Fault Detection of Cooling System Based on Long Short-Term Memory at Nam Ngum 1 Hydropower Plant

• Published in: International Conference on Advanced Mechatronic Systems pp. 77 - 81
• Main Authors: Thansouphanh, Bounpone, Premrudeepreechacham, Suttichai, Ngamsanroaj, Kanchit, Srirattanawichaikul, Watcharin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.11.2024
• Subjects: Accuracy; Cooling; Deep Learning; Fault detection; Feature extraction; Hydroelectric power generation; Load modeling; Long short term memory; LSTM; Maintenance; Nam Ngum-1 Hydropower Plant; Predictive Maintenance; Support vector machines; Synchronous Generator; Synchronous generators
• ISSN: 2325-0690
• DOI: 10.1109/ICAMechS63130.2024.10818838

Synchronous generators play a vital role in hydropower plants, making reliable cooling systems essential to prevent severe performance issues and potential failures. Traditional maintenance methods for these cooling systems often rely on periodic inspections or time-based strategies, which may be limited in their ability to detect subtle or evolving faults. This paper introduces a deep-learning approach using Long Short-Term Memory (LSTM) networks for fault detection in the cooling system of the Nam Ngum-1 (NNG-1) hydropower plant in Lao PDR. The methodology involves preprocessing historical operational data to extract key features, which are then used to train the LSTM network. This network is designed to learn complex temporal patterns associated with various fault conditions. The model's effectiveness is assessed through accuracy, precision, and recall metrics. The proposed approach not only enhances fault detection but also minimizes downtime, and optimizes maintenance schedules, ultimately improving the overall reliability and efficiency of the hydropower plant.