A deep learning-based strategy for fault detection and isolation in parabolic-trough collectors

Solar plants are exposed to the appearance of faults in some of their components, as they are vulnerable to the action of external agents (wind, rain, dust, birds …) and internal defects. However, it is necessary to ensure a satisfactory operation when these factors affect the plant. Fault detection...

Descripción completa

Detalles Bibliográficos
Autores: Ruiz-Moreno, Sara, Sánchez, Adolfo J., Gallego Len, Antonio Javier, Camacho, Eduardo F.
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2022
País:España
Institución:Universidad de Sevilla (US)
Repositorio:idUS. Depósito de Investigación de la Universidad de Sevilla
OAI Identifier:oai:idus.us.es:11441/133884
Acceso en línea:https://hdl.handle.net/11441/133884
https://doi.org/10.1016/j.renene.2022.01.029
Access Level:acceso abierto
Palabra clave:Solar energy
Parabolic-trough collectors
Artificial intelligence
Fault detection
Fault diagnosis
Descripción
Sumario:Solar plants are exposed to the appearance of faults in some of their components, as they are vulnerable to the action of external agents (wind, rain, dust, birds …) and internal defects. However, it is necessary to ensure a satisfactory operation when these factors affect the plant. Fault detection and diagnosis methods are essential to detecting and locating the faults, maintaining efficiency and safety in the plant. This work proposes a methodology for detecting and isolating faults in parabolic-trough plants. It is based on a three-layer methodology composed of a neural network to obtain a preliminary detection and classification between three types of fault, a second stage analyzing the flow rate dynamics, and a third stage defocusing the first collector to analyze thermal losses. The methodology has been applied by simulation to a model of the ACUREX plant, which was located at the Plataforma Solar de Almería. The confusion matrices have been obtained, with accuracies over 80% when using the three layers in a hierarchical structure. By forcing all the three layers, the accuracies exceed 90%.