• Earthquakes and Structures
• /
• 제14권5호
• /
• pp.425-436
• /
• 2018

The design of seismically isolated structures considering the stochastic nature of excitations, base isolators' design parameters, and superstructure properties requires robust reliability analysis methods to calculate the failure probability of the entire system. Here, by applying artificial neural networks, we proposed a robust technique to accelerate the estimation of failure probability of equipped isolated structures. A three-story isolated building with susceptible facilities is considered as the analytical model to evaluate our technique. First, we employed a sensitivity analysis method to identify the critical sources of uncertainty. Next, we calculated the probability of failure for a particular set of random variables, performing Monte Carlo simulations based on the dynamic nonlinear time-history analysis. Finally, using a set of designed neural networks as a surrogate model for the structural analysis, we assessed once again the probability of the failure. Comparing the obtained results demonstrates that the surrogate model can attain precise estimations of the probability of failure. Moreover, our proposed approach significantly increases the computational efficiency corresponding to the dynamic time-history analysis of the structure.

• 대한임베디드공학회논문지
• /
• 제17권4호
• /
• pp.217-228
• /
• 2022

Currently, there are increasing demands for applying deep neural networks (DNNs) in the embedded domain such as classification and object detection. The DNN processing in embedded domain often requires custom hardware such as NPU for acceleration due to the constraints in power, performance, and area. Processing DNN models requires a large amount of data, and its seamless transfer to NPU is crucial for performance. In this paper, we developed a cycle-accurate NPU simulator to evaluate diverse NPU microarchitectures. In addition, we propose a novel technique for reducing the number of memory accesses when processing convolutional layers in convolutional neural networks (CNNs) on the NPU. The main idea is to reuse data with memory interleaving, which recycles the overlapping data between previous and current input windows. Data memory interleaving makes it possible to quickly read consecutive data in unaligned locations. We implemented the proposed technique to the cycle-accurate NPU simulator and measured the performance with LeNet-5, VGGNet-16, and ResNet-50. The experiment shows up to 2.08x speedup in processing one convolutional layer, compared to the baseline.

• 대한산업공학회지
• /
• 제27권3호
• /
• pp.305-314
• /
• 2001

In this paper, we construct an automatic evaluation system of psychological feeling for corporate identity (CI) symbol mark based on a fuzzy neural network technique. The system is modelled by trainable fuzzy inference rules with several input variables (qualitative and quantitative design components of CI symbol mark) and a single output variable (consumer's feeling). The back propagation learning algorithm, which is a conventional learning method of multilayer feedforward neural networks, is used for parameter identification of the fuzzy inference system. The learning ability to train data and the generalization ability to test data are evaluated for the proposed evaluation system by computer simulations.

• International Journal of Control, Automation, and Systems
• /
• 제2권1호
• /
• pp.92-99
• /
• 2004

In this paper, the decoupled neural network reference compensation technique (DRCT) is applied to the control of a two degrees-of-freedom inverted pendulum mounted on an x-y table. Neural networks are used as auxiliary controllers for both the x axis and y axis of the PD controlled inverted pendulum. The DRCT method known to compensate for uncertainties at the trajectory level is used to control both the angle of a pendulum and the position of a cart simultaneously. Implementation of an on-line neural network learning algorithm has been implemented on the DSP board of the dSpace DSP system. Experimental studies have shown successful balancing of a pendulum on an x-y plane and good position control under external disturbances as well.

• ETRI Journal
• /
• 제41권6호
• /
• pp.760-770
• /
• 2019

Based on the growing demand for neural network technologies, various neural network inference engines are being developed. However, each inference engine has its own neural network storage format. There is a growing demand for standardization to solve this problem. This study presents interworking techniques for ensuring the compatibility of neural networks and data among the various deep learning frameworks. The proposed technique standardizes the graphic expression grammar and learning data storage format using the Neural Network Exchange Format (NNEF) of Khronos. The proposed converter includes a lexical, syntax, and parser. This NNEF parser converts neural network information into a parsing tree and quantizes data. To validate the proposed system, we verified that MNIST is immediately executed by importing AlexNet's neural network and learned data. Therefore, this study contributes an efficient design technique for a converter that can execute a neural network and learned data in various frameworks regardless of the storage format of each framework.

• Journal of Advanced Marine Engineering and Technology
• /
• 제28권1호
• /
• pp.90-97
• /
• 2004

DC motor requires a rotor speed sensor for accurate speed control. The speed sensors such as resolvers and encoders are used as speed detectors. but they increase cost and size of the motor and restrict the industrial drive applications. So in these days. many Papers have reported on the sensorless operation or DC motor(3)-(5). This paper Presents a new sensorless strategy using neural networks(6)-(8). Neural network structure has three layers which are input layer. hidden layer and output layer. The optimal neural network structure was tracked down by trial and error and it was found that 4-16-1 neural network has given suitable results for the instantaneous rotor speed. Also. learning method is very important in neural network. Supervised learning methods(8) are typically used to train the neural network for learning the input/output pattern presented. The back-propagation technique adjusts the neural network weights during training. The rotor speed is gained by weights and four inputs to the neural network. The experimental results were found satisfactory in both the independency on machine parameters and the insensitivity to the load condition.

• 보건행정학회지
• /
• 제13권3호
• /
• pp.21-34
• /
• 2003

Patient revisit to used hospital is a key factor in determining a health care organization's competitive advantage and survival. This article examines the relationship between customer's satisfaction and his/her revisit associated with three different methods which are the Chi Square Automatic Interaction Detection(CHAID) for segmenting the outpatient group, logistic regression and neural networks for addressing the outpatient's revisit. The main findings indicate that the important factors on outpatient's revisit are physician's kindness, nurse's skill, overall level of satisfaction, hospital reputation, recommendation, level of diagnoses and outpatient's age. Among these ones, physician's kindness is the most important factor as guidelines for decision of their revisit. The decision maker of hospital should select the strategy containing the variable amount of the level of revisit and size of outpatient's group under the constraint on the hospital's time, budget and manpower given. Finally, this study shows that neural networks, as non-parametric technique, appear to more correctly predict revisit than does logistic regression as a parametric estimation technique.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2005년도 ICCAS
• /
• pp.1637-1640
• /
• 2005

This paper presents a technique for electromagnetic field analysis on surge response due to Mid-span back-flashovers effects in lightning protection system of 500 kV EHV single circuit transmission tower by the neural networks method. These analyses are based on modeling lightning return stroke as well as on coupling the electromagnetic fields of the stroke channel to the line. The ground conductivity influences both the electric field as well as the coupling mechanism and hence the magnitude and wave shape of the induced voltage. The technique can be used to analyzed the corona voltage effect, the effective of stroke to the span tower, the surge impedance of transmission lines. The maximum voltage from flashovers effects in the lines. The model is compatible with general electromagnetic transients programs such as the ATP-EMTP. The simulation results show that this study analyses for time-domain with those produced by a cascade multi-section model, the surge impedance of a full-sized tower hit directly by a lightning stroke is discussed.

• Wind and Structures
• /
• 제8권3호
• /
• pp.147-161
• /
• 2005

Two identical tall building models with square cross-sections are experimentally studied in a wind tunnel with high-frequency-force-balance (HFFB) technique to investigate the interference effects on wind loads and dynamic responses of the interfered building. Another wind tunnel test, in which the interfered model is an aeroelastic one, is also carried out to further study the interference effects. The results from the two kinds of tests are compared with each other. Then the influences of turbulence in oncoming wind on dynamic interference factors are analyzed. At last the artificial neural networks method is used to deal with the experimental data and the along-wind and across-wind dynamic interference factor $IF_{dx}$ & $IF_{dy}$ contour maps are obtained, which could be used as references for wind load codes of buildings.

• 한국전기전자재료학회논문지
• /
• 제14권3호
• /
• pp.241-245
• /
• 2001

In this paper, we present parameter optimization technique for GaAs/AlGaAs multiple quantum well avalanche photodiodes used for image capture mechanism in high-definition system. Even under flawless environment in semiconductor manufacturing process, random variation in process parameters can bring the fluctuation to device performance. The precise modeling for this variation is thus required for accurate prediction of device performance. The precise modeling for this variation is thus required for accurate prediction of device performance. This paper will first use experimental design and neural networks to model the nonlinear relationship between device process parameters and device performance parameters. The derived model was then put into genetic algorithms to acquire optimized device process parameters. From the optimized technique, we can predict device performance before high-volume manufacturign, and also increase production efficiency.