Multivariate fault detection with convex hull

Abstract

We propose a multivariate trending for aircraft fault detection. Multivariate trending generate fault indicators using output sensor data, is one of black-box approach. We use convex polygon for the computation of a rough shape or extent of the normal data set. Quickhull algorithm is used for the hull finding because it is simpler and uses less memory. It is assumed that the normal data points are in general position, so that their convex hull is a simple complex. We represent a d-dimensional convex hull by its vertices and (d-1)-dimensional faces. From multivariate trend analysis, if we find the measurements have the tendency to leave the convex polygon, this measurement can be labeled as a fault. If a new point is above all hyperplane of the convex hull, it is outside the convex polygon.