Fault detection using online selected data and updated regression models

Doris Schadler, Ernst Stadlober

Research output: Contribution to journalArticlepeer-review

Abstract

Since their increasing complexity and the impossibility of monitoring the data manually, fault detection is of special interest at engine test beds. The work with research data presents yet another challenge. First, it is hard to establish physical models, at least for all variables of interest. Second, prior knowledge of all data settings is not available. Hence, this paper introduces statistical models with no a priori knowledge which are updated online to adapt to new data settings. In order to reduce the amount of time and memory required, incoming data is filtered according to different criteria. Several data selection criteria based on well-known statistical measures such as leverage or Cook's distance have been tested on four different data sets. Our new proposal, a proper combination of the leverage measure and the forecast residual, performs the best. The models are still applicable to forecasting yet even more suitable for error detection.

Translated title of the contributionFehlererkennung durch online selektierte Daten und aktualisierte Regressionsmodelle
Original languageEnglish
Pages (from-to)437-449
Number of pages13
JournalMeasurement
Volume140
DOIs
Publication statusPublished - Jul 2019

Keywords

  • Data selection
  • Fault detection
  • Online model updating
  • Regression model

ASJC Scopus subject areas

  • Statistics and Probability
  • Mechanical Engineering
  • Instrumentation
  • Electrical and Electronic Engineering

Fields of Expertise

  • Information, Communication & Computing

Treatment code (Nähere Zuordnung)

  • Experimental

Fingerprint

Dive into the research topics of 'Fault detection using online selected data and updated regression models'. Together they form a unique fingerprint.

Cite this