A Sequential Experimental Design Method Based on Binary Response Prediction

In industrial production, there are many problems concerning the experimental design of product performance. This kind of problem can be translated into the problem of studying the product performance boundary through binary response modeling. Collecting data through simulation for such experiments is computationally expensive, and conditions and time are limited. Therefore, a sequential experimental design method based on binary response prediction is proposed by collecting as little data as possible to achieve the purpose of the experiment. The initial experimental scheme is generated by the Maximum Projected Coordinate Exchange (MP-CE) algorithm, and the sequential process is combined with the local criterion based on the binary response prediction boundary and the global optimization method for sampling. Example verification results show that the proposed method is superior to random sampling and space-filled sampling. In addition, sensitivity analysis and some algorithm parameters are discussed.