Support Vector Machine-based Modified Sp Statistic for Subset Selection with Non-Normal Error Terms

Support vector machine (SVM) is used for estimation of regression parameters to modify the sum of cross products (Sp). It works well for some nonnormal error distributions. The performance of existing robust methods and the modified Sp is evaluated through simulated and real data. The results show the performance of the modified Sp is good.

An improved path discretization method for automated fiber placement

In the automated fiber placement process, the continuous placement paths need to be discretized into a finite number of path points because the laying head cannot continuously trace the predetermined curved path. However, the discretization of the placement path, which is a spatial curve, will inevitably introduce error. In this paper, an improved path discretization algorithm is proposed for the fiber placement of complex double-curved structures. Firstly, the discrete error was decomposed into normal direction and binormal direction, and they are correlated with the laying process and their influences on the laying quality are discussed, respectively. Secondly, the relationship between the binormal error and the overlap of the tow is analyzed with differential geometry, and the influence of the normal error on laying force is discussed by the pressure experiment and the finite element method. Finally, the improved path discretization algorithm has been verified on double-curved surface and compared with the traditional path discrete algorithms. The results showed that the number of discrete path points decreases by 45.8% on average compared with the chordal deviation discretization algorithm and by 63.1% compared with the equal-arc discretization algorithm.

A Bayesian Logistic Growth Model for the Spread of COVID-19 in New York

We use Bayesian Estimation for the logistic growth model in order to estimate the spread of the coronavirus epidemic in the state of New York. Models weighting all data points equally as well as models with normal error structure prove inadequate to model the process accurately. On the other hand, a model with larger weights for more recent data points and with t-distributed errors seems reasonably capable of making at least short term predictions.