Tecnalia – WTURBO Wind Turbine Power Train Digital Twin
Tecnalia
Sector: Energy
Business Case
Operation and maintenance (O&M) costs represent 25 to 35% of the LCOE (levelised cost of energy) of wind turbines. Corrective maintenance is responsible for 30 to 60% of O&M costs. This challenge involves developing a digital twin for the electrical components within the power train of a doubly fed wind turbine. In this case, onshore wind. Using the digital twin, synthetic defect data can be generated under different operating conditions, and diagnostic classifiers can be trained to diagnose defects in situations that have not yet occurred. The diagnosed issue pertains to the cooling of the electric generator, causing a temperature impact on the stator windings.
Objectives
Characterise and learn theoretical and real machine behaviour. Identify stator winding overheating events and analyse the cause. Diagnose problems in the generator cooling system. Minimise maintenance costs.
Use case
The solution is based on designing a methodology for the development of digital twins. We understand digital twins as mathematical models that represent physical phenomena fed back with real operating data. The improvement of such physical models is achieved through surrogate optimisation techniques. The improved models allow simulating scenarios under normal and defect conditions and generating synthetic data. The generation of synthetic data is enriched by statistical models or other techniques such as adversarial generation networks (AGN). The diagnostic classifier trained with real operating and synthetic data is easily replicable through transfer learning.
Infrastructure
Hybrid On premise Cloud
Technology
Machine learning and deep learning RPA
Data
Static wind turbine power train design data. Time series of SCADA operating data from doubly fed asynchronous generator (DFIG) wind turbines: Time series of generator stator temperature data. Labelled faults and maintenance actions concerning the refrigeration system.
Resources
Multidisciplinary team of: domain experts (wind, electric machine), data science experts, data engineers, knowledge of the wind business model.
Difficulties and learning
Difficulties include the availability of wind turbine operation and maintenance data, the selection of the specific defect type to focus on in the use case study, the coordination of a multidisciplinary team, and the design of the methodology for developing the digital twin.
KPIs (business impact and metrics of the model)
KPIs validated in different ENGIE wind farms: OPEX: Reduced maintenance costs at portfolio level: €5M. Diagnosis: 10% reduction of false positives. ADDITIONALLY, it facilitates troubleshooting (knowledge of cause-effect defect). Precision of the digital twin (MAPE): Active power 2.33%, current 2.66%, stator winding temperature 4.33%.
Funding
Elkartek VIRTUAL. EU-H2020 PLATOON. https://platoon-project.eu/ TECNALIA’s own funds.
Collaborators, Partners
ENGIE. TECNALIA’s own development (patented). It is being transferred to several companies in the sector.
