• Journal of Magnetics
• /
• v.19 no.4
• /
• pp.385-392
• /
• 2014

Uncertainties in loads, materials and manufacturing quality must be considered during electromagnetic devices design. This paper presents an effective methodology for robust optimization design based on the variance decomposition in order to keep higher accuracy of the robustness prediction. Sobol' theory is employed to estimate the response variance under some specific tolerance in design variables. Then, an optimal design is obtained by adding a criterion of response variance upon typical optimization problems as a constraint of the optimization. The main contribution of this paper is that the proposed method applies the variance decomposition to obtain a more accurate variance of the response, as well save the computational cost. The performance and robustness of the proposed algorithms are investigated through a numerical experiment with both an analytic function and the TEAM 22 problem.

• Journal of the Korean Statistical Society
• /
• v.24 no.2
• /
• pp.563-568
• /
• 1995

It is well-known that the sample mean and the sample variance are jointly sufficient under normality assumption. In this paper a proof of the joint sufficiency is given without using the factorization criterion. It is related to a finite Sanov-type conditional theorem, i.e., the conditional probability density of $Y_1$ given sample mean $\mu$ and sample variance $\sigma^2$, where $Y_1, Y_2, \cdots, Y_n$ are independently and identically distributed (i.i.d.) normal random variables with mean m and variance $\delta^2$, equals that of $Y_1$ given sample mean $\mu$ and sample variance $\sigma^2$, where $Y_1, Y_2, \cdots, Y_n$ are i.i.d. normal random variables with mean $\mu$ and variance $\sigma^2$.

• Journal of the Korean Statistical Society
• /
• v.24 no.1
• /
• pp.31-44
• /
• 1995

In fitting a model, there always exists a discrepancy between the fitted model and the true functional relationship. In measuring this discrepancy, Box and Drapper (1959) used the criterion dividing the discrepancy into two parts which are the bias error part and the variance error one in single response case. In this paper, an optimum design which makes these two types of errors as small as possible is found by extending the Box and Drapper criterion to multiple response situation. Especially, a design is found to meat rotatability conditions when we fit a quadratic model to each response fearing cubic bias. Using the central composite design, an application of general results to a specific case is shown to help understanding the material.

• International Journal of Reliability and Applications
• /
• v.17 no.1
• /
• pp.21-49
• /
• 2016

Many mechanical, electrical and electronic products have more than one performance characteristics (PCs). For example the performance degradation of rubidium discharge lamps can be characterized by the rubidium consumption or the decreasing intensity the lamp. The product may degrade due to all the PCs which may be independent or dependent. This paper deals with the design of optimal bivariate step-stress partially accelerated degradation test (PADT) with degradation paths modelled by gamma process. The dependency between PCs has been modelled through Frank copula function. In partial step-stress loading, the unit is tested at usual stress for some time, and then the stress is accelerated. This helps in preventing over-stressing of the test specimens. Failure occurs when the performance characteristic crosses the critical value the first time. Under the constraint of total experimental cost, the optimal test duration and the optimal number of inspections at each intermediate stress level are obtained using variance optimality criterion.

• Communications for Statistical Applications and Methods
• /
• v.14 no.1
• /
• pp.11-21
• /
• 2007

In this paper, we consider a decision-theoretic oriented, objective Bayesian inference for the difference between two normal means with unknown com-mon variance. We derive the Bayesian reference criterion as well as the intrinsic estimator and the credible region which correspond to the intrinsic discrepancy loss and the reference prior. We illustrate our results using real data analysis as well as simulation study.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.43 no.9 s.351
• /
• pp.149-157
• /
• 2006

Turbo code, a kind of error correction coding technique, has been used in the field of digital mobile communication system. As the number of iterations increases, it can achieves remarkable BER performance over AWGN channel environment. However, if the number of iterations is increased in the several channel environments, any further iteration results in very little improvement, and requires much delay and computation in proportion to the number of iterations. To solve this problems, it is necessary to device an efficient criterion to stop the iteration process and prevent unnecessary delay and computation. In this paper, it proposes an efficient and simple criterion for stopping the iteration process in turbo decoding. By using variance values of LLR in turbo decoder, the proposed algerian can largely reduce the average number of iterations without BER performance degradation in all SNR regions. As a result of simulation, the average number of iterations in the upper SNR region is reduced by about $34.66%{\sim}41.33%$ compared to method using variance values of extrinsic information. the average number of iterations in the lower SNR region is reduced by about $13.93%{\sim}14.45%$ compared to CE algorithm and about $13.23%{\sim}14.26%$ compared to SDR algorithm.

• Journal of the Korean Data and Information Science Society
• /
• v.17 no.2
• /
• pp.291-300
• /
• 2006

In this research, the performance of widely used Bayesian estimators such as Bayes estimator, empirical Bayes estimator, constrained Bayes estimator and constrained empirical Bayes estimator are compared by means of a measurement under balanced loss function for the typical normal-normal situation. The proposed measurement is a weighted sum of the precisions of first and second moments. As a result, one can gets the criterion according to the size of prior variance against the population variance.

• Journal of the Korean Statistical Society
• /
• v.25 no.2
• /
• pp.277-288
• /
• 1996

The minimum aberration criterion is commonly used for selecting good fractional factorial designs. In this paper we give same necessary conditions for $3^{n-k}$ fractional factorial designs. We obtain minimum aberration $3^{n-k}$ designs for k = 2 and any n. For k > 2, minimum aberration designs have not found yet. As an alternative, we select a design with minimum aberration among minimum-variance designs.

• Journal of the Korean Data and Information Science Society
• /
• v.16 no.2
• /
• pp.371-382
• /
• 2005

The paper compares the performance of some widely used Bayesian estimators such as Bayes estimator, empirical Bayes estimator, constrained Bayes estimator and constrained Bayes estimator by means of a new measurement under squared error loss function for the typical normal-normal situation. The proposed measurement is a weighted sum of the precisions of first and second moments. As a result, one can gets the criterion according to the size of prior variance against the population variance.

• Speech Sciences
• /
• v.7 no.4
• /
• pp.11-26
• /
• 2000

In this paper, we propose the HMM with the MIN module. Because initial and re-estimated variance vectors are important elements for performance in HMM recognition systems, we propose a method which compensates for the mismatched statistical feature of training and test data. The MIN module function is a differentiable function similar to the sigmoid function. Unlike a continuous density function, it does not include variance vectors of the data set. The proposed hybrid HMM/MIN module is a unified network in which the observation probability in the HMM is replaced by the MIN module neural network. The parameters in the unified network are re-estimated by the gradient descent method for the Maximum Likelihood (ML) criterion. In estimating parameters, the variance vector is not estimated because there is no variance element in the MIN module function. The experiment was performed to compare the performance of the proposed HMM and the conventional HMM. The experiment measured an isolated number for speaker independent recognition.