• IEMEK Journal of Embedded Systems and Applications
• /
• v.10 no.2
• /
• pp.65-71
• /
• 2015

It has been known that GMM (Gaussian Mixture Model) based speaker identification systems using ML (Maximum Likelihood) and WMR (Weighting Model Rank) demonstrate very high performances. However, such systems are not so effective under practical environments, in terms of real time processing, because of their high calculation costs. In this paper, we propose a new speaker-pruning algorithm that effectively reduces the calculation cost. In this algorithm, we select 20% of speaker models having higher likelihood with a part of input speech and apply MWMR (Modified Weighted Model Rank) to these selected speaker models to find out identified speaker. To verify the effectiveness of the proposed algorithm, we performed speaker identification experiments using TIMIT database. The proposed method shows more than 60% improvement of reduced processing time than the conventional GMM based system with no pruning, while maintaining the recognition accuracy.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.49 no.8
• /
• pp.407-414
• /
• 2000

This paper addresses the analysis and design of fuzzy control systems for a class of complex uncertain single-input single-output nonlinear systems. The proposed method represents the nonlinear system using a Takagi-Cugeno fuzzy model and construct a global fuzzy logic controller by blending all local state feedback controllers with a sliding mode controller. Unlike the commonly used parallel distributed compensation technique, we can design a global stable fuzzy controller without finding a common Lyapunov function for all local control systems, and can obtain good tracking performance by using sliding mode control theory. Furthermore, stability analysis is carried out not for the fuzzy model but for the real nonlinear system with uncertainties. Duffing forced oscillation sysmte is used as an example to show the effectiveness and feasibility of the proposed method.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.8 no.3
• /
• pp.80-87
• /
• 1998

This paper presents and online scheme for fuzzy modelling of nonlinear systems, based on the model adjustment technique and the genetic algorithm technique. The fuzzy model is characterized by fuzzy "if-then" rules which represent locally linear input-output relations whose consequence parts are defined as subsystems of a nonlinear sysem. The discrete-time model for each subsystem is obtained to deal with initalization and unmeasurable signal problems in online estimation and the final output of the fuzzy model is computed from the outputs of the discrete-time models. Then, the parameters of both the premise and consequence parts of the fuzzy model are adjusted by a genetic algorithm. A set of simulation works is carried out to demonstrate the effectiveness of the proposed method.ed method.

• Proceedings of the KSME Conference
• /
• 2001.06b
• /
• pp.335-340
• /
• 2001

It is important to model the mechanical structure precisely and reasonably in predicting the dynamic characteristics, controlling the vibration, and designing the structure dynamics. In the finite element modeling, the errors can be contained from the physical parameters, the approximation of the boundary conditions, and the element modeling. From the dynamic test, more precise dynamic characteristics can be obtained. Model updating using parameter modification is appropriate when the design parameter is used to analyze the input parameter like finite element method. Finite element analysis for cantilever and simply supported beams with uniform area and shape change are carried out as model updating examples. Mass and stiffness matrices are updated by comparing test and analytical modal frequencies. The result shows that the updated frequencies become closer to the test frequencies.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2001.11b
• /
• pp.1260-1265
• /
• 2001

It is important to model the mechanical structure precisely and reasonably in predicting the dynamic characteristics, controlling the vibration, and designing the structural dynamics. In the finite element modeling, the errors can be contained from the physical parameters, the approximation of the boundary conditions, and the element modeling, From the dynamic test. more precise dynamic characteristics can be obtained. Model updating using parameter modification is appropriate when the design parameter is used to analyze the input parameter like finite element method. Finite element analysis for free-free-free-free(FFFF) and clamped-free-free-free(CFFF) plate with uniform area and shape change are carried out as model updating examples, Mass and stiffness matrices are updated by comparing test and analytical modal frequencies. The result shows that the updated frequencies become closer to the test frequencies.

• Journal of Korean Institute of Industrial Engineers
• /
• v.40 no.1
• /
• pp.84-99
• /
• 2014

In this study, determinant input-output variables are identified for calculating Data Envelopment Analysis (DEA) efficiency scores relating to evaluating the efficiency of government-sponsored research and development (R&D) projects. In particular, this study proposes a systematic framework of design and analysis of experiments, called "all possible DEAs", for pinpointing DEA determinant input-output variables. In addition to correlation analyses, two modified measures of time series analysis are developed in order to check the similarities between a DEA complete data structure (CDS) versus the rest of incomplete data structures (IDSs). In this empirical analysis, a few DEA determinant input-output variables are found to be associated with a typical public R&D performance evaluation logic model, especially oriented to a mid- and long-term performance perspective. Among four variables, only two determinants are identified : "R&D manpower" ($x_2$) and "Sales revenue" ($y_1$). However, it should be pointed out that the input variable "R&D funds" ($x_1$) is insignificant for calculating DEA efficiency score even if it is a critical input for measuring efficiency of a government-sonsored R&D project from a practical point of view a priori. In this context, if practitioners' top priority is to see the efficiency between "R&D funds" ($x_1$) and "Sales revenue" ($y_1$), the DEA efficiency score cannot properly meet their expectations. Therefore, meticulous attention is required when using the DEA application for public R&D performance evaluation, considering that discrepancies can occur between practitioners' expectations and DEA efficiency scores.

• Proceedings of the Korean Geotechical Society Conference
• /
• 2004.03b
• /
• pp.714-721
• /
• 2004

The purpose of this study is to find out the similitude laws for dissipation velocity of excess pore pressure after liquefaction according to magnitude of input accelerations and height of model soils from the results of impulse load tests. In impulse load tests, model soils were constructed to the height of 25cm, 50cm, and 100cm in acrylic tubes whose inside diameters were 19cm and 38cm respectively, and impulse loads were applied at the bottom of each model soil to liquefy the entire model soil. Excess pore pressure distribution by depth and settlement of soil surface were measured in each test. Dissipation curves of excess pore pressure measured in each tests were simulated by solidification theory, and dissipation velocities of excess pore pressure were determined from the slope of simulated dissipation curves. From the results of impulse load tests, dissipation velocity of excess pore pressure was not affected by magnitude of input acceleration, and from this fact, dissipation process was proved to be different from dynamic phenomenon. However, dissipation velocity of excess pore pressure increased as height of model soil increased and showed little difference as diameter of model soil increased. Therefore, the similitude law for dissipation velocity could be expressed by the similitude law for model height to 0.2 without regard to the diameter of model soil.

• Smart Structures and Systems
• /
• v.14 no.6
• /
• pp.1173-1196
• /
• 2014

Real-time dynamic substructuring tests have been conducted on a cable-deck system. The cable is representative of a full scale cable for a cable-stayed bridge and it interacts with a deck, numerically modelled as a single-degree-of-freedom system. The purpose of exciting the inclined cable at the bottom is to identify its nonlinear dynamics and to mark the stability boundary of the semi-trivial solution. The latter physically corresponds to the point at which the cable starts to have an out-of-plane response when both input and previous response were in-plane. The numerical and the physical parts of the system interact through a transfer system, which is an actuator, and the input signal generated by the numerical model is assumed to interact instantaneously with the system. However, only an ideal system manifests a perfect correspondence between the desired signal and the applied signal. In fact, the transfer system introduces into the desired input signal a delay, which considerably affects the feedback force that, in turn, is processed to generate a new input. The effectiveness of the control algorithm is measured by using the synchronization technique, while the online adaptive forward prediction algorithm is used to compensate for the delay error, which is present in the performed tests. The response of the cable interacting with the deck has been experimentally observed, both in the presence of delay and when delay is compensated for, and it has been compared with the analytical model. The effects of the interface delay in real-time dynamic substructuring tests conducted on the cable-deck system are extensively discussed.

• Proceedings of the KIEE Conference
• /
• 2002.07d
• /
• pp.2821-2823
• /
• 2002

In this paper we propose the Multi-FNN (Fuzzy-Neural Networks) for optimal identification modeling of complex system. The proposed Multi-FNNs is based on a concept of FNNs and exploit linear inference being treated as generic inference mechanisms. In the networks learning, backpropagation(BP) algorithm of neural networks is used to updata the parameters of the network in order to control of nonlinear process with complexity and uncertainty of data, proposed model use a HCM(Hard C-Means)clustering algorithm which carry out the input-output dat a preprocessing function and Genetic Algorithm which carry out optimization of model The HCM clustering method is utilized to determine the structure of Multi-FNNs. The parameters of Multi-FNN model such as apexes of membership function, learning rates, and momentum coefficients are adjusted using genetic algorithms. An aggregate performance index with a weighting factor is proposed in order to achieve a sound balance between approximation and generalization abilities of the model. NOx emission process data of gas turbine power plant is simulated in order to confirm the efficiency and feasibility of the proposed approach in this paper.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.17 no.8
• /
• pp.779-788
• /
• 1992

The Input /output (I /0) subsystem is often the bottleneck in high performance computer system. Generally, system performance evaluation models were enhanced to include the effect of the I/0 system. In this paper, we modeled the terminal servers which are Indispensable devices In distribution of computer resources. We use M /M /1 Queueing model for find out the point of the system performance FIFO buffer sizes In the terminal server arc the Important fanctions of the system design and could be effected to the overall system functions. We have proposed the of optimal buffer sizes in the model of terminal server for increasing the system performance. We analizing the vatting time for terruanl server using Queueing model. and We find out the reference model result from simulation.