The development of statistical models to detect degradation and predict failures of Wind Turbines (WT) components was a key step in reducing the cost of O&M of wind farms. IBM further (a) developed feature sets from the collected data, including the results of the physical models; (b) deployed a selected set of Machine Learning (ML) models e.g., LSTMs; and (c) trained and evaluated the most appropriate models.
The ML models allowed to improve the capacity of detecting degradation patterns, as well as the capacity of predicting failures in the following key components of mentioned WTs: Gearbox, converter, generator, blade bearing, main shaft, main transformer, and pitch system.
Oceans, seas and watersAddressing target audiences and expressing needs
- Collaboration
Potential early adopters and/or customers:
– Wind Farm operator
– O&M suppliers and their partners
- Public or private funding institutions
- Academia/ Universities
- Private Investors
R&D, Technology and Innovation aspects
This result achieved a TRL-7 “actual system proven in operational environment”. Next steps: broad scale deployment of the ML models in the offshore wind industry will require pilot projects which demonstrate their benefits.
Since most components of our ML pipeline can be easily re-used for other wind turbines, and since through ROMEO we showed the possibility to deploy these models on IBM Cloud. The scalability of such a solution would only mean re-configuring the cloud deployment environment to provision more or less resources as needed and therefore accommodate different windfarms almost seamlessly.
The bulk of the models developed focus on Anomaly Detection for offshore wind turbines. These models can be re-used by retraining them using SCADA data that represents normal behavior. This can be done seamlessly without any modifications to the models themselves, therefore any other wind turbine that uses SCADA can re-use these models. Furthermore, any results which are generated can be back tested given historical SCADA data accompanied with normal and abnormal operation periods, hence easily replicating and confirming the results.
Given the enhanced cost-efficiency described above: The deployment of a new generation of key components of wind turbines will allow improved performance of new and existing off-shore wind power plants and therefore reduce the cost of energy.
