A fault detector/classifier for closed-ring power generators using machine learning

• Published in: Reliability engineering & system safety Vol. 212; p. 107614
• Main Authors: Quintanilha, Igor M., Elias, Vitor R.M., da Silva, Felipe B., Fonini, Pedro A.M., da Silva, Eduardo A.B., Netto, Sergio L., Apolinário, José A., de Campos, Marcello L.R., Martins, Wallace A., Wold, Lars E., Andersen, Rune B.
• Format: Journal Article
• Language: English
• Published: Barking Elsevier Ltd 01.08.2021; Elsevier BV
• Subjects: Algorithms; Classification; Condition-based monitoring; Detection; Fault detection; Information processing; Learning algorithms; Machine learning; Machinery condition monitoring; Principal components; Principal components analysis; Random forests; Reliability engineering; Condition-based monitoring; Detection; Random forests; Classification; Machine learning; Principal components
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2021.107614

Condition-based monitoring of power-generation systems is naturally becoming a standard approach in industry due to its inherent capability of fast fault detection, thus improving system efficiency and reducing operational costs. Most such systems employ expertise-reliant rule-based methods. This work proposes a different framework, in which machine-learning algorithms are used for detecting and classifying several fault types in a power-generation system of dynamically positioned vessels. First, principal component analysis is used to extract relevant information from labeled data. A random-forest algorithm then learns hidden patterns from faulty behavior in order to infer fault detection from unlabeled data. Results on fault detection and classification for the proposed approach show significant improvement on accuracy and speed when compared to results from rule-based methods over a comprehensive database. •Condition-based monitoring of power systems in dynamic positioning vessels.•Fault detection/classification using machine learning techniques.•Significant improvement in accuracy and speed when compared to rule-based methods.•Proposed system easily updated for new data, independently of expert knowledge.•Deployed database emulating real power generation systems.