AI predicts failures in wind turbines

A defect in a wind turbine component, such as the failure of an air blowing fan within a cooling system of a converter, can shut down energy production for weeks and calls for costly on-site repair work. Now, thanks to a joint project carried out by Proxima Solutions and CSEM, multiple AI algorithms have been developed that can early detect anomalies in components, so expensive machinery breakdowns and production downtime can be avoided. The algorithms can detect, for example, repetitive errors that cannot be remotely acknowledged. Their application was designed over one year based on the feedback and technical knowledge of technicians in the field. This collected knowledge was then reviewed by wind experts who co-operated with data scientists, so they could extract the right information and covert this via the AI algorithms into useful insights.

How it works

“Many functional failures in wind turbines take time to develop. Within that development period, “symptoms” such as anomalous temperature values or vibrations, can be observed in the sensor signals readings”

notes Baptiste Schubnel, Data Scientist at CSEM.

“Our software continuously analyzes readings from SCADA sensors in order to detect mechanical problems before they lead to major shutdowns.”

Finally, this whole process was grown organically based on the direct outputs of technical discussions between the field technicians, wind engineers and data scientists. The more anomalies are detected, the richer the root causes analyses are and the better the outputs will be.

Hailed as a breakthrough, this solution combines leading-edge data science methods with the technical knowledge of wind turbine engineers.

Launched on the market

“Of course, artificial intelligence won’t replace technicians and operators. Our objective is to support them while improving efficiency.”