Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Artificial intelligence for fault isolation in wind energy conversion systems
Submission Author:
César Tadeu Nasser Medeiros Branco , SC , Brazil
Co-Authors:
César Tadeu Nasser Medeiros Branco, Jonny Silva
Presenter: César Tadeu Nasser Medeiros Branco
doi://10.26678/ABCM.COBEM2023.COB2023-1348
Abstract
Installed global wind capacity grows steadily each year in an effort to harness wind power and generate clean electricity from renewable energy sources. A trend in offshore wind turbine installations presents new challenges for operation and maintenance teams, aiming to enhance reliability and minimize downtime. The identification and classification of faults in wind energy conversion systems are crucial factors in extending the lifespan of wind turbines and launch bases by isolating faults based on failure modes. Geared drivetrain systems in offshore wind turbine facilities are being replaced by direct-drive systems, where the rotor and multi-pole synchronous generators are coupled together. Due to the harsh environment, the generator is more susceptible to extended downtime, with the pitch control system exhibiting one of the highest failure rates. Thus, this study aims to develop an intelligent decision-making system to classify the types of faults present in the pitch control system. The dataset was generated using AMESim software, simulating both healthy and faulty signals. The variables used in the fault modeling included pitch angle, rotor speed, and active power. The pitch system was modeled using a PI controller, while the permanent magnet synchronous generator was represented using the dq0 representation of the three-phase system. Four different types of faults were simulated within the wind energy conversion system, comprising three faults in the speed sensor and one in the pitch actuator. In terms of artificial intelligence, a knowledge-based approach was developed using object-oriented programming and compared to three machine learning algorithms: random forest, gradient boosting, and k-nearest neighbor. All four faults were effectively identified and isolated by the rule-based system, demonstrating an accuracy of 95%, slightly lower than the machine learning algorithms. Thus, the knowledge-based system proved to be proficient in fault detection and explanation, comparable to ML models. However, the fact that a knowledge-based system has the ability to explain its solutions, that approach fosters the exploration of new avenues in diagnosing faults in wind turbines.
Keywords
Artificial Intelligence, wind turbine, Fault modeling, Fault Detection
