• 품질경영학회지
• /
• 제40권4호
• /
• pp.481-496
• /
• 2012

Purpose: This paper aims at improving inefficiency of an existing posterior preference articulation method proposed for dual response surface optimization. The method generates a set of non-dominated solutions and then allows a decision maker (DM) to select the best solution among them through an interval selection strategy. Methods: This paper proposes an iterative posterior preference articulation method, which repeatedly generates the predetermined number of non-dominated solutions in an interval which becomes gradually narrower over rounds. Results: The existing method generates a good number of non-dominated solutions not used in the DM's selection process, while the proposed method generates the minimal number of non-dominated solutions necessitated in the selection process. Conclusion: The proposed method enables a satisfactory compromise solution to be achieved with minimal cognitive burden of the DM as well as with light computation load in generating non-dominated solutions.

• 지식경영연구
• /
• 제19권2호
• /
• pp.151-162
• /
• 2018

Response surface methodology (RSM) is one of popular tools to support a systematic improvement of quality of design in the product and process development stages. It consists of statistical modeling and optimization tools. RSM can be viewed as a knowledge management tool in that it systemizes knowledge about a manufacturing process through a big data analysis on products and processes. The conventional RSM aims to optimize the mean of a response, whereas dual-response surface optimization (DRSO), a special case of RSM, considers not only the mean of a response but also its variability or standard deviation for optimization. Recently, a posterior preference articulation approach receives attention in the DRSO literature. The posterior approach first seeks all (or most) of the nondominated solutions with no articulation of a decision maker (DM)'s preference. The DM then selects the best one from the set of nondominated solutions a posteriori. This method has a strength that the DM can understand the trade-off between the mean and standard deviation well by looking around the nondominated solutions. A posterior method has been proposed for DRSO. It employs an interval selection strategy for the selection step. This strategy has a limitation increasing inefficiency and complexity due to too many iterations when handling a great number (e.g., thousands ~ tens of thousands) of nondominated solutions. In this paper, a TOPSIS-based method is proposed to support a simple and efficient selection of the most preferred solution. The proposed method is illustrated through a typical DRSO problem and compared with the existing posterior method.

• 한국산학기술학회논문지
• /
• 제16권10호
• /
• pp.7061-7070
• /
• 2015

다중반응표면 최적화는 다수의 반응변수(품질특성치)를 동시에 고려하여 최적의 입력변수 조건을 찾는 반응표면분석의 세부 분야이다. 가중평균제곱오차(Weighted Mean Squared Error, WMSE) 최소화법은 평균제곱오차의 두 구성 요소인 제곱편차와 분산에 가중치를 부여한 WMSE를 활용하는데, 반응변수별로 WMSE를 구하여 이들을 종합적으로 최소화한다. 지금까지 WMSE 최소화법과 관련하여 개발된 기법은 대부분 의사결정자의 선호도 정보를 문제풀이 이전에 결정할 것을 요구하는 선호도사전제시법에 해당된다. 그러나 현실적으로 의사결정자가 자신의 선호도 정보를 사전에 정확히 제공하는 것은 매우 어렵다. 본 논문에서는 이러한 한계점을 개선하기 위하여 WMSE 최소화를 위한 선호도사후제시법을 제안한다. 제안된 방법은 의사결정자의 선호도 정보 없이 다수의 비지배적해를 생성한 후, 의사결정자가 생성된 비지배해 중 최고선호해를 선택하는 단계로 진행된다. 제안된 방법은 의사결정자로 하여금 전체 해집합의 트레이드오프 관계를 보다 폭넓은 시각으로 이해한 후 선호도 정보를 제시할 수 있도록 함으로써, 의사결정자의 선호도에 부합하는 최고선호해를 효과적으로 도출할 수 있다.

• 대한산업공학회지
• /
• 제43권3호
• /
• pp.164-175
• /
• 2017

Dual response surface optimization (DRSO) attempts to optimize mean and variability of a process response variable using a response surface methodology. In general, mean and variability of the response variable are often in conflict. In such a case, the process engineer need to understand the tradeoffs between the mean and variability in order to obtain a satisfactory solution. Recently, a Posterior preference articulation approach to DRSO (P-DRSO) has been proposed. P-DRSO generates a number of non-dominated solutions and allows the process engineer to select the most preferred solution. By observing the non-dominated solutions, the DM can explore and better understand the trade-offs between the mean and variability. However, the non-dominated solutions generated by the existing P-DRSO is often incomprehensive and unevenly distributed which limits the practicability of the method. In this regard, we propose a modified P-DRSO using multiple objective genetic algorithms. The proposed method has an advantage in that it generates comprehensive and evenly distributed non-dominated solutions.