Fault Detection Using Nonlinear Low-Dimensional Representation of Sensor Data

TL;DR

This paper explores the use of nonlinear dimension reduction techniques like t-SNE and KPCA for fault detection in sensor data, improving interpretability and enabling edge processing in IoT systems.

Contribution

It introduces a novel approach combining nonlinear dimension reduction with anomaly detection for more effective fault detection in multivariate sensor data.

Findings

Nonlinear dimension reduction improves fault detection accuracy.

Low-dimensional representations enhance interpretability.

Method supports edge processing in IoT applications.

Abstract

Sensor data analysis plays a key role in health assessment of critical equipment. Such data are multivariate and exhibit nonlinear relationships. This paper describes how one can exploit nonlinear dimension reduction techniques, such as the t-distributed stochastic neighbor embedding (t-SNE) and kernel principal component analysis (KPCA) for fault detection. We show that using anomaly detection with low dimensional representations provides better interpretability and is conducive to edge processing in IoT applications.