VENTUS - Causal, Probabilistic and Physics-Informed Machine Learning for Diagnosis and Predictive Maintenance in Wind Turbines

Austrian's wind power expansion target for 2030 demands newly installed wind power capacity of 500 MW/year. At the same time, it is known that operation and maintenance of wind turbines can cost up to 32% of the total cost of energy, calling for radical new approaches towards AI, digital twins and predictive maintenance. In the VENTUS project, we follow recent advances in the fields of physics-informed AI and probabilistic-causal AI for various reasons: first these approaches allow to augment traditional data-based learning with (physical, causal) domain knowledge, leading to often dramatic higher data efficiency and transferability to new scenarios, e.g. different types of turbines. Additionally, these approaches show a substantially higher degree of explainability than traditional AI systems. Based on a fail-case and performance-degradation analysis conducted together with the relevant stakeholder, we will target an explainable AI system with the potential to reduce losses due to downtime and maintenance by 50%.