Optimal Experimental Design for Training of a Fault Detection Algorithm

Abstract

Abstract This paper focuses on Optimal Experimental Design to train a Projection Pursuit Regression (PPR) model used for fault detection. A novel experimental design method, referred to as Gaussian Probability Design, is proposed and compared with the conventional Factorial Design. The Gaussian Probability Design automatically searches for the sparseness of the data, and adds pairs of training data on both sides of a class boundary in areas where the data density is the lowest. This design method outperforms the Factorial Design in reducing the fault misclassification more effectively with the same amount of new training data.