Machine Learning can predict failure of wind turbine components

July 12, 2019

To avoid unscheduled downtime and decrease unforeseen expenditures, wind farm operators can reduce the life-cycle cost of wind farms by up to 30% by reducing component failures and maintenance using machine learning models.

Some component failure can be up to 9 months in advance allowing much more forward planning to be carried out.

This early indication for component failure can reduce downtime, maintenance cost and increase component life and enables operators and managers to act with a plan instead of acting within a crisis to make informed maintenance decisions.

The very tough and rigorous conditions a wind turbine is put through, essentially being operated continuously, it is unsurprising that with all the thousands of mechanical and electronic components they will fail.

By monitoring SCADA data such as temperature profiles and detecting chronic components failures it is easy to plan to stock the right spares, schedule maintenance in time ( especially tricky with offshore wind farms where maintenance is hampered by the weather ) and reduce O&M (Operations & Maintenance) costs.

The Heimdall Machine Learning platform can analyse various streams of data, and unlike systems that use neural networks which claim to have 94% accuracy, won’t get overfitted. So if there are slight changes to the environment the predict models won’t fail. Also using GLM and making the models easily explainable on the weightings of each predictive function it is much easier for an operator and senior management to understand where failures are occurring and to have simple and transparent data and models.

If you are a wind farm operator or owner feel free to contact me at we would love to work with you to reduce your failures and cut costs.

Predictive Crime product launched
New product release - Heimdall