Unsupervised Fault Detection Based on Laplacian Score and TEDA

Abstract

The drawback to Typicality and Eccentricity Data Analytics(TEDA), a classic unsupervised learning algorithm, is that TEDA requires strict priori knowledge during the stage of data preprocessing. In view of the disadvantage, a method of unsupervised fault detection called Laplacian Score with TEDA (LS-TEDA) is proposed. Features are selected by LS and unsupervised fault detection is realized by using TEDA in this method. LS-TEDA has been applied with Lublin Sugar Factory and the result shows high accuracy in fault detection.