APPLICATION OF A PCA-BASED FAULT DETECTION AND DIAGNOSIS METHOD IN A POWER GENERATION SYSTEM WITH A 2 MW NATURAL GAS ENGINE
Based on increasing global energy demand, electric power generation from Internal Combustion Engines (ICE) has increased over the years. On this idea, the industries have adopted different methods and procedures to prevent failures in these engines, achieve an extension of the life cycle of the mach...
- Autores:
-
Cardenas, Yulineth
Carrillo, Gaylord
Alviz, Anibal
Alviz, Antistio
Portnoy, Ivan
Fajardo, Juan
Ocampo, Eric
Da-Costa, Edson
- Tipo de recurso:
- Fecha de publicación:
- 2022
- Institución:
- Universidad Tecnológica de Bolívar
- Repositorio:
- Repositorio Institucional UTB
- Idioma:
- eng
- OAI Identifier:
- oai:repositorio.utb.edu.co:20.500.12585/12270
- Acceso en línea:
- https://hdl.handle.net/20.500.12585/12270
- Palabra clave:
- Batch Process;
Fault Detection;
Canonical Variate Analysis
LEMB
- Rights
- openAccess
- License
- http://creativecommons.org/licenses/by-nc-nd/4.0/
| Summary: | Based on increasing global energy demand, electric power generation from Internal Combustion Engines (ICE) has increased over the years. On this idea, the industries have adopted different methods and procedures to prevent failures in these engines, achieve an extension of the life cycle of the machines, improve their safety, and provide financial savings. For this reason, this work implements a methodology for detecting and identifying failures in a natural gas engine (JGS 612 GS-N. L), based on the integration of Principal Component Analysis (PCA) and alarm streak analysis. A method used to describe a data set in terms of new uncorrelated variables or components. The components are ordered by the amount of original variance they describe, making the technique useful for reducing the dimensionality of a data set. Technically, PCA searches for the projection according to which the data are best represented in terms of least squares, using the T2 and Q statistics. In the initial stage, a PCA-based algorithm was developed to detect abnormal process trends and identify the variables of greater impact when these anomalies arise. In the next stage, an algorithm was developed and implemented, based on the analysis of alarm streaks, to identify the system’s behavior and thus classify f luctuations into either normal operating condition drifts or system failures. The application of the proposed methodology with real operation data of the engine (JGS 612 GS-N. L) shows that the method outperforms operators in detecting and identifying faults, as it performs these tasks considerably earlier than operators. © The Author(s) 2022. |
|---|