Predictive models of failures in wind turbines through machine learning techniques based on Artificial Intelligence
The objective of the WinDIAG project is to provide large-scale wind farm managers (“utility scale”) with a new generation of tools that allow wind turbines to anticipate and avoid outages, reduce downtime for maintenance and extend their useful life, without making additional investments for the detection and prediction of said failures.
To do this, functions based on Deep Learning are incorporated that allow operator learning and feedback in decision making. The proposed line of research is that of Reinforcement Learning, in which, in the form of “rewards” or “punishments” with the feedback of the system user (“operator”), it allows training for an accurate diagnosis. The use of Machine Learning is critical for the generation of training data, given the difficulty of having them for the same wind farm; the idea is to use advanced techniques of Adversarial Generative Neural Networks (GANs), for this generation.
The result is an analysis infrastructure and a set of algorithms that, based on the data from the SCADA systems for monitoring and controlling wind turbines, and after a process of configuration, data cleaning and training, are capable of detecting situations that are out of normal, performing a diagnosis based on the possible failure modes of the critical components of the wind turbines, and providing a recommendation for action by the O&M teams based on learning.
Implementation period: 2022 – 2024
Funding: Project financed by the public business entity Red.es, attached to the Ministry of Economic Affairs and Digital Transformation, charged to funds from the Recovery and Resilience Mechanism within the European Recovery Instrument (“Next Generation EU”), in accordance with the Recovery, Transformation and Resilience Plan
Call: 2021 Grants Aimed at Research and Development Projects in Artificial Intelligence and Other Digital Technologies and their Integration into Value Chains. /21-ED
Total budget: 535,948 €