• The Transactions of the Korean Institute of Electrical Engineers B
• /
• v.53 no.3
• /
• pp.188-195
• /
• 2004

This paper presents neural load torque observer that is used to deadbeat load torque observer and gain compensation by parameter estimator As a result, the response of the PMSM(permanent magnet synchronous motor) follows that nominal plant. The load torque compensation method is composed of a neural deadbeat observer To reduce the noise effect, the post-filter implemented by MA(moving average) process, is adopted. The parameter compensator with RLSM (recursive least square method) parameter estimator is adopted to increase the performance of the load torque observer and main controller The parameter estimator is combined with a high performance neural load torque observer to resolve the problems. The neural network is trained in on-line phases and it is composed by a feed forward recall and error back-propagation training. During the normal operation, the input-output response is sampled and the weighting value is trained multi-times by error back-propagation method at each sample period to accommodate the possible variations in the parameters or load torque. As a result, the proposed control system has a robust and precise system against the load torque and the Parameter variation. A stability and usefulness are verified by computer simulation and experiment.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.1 no.1
• /
• pp.87-94
• /
• 2001

Silicon nitride films grown by plasma-enhanced chemical vapor deposition (PECVD) are useful for a variety of applications, including anti-reflecting coatings in solar cells, passivation layers, dielectric layers in metal/insulator structures, and diffusion masks. PECVD systems are controlled by many operating variables, including RF power, pressure, gas flow rate, reactant composition, and substrate temperature. The wide variety of processing conditions, as well as the complex nature of particle dynamics within a plasma, makes tailoring SiN film properties very challenging, since it is difficult to determine the exact relationship between desired film properties and controllable deposition conditions. In this study, SiN PECVD modeling using optimized neural networks has been investigated. The deposition of SiN was characterized via a central composite experimental design, and data from this experiment was used to train and optimize feed-forward neural networks using the back-propagation algorithm. From these neural process models, the effect of deposition conditions on film properties has been studied. A recipe synthesis (optimization) procedure was then performed using the optimized neural network models to generate the necessary deposition conditions to obtain several novel film qualities including high charge density and long lifetime. This optimization procedure utilized genetic algorithms, hybrid combinations of genetic algorithm and Powells algorithm, and hybrid combinations of genetic algorithm and simplex algorithm. Recipes predicted by these techniques were verified by experiment, and the performance of each optimization method are compared. It was found that the hybrid combinations of genetic algorithm and simplex algorithm generated recipes produced films of superior quality.

• Journal of the Korean Data and Information Science Society
• /
• v.20 no.5
• /
• pp.829-835
• /
• 2009

Parameter design is an approach to reducing performance variation of quality characteristic value in products and processes. Taguchi has used the signal-to-noise ratio to achieve the appropriate set of operating conditions where variability around target is low in the Taguchi parameter design. However, there are difficulties in practical application, such as complexity and nonlinear relationships among quality characteristics and control factors (design factors), and interactions occurred among control factors. Neural networks have a learning capability and model free characteristics. There characteristics support neural networks as a competitive tool in processing multivariable input-output implementation. In this paper we propose a substantially simpler optimization procedure for parameter design using neural network. An example is illustrated to compare the difference between the Taguchi method and neural network method.

• International Journal of Advanced Culture Technology
• /
• v.4 no.2
• /
• pp.47-56
• /
• 2016

In transformer cooling systems, oil temperature is controlled through the use of a blower and oil pump. For this paper, set-point algorithms, a reset algorithm and control algorithms of the cooling system were developed by neural networks and fuzzy logics. The oil inlet temperature was set by a $2{\times}2{\times}1$ neural network, and the oil temperature difference was set by a $2{\times}3{\times}1$ neural network. Inputs used for these neural networks were the transformer operating ratio and the air inlet temperature. The inlet set temperature was reset by a fuzzy logic based on the transformer operating ratio and the oil outlet temperature. A blower was used to control the inlet oil temperature while the oil pump was used to control the oil temperature difference by fuzzy logics. In order to analysis the performance of these algorithms, the initial start-up test and the step change test were performed by using the dynamic model of a transformer cooling system. Test results showed that algorithms developed for this study were effective in controlling the oil temperature of a transformer cooling system.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.6 no.6
• /
• pp.819-826
• /
• 2002

Cellular neural networks(CNN) are neural networks that have locally connected characteristics and real-time image processing. Locally connected characteristics are suitable for VLSI implementation. It also has applications in such areas as image processing and pattern recognition. In this thesis cellular neural networks are used for feature detection in number recognition at the stage of re-processing. The four or six directional shadow detectors are used in numbers recognition. At the stage of classification, this result of feature detection was simulated by using a multi-layer back Propagation neural network. The experiments indicate that the CNN feature detectors capture good features for number recognition tasks.

• The Transactions of the Korea Information Processing Society
• /
• v.5 no.1
• /
• pp.119-127
• /
• 1998

In this paper, the design of a VLSI-based multilayer neural network is presented, which can be used as a dedicated hardware for character-type segmentation and character-element recogniti on consuming large processing time in conventional software-based Hangul printed-character recognition systems. Also the architecture and its design of a neural coprocessor interfacing the neural network with a host computcr and controlling thc neural network are presented. The architecture, behavior, and performance of the proposed neural coprocessor are justified using VHDL modeling and simulation. Experimental results show the successful rates of character-type segmentation and character-element recognition is competitive to those of software-based Hangul printed-character recognition systems with retaining high-speed.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2013.10a
• /
• pp.320-324
• /
• 2013

A crack identification method using an equivalent bending stiffness for cracked beam and committee of neural networks is presented. The equivalent bending stiffness is constructed based on an energy method for a straight thin-walled pipe, which has a through-the-thickness crack, subjected to bending. Several numerical analysis for a steel cantilever pipe using the equivalent bending stiffness are carried out to extract the natural frequencies and mode shapes of the cracked beam. The extracted modal properties are used in constructing a training patterns of a neural network. The input to the neural network consists of the modal properties and the output is composed of the crack location and size. Multiple neural networks are constructed and each individual network is trained independently with different initial synaptic weights. Then, the estimated crack locations and sizes from different neural networks are averaged. Experimental crack detection is carried out for 3 damage cases using the proposed method, and the identified crack locations and sizes agree reasonably well with the exact values.

• Nuclear Engineering and Technology
• /
• v.56 no.10
• /
• pp.4149-4157
• /
• 2024

Welding processes are used to connect several components in nuclear power plants. These welding processes can induce residual stress in welding joints, which has been identified as a significant factor in primary water stress corrosion cracking. Consequently, the assessment of welding residual stress plays a crucial role in determining the structural integrity of welded joints. In this study, a deep fuzzy neural networks (DFNN) with a rule-dropout method, which is an artificial intelligence (AI) method, was used to predict the residual stress of dissimilar metal welding. ABAQUS, a finite element analysis program, was used as the data collection tool to develop the AI model, and 6300 data instances were collected under 150 analysis conditions. A rule-dropout method and genetic algorithm were used to optimize the estimation performance of the DFNN model. DFNN with the rule-dropout model was compared to a deep neural network method, known as a general deep learning method, to evaluate the estimation performance of DFNN. In addition, a fuzzy neural network method and a cascaded support vector regression method conducted in previous studies were compared. Consequently, the estimation performance of the DFNN with the rule-dropout model was better than those of the comparison methods. The welding residual stress estimation results of this study are expected to contribute to the evaluation of the structural integrity of welded joints.

• The Journal of the Convergence on Culture Technology
• /
• v.6 no.3
• /
• pp.463-468
• /
• 2020

Artificial intelligence researches are being applied and developed in various fields. In this paper, we build a neural network by using the method of implementing artificial intelligence in the form of spiking natural networks (SNN), the next-generation of artificial intelligence research, and analyze how the number of neurons in that neural networks affect the performance of the neural networks. We also analyze how the performance of neural networks changes while increasing the amount of neural network learning. The findings will help optimize SNN-based neural networks used in each field.

• Journal of KIISE
• /
• v.44 no.7
• /
• pp.668-673
• /
• 2017

Deep neural networks have shown remarkable performance in various fields of pattern recognition such as voice recognition, image recognition and object detection. However, underlying mechanisms of the network have not been fully revealed. In this paper, we focused on empirical analysis of the network parameters. The Faster R-CNN(region-based convolutional neural network) was used as a baseline network of our work and three important parameters were analyzed: the dropout ratio which prevents the overfitting of the neural network, the size of the anchor boxes and the activation function. We also compared the performance of dropout and batch normalization. The network performed favorably when the dropout ratio was 0.3 and the size of the anchor box had not shown notable relation to the performance of the network. The result showed that batch normalization can't entirely substitute the dropout method. The used leaky ReLU(rectified linear unit) with a negative domain slope of 0.02 showed comparably good performance.