• Journal of the Korean Institute of Intelligent Systems
• /
• v.6 no.3
• /
• pp.63-72
• /
• 1996

In this paper, we propose a design technique for a neural network controller and apply it to power system stabilization. Using a learning algorithm of error back propagation that accepts error and change of error as inputs, the momentum learning technique is used by which reduction of learning time is possible for real time control. The related simulation results show that the proposed control techinque is more powerful than the conventional ones for dynamic responses.

• Journal of KSNVE
• /
• v.6 no.6
• /
• pp.851-856
• /
• 1996

The active noise cancellation system showing the effective convergence and stability has been studied by simplifying the controller structures using the weighting factors of control error path to the multi-channel filtered-x LMS algorithm which needs a lot of calculations and the performance has been verified experimentally. Besides, to implement the system performance in a vehicle cabin, experimental work for selecting the suitable numbers and positions of the microphones and speakers was accomplished. Effectively combining a TMS 320C 31 main processor conducting real number calculations and having various functions with other components, the purpose-built system board for active noise cancellation has been designed and with this board, car active noise cancellation system showing maximum stable 10dB noise reduction has been obtained at the car idling conditions above 3000rpm range.

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.1E
• /
• pp.64-70
• /
• 1994

Predictive neural network models are powerful speech recognition models based on a nonlinear pattern prediction. But those models suffer from poor discrimination between acoustically similar words. In this paper we propose an discriminative training algorithm for predictive neural network models. This algorithm is derived from GPD (Generalized Probabilistic Descent) algorithm coupled with MCEF(Minimum Classification Error Formulation). It allows direct minimization of a recognition error rate. Evaluation of our training algoritym on ten Korean digits shows its effectiveness by 30% reduction of recognition error.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.19 no.1
• /
• pp.38-44
• /
• 1995

We made a counter to measure the output of motor encoders for the motion error analysis of a CNC machining center, and have measured the dynamic characteristics and the position errors experimentally. Especially, we measured the radius errors for different feedrates and different radii when the CNC machining center performed a circular interpolation. We have also used an iterative learning method to reduce the radius errors and stick motion errors generated by the CNC machining center performing a circular interpolation. The results show that the proposed learning scheme can reduce the radius error and stick motion error significantly. The reduction of errors becomes more pronounced for higher feedrate and smaller radius.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.12
• /
• pp.108-115
• /
• 1994

Most pattern classifiers have been designed based on the ML (Maximum Likelihood) training algorithm which is simple and relatively powerful. The ML training is an efficient algorithm to individually estimate the model parameters of each class under the assumption that all class models in a classifier are statistically independent. That assumption, however, is not valid in many real situations, which degrades the performance of the classifier. In this paper, we propose a minimum-error-rate training algorithm based on the MAP (Maximum a Posteriori) approach. The algorithm regards the normalized outputs of the classifier as estimates of the a posteriori probability, and tries to maximize those estimates. According to Bayes decision theory, the proposed algorithm satisfies the condition of minimum-error-rate classificatin. We apply this algorithm to NPM (Neural Prediction Model) for speech recognition, and derive new disrminative training algorithms. Experimental results on ten Korean digits recognition have shown the reduction of 37.5% of the number of recognition errors.

• Proceedings of the IEEK Conference
• /
• 2000.07b
• /
• pp.1080-1083
• /
• 2000

In this paper, we show that a variation in sampling rate give rise to system performance degradation and propose a method to effectively reduce the error. We first capture the variation as a first order autoregressive (AR) model and project it as an additional sensor measurement noise. By considering that the sensor measurements include correlated noise, we perform a decorrelation process and then apply a standard Kalman filter (SKF) to estimate the target-state. As a result of the two-step procedure, we achieve a significant reduction in the target state estimation error.

• The Transactions of the Korean Institute of Electrical Engineers B
• /
• v.48 no.9
• /
• pp.517-522
• /
• 1999

This paper addresses the issue of current sampling in a high performance AC drive system fed by a discontinuous PWM inverter. The effect of the sampling error due to the measurement delay produced by an input stage low pass filter and an A/D converter is described in the case of discontinuous PWM. To compensate for the sampling error, a method to estimate the delay time of the whole measurement system based on the measured current is proposed and its effectiveness is verified by experimental results. The proposed algorithm can automatically estimate the system delay introduced by the low pass filter and the A/D converter at the commissioning stage. By delaying the current sampling by the estimated value, experimental results indicate that more than 50% reduction of current ripple can be achieved.

• Proceedings of the KIEE Conference
• /
• 1995.07a
• /
• pp.247-250
• /
• 1995

This paper proposes a new method to reduce the torque ripple in vector controlled inverter fed synchronous motor systems. In three phase voltage source inverter systems, all the three line currents are generally not measured and the currents of two lines are measured through two sensors and two A/D converters. The measured currents may contain some error due to the non-ideality of the current sensors and A/D converters, and the error coefficient of two line currents are not same. As a result, the developed torque contains the torque ripple. The proposed method can eliminate the torque ripple by setting the error coefficient to same value. To verify the proposed method, digital simulations are carried out.

• Journal of the Korean Vacuum Society
• /
• v.9 no.1
• /
• pp.1-6
• /
• 2000

An unexpected error is produced in calculating the transmission probability of a multipartite duct because of beam and shadow effects, if using a simple summation rule like the Oatley's equation. Particles moving in a tube are directed more or less towards the axis of the tube by the beam effect, and the length of a compound tube shortens virtually by the shadow effect originated from a reduction in the number of particles reaching the corner between two tubes of different cross-sections. Both effects make the transmission probability of the tube connected behind and consequently of the whole duct increase slightly. In this paper sources of the error in the calculation of the transmission probability are analyzed quantitatively and variations in the error depending on the dimensions of cylindrical tubes are calculated.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2005.11a
• /
• pp.760-763
• /
• 2005

Optimum design of the tooth surface for the reduction of transmission error is very difficult to determine analytically due to nonlinearity of transmission error under the several load condition. The design of tooth surface that can give a low noise under the various load condition is very important. Therefore, this study proposes the method to determine the optimal lead curve and robust design of the tooth surface by using the response surface method. To do so, the design variables are selected by a screening experiment. Then the fitted regression model Is built with the check of the usefulness of the model. The model with constraints is solved to obtain the optimum values for the lead curve and the robust design fur the tooth surface.