A survey of sequential adaptive sampling strategy for transmission power loss measurement

Abstract

Laboratory experiments for characterizing the power loss behavior in transmissions causes time and energy costs when performing the traditional factorial design strategy. This is because the space filling strategy may locate redundant samples with low information regarding the measurement targets. This article proposes a sequential adaptive sampling methodology for performing efficient and informative power loss measurements. The presented sampling methodology associates surrogate modeling based on a Gaussian Process approach with Subset Simulation. Gaussian Process Regression creates a statistical model with the estimation targets and the expression of uncertainty. Subset Simulation is applied to efficiently identify the regions where an objective function reaches a predefined critical threshold. The proposed adaptive sampling method is implemented for the real-time measurement of a 7-speed DCT on a drivetrain test bench. Compared to the traditional factorial design, the iterative adaption of the proposed method ensures an informative and effective measurement and an automatic termination with reduced time and energy cost.