• 제어로봇시스템학회:학술대회논문집
• /
• 1995.10a
• /
• pp.215-218
• /
• 1995

In this paper, a sliding mode controller (SMC) which can be characterized by high accuracy, fast response and robustness is applied to speed control of AC-SERVO motor. The control input is changed to continuous one in the boundary layer to reduce the chattering phenomenon, and the boundary layer converges to zero when the state variables of system reach to steady state values. The integral compensator is added to reduce steady state error and to provide the continuous torque reference. The acceleration which is necessary to get the sliding plane is estimated by an observer. Sliding surface is included in control input to enhance the robustness and transient response without increasing sliding mode controller gain. The proposed controller is implemented by DSP(digital signal processor). The effectiveness of the proposed control scheme for speed controller is shown by the real-time experimental results in the paper.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.14 no.8
• /
• pp.658-662
• /
• 2001

This paper describes on the fabrication and characteristics of hot-film type micro-flowsensors integrated with Pt-RTD(resistance thermometer device) and micro-heater on the SOI(Si-on-insulator) membrane and trench structures, in which MGO thin-film was used as medium layer in order to improve adhesion of Pt thin-film to SiO$_2$ layer. Output voltages increased due to increase of heat-loss from sensor to external. The output voltage was 250 nV at N$_2$ flow rate of 2000 sccm/min, heating power of 0.3 W. The response time($\tau$:63%) was about 42 msec when input flow was step-input. The results indicated that micro-flowsensors with the SOI membrane and trench structures have properties of a high-resolution and ow consume power.

• Journal of Biomedical Engineering Research
• /
• v.17 no.3
• /
• pp.395-401
• /
• 1996

In this paper a new neural network model, time delayed counterpropagation neural networks (TDCPN) which have high recognition rate and short total learning time, is proposed for electromyogram(EMG) recognition. Signals the proposed model increases the recognition rates after learned the regional temporal correlation of patterns using time delay properties in input layer, and decreases the learning time by using winner-takes-all learning rule. The ouotar learning rule is put at the output layer so that the input pattern is able to map a desired output. We test the performance of this model with EMG signals collected from a normal subject. Experimental results show that the recognition rates of the suggested model is better and the learning time is shorter than those of TDNN and CPN.

• Journal of Information Processing Systems
• /
• v.13 no.5
• /
• pp.1052-1070
• /
• 2017

For the autoencoder (AE) implemented as a construction component, this paper uses the method of greedy layer-by-layer pre-training without supervision to construct the stacked autoencoder (SAE) to extract the abstract features of the original input data, which is regarded as the input of the logistic regression (LR) model, after which the click-through rate (CTR) of the user to the advertisement under the contextual environment can be obtained. These experiments show that, compared with the usual logistic regression model and support vector regression model used in the field of predicting the advertising CTR in the industry, the SAE-LR model has a relatively large promotion in the AUC value. Based on the improvement of accuracy of advertising CTR prediction, the enterprises can accurately understand and have cognition for the needs of their customers, which promotes the multi-path development with high efficiency and low cost under the condition of internet finance.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.88.5-88
• /
• 2001

We propose a neural network solver for an inverse problem. The problem is that input data with complete teaching include defects and predict the defect value. The solver is constructed of a three layer neural network whose learning method is combined from BP and reconstruction learning. The input data for the defects are unknown; therefore, the circulation of an arithmetic progression replaces them; rightly, the learning procedure is not converged for the circulation data vut for the normal data. The learning is quitted after such a learning status id kept. Then, we search a minimum of the differences between teaching data and output of the circulation. Then, we search a minimum of the ...

• The Journal of Korea Robotics Society
• /
• v.16 no.1
• /
• pp.64-71
• /
• 2021

This paper describes a robotic teleoperation system to perform an accurate needle insertion into a cornea for a separation between the stromal layer and Descemet's membrane during deep anterior lamellar Keratoplasty (DALK). The system can reduce the hand tremor of a surgeon by scaling the input motion, which is the control input of the slave robot. Moreover, we utilize corneal applanation to estimate the insertion depth. The proposed system was validated by performing the layer separation using 25 porcine eyes. The average depth of needle insertion was 742 ± 39.8 ㎛ while the target insertion depth was 750 ㎛. Tremor error was reduced from 402 ± 248 ㎛ in the master device to 28.5 ± 21.0 ㎛ in the slave robot. The rate of complete success, partial success, and failure were 60, 28, and 12%, respectively. The experimental results showed that the proposed system was able to reduce the hand tremor of surgeons and perform precise needle insertion during DALK.

• Journal of the Korean Society for Precision Engineering
• /
• v.14 no.2
• /
• pp.136-144
• /
• 1997

Usually triangular patches are used to transfer geometric shape in Rapid Prototyping CAM system. STL, a list of triangles, is de facto standard in RP industry. Because STL does not have topological infoma- tion, it can cause errornous results. So STL should be verified before using. After adding support structures to anchor the part to the platform and to prevent sagging or distortion, slicing and layer by layer manufactur- ing process are done. But triangular patch is surface model and cannot provide sufficient information on geometry in the above processes. So, geometric modeling is necessary in verifying STL, adding support structures and slicing. It is natural that triangle based modeling is the best when tringular patches are used as input. Considering support structures, solid and faces coexist in RP process. Therefore non-manifold modeler is required. In this study, triangle based non-manifold geometric modeling is proposed for RP sys- tem consistent with STL input.

• Journal of Information Processing Systems
• /
• v.19 no.6
• /
• pp.803-816
• /
• 2023

Explainable artificial intelligence is a method that explains how a complex model (e.g., a deep neural network) yields its output from a given input. Recently, graph-type data have been widely used in various fields, and diverse graph neural networks (GNNs) have been developed for graph-type data. However, methods to explain the behavior of GNNs have not been studied much, and only a limited understanding of GNNs is currently available. Therefore, in this paper, we propose an explanation method for node classification using graph convolutional networks (GCNs), which is a representative type of GNN. The proposed method finds out which features of each node have the greatest influence on the classification of that node using GCN. The proposed method identifies influential features by backtracking the layers of the GCN from the output layer to the input layer using the gradients. The experimental results on both synthetic and real datasets demonstrate that the proposed explanation method accurately identifies the features of each node that have the greatest influence on its classification.

• Journal of information and communication convergence engineering
• /
• v.6 no.2
• /
• pp.204-206
• /
• 2008

In this paper, we proposed the recognition method that automatically extracts cracks from a surface image acquired by a digital camera and recognizes the directions (horizontal, vertical, -45 degree, and 45 degree) of cracks using the fuzzy single layer perceptron. We compensate an effect of light on a concrete surface image by applying the closing operation, which is one of the morphological techniques, extract the edges of cracks by Sobel masking, and binarize the image by applying the iterated binarization technique. Two times of noise reduction are applied to the binary image for effective noise elimination. After the specific regions of cracks are automatically extracted from the preprocessed image by applying Glassfire labeling algorithm to the extracted crack image, the cracks of the specific region are enlarged or reduced to $30{\times}30$ pixels and then used as input patterns to the fuzzy single layer perceptron. The experiments using concrete crack images showed that the cracks in the concrete crack images were effectively extracted and the fuzzy single layer perceptron was effective in the recognition of the extracted cracks directions.

• 제어로봇시스템학회:학술대회논문집
• /
• 2002.10a
• /
• pp.48.1-48
• /
• 2002

The method of endowing the controller with the strengthened robustness within the boundary layer is presented for controlling the uncertain nonlinear systems in which the variations of the uncertainties are slow. From this controller, the width of the boundary layer where the robust control input is smoothened out can be given by an appropriate value but a better control performance within the boundary layer can be achieved without the control chattering because the role of adaptive control is to compensate for the uncovered portions of the robust control occurred from the continuous approximation within the boundary layer.