Dynamic robust design with missing data

Robust parameter design is conventionally analyzed by means of statistical techniques. However, the statistical‐based approach is inefficient when optimizing a dynamic system in regards to quantitative control factors and missing observations. The aim of this paper is to propose an alternative approach based on data mining tools to model and optimize dynamic robust design with missing data.

A three‐phase approach based on data mining techniques is proposed. First, a back‐propagation network is trained to construct the response model of a dynamic system. Second, three formulas of performance measures are developed to evaluate the predicted responses of the response model. Finally, a genetic algorithm is then performed to obtain the optimal parameter combination via the response model.

The proposed approach is capable of dealing with both qualitative and quantitative control factors for dynamic systems as well as static systems. In addition, the proposed approach can efficiently treat parameter experiments with missing data. The proposed approach is demonstrated with a numerical example. Results show that this three‐phase data mining approach outperforms the conventional statistic‐based approaches.

This work provides a relatively effective approach to optimize the three types of dynamic robust parameter design. Performing the approach, practitioners do not require much background in statistics but instead rely on their knowledge of engineering.