求翻译：In this work, projection pursuit regression (PPR) has been chosen as the modeling algorithm for fault diagnosis tasks. PPR is a multivariate statistical technique (Friedman & Stuetzle, 1981; Hwang, Lay, Maechler, Martin, & Schimert, 1994; Utojo & Bakshi, 1995), ideally suited for nonlinear systems. Flick, Jones, Pries and Herman (1990) have applied this technique to a simple fault detection problem in a mathematical sense. Similar to other multivariate methods it is based on the decomposition of the inputs along principal components. However the basis functions referred to as hidden functions are not fixed a priori but determined by the training data and the output calculation is based on a nearest neighborhood approach applied in the hidden function space. In previous work by Lou, Budman, and Duever (2002), PPR has been found to have a good trade-off, compared to other techniques, in dealing with extrapolation error due to insufficient training data, and sensitivity to noise. This paper deals with two new methods to design optimal experiments for the training of a PPR-based fault diagnosis algorithm. The objective is to design experimental data in some predetermined window of operating conditions to minimize fault misclassification during testing. The incentive is to minimize the overall number of experiments for the sake of economy. Three different sets of data are considered in this study: (1) an initial training data set to be referred to as data set #1 is obtained using a conventional factorial design to obtain a preliminary model; (2) a second training data set #2 is designed based on the knowledge of the first data set; (3) finally, a testing data set is generated to assess the classification accuracy of the algorithm trained with the two previous training data sets. The current work presents two methodologies to design the second training data set. Both proposed design methods follow a sequential approach where the optimal design of training data set #2 is based on a priori knowledge of training data set #1.是什么意思？

待解决 悬赏分：1 - 离问题结束还有