RT Journal Article
T1 An Agentic Artificial Intelligence Observer for Predictive Maintenance in Electrolysers
A1 Abiodun Abiola, 
A1 Barragán Piña, Antonio Javier
A1 Andújar Márquez, José Manuel
A1 Segura Manzano, Francisca
K1 Electrolysis plant
K1 predictive maintenance
K1 artificial intelligence-based observer
K1 fuzzy system
K1 long short-term memory (LSTM)
K1 neural network
AB This paper presents an artificial intelligence (AI)-based observer that combines fuzzy logic and neural networks to detect abnormalities in sensors embedded in an electrolyser. Electrolysers are hydrogen production plants that require effective maintenance to guarantee suitable operation, prevent degradation, and avoid loss of efficiency. In this sense, predictive maintenance arises as one of the most advisable techniques for maintenance in electrolysers by using sensor data to predict potential abnormalities. However, if the sensor fails, there will be an incorrect forecasting of abnormalities. Among the different types of operational faults that sensors can present are drift-related faults, which are probably the most difficult to detect due to a slow but progressive loss of accuracy in measurements. Another problem with predictive maintenance is that it often requires enormous training data, which is not available at the early stage of plant operation. The developed fuzzy system is responsible for detecting faulty readings arising from drift sensor signals, while the neural network complements the function of the fuzzy system by predicting sensor signals when enough training data are available. The AI-based observer and the fuzzy rules are validated in an experimental case study with a 1 Nm 3 /h electrolyser. The selected variables are electrolyser temperature and efficiency. Experimental results show that the rules of the fuzzy component of the AI-based observer guarantee an accuracy of ±0.25 within the range of 0 to 1, and the neural network component predicted correct sensor values with a root mean square error (RMSE) as low as 0.0016. The authors’ approach helps to determine drift faults without additional sensors or components installed in the plant.
PB The Tech Science Press
SN 1526-1492
YR 2026
FD 2026
LK https://hdl.handle.net/10272/28084
UL https://hdl.handle.net/10272/28084
LA eng
NO This work has been partially carried out thanks tothe supportof (1) GrantRef. Ref. PID2023- 148456OB-C41 and (2) GrandRef. RED2022-134588-T found bi MICIU/AEI/10.13039/501100011033
DS Repositorio Institucional de la Universidad de Huelva
RD 30 may 2026