• Proceedings of the KIEE Conference
• /
• 1996.07b
• /
• pp.1186-1188
• /
• 1996

Neural networks and fuzzy systems have attracted the attention of many researehers recently. In general, neural networks are used to obtain information about systems from input/output observation and learning procedure. On the other hand, fuzzy systems use fuzzy rules to identify or control systems. In this paper we present a generalized FCMAC(Fuzzified Cerebellar Model Articulation Controller) networks, by integrating fuzzy systems with the CMAC(Cerebellar Model Articulation Controller) networks. We propose a direct adaptive controller design based on FCMAC(fuzzified CMAC) networks. Simulation results reveal that the proposed adaptive controller is practically feasible in nonlinear plant control.

• Journal of Welding and Joining
• /
• v.11 no.2
• /
• pp.27-41
• /
• 1993

The estimation of nugget sizes was attempted by utilizing the artificial neural networks method. Artificial neural networks is a highly simplified model of the biological nervous system. Artificial neural networks is composed of a large number of elemental processors connected like biological neurons. Although the elemental processors have only simple computation functions, because they are connected massively, they can describe any complex functional relationship between an input-output pair in an autonomous manner. The electrode head movement signal, which is a good indicator of corresponding nugget size was determined by measuring the each test specimen. The sampled electrode movement data and the corresponding nugget sizes were fed into the artificial neural networks as input-output pairs to train the networks. In the training phase for the networks, the artificial neural networks constructs a fuctional relationship between the input-output pairs autonomusly by adjusting the set of weights. In the production(estimation) phase when new inputs are sampled and presented, the artificial neural networks produces appropriate outputs(the estimates of the nugget size) based upon the transfer characteristics learned during the training mode. Experimental verification of the proposed estimation method using artificial neural networks was done by actual destructive testing of welds. The predicted result by the artifficial neural networks were found to be in a good agreement with the actual nugget size. The results are quite promising in that the real-time estimation of the invisible nugget size can be achieved by analyzing the process variable without any conventional destructive testing of welds.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.30 no.4B
• /
• pp.399-412
• /
• 2010

The goal of this research is to apply the neural networks models for the temporal disaggregation of the yearly pan evaporation (PE) data, Republic of Korea. The neural networks models consist of multilayer perceptron neural networks model (MLP-NNM) and generalized regression neural networks model (GRNNM), respectively. And, for the performances evaluation of the neural networks models, they are composed of training and test performances, respectively. The three types of data such as the historic, the generated, and the mixed data are used for the training performance. The only historic data, however, is used for the testing performance. From this research, we evaluate the application of MLP-NNM and GRNNM for the temporal disaggregation of nonlinear time series data. We should, furthermore, construct the credible monthly PE data from the temporal disaggregation of the yearly PE data, and can suggest the available data for the evaluation of irrigation and drainage networks system.

• Journal of the Korean Data and Information Science Society
• /
• v.15 no.4
• /
• pp.911-920
• /
• 2004

We try receiver operating characteristic(ROC) curves by neural networks of logistic function. The models are shown to arise from model classification for normal (diseased) and abnormal (nondiseased) groups in medical research. A few goodness-of-fit test statistics using normality curves are discussed and the performances using neural networks of logistic function are conducted.

• Journal of Electrical Engineering and Technology
• /
• v.10 no.4
• /
• pp.1822-1829
• /
• 2015

In this study, artificial neural networks were used to determine the intensity of brightness in interior spaces. The illumination elements to illuminate indoor spaces were considered, not individually, but as a system. So, during the planned maintenance periods of an illumination system, after its design and installation, simple brightness level measurements must be taken. For a three-dimensional evaluation of the brightness level in indoor spaces in a speedy and accurate manner, the obtained brightness level measurement results and artificial neural network model were used. Upon estimation of the most suitable brightness level for indoor spaces by using the artificial neutral network model, the energy demands required by the illumination elements decreased. Consequently, in this study, with estimations of brightness levels, the extent to which the artificial neutral networks become successful was observed and more correct results have been obtained in terms of both economy and usage.

• Journal of the Korean Society for Aviation and Aeronautics
• /
• v.21 no.3
• /
• pp.1-8
• /
• 2013

An application of adaptive flight controller is required for the non-linear and high uncertain system that configuration of tiltrotor aircraft is dramatically changed from rotary wing mode to fixed wing mode. In this paper, the applicable adaptive controller for the tiltrotor aircraft was designed using Neural Networks and DMI (Dynamic Model Inversion). The performance of the SCAS (Stability and Control Augmentation System) was simulated against manned military specification, using the fullscale model of 'Smart UAV(Unmanned Aerial Vehicle)' developed by Korea Aerospace Research Institute. And Neural Networks based adaptive controller was verified through its whole operating envelope using the established HQ (Handling Quality) criteria.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.26 no.4
• /
• pp.263-279
• /
• 2022

This paper presents an automatic inspection of defects in semiconductor images. We devise a statistical method to find defects on homogeneous background from the observation that it has a log-normal distribution. If computer aided design (CAD) data is available, we use it to construct a signed distance function (SDF) and change the pixel values so that the average of pixel values along the level curve of the SDF is zero, so that the image has a homogeneous background. In the absence of CAD data, we devise a hybrid method consisting of a model-based algorithm and two neural networks. The model-based algorithm uses the first right singular vector to determine whether the image has a linear or complex structure. For an image with a linear structure, we remove the structure using the rank 1 approximation so that it has a homogeneous background. An image with a complex structure is inspected by two neural networks. We provide results of numerical experiments for the proposed methods.

• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.485-485
• /
• 2000

A design method of model following control system using neural networks is proposed. An unknown nonlinear single-input single-output plant is identified using a multilayer neural networks. A linear controller is designed fer the linear approximation model obtained by linearinzing the identification model. The identification model is also used as a plant emulator to obtain the prediction error. Deficient servo performance due to controlling nonlinear plant with only linear controller is mended by adjusting the linear controller output using the prediction output and the parameters of the identification model. An optimal preview controller is adopted as the linear controller by reason of having good servo performance lowering the peak of control input. Validity of proposed method is illustrated through a numerical simulation.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Proceedings of the KIEE Conference
• /
• 2005.10b
• /
• pp.322-324
• /
• 2005

In this paper, we introduce the advanced Self-Organizing Fuzzy Polynomial Neural Network based on optimized FPN by evolutionary algorithm and discuss its comprehensive design methodology involving mechanisms of genetic optimization, especially genetic algorithms (GAs). The proposed model gives rise to a structurally and parametrically optimized network through an optimal parameters design available within Fuzzy Polynomial Neuron(FPN) by means of GA. Through the consecutive process of such structural and parametric optimization, an optimized and flexible the proposed model is generated in a dynamic fashion. The performance of the proposed model is quantified through experimentation that exploits standard data already used in fuzzy modeling. These results reveal superiority of the proposed networks over the existing fuzzy and neural models.