457-467
• /
• 2001

It is well known that the performance of a fuzzy neural network strongly depends on the input features selected for its training. In its applications to sensor signal estimation, there are a large number of input variables related with an output As the number of input variables increases, the training time of fuzzy neural networks required increases exponentially. Thus, it is essential to reduce the number of inputs to a fuzzy neural network and to select the optimum number of mutually independent inputs that are able to clearly define the input-output mapping. In this work, principal component analysis (PCA), genetic algorithms (CA) and probability theory are combined to select new important input features. A proposed feature selection method is applied to the signal estimation of the steam generator water level, the hot-leg flowrate, the pressurizer water level and the pressurizer pressure sensors in pressurized water reactors and compared with other input feature selection methods.

• 전자공학회논문지B
• /
• 제33B권7호
• /
• pp.173-182
• /
• 1996

In this paper, we propose the parallel, self-organizing, hierarchical neural netowrks as a handwritten digit recognition system. This system can absorb the various shape variations of handwritten digits by using the different methods of extracting the features in each stage neural network (SNN) of the PSHNN, and can reduce training time by using the single layer neural network as the SNN, and can obtain high rate of correct recognition by using the certainty area in all the output nodes individually. experiments have been performed with NIST database. In which we use 21, 315 digits (10, 625 digits for training and 10,663 digits for testing). The results show that the correct rate is 97.48% the error rate is 1.72% and the reject rate is 0.78%.