Sensors Validation Based on Bayesian Classifiers

Abstract

Validation of Sensors has very important effects on the consequences of structural experiments and subsequent analyzing works. This article focus on the problem that if the data collected from the sensors are valid or not. It tested the validation of an target acceleration sensor on a truss structure by using Naive Bayesian Classifier and Tree Augmented Naive Bayesian Classifier which are based on machine learning technology whose theory basis is probability statistics. In the course of data analyzing, the theoretical values modified by Finite Element Modeling are taken as an criterion of data collected from sensors. The continuous data are discretized by several different discretization methods. Both of the classifiers are created by discretized training data and used to test the validation of the specified sensor. The comparison between two experiments based on NBC and TAN is presented. It is proved that both the testing methods are effective.