Dr Juan Albarracín
SyMeCo project: “Enhancing Wind Turbine Condition Monitoring Performance Through Signal Super-resolution with Deep Generative Models”
Supervisor: Prof Conor Ryan
Host University: University of Limerick (UL)
Email: juan.albarracin@ul.ie
LinkedIn: https://www.linkedin.com/in/juan-albarrac%C3%ADn-124440142/
Dr. Juan Albarracín is a SyMeCo postdoctoral fellow with Lero@UL. Juan obtained his bachelor’s in computer engineering from Universidad Nacional de Colombia (2014) and his master’s and Ph.D. from Universidade Estadual de Campinas, Brazil (2023). He has extensive experience in both industry and academia in the field of machine learning and data science. He is specialised in first-generation deep generative models, and currently is working with the wind energy sector, applying machine learning and data science algorithms for wind turbine condition monitoring.
Juan’s SyMeCo research project, titled “Enhancing Wind Turbine Condition Monitoring Performance Through Signal Super-resolution with Deep Generative Models” aims to develop a Deep Generative Model for signal super-resolution, which can transform low-frequency SCADA data (Supervisory Control and Data Acquisition, typically 10-minute intervals) into synthetic high-frequency SCADA and vibration signals. This innovation aims to overcome the high cost and storage limitations associated with collecting high-frequency sensor data, thereby allowing for more accurate and widespread application of Machine Learning (ML) algorithms for CM tasks like fault prediction.
Project impact – The expected impact of the project is multifaceted, promising significant advancements in both the wind industry and the field of Artificial Intelligence. For the wind industry, the development of a Deep Super-resolution (SR) model that infers high-frequency signals from standard 10-minute Supervisory Control and Data Acquisition (SCADA) data offers two major advantages: cost reduction by eliminating the need for specialized high-frequency sensors and mitigating storage issues, as the high-frequency data can be generated under demand and deleted afterward. These benefits are expected to greatly reduce the costs associated with preventing wind turbine failures. Furthermore, high-frequency signals significantly enhance the accuracy of Condition Monitoring (CM) systems, and the project intends to conduct the first massive-scale study on the impact of high-frequency signals for CM of wind turbines by easing access to this data in real-world wind farms. From an Artificial Intelligence perspective, the research is groundbreaking because it extends the state of the art in generative models for super-resolution beyond typical perceptual data like images and audio to address the challenge of time series synthesis with operational value.
Interdisciplinary aspects: The project is inherently interdisciplinary, designed to merge cutting-edge Deep Generative Models (from the domain of Artificial Intelligence) with highly specialized knowledge in Wind Energy Engineering for Condition Monitoring (CM). The core task is to develop Deep Super-resolution models that transform low-frequency 10-minute SCADA signals into synthetic high-frequency signals, a process that extends Generative Models, which have traditionally been limited to perceptual data like images and audio, into the complex domain of operational time series. This involves tackling the technically different task of Cross Modality, generating vibration signals from SCADA data, which is completely unexplored in the current literature about time series generative models..